Issuu on Google+



The Nature of Robots R o b o t i c

F a b r i c a t i o n

i n

A r c h i t e c t u r e

CONTENTS // Part I - keynotes Robots and Architecture...................6 Computation Or Revolution............12 Changing Building Sites.................21 Enterpreneurship In Architecture Robotics..........................................28 Oddico Formwork Robotics............34 Robofold And Robots.IO.................36 Machineous....................................38 Rob Technologies...........................40 Greyshed........................................42 Part II - Research and Projects Building a Bridge with Flying Robots..................................45

The Nature of Robots

Autonomus Robotic Assembly With Variable Material Propties......52


An Integrated Modelling and Toolpathing Approach for a Frameless Stressed Skin Structure, Fabricated Using Robotic Incremental Sheet Forming...........................................59 Robotic Lattice Smock....................68

Robotic Multu-Dimensional Printing Based On Structural Performance...................................75 Fabric Forms: The Robotic Positioning Of Fabric Formwork.....82 Path Planning For Robotic Artistic Stone Surface Production...............89 Towards A Macro Design Of Acustic Surfaces.............................97 Robotic Hot-Blade Cutting............105 Fabrication-Avare Design Of Timber Folded Plate Shells With Double Through Rendon Joints....113 Rbdm_Robodome.........................120 Topology Optimization And Robotic Fabrication Of Advanced Timber Space-Frame Structure.......................................126 Mobile Robotic Brickwork.............133 Closeness: On The Relationship Of Multi-Agent And Robotic Fabrication....................................140 The SPIDERobot: A Cable-Robot System for On-Site Construction in Architecture...................................147 Developing Architectural Geometry Through Robotic Assembly And Material Sensing...153

Botbar: A Platform For Multy-Disciplinary Design Education......................................158

Part III - Real-Time Control

RECONstruction...........................166 Robotics-Based Prefabrication In

Towards On-Site Collaborative Robotics..................235


Stigmergic Accretion.....................242

Stereotomy Of Wave Joined Blocks................................177 Crafting Robustness: Rapidly Fabricating Ruled Surface Acoustic Panels............................183 From Analysis To Production And Back Attempts Result Of Reusable Adaptive Freeform Production Strategies For Double Curved Concrete Construction Elements..189 Free Form Clay Deposition In Custom Generated Molds.............195

Robot UI.......................................229

Sensor And Workflow Evolutions: Developing a Framework for Instant Robotic Toolpath Revision.........................250 Towards Real-Time Adaptive Fabrication-Avare Form Finding In Architecture..................259 Temporary conclusion The Robot...................267

Materially Informed Design To Robotic Production: A Robotic 3D Printing System For Informed Material Deposition.......................207 Robotics-Enabled Stress Line Additive Manufacturing..........214 Build-Ing The Mass Lo-Fab Pavillion.........................................221

The Nature of Robots

Solar Bytes Pavillion.....................201


ACKNOWLEDGEMENTS // The Author gratefully acknowledge the people who gave their permission to reproduce material in this book. While every effort has been made to contact copyright holders for their permission to reprint material, the Author would be greatful to hear from any copyright holder who is not acknowledged here and will untertake to rectify any errors or omission in future editions. Editorial Note on Presentation and Editing of Texts // This anthology aims to be inclusive, therefore some texts must appear in abbreviated form in the course of presenting others in their enirety. Some texts have been edited to shorten them, to exlude references that require more space for a full explanation and to preserve the flow of argument This is a collective work. The credits are indicated at the beginning of each text. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by anu means, electronic, mechanical, photocopying, recording or otherwise, exept as permitted by the UK Copyright, Design and Patents Act 1998, without the prior permission of the Author. This publication is designed to provide accurate and authorative information in regard to the subject matter covered.

This etition first published 2016

The Nature of Robots

ISBN-13: 978-1534931596 ISBN-10: 1534931597


The Nature of Robots

Part I - Keynotes


ROBOTS and ARCHITECTURE Experiments, Fiction, Epistemology Credits // Antonie Picon //

The Nature of Robots

ABSTRACT // How ready we are to receive robots on our building sites? Antonie Picon, G Ware Travel stead Professor of the History of Architecture and Technology at Harvard Graduate School Design (GSD), highlights the mixed cultural reception that robotics in architecture has received, veering from techno-utopianism to techno-pessimism. Is the greatest value in robotics for architecture in fact contained within the discipline, residing in the way that it forces architects to think differently, shifting their mental landscape and making them design truly three-dimensional space? Are we though, in danger of neglecting to explore and re-imagine the fundamental relationship between man, designers, workers, machines, computers and robots?


INTRODUCTION // There are two very different but prevailing attitudes to technology, in regards to its capacity to innovate and make a difference. The first of these two attitudes is techno-utopianism. Innovators tend to share with entrepreneurs an unbridled optimism. You cannot be an entrepreneur if you are not confident that things will turn out for the better. Beyond a merely positive outlook, techno-utopianism is generally typified by a belief that innovation has an immediate and ben-

eficial impact on society. There are numerous examples of this kind of approach. Richard Buckminster Fuller comes immediately to mind, with his unflagging endorsement of technological progress that he saw as a means to redesign society. According to Fuller, the Dymaxion house, car, and prefabricated bathroom that he developed in the 1920s and 1930s were not only ingenious devices meant to revolutionize the building industry transportation and everyday life; they were also intended to pave the way for a radically different future in which men would roam free on the surface of the globe, live everywhere and fully take advantage of their intellectual capacities. ln contrast to techno-utopianism, the opposing view, techno pessimism, is not that prevalent. It is largely limited to domains like literature and philosophy. You have only to think of the condemnation of railways by so many 19th-century writers, or the distrust expressed by 20th century phenomenologists, such as Martin Heidegger, in the power of technology to improve the human condition. A more nuanced assessment of the real impact of technological innovation on the structures of production and society is far more common. Wary of discourses, which they often perceive as crude simplifications, historians have specialized in a more subtle and perceptive take on innovation. In his book The Shock Of The Old: Technology And Global Hystory Since 1900, British historian David Edgerton has tried, for instance, to reassess the re-

01 — ROBOTS AND FICTION // In the last few decades, the digital and its impact on architecture and construction have given birth to a wide array of discourses and statements that can generally be categorized as either techno-utopianism or techno-pessimism. While vibrant praise for digital technologies with their potential for positive change has become ubiquitous, more nuanced assessments of any ensuing transformations have also been produced. Recently, digital fabrication in architecture has become the focus of many of these unabashed eulogies and critical evaluations. Are we facing a revolution comparable in scope with the invention of printing in the Renaissance? Or does it represent a more finite evolution? A more tentative hypothesis is after all tenable, given the gap between the still highly experimental character of digital technologies and the predominance of traditional mass-production techniques, if not their further diffusion, in countries like China. This dual regime is even more pronounced when automating construction through the use of robots. On the one hand, following the pioneering experiments of Fabio Gramazio and Matthias Kohler at ETH

Zurich, where they share the Chair of Architecture and Digital Fabrication, robots appear as a key element of future architectural development. Many schools of architecture are now equipped with robotic arms that are used for structural investigations as well as for research on surfacing and patternig. On the other hand, it is easy to measure the limitations of the use of robots on ordinary construction sites, beginning with the innumerable problems linked to security and maintenance that they raise. In addition, they still possess an unassailable association with science fiction; a connotation that is especially marked in the experiments involving flying robots led by Gramazio and Kohler. The aerial ballet of their insect-like machines assembling blocks with superhuman precision is evocative of aerial traffic patterns in cities of the future where flying vehicles have become common. Already present in Albert Robida’s late 19th - and early 2Oth – century engravings and novels, this vision of the urban future would later become a standard of science-fiction movies, from Fritz Lang’s Metropolis (1927) to Ridley Scott’s Blade Runner (1982). The science-fiction overtones of robotics may well serve as a convenient transition towards one of the key roles of robots these days, namely their supporting part in a narrative regarding the future of the architectural discipline and the rising importance of automated fabrication. This type of narrative dimension is by no means a recent phenomenon. The various attempts to industrialize building activ-

The Nature of Robots

spective roles of traditional techniques and highly visible breakthroughs like nuclear power in our contemporary world. His line of argument is that everyday techniques have been more instrumental in shaping what lies under our eyes than more widely and highly acclaimed innovations.


The Nature of Robots

ity throughout the 20th century were intimately related to a grand narrative regarding the necessity to adapt architecture to the age of the machine. One of its interests was to enable its proponents to occupy a middle ground between overt and thus disputable techno-optimism, the fictional mode suspending questions of immediate feasibility, and the opposite skeptical attitude. Experiments extended into fiction as a way to suggest, without unnecessary heaviness, that they could lead to widespread change.


02 — UTOPIAN PERSPECTIVES // Despite its attempt to distance itself from technoutopianism, the narrative of the industrialization of construction remained permeated by utopian concerns such as the desire to reconcile nature and technology, the project to free man of unnecessarily harsh work, or the ambition to enable man to live everywhere on the planet, the latter being especially present in Buckminster Fuller’s approach to industrialization. In many respects, the advent of robotics in architecture is marked by the emergence of a similar type of narrative. On a certain number of points, this new narrative appears as the direct inheritor of the industrial one. Like its forerunner, it is permeated by utopian themes, some relatively traditional, others without equivalent in the history of industrial ideals. Whereas the project to relieve man of painful tasks is by no means original, the quest for a new immediacy based on computation between the designer’s mind and the built reality is with-

out precedent. Throughout the 20th century, machines used to prefabricate or customize, and to assemble parts, had been interpreted as tools radically distinct from the mind that put them in motion. Equations and flows of data seem to constitute, in contrast, a fluid milieu that tends to unite the human brain and its mechanical extensions. This new intimacy could be described as the advent of a cyborg designer whose intentions are materialized through the action of powerful artificial arms. But this perspective may be misleading insofar that the best way to envisage what robots do is not necessarily to consider them as extensions of the human mind and body. For they do not exactly replace human arms and hands; they follow principles of their own, often different from the rules that govern human productive gestures. Coupled with the readiness with which they obey the designer’s instructions, such difference increases their epistemic potential, as we shall see towards the end of this article. Besides the new immediacy between the mind and the built reality, robotics announces the possibility to radically overcome the constraint of large series that had hampered so many former attempts to industrialize construction. The utopian perspective of a world of ‘makers’, to use Chris Anderson’s term, becomes almost inescapable. In this world, prototyping and small-scale production of sophisticated components would advantageously compete with repetition and mass production. The enthralling spectacle of a harmo-

left by robots’ shortcomings and their potential otherness that forbids considering them as mere extensions of the designer’s hand. 03 – FROM FICTION TO CONVERSATION // Will the fiction one day become realiry? Although the grand industrial narrative of the 20th century never came fully to fruition, its legacy was considerable, from new materials like plastics to key techniques of dry assemblage. The robotics narrative will probably have equally enduring effects on the built environment, and this is all the more feasible given that the digital age is marked by the multiplication of autorealising fictions. Before the multiple connected devices that surround us were realized, ubiquitous computing was one of these fictions. But the impact of current robotic fabrication experiments on architecture may extend beyond improved materials and innovative techniques. Returning again to modernist industrialization, it is striking to observe how it influenced the fundamentals of the architectural discipline by redefining both what design was about and the effects and affects it was supposed to produce. To use Walter Gropius’s characterization, industrialization was not only meant to produce buildings differently; it was to represent a ‘purifying agency’ liberating architecture from outmoded technological as well as aesthetic values. In other words, it played an epistemological role and forced designers to think differently. Such might be the most important

The Nature of Robots

nious ballet of productive forces and machines, of a streamlined universe of efficiency and elegance, appears also inescapable. On this ground again, the robotic fabrication narrative is not entirely original. Modernist industrialization had already dreamt of streamlining projection. The possible autonomy of robots could constitute nevertheless a striking point of departure from the vision developed throughout the 20th century by the advocates of rationalization. In science-fiction novels and films, robots appear sometimes as obedient slaves, sometimes as dangerously independent life forms. Recent and spectacular advances in intelligent automation ate, day after day, making the latter alternative more plausible. Robotic fabrication may confront us for the first time directly with the need to cooperate with our technological auxiliaries rather than simply use them. The human workforce seems to be so far missing from this narrative, as if an exclusive dialogue between designers and robots were the only development worth exploring. More generally, discourses on digital fabrication in architecture often indulge in a strange kind of Ruskinianism. In their neo-Ruskinian perspective, designers tend to occupy the place formerly devoted to craftsmen, that of inspired artisans shaping the world with their hands - digitally augmented hands that is. Understanding the new immediacy between mind and matter in such away is doubly misleading. It forgets both the persistent need for a human workforce to step into the gap


The Nature of Robots 10

consequence of the introduction of robots in architecture, at least for now. The large success met by the experiments conducted by Gramazio and Kohler at ETH Zurich has to do with the clarity with which they have outlined this epistemological dimension. This led them to propose in their first book the notion of a new digital materiality redefining the way designers address the fundamental problems of their discipline. In addition to the characterizations of this new materiality they proposed – like the simultaneous rise of sensuality and of programming, the growing importance of variation and multiplicity and the key role of processes, or rather in a slightly different perspective — the following observations can be made. First, robots do force designers to think in a truly three-dimensional space in which there are no longer any privileged directions. Moreover, they remind us of the foundational character of rotation in the analysis of motion. The movements of our body are themselves based on the various rotations of our members. But we have for a very long time forgotten this simple fact and used rectilinear motion as the standard spatial operation, from mechanics to design. Modernist industrialization itself relied heavily on repetition by rectilinear translational motion. In short, robots introduce us to a profoundly different geometric world. Even if gravity will continue to make us distinguish between the vertical and the horizontal. While translational motion will remain an important feature of mechanics, the mental

landscape of design is about to shift. Second, we have known for awhile that the emergent digital materiality was based on the association of notions that used to be antagonistic: the physical and the electronic, the sensory and the computational, the concrete and the abstract. Robots teach us that there is also a thinner and thinner line between objects and processes, as well as between stability and instability. As Gramazio and Kohler have once again demonstrated with experiments like their 2008 Structural oscillations installation, robots enable designers to play at the very frontier that used to separate stable assemblages from unstable combinations. Third, robotic fabrication in architecture induces a change in efficiency as well as in the beauty that is linked to it. There used to be a superiority attached to the simple and the scarce over the complex and the multiple, a superiority rooted in technological reminiscences. According to scholastic philosophy, simplicity and unity represented fundamental attributes of the divinity. Digital technology and robots are radically challenging such a perception. The complex and the multiple appear more and more as the natural condition from which designers should start. What remains to be explored is the potential of the machine to emancipate itself, at least partially, from the instructions of the designer in order to appear as a significant other in the conception of the project. Whereas architects like Cedric Price had tried to follow this path in the early

chitecture. Why not imagine a unified design and fabrication process based on a series of conversations between men, designers and workers, and machines, computers and robots? A truly different architecture could rise from such extended conversations.

Gramazio & Koeller and Raffaello D’Andrea with ETH Zurich. Flyigh Assembled Architecture, FRAC Centre, Orleans, France, 2012; Flying robots assembling the Vertical Village, a megastructural project designed by Gramazio & Koeller. The design of Vertical Village is intimately dependent on the project to assemble it using flyig robots.

The Nature of Robots

1970s - probably too early in terms of technological feasibility - the possibility of such otherness is often lacking today from the speculations and experiments regarding digital design, just like the role devoted to the workforce is generally minimized. Everything happens as if computers and robots were to remain forever obedient slaves, while the role of workers is steadily diminishing. Overcoming this strange shortsightedness could very well represent the next step in the investigations regarding the use of robots in ar-



The Nature of Robots

Credits // Philippe Morel


INTRODUCTION // In Le Corbusier’s key work entitled Vers une Architecture [Towards an Architecture], published in 1923, he maintained that the world’s transformation by machines would lead either to ‘an amelioration, of historical importance’, or to ‘catastrophe’. Today, it is tempting to use these same terms when referring to the impact of the information technology revolution and of linking computation and machines in what we consider here, in general terms and without distinctions, as ‘robotics’ or ‘mechatronics’ – the latter term referring to a fusion of mechanical engineering, electronic engineering and computer engineering. In fact, Le Corbusier’s opinions, and more generally the French architect and theorist’s studies on the connection between technology and society, are still so compelling that not a week goes without a scientist, futurist or economist publishing on the social impact of technological change, as demonstrated by articles and works by Erik Brynjolfsson, Kevin Kelly and Paul Krugman . All three seek, amongst other things, to revive certain Marxist theories on technology including Krugman, whom no one would consider leftist. However, although the debate on the role of technology and, more specifically, robotics in the multiple economic and social transformations taking place today is very topical, due

mainly to exponential growth in the quantity of products available and a drop in production costs, it remains blurred. Firstly, it was recently subject a short-term productivity analysis, which implicitly identified the IT revolution with human computer users rather than with the machine themselves. This obscured the defining feature of a technology, which has since opened up the path to full automation in production and to algorithmic capitalism in finance. To a large extent, this is a recent form of capitalism and its development, conversely, tends to validate the assertion by Paul Krugman that “Productivity isn’t everything, but in the long run it is almost everything” . Following on from this, the debate on the social impact of computation came up against the remarkable lack of knowledge concerning the already distant origins of information technology and consequently the concepts which relate to it. This lack of knowledge can be attributed to disdain for the technology or to the difficulty in accessing an aspect of modern culture based largely on science. Although the conceptual basis was laid prior to the 20th century, the operational aspects of information technology and electronics were developed in an era much closer to the publication of Towards an Architecture than to the modern day. This was an era associated with the Second World War, whose importance can scarcely be overestimated: anni mirabiles 1947 and 1948 alone saw the inven-

01 — The Market of Robotics // The robotics market is currently growing at phenomenal rate, driven by progress in all areas of industry from electronics and precision machining to material science and system engineering. This impacts on the performance and costs of robots, paving the way for a whole range of new applications. For example, the costs of industrial robots, adjusted according to quality decreased from a factor of 100 to 20 between 1990 and 2004, while sales of these same robots increased by a yearly average of 9 per cent between 2008 and 2012, with a record total of 166,000 robots sold in 2011. As regards non-industrial Robots, for instance unmanned aircraft and vehicle system (UASs and UVSs) used in precision agriculture, telecommunications, transport, surveillance and monitoring, a recent study estimates at $27.6 million the loss per day to the United States of failure to integrate these into the National Airspace System (NAS), which gives a sense of the issues involved. Although architecture is one of these issues and is the main focus here, assessing robotics in a wider context provides insights that are not visible in the early stages of development of architecture-related robotics. Firstly, the actual definition of the word “robot” and the different types of robot should be clarified. Until 2011, ISO standard 8373, issued by the International Standard Organization (ISO), that defines terms used in relation to robots and robotics devic-

The Nature of Robots

tion of the transistor and the publication of the founding theories of contemporary science and technology by John von Neumann, Claude Shannon and Norbert Wiener . The technological revolution following the Second World War has resulted from a very deep intricacy between fundamental science (including mathematics and quantum physics) and industry. Our Robotic age is also characterized by such an association. probably at an even deeper level. Lastly, agreement, even among bona fide experts, on the future development of a society, indeed a whole civilization, is not necessarily a given, as shown by the differences of opinion between Arthur W Burks and Douglas Hartree on the future of information technology. The former predicted, after the creation of Electronic Numerical Integrator and Computer (EINAC) – the first general-purpose computer, developed by the United States Army’s Ballistic Research Laboratory to calculate artillery firing tables in 1946 - that “a new era, an era of electronic calculation” was starting ; while the latter, who took a more reserved approach, foresaw the commercial failure of computers, saying that no country could eve train the number of programmers needed to sell computers . These two attitudes still featured in the debates that shook the 1960s ant 1970s and which, added to those addressed further on in this piece that examine the impact of technology from the work point of view, mirror the current discussion on robotics.


The Nature of Robots 14

es, defined an industrial robot as ‘an automatically controlled, reprogrammable, multipurpose manipulator programmable in three or more axes, which may be either fixed in place or mobile for use in industrial automation applications’. Service robots, conversely, had no strict or formally accepted definition. This definition of industrial robots – still very similar to that of the Robot Institute of America used until 1993, which held that ‘a robot is a reprogrammable, multifunctional manipulator designed to move materials, parts, tools, or specialized devices through variable programmed motions in the performance of a variety of tasks’ – only partially reflected the development of robotics. Leaving aside service robots and non-standard robots, the definition continued to be influenced by the industrial context of a robotic technology still in its infancy. This era was symbolized by the term ‘Article Transfer’, taken from the title of a patent field on 10 December 1954 by George Devol named Programmed Article Transfer. (Granted by the United States Patent and Trademark office on 13 June 1961, the patent gave rise to the first industrial robot, Unimate, manufactured by Devol’s company Unimation and put into use latter year. The term ‘programmed’ confirm the stronger link with computers. While the key notions of “automatic” and ‘reprogrammable’ were maintained in the standard definition of robots following on form this firs patent, the other terms excluding service robots gradually became increasingly problematic. These robots,

which do not have the kinematic or generic design of industrial robots, are nevertheless more popular today. This was one of the factors that led to the update in 2012 of ISO 8373, whose general definition of a robot is an: ‘ actuated mechanism programmable in two or more axes, with a degree of autonomy, moving within its environment, to perform intended tasks. Note 1 to entry: A robot includes the control system and interface of the control system. Note two to entry: The classification of robot into industrial robot or service robot is done according to its independent application.’ These remarks could be considered linguistic subtleties if they did not show how great the trend now is to class any object, be it just slightly mobile or intelligent, as a robot, leading to Suprematist-type world without objects. If the universal Turing machine was the reference point used to evaluate the computational capacity of different computers and therefore their classes, the industrial robot, whose 6 degrees of freedom in linear movements and rotations (3+3) allow it to reach any possible point within its work envelope, represents a reference electromechanical machine, a mechatronic benchmark for the robotics era. Although the old distinction between industrial and service robot still exist to certain extend, the boundary is, as mentioned, porous in the sense that it is the application that determines the classification. For example, Section 2.10 of ISO 8373 (2012) states: ‘while articulated robots used in production lines are industrial robots, similar ar-

02 — Specificity of Robots and Specificities of Architecture // Architecture-related robotics could potentially be split into ‘architectural robotics’ and ‘robotic architecture’; but does this first branch of robotics actually exist? If logistics and transport have, for example, led to the development of new kinds of robots or, at very least, to the classification of mobile machines (which until recently could only be called cars, tractors or planes) as robots, the same innovation process seems much slower in architecture. To date, experimets in automating prefabrication carried out in Japan mainly in the 1970s and 1980s, more recently, in Western so called ‘avant – garde’ architecture have mainly relied either on standard industrial robots or non – standard but existing robots, such as unmanned aerial vehicles (UAVs). In the first case the end result is still similar to what it would be without the use of automated process, while in the second case, with more innovatory features, innovation is more tangible in architecture than in robotics itself. Although 159,000 robots were sold throughout the world in 2012 alone and the number of service robots in use has exceeded the 10 mil-

lion mark, a very small proportion of them have found their way into architecture. If the construction materials and parts industry is highly, indeed for certain plants entirely, automated and if service robots are already being used on construction site, an increase in productivity, within a fully liberal economic system, is surely the likely result. So which criteria distinguish between the advanced architectural researches mentioned here and the simple opportunistic application used by the industry? First of all, steps should be taken to combine a revolution of architectural and constructive language with a revolution of social organization, which is now out of step with the general transformations taking place in science and technology. If possible, this should be done, as stated by Paul Krugman, ‘before the robots and the robber barons turn our society into something unrecognizable’. Secondly there should be a clear awareness that robotics in architecture cannot be isolated from this same society - that, despite appearances and obviously beyond what it represents as a specific area of science, robotics constitutes neither a procedural problem, nor a discipline problem, nor even a work problem, but rather an aspect of generic problem of our era: computation. Although, when returning to the source of cybernetics, robotics appears as a control problem and, when adding an additional level of abstraction as a logical mathematical problem, in both cases they require their own epistemology and historical understanding. Yet, more often

The Nature of Robots

ticulated robots used for serving food are service robots.’ It is not difficult to transpose this kind of distinction to architecture, but in doing so, answering this simple question becomes a challenge: what do we mean when we talk about robotics in architecture and how will it or should it be distinctive?


The Nature of Robots 16

than not, this is quite simply missing in architecture. While Charles Babbage’s theorizing on the unavoidable replacement of work by machines in his 1832 On the Economy of Machinery and Manufactures shows that had learnt from the practices of division of high-skill labor advocated by French mathematician and engineer Gaspard de Prony, 36 years his senior, it seems little has been learnt since his time. This in spite of the repeated warnings given by those closest to the scientific foundation of technical development. As Kevin Kelly cautions: ‘The rote tasks for any information-intensive job can be automated. It doesn’t matter if you are a doctor, lawyer, architect, reporter or even programmer: The robot takeover will be epic’. Does this mean that there is no specifically architectural problem linked to robotics? Of course it does not. All those who work with robots know, for example, that programming numerous non-repetitive trajectories in reasonable time frame, however, is a problem. Developing software to reduce this work and use trajectories from standard CAD software, with automatic code generation if possible for several brands and types of robot is another. Similarly, the incompatibility of construction materials and techniques from a pre-robotics era also presents its own difficulties, as does large-scale construction using robots most of which were not developed for this purpose. Nevertheless, without understanding the work and intelligence needed to overcome these difficulties, they still appear to provide

little indication about the development of robotic architecture, since they can be resolved in the short term. An example is the work towards simplifying communication between industrial robots and the types of CAD software most widely used in architecture that has been carried at EZCT Architecture &Design Research since July 2010, when ABB IRB 120 robot was acquired for the purpose. The consequent development of HAL, a Grasshopper® plug-in for industrial robot programming and control, by EZCT intern Thibault Schwartz goes long way to addressing the automatic code generation issue. The research conduct in the DFAB facility at ETH Zurich on adaptability to existing and noncalibrated materials has also been conclusive, as has the use of artificial intelligence and vision systems – now found in most service robots from the vacuum cleaner to the driverless car and including the drones used in precision agriculture – to manage dynamic tolerance problems. Scientific and technological discoveries have played a major role in the development of robotics, over and above the applications intended for them today. Artificial intelligence and computation as ‘Communism of Genius’ // The romantic associations between architecture and technology deflect industrial economy problems onto an aesthetic theory of widespread creativity. Th. theory holds that robots and machines are the Neo-Ruskinian gadgets of those types of architect who are incapable of accepting their

from Reuters and Associated Press wires, articles which could not be distinguished from pieces written by humans and which recall Turing’s work on the Imitation Game, or the recent developments in automated verification of mathematical theorems which even a ‘traditional’ mathematician like Vladimir Voevodsky considers fundamental. This Progress, which led Voevodsky to claim that ‘soon mathematicians won’t consider a theorem proven until a computer has verified it’, validates the long-standing opinion, which has been taken up by epistemologist Franck Varenne in an engineering context, that simulation models are better than real models. To sum up, in architecture, most of us still view the relationship between architecture and machines, more specifically robots, in a traditional framework at the very moment when it is disappearing. Rather than wondering how robotics can play a part in an architectural discipline that is still romantically defined by a focus on ornament or architectural composition - which any well -programmed smartphone can determine in a fraction of a second without distinction as to the authorship - it has become clear that we must address artificial intelligence and accept that robots are first and foremost computers. Therefore, due to the capacity robots have to perform calculations, we can and must consider their potential in much more detail than we are currently. Identifying the true nature of technologies has been a trademark of the

The Nature of Robots

own obsolescence, as convinced of their own infallibility as booksellers as they are of the need for a greater human touch than the algorithms of Google and Amazon. Of course, reality proves every minute that the opposite is true, and a mere glancing at the situation architecture finds itself in shows the extent of this misjudgment. Indeed, virtually all architecture can be described as a pastiche business, in which a few often basic outlines or elementary lines of scripts result in tens of thousands of buildings, just as the sperm of a few prime bulls results in millions of other bulls intended for the food chain. Saying that most political lessons of the emergence of computation have not been learnt is a euphemism in today’s architectural practice and theory. Whether it is in terms of pure ability to perform calculations, as reflected in the word ‘computer’, or by means of artificial intelligence, which has a close but complex connection with calculation, computation - as an authentic product of the knowledge democracy of, as the Surrealists might have called it, the ‘communism of genius’ - ‘overturns all our concepts of (architectural) culture’. Most of what we considered or nostalgically still consider an architectural issue no longer appears as such when examining the link between artificial intelligence and automated production. For this, it is simply a question of changing perspective and observing a few recent phenomena within clear epistemological reach. Examples include the automated drafting of press articles


The Nature of Robots 18

best architects/theorists of the 20th century, for example Le Corbusier, Buckminster Fuller, and indeed Marshall McLuhan, who, although not directly concerned with the production of objects, wrote in 1969 - well before architects began to express a similar opinion – that ‘once electronically controlled production has been perfected, it will be practically just as easy and affordable to produce a million different objects as to create a million copies of the same object’. So, due to an exponential drop in the cost of electronics and thus technologies, a whole range of new approaches to robotics in architecture will start to appear. The opportunities presented by UAVs for transporting heavy loads are already among them, even though, as we have already said, most applications remain similar to what robotics offered six decades ago, namely Programmed Article Transfer. In fact, real developments require a conceptual leap, for instance when the modern car was assembled from separate elements - horses and carriage - into a single object or when the flight of birds stopped being used as inspiration for airplanes and was replaced by a unique application of physical principles, which Le Corbusier delighted in reiterating. Automation and the constantly developing trend towards autonomous objects now give full meaning to his term ‘machine for living’, since the home is no longer an object built by robots but is itself potentially part of the ‘robots for living’ class, a class which was partly anticipated during the 1960s. The process

started by LeCorbusier and the Russian productionists would therefore be completed, whereas at the same time the process started by the Futurists is constantly re-starting, each minute confirming the extent to which futurism is the unspoken and unwritten doctrine of our civilization. A doctrine which the ‘science of calculation’ or to use the current term ‘computer simulation’ has implemented through ‘its own condensed language, expressions, which are to the past as history, to the future as prophecy’. A Temporary Conclusion on Computational Literacy and Politics // The time will come when architectural design and construction will only be taken seriously if entirely automated and checked by a computer. The prospect of this moment will be daunting to certain people, even though only a tiny percentage of buildings can actually be called architecture and global developments are governed not by this percentage but by the general majority. So architectural robotics should mainly target this group, not only because it can, but because it has no choice: as mentioned above, since ‘the rote tasks of any information-intensive job can be automated’, they will be. In order to properly meet needs, the whole discipline should already be reviewing its practices and its bases, including the method of teaching, where the deficit in scientific knowledge is difficult to understand. If it takes a real author to write THE BROTHERS KARAMAZOV or a real architect to plan the Villa Savoye and

facing automation in architecture and indeed automation in general. The first challenge is to ensure that digital and computational literacy is properly integrated into teaching at architecture schools, as of first year and then at various subsequent stages of learning. The need to acquire and transfer this new type of knowledge in the education system led me, among other reasons, over 10 years ago to stop using architecture software in favor of scientific computing software, i.e. Mathematica. Following on from the key work carried out by DEAB at ETH Zürich, the second challenge is to test and develop new types of robot which would not only allow such a work to be replicated but which would move architecture forward; for although it is an age-old discipline, architecture is needed now more than ever. As a result, together with the COPRIN team at INRIA Sophia Antipolis and ENSAM Paris, we are currently testing a Stewart platformtype parallel robot. The robot, which is portable and cost-effective, can in theory cover an area of several thousands of square meters while bearing a load of more than 500 kilograms (1,100 pounds). Its design involves closed-loop kinematic chains that ensure good rigidity and high repeatability; however, it is difficult to model the behavior of this kind of parallel robot since the computation of a configuration requires the solving of a set of non-linear equations which theoretically may have up to 40 solutions, this computational complexity involving a couple of non-trivial problems for

The Nature of Robots

to invent the concepts of which it is an ultimate proof what is there to say about mere ‘newspaper article architectures’ where each one is desperately seeking an inaccessible originality of style? In an era of complexity, should we not be encouraging the only intelligence capable of tackling this complexity, namely artificial intelligence? While no one nowadays could imagine looking for a document without the help of a search engine, many people still think that problems which are vastly more complex, for example in politics, architecture and urban planning, can be resolved ‘traditionally’, in other words by humans. It is a disastrous belief, which the Russian theorists denounced almost a century ago when they asserted that ‘talent is no longer random but artificially cultivated’; the ‘reform of science and education’ would therefore need to be’ viewed and implemented from a Constructivist angle’ , which, today, translates as foregoing the simple algorithmic aesthetics approach for the assimilation of these principles. If architects had the opportunity to use both automated construction and design, they would place themselves much higher up the scale of production.” They would also adapt their current superficial use of technology, for example that of electronics which is becoming more financially accessible just as architecture is moving at a forced march in the opposite direction (which incidentally invalidates cynical low-tech approaches) . . . Based on these various points, it would appear that there are three challenges


The Nature of Robots 20

low-cost implementations. The third, much less specific challenge implies a general understanding of the nature of robotics as computation applied to objects. If the refusal to adopt this wider application of computation in architecture sometimes appears to be a stumbling block, more often than not this is the result of a pragmatic or more precisely opportunistic use of the latest technological advances in an unsuited social context. This is a ‘prehistoric era with oversupply’, where false political and social foundations are upheld by capitalism itself or by subjects whose acceptance of traditional politics, despite the accumulated evidence of failure, is a mystery. In most areas, artificial intelligence and automation are still only tentatively promoted and indeed kept separate from each other. The illusory need for certain intellectual skills is artificially maintained by entrusting design to ‘human robots’, that are underpaid but still profitable. The already effective automation of physical tasks combined with the intrinsic capacity of computers shows that architecture is lagging behind. In the same way that traditional trading floors represent the folklore of finance rather than finance itself when compared with robot-traders, architecture as we regrettably continue to conceive of it will unfortunately also become folklore. That being the case, we would have everything to gain - or at least not much to lose – in architecture as well as in all economy-related questions, by heeding Charles Babbage, and others

more recently, and entrusting operations to calculation (Babbage’s use of which term corresponds to today’s ‘computation’). Indeed, by solving the arithmetic side of the economic calculation problem famously considered by Ludwig von Mises as the cause of the failure of socialism, the computational power of Google servers and personal supercomputers proves to

Industrialization and Automation of the Building Process Credits // Thomas Block // Silke Langerberg // ABSTRACT // The introduction of robotics in construction is part of a much longer his tory of industrialization and automation on the building site. Thomas Bock of the Technical University of Munich and Silke Langenberg of the University of Applied Sciences, Munich, highlight how the Industrial Revolution and the development of a transport infrastructure in the 18th and 19th centuries in Europe first triggered the building trade from a largely localized industry into a national and mechanized one, leading to the highly advanced automated construction techniques that continue to be developed in Japan and other Asian countries to this day. INTRODUCTION // The Industrial Revolution changed the building process, by then largely dependent on a local base of materials, skills, building knowledge and tradition, irrevocably. In the late 18th century and throughout the 19th century, new machinery serial-produced elements and industrially fabricated materials started to appear on building sites, complementing long term approved construction techniques. It was not, however, until the 20th century that there was a real attempt to adopt industrial manufacturing processes. By the 1920s and

1930s a few prototypical buildings had been realized, anticipating the so-called ‘industrialized construction’ processes that were rolled out at a larger scale during the second half of the century: from the serial prefabrication of building elements to the mass production of standardized housing estates and system buildings. The specialized robotic machinery and automated high-rise construction sites that were developed in Asia during the 1990s can be viewed in the context of a greater trajectory of mechanization and industrialization. The beginning of the 21st century could again prove pivotal for the building process, with robotic fabrication having the potential to change the building site once more. On that account it seems crucial to take a look at the development, influence and results of some historical precursors in order to understand that the implementation of robotics in architecture at a larger scale may not just require a first phase of experimental research and prototyping, but also a fundamental change in the early design stages as well as in the construction process that goes far beyond imitating existing building technologies. 01 — The Establishment of Industrially Produced Elements and New Materials // The mechanization that was a consequence of the Industrial Revolution has, by the expansion of the railway networks, thus increasing mobility and transportation of goods and materials, directly resulted in the growth and congestion of industrial

The Nature of Robots



The Nature of Robots 22

towns, and the fast development of their building stock and infrastructure. It has also directly affected the building industry and construction process: industrially produced materials and prefabricated building elements such as cast-iron beams and columns, glass, factory-made bricks or artificial stone decorative elements were increasingly used during the 19th century and their potential in the erection of large industrial, infrastructural end representational buildings explored. Different construction techniques were thus required, and specially developed building machinery and cranes began to change the organization of the building site; the construction of the Crystal Palace for the Great Exhibition in 1851, for example, required a whole series of different machines, powered by a steam engine. As a large number of identical or similar parts were essential for making industrial fabrication and serial production economically viable, the number of different building elements was reduced. At the same time their lot size was further increased by using them en masse for the erection of large building volumes. Thus the application of industrially mass-produced building elements first manifested itself in the construction of large-scale steel structures - bridges, train stations, towers, exhibition halls, the glass roofs of galleries and department stores, which were assembled from a, manageable amount of columns and beams in available standardized cross sections. Around the middle of the 19th century

concrete also gained in importance as a result of the industrial production of Portland cement. By the beginning of the 20th century, the general availability and usability of serial mass-produced building elements in steel and concrete, combined with opportunities to transport them over longer distances, made the use of prefabricated elements much more common even in smaller individual buildings. However, the architectural design of the numerous town houses and office buildings constructed during that time rarely represented the use of industrially produced materials or components, which had somehow simply become state of the art. 02 — RATIONALIZATION and INDUSTRIALIZATION of the BUILDING PROCESS // The building industry started to adopt industrial production methods during the 1920s and 1930s in a push to solve the housing shortage in the growing towns after the First World War: The design of a limited number of identical building elements to construct slightly different housing types aimed to enable serial mass production and time and cost savings akin to those realized in other sectors of industry. At the same time there was an attempt to reduce and simplify the number of stages involved in building on the construction site, to increase the employment of unskilled labour and to shorten the completion time. Walter Gropius’s TÜrten housing estate in Dessau, Germany (1928) is maybe one of the best-known examples, along with the Hausbau-

and industrial sectors. The design of these structures was often subordinate to their production and construction principles, indicating a paradigm shift. At the beginning of the 1960s, the different elements of large-scale projects often had to be prefabricated in field factories on or near the building site. In the ensuing years, however, an increasing number of independent prefabrication factories began to be built in all industrialized countries. The resulting reduction in the distances travelled to transport materials led to industrially mass-produced building elements being employed at a previously unprecedented scale. The subsequent development and application of different casting techniques, such as slip casting or lift slab constructions, were also significant, as they not only aided the advancement of the industrial prefabrication of elements for later assembly, but also the automation of the building process and the construction site itself. The 1973 oil crisis, however, proved an impasse for these developments; the Organization of Arab Petroleum Exporting Countries (OAPEC) proclaimed an oil embargo on the US, disrupting the energy supply, triggering recession and unleashing the very real long-term possibility of high oil prices. This created a new awareness of the limits on economic growth, while population rates concurrently stagnated. By the mid-1970s, attempts to fully industrialize the building process declined or were abandoned in both Europe and the US. At the same time,

The Nature of Robots

maschine (‘House Building Machine’) developed and published during the Second World War by Ernst Neufert. This process-oriented initiative differed completely from the approaches of the 19th century in its ambition to change the organization of the building site instead of just responding to, and borrowing from, the innovations and products developed by other sectors of industry. The ideas behind the rationalization of the production of building elements and industrialization of the building process were first propagated in Europe after the Second Word War, when there was a concerted effort to realize them at a larger scale. For the first time the serial mass-production of elements and use of industrial fabrication methods in the building industry seemed to make sense, because of the tremendous amount of buildings required to meet the urgent task of reconstruction and demand for housing in the postwar period, as well as during the following boom years between 1950 and 1970. While in the 1950s building construction remained quite conventional, with advances largely limited to the use of new materials - plastics, aluminium and composites - as well as larger and stronger machinery, by the 1960s and 1970s a distinct change had come about in the design process. Standardized buildings and building systems were increasingly developed, enabling time and cost savings, as elements were commonly mass produced for the construction of buildings in the housing, education, commercial


The Nature of Robots

the social problems of towns, which had been planned to be mono-functional, and their mass-housing estates became apparent, and the planning principles of the boom years, based on growth, progress, technology and prosperity were mostly replaced by ecological strategies and economic considerations. In contrast, Asia did not experience the same dramatic turnaround. Its increasing population and growing cities, resulting from rural depopulation, continued to create demand for the construction of large buildings as well as large building masses. Simultaneously, the lack of skilled labour, especially in Japan, led to the promotion of automation in prefabrication and construction as an alternative to common construction practices.


03 — TOWARDS AUTOMATED HOUSING PREFABRICATION // Automation in housing construction started in Japan in the 1960s, with large prefabrication companies such as Sekisui House, Toyota Home and Pana(sonic)Home, which were all descendants of firms that had already successfully employed automation in other sectors. Their manufacturing processes were characterized by a shift away from the construction site to a structured and automated factory-based work environment. In the case of Toyota, for example, 85 per cent of work was pre-executed off site. Nevertheless, the production process in these factories was, for the most part, still conducted by human labour, so it

owed more to the organization of the assembly line than to real automation. In contrast to European approaches, where prefabrication was primarily optimized to achieve fast and cheap production of large numbers of identical elements, the major achievement of the Japanese prefabrication industry was their quite early success with customization and personalization, as well as their ensuing knowledge of users’ demands. The structured assembly-line work, combined with the advantages of human labour in a factory environment, allowed for the individual adaptation of single parts meeting customer demand without disturbing the production chain. They could simply be taken out of the assembly line and replaced manually, to be reworked or finished, before being introduced back into the next stage of the production process, causing minimal disruption to the overall productivity. This approach can be understood as a direct ideational precursor of today’s promotion of robotics in architecture even if it was, of course, far from real automation and levels of productivity a modern industrial robot can achieve. A common characteristic of the early manufacturing systems of the Asian housing prefabrication industry, which is quite distinct from conventional or traditional product production, was the focus on ongoing development. It was this that optimized them for automated manufacturing. Building systems and manufacturing technologies were mutually adapting to each other.

construction robots, and safety measures were required because of the inferior parallel execution of human work tasks in their operation area, productivity gains were often counterbalanced. The evaluation of the first such robots therefore resulted in the conclusion that an off-site approach would be most suited to the organization of on-site environments. Sites would be better structured and designed like factories, and the final goal was the implementation of automated manufacturing and construction technologies. Hence research in automated construction was intensified in Japan, leading to the development of socalled integrated automated construction sites. 05 — INTEGRATED AUTOMATED CONSTRUCTION SITES // The first concepts for such structured environments for larger automated construction emerged from 1985 onwards, integrating the earlier single-task construction robots as well as other elementary control and steering technologies as subsystems. These integrated automated construction sites were organized as partly automated, vertically moving on-site factories providing a shelter for on-site assembly, which was controlled, structured and systemized, and unaffected by the weather, as well as for a disassembly process of prefabricated, modular low-,medium-, and high-level detailed building components. Robot technology was thus facilitated by the creation of the right conditions to install

The Nature of Robots

04 — SINGLE-TASK CONSTRUCTION ROBOTS // ln 1975, after the first experiments in the industrialized prefabrication of ‘system houses’(consisting of various structural, exterior or interior wall subsystems etc.) were conducted in larger series in Japan, and the first range of products, such as Sekisui M1, the first industrially mass produced house type, achieved market success, the main building contractor, Shimizu Corporation, set up a research Soup for construction robots in Tokyo. The intensification of research in this field during the following decade was based on the 1970s ‘robot boom’ in the general manufacturing industry. The adoption of robots was thus a logical approach for Japanese construction firms. The single-task construction robots that were subsequently developed were a distinct departure. Rather than merely shifting complexity from the construction site into a structured prefabrication environment, they deployed robotic systems locally on site for demolition, surveying excavation, paving, tunneling, concrete transportation and distribution, concrete slab, seeding and finishing, welding and positioning of structural steel members, fire-resistance and paint spraying, inspection and maintenance. The initial focus was on simple systems that could execute a single, specific construction task in a repetitive manner. Their steering was, in most cases, conducted manually, and was only rarely automated. Since upstream and downstream processes were also not usually integrated in these single-task


The Nature of Robots 26

automated overhead cranes, vision systems and other real-time control equipment. The conceptual and technological reorientation from single-task construction robots towards integrated automated construction sites was instigated in 1982 by the Waseda Construction Robot Group (WASCOR), which brought together researchers from major Japanese construction and equipment firms in a single initiative. In total, about 30 sites were developed, some as prototypes and others as commercially applied systems. However, their market share and application was limited due to relatively high initial installation costs. These integrated automated sites and their subsystems, such as automated logistics, alignment, welding, etc, were thus used mainly when the special conditions (land prices, high labour costs, traffic, noise and waste restrictions) of a project required them. Since 2008,Japans major contractors have also developed mechanized and partly automated deconstruction systems, which generally follow the same approach as the automated construction sites, in reverse. The advantage of being able to reduce noise, dust and disturbance of the surrounding environment that these deconstruction systems afford strongly supports contractors’ new project acquisition strategies. 06 — IMPLEMENTATION AT A LARGER SCALE // The historical development of the building industry shows that every innovation in con-

struction technology needs at least one generation to establish itself, no matter how groundbreaking the first experiments or prototypes may have been. While early attempts by the building industry to use industrial materials and production methods were accepted bit by bit (with all their pros and cons) and subsequently changed the organization of the construction site, the time has perhaps come for automation and robotics to establish themselves in architecture at a larger scaleAdvances in automated construction continue to be developed, especially in Japan and other Asian countries, and are slowly starting to consider the need for customization of an increasingly individualizing society as well as of the intrinsic conditions of architecture. At the same time, the use of flexible industrial robots in the prefabrication of building elements, as well as in architectural research institutions, is becoming more widespread. However, instead of merely trying to copy and perform long-established construction technologies or prevailing factory automation methods, in order to achieve their inherent performance potential new robotic tools require appropriate conditions, design strategies, kinematics, programming and control. Strong complementarities exist between the actual building - its design, manufacturing and information technology - and its construction and organization strategy. The next real change will only occur on the construction site once design, management and engineering comply with the robot as a new tool.


The Nature of Robots

ENTERPRENEURSHIP in ARCHITECTURE ROBOTICS The Simultaneity of Craft, Economics and Design Credits // Jelle feringa //

The Nature of Robots

ABSTRACT // What do Odico Formwork Robotics, RoboFold, Machineous, ROB Technologies and GREYSHED share in common? They are all arch itectural robotics startups. Jelle Feringa, Chief Technology Officer at Odico, places the phenomenon of the architectural robotics entrepreneur in a historical and cultural context while highlighting the very practical role startups are poised to play in bridging the gap between academic research and industry, by providing the building industry with much needed new software tools.


INTRODUCTION // What are the prospects of entrepreneurship in architectural robotics? Do forward-looking, entrepreneurial architects share kinship with the Renaissance architect Filippo Brunelleschi, who in the 1420s engineered a giant, three-speed, reversing ox-driven hoist, enabling him to span the massive crossing of Florence’s cathedral with a dome? Is Jean Prouvé, an architect who ran a factory and who played a pivotal role in developing cutting edge production technology and modular systems, a 20th-century analogy of how architects will operate in the 21st? Or will education tragically monopolize patronage of architectural robotics?

Sharing the technological platform of robotics with industry encourages the transfer of knowledge from academia to industry. Researchers share the parlance of industry - code interpreted by the robot controller - with dialects differing from ABB’s RAPID, Kuka’s KRL or Straütibli’s VAL3 robot language. Operating on a similar platform challenges where technological innovation stops and industrial adoption starts; industrial robotics allows new manufacturing technologies to gradually evolve from initial experiments to industrial processes. The modest investment required to bootstrap a startup company, given the cost of a new or refurbished robot seen in relation to its potential returns and production capacity coincides both with a renaissance in manufacturing and a need to renew the architectural profession. STARTUPS // A number of promising startup companies have been surfacing over recent years, leveraging architectural robotics beyond mere conceptual merit and stepping into the industrial arena - where startups play a pivotal role. While the robot research work is creating considerable interest, without the filter of practical adoption - technology you can buy expectations go unfulfilled. This is why the startups presented over the following pages occupy an essential position in bridging the domains of academia and industry. Progress is affected by the cyni-

to architecture, where the umbilical cord to CAD geometry cannot be cut. RECOVERING LOST GROUND // This persisting CAD connection is why all of the startups presented here are so focused on software development where the limits of where design stops and production starts are becoming increasingly intertwined. ROB Technologies and RoboFold develop both design and production tools. Are these companies forming a second wave after seeing the rise of the likes of Design-to-production, 1 :One and CASE?Rather than making the most of the array of conventional CNC fabrication methods, these companies take up the role of introducing novel fabrication processes, ranging from stacking bricks to folding aluminium panels and cutting massive blocks of foam. By making the software and design tools available to architects and industry, the threshold for market adoption is lowered considerably. Is it possible that, when an earlier generation of architects ceded greater responsibility for the realization of buildings to engineering offices, architecture lost some of its professional authority? The success of design rationalization offices, file-to-factory processes and architectural robotics suggests so. Are the aforementioned companies and the robotic startups stepping into this void and recovering lost ground? If anything, the discrepancy between what is practically and economically feasible, in comparison to the daily routine of building produc-

The Nature of Robots

cism sometimes observed in these domains, while institutionalism fuels the opportunity. The arguments are that from an academic or a computer science perspective there is little novelty. Inverse kinematics solvers and offline-programming software are sometimes considered to be ‘known’ problems. Or from an industry point of view; the approaches developed in architectural robotics are far-fetched or perceived as too costly to develop, and as yet have an uncertain future in terms of their economic yield. The startups here challenge such long-proliferated misconceptions. The lack of software tools suited to architectural processes might explain the relative and creative dilettantism in developing robotics software for architecture, developed by architects. However, the robotics entrepreneurs featured here focus on pushing novel approaches beyond their conceptual infancy, bridging the gap between academic merit and industrial adoption, providing architecture and the building industry with an important new set of tools. Traditional robot integrators – often coming from a background in the automotive industry - are yet to step into this domain. The fabrication processes developed in architectural robotics differ considerably from the repetitive routines of industrial automation. For example, industrial processes are often programmed in-situ, where a robot programmer is explicitly jogging and feeding the robot with instructions. Once instructed, a robot is able to repeat this procedure. This mechanical approach does not extend


The Nature of Robots

tion, is enormous; a shared thread is found in the ambition of closing this rift. An economical perspective brings about the promise of novel fabrication methods and architectural robotics. However oversimplistic it is to consider architecture as the difference between the raw building materials and the cost of their assembly, it is here where the difference is made. Agriculture and construction are both the largest and least automated industries - where in the latter the cost of assembly essentially equates to the cost of labor. The economics of building have marginalized architectural ambitions, since craft is costly. However, as the cost of mechanized labor continues to drop dramatically, as automation and robotics become so ubiquitous, the moment has come to revitalize the architectural profession. Gramazio &Kohler’s early Gantenbein vineyard façade (Fläsch, Switzerland, 2006) was seminal in this respect, simultaneously exploring the approach of automated bricklaying and its architectural potential. It is precisely this simultaneity of craft, economics and design that is so striking.


DESIGN AND PRODUCTION // The BrickDesign tool developed by ROB Technologies further blurs the false schism between design and production. BrickDesign is a Rhinoceros plug-in encapsulating the expertise required to design and plan large assemblies of discrete elements - bricks. To bring ROB Technologies’ building process of automated brick-

laying in non-standard formations to the market, it was necessary to facilitate the means of design. A lack of proper software tools means it is virtually impossible to design such a number of discrete parts without falling back into traditional bonding rules, which would not exploit the potentialities of the robotic process. As one of the pioneering architectural robotics technologies, developed by architects for architects, BrickDesign is an important precedent of what is required before a fabrication concept can reach the levels necessary for application at the industrial scale. It has both democratized and opened up the approach to towards market adoption. This is a significant achievement, since traditionally industrial robot integrators have not been able to develop the required design tools, and as a result have only provided the required process automation. The design and realization of projects and prototypes built over the years by means of robotic bricklaying is an integral part of the push towards industrial readiness. For example, the client for a recent project – the brick facade of the Keller AG Headquarters (Ofenhalle, Pfungen, Zurich, 2012) - is in fact a central industrial partner of ROB Technologies, not only demonstrating confidence in the method developed, but also challenging the clich6 of industry not being willing to commit to an investment in novel technology. Here is a partner that is pushing architects to reach new heights, both architecturally and in terms of technology, renewing the relationship between

crete. By upscaling wire-cutting technology to a robot with a reach of over 25 metres (82 feet), the company can produce intricate polystyrene moulds for sophisticated concrete structures in a short timeframe and within moderate budgets that challenge existing approaches to formwork. It provides solutions for the scaling of fabrication processes to architectural dimensions, developing technology that is congruent with the dominant building method - casting concrete. By moving beyond the limited production capacity of a CNC router, a cost-effective approach for the realization of large sophisticated concrete structures becomes possible. Alongside its efforts for the building industry, Odico also produces formwork elements for a number of companies in the clean tech industry, such as Siemens Wind Power. Recent explorations in stone-cutting technology by the Hyperbody research group at the Delft University of Technology underscore the interwoven relationship between manufacturing technology, economy and architecture. Being able to process stone with the modest means of an industrial robot rapidly eludes the idea - an archaic notion - of building in massive stone elements. In addition, a salient realization early on in the project is that the cost of marble has considerable scope. Since only up to 75 per cent of the stone quarried yields material that is of pristine quality there is a large volume of second- and third-rate quality that is often ground down to a polishing agent used in toothpaste. The difference in quality is roughly

The Nature of Robots

architect and industry. Regrettably, it is rare to see architects and their industrial partners changing the course of their industry in this way hence the significant role startups such as ROB Technologies and the others featured here might play in progressing the field of architecture. These startups are not just passively facilitating a construction method: an architectural ambition precedes an architectural technology. The recent Arum installation by Zaha Hadid Architects, RoboFold and Philippe Block, presented at the ‘Common Ground’ Venice Architecture Biennale in 2012, is insightful here. While a number of authors are credited, without RoboFold providing its own design software and fabrication process the constructive and aesthetic potential of the approach would not have been realized. The project was influenced by RoboFold’s earlier developments, and as such this robotic startup company can be considered as both design and production partner. Their software tools aid in exploring the possibilities opened up by the robotic folding process - much like ROB technologies’ BrickDesign software allows exploration of the design space the manufacturing technique offers. The question of economy, of developing an affordable approach to the fabrication of formwork, is addressed by Odico Formwork Robotics. The manufacturing of sophisticated formwork accounts for more than three-quarters of the costs associated with the realization of sophisticated geometry in con-


The Nature of Robots 32

based on the whiteness and homogeneity of the color of the material, and translates into a factor of over 40 in the price per tonne. Given this economic bandwidth, an efficient way to shape the raw material into ‘traits’ suggests that building in stone may be a more viable option than is often considered. This understanding has been one of the motivations driving the project, raising the question whether aspects of economy are an inherent element of design. Just as CNC and robotic technologies have made fabrication a more central design aspect, are such manufacturing technologies in turn bringing economic considerations into focus? ROB Technologies creates software, and RoboFold and Odico both develop software and produce building products. What will surface when the approach is extended to design? Dutch designer Dirk Vander Kooij offers an interesting analogy. By mounting an extruder to an old FANUC robot, original and affordable chairs are produced by stacking contours layer by Layer, much like an oversized 3D printer. What is interesting here is not only the quality of the end product, but that the designer is not relying on a third party to develop the process or produce the chairs, challenging the traditional boundaries of the profession, and certainly so from an economic point of view: ‘If he is lucky, the designer gets 3% ex factory. The brand adds 300% and the shop doubles that again. It’s ridiculous how little of the cut a designer gets. If we used digital tools and changed the way stores

work, the ratio would be able to favor creativity and the craftsman. DESIGN AND BUILD // It would be misleading to mistake architecture for industrial design; even at the scale of the house, it is over-simplistic to think of architecture as a product. This said, the raw production capacity of industrial robotics does bring ‘design and build’ approaches to construction into view. Are the startups in architectural robotics revisiting the idea of an architect with a method of design and the means of production? ln 1996, Bernard Cache’s company Objectile set up a factory utilizing CNC milling machines. In 2000, architect Bill Massie built the Big Belt house, and more than a decade later companies like Facit Homes are revisiting the idea of the house as a product, where CNC is the enabling technology. Do these projects suggest a reconsideration of the early objectives of Modernism, to provide affordable and modern houses of architectural ambition? To what extent the new-found vicinity of construction is desirable remains an open question. The emerging ecology of knowledge and new possibilities from which both the architect and contractor profit is good news for architecture. Combining architectural ambition with a sense of economical pragmatism, robotic entrepreneurship challenges preconceived ideas of what it is possible to realize given a building budget - an essentially architectural agenda is shared by these young companies. Machineous is a fabrication and R&D

and market belief in the ‘third industrial revolution’, the impetus felt throughout the architecture, engineering and construction industry, the potential return on investments, and the low investment required to start developing architectural robotics, it is surprising that there are so few startups currently active. Are robotics and fabrication offering an apt bypassing of the maelstrom of architectural competitions, focusing on what is essential in architecture, or is there a derailing effect from taking up such a central role in the building process? Could it be Brunelleschi, the architect-engineer, versusLeon Battista Alberti, the intellectual architect, all over again?

The Nature of Robots

company that has the most experience in this domain. Since its inception in 2008, the company has sized up its operation to five robot stations, and recently moved to a large 1800 sqm facility in Los Angeles, completing the transition from a robotic artisanal workshop to an industrial operation. The company’s rapid expansion is matched by its increased scaling, from the production of bespoke furniture, installations and public sculptures to the production of building elements ranging from stairs to window apertures and facade panels. The crises of the American automotive industry be’ween 2008 and 2010 made plenty of inexpensive but capable robots available at a fraction of the cost, and Machineous has adopted a number of these from a former Chrysler plant. The automotive industry’s technological compost heap also fostered another startup: GREYSHED’s a design-research collaborative focused on robotic fabrication within art, architecture and industrial design. GREYSHED’s research has seen the integration of technologies such as augmented reality, gestural and sensor feedback in closely coupled design and fabrication processes. In its anti-institutional approach, experimentation and development of fabrication strategies are fuelled by the re-appropriation of affordable, off theshelf technological commodities ranging from smartphones and Kinects to chainsaws, cordless drills and refrigerator parts. Given the current economic climate of the building industry, the momentum



The Nature of Robots

Credits // Asbjørn Søndergaard //


ODICO // When Odico Formwork Robotics was founded in April 2012, it was apparent that to bring the fabrication principles of architectural robotics beyond the laboratory and prototype project, it would be necessary to fully confront market conditions. The founding partners – two architects, one engineer and one investor - thus set out to develop a manufacturing concept capable of providing low-cost production of advanced formwork at industrial volumes of scale. They were convinced that the internal, comparative tests of the Robotic Hotwire Cutting (RHWC) technique they had developed through their architectural research in 2009, exhibited a greater production capacity than equivalently scaled CNC-milling processes. When Odico began the commercialization of this technology in 2012, at was an international first. Currently hosting 15 employees in its Funen-based facilities in Denmark six of whom are industrial ABB robots – Odico has a business model that revolves around two main activities: formwork production, and research

and development (R&D) technology. Targeted at the concrete industry, the ongoing manufacturing provides advanced polystyrene moulds for architectural production. Replacing in a matter of minutes formwork previously requiring days of handcrafted labor, the factory production covers widespread, high-volume applications such as stairs and facade components, as well as bespoke, cultural artefacts in the form of sculptures and architectural structures. Complementing the manufacturing services provided for the construction industry, a significant part of Odico’s production resources are occupied by architectural-scale industrial prototyping and manufacturing for the wind-turbine and wave-energy industries. Alternating between physical production experience and architectural ambition, the second strand of the firm’s business strategy - the R&D technology – facilitates the maturation of manufacturing concepts along with incubation of new technological ideas. The activities are focused in two primary areas: the provision of robotic cell design licences and fully automated, lights out manufacturing processes. The hardware development laboratory is able to fully exploit the entirety of the factory space for 1:1 concept testing, while software development engages the sophistication of Odico’s core enabling technology: PyRAPID, a pythonOCC-based standalone RHWC-CAM application. Written by a chief technology officer with a purely architectural background, it exemplifies Odico’s founding spirit; that a

notion in a wider tectonic context, the applicability of the stereotomic experiences from Odico’s RHWC to robotic diamond wire cutting of marble in recent experiments, indicate that one measure of the robustness of a technological idea lies in its transferability across materials and platforms.

The Nature of Robots

technology innovation led by architectural aspiration can provide practical and economical industrial solutions that liberate, rather than delimit, architectural vision. The R&D technology is nurtured through strong relations with academia. Within its first year of operation, Odico headed the three-year BladeRunner research project funded by the Danish National Advanced Technology Foundation. Conducted in collaboration with leading practice, university and industry partners, the project investigates the low-cost production of large-scale double-curved freeform geometries through the development of robotic hot-blade cutting technology. Complementing this is the two-year Opticut pilot project in which cost-efficient realization of topology is being optimized, and concrete structures explored through ruled surface rationalization and RHWC fabrication. Through the fabrication of a 20 x 3 x 5m prototype structure at Arhus Bay, preliminary results indicate an approximate 24-fold increase in production capacity over comparable CNC techniques, leading to an 80 per cent reduction in machining time costs, thereby paving the way for realizing greater architectural ambitions within moderate budgets. Inspired by the ideas of Jean ProuvĂŠ that to renew the profession architects must engage deeply in the design and development of construction technology itself, Odico continues to pursue the objective of introducing disruptive CAM technologies to the architectural industry. Tentatively exploring this



The Nature of Robots

Credits // Gregory Epps //


ROBOFOLD // RoboFold was founded by Gregory Epps in 2008, after using curved folding obsessively as a method of design for more than 17 years. The technology translates the intuitive process of folding paper by hand into an industrialised system designed to fold metal using multiple industrial robots. The company builds on the knowledge of how to work with the aesthetics and physics of sheet metal to achieve effective design solutions. This novel approach demonstrates that innovation can reverse the ingrained concepts of conventional metal forming, a dichotomy that can be summarized through two opposing approaches. The conventional approach requires up-front financial commitment to the mould tooling used to form parts, which in turn requires production runs of thousands of identical parts to recoup the high costs involved. ln contrast to this is a more transient approach of evaluating multiple design options in digital design software, followed by direct manufacturing of one or more self-similar parts from the digital data. In the RoboFold

process speeds are comparable to mass production, however every part can be different without any financial penalty; in essence it is a form of rapid prototyping similar to 3D printing. Variation as Standard // At the 2012 Venice Architecture Biennale, Zaha Hadid Architects embraced the RoboFold technology with their dramatic sculpture Arum by creating an exciting new aesthetic using the system. The architects quickly realized the possibilities of a technology built from the ground up to embed the manufacturing characteristics in the design and control software. Output from generative and parametric design often seemingly offers ‘unlimited’ possibilities for architects, but all too easily defies the laws of physics and economics. Having developed a revolutionary new process for dealing with the physical constraints of folding metal with industrial robots, RoboFold must continue to advance research and development of the technology while maintaining an understanding that a balance between aesthetics, manufacturing and financial feasibility all contribute to its viability as an effective industrial process. The Process // RoboFold has developed and released a suite of CAD software plug-ins to manage the design-to-production workflow. The software is based in the popular Rhinoceros and Grasshopper platform, and manages each stage of the workflow to enable a parametric link from start

Robots to the Core // RoboFold’s client services are focused on developing production solutions that include software development, design consultancy, prototyping and licensing of its manufacturing technology for designers, artists, architects, and automotive, electronics and multinational industrials. The company’s experience in developing an end-to-end robotic fabrication

workflow for the patented RoboFold process has now been encapsulated in its new Robots.lO (short for Robots: lnputOutput) consultancy, which has an increased focus on high-value software and solutions for robotics professionals, again using Rhino and Grasshopper as a CAD platform. Accessing robots has never been easier, and their possibilities are becoming more apparent. Robots.IO creates custom robotic solutions and continues to provide Godzilla for robot owners to do the same. Recent projects range from web-app controlled robots to 3D scan-driven CNC-milling plug-ins and robot installations in industrial settings. Though this area of business is separate from the RoboFold brand’s cutting-edge metal-forming process which is now also a customer of Robots.IO - both will continue to benefit from the company’s continuously high level of investment in robotics research and development.

The Nature of Robots

to finish. Design starts with quickly realizing paper folding studies to ensure sheet material can be used from the outset. The development continues with a process of manual surface analysis that extracts the necessary data to simulate folding in KingKong, a Grasshopper plug-in using the Kangaroo physics engine, to create computational folding simulations as well as facade studies. The KingKong plug-in outputs the varied data in two forms: as flat patterns for cutting and as folding animations to drive the robot simulation. Cutting on the CNC router is facilitated by Unicorn, another Grasshopper plug-in, to generate the G-code CNC programming language. The Godzilla six-axis robot simulation plug-in is a powerful yet intuitive robot simulation environment in Rhino and Grasshopper where all the necessary checks for production feasibility occur. The final software stage sees Mechagodzilla take over and generate code for the robot on a remote Raspberry Pi. Once the metal is cut and scored on the router, it is positioned below the robots that pick it up with vacuum end effectors, and the folding begins.



The Nature of Robots

Credits // Andreas Froech //


MACHINEOUS // Machineous is a specialized fabrication facility that has made the use of CNC equipment the focal point of its activity and exploration in the making of architecture. The process begins with an architect’s CAD model and includes development of the design details, engineering, prefabrication, assembly and installation logic. Large industrial robots are the centerpieces on the factory floor and perform many of the initial steps to produce custom parts at many scales. The company has manufactured complex facade panels for Patrick Tighe Architecture and Kevin Daly Architects, experimental installation projects for Greg Lynn, Zaha Hadid, Hitoshi Abe and Tom Wiscombe, and furniture-scale products for Ammar Eloueini, Aranda\Lasch and Jeffrey lnaba. Machineous, based in Los Angeles, was founded by Andreas Froech in 2008. Trained as an architect in Vienna, Austria and at Columbia University in New York, Froech developed

his interest in the digitally supported fabrication of architecture when working with Greg Lynn and teaching at the University of California, Los Angeles (UCLA) School of Architecture between 1997 and 1999. Self-taught in the operation of CNC equipment, he was among the first to explore its potential for architectural fabrication. Prior to founding Machineous, he was Director of Material Development at Panelite where he was responsible for the development of several of the company’s patented and award-winning composite honeycomb panels and systems, as well as for specialized material research contracted by OMA and Prada. Robots are the company’s equipment of choice due to their outstanding strength, speed, precision and reliability. They are used to operate plasma cutters, spindle rotary cutters and large circular saws. A single operator and a single PC station can interact with any of the robots to produce a large variety of parts efficiently. Many parts are one of a kind but share a common setup with respect to the equipment. An intense front-end software process allows the operator to identify common process methods, developing cost and material savings to fit most construction budgets. Up to 20 staff execute processes including wood- and metal-working, industrial painting and project management. The extensive evolution of computer software-driven design allows architects to produce a wide variety of design solutions and to develop highly

Machineous operates with a deep Understanding of the nature and behavior of different materials, whether wood, plastic, metal or other, and of the behaviors, imitations and expressions of the robot. Fabrication processes and project-specific solutions are developed to exploit and embrace the interaction of specific material qualities with robotic technology. In the past, such high-level technology and fabrication processes were mainly only available to projects with substantial budgets. However, recent reductions in equipment costs combined with Machineous’s own efficient production and setup methods now enable affordable design solutions at any scale, from large metal facade screens with complex patterns, to freeform furniture and experimental installations.

Machineous is now enjoying rapid growth. It currently operates five large robotic stations integrated with a full service fabrication and finishing facility under one roof, and is geared towards further expansion to become one of the premier ornamental metal and specialist design fabricators in the US.

The Nature of Robots

complex forms. The x,y,z data required to generate CAD construction drawings and presentation renderings can also be used to drive robotic equipment. Machineous has developed many proprietary conversion postscripts to translate that same data into actual robotic arm movement for the production of architectural parts.



The Nature of Robots

Credits // Tobias Bowetsch // Ralph Bärtschi //


ROB TECHNOLOGIES // ROB Technologies provides software solutions that enable highly flexible digital fabrication processes for the efficient production of small batches of building components as well as individual nonstandard parts. Founded in 2010, the firm has a staff of three who combine expertise in architecture and building knowledge, robotics and software development to offer software environments that drive the robots of different customers from industry as well as research institutions. Based in Zurich, it builds upon the knowledge and experience of its founders, To-

bias Bonwetsch and Ralph Bärtschi, who for more than six years undertook pioneering research on robotic processes in architecture with Fabio Gramazio and Matthias Kohler at ETH Zurich, where together they hold the Chair of Architecture and Digital Fabrication. At ROB Technologies, the industrial robot is understood not only as a means of achieving greater efficiency in production processes, but also, due to its inherent flexibility, as a tool that holds the potential to enhance a greater degree of freedom in design and construction. This potential combined with the ability to adapt to diverse production jobs and to realize a multitude of different fabrication processes, should make industrial robots the tool of choice within the building industry. However, in relation to the size of the market, the number of industrial robots actually performing building tasks is negligible. Generally, the arguments for highly flexible automated fabrication processes are indisputable: on the one hand sophisticated digital design tools are available, while on the other the dexterity of industrial robots enables the performance of arbitrary fabrication tasks. Unfortunately, though, in reality a gap still exists between the conceiving and planning of a design and its execution by (just in theory) highly flexible industrial robots. The problem seems to be located in the actual activation and utilization of this flexibility. Robot manufacturers only provide proprietary old-style robotic programming languages that have a limited level of abstraction. Industrial

non-standard brick façade elements is thus realized. Aside from fully realized prototypical facade projects like the Ofenhalle in Pfungen by architects Gramazio & Kohler (2012), the first large scale commercial projects based on the BrickDesign software, in 2014, will be the Le Stelle di Locarno residential building built in Ticino, Switzerland, by Buzzi studio d’Architettura, a further residential building in London, and sports facilities in Manchester, in 2014. ROB Technologies’ CAD-based URStudio software environment for off and online programming of universal robots offers bi-directional communication. The goal is to ease complex task programming of the robot through a unique combination of teaching and manipulating virtual geometries, without the need to descend into programming machine code. The aim of ROB Technologies is to foster the application of industrial robots in architecture by helping to activate their intrinsic potential, so that robots can become a truly powerful tool for innovation in the future of building and manufacturing. The Nature of Robots

robots need to become easier to control and more intelligent in perceiving their environment. As the programmability of industrial robots is pivotal, ROB Technologies concentrates on the development and provision of software solutions. To make the control of robotic systems more flexible and easier to use, thereby enabling companies to exploit these capabilities in their manufacturing tasks, ROB Technologies locates the starting as the design phase. The company provides fabrication-specific design tools that are combined with flexible control of robotic systems. Programming the robot and adjusting the system to modified outputs and processes can therefore be performed by non-experts in a fast and efficient way. By primarily targeting the construction industry, its core technology is also of high interests to other manufacturing industries that require automation but are reluctant to invest in traditional robotic solutions due to their complexity and costs. One of ROB Technologies’ exemplary products is BrickDesign, a comprehensive approach to the design and robotic fabrication of brick facades. The software incorporates parameters of the robotic fabrication process already in the design phase, exploiting the capabilities of the robot to effortlessly position each individual brick differently. The data produced is directly used to control the fabrication process without the need for additional process programming of the robotic system. Efficient and highly flexible automated production of



The Nature of Robots

Credits // Ryan Luke Johns //


GREYSHED // is just that: a 4 x 8 metre breeze block garage. From behind a rolling door in a row of postwar homes in suburban Princeton, New Jersey, a dull machine hum emanates. Inside, an assemblage of components salvaged from local scrapyards work to convert the common domestic power supply to the three-phase, 480-volt demands of a two-and-a-half-tonne industrial robot. ‘Abraham’ is a late1990s welding robot that was stripped from a Ford factory line and acquired for less than $10,000 from a second-hand robot dealer found on eBay. While the complications of transporting, recommissioning and operating such a behemoth are far from trivial, the unfamiliarity and scale of the machine belie its relative simplicity. GREYSHED aims to diminish the mystique surrounding this technology by expanding the territory of architectural robotics from the advanced institution into the unremarkable and unsupported suburban landscape. Just as the domestic appropriation

of military technologies in the 1950s turned the marvelous to the mundane, GREYSHED takes the robot off the pedestal and puts it in the carport. The robot is not the future: it is already here. Under that assumption, GREYSHED explores complex design experiments with commonplace components. Recent projects turn used smartphones into augmented-reality headsets, appropriate game console hardware to create gestural design platforms, and enable mediated toolpath manipulation with touchscreen tablets. Likewise, the robot’s end effectors are hacked together from cheap, secondhand power tools. Precise fabrication is executed with an electric chainsaw, handheld router or bandsaw, glue or resin extruders are powered by a cordless drill, and the vacuum-gripper runs on refrigerator parts. While the low-budget nature of such independent research can be limiting, the associated flexibility is simultaneously liberating. Essentially, GREYSHED is a decision to exchange institutional red tape for duct tape. The balance between limited resources and resourcefulness characterizes not only the means by which GREYSHED approaches its work, but the work itself. It is not the lack of constraints, but the variation of traditional constraints that enables a novel approach to design. The Robotic Poché project, for example, explores ‘slow fabrication’ procedures while simultaneously dealing with the practical problems of a poorly insulated, garage-based studio. In order to de-

Operating somewhere between research and practice, GREYSHED fluctuates as it must from laboratory to design studio, consultancy and fabrication shop. Founded in 2011, it follows the entrepreneurial spirit of garage innovators by balancing research, play, production and collaboration. Propelled by polarity, it explores the space dividing the traditional architectural dichotomies of design/construction, digital /analogue, stochastic/ deterministic, man/machine,simulation/execution and amateur/professional. Through the simultaneous occupation of multiple phases of the design-production spectrum, GREYSHED seeks to create not only ‘highly informed’ architecture, but highly informed architects. Operating at a localized scale of ‘byte to robot’ rather than ‘file to fac-

tory’, the design-prototype-production sequence is compressed into a feedback loop that empowers the designer to preempt problems generally faced by engineers and contractors long after the initial design impetus has passed. By fostering concurrent computation, construction, craftsmanship and design, GREYSHED works to advance digital fabrication while revitalizing the role of the human designer.

The Nature of Robots

crease heat loss during winter research while reducing the noise transmission associated with fabrication work, the project engages material assemblies that provide both thermal and acoustic isolation - decoupled surfaces with foam infill. Here, the robot is used as a reconfigurable formwork for laying a complex configuration of ceramic tiles over the imprecise, pre-existing ceiling structure. By holding each tile in place while the void between the tile and the ceiling is filled with expanding polyurethane foam (and remaining in place until the foam cures), the finely tuned and inhumanly patient manipulator makes traditionally crude materials and processes viable tools for digital fabrication.


The Nature of Robots

Part II - Research and Projects


Building a Bridge with Flying Robots

challenges for implementing the construction method on a full-scale, loadbearing, architectural artefact. Firstly, a series of tensile links is fabricated at defined lengths between two distant support structures to build the primary elements of the bridge. Secondly, cooperating flying robots brace the assembly by braiding the primary elements to one another. And finally, the structure is stabilized through the fabrication of additional connections by robots flying around existing elements within the porous structure.

Credits // Ammar Mirjan // Federico Augugliaro // Raffaello D’Andrea // Fabio Gramazio // Matthias Kohler // ABSTRACT // The research presented here investigates techniques and tools for design and fabrication of tensile structures with flying robots. Tensile aggregations are described as a concatenation of nodes and links. Computational tools provide the designer of such a structure with the necessary aid to simulate, sequence and evaluate a design before fabrication. Using a prototypical suspension footbridge as an example, this paper describes the techniques and

01 - INTRODUCTION // Today, digital fabrication is predominately realized with devices that are fixed to the ground. The solid fastening of a robotic arm or the movable parts of a CNC-machine to a base ensure precision in material manipulation. A static environment is assumed in order to calculate the spatial situation of an end-effector for trajectory generation and position control. Recent developments in sensing, computation and control, however, allow the creation of autonomous construction machines that are mobile and have the ability to localize themselves in unstructured environments. Flying robots are examples of this type of machine and are becoming increasingly relevant in robotic construction. Aerial robots can be used to move construction elements to locations not accessible by ground robots, they can maneuver

The Nature of Robots

KEYWORDS // Aerial robotic construction // Tensile structures // Cooperative fabrication


The Nature of Robots 46

around existing objects to aggregate construction elements and they can fly in or around already built structures and manipulate them. Flying robots have profoundly different capabilities to established mechanical fabrication devices and, as such, may disrupt the conditions for how architecture is designed and materialized. From the assembly of an architectural scale tower structure built from discrete foam elements, the construction of experimental truss structures the assembly of space frame structures, to the 3D printing of structures the past few years have seen various robotic construction projects incorporating flying robots. The authors of this paper are specifically interested in the fabrication of tensile structures with hover capable Unmanned Aerial Vehicles (UAVs), such as quadcopters. This fabrication method fully exploits the capabilities of the flying machines and allows fabricating loadbearing architectural structures that no other machine could build. Flying vehicles such as helicopters are applied on construction sites since the 1950s. They are used in bridge construction to transport prefabricated building elements to the site and to string pilot cables between the two sides to later pull suspension cables across. Here, robotic control of flying machines offers potential in the realization of structures. The machine not only substitutes a crane with almost unlimited reach, but, through the linking of computational design with fabrication, allows new forms of material interaction.

02 - TECHNIQUES // In natural formations, as well as in manmade constructions, tensile structures, such as cable net structures, usually act continuous in material but are structurally non-linear. This is also the case in this project, where a continuous, flexible building material is used to aggregate an interconnected assembly, however, the research presented here regards the actual fabrication of such a structure as a sequence of discrete building modules connected to one another. The aerial aggregation of tensile structures can therefore be summarized as choreography of two basic modules, a concatenation of nodes and links. Here, we describe their parameters, as well as computational tools that enable the design of aerially buildable structures. 02.1 - NODES // A node is a point of intersection where a tensile construction element, such as a rope or a cable, interacts with another object or with itself. A node can be a solid fastening to an element, a knot, or it can be a sliding connection. In previous work, we have presented a general framework that permits descriptions of nodes that can be realized by flying machines. This framework consists of three parts. Firstly, in the knot theory, a branch of topology that studies mathematical knots, a node is represented as a knot diagram, where the node is projected onto a plane and crossings are identified and numbered. These numbers listed in a matrix define the node as code. Secondly, since knot theory only math-

02.2 - LINKS // A rope spanned between two nodes generates a link that resembles a catenary curve. The parameters of a link are its two support points and its length or tension, defining the sagging of the link. While a node, once it is built, does not change its characteristics, a link might change

its shape over time, since every intersection with another link results in a new equilibrium state. The flying machine is equipped with an active rope dispenser; a motorized roller, which allows the force on the rope to be controlled during its deployment. This allows links to be created with different lengths or tensions. The payload capacities of flying machines are constrained and their maneuverability is greatly influenced by the load. These mechanical constraints limit the solution space of buildable links in relation to the weight of the rope and the tension applied (the flying machines used here can apply approximately 2 N-3 N of force while still being controllable). The shorter the link, the higher tension can be applied and vice versa. 02.3 - IRREGULAR MODULES // Most tensile structures buildable by flying machines can be realized, as a concatenation of nodes and links. As described above, they can be generally defined as basic building modules. The amalgamation of these modules permits the creation of manifold physical realizations. The unique capabilities of the flying machine, however, also allow the fabrication of tensile elements that are neither nodes nor links; they are situated in between, requiring specific methods and tools to be utilized. The density of this surface was limited by the size of the flying machine and the distance between the support points. In contrast to this earlier work, the weaving module makes use of the ability of the material to slide.

The Nature of Robots

ematically describes a closed knot with joined ends, the knot representation has to be modified in order to take into account the actual fabrication of the node as a sequence of moves on a support element. This topological representation of the node does not incorporate spatial information (such as scale, position and orientation) of the node, which is required to generate the trajectory for the flying machines. Hence, finally, a three-dimensional trajectory is generated from the knot code by incorporating the following parameters: the position of the knot (3D-point), the orientation of the support element (3D-vector) and the approach direction of the flying machine (3D-vector). These three parameters define a 3D plane and add the needed spatial information to the node. Following the knot code, a node is realized as a series of circular movements around the center point. The described generalized method allows the fabrication of any node with constant tension (winding knots), assuming there is enough space to fly the node. An interesting challenge will be to incorporate the realization of nodes with loops (looping knots), where the vehicle creates a hanging segment to fly through.


The Nature of Robots

Therefore, in this method, the flying machine creates a surface-like structure by flying a figure eight like trajectory around already built structural members (similar to the zigzag) with an additional pullback movement after every crossing. This pullback movement closes the space that was needed for the machine to fly through, while creating a dense filling. The density of the filling can be adjusted by flying additional circles around the support elements, similar to woven vinyl cord of the Acapulco Chair. Another building module for the aerial fabrication of tensile structures, the braiding module, is both a link and a node. Interlacing multiple strands of tensile material, overlapping and crossing each other, in an intertwined, often linear, manner forms a Braid. While most of the elements described above can be built by one vehicle, the fabrication of braids usually requires multivehicle cooperation. The construction of a braid with more than two strands cannot be sequenced for a single vehicle and therefore requires the interaction of minimum two flying machines.


02.4 - SIMULATION // SEQUENCING // EVALUATION // In order to be able to design tensile structures that are buildable with flying robots, a series of computational tools have been developed, specifically addressing the characteristics of the building method. These tools allow the simulation, sequencing and evaluation of structures incorporating material, machine and fabrication constraints.

02.4.1 - SIMULATION TOOL // Designing the form of an active structure, like a cable net structure, is challenging since its shape is not known in advance. Designing with linear tensile elements, such as ropes or cables, demands-the aid of form finding techniques to statically determinate the structure acting in pure tension under self-weight. Furthermore, a tensile structure might contain nodes that are not fixed (a simple turn node), sliding on structural members. Hence, this work proposes a tool for the simultaneous physical simulation of tensile elements, as they act under gravity and collisions. The tool combines a design environment with a physics engine: the design information, such as support points, link length or node type, are defined in McNeel Rhinoceros 3D, while the physical simulation in Maya Nucleus runs in the background. 02.4.2 - SEQUENCING TOOL // The fabrication of tensile structures with flying robots does not require building from the ground up. The order of when and where a link is constructed does not have to be linear. This design freedom implies additional complexity. The form of an active structure changes its shape with every newly built interacting link. The spatial situation alters over time, constraining the path a vehicle can take. The design therefore has to incorporate the spatiotemporal performance of the structure and simulate it step by step. To take this into account, we propose

02.4.2 EVALUATION TOOL // Prior to a material realization of the design, the digital model has to be evaluated on whether it is buildable according to environmental, physical and mechanical constraints. These constraints influence the design of the structure and have to be integrated into the design process. Each node type has a specific solution space in relation to the orientation of the support element and the approach direction. Alongside the sequencing and the simulation of the structure, each node is evaluated firstly regarding its orientation and secondly regarding collisions with the environment and already fabricated elements, respectively. The vehicle size and maneuverability influence the solution space. 03 - IMPLEMENTATION // The building modules and design tools described above have been individually tested and adopted in separate experiments. Here, the research takes the important step of interlinking these

single elements and testing heir synthesis in a proto typical architectural context. Multiple flying robots fabricate a full scale, loadbearing footbridge, spanning 1.5 m. 03.1 - EXPERIMENTAL SET UP // The bridge is constructed in the Flying Machine Arena,3 a l0 x l0 x l0 m indoor space for aerial robotic research. The space is equipped with a motion capture system that provides vehicle position and attitude measurements. This information is sent to a PC, which runs algorithms and control strategies and sends commands to the quadcopter (acceleration and body rates). As previously described, the vehicles are equipped with a motorized rope dispenser, enabling the dynamical adaption of rope tension during its deployment, as well as estimating the length of already placed rope. The bridge is fabricated from ultrahigh-molecular weight polyethylene rope (Dyneema) with 3 mm and 4 mm diameters. The material distinguishes itself for aerial manipulation due to its high strength and low weight. Its weight-to-strength ratio is around 8-15 times lower (better) than that of steel. A 100 m long rope with a diameter of 4 mm weighs 700 g and can support 1300 kg. Its low stretch and positive durability properties (water, chemical and UV-resistance) make it useable for architectural applications. Additionally, Dyneema has a low coefficient of friction, allowing the material to slide easily against itself. This is beneficial when the rope has to slide under load to find a structural equilib-

The Nature of Robots

designing aerially buildable tensile structures sequentially, according to the actual fabrication order (not as a global representation). A design usually starts with a single node on a support point, followed by a link to another support point, and so on. The whole structure is simulated with each new link, creating a time-fabrication based digital model of the artefact. All steps are recorded, allowing jumping back to specific steps in order to implement design changes.


The Nature of Robots

rium, but it is also challenging during the fabrication of a node when minimum sliding is required. The white cube nature of the building space offers few options for support points for building. In response to that, two scaffolding towers were erected and solidly fastened on either side of the space. The round horizontal, vertical and diagonal bars offer multiple options for support, allowing the realization of a variety of structures in the space.


03.2 - REALIZATION // The bridge is fabricated in three consecutive steps. Firstly, three tensile links are erected at defined lengths between the two distant support structures to build the primary elements of the bridge, responsible for sustaining the majority of the loads. Three vehicles simultaneously construct one tensile link each, using a variation of the boom hitch as a start node, flying to the other side and fastening the link by constructing a Munter hitch. These primary tensile links could have been fabricated by a single machine, building one link after another, however, parallelization of the task substantially speeds up the fabrication process. When viewed in section, the links are arranged in a V-shape. The bottom link supports the feet of a user when crossing, while the two parallel upper links provide the handrails. As mentioned, the building material is low stretch, however, when a link is loaded the nodes tighten, resulting in the sagging of the link. Therefore, the three links are

constructed with the maximum possible tension to make the crossing more comfortable. After the fabrication of the main links, two vehicles brace the assembly by braiding the primary elements to one another. First, the machines simultaneously erect a node on the handrail support points on one of the scaffolding towers. Then, a series of braids at the center link and single tum nodes at the handrail links are realized. The vehicles navigate to the center link and construct an 800 mm long braid crossing each other before making a turn node around the respective handrail link. This secondary structure braces the bridge and joins all the elements to a structural whole. The connections between the central link and the handrails distribute the forces when the bridge is loaded, while the sliding of the single tums at the handrails allow the bridge to dynamically adapt its shape and find an equilibrium according to different load cases. Finally, in the last step, the bridge is stabilized by adding additional links to the structure. First, two links are erected between the scaffolding towers below the central link. Then these two connections are joined to the central link one after another by flying a zigzag trajectory around them and through the openings between the central link and the handrail links. The size of these openings, defined by the length of the braids and the gaps between them, is dimensioned for the vehicles to fly through. The bridge can be crossed securely without the addition of the stabilizers.

04 - CONCLUSION // The work presented here demonstrates for the first time the use of flying robots for the construction of a full scale, loadbearing architectural structure. A framework for representing and building tensile joints, a method to fabricate links at defined lengths, as well as computational tools that allow the simulating, sequencing and evaluation of structures enable the design and fabrication of aerially buildable suspension structures. The prototypical artifact described in this paper showcases the ability of the vehicles to architecturally aggregate material independently of the ground conditions and the machine size. The ability of the vehicles to fly in and around existing objects is utilized to interconnect existing members of the assembly and create a structural ensemble. Cooperation between machines through parallelization accelerates the production process, while the cooperation through interaction allows the fabrication of structures that a single vehicle could not realize. The V-shape of the footbridge allows the safe crossing of the structure. However, the traversing could be enhanced with the integration of the winding module by introducing a surface structure to walk onto, instead of a linear tensile element. A further

interesting challenge is the aerial robotic fabrication of architectural structures in an outdoor environment. The work demonstrates that flying robots are not constrained to aggregate material layer by layer from the ground up and proposes thinking about robotic construction as a nonlinear, sequential set of operations in material interaction. As such, the realized bridge structure does not mimic the usual manual process of building such a structure but reinterprets it using the unique abilities of the flying machine.


presented here is based on a collaboration between the Institute for Dynamic Systems and Control and Gramazio Kohler Research, ETH Zurich. The experiments shown here are performed in the Flying Machine Arena [3] at the Institute for Dynamic Systems and Control at ETH Zurich. The work is supported by the Hartmann Mueller-Fonds on ETH Research Grant ETH-30 12-1. A special thanks goes to Augusto Gandia and Maximilian Schulzu who have contributed to the work.

The Nature of Robots

However, integrating them helps to absorb non-uniform loads, and possible lateral and uplift forces. It alters the artefact from a passable structure to a usable footbridge.



they deviate from a predictive virtual model. Establishing communication strategies for a live-control pipeline as the infrastructure for this system allows the system to respond to prebuild scans of part dimensions, as well as update the virtual model when post-build scanning detected deviation. In the worst-case scenario - if preconditions were not met - the postbuild scan would be unsuccessful and the system would self-terminate. Otherwise, deviations would update and influence future actions. This influence is what leads to the indeterminate nature of the resultant forms. KEYWORDS // Autonomous robotics // Adaptive control system // Sensor feedback // Material variability // Assembly

The Nature of Robots

Credits // Michael Jeffers


ABSTRACT // This research discusses the problems within autonomous robotic assembly workflows as they encounter a variable property of assembly parts or materials. This is shown through a case study with an industrial robot in an enclosed work cell and a simple assembly task with wooden sticks of variable lengths, designed as an adaptive feedback control system. To perform the study, the development of a virtual model for the persistent storage of material data and computation of next build-actions is required. Different sensing strategies are used to address issues of substantial, and minute, material variability of dimensional properties as

01 - INTRODUCTION // This research illustrates the problems within and methods for addressing autonomous robotic assembly using a case study and, by its current implementation, a set of constraints as proxies for real world conditions. The purpose of the case study is to demonstrate viability in projected real-world conditions. Incorporating wood materials, eye-inhand sensing, and adaptive feedback control within autonomous robotic assembly processes are each targets for providing evidence and testing of some advantages and disadvantages of this approach. The case for autonomous robotics has largely been made in the context of dangerous or difficult tasks where human actors are supplanted by robots.

01.1 – MATERIAL VARIANCE // This case study is designed to use length as a highly variable factor, and to include other properties and uncertainties of the wood material which might introduce noise to the system that could neither be preemptively determined nor engineered out. The domain of variability is the variance in property. Member length has high variance, whereas slight irregu-

larities along the profile have low variance. It is found that highly variant properties, those that exceed design tolerances of the system, must be known prior to build-action. Low variances, on the other hand, might not be detectable within the resolution of the sensing techniques or devices. In the case of assembly, error may accumulate beyond tolerances and therefore requires the introduction of post-build verification and error recovery. Similar work explores the same single dimensional variance but with focus only on the initial analysis of the parts and its placement in order to minimize subsequent error. This case study demonstrates a number of methods to overcome both high and low material variances. An experiment was designed to replicate practical scenarios of standard cuts or drops of material that are delivered within some tolerance; nominal versus actual dimension. Handling, moisture, and storage life can con tribute to more deviation between expected and actual dimensional properties especially in the case of wood. The robotic process, even when standardized, has to adapt to such alterations if tooling and manipulation requires non-compliant registration. The difference here is that length is highly variable and the system is designed to take into account any length. This value, once known, strongly characterizes subsequent actions that depend on this data. The low variance of material irregularities plays itself out only over the course of the assembly task.

The Nature of Robots

This has resulted in certain materials and processes being quickly adopted and heavily engineered into a precise process to minimize error. The increasing need for varying tolerance or adaptability of assembly onsite limits the implementation of robotics in construction applications. Fewer factors can be ensured as reliable preconditions and are usually compensated by on-site decision-making based on human observation and measurements of actual conditions. If the task is known but the nature of subtasks includes factors that vary this can still be a closed system with feedback. This is the problem considered in this research: How can an autonomous robotic system accomplish a known task/goal if some subtasks are known to be variable? Other questions that have emerged and worth considering are: Can we assemble a known form with unknown parts? Or rather, can we assemble unknown parts in a known manner? The latter is used to isolate behavior, although ensuring that local rules are satisfied leads to difficulty in determining outcome.



The Nature of Robots

02.1 – COMMUNICATION // An autonomous system with feedback needs the ability to compute next actions based on the feedback. Communication between the external computer (client) and the robot controller (server) is via TCP/lP. Other projects, for example Hal, Scorpion, Robo.Op, the server and client described in “Interlacing” and ROS, tackle the same functionality with each adapting to different client-side or robot controller environments. The client in this case processes and communicates with the sensor devices, parses inputs, computes a virtual model, and produces commands for the robot. This was developed in Java using the standard library with graphical elements leveraging the Processing API and associated libraries. An important criterion in the design and implementation is that this communication must be synchronous. The server is often occupied with executing motion commands, while the client could easily overflow a queue of sent messages. A send-receive pattern is enforced by a handshake protocol.


02.2 – VIRTUAL MODEL // The virtual model is the persistent data structure used to store and update data relevant to the assembly building process. The build-algorithm parses this data and determines next actions. Its design is strongly linked to the nature of the assembly process. The more abstract the model, the fewer input

variables that can be detected and taken into account. It is therefore assumed to be constant. The stacking model is 2.5D. At each ‘level’, including the base-plane, construction is planar and parallel. There are a number of factors about wooden sticks, which obscure and defeat this ideal. This discrepancy serves as a proxy for other on-site issues of dynamic environments, materials, and noise in the system that post-build scanning processes are designed to compensate for and maintain validity of the virtual model. Consequently, the purpose of the case study is to determine what data can be predicted, to what extent, and when is revision required. For the virtual representation, any stick is reduced to: a pair of endpoints, a length, and gripper location as a factor of length. This is sufficient for generating higher-level information for subsequent calculations. Endpoints are first determined and then used to iteratively test potential placements of the stick for a valid assembly. Stick length is the seed for this process. Boundary geometry, for representation or collision detection, is generated from this information. Within these rules different geometries are used to perform different relational calculations. The centerline is used when trying to determine bearing conditions, whereas a scaled profile is used or collision predictions. The gripper boundary is computed for collision only for the current stick that is being added to the assembly. This multiplicity of representations allows for more economical computation of otherwise

02.3 – SCANNING METHOD // PreBuild: Measuring // Sensor feedback is isolated to two specific steps in the build-cycle: pre- and post-build. Prebuild scanning is used to acquire stick length, a process that involves signal processing to ensure that certain preconditions are met before the start of each robot action. A signal is tripped when ‘registering’ a material or part against a known location. Typically,

changes in input signal can be used to stop a robot motion, or extract its current position etc. To calculate stick length, if location of the limit switch and robot approach vector are known, one can compute the difference in distance from where the switch is and the distance away where the Tool Center Point (TCP) is located at the time the switch is tripped. Measuring both sides of the stick gives us the length and position of the TCP as the dispenser does not center the stick at the location of grasping. Post-Build: Verify and Search // Different devices afford different advantages, however some require developing a process to acquire higher-level data beyond what the given sensor is tailored to provide. Positioning these within the build process (pre- or post-) is weighted according to their sensor attributes. The Parallax Laser Range Finder (LRF) or Sharp GP2Y0A02YK0F Infra-Red (IR) sensor provides depth from the point of emission, within a degree of fidelity and error, delivering new readings at an approximate rate. Hosting sensor devices at the End of Arm (EOA) has the additional advantage of taking a single 1-Dimensional sensor and allowing it to acquire data about any non-occluded surface in the robot’s work envelope. To verify the depth of a stick on its centerline, the LRF is used to hit multiple points on the length of the most recently placed stick. Comparatively, the LRF gives more reliable measure of real depth than the IR sensor. The latter fluctuates with the analog signal

The Nature of Robots

complex 3D relationships by leveraging the simplest representation required. The build-algorithm produces not only tall but also relatively stable stacks. Height was a simple metric that would present more challenges with respect to stability when faced with accumulating error from sticks below. Looking for a maximum span condition helps reduce cantilevered ends. Otherwise, should sticks accumulate heavily to one side, it will cause the stack to topple. This would produce an error-state that would later be detected in scanning. The current rules and build-algorithm only take into account the current stick as the object that is manipulated. Thus, the remaining virtual model is effectively frozen. This is designed to allow for subsequent incremental mechanical fastening of parts in the assembly to reflect more closely a real assembly process and to reduce the search-space for scanning processes. Economizing what data needed to be recovered, updated, and stored is vital for reducing the complexity of the scanning problem and the resources it requires.


The Nature of Robots 56

processing and is slightly affected by the color of the target. However, the LRF scans at about once per second, which is slow for a single reading. As a result, IR sensing is used to populate depth values in a grid over a known location, despite its increased output noise. There are two major components to the depth grid. First, previous as for example “Stock Finder� utilizes a quad-tree like method of self-subdivision that has proven advantageous for selectively refined grid-based processing. Notably, statistical culling of outliers and determining which cells are significant for further examination are of value when paired with an expected depth value. The second is to derive a new endpoint of a stick that is either within the search radius or not. If it is not present, a threshold needs to be determined at which the search should abort. If it is present, the question then is how to use a matrix of values to indicate an endpoint of a stick. Of all the implementations, linear regression proved the most reliable at low resolutions. Linear regression is used to model trends in the data assuming a dependence of one value along an axis against an independent variable on the perpendicular axis. Understanding this assumption does not hold, linear regression could be performed over both the x- and the y-axis, then compared for the best match or averaged as they converge (Fig. 7). The line segment generated better approximated an endpoint at lower resolutions than other algorithms like the convex-hull, which required higher levels of refinement and further analy-

sis to determine the endpoint. 03 – RESULTS // The system is able to construct stacks of sticks without collision, with proper bearing conditions, and able to determine at each step if and where it could build higher. Furthermore, it can detect and compensate for disturbances, either as a result of accumulated error from subtle material variance or from a dynamic environment. Tolerances of the system are relatively low, as first pass post-build scanning would allow for certain amount of shifting. This tolerance however is incorporated on the front-end of the build-algorithm when detecting collision conditions. Establishing this tolerance range as a baseline lets the system decide when further scanning is needed. Scanning processes are targeted to the scope of material variance they are designed to compensate for. Pre-build scanning is used to acquire the length of the stick, which largely determines where such a shape could be placed. Post-build scanning had two phases, the first of which is to verify the placed-stick. Pinging from a known depth, the virtual model produces an expected distance that should be acquirable from the top of any given part in the assembly. Should measured distances deviate from expected distances, a second-phase recovery process is initiated because the stick is not present in the expected location. In its current implementation, this can only account for the stick being slightly rotated or translated from its intended

04 – DISCUSSION // Design processes cannot predict on-site realities, but they can provide a framework for a decision-making process with respect to these variables. The role of the designer here is not one of user interaction, but of establishing the rules that govern the autonomous system. Furthermore, the design here has no formal value as the parts are assumed to be entirely unknown. This indeterminacy of resultant assemblies relates directly to the rules that govern the stacking and the recovery from deviations that occur within the process as a result of material irregularities. Therefore trying to force this to achieve a desired form is futile, but providing metrics instead has the opposite effect. 04.1 – FUTURE DEVELOPMENT // The case study demonstrates the system’s scope of behavior in the context of variable materials. Variable context, or dynamic environment, demands attention next to fully address the problem of on-site robotics. Sensing in

this case expands from analyzing a task at hand to constant monitoring of surroundings. Many variables can be ruled out with enclosed work cells for repeatable success and safety. Short of developing a mobile platform that can interface with just one task on-site there is additional work with creating automated workflows involving multiple tasks. Integrating complementary processes such as automated mechanical fastening will complete the ambitions of a truly autonomous assembly process. Working with standard and reclaimed wood products will also push the precision sensing of this system forward to be able to adapt to highly variable materials. Even within the problems found in the case study, the boundary where unrecoverable failure occurs can be pushed further and additional cases for error recovery can be adopted. Understanding this failure-state as a moving boundary between the known, the unknown, and up to the unknowable is a critical observation for analyzing the viability, repeatability, and safety of an autonomous system. 05 – CONCLUSION // This research presents a case study, which demonstrates an implementation of a closed adaptive feedback control system for autonomous robotic assembly with known material variance. The four fundamental components of this system are: memory, feedback, decision, and actuation. Construction processes can be described through this framework. Construction documents (mem-

The Nature of Robots

location. Should the stick exist within the search-space of the recovery, the data acquired is processed to create revised endpoints of the new location of the stick object. This subsequently updates the virtual model and a first phase verification scan then confirms the update. When the stick is not found within this search-space, we cannot guarantee the success or safety of further actions and therefore the system enters a failure-state.


The Nature of Robots 58

ory) detail a task at hand. One may observe discrepancies (feedback) that are not present in the document. Action based on information given and observed must be taken to best satisfy any directives in the document (decision and actuation). The system developed for the case study follows the same form. Memory is contained in the virtual model. Feedback comes from the scanning processes that supply the real-data to revise model information. Decision-making is shown as the build-algorithms. This framework can therefore apply to a construction task that can be described in the same manner. On a high level, most construction tasks may sound like procedures. When examined, they are described more precisely as an algorithm with many low level rules that may be referred to as intuition or skill. If this methodology is adopted and additional sources of variance can be identified and incorporated, autonomous robotic systems in construction becomes a more immediate future. This opens the possibility of not just precise execution, but rather autonomous construction systems that can operate with little oversight – observing and adapting to a dynamic environment. The result may satisfy the task, but the exact nature of the form will be a consequence of on-site decisions, as it is with construction today.

ACKNOWLEDGERNTS // This research was generously supported by Carnegie Mellon University’s school of architecture ad associated staff and faculty. The authors would like to express their gratitude to Josh Bard, Ramesh krishnamurti and Richard Tursky for their generous advice ad support.


rication, and introduces the context of structures where the skin plays an integral role. It describes the development of an integrated approach tor the modelling and fabrication of stressed skins, an incrementally formed sheet metal structure. The research then focus upon the use of prototypes and empirical testing as means to inform digital models about fabrication and material parameters including: material forming limits and thinning; the parameterization of macro and meso simulations with calculated and observed micro behavior; the organization and extraction of toolpaths; and rig setup logics for fabrication. Finally, the validity of these models is evaluated for structural performance, and for geometric accuracy at multiple scales.

Credits // Paul Nicholas // David Stasiuk // Esben Norgaard // Christopher Hutchinson // Mette Ramsgaard Thomsen // ABSTRACT // For structural assemblies that depend upon robotic incremental sheet forming (ISF) the rigidity, connectivity, customization and aesthetics play an important role for an integrated and accurate modeling process. Furthermore, it is critical to consider fabrication and forming parameters jointly with performance implications at material, element and structural scales. This research briefly presents ISF as a method of fab-

01 - ISF // Incremental sheet forming (ISF) is a fabrication method that imparts 3D form onto 2D metal sheets. It is driven by 3D CAD models and has been developed for the purpose of industrial prototyping within the automotive industry. In the most typical ISF method, a ball-head tool is moved over the surface of a thin metal sheet, causing a progression of localized plastic deformation. ISF is useful for three reasons. First, it negates the need for time-intensive creation of costly dies (negative forming), instead directly machining semi-finished pieces of metal. Secondly, because form-

The Nature of Robots

KEYWORDS // Incremental sheet forming // Mass customization // Robotic fabrication // Toolpath optimization //


The Nature of Robots 60

ABSTRACT // For structural assemblies that depend upon robotic incremental sheet forming (ISF) the rigidity, connectivity, customization and aesthetics play an important role for an integrated and accurate modeling process. Furthermore, it is critical to consider fabrication and forming parameters jointly with performance implications at material, element and structural scales. This research briefly presents ISF as a method of fabrication, and introduces the context of structures where the skin plays an integral role. It describes the development of an integrated approach tor the modelling and fabrication of stressed skins, an incrementally formed sheet metal structure. The research then focus upon the use of prototypes and empirical testing as means to inform digital models about fabrication and material parameters including: material forming limits and thinning; the parameterization of macro and meso simulations with calculated and observed micro behavior; the organization and extraction of toolpaths; and rig setup logics for fabrication. Finally, the validity of these models is evaluated for structural performance, and for geometric accuracy at multiple scales. KEYWORDS // Incremental sheet forming // Mass customization // Robotic fabrication // Toolpath optimization // 01 - ISF // Incremental sheet forming (ISF) is a fabrication method that imparts 3D form onto 2D metal sheets. It is driven by 3D CAD models and

has been developed for the purpose of industrial prototyping within the automotive industry. In the most typical ISF method, a ball-head tool is moved over the surface of a thin metal sheet, causing a progression of localized plastic deformation. ISF is useful for three reasons. First, it negates the need for time-intensive creation of costly dies (negative forming), instead directly machining semi-finished pieces of metal. Secondly, because forming is highly localized, the force required does not increase with scale, meaning that there is no theoretical limit to formed sheet size. Lastly, ISF extends the formability of metals beyond conventional methods, such as stamping or deep drawing Drivers of new research in this field include the exploration of larger scale applications, typically in the automotive and aerospace industries, and improving forming accuracy. The geometry change impacted on the steel sheet is achieved through a local tensile or biaxial stretching of the metal, and is dependent upon a connection between geometric considerations, processing parameters and material properties. As it is stretched the metal undergoes strain hardening or cold working, which increases its strength locally through the accumulation of plastic deformation. This metallurgical trans formation attends geometric change, as the sectional thickness of the sheet diminishes relative to stretching. In the context of a lightweight skin, these changes are not insignificant. For example, in Stressed Skins - which uses low carbon mild steel formed at room

02 – THE ARCHITECTURAL RELEVANCE of ISF // Transferred into architecture, ISF graduates from a prototyping to a production technology that supports mass customization. As has been noted, potential architectural applications include for example folded plate thin metal sheet structures. We have further identified an application for ISF in customized, load-adapted architectural designs. Architects use thin metal sheets as cladding panels to provide integrated enclosure, structure and form. As loads vary in building system, so do performance requirements, so that the customization of elements becomes a key concern. Using ISF on pre-cut metal cladding panels to add features that locally stiffen the panel (in the locations and to the extents needed) can significantly increase efficiencies of material use and reductions for supporting structural systems. 03 – ISF for STRESSED-SKIN STRUCTURES // In this research, the ISF process is used to fabricate an architectural stressed skin structure. Such systems are typically a hybrid assembly in which a thin skin is structurally active, bearing both planar and shear forces and providing significant rigidity by continuously wrapping an underlying, compressive frame. They are an intermediate between monocoque and rigid frame assemblies,

and have been particularly associated with the early application of metals in lightweight structures. In their design, rigidity is a central concern at multiple scales: rigidity against instability in the whole structure, against local buckling of the components that carry compressive load, and against micro buckling or ‘wrinkling’. The research Stressed Skin develops a structural approach in which the skin carries planar and shear forces, but without the use of an additional framing system, at the scale of a pavilion. Research at RWTH Aachen has established ISF as structurally feasible at this scale, in the case of formed panels spanning between a hexagonal continuous framing structure. Recent research explores doubly curved sheet metal panels for free-form metal skins and self-supporting structures, which utilize cone geometries as means to reach from one skin to another. These have been developed to prototype scale. Stressed Skins is designed as an asymmetric tunnel, which cantilevers at one end. The structure consists of 186 unique planar, pentagonal panels. These are arranged into an inner and outer skin. The framing system for typical stressed skin assemblies is replaced by the introduction of geometric features for resisting local buckling and structural connections, both continuous within each skin and for managing shear across inner and outer skins. These are produced through the custom robotic ISF of individual panels.

The Nature of Robots

temperature-sectional thickness reduces in places from 0.5 to 0.15 mm, and strengths increase from 220 to 410 MPa.


The Nature of Robots

04 - PROTOTYPE, MODELLING and FABRICATION: INTEGRATED APPROACH // stressed skins is developed through multiple iterations of physical prototypes and computational models, and integrates observations from physical prototypes towards the digital environment to addresses the multiscale nature of the forming process and the structural assembly. In this approach, different computational models, specific to particular scales of parameterization, behavior and decision making, are made critically interdependent upon one another. A number of considerations are addressed here:


04.1 - COMPUTATION // The digital modelling of stressed skins is informed by parameters and limits derived directly from these physical prototypes. Three modeling scales-macro, meso and micro-are considered to be markers along a structural continuum. In general, the macro scale refers to overall geometric configurations and predictions of its structural performance. The meso scale considers the level of the panel and its detailing, and implements geometric transformations related to connectivity, stiffening, and component-level tectonic expression. Finally, the micro scale relates to the calculation of material implications at the most discrete level, which includes the thinning and hardening of the steel sheet that results from forming. The modeling does not include the actual simulation of the ISF fabrication process, but only the expected material transformations in-

troduced through it. The multi-scale modelling approach is thus comprised of multiple techniques that enable the information generated at each scale to flow both up and down the continuum. Here, an adaptive mesh refinement method is used to support localized variations in resolution and information flow. From a perspective of design development, these include the overall form-finding and panelization operations; global structural analysis and adaptive specification of connectivity arrays; and recursive local tectonic pattern formation which depends upon finite element analyses and is further informed through the calculation of forming strains and material thinning. The features of a half edge mesh-its vertices, half - edges and faces - are coupled with a series of lists, dictionaries and Grasshopper data trees that effectively bundle within mesh elements critical design data related to: topology, form-finding and geometry; structural behavior; material characteristics; connection detailing; and patterning and tectonic expression. 04.2 – PHYSICAL CONSIDERATIONS // Physical prototypes have played a key role in determining the parameters and limits of mass customization. Parameters that informed the digital modelling include individual panel constraints related to size, orientation, and formable territory; the development of connection and assembly strategies; change in material properties; and forming limits in regards to both feature geometry

04.3 – ISF SETUP // An ABB IRB 140 six-axis multipurpose industrial robot is used to fabricate Stressed Skins. The wall-mounted robot arm is situated above a flat table, which bears a clamped, re-orientable MDF jig. MDF dies are fixed to this jig and where necessary are supported from below using a collection of standard elements. The dies are laser-cut templates that define the outlines of desired formed geometries, where those geometries cut the plane, and provide resistance for the steel sheets in areas intended to remain planar. Steel sheet blanks are fixed to the dies along their edges with bolted MDF blocks. 04.4 - POSITION OF FEATURES // Through empirical testing, the capacity of the ABB IRB 140 to exert a downward force was established for the working area. Because of this varying strength capacity, the position of features proved to have an impact on the forming accuracy, and in some cases resulted in the robot’s failure to apply sufficient force to form the steel. To counter this problem, which is to a large extent linked to the size

and specification of the robot system, 6 different forming positions were defined. An analysis of target locations per feature enabled toolpath distribution across locations. As a result, forming was concentrated in areas of maximum strength. 04.5 – PROGUCTION of SAMPLES // Samples were prepared using two alternative forming methods. The first of these was a pneumatic hammering tool, where force was generated via the stroke of the tooltip. The second method, a single-point pressing approach that utilizes the robot’s capacity to impart force, was understood and developed based on access to an ISF-designated CNC setup at DTU Mekanik. Several processing parameters were varied in a systematic fashion across five samples-the forming method, the tool speed and the wall angle. Both forming methods were tested in order to understand their relation to strain rate. The relation between tool speed and wall angle was tested to understand how these parameters affected formability. The speeds varied between 10-65 mm/s, and the wall angle was tested at 15, 35 and 50 degrees. 04.6 – TESTING of SAMPLES // Within the range of speeds and angles tested, all samples were successfully formed. Visual monitoring of the grains and measurement of thickness at the same points was achieved using optical microscopy at 5 points on the cross sectional thickness of each sample. Their local increase in

The Nature of Robots

and tooling time. Samples were prepared to systematically vary multiple processing parameters: tool type (either hammer or point), tool movement speed, feature angle, and to measure strain hardening and thinning. To integrate factors inherent to the fabrication setup, all samples are produced using the same rig used for final production.


The Nature of Robots

strength was monitored using Vickers Hardness tests with a 5 kg load measured at 50 points on the cross sectional thickness of the sample. The resulting hardness were converted to estimated flow stresses and correlated with the local strains. Flow stress is the yield strength of the metal as a function of strain, and describes the point at which the material enters plastic deformation.


04.7 – RESUTS // The hammering technique imparts plastic deformation through the rapid sequential impacting of a tool against the surface. This involves a much higher speed of tool motion and therefore produces larger strain rates and leads to a greater straining of the sheet but not, in the samples tested, to higher flow stresses. In this respect, hammering does not offer obvious advantages compared to the pressing technique. Furthermore, according to our observations hammering was associated with greater springback, and we were able to better achieve the target form through the pressing technique. Based on these observations as well as results from the results of the Vickers hardness tests, the pressing mode of forming was identified as the fabrication method of interest. It was found that in the range of 30-60 mm/s the tooling speed had a large impact in both the surface quality and formability of the steel sheets and that the amount of impact had a direct connection to the wall angle. However there was no observed impact on the material properties and that at this range,

the impact of tool speed is negligible to resultant strain hardening. 05.1 - SYNTHESIS of COMPUTATIONAL and PHYSICAL CONSIDERATIONS IN THE EXAMPLE of a PANEL // The panel arrays for both layers of stressed skins were developed through a stepwise accumulation of panels onto two respective target surfaces according to a pentagonal planar tiling strategy. The targets were derived as being variably offset from a baseline surface that was developed both to accommodate occupancy requirements on the site and to generate suitability challenging structural performance demands through its spanning and cantilevering. The variable offset was calculated based on an initial shell FEA (Finite Element analysis), with greater offset -and structural depths-assigned to areas of high utilization. A constraint-based dynamic form finding system -a beta scripting library of the Grasshopper plug - in Kangaroo2 - was used to adapt and planarize each panel as it adhered to its respective offset target design surface. Following this initial panel organization, a series of connection cones were solved between the two skins. These cones form the primary basis for managing the structural shear requirements, taking on much of the role played by the compressive frame in traditional stressed skin structures. Here they were located to maximize diverse connectivity across multiple panels between the upper and lower skins, and oriented in response to a

05.2 – INFORMED PATTERNING and LOCAL FEA SIMULATION // The prior understanding of the relationship between geometric forming and the consequential material hardening was then integrated with these connection point translation and rotation vectors into an iterative feedback design and analysis cycle for the purpose of locally introducing performance improvements into individual panels specifically to resist in-panel bending forces. This was achieved through the variable-depth forming of a pattern onto each panel, integrated with the base inter-skin connection cones that provide primary structural depth and accommodate the transfer of shear forces within the assembly. The pattern was first generated as a flat, graphic element over all panels on the mesh. An implementation of the Gray-Scott reaction-diffusion algorithm on the design mesh was used to achieve this. Beyond its aesthetic, this algorithm was selected for a two key reasons: its generally isotropic nature enabled resistance to bending in multiple directions, and its form could reliably be cut into the MDF dies used during forming. This baseline pattern worked as a scaffold to receive addi-

tional depth. This additional depth was realized through an iterative process. First, each panel begins as flat in all areas except for the features used for both inter and intra skin panel connections. Over each panel sub-mesh, an adaptive quad mesh is arrayed and inscribed ellipses used to determine the local strain introduced in the forming of these baseline features. Equations are deployed to locally differentiate yield-strength material settings for each face in the primary mesh, and an FEA is executed for the individual panel, using the translation and rotation force vectors derived from the global FEA. Resulting utilizations are extracted. Here, note that areas within each panel that have been hardened due to forming (as in the deep connection cones) tend to have significantly higher strengths, and therefore lower utilizations. High utilization areas then drive the local introduction of incremental depth, which is here visualized as incremental changes from baseline features, with black being zero change. This process of transforming material settings, applying connection nodal translations, and adding local pattern depth is then iterated up to fifteen times per panel, resulting in a steady decrease in utilizations for each panel due to strain hardening, and greater bending energy due to geometric stiffening resulting from added depth where it is useful. Finally, each panel is subdivided to a finer level of resolution, and initial contours are extracted for toolpath generation.

The Nature of Robots

second FEA that identified shear force vector lengths and directions. A third FEA was performed following this precise locating of the connection cones, and translation and rotation nodal displacements were extracted from the model at all connection points between panels, both within and across skins.


The Nature of Robots 66

05.3 – INFORMED PATTERNING and LOCAL FEA SIMULATION // An algorithm was developed to derive tool paths from the model geometry, based on contours cut from a design mesh. The progression of the algorithm was: 1. Grouping of features // 2. Position of features // 3. Tooting speed in relation to wall angle// This order was developed to firstly calculate the entire toolpath, then divide it into modules, and lastly to add information about speed to the target positions. To improve control over tooling time and surface quality, the toolpathing algorithm is developed based on the creation of spirals, an established approach for ISF. The grouping and position of features, and the tooling speed in relation to wall angle are discussed in the following: 1. Grouping of Features // with the basis being horizontal contours, a strategy for grouping the section curves was developed. By checking the section curves for inclusion in the domains of the previous layer, groups of curves were created in order to make multiple continuous spiraling toolpaths that allowed forming of complex geometry, but ensured no metal was being formed unconsciously or tried forming twice. The figure shows an example of a color coded grouping, revealing the complexity in both geometry and toolpath. 2. Position of Features // To ensure features being formed in optimal position in relation to the strength of the robot, each spiraling toolpath was divided into 1000 points. Each point was

checked for inclusion in predefined areas. The toolpath was then transformed into the area that recorded the highest percentage of inclusion. 3. Tooling speed in Relation to wall Angle // The production schedule for Stressed Skins required working at the maximum permissible speed, with consideration to formability and surface quality. Building on the knowledge obtained from the tested samples, further exploration showed that if the wall angle did not exceed 45 degrees a speed of 65 mm/s could be used, but as wall angle got higher the speed needed to be lower to ensure both surface quality and formability. with our setup we reached a limit of a 60 degrees wall angle which could be achieved with a tool speed of 30 mm/s. Based on this testing, a linear relation between angle and tool speed was used to set a unique tool speed for each target point along a toolpath: S[mm/s] = -2.33 * A [ ° ] 06 – ASSEMBLY and EVALUATION // The research extended towards a research project, with the fabrication of 187 panels for structural assembly. The structure is characterized by a high degree of connectivity for successful assembly, and thus relied upon accurate forming and low tolerances. Geometric accuracy has been a key concern regarding ISF since its inception, with typical geometric tolerances of more than 3 mm within a part. Though traditional architectural practice accepts such tolerances, in the case of Stressed Skins these needed to be tightly managed and not

06 – RESULTS and CONCLUSION // This research has discussed the modeling and fabrication of an incrementally formed, stressed skin architectural structure. A robotic ISF process has been used to increase rigidity through geometries within a surface of thin steel panels, and through connections between those panels. ISF possesses significant architectural potential in the area of mass customization, but as has been discussed, this requires a tight coupling between fabrication process, material properties and the digital design model.This research paper aimed to contribute to ongoing research in the robotic fabrication of single elements, towards highly inte-

grated structural assemblies. In doing so, it aimed to extend the scope of architectural applications by developing a highly integrated structural assembly and an associated digital modeling method.The resarch discussed a modeling method that is informed by fabrication parameters and material properties, which are established through prototyping and empirical testing. This method incorporates the specifics of a fabrication environment, and integrates empirically derived material hardening and thinning data. Empirical data have been used to define parameters of macro and meso finite element simulations with calculated micro behavior; in order to set and extract toolpath information; and to inform ring setup logics for fabrication. The high level of integration between modeling and prototyping enabled the simulation to incorporate a level of information not typical within architectural modeling, and a fabrication process where the relationship between tool speed and wall angle was optimized. The successful assembly of the panels, some of which support up to ten unique connections, demonstrates that incrementally formed frameless structural assemblies can be made available at an architectural scale. ACKNOWLEDGEMENTS // This re-

search was undertaken as part of the Sapere Aude Advanced Grant research project ‘Complex Modelling’, supported by The Danish Council for Independent Research (DFF). The authors would like to acknowledge the collaboration of Bollinger Grohmann consulting engineers, Daniel Piker and Will Pearson, the research departments DTU Mekanik and Monash Materials Science and Engineering.

The Nature of Robots

extrapolated over the entire structure. The research thus evaluated the built structure for its performance. A Faro Focus 3D 120 scanner was used as a means to measure geometric accuracy and structural performance. At the scale of panel, forming accuracies were measured to have a 2 mm standard deviation. This was accounted for via the inclusion of a 4 mm spacer at connection points between inner and outer skin. After assembly and a setting period of one month, the structure was scanned and evaluated. The maximum deviation to the geometry predicted using the finite element model was 10 cm, recorded at the extremity of the cantilever, which illustrates the contours color-coded to distinguish unique groupings of curves for creating individual spirals, enabling the translation of complex geometries to discrete, sortable toolpaths.


ROBOTIC LATTICE SMOCK A Method for Transposing Pliable Textile Smocking Techniques Through Robotic Curved Folding and Bending of Sheet Metal.

Smock (RLS) presents a method for transposing pliable fabric folding techniques of smocking to an architectural scale through robotic bending and folding of rigid planar sheet metal. Building on the limitations of three-axis CAD/CAM fabrication techniques for unfolding and cutting planar pieces, RLS explores the process of six-axis robotic curved folding and bending to ‘gather’ or “smock” planar developable surfaces to overcome brute force assembly, build volume through more efficient material use of planar sheet material and generate novel material aesthetics through the hard constraint of disciplined material transposition. KEYWORDS // Robotic fabrication // Smocking // Curved folding // Textiles // Gottfried Semper // Transposition // Physical computing // Developable surface // Torsal ruled surface // Rulings // Aesthetics//

Credits // Andrew Saunders //

The Nature of Robots

Gregory Epps //


ABSTRACT // Architect Gottfried Semper built a discourse on architectural aesthetics based on his belief that textiles were the mother of all arts, and the initial motivation for all architectural form. Inherent in this evolutionary premise is the concept that cultural development begins with pliable and easily manipulated materials, and can be extended and transformed through technological advances for crafting more robust and permanent materials. As a contemporary projection of this framework, Robotic Lattice

01 – INTRODUCTION: SEMPER, TEXTILES and TRANSPOSITION // The nineteenth-century German architect and historian asserted that textiles are the mother of all arts, influencing every branch of the technical arts, thus the origins of all basic architectural forms. His lexicon ‘Der Stil’ (transl. ‘Style’ in the Technical and Tectonic Arts; or, ‘Practical Aesthetics’) methodically traces the influence of textile motifs on various forms and manufacturing procedures for more permeant material including ceramics, tectonics (carpentry) and stereotomy. The motivation was twofold: 1. To prove polychromatic ornamen-

02 – LATTICE SMOCKING // It is interesting to note the etymological game that Semper plays when choosing architectural terminology. ln ‘Die vier Elemente der Baukunst’ (transl. The Four Elements of Architecture) he uses the term die Wand for enclosure or wall, a word that is strikingly similar to Gewand and Winden (German for ‘dress’ and ‘embroidery’).

Building on this, we selected the technique of smocking as one of the most basic forms of embroidery to transpose. Lattice smocking was further chosen for the project due to its flexibility, deep relief and variation of motifs, as is discussed in the following: 1. Flexibility. The base grid for the lattice smock pattern can remain regular or accommodate increase and decrease in the size of the grid modules, supporting topological variation in the stich pattern. In anticipation of a facade system, panels following this logic would be easily adapted to a grid or diagrid substructure. 2. Deep relief. The relatively low density of stitches allows for more fabric to be gathered between stiches producing deep volumetric folds with recognizable signatures. In addition, the longer distance between stitches accepts much thicker fabric. In anticipation of folding and bending thin gage metal, thick felt was chosen for smocking to produce larger and more consistent folding radii. 3. Variation. Very basic grid stitch patterns of crenulations, zigzags and branching produce amazingly complex folded formations in the fabric when gathered. The resulting manifolds include patterns identified as lattice, lozenge, flower and arrow. An arrow pattern was chosen due to its triangular relief. The three points integrated with the robotic lab environment that included a station with two 6-axis ABB robotic arms and one stationary arm.

The Nature of Robots

tal covering of Greek architecture through the persistence of certain symbolic motifs carried over from nomadic textiles; 2. To critique the cheap 19th century industrial simulation of one material by another, specifically casting, stamping, and molding. For Semper, these fabrication processes were paradoxically indifferent to the symbolic continuity essential to the recreation of tectonic form. Semper’s criticism finds new relevance as the current integration of industrial robotic arms in bespoke design and fabrication gives rise to new methods for challenging industrial standardization and construction processes at the architectural scale. The expanded scope of six-axis movement offers increased agility to work material to the full extent of its expressive scope As a contemporary projection of Semper’s tectonic framework, Robotic Lattice Smock (RLS) explores the expressive qualities of folded sheet metal through the transposition of pliable textile smocking techniques to robotic folding routines.


The Nature of Robots 70

03 – SURFACE RATIONALIZATION: GEOMETRY // RLS offered an opportunity to deploy six-axis robotic fabrication and physical computation to build on the research of Sartorial Tectonics that similarly addressed tectonic relationships of textile analogs and three-axis computer aided manufacturing. Although the previous research was successful at transposing affects of fabric manipulation to more rigid materials from computational aided manufacturing, the complexity of the developable surfaces were limited to tangential surfaces and the re-rolling of unrolled surface components was entirely by manual brute force. The metal folding and bending process requires very specific surface rationalization. Because both processes begin with flat material and are manipulated to build volume and pattern, they are subjected to the rules of developable surfaces. Developable surfaces are a subset of ruled surfaces with zero Gaussian curvature, able to be unrolled onto a flat plane. A ruled surface is a surface generated by a straight line moving along a curve. The straight lines that make up the surface are rulings. If the ruling has different tangent planes at each end point, it is scrolar and the points on the scrolar ruling are hyperbolic (non-developable). If the ruling is touched by one unique tangent plane, the ruling is torsal and the points on a torsal ruling are parabolic (developable). For RLS to be robotically folded, all rulings on the surface needed to be torsal guaranteeing the surfaces were developable. As a developable sur-

face, all governing profile lines could be unrolled for cutting and scoring in planar material. 04 – METHODOLOGY // The RLS facade system prototype is fabricated from flat thin gauge 1.5 mm aluminum panels with milled profiles and curved fold line scores. These panels are then placed flat on a workstation with two six-axis ABB IRB6400 industrial robotic arms and one stationary arm. All three are equipped with vacuum end effectors to grip the panel. The end-to-end process outlined in the following involves smocking felt and developing unfolded patterns for the gathered smock manifold, constructing physical mockups in cardboard, paper and at 1:1 in hand-folded aluminum, conducting physical and digital simulation to obtain three-dimensional folding and bending paths and pinning bent panels at the fully gathered position before release from robotic arms. The initial research had the following objectives: 1. To explore the novel expressive aesthetic qualities generated by transposing textile smocking to robotic curved folding and bending; 2. To explore the manipulation of flat fabric as an analog model for robotic folding and bending of planar sheet metal; 3. To examine the explicit feedback loop between physical and digital simulation of folding developable surfaces through robotic manufacturing; 4. To explore lattice smocking as a generative process for the creation

04.1 – LATTICE SMOCKING FELT // As a departure point, the transfer of fabric to harder substrates was investigated in order to improve the process of folding paper, which has a limited amount of changes that can be made before it deteriorates. Building on this process, RLS began by lattice smocking an arrow pattern with thread, needle and a (40.64 cm by 22.86 cm) piece of felt. Felt was chosen due to its heavy density, and also the defined and semi-regular fold lines and curved creases that emerge when the textile is gathered. The resulting felt arrow smocks were carefully analyzed and distinct curved crease lines were identified and traced on the felt in its gathered state. When the pulled stitches were released, the felt returned to a flat position revealing new intricate curved crease patterns on the original arrow lattice smock grid pattern of pulled stitches. Mountain and valley coloring of red and blue curves are used during the curve acquisition as per the origami conventions, to indicate a positive (mountain) or negative (valley) fold. 04.2 – FOLDED // The new flattened patterns of curved crease lines were scanned in 2D to create digital tem-

plates. Flat heavyweight paper mockups panels were scored and cut using a die cast paper cutting machine. The paper mock-up panels were folded to form more rigid versions of the arrow smock. The paper mock-ups were used as analog models for the metal folding process. They were helpful to examine bending tension and rationalization of the surface geometry. Due to the stiffness of the heavyweight paper, the resultant surface embodied the developable properties of a torsal ruled surface. As the folding was enacted, regions between the curved creases revealed composition of partial planes, cylinders, cones or tangent surfaces. The points where the gathering stitches were previously became points of tight radii cones and were the points in the paper panel with the most intense bending tension when folded. The greater the degree of curvature, the greater degree of surface tension when folded. There are two options for relieving the bending pressure: decrease in curvature of folding crease or removal of material closest to where the ruling lines converge. By using the points of the cones as center points to construct arc cut lines on the template, additional material was removed enabling the new paper panel to fully fold mimicking the original gathered position in the felt smock. Once a developable surface version of the lattice smock panel had been established, the template was calibrated to accommodate specific dimensional constraints of the robotic fabrication environment. Key criteria in adjusting

The Nature of Robots

of complex torsal ruled surfaces (rationalized through partial planes, cylinders, cones ortangent surfaces) to be robotically folded; 5. To examine robotic folding as a method to automate curved folding and bending of metal panels to counter brute force assembly tactics.


The Nature of Robots

the template were the identification of planar regions for vacuum end effectors attachment; maximum bending angles; and space for manual pinning of a fully bent metal panel. With final adjustments being made to the local module of one “arrow” fold of the smock pattern, focus returned to the global configuration. Additional paper mock-ups were produced to study how modules would nest as an assembly and how edge conditions of the field would terminate. It was important that the design embodied a flexible rule set to accommodate edge conditions as terminus. A “sole” plate was developed to provide stability or the module. The “sole” plate is a flat profile traced from the gathered position of the panel edges. The sole plate panel was attached after the robotic arms bent the panel to the fully gathered position, keeping the panel from returning to the maximum fold angle (which is less than the maximum bending angle). In addition, the bottom edges and sole plate extend further at any edge condition to create a fringe condition terminating the continuity of the lattice smock pattern.


04.3 – PHYSICAL SIMULATION of FOLDING AND BENDING METAL // The final physical mock-up phase, included cutting and scoring of aluminum panels, conducting manual folding and bending tests. Through the process of manually bending each metal mock-up, sharpness of curved fold creases were finessed and termination points of fold scores were adjusted to calibrate the

hardness and fading of creases within the metal panel. To simulate the fully bent form of the panels, zip-ties were used to incrementally “gather” the panel under tension into the final position of the arrow smock. The final adjusted aluminum mockup was analyzed by sweeping a straight edge over the surface to identify and trace torsal rulings between curved creases. The locations of all major rulings were scanned and added back to re-inform the digital template. Together with the major curved fold creases, they would form the critical network of components necessary to construct the digital simulation model. 04.4 – DIGITAL SIMULATION of FOLDING and BENDING // To accurately model and simulate the transition from flat planar material to final folded form without distortion in the digital environment, the King Kong plug-in For Rhino Grasshopper that integrates a live physics engine was used. The curved creases and torsal rulings traced from the physical models were defined as a network of springs and hinge forces. Similar to the parameters for simulating rigid origami, curved creases are defined as hinge forces swinging in one of two directions, mountain or valley. Rulings are defining as springs with rest length equal to actual length to eliminate distortion through the folding process. After establishing a digital simulation of the folding and bending behavior, the positions of each face of the model were tracked. The process

04.5 – DIGITAL SIMULATION of FOLDING and BENDING // The final phase of transposing smocking techniques to curved folding and bending sheet metal deployed robotic fabrication. The robotic arm routines were checked self-intersection, for reach and rotation limits in the IO plug-in. Once the code was verified as safe, the software allowed it to be uploaded and enacted. The final fabrication of six RLS panels involved manually placing a precut and pre-scored flat panel in a known position on the workstation that corresponded with the simulation. After the robotic arms had gripped, folded and bent the piece into position, the sole plate was pinned manually to prohibit the panel from returning to the folded position and the arms released the piece. The robot was programmed to pick up, move into position, fold, and return to home position, with each distinct stage indicated in a named section of the timeline component in the IO plug-in. This level of clarity allowed for fast iterations through modification or basic robot parameters such as speed. The timeline features two robots, which are synchronized using an algorithm to determine equal time gaps between each individual plane. Two entire sequences are the same length of time,

but follow a complex curve as determined by the non-linear folding animation (simply matching the overall time will not suffice). Various aspects of the end-effectors such as the angle and position of the vacuum grippers are adjusted through a series of sliders in Grasshopper, in order to fine tune the location that they are imparting force on the metal surface. This adjustment is used to determine the best location to achieve a good result-this process occurs through some know-how and trial and error. Very little room for error was allowed, with only a valid solution occurring in space vertical region of 20 cm and within a rotation of 30 degrees. As an integral plug-in of Grasshopper (McNeel), RoboFold enables control over different sections of the code: Mesh Import, Robot and Tool adjustment, simulation and control and outputs including: Code, Synchronization Codes, Fold Preview, Axis Angles, Camera Animation, Render Mesh Baking, or Simulation Timing. 05 – ANALYSIS // In the case of bending metal, where the resulting surface must be developable, the smocking analog presents a flexible method for developing torsal ruled surfaces, and can thus become useful for guiding robotic fabrication of more rigid material at the architectural scale. The act of transposition champions qualities of each material (or medium) and fabrication processes of each specific material. Achieving the deep undercut relief and supple intricacy of folding in fabric has presented the largest chal-

The Nature of Robots

was critical in identifying the exact three-dimensional path of the faces dedicated for vacuum gripper contact. Once defined, the digital simulation was used to choreograph the motion of two six-axis robotic arms.


The Nature of Robots

lenge. The first stage of this research successfully examined the local scale of individual panels, whereby six identical arrow smocked panels were robotically fabricated. Curved creases with high curvature and bending against the embodied energy of the bent metal produce recognizable thresholds and limitations in the process. The extreme pressure tested the strength of the vacuum gripper end effectors and created dangerous conditions for the manual pinning of gathered position. This could be overcome by introducing more secure gripping mechanisms or bending thinner gauge metal (or other planar material). Ideally, the entire process, even the pinning of bent panel would be automated robotically.


05 – CONCLUSION // RLS presents a method for transposing textile manipulation techniques into more rigid material at the architectural scale through robotic curved folding and bending of planar sheet metal. By using the process of lattice smocking, flexible textiles serve as an analog model for producing deep relief and complex curvature from a flat sheet material. Torsal ruled surface properties are intrinsic in the surface manipulation from the initial starting point and lead to a fluid transposition into a rationalized surface for robotic simulation and fabrication. In terms of future research, one area of further investigation would be topological variations of the arrow smock panel, and the integration into a seamless fabrication workflow. In addition, the

flexibility of the lattice or grid will be examined, first in two dimensions with warped grids, and then in three dimensions as a diagrid with the capability of accommodating predefined complex forms. As Semper discussed, this process then enables a transfer of cultural knowledge from one medium to another. The process of transposition champions matter and the investment of techniques and rules unique to each medium. The resulting design is not evaluated on whether or not it ‘looks’ like, instead on whether the metal ‘behaves’ like the lattice smock. In the direct process of transposition it gains unique aesthetic expression as the embodiment of particular geometric and physical signatures of the material and process of fabrication. ACKNOWLEDGMENTS

// Robotic Lattice Smock is a collaboration of Andrew Saunders and RoboFold Ltd. sponsored by the Rensselaer Roberl S. Brown’s 52 Fellows Program. Design Team: Andrew Saunders, Sahar Mihandoust, Guo Huanyu, Jessica Collier, Elizabeth Sammartino, Matthew Vogel. RoboFold Team: Gregory Epps, Ema Epps, Florent Michel, Jeg Dudley.

Credits // Philip F. Yuan // Hao Meng // Lei Yu // Liming Zhang// ABSTRACT // The research discusses a robotic multi-dimensional printing design methodology based on a material’s structural performance. Through research on the process of a spider’s behavior, e.g., spinning and weaving, the designers simulate natural construction principles and apply them to the optimization of traditional 3D printing techniques. A 6-axis robot is programmed to carry a customized printing end effector to create freestanding geometries in space. The structural behavior of the design is optimized through the consistent negotiation between material analysis and structural simulation in both virtual and physical environment, together with the implementation of sensor input and real-time feedback

between construction tools and simulation interfaces. The printing tools are designed with additional extruder sand nozzles of various dimensions to adapt to different materials and design requirements. In this way, a flexible and adaptive additive manufacturing methodology is established, which integrates the material and structural information with design initiatives. Displaying a high degree of spatial and structural complexity, the alliance between 3D printing and robotic technology opens new possibilities to sophisticated architectural structures. KEYWORDS // Multi-dimensional printing // Robotic fabrication // Structural performance // Material performance // Tool development // 01 - INTRODUCTION // 01.1 – FROM 3D PRINTING to MULTY-DIMENSIONAL PRINTING // 3D printing has been developing or decades. So far, there are two types of commonly used and representative printing technologies, namely FDM (Fused Deposition Modeling) and SLS (Selective Laser Sintering). FDM melts printing material and extrudes melted material through fine printing nozzles for deposition modeling while SLS provides selective sintering of powdered 3D printing material for modeling and remove the un-sintered part. These two representative 3D printing technologies have shown certain limitations in actual application. On one hand, both of these two 3D printing technologies are rapid

The Nature of Robots



The Nature of Robots 76

prototyping with given materials. The printing process is a simple formation process and excludes inherent structural logic of the product. Such contour printing method also has certain limitations in printing capability, especially in FDM. On the other hand, due to the limitation of the printing method (material deposition modeling), the products printed using such technology will inevitably sacrifice the structural strength at the interface between different layers. Hence the strength of the material cannot be sufficiently exerted. Although industrial standard SLS 3D printers with high precision can perform printing with extreme strength, the cost is very high. In this context, the multi-dimensional printing concept is introduced to overcome the limitations of traditional 3D printing. This has two advantages. The first is higher printing freedom in space. Although the current 3D printing technology is called three dimensional, it only refers to the three dimensional volume of end results. In terms of formation technology, the current 3D printing mostly adopts a planer contour stacking technique. Multi-dimensional printing is expected to realize 3D modeling in space through multiple axes just like a person using their hands. The second advantage is the ability to integrate various performative optimizations within design and the fabrication process. As mentioned before, the material deposition modeling will cause structural defects in printed products. The flexible printing strategies and the constant data feedback between the design and

fabrication platform improve the structural rigidity of the material. 01.2 – MULTY-DIMENSIONAL PRINTING with INTEGRATION of STRUCTURAL and MATERIAL PERFORMANCE // Existing studies on multi-dimensional printing are mainly focused on the following two aspects, namely, the integration of a structural logic, and the improvement of strength. The integration of a structural logic with the printing process has been explored illustrated in the Mesh Mould project from the team of Gramazio & Kohler, ETH. In this research project, the geometry to be printed was first transformed to a spatial truss, which is an optimized structural skeleton of the design. The designer built a rapid cooling device inside the traditional 3D printing extruder to ensure instant cooling and solidification of extruded printing material. Though the material used in this project is a traditional 3D printing material, the printed geometry has specific self-support abilities to keep its form stable after the material being cooled down. Thus, a spatial structure can be steadily built up following the extruder’s movement along the preset grid pattern. The weaving Printing project conducted by Digital Design Research Center at Tongji University carried similar research in testing more sophisticated geometry and spatial frame with transitional grid density. This type of research indicates the possibility of rapidly producing large structural prototypes and architectur-

01.3 FROM MULTY-DIMENSIONAL PRINTING to ARCHITECTURAL 3D PRINTING // In recent years, architectural 3D printing has become an avenue for novel research. Several experimental 3D printed architectures have been produced, such as the 3D printed residential house of Yingchuang Technology Company, Shanghai. Integrated structural reinforcement has been applied innovatively in the printing process of the residential products, e.g. in the wall section of the building, slant braces similar to a truss structure are printed to ensure the strength of the wall. Yet there exists a significant difference between the building industry and current 3D printing industry: traditional 3D printing is a simple model making process, whereas in building construction materials, structure, and construction processes need to be considered. As

a result, traditional 3D printing technology is far from meeting the requirements of building construction when the building is expected to be 3D printed. Therefore, only through integration of fabrication techniques in traditional 3D printing industry may the technology be used in building construction field. The multi-dimensional printing concept is consistent with the demand for developing large scale 3d printing architecture. The study of multi-dimensional printing will have decisive significance in architectural 3D printing research. Albeit a good example of multi-dimensional printing, in the project MATAERIAL the multi-dimensional printing is only realized by enhancing the physical strength of print material, i.e., the structural logic of the built geometry is not a primary concern in the printing process. This paper introduces a new research that tries to incorporate structural behavior of the material into the printing process and employs a customized printing end effector with greater freedom and precision through one 6-axis robot arm to realize multi-dimensional printing of complex geometries. This method makes it possible to create flexible and versatile nonstandard space structures with no formwork or additional support, demonstrating a sustainable construction strategy that is materially efficient and cost effective.

The Nature of Robots

al moulds for building components with complicated geometry. Secondly, research has been undertaken for improving the strength of printing materials. In the project MATAERIAL by Advanced Institute for Advanced Architecture of Catalonia (IAAC) and Joris Laarman Studio, the research team developed a printing material that can solidify rapidly. This material has excellent strength after solidification. Thus, it can ensure accurate shaping and self-support ability of the geometry. The significance of this research project is the realization of free curvilinear printing through material engineering.


The Nature of Robots



02.1 – BIONIC DESIGN INSPIRED by SPIDER SILK // One of the most widely used 3D printing technology, FDM (a Linear deposition o melted material), is the basis for this project. Therefore, if the linear structure can be strengthened during the printing process to self-support its own weight, multi-dimensional printing can be achieved with various materials. Researchers have looked to spider webs for answers, however, the structural rigidity of a cobweb is its resistance to tension while the printing structure in our case is mainly subjected to self-weight and bending moment. We were also inspired by the sectional morphology of the spider silk, data shows that the sectional Diameter of the spider silk is about 1/1000 mm, but its strength is four times stronger than that of steel in the same diameter. In terms of geometry, a spider web consists of two parts, namely a spindle-knot and a joint. A spindle-knot surface has a stretched porous structure while the joint surface is made of a random porous structure. These two different structures join together to guarantee the structure performance of the cobweb. The sectional variation from spindle-knot to alternate joint brings insight to self-standing linear structures. The bending moment of 3D printed geometry can be overcome by reproducing different sections like spider silk based on a structural reinforcement.

Integrating the geometric morphology of the spider silk into 3D printing process is the key to the realization of spatial printing. We have developed a new printing profile by adding multiple wavy auxiliary curves to the main curve. The area where auxiliary curves are in contact with the main curve forms joint, while the place where the supplementary structures deviate from the main structure forms spindle-knot. Therefore, the combination of primary structure and secondary auxiliary structure inherits the morphology from the spider silk, and achieves greater structural performance in resisting bending force. 02.2 – SECTION OPTIMIZATION // In order to achieve convincing spatial prototypes, the number of auxiliary curves that integrate with the main curve needed to be further determined. Common sense would suggest that the more auxiliary structures are added, the better self-support ability that the overall geometry can achieve. Excessive auxiliary support, however, may increase dead weight of the structure and thus lead to negative effects. Considering the combination of multiple extruders will inevitably increase the possibility of collisions among the mechanical devices, combinations from zero to four auxiliary curves with different compositional positions have been tested and a total number of seven possibilities are simulated and compared. Five different load scenarios were simulated in the experiments; selfweight load, point load of 200 N ver-

02.2 – MATERIAL PERFORMANCE STUDY // The materials tested here are ABS and PLA. These two 3D printing materials have their own advantages-the main characteristic this research investigates is rapid solidification, which is of critical importance because the time that a material takes to solidify directly relates to whether the profile of wavy auxiliary curves can be accurately printed. The greatest difference between PLA and ABS is that PLA is crystal material, while ABS is non-crystal material. Crystal materials have a fixed melting temperature. When the crystal is heated to a specific temperature, it begins to melt, and the temperature remains stable during the process until the crystal is completely melted. After that, the temperature starts to rise again. The same goes for the solidification process. The heat has to be significantly decreased in a short time for the solidification of PLA material, which is a significant challenge to the cooling device. ABS is a non-crystal material, which does not have a fixed melting point; it is able to gradually solidify with the decrease of temperature. Hence, it can take more cooling time for the material solidification process. Meanwhile, the printing temperature of ABS (over 230 C) is much higher than PLA (200 C) and the solidified temperature of ABS is much easier to

reach. Therefore, the material performance of ABS is superior to that of PLA for this project. Our cooling test verifies this result. In terms of material strength, the flexural behavior of ABS material is stronger than PLA. Under the situation of bending force, PLA material is more fragile and easier to be broken. After comprehensive consideration of various aspects, ABS was chosen as the printing material in this study. 02.4 – MECHANISM SYSTEM // The mechanical system refers to the device controlling the collective motion of all the printing nozzles. The result of section optimization is the combination of one primary structure plus three auxiliary structures. Therefore, one fixed printing nozzle and three flexible nozzles that can move consistently and synchronously are required. There are two options or the design of the mechanical system: one is to use three drivers (e.g. stepper motors) to control the motion of each movable nozzle respectively. The other is to use one driver and a set of the linkage system to monitor the movement of three secondary nozzles together. Since the action of the three movable nozzles is not complicated, but requires relatively high consistency, the second option is more sophisticated but avoids unnecessary deviation caused by desynchronization of three separate drivers. The final mechanical system consists of a central turn plate, three nozzle operators, stepping motors and a gear set. The central turn plate is in hex-

The Nature of Robots

tical force, 40 N*m external torque, point load of 200 N vertical external force of different sectional scale, and 40 N*m external torque load of different sectional scale.


The Nature of Robots

agonal shape with fillet corners. During the printing process, one stepping motor drives the rotation of the central tum plate through a gear set. Because the distance from the central point of the turn plate to the six vertexes and six edges are different, the three angle switches can be controlled to open and close periodically. In this way, one stepping motor could control the three nozzles work synchronously, and collaboratively produce wavy sub structures.


02.5 – ELECTRONIC CONTROL SYSTEM // The electronic control system of the spatial printing end effector is developed based on the structure of a conventional FDM 3D printer. The system consists of an operating interface and a central controller. The core controller includes one main control panel (Arduino Mega 2560), one extension board (RAMPS1.4), five stepper motor drivers (A4988), four of which control the stepping of the motors of the material extruders while the other one controls the stepping motor of the central rotation plate. The operating interface includes one control panel and one display panel. The control panel controls the rotation speed of the central turn plate, the switches of material extruders, central tum plate and cooling unit, and the display panel shows the present temperature and material distribution status. The working process of the electronic control system is as follows: after turning on the printing switch on the control panel, the heating device firstly heats up the four 3D printing noz-

zles to the melting temperature of ABS material. Then the panel controlling material extrusion sends out signal to command printing nozzles to extrude the melted ABS material; while the cooling device cools down and solidify the extruded ABS simultaneously. Meanwhile, the rotation control panel operates the stepping motor to drive the rotation of central tum plate; the rotating hexagon takes the three extra nozzles to move towards and deviate from the central nozzle in a regular manner. A Kuka 6-axis robot arm carries the printing end effector steadily along the pre-designed route. As a result, a spatial geometry with variable cross-sections is produced through the collaboration between robot and customized printing tool. 2.6 Printing Experiments // The designers conducted a series of multi-dimensional printing test with various geometrical inputs, including straight curves with different inclinations and multi-dimensional curvilinear structures with different section profiles. Experimental results have shown that the printing process of this integrated robotic system is very stable. It can realize various self-supported spatial geometries within the accessible area of the 6-axis robot. The printed structure is also proven to be rigid and effective. The research successfully introduces new methods in spatial printing through designing variable cross sectional structures. 2.6 CONCLUSION and FURTHER RESEARH GOAL // While first testing

bination of first and secondary structure. As this research discussed, spatial printing with variable sectional profile is feasible. However, the research should not be limited to curvilinear printing; different spatial structures could be developed based on the curvilinear geometry. With further improvements of the mechanics of the end-effector it will be possible to handle volumetric structure printing. The research indicates new possibilities for robotically printed, innovative structures for architecture and the building industry.

ACKNOWLEDGMENTS // The authors would like to acknowledge project group information. Project Name: Robotic Extrusion (Robotic 6-Axis 3D Printing); Brief Info: 3-week group work of ‘Digital Future’ Shanghai Summer workshop 2014, Shanghai; Design Team: SHI Ji in collaboration with LIU Xun/LUo Ruihua/CUI Yuqi; Instructor: YU Lei (Project Instructor, from Tsinghua)/ Philip. F. YUAN (Workshop Leader, form Tongji)/Panagiotis Michalatos (Software Turorial, from GSD); Photography (Filming) and Editing: SHI Ji.

The Nature of Robots

was successful, there are challenges that will be further investigated by the research. These include material and geometry. Firstly, the material in this study is traditional 3D printing material ABS with very limited strength in resisting bending and shearing forces. Therefore, it cannot be applied into larger scale architectural fabrication since the material itself does not have sufficient structural rigidity compared to building material. Secondly, the printed result of this research is a complex combination of curvilinear geometries. The scale and complexity of the structural component limits the design development in the study of surface or volume printing. Although this research is at preliminary stage, the result has proved its potential. Further research will be conducted in order to apply this structure performance-oriented multi-dimensional printing strategy to architectural fabrication. This will include further research of printing materials, respectively high strength and rapidly solidified materials that would expand the range and scale of design products. As long as the material strength meets the demand in architectural industry, this printing technology could be rapidly used for customized building components, in particular non-linear structural systems. This structure performance-based printing technology could easily realize customized fabrication of complex spatial structure with greater material efficiency and smaller energy consumption. Furthermore, diversified printing strategies will be developed with new com-



structural weaknesses, and load-path optimization to achieve a digitally informed final geometry. KEYWORDS // Robotics // Flexible fabric // Casting // Parametric design // Scripting // Gravity simulation // Nodal connectivity // Mesh relaxation //

The Nature of Robots

Credits // Ron Culver // Julia Koerner // Joseph Sarafian //


ABSTRACT // The novel, robotically-controlled system delineated by this research facilitates a rapid and economical workflow realizing a complex network of parametric geometry. The method of concrete fabrication proposed here removes the traditional limitations of rigid formwork and satisfies the need for variation in the realization of parametric design. Lycra is stretched and positioned by robot arms as a formwork into which concrete is poured. Thus, the flexibility of fabric is translated into flexibility in design permutations. The prototyping considers material constraints,

01 - INTRODUCTION // Traditional rigid formwork has distinct disadvantages for casting complex forms from concrete. Time-intensive computer numerically controlled milling and subsequent form assembly fail to adequately replicate the compound shapes and undercuts required of complex geometries. Moreover, the casting of multiple parts with even slight variation is often cost-prohibitive. The proposed robotic system facilitates a faster, more precise and more economical workflow to realize complex or truncated parametric geometry from unique cast masonry components. Given the shortcomings described above, this project focuses on developing a fabrication technique utilizing motion to create 3D space and components. In this respect, robots are used as precision, time-based tools to generate motion for variability between the individual prototypes. This independent research project is part of a UCLA technology seminar that focuses on achieving a robotic casting system for the fabrication of 3D concrete component typologies. Robotically-controlled, flexible fabric formwork is explored as a means of accurate, replicable and cost-effective

of 1:12 scale, standing 32” (81.3 cm) tall. One relevant precedent is the American cement Building in Los Angeles by DMJM. This building showcases the potential for precast concrete as a structural as well as aesthetic element. The original design concept for the façade of the Los Angeles Broad Museum by Diller Scofidio + Renfro would be a closer example of the true potential of this system. The far less ambitious façade that was actually realized is a prime example of the limitations of design/fabrication systems currently employed and how this robotically controlled flexible fabric system can dramatically expand the vocabulary of designers worldwide. 02 – DIGITAL PROCESS // The proposed system allows for a complex design to be rationalized into discrete elements that are fabricated and assembled into the final composition. A design of either digital or analog origins is simplified into discrete elements that are analyzed using a parametric feedback loop for structural performance, and then re-engineered for optimization. Using Karamba for Grasshopper3D, load paths that act on each member in the system are calculated to understand their behavior prior to fabrication. Subsequently, each element is refined to meet the structural and performance criteria then re-generated. Once the components are analyzed, their endpoints are determined and optimized. These coordinates are sent to the robotic arms, which translate the Euclid-

The Nature of Robots

production. Geometrically complex concrete objects can be fabricated with practically infinite organic variation and texture. A pair of six-axis robotic arms attached to identical flexible fabric sleeves acts as an adjustable formwork for concrete. The robotic arms can position the endpoints of the limbs accurately and quickly, enabling the composition of an intricate series of unique objects as dictated by the design. For this research, branched objects are designated as a constant starting shape to be manipulated by the robot arms. This allows for one branched limb to be fixed to a stationary formwork at a filling point and two others to be stretched to the desired geometry by two robotic arms working synchronously. Custom robotic assemblies can be manufactured to meet specific project needs beyond the scope of this research. The robotically cast components have the potential for deployment as constituents of a compound structure that can also be realized on a building scale. Load-bearing facades, walls, glazing modules and freestanding sculptures are all possible applications. The composition of the specific project will dictate many factors in the design of the components including for example overall scale of the objects, thickness of casting, or density of the fabric relative to elasticity. For the purpose of evaluating the system, a small-scale prototype is deployed as a case study. This self-supporting structural composition consists of 13 individual pieces


ean coordinates into physical space.

The Nature of Robots

02.1 – SHAPE OPTIMIZATION // The robot lab at UCLA features the arrangement of two industrial robots in tandem. Therefore, this prototype project is constrained to three controlled end points, one of which is fixed in space and functions as the concrete fill point, while the other two are attached to robot arms. The shape of the object between the three end points is optimized for volume, gravity, load path and structural stress points. Increased volume leads to high gravitational deformation which can be alleviated by removing either the center or perimeter mass. Removing mass at the perimeter simplifies the object, the load path and results in the least amount of material per piece. Curves are introduced between the end points to effectively reduce structurally vulnerable sharp turns and provide additional mass where it is needed at the connectors.


02.2 – FROM AGENCY to ANALYTICS // The design process begins by coding in the input parameters necessary for a specific morphology. These include constraints such as the bounding box dimensions, build volume, the size of each component and the characteristics of the casting material. The size of the original fabric formwork is also considered, as its stretch limits will determine the ultimate size of each component. These constraints are written into the Grasshopper definition to form a quantitative basis for the design.

The next phase requires intuition as well as digital modeling. By manipulating an array of nodes connected by line segments, a form emerges that best demonstrates the unique characteristics of the fabric formwork system. Each node represents a connective piece in the composition while the lines represent the cast concrete elements. Each component is analyzed as a discrete agent in a larger structural system, on local then macro levels. When the position of the nodes are established first, the interstitial connected network of line segments are generated by an algorithm which groups the array of points into clusters of three-legged objects. Each set of objects can be considered in isolation. Once the wireframe lattice of nodal connections is built, each line segment is given thickness independently, simulating the fabric stretching of the Lycra formwork. Consideration is given to the fabric elasticity limits and thermochemical curing of the cement when evaluating the composition. Vertical load paths and bending moments within the array are evaluated and the model is recursively adjusted as necessary to attain the design goal. A similar technique of clustering nodes into discreet elements has been implemented in the installation “Cast Thicket” by Kenneth Tracy and Christine Yogiaman. This prototype uses nodes and a connectivity network as the design generators for a cast-in-place concrete technique using plastic formwork.

02.4 – MATRIX COMPOSITION // Early composition studies of the design show that elements with three limbs have limited design capabilities. Without employing weaving strategies, the resultant matrix is predominantly two-dimensional with restricted opportunity in composition depth. To counter this, a truncated tetrahedral ‘coupler’ acts as the interface between concrete elements. This not only affords design flexibility but also facilitates a logical means of attachment by inserting bolts through the truncated faces into sleeve inserts cast into the end of each limb. In addition, the connector adds resilience to the structural assembly while allowing flexibility in the connective nodes. This coupler can be built with a flexible material to allow bending moments or it can be rigid to increase the structural integrity of the composition.

The success of robotic positioned fabric formwork does not reside with this connector, however. It is employed as pan of the composition of the installation, and designed to be a proof of concept for the system. The system itself is intended as a vehicle to solve a myriad of design and construction challenges. It is eminently conceivable that a composition be designed without a similar connector, where each cast member connects directly to the adjacent ones as shown in DMJM’s American Cement Building Facade. It is also possible that the cast piece itself acts as a bespoke coupler for highly variable curtain wall outrigger systems that engage as interstitial elements between standardized tube sections. 02.5 – ROBOTIC ARMATURE and END-ARM TOOL // An armature capable of anchoring one cast object is installed between two 6-axis robotic arms such that the forces applied to the fabric formwork will not disturb the armature. One fabric limb is affixed to each robotic arm and the remaining limb is affixed to the top of the armature. This top limb is attached so that it can also serve as the cement filling point with a retaining cutout that constrains the end to the shape of the nodal connector. After filling, the open end is capped with an acrylic form with inserted nut attached for bolting. Each of the side limbs similarly has an end-arm tool in the shape of the matrix connector. It includes an insert nut attached that allows the fabric to be tensioned prior to pouring the ce-

The Nature of Robots

02.3 – DIGITAL to ANALOG // The translation from Euclidean space to physical space is achieved through BD Move software and the use of robotic arms. Each limb endpoint is assigned its coordinates in space and inclination angle to position its end-arm tool so that the surface normal of the end nodes always faces the concrete center. Each position is systematically sent to the robots as the pieces are cast. Every physical constraint is modeled digitally from the armature and its pedestal to the custom end-arm tool. It is essential to have an accurate 3D environment to minimize error. Any slight omission in the digital file would result in a misaligned final product.


The Nature of Robots

ment mixture. The tools allow secure and rapid attachment/removal from the robotic arms for more consistent casting.


02.6 – FLEXIBLE FABRIC FORM // While the original patent held by Joseph Shivers on segmented copolyetherester elastomers (otherwise known as Spandex) indicates a potential 325% elasticity and spandex yarn has 650% Min/Plus 30 elongation, it is invariably blended up to 80% with other natural and man-made fabrics in order to make it more practical in the manufacture of garments. The ultimate elasticity of spandex materials depends on the specific blend type and ratio. When stretched on the robots to maximum elasticity in field tests, a seam length of 14.25” reached a maximum length of 31.0” before failure, representing elasticity of 217.5%. The fabric itself could be stressed further but the practical assembly requires stitching, so it was tested with a flexible zigzag stitching using polyester thread. Even at maximum tension, the form filled with casting material naturally and with sufficient volume to produce usable results. While other fabric casting methods seek to constrain the fabric to specific shapes without undue deviation, or tailor the fabric specifically to the cast object the Lycra used in this project specifically enables the radical manipulation of the fabric formwork. This allows an almost infinite number of potential cast concrete geometries with a singular fabric starting profile. The use of fabric as a formwork also plays

a role in the compressive strength of the member. Because Lycra allows moisture to wick through its fabric while retaining the cement, a high-strength concrete is produced with minimal air pockets entrained in the object. 02.6 – PROTOTYPING PROCEDURE // The combination of a nodal coupler and concrete element has distinct advantages over other flexible formwork approaches such as that employed in Crease, Fold, Pour. The installation utilized a monolithic pour with flexible formwork but had no means of avoiding form failure other than pouring in stages with resultant cold joints that weaken the structure. By breaking the composition down into individual pieces attached with nodal matrix connectors, the design can be rapidly assembled and disassembled, and the node strength increases. To achieve these advantages, the nodal matrix connector must be rapidly mountable, easy to assemble and capable of handling temporary cantilevered loading during assembly as each new cast piece is initially attached. Making the attachment screws bypass the truncated tetrahedral center and attach to the wall adjacent to the node of the cast object enables the positioning of all bolts through the connector center but adds torsional stress to the wall. As a flexible fabric that has a natural form-found rather than explicit geometry Lycra formwork produces a unique shape and texture. This system not

04 – LEARNING FROM FAILLURE // 04.1 – STRUCTURAL WEAKNESS // Initial casting experiments ranged from plaster of Paris to high-strength cementitious grouts. Experiments with ready-mixed 9,000 psi constriction, grout yielded good results with fast cure time of 3,000 psi within one hour, enabling removal from the robotic arms with an initial set time of 30 min. Despite the high compressive strength, the final cast objects were still highly susceptible to fracturing at finite edges and breakage from sudden impact. Introduction of ½” nylon monofilament fiber at the rate of 0.032 oz. per pound of cement dramatically improved the tensile strength and edge efinition without compromising surface appearance. Further reinforcements of the cast objects can be achieved in larger scale casting with a slip-jointed steel reinforcing bar installed prior to casting. Post-tensioning can also be achieved by introducing flexible polyethylene tubing through the objects prior to casting that will accept a steel cable strung through the entire assembly after casting. An alternate method of support has reinforcing bar attached to the end-arm tools drawn into the fabric forms during tensioning. The assembly for this research employs this method of reinforcing at 1:12 scale. A 16.5 gauge form tie-wire at the pro-

totype scale represents the equivalent of a 5/8” (12.7/20.32 cm) diameter reinforcing bar. 04.2 – FLEXIBLE FABRIC, GRAVITY and MINIMAL DIMENSIONS // Early casting tests reveal that restricting the fabric to less than 1” (1.54 cm) thick leads to breakage, even with 9,000 psi cement. Objects that come to a minimal edge are similarly problematic. Maintaining adequate dimensions throughout the entire cast object becomes a critical requirement to successful production of the pieces. Increasing fabric dimensions beyond our given size results in uncontrollable, excessive masses. If the fabric is not pre-tensioned to at least 60% of maximum elasticity and dimensions exceed allowable limits at any point, runaway loading occurs at the point of greatest volume. Therefore, close control of the potential volume must be maintained through appropriate pre-tensioning. Failures of this type occur most frequently when using 4-way (biaxial) stretch material (designated as “Spandex” brand) but also occur with 2-way (monaxial) stretch material (designated as “Lycra” brand) when limits are exceeded. 05 – CONSTRUCTION IMPLICATIONS // Product applicability includes self-supporting structural facade systems, sculptural breezeblock or wall applications, glazing modules and freestanding sculptural elements. Another potential use includes uniquely-shaped connectors between standardized tube sections for facade

The Nature of Robots

only creates accurately positioned geometries, it also allows for organic variation, texture and a natural unpredictability.


The Nature of Robots

or curtain wall systems. The prefabrication of individual elements speeds time-intensive site-work and diversifies the supply of materials to a building site. As a result, formwork costs can be dramatically reduced and construction waste is virtually eliminated. This is a highly sustainable manufacturing method when compared to alternative processes such as computer numerically controlled milling of single-use forms that will subsequently be destroyed after use. This research has further implications in architectural screen walls. Erwin Hauer pioneered the interlocking concrete wall as a modular facade system. These were used as interior porous partitions whose interlocking character and elegance are considered one of the “quintessential works of modernism” in Domus 1928-1999. Hauer’s exploration of the curvature, plasticity and weaving of concrete are echoed in the potential of this system.


06 – CONCLUSION // The dissemination of robotic fabric formwork into the construction industry would not only allow for the realization of the next generation of parametric, node-based lattice structures, but would significantly reduce concrete construction costs for conventional components. Multi-directional pieces can be cast without rigid formwork and their destructive removal. The robotically-controlled system resulting from this research proved its ability to facilitate a rapid and economical workflow to realize complex or

truncated parametric geometry. Most importantly, this method can help the construction industry adapt to emerging digital fabrication tools and allow for rapid design to production cycles that go beyond rapid prototyping. The effect is a digital-to-physical workflow that is abundantly more flexible in its dimensional freedom and more economical than the industry standard. Future development of this system would likely include the use of custom-built robots tailored specifically to the configurations of the pieces to be cast, allowing for many more points and limbs if desired. Customized robotic work processes could one day replace human labor in many professions including those on the construction site. ACKNOWLEDGMENTS // The authors gratefully acknowledge the technical and material Teaching of Julia Koerner and Peter Vikar from conception to completion. This work would not have been possible without their expertise at the UCLA School of Architecture and Urban Design. Special thanks to Greg Lynn and Guvenc Ozel for their guidance towards the completion of the project. The early experiments in casting with fabric for this project were conducted with fellow UCLA students Shobitha Jacob, Oscar Li and Qi Zhang who were a part of the initial research.

Credits // Gregor Steinhagen // Johannes Braumann// Jan Brüninghaus, Matthias Neuhaus // Sigrid Brell-Cokcan // Bernd Kuhlenkötter // ABSTRACT // Traditional artistic stone processing techniques offer vast possibilities for finishing stone products. However, stone processing is physically highly demanding work requiring stamina as well as skill. This makes products expensive to produce and the detailed design only accessible for skilled masons as an efficient communication between designers and masons is difficult. We introduce a robot-based approach to produce “artistic” surfaces for individualized stone products. First, distinctive traditional,

manual processing techniques will be introduced and analyzed towards enabling us to specify the necessary requirements of an adaption to an industrial robot. These requirements are then implemented in an automated tool and an automated path-planning algorithm. Building upon a visual programming environment we will present an accessible interface that allows the user to apply customizable stone structuring patterns to an individual stone product. 01 - INTRODUCTION // Stonemasonry is an old craft, which developed over centuries. It is characterized by a high number of different techniques, which make use of both tool geometry, as well as the tool’s handling. There are also a vast number of different stones with different heterogeneous behavior: Not every technique is applicable for every stone. While some techniques are easy to produce, other techniques are demanding and require high manual skills and long training. Thus the production of artistic stone surfaces with such techniques requires knowledge and talent. This restricts the accessibility of techniques for designers and architects - the sole possibility is the communication with a skilled mason. However, the number of techniques and their variability combined with the many different kinds of stones make it impossible to communicate the vision of one’s design idea in detail. This results in the re-application of tradi-

The Nature of Robots



The Nature of Robots 90

tional old patterns whereas new design strategies do not often find their way into architectural applications. As such, the main field of application for stone-structuring techniques is still the restoration of old buildings. Another drawback is that most techniques are physically demanding, so that even a skilled mason needs a long time to produce large surfaces-making the application of stone structuring to building fronts very expensive. While there exist some large specialized machines for stone structuring, they either rely on very simple, fixed chisel-strategies or simply aim to imitate traditional techniques with grinding tools. As such, they do not allow the user to go beyond traditional structuring strategies or to work with three-dimensional, curved, surfaces. Generic methods that can also be applied to stone structuring such as milling sawing and waterjet-cutting are being used in industry but are either very time consuming or do not closely emulate the finish of traditional stone structuring techniques. Current research in the field of architecture and design focuses mostly on cutting, rather than surface processing. All these problems are addressed in a new robot based approach, where a robot-mounted tool performs the traditional techniques and then goes beyond the scope of manual processes. The freedom of the robot allows variation in the technique parameters and results in huge design possibilities. Furthermore, a design interface is implemented which allows designers to

model their ideas and carve them into stone as the special masonry knowledge is embedded within the code and thus not required of the user. First we will describe the techniques, which have been analyzed. We will then describe the adaption of these techniques to the robot. The path planning which connects the design interface with real robot cell will be shown. In conclusion, we will give an overview to the design interface and show first examples of the design possibilities. 02 – DEFINING a STONE STRUCTURING PROCESS // Robotic arms are highly multifunctional machines with a power and precision that - at least in combination - by far exceed the capabilities of human arms. However, in order to be able to program a particular robotic task, we have to be able to clearly define it. This task definition is one of the main challenges towards applying robotic labor to stone surface structuring: The most common robotic applications are well defined and structured (e.g. pickand-placing or spot-welding in the automotive industry), while some more complex processes such as milling can at least be quantified, measured, and evaluated based on static criteria such as the minimization of difference between the digital data and resulting physical output. Stone surfacing is much less defined: two stone masons utilizing the same technique and an identical tool can produce very different surfaces that are the result of small variations in

03 – ANALYSIS of TRADITIONAL TECHNIQUES // Due to the high number of different techniques, a first selection of popular strategies was made with the help of the masons of Bamberger Natursteinwerke, who later also supported the analysis of the different techniques. The selection was based on the dissemination of techniques and an estimation of their potential for automation. Tooling, aligned tooling, punching and bush hammering were selected as first experiments. When not considering the

complete imitation of the mason’s movement, the most important aspects of the techniques are the chisel movement and the kinetic energy, which is applied to the chisel with the hammer. Both are analyzed with the high-speed camera system GOM Pontos HS capturing the manually performed techniques with 20,000 fps. The videos can afterwards be analyzed with the help of markers placed on the chisel and the hammer. Thus we can analyze the relevant movements in X and Y direction as well as the angle around the Z axis. The energy was estimated by the speed of the hammer before impact and its weight. In the process of analysis we also analyzed further aspects of automation such as predictability of results and necessary positioning precision Similar methods have so far only been applied in the field of anthropology and archaeology to understand prehistoric tooling processes. 1. Tooling // This technique is very common for finishing traditional stone surfaces, creating highly characteristic rounded grooves. It is performed with a drove chisel and a mallet. While moving through the stone the chisel is rotating slightly. At the end of the movement it leaves the stone. This movement results in the characteristic appearance. The Energy was estimated to be 19-46 J. This energy can be applied with a robot-mounted actor. Based on these first measurements we can say with a high confidence that the movement of the chisel is challenging but reproducible. Positioning accuracy and predictability of

The Nature of Robots

force, angle, speed, and other parameters. These variations are then not defects, but rather make up the aesthetic appeal of traditionally structured stone surfaces. Previous research and literature reviews show that a direct translation of human movements to robotic movements is highly challenging even for tasks that are considered easy for humans. The precise timing and force control needed to exactly emulate a human mason was therefore considered to be beyond the scope of the research project. As such, the idea was that rather than trying to emulate human processes directly, we would attempt to create optimized robotic processes that incorporate traditional masonry strategies but augment them in the areas where a robot exceeds a stonemason, namely, in speed and accuracy. The first step towards implementing new movement strategies is to analyze existing, manual techniques and to evaluate them in regards to “compatibility” with robotic processes.


The Nature of Robots 92

the results are manageable as well. 2. Aligned Tooling // Aligned tooling is a variation of tooling which main application can be found at the edges and comers of complex stone products where there is no space for performing the original tooling. The chisel does not reave the stone and moves linear into the stone, resulting in triangular, rather than rounded grooves. The movement analysis exhibits only a small rotation of the chisel. Since smaller drove chisels are commonly used, the energy revel is smaller than what we observed with normal tooling, while other values are mostly within the same range. Resulting from all this aspects we conclude that the automation of this technique should be even easier than normal tooling. 3. Punching // This technique is performed with a punch and a hammer. There are different variations of this technique: either single punches are performed in point punching or concluding punches are performed to produce a linear pattern through in line punching. However, especially the second one is difficult to automate since the predictability of the punches is row. For point-punching, the movement is relatively simple since it literally punches into the stone with an estimated energy of 23.99 J. 4. Bush Hammering // This technique uses a hammer head with a number of tips and can be either performed with a special bush hammer or pneumatic tools. The hammer hits the surface but does not move deep into the stone, producing a shallow pattern

on the surface. This results in a movement that is relatively easy to perform and requires just 1.14 J, but only offers a limited degree of variability. 04 – DESIGN PARAMETERS // Building upon the analysis of human stone masons using high-speed cameras a custom, a modular tool was developed that emulates the stone masons micro-movements and decouples its chisel from the robot itself, preventing backlash to the robot’s gear system. Therefore, the robot acts in the macro-scale as a high-speed, accurate spatial positioner while the modular tool interacts with the stone. This modular approach allowed us to pursue the time-constrained process development in two parallel tracks, while also facilitating future maintenance and trouble-shooting. Based on the custom-developed stone-structuring tool, we have identified a number of design parameters that can be adjusted to greatly influence the surface finish and process time. These have been grouped into three different layers, with the first one defining the general layout of the design and the second and third one fine-tuning the structure and depth of the processing. Layer 1. Tool position and rotation in the material’s surface // The general tool position has got the most significant impact on the surface as it defines its overlaying structure, e.g. by following the isocurves of a surface, tracing lines, or being aligned according to the raster information of an image. This positioning happens exclu-

For the initial experiments we have created a number of different strategies that utilize the three parameters previously described, depending on the design intent. These first structures are based on patterns, curves, raster images, or a combination of the above. 1. Patterns // Manual stone structuring is based on regular patterns that are applied to a surface. As such, we have created a set of adjustable, regular patterns that can then also be perturbed through attractor points

or other parameters. A sine curve adjusts the rotation both within the stone surface plane, as well as around the tool’s X axis, resulting in a 3D form which would not be possible with a manual process. 2. Curves // Here the tool path is provided as a list of curves that have to be chiseled into the stone. A significant challenge is the division of a curve into a number of linear segments with a fixed length (according to the used tooltip). To divide the curve according to different chisel sizes we assessed the chord, which connects the starting point on the surface with points on the curve. The length has to correspond to the chosen chisel length. This approach works in cases where the curvature is not too high, and there are no complex forms, small grooves or waveforms. Based on the chord middle the chisel position is then derived. If a long curve is divided in different chisel segments, the overlap of two chisel positions is a parameter, which can vary. Thus a curve can be build up from a number of chisel positions with a fixed chisel size. The inclination of the chisel is derived from the normal surface vector of this chisel position and the design inclination defined by the user. 3. Raster Images // In previous research we have already explored the possibilities of using raster images for surface structuring. In the case of stone structuring, the brightness values of an image are sampled and then turned into toolpaths. Depending on the (processed) surface of each stone, the stone becomes then either bright-

The Nature of Robots

sively in the XY plane of flat material, or alternatively on the UV/mesh parametrization of curved surfaces. Layer 2. Relative tool inclination // The tool’s inclination in relation to the stone-surface is performed around the chisel edge by the internal tool mechanism. Due to the optimized movement of the tool, following the predefined path rather than a straight line, the tool inclination has got a distinctive effect on the surface finish, allowing us to accurately fine-tune the width of each stroke. Layer 3. Hardware: width and Force // The third strategy to influence the surface finish is through adjusting the hardware parameters, most notably the tool dimensions and the force with which the tooltip is applied to the stone. In particular, the tool-width has a significant impact on process time however, while larger tools allow for quicker processing of structures, they are also more geometrically limited when not following the rulings of a surface.


The Nature of Robots

er or darker where it is hit. In addition to locally adjusting the brightness, the orientation of the tool geometry can also be informed by the process, e.g. based on vectorized geometry.


05 GEOMETRY and REPRESENTATION // The representation of the developed strategies is a very important topic as it is crucial for allowing a new user to estimate the result of the programmed process. An ideal way to accurately represent the effect of the tool would be to perform Boolean operations for each hit. However, such a process would be extremely computationally expensive, and would thus interrupt the fluid design process. As an alternative, the most basic representation of the tool would be a line, representing its chisel edge. While this visualizes the effect of the first fabrication layer it would not show the effect of the tool’s inclination. We therefore implemented a geometric model that calculates the approximate shape of each chiseling movement based on the tool size, inclination, force and stone without having to “physically” intersect two geometries. Therefore, the geometric footprint and memory requirements are kept low, even for complex structures. However, significant challenges arise once doubly-curved surfaces are processed (or when the tool’s y-axis moves away from the straight rulings of the surface), both in regards to the physical process and the initial preview: The contact surface between the tool and the stone then covers only a part of the length of the tool,

which can possibly chip either tool or stone. Depending on the stone that is being used, the user has to evaluate the effect based on the visual feedback provided by our software, and either takes the risk with soft stone, or use an automated process to optimize the tool’s orientation with the goal of lessening the irregular effect. 06 PROGRAMMING STONE STRUCTURING DESIGN PATTERNS // We expect that by solving the kinematic complexity and timing of a stone structuring process with an intelligent, modular tool, the robot can be used much more freely as a design tool with a comparably low stone-specific overhead. This macro-path planning is however still highly challenging, as no commercial solutions for the dynamic structuring of stone are available on the market. Initial virtual simulations of early stone-structuring approaches showed that the largesize of the structuring tool, along with the geometric constraints of traditional chiselling tools, can very quickly lead to unreachable positions, and singularities-thus requiring a capable, dynamic robot simulation environment where such processes can be reliably defined, simulated, and optimized. As such we decided to build upon the flexible KUKA | prc framework and expanded it at the source code level with new simulation capabilities, improved interpolation algorithms, and multi-core optimizations to more efficiently optimize stone structuring processes. KUKA l prc itself builds upon the visual programming environment

in the process are poses where the fourth axis and sixth axis are aligned. These positions can be resolved if the inclination of the chisel is slightly changed for the corresponding movement instruction. This approach builds upon the new monitoring function of KUKA l prc which exposes much of the simulated robot data for analysis and optimization. For an iterative update of the chisel positions we then coupled the robot simulation with the plugin Hoopsnake. Thus the angle is automatically altered only as much as is absolutely necessary. As a further step, we plan to implement a module that optimizes the sequence of the single chisel poses in regard to the process time, so that even complex stone products with a high number of chisel positions can be produced efficiently. 07 CONCLUSION // In this chapter, we have presented our work on the analysis of traditional stone processing techniques and their adaption to an industrial robot. By implementing this knowhow and strategies into the custom soft- and hardware, our research allows designers to be directly involved in the structuring of stone. Changeable parameters such as tool inclination and different design patterns can be adjusted or new stone products can be defined from scratch. The automated path planning also enables the designer to focus on the design rather than on the intricacies of robotic path planning such as singularities. Thus we have developed a competi-

The Nature of Robots

Grasshopper and expands the standard modules with a range of robot simulation and control components. As such it is ideal for the usage as a path-planning tool for complex technical systems with an easy to use interface. The path planning is divided in different steps. First, the user designs an idea as described above and receives feedback regarding the applicability of different chisel sizes to the curvature by color-coding each preview position. With this information the designer can decide when to apply a distinct chisel size to a surface point. For non-planar surfaces, the whole path planning is performed on a scanned mesh representation of the stone. This is necessary because of high production deviations in the machining of stone products, where the tool is continuously ground down by the abrasive stone. These initial offline steps can be performed without a robot. The first step where the robot is needed is the measuring of the initial stone position. We implemented a measurement system, which was developed in a previous research project. It is based on a laser triangulation sensor, which captures a number of points on the stone. The points are than matched with a point cloud of the stone and thus the stone is virtually placed in the robot’s simulated workspace. The robot’s poses are then updated and checked for singularities and other conflicts such as reachability. Due to the geometries of stone processing, the only probable singularities


tive approach that will enable a broader user group to apply traditional stone processing techniques. Our further work will focus on the optimization of the shown technique, the adaption of other techniques, and new robot control interfaces for dynamic processes Furthermore, we will optimize the process speed by analyzing and optimizing the dynamic behavior of the robot and tool.

The Nature of Robots

ACKOWLEDGMENTS // The research


leading to these results has received funding from the European Union Seventh Framework Programme (FY7120fJ7-2013) under grant agreement n.606453. SME consortium partner: Klero GmbH, Bamberger Natursteinwerk, G. Gibson & Co Ltd, II Architects. Research consortium partners: TU Dortmund University, Association for Robots in Architecture,Labor. Associate consortium partner: KUKA. Web:

TOWARDS a MACRO DESIGN of ACUSTIC SURFACES Robotic Fabrication of Complex Pattern Geometries

fabrication, specifically when a series of tests is required with a high degree of detail. Whereas 3D printed samples are impractically small, and CNC fabrication is limited by tool path axis, robotic fabrication enables precision for 1:10 scale model prototypes such as the quick sampling of sound discs that can be used to analyze acoustic scattering. Through a process of reverse engineering from parametric modeling to scale model production to physical simulation, the acoustic reflective properties of surface patterns are investigated for scattering coefficients, in order to derive statistical data on acoustic properties of these surfaces, and to deduce design rules.

Credits // Dagmar Reinhardt // Densil Cabrera // Alexander Jung // Rod Watt/ ABSTRACT // In the context of acoustic performance in architecture, this paper presents research into the computational design and robotic fabrication of surfaces with micro-geometries that can change the acoustic response of space. It explores the design affordances for acoustically efficient patterns for sound scattering - between complex geometries, acoustical effects, and robotic fabrication. Spline curves pose a problem for the translation between geometry and material

01 - INTRODUCTION // Architectural surfaces shape the way in which we hear space. Sounds such as speech can be thought of as an acoustic signal that is heard through architecture, which forms an extended acoustic system. Architecture transforms a sound that travels, but also influences the way in which signals are produced by people: speech projection depends on the visual and auditory environment experienced by the talker. On a very basic level, all architectural surfaces express acoustic space through the way in which sound is reflected at each surface, depending on material properties, surface geometry, and the sound field’s spatial, spectral

The Nature of Robots

KEYWORDS // Subtractive manufacturing // Complex geometries // Parametric design // Robotic fabrication // Sound scattering


The Nature of Robots 98

and temporal features. Acoustic performance results from the complex combination of spatial volume, building envelope and surface properties, which combine to affect speech transmission, in many contexts yielding reduced intelligibility. Through diffusion, scattering or appropriate reflection/ absorption of sound this degradation can be improved. And while the discourse of performance as a key design factor of the built environment has been associated with design computation, geometry, material, or structural performance, recent studies have just begun to integrate acoustic performance. This paper reports on ongoing research into the acoustic effects of complex architectural geometries, with a focus here on robotic fabrication of micro-geometric surfaces that could be used to improve acoustic performance by scattering. In doing so, it expands previous research into the robotic fabrication of spatial geometries that change the coloration of sound. We are presenting here ongoing research and workflow shared between architecture and acoustics, for the design of robotically fabricated scattering surfaces as scale models for acoustic testing. In the following, the paper introduces parameters of sound reflections; reports about a series of preliminary design and physical tests of acoustic patterns. It further discusses a transfer from generative tools to 6-axis robotic fabrication, linked to the angle and cavity depth in a surface medium that impact on acoustic performance.

01 – SPECULAR REFLECTIONS and ACUSTIC SCATTERING // Like light, the propagation of sound in space can be understood through ray and wave theory, which is more important on a human scale because sound wavelengths can be significantly large compared to the size of objects, surfaces, and surface elements in the human environment. To address this complexity, architectural acoustics employs a variety of theoretical paradigms for modeling the behavior of sound: including analytic, statistical and numerical methods; based on ray, wave and particle propagation; in any or all of the domains of time, frequency and space. In architectural spaces, sound focusing, discrete and flutter echoes, and sound coloration can strongly detract from the space’s usability for speech communication or music performance. These problems can be avoided without deadening the acoustics by introducing scattering, whereby sound is reflected irregularly over a wide range of directions. In general terms, scattering can be created by variation in the physical surface such as curvature, relief forms or textures, and changes in contrasts in material acoustic properties. The angle of a reflection may be influenced by the incidence angle onto the particular part of a surface and the incidence angle on the overall surface, which can be conceptualized as specular reflection. Even if specular reflection occurs on a small scale, there may be phase interference between reflections across the surface, yielding a much more

of this, and similarly at odd multiples of the frequency). Yet for 100 Hz the wavelength is 10 times longer, so the phase change due to this well is only 18°, hence the surface can be considered to be almost flat at that frequency, yielding an essentially specular reflection. When there are multiple depths across a surface, the reflection pattern develops from interference between a multitude of phase shifts, so that scattering occurs from a macroscopic perspective. Increasing the contrast in surface depth extends scattering to lower frequencies, although in practical implementation there are limits to the available depth (due to cost and/or available space). In the high frequency range, surface elements may be sufficiently large (relative to wavelength) for their angle to affect the reflection angle. Periodicity in the surface reduces the complexity of the reflected sound field, and so aperiodic patterns are preferred for scattering surface treatments to achieve high scattering. 03 - SHAPING ANGLES: SCATTERING DISKS // The complexity of acoustic reflections from micro-geometric surfaces provides a workflow rationale that extends from scripting surfaces towards the physical measurement of scale prototypes as an important part of the design and validation process. In order to identify the potential of acoustic behavior, and to derive threshold criteria, the research employed an iterative collaborative process with the following parts:

The Nature of Robots

complex reflected sound field, so that sound is scattered. Non-flat surfaces introduce small time delays, which vary with the depth of the cavity or well in the surface, and which can result in significant phase shifts introduced by a reflected wave front so scattering occurs. As this is non-trivial to predict, physical measurement of prototypes plays an important role in the development of scattering surfaces. Specular acoustic reflections occur on large and small scales, where the angle of reflection is equal to the incidence angle mirrored by the surface’s normal (Sn) at the point of reflection. While the reflected rays radiate in the same direction, in practice this will only occur in this simple way when the phase shift due to varying depth is small - otherwise phase interference between reflections will yield scattering. Considering that the wavelengths of audible sound range between 17 m (for extremely low frequency) to 17 mm (extremely high frequency), it is evident that for a given variable depth surface the extent of scattering is likely to be strongly dependent on frequency. The wavelength of sound is equal to the speed of sound (typically about 344 m/s) divided by frequency. For example, for a frequency of1 kHz, the wavelength is 0.344 m: if a well depth is ¼ of this (0.086 m), this would result in a 180° phase shift relative to a part of the surface with no well, potentially creating local sound cancellation (and complex reflection patterns for frequencies in the vicinity


The Nature of Robots

1. Specification of the architectural design parameters, along with the acoustic design aims (e.g. scattering coefficient spectrum); 2. Computational design of specific surface micro-geometries; 3. Fabrication of physical scale model test samples in the form of discs; 4. Acoustic measurement and analysis of sample performance; 5. Refinement of the design with potential further iteration. This approach concatenates computational design, acoustic analysis and robotic fabrication, which expands the potential scope of micro-geometric surfaces by integrating scripting logic, surface angles and depth, and toolpath, thus enabling successive acoustic design variations that can be tested for proficiency.


03.1 - PATTERN SCRIPTS: HEXDF and FLOWL // As an initial departure point, two pattern variations were generated in Grasshopper (a plug-into McNeel Rhino/visual scripting environment). The first is a hexagonal periodic (HexNDef-S1) and deformed pattern (HexDef-S2), the second a vector based pattern (Flowl-S1) which adopted the customized script Flowl. Both develop zones of highly differentiated depth across the surface. The first sample uses a parametric pattern of tessellated hexagons varied between two primary points, so that the initially periodic tiling is deformed in attraction/repulsion. As a result, individual facets vary in depth, height, directionality, so that diversity is created

for testing sound scattering and sound diffusive properties of surfaces. The second geometry orients curvilinear splines in relation to an increasing number of attractor points. It uses a parametric vector field of streams between a variable number of up to 9 of attractor points, and relative adjacency between neighboring fields. Through control over number of attractor points in the scripting environment, the overall depth of surface resulting from isocurves can be manipulated and used as adjustable toolpath in KUKA l prc (a Rhino McNeel plug-in for robotic fabrication). 03.2 - PROTOTYPING 1:1 SCALE MODELS for ACOUSTIC TESTING // Computational prediction of performance describes sound through mathematical models, but scale models or prototypes monitor the physical phenomenon itself. In wave acoustics, accurate computational prediction can be very expensive, whereas physical modeling is comparatively efficient. The test surface samples are disks (designed circular so they can be rotated without any change in their outer edge, which is important for scattering measurement), fabricated at a scale of 1:10, and tested in a scale model reverberation chamber. As a shared geometry base, the patterns were thus adapted to circular disks. The sound discs were thus designed with 310 mm diameter and 19 mm depth, and prototyped in subtractive cutting processes in XPS Styrofoam

03.3 – RANDOM INCIDENCE MEASUREMENT // The acoustic behavior of the 1:10 scale model prototype surfaces was then tested by random incidence measurement in a scale-model reverberant room. Each disk is placed on a turntable, and synchronously averaged impulse responses are obtained for different source and receiver positions from the material sample. The acoustic performance is measured as apparent reverberation time: with and without the sample; and in stasis and rotation, yielding a spectrum of random incidence scattering coefficients (per ISO 17497-1:2004). From these, the scattering coefficient is calculated, which describes the ratio of acoustic energy reflected in a non-specular manner to the total reflected acoustic energy. This provides a summary parameter by which fabricated prototypes can be evaluated for their effectiveness over the frequency range of interest, and presents criteria for further design developments. Prototypes evaluated included a non-patterned reference disk, a non - (HexNF) and deformed hexagonal tessellation (HexDef-Sl red), and the flowl (blue), which showed better results for the lat-

ter. The normal incidence absorption coefficient of the material was measured (in an impedance tube), showing values of less than 0.3 across the frequency range. Results measured for HexNF (Undeformed) were insignificant, which is likely to be due to a combination of the periodicity of its pattern and the shallow depth across each hexagon. In contrast, the second and deformed pattern (HexDef) resulted in significant scattering at frequencies above I kHz (after frequency scaling), thereby proving to be effective in sound scattering for important sounds such as human speech. But better scattering effects resulted from the Flowl surface, due to depth and number of ridges/valley volumes. The physical test series resulted here in two important determining factors for fabrication of further surfaces in the Flowl series: firstly, results are impacted by the structure of the discs; and secondly in relation to width, depth, and variation of depth across each flow line of the design. This is important, because these two key criteria can be parametrically controlled through computational design. Furthermore, the particular benefit of robotic fabrication (compared to other options such as CNC fabrication) is the ability to develop forms using a wide range of tool-path angles, which can expand the effective surface area of a prototype scattering sample, with the potential to increase its effective depth by cutting at angles significantly different to the surface’s normal.

The Nature of Robots

and wood. While the dimensions of the scattering discs were selected to fit the acoustic scale reverberation room, height field variations in the prototypical scale models were also dependent on the CNC milling path of the tooling head (4 mm/1 mm spindle head) and cutting time, and the directionality of material, resulting in different roughness of disc surfaces.


The Nature of Robots 102

04 – REVERSE DESIGN of SOUND: PATTERN GEOMETRIES RELATIVE TO 6-AXIS FABRICATION ANGLE// The research then continued in a reverse design process, by which the angle of sound reflection was linked to the robotic toolpath, thereby creating depth in the material that impacts the sound reflection. This was explored focusing on two aspects to increased depth: firstly, by increasing depth of available angle per singular isocurve, and secondly by increasing the sum of lines resulting in the total well area that sweep along each isocurve. In this further development of the pattern, we prioritized the Flowl’s geometrical logic where multiple surfaces result from directional robotic milling of isocurves. In combination with the controlled tooling path, this results in overall depth for the acoustic disk. Effectively, the mother geome0ry is then reduced to parametric scripting of attractor points (GH and plug-in flowl), in combination with robot simulation in KUKA l prc. The 6-axis robotic fabrication offers degrees of freedom over CNC (MultiCAM CNC 3-axis router) milling that allows to adopt swarf or flank machining techniques in which the side of the tool is used to produce the desired patterns in a single pass, the pattern adapted to the scale of the surface texture, material and tool selection. 04.1 – ROBOTIC FABRICATION: INITIAL PARAMETERS // The particular Flowl prototype uses the advantage of a relatively simple geometrical rule for deforming a collection of individ-

ual lines relative to one attractor and its adjacent neighbors (depending on the GH definition). Each single line is a spline, but can be directly linked to the robotic toolpath, with the angle of the milling tool predefined and variable along the curve, resulting in the depth of valley that must be achieved to provide scattering. Instead of a workflow with multiple passes along splines, a pair of 2 passes can produce the valley and thus effect. Our initial studies focused on robotic milling of a single well resulting from two parallel isocurves. The robotic parameters include here toolpath and defined angle of the milling tool; multiple passes along isocurve; distance of end effector to material surface, depth and surface angles of well variable along the curve. The robotic cutting path is set as a series of paths between: (a) two edge curves (top and bottom surface), with (b) subdivision between points on each curve with distance 20 mm, and the defined limit of 19 mm cutting depth relative to the 3.2 mm ballpoint toolbit (120 mm tooldepth). 04.2 – WORKFLOW and FABRICATION PROCESS // This system was then parametrized for the disc system, and tested on the more complex geometry of the Flowl series. The robotically fabricated prototype includes 414 faces that are again resulting from isocurves from XPS extruded Styrofoam, which allows precise and fine milling due to its close packed and non-directional material characteristics. The transfer from

04.2 – CONCLUSION and FUTURE RESERCH // Acoustic surface design is a specialized area of architecture, which has traditionally involved relatively simple structures exploiting the bulk properties of materials and resonance phenomena. The conceptual framework and pattern language available for acoustic scattering discs can be significantly expanded through highly flexible robotic fabrication that allows the efficient production of 1:10 scale model prototypes. This research project has presented an example of this, applied to acoustic scattering. Scattering is not the only acoustic surface characteristic that can benefit from high degrees of freedom digital fabrication. The micro-design of special patterns of reflection, highly tuned absorption, and potentially other unconventional acoustic surface behaviors can be investigated in similar ways. However, scattering is of particular interest as a case study because there is a standard and efficient method to measure it. The scattering coefficient is a single number (as a function of frequency) relating to statistically-defined (random) sound fields. Future research will study reflections more comprehensively with a 196-hemispherical loudspeaker array in a sound absorptive room, which, in conjunction with a microphone array, could comprehensively describe reflection phenomena (including direction-specific absorption and diffusion) from micro-geometric surface designs. Future research can then extend criteria, conceptual framework and robotic

The Nature of Robots

the scripting in Grasshopper towards KUKA l prc took into account several adaptations to the robotic fabrication process. The positioning of desired object surface in relation to the robotic zero point required a vertical orientation of the material (EPSX Styrofoam), due to the local positioning of the robot endefector. This proved to be beneficial as material cut-offs were spontaneously removed from the toolpath by gravity. Out of the total number of faces, 34 faces needed to be adjusted manually (>8%), due to the intersections of isocurves in concentration areas around attractor points. Each face is controlled in KUKA l prc as each double curved surface path with 4 stepovers, in closely packed linear moves. All faces are cut at a maximum 40° angle of the toolbit relative to the surface’s normal. To date, the research has continued the robotic manufacturing of a series of sound effective acoustic discs as scale model prototypes for acoustic analysis in different degrees of overall surface depth. In addition, the parametric GH code and KUKA l prc have been aligned, and thus enabled an upscaling towards three times the original size (930 mm), cut to a maximum size of milling depth of 65 mm with a 6 mm ballpoint toolbit. In a continued work series, patterns can be further manufactured in plaster, using a KUKA KR10 onsite, with again toolpaths scaled to the required robotic reach and work envelope.


fabrication processes to the acoustic surface treatment of existing surface geometries, or to the conditioning of complex curved surfaces that can be sound effective on a larger scale in architectural space(s).

The Nature of Robots

ACKNOWLEDGMENTS // This project is part of an ongoing interdisciplinary research collaboration (Architecture and Audio & Acoustics) into complex curved geometries and their acoustic behavior, undertaken at the Faculty of Architecture, Design and Planning, The University of Sydney. Research assistance for coding of geometric patterns by Iain Blampied and Mitchell R page, for digital fabrication by Celeste Raanoja (2014), with acoustic behavior and sound measuring undertaken by James R Colla, Jesse H L,oweke and David S O’Brien. The research has been supported by 2014 ECR SEED fund, and was produced at DMaF, The University of Sydney.


ROBOTIC HOT-BLADE CUTTING An Industrial Approach to Cost-Effective Production of Doubled Curved Concrete

blade, mounted on the flanges of two manipulators. Re-orienting or translating either end of the blade dynamically deforms the blade’s curvature. The blade follows the contours of the rationalized surface by continuous change in position and orientation of the end-effectors. The concept’s potential is studied by a pilot production of a full-scale demonstrator panel assembly.

Credits // Asbjørn Søndergaard // Jelle Feringa // Toke Nørbjerg // Kasper Steenstrup // David Brander // Jens Graversen // Steen Markvorsen // Andreas Brerentzen // Kiril Petkov // Jesper Hattel // Kenn Clausen // Kasper Jensen // Lars Knudsen // Jacob Kortbek // ABSTRACT // This research presents a novel method for cost-effective, robotic production of double curved formwork in Expanded Polystyrene (EPS) for in situ and prefabricated concrete construction. A rationalization and segmentation procedure is developed, which allows for the transliteration of double curved NURBS surfaces to Euler elastica surface segments, while respecting various constraints of production. An 18 axis, tri-robot system approximates double curved NURBS surfaces by means of an elastically deformed and heated

01 - INTRODUCTION // The vast majority of contemporary building designs are restrained to a formal language of planar surfaces and derivative geometric constructs; a constraint that stems from the practicalities of construction, which favors the use of mass-produced semi-manufactures and-for concrete in particular-modular, reusable formwork systems. An increasing number of high-profile project designs challenge the dominant paradigm. The challenge is posed by advanced building design projects, such as the Kagamigahara Crematorium (Toyo Ito Architects 2006) and Waalbridge Extension (Zwart & Jansma, under construction), which utilize manual production of formwork to achieve complex curvatures; and building projects which employ large scale CNC-milling to realize advanced structures, such as the Museum Foundation Louis Vuitton by Gehry & Associates (Paris 2014); the Nordpark cable railway by Zaha Hadid Architects

The Nature of Robots

KEYWORDS // Robotic fabrication // Concrete structures // Hot-Blade EPsmolds // Cost-efficiency //


The Nature of Robots 106

(Nordpark 2007), and the Metz Pompidou by Shigeru Ban (Metz 2010). However, neither manual formwork production nor large scale CNC-milling provide a cost-effective option for general construction, and projects of this type therefore require extraordinary budget frameworks for realization. Recent developments in architectural robotics by authors of this paper have demonstrated novel, cost-effective means of producing bespoke formwork with the constraint of being limited to ruled surface. The Robotic Hotwire Cutting (RHWC) utilized to approach is concrete casting in Expanded Polystyrene that has been developed to industrial scale. Currently, Odico Aps is putting forward RHWC in relation to a project design by the Danish artist Olafur Eliasson, for the Kirk Kapital Hq in Vejle. Here over 4000 sqm of formwork are produced, achieving production speeds order of magnitudes faster than CNC-milling through the principal mechanics of the method. In extension of these developments, experiments at Odico are performed in abrasive wire-sawing. Through this technique, the same digital procedures – facilitated by the internally developed control software, PyRapid - is applied to direct processing of construction materials, such as industrial marble. In further maturation of the concept, the method is being adapted in partnership development with Baümer AG for industrial machining. Prototype production have revealed further significant reduction in machining times, in which full scale elements may be cut in matter of sec-

onds. However, for a number of projects, the realization of general double curved structures is imperative. Here,no effective methods currently exist for architectural scale in industrial production. In 2012, Odico Aps. tendered as part of a consortium for the realization of the aforementioned Extended Waalbridge project. Here, the double curvature of the columns of the bridge elegantly blending with the bridge slab are dominated in a single direction. The considerable scale of the project implied large local radii (between, 1 and 2 m) of the surfaces. Since, for this scale, CNC milling molds from EPS would have been a prohibitively ineffective method, digital manufacturing would not be economically competitive with the more traditional approach that was chosen. While developing the tender documents, Odico Aps. realized that the Hot Blade cutting method discussed in this paper would represent a competitive solution. 02 – STATE of the ART // Contemporary construction currently employs either manually produced, bespoke formwork or CNC-milling of foam molds for the realization of complex concrete structures. In addition to these techniques, actuated mold systems have been explored by Danish Adapa and in the EU FP7 project TailorCrete. This technique employs actuation of a flexible membrane as a casting surface; however, the method is limited to concrete prefabrication; by the casting pressure the individual systems can take; and the need for

jority of cases be described via swept splines. The term ‘splines’, nowadays refers to piecewise polynomial or rational functions used in CAD systems to model curves and surfaces. However, prior to the introduction of computers in the 1950s the term was used for thin wooden rods the shapes of, which were manipulated by the placement of so-called “ducks” at various points to create a naturally smooth curve for drawing designs. These were used in ship building and, later, in the aviation and automotive industries. The placement of the ducks simulates the placement of ribs in the hull of the ship, and hence the curve drawn by following the spline is an accurate reflection of the natural shape adopted by the planks forming the ship’s hull. The use of splines for the storage and transmission of a design goes-back to the Romans, in the form of physical templates for the ribs of ships. Splines and ducks suitable for drawings of ship designs were developed later, perhaps in Hull in the 1600s. The mathematical shape of a physical spline can be described exactly, although it requires the use of so-called elliptic functions, which are nonlinear in nature. The correct mathematical model for an elastic rod bent by a force at one end with the other end fixed was given by James Bernoulli in 1691. In his approximation of the solution for the case that the ends of the rod are at right angles to each other, he recognized that the solutions would require non-standard functions. Later, in 1743, Bemoulli’s nephew, Daniel, suggested the problem to Euler, who

The Nature of Robots

multiple casting aggregates for large volume production due the curing time for concrete elements. In addition, dynamic slip-casting for column elements is being explored, as a variant of the additive manufacturing or concrete structures. These and related methods attempt avoiding the need for formwork altogether - however do so at the cost of significant degrees of freedom, such as the capacity to realize cantilevered designs. Finally, fabric formworks have been proposed and experimentally applied as an alternative technique for the casting of advanced designs. This approach is challenged by the capacity of the fabric to achieve desired designs, as well as the unpredictability of the fabric behavior in combination with the required complexity of creating bespoke molds. Common denominator of the described developments is the requirement of shifting to entirely new modes of construction, which creates a high barrier for full scale implementation; or limits the degrees of freedom achievable compared to existing means of realization. In contrast, the method presented here proposes a production cycle, which is fully compatible with current in situ and prefabrication in concrete construction, while achieving doubly curved formwork designs at machining times more than a hundred times faster than comparable CNC-milling, the most developed and applied strategy for industrial scale production. Double curved surfaces with positive Gaussian curvature can in a vast ma-


The Nature of Robots

then, in an appendix to his famous treatise on the calculus of variations found all possible shapes for these so-called Euler elastic.


03 – GEOMETRY RATIONALIZATION // The presented geometry rationalization approximates the physical behavior of the Hot-Blade in order to convert arbitrary input surfaces to producible geometry. The Hot-Blade is fixed between two robot arms, which enable us to choose the location and rotation of the blade’s ends. The shape of the blade is the curve that, subject to the endpoint constraints, minimizes the elastic energy. These curves are the above-mentioned Euler elastica or elastic curves. Before discussing the approximation of a CAD surface, let us consider the class of surfaces defined by this cutting process, namely the surfaces swept out by continuously varying families of planar Euler elastica. A planar curve is geometrically determined by its curvature function. Applying all possible dilations, translations and rotations, one obtains all possible elastic curve segments. Allowing all of these parameters to vary with time, and then generating the time sweep so defined, one obtains all possible elastica-swept surface patches. When rationalizing a CAD surface to Euler-elastica for Hot-Blade cutting, the surface is segmented into patches that can be approximated by surfaces. We essentially do this simply by finding planar curves on the original surface and then approximating these

by segments of planar elastic curves. 03.1 – CURVE APPROXIMATION // Given a parameterized planar curve segment we wish to find a piece of an elastic curve which has the same shape. We do this via an optimization algorithm that minimizes the distance between two curves. By choosing a standard parameterization, we are able to describe any elastic curve segment by four control parameters, which determine the length and shape of the segment. Three more parameters determine the position and rotation of the curve in the plane. The distance between the given curve and any elastic curve is thus a function of the seven control parameters. The approximation algorithm has two steps: first, we analyze the geometry of the given curve in order to find control parameters for an elastic curve segment for an elastic, which has the same overall shape. Then, starting from this initial guess, we tweak the parameters, using the optimization tool IPOPT, until we get the closest fit. We can do this either with or without endpoints fixed. 03.2 – SURFACE APPROXIMATION // we now consider a given CAD surface, and we want to approximate it by a surface that can be obtained by moving elastic curves through space. From the CAD design we extract planar curves on the surface and approximate each of these by an elastic curve. By interpolating the control parameters we obtain a rationalized design-a new surface, which is swept

04 – SURFACE APPROXIMATION // A number of segmentation procedures are developed, targeting three production constraints: (a) plane segmentation when exceeding the dimensions of the input EPS work object; (b) instability of the blade due to multiple inflection points, or (c) cutting the same area multiple times due to rotation of the blade profile. An inflection point is a point where the sign of the curvature changes; in other words the tangent at the point will cut the curve in two. We use a subdivision scheme to find the inflections. Analysis of one of the curves shows six inflection points and since many inflection points on the curve make the blade less stable, segmentation is required. Assuming the rationalization of each cut is curvature the continuous, there win be same number of inflection points on the cut and the rationalization. Two exceptions to this are inflection points near the edge of the cut that may disappear, and pairwise inflections close to each other, which may cancel out, just like pushing out a small dent. Taking the above into account, we propose the following algorithm:

1. Find the planar curves on the surface; 2. Calculate inflection points for each curve; 3. Segment the surface into a grid of blocks; 4. For each block test if there are more than two inflection points; if so try to: a) b) 5.

Move the block if there are overlaps to improve; Remove inflection points close to each other; Take two new brocks, each of the same size as the original block, and place them so that they overlap both each other and the two adjacent blocks in the row. Go to step 4.

In this algorithm we can control whether we keep the same number of blocks in all rows or not. This affects the aesthetics of the segmentation. In the overlap of the blocks we choose a cutting plane such that the segmentation follows the geometry. The problem of cutting the same area multiple times arises when rotation of the blade-in the cutting direction is allowed. We see here that the curves intersect each other, and thus, part of the surface will be cut multiple times, which is undesirable. In most cases this problem can be solved by segmenting the surface, as described above. We only need to add a test for intersecting curves in step 4.

The Nature of Robots

out by elastic curves moving through space. For larger designs we need to segment the surface into pieces that can be cut individually. Because we control the endpoints and directions of the blade, we can ensure smooth transition from one piece to another.


The Nature of Robots 110

05 – DATAFLOW and ROBOTIC SYSTEM CONFIGURATION // The experimental setup consists of three robots. Robot t holds the EPS work object, which is to be cut, and moves the block linearly through space, thus acting in principle like a conveyer belt. Robots 2 and 3 control the ends of the Hot Blade thereby determining its shape and its position in relation to the EPS block. When the geometry rationalization is completed, we know a set of planar elastic curves on the rationalized surface. The curve segments which lie on the surface are shorter than he Hot Blade cutting tool, but since we know not just the curve segments, but the entire curves we can easily extend the curves to the required length, i.e. the length of the Hot Blade. These extended curves are the target shapes for the Hot Blade during the cutting. We extract the relevant data for the extended curves, that is, we find the coordinates for the endpoints and the tangents at the endpoints. The endpoint coordinates determine the position of the tools of robots 2 and 3 relative to the EPS block. The tangents determine the rotation of the tools, which in turn controls the shape of the blade. For our experiments the robots were given 51 targets. That is, for each block that was to be cut, we provided 51 sets of positions and rotations for the tools of robots 2 and 3. The robot program then interpolates between these targets to follow a smooth path from the first to the last target, thus moving the blade while changing its shape, resulting in an EPS surface of

the rationalized design. 06 – BLADE MECHANICS and CUTTING EXPERIMENTS // The main cutting tool used in the process is a thin metal strip-usually referred to as a blade-made of a nickel-chromium super alloy. The blade is pre-heated to a temperature of 300-400 °C by means of Joule heating and then it is slowly brought into contact with an EPS block to produce melting, and subsequently to form or cut the block into a desired shape (also referred to as thermal cutting). At such high operating temperatures, the blade has to be displaced (or bent) into an elastic shape with predefined curvature and at the same time maintain its elastic and flexibility properties. Using FEM simulations, the effect of mechanical properties on the target geometry was investigated and a particular material was chosen to ensure smooth cutting. The blade is attached to two robots, one at each end, by specially designed sandwich based holders to ensure strong and safe supports during all cutting operations. The physical displacement of the blade is achieved by moving the robots into an appropriate position, at the same time maintaining the elastica-strain-curvature relations. The temperature dependent variations of the blade shape are to be incorporated in the computational algorithm to secure proper shape representation. Two experiments were designed and performed in order to test the utility of the setup. In the first experiment a convex doubly curved surface was cut. The curvature of the blade was

07 – FORMWORK SYSTEM and PRODUCTION WORKFLOW // The efforts described in the previous chapters outline the general method for the cost-effective production of doubly curved formwork in Expanded Polystyrene. From this, the following process is developed (Fig. 12): The cyclical workflow links conven-

tional CAD-modelling operations with the robotic Hot Blade fabrication and standard concrete casting techniques. This requires the rationalization and segmentation of geometry types before rebuilding the geometry to the constraints of the blade, robot work envelope, work object dimensions and tolerances. After the input geometry has been translated to segments of swept Euler elastica surfaces and data deducted for tri-robot motion, EPS-mold pieces are produced. The mold pieces are subsequently used in combination with existing pre-fabrication and in situ workflows. For element pre-fabrication, molds are mounted on vibration tables and sides enclosed with metal or wooden frames. For in situ applications, mold pieces are used in combination with standard scaffolding modules for casting pressure support. These applications ensure a full compatibility of the end products of the Hot Blade with established industry workflows, critically ensuring a low barrier to adoption. 08 – CONCLUSION // A general purpose robotic fabrication method for producing doubly curved formwork has been presented. The efficacy of the method has been demonstrated through geometry rationalization and pilot production of a sample formwork panel design. The method is being implemented for industrial scale fabrication by one of partners of the research consortium, and the identified challenges are being addressed through this work.

The Nature of Robots

continually changing during cutting in order to test the limit of complexity that can be achieved and ensure proper geometrical representation. The presence of two inflection points on the discretized surface was considered as a possible problem, but the experiments showed that it does not make the blade unstable, since the robots compensate with the angles of the holders and the curvatures involved were moderate. Good surface quality was achieved at cutting with an absolute speed of motion of 7 mm/s. The EPS block to be cut had the dimensions of 600 x 600 x 600 mm. The second test aimed to cut a number of EPS blocks and then assemble them into a single structure that should represent a ready-made mold for concrete casting. Different discretized pieces of doubly curved surfaces of both convex and concave types, as well as hyperbolic surfaces (negative Gauss curvature), were successfully cut with the setup. The size of each individual block was approx. 600 x 785 x 600 mm, resulting in an assembly of size 1800 x 2345 mm, comparable to the size of production frame molds.



The Nature of Robots

presented in this paper is part of the larger 3-year research effort, ‘BladeRunner’ established and generously supported under the program of tire Innovation Fund Denmark for advanced technology projects. The project is conducted by the partners Odico Aps (project lead), the Technical University of Denmark, Department of Applied Mathematics and Computer Science and Department of Mechanical Engineering, the Danish Institute of Technology; GXN A/S and Confac A/S.



tures with two interconnected layers of cross - laminated engineered wood panels. The shape of the plates and the assembly sequence allow for an attachment without additional connectors or adhesives. The fabrication and assembly constraint based design is achieved through algorithms, which automatically generate the geometry of the parts and the G-code for the fabrication. The project presents the fabrication and assembly of prototypes fabricated with 3D CNC milling and laser cutting systems, comparing and discussing the advantages and disadvantages of the individual techniques.

Credits // Christopher Robeller // Yves Weinand

Integral attachment, the joining of parts through their form rather than additional connectors or adhesives, is a common technique in many industry sectors. Following a renaissance of integral joints for timber frame structures, recent research investigates techniques for the attachment of timber plate structures. This paper introduces double through tenon joints, which allow for - the rapid, precise and fully integral assembly of doubly-curved folded surface struc-

01 – INTRODUCTION // In the design of smooth segmented plate shells, methods such as the Tangent Plane Intersection TPI can be used for the panelization of doubly-curved freeform surfaces. Different methods are required for the design of irregular and freeform folded surface structures. Previous techniques have been presented using and origami inspired techniques using reflection planes and cross-section profiles or mathematical models. In parallel, research is being undertaken for the construction of shells with lightweight and sustainable timber plates using laminated veneer lumber (LVL). The easy machining of wood com-

The Nature of Robots

KEYWORDS // Folded surface structures // Integral Attachment // Laminated Veneer lumber // Miura ori fold // 5-axis CNC // 2D-3D laser cutting //


The Nature of Robots

bined with numerically controlled machines allows for the integration of joints into the geometry of the plates, allowing for a rapid and precise assembly, aesthetic and easy-to-recycle mono-material structures. This paper builds upon previous research in the field and presents a new method that integrates fabrication and assembly constraints specific to folded surface structures built from LVL panels and assembled with integral attachment techniques.


02 – 2-LAYER ASSEMBLY with THROUGH TENON JOINTS // Integral multiple-tab-and-slot joints (MTSJ) such as finger joints provide geometric features for a fast and precise alignment and assembly of the plates, as well as a high resistance to compression and shear, which are the primary forces in segmented and folded timber plate structures. However the joints between the plates receive not only shear, but also traction and bending forces. These forces are typically supported by metal connectors. Alternative solutions are hybrid finger/ screw joints, such as in the ICD/ITKE LaGa Shell or prismatic integral joints such as dovetails, which provide additional features for the assembly and a resistance to bending and traction forces. A comparison of the bending moment resistance of different edgewise joints for laminated veneer lumber (LVL) plates has recently been provided by the authors, including screwed, finger, dovetail, nejiri arigata and through tenon joints. This comparison showed

that the strength of the through tenon joints was the highest, which comes at the cost of a short protrusion beyond the jointed comers. A design constraint of the through tenon joints is their restriction to connections of plates in two planes. A connection of plates in one plane is impossible due to the joint geometry. In consequence, these joints are not applicable to smooth manifolds, however they can be used for the design of folded timber plate structures. In these designs, plates are always connected in two planes, where an orthogonal dihedral angle µ = 90° between the plates is beneficial for the structural performance as well as for the fabrication process. However, a deviation ß from this orthogonal angle is required for the design of curved and irregular shell structures. When using through tenon joints, ∂ is equivalent to the inclination of the cuts, which are required for the fabrication of the joints. Such cuts can be fabricated with multi-axis cuffing machines such as gantry or robot routers or laser cutters, however the inclination ßmax of these machines is limited, which sets a hard fabrication-constraint that must not be exceeded anywhere in the design. Tenon joints have the ability to connect to multiple adjacent plates through intersection. An entirely integral attachment of four plates is possible following the illustrated method. On a mountain fold the lower plate intersects both counterparts with a double through tenon joint. Then, the upper plate is inserted onto the tenons on its

1. A direct connection of the lower layers to the upper layers, without transferring the forces through additional elements such as connectors; 2. Integral spacing of the two layers, which are kept at the correct distance; 3. Blocking of elements: In such an assembly, only the last segment (two plates) can be removed. All other plates are blocked and firmly held in place by other parts, which must be removed before. A disassembly is only possible in the reversed piecewise order of assembly. Therefore, no additional connectors are required to fix the plates. This does not only bring aesthetic advantages, rapid assembly and cost savings, but it also allows the use of thin plates, on which the use of edgewise screwed joints may not be permitted. 03 - SEGMENTATION of DOUBLY-CURVED FOLDED PLATE SHELLS // For the construction of self-supporting, doubly-curved surface structures with discrete plate elements, we must find a segmentation that satisfies the previously mentioned constraints. Figure 1 illustrates this procedure on a target surface with a span of l0 m in the V-direction and a

span-to-rise ratio of 3. In a first step we discretize this surface into quadratic quadrilateral polygon mesh faces. The resulting value ßmean = 85.6° indicates that we cannot join the plates with through tenon joints, because our 3D cutting techniques are limited to ßmax = 45°. Instead, we will use two folding patterns known from Japanese Origami paper folding: Pattern 1 // Yoshimura Fold Pattern, is a triangulation of the previous quadrilateral mesh. The deviation ß is still very large, but can be reduced through a reduction of segments in the V-direction. This results in deformed thin triangles and large plates with Lmax, being larger than ½ of the span of the structure. The assembly of such large parts with integral joints is difficult, because the edges of the plates must be kept parallel during the insertion of the joints. Pattern 2 // Miura-Ori Pattern, presents an alternative approach. While previous Origami-related methods aim at flat-foldable designs, we plan to produce our structure from discrete components and do not include a flat-foldability constraint. Instead, we generate a pattern through the evaluation of a point grid on the parametric base surface, where every second vertex in the U-direction is shifted by a half segment length along V, and every second vertex in the V-direction is raised by the offset length h along the surface normal. With this method, we can reduce the global de-

The Nature of Robots

counterparts like a splice plate, which we call inverse assembly. On a valley fold, the upper plate is inserted with a double-tenon and the lower plate is inserted inversely. Major benefits of this connection include:


The Nature of Robots

viation ßmean = 11° at an offset height of h = 0.75 m, which satisfies the fabrication constraints of our 3D cutting methods. However the quadrilaterals generated with this method are not fully planar. D denotes the closest distance between face diagonals. We reduce Dmean to 0.004 mm in a second step using an external optimization framework, which flattens the faces while it preserves the surface boundary. ßmax increases slightly through this step, which could be reduced through an integration of the dihedral angle constraint into the external optimization framework. However, the solver cannot find a fold pattern that satisfies the dihedral angle constraints without an initialization mesh with the correct mountain and valley folds. Therefore we have chosen our strategy in two separate steps.


04.1 – ASSEMBLY ORDER and JOINT CONFIGURATION // We base our algorithm on a polygon mesh with a uniform sampling of the unit circle. For each plate, up to 2 edges must be joined simultaneously. The common insertion direction v for these edges is found at their bisector. This causes a deviation ∂ from a line on the plane perpendicular to the edge. It is assembly-constrained to ∂max = ± 30°. A piecewise assembly of our structure is only possible as illustrated, with the x-direction changing in every second row (due to the opposite obtuse angles of the faces changing in every second row in the Miura Ori pattern). The tab-and-slot configuration was described, as well as its inversion

based on mountain and valley folds. 04.2 – INSERTION DIRECTIONS // In addition to the two simultaneously assembled edges with outgoing through tenons (V0 and V1 in the positive x-direction and V0 and V3 in the negative x-direction), the tenons on the other two edges connect to inversely inserted incoming parts. Generally, all insertion directions for incoming, inversely assembled parts are determined through a cross product with the face normals of the diagonal neighbours in the direction of assembly: If a row is assembled in the positive x-directiofl, V3 = n0 x n4 and V2 = n0 x n3. In the negative x direction the incoming through tenon directions are V2 = n0 x n2 and V1 = n0 x n1. 04.3 – CONNECTIVITY and BLOCKING // Generally joints in timber constructions are semi-rigid, introducing a certain weakness in the structure. Apart from improving the strength of the joints, it is beneficial to attach each plate to multiple adjacent plates. In a regular single-layer assembly with quadrilaterals, each plate is connected to 4 adjacent plates, in the 2-layer folded structure with through tenon joints, each plate is attached to 8 adjacent plates. 05 – PROTOTYPE FABRICATION // 05.1 – MILLING SYSTEM // The prototype with a span of 3.250 mm, a width of 295 mm and a weight of 82 kg was built from 9-layer birch plywood panels with a thickness of t = 12

05.2 – 3D LASER SYSTEM // For the production of small-scale prototypes of doubly curved shells with medium density fiberboard (MDF) and construction paper, laser cutting proved to be an efficient technique. However, widely available 2D laser systems cannot cut angular slots for our through tenon joints with two rotations ß and ∂. In the automated production of furniture with 3-axis milling machines, non-orthogonal joints are often realized through an increased slot width, which allows the inserted part to rotate to its predetermined rotated position. The contact between the two parts is along the edges of the slot rather than its side faces. We have integrated this

method into the joint generation algorithm. Table I shows the joint processing for one of the segments, 8 edges are being processed on the upper and lower plate. The first four columns show the joint configuration and the assignment of slots to the adjacent plates, followed by the dihedral angle and the joint rotations. From these rotations, as well the thickness t and the offset o, we can calculate corrections for the shortening of the tenon base L tenon, the extension of the slot width Wslot, and the extension of the slot length Lslot. This method allows for the rapid production of precise 3D models based on doubly curved target surfaces. 05.3 – 3D LASER SYSTEM // Due to the contactless operation of laser cutting systems, there is a constant cut quality, no tool wear and the multi-axis robot or gantry system is not exposed to mechanical forces generated by the cutting. The resulting ease-and rapidity of production of this method raises the question whether it could also be used for full-scale applications. Additionally, larger-scale laser systems in the automotive industry are commonly used for the trimming of deep-drawn, curved sheets of hardened steel and therefore capable of 3D simultaneous cutting similar to the 3D milling system used in Sect. 5.1. We have performed tests with such a system, cutting through tenon joints and slots on structural grade spruce LVL panels with a thickness of up to t = 38 mm and a 3D rotation of up

The Nature of Robots

mm. With a span-to-rise-ratio of 9, the arch demonstrates the construction of a shell with a low curvature like in a typical roof structure. The maximum tool inclination for the joint fabrication is reduced to ßmax = 11.5°. The parts were fabricated with a 5-axis gantry router equipped with a 10 kW electro spindle, operated at 16.000 rpm and a feed rate of 5 m/min in 2 vertical infeeds. The G-Code for the fabrication was generated with a custom script, based on a Loft-type synchronization between upper and lower polygon outlines of the plates and a conversion of the 3D tool vector into cardan rotations using an arctangent function with two arguments. All joints and slots were cut without additional gaps or tolerances. The tight fitting pieces were inserted quickly and precisely, the insertion force was applied with a rubber mallet.


The Nature of Robots

to ß = 45° using a gantry machine equipped with a 6 kW CO2 laser. Cutting at a feed rate of 11 m/min (in a single infeed) with N2 gas and 5 kW power, the accuracy of the joints was high and independent from the rotation ß. The cut width of only 0.6 mm allows for thin cuts and small radii on comers in the cuffing contours. However, a disadvantage of the method is the charring and odor of the laser-cut edge surfaces. This can be decreased through higher feed rates but remains noticeable.


05.3 – CONCLUSION // Through tenon joints for LVL panels combine the shear strengh of finger joints with a high resistance to bending moments and out-of-plane traction. The prototypes presented in this paper demonstrate the additional possibility of using double through tenon joints for the integral attachment and spacing of double-layered timber plate structures. The plate configuration based on the Miura-Ori pattem allows for the design of fabrication - and assembly - aware doubly curved folded surface structures. While the Yoshimura pattern is constrained to target surfaces with a high curvature and results in large plate sizes, the Miura-Ori pattern can also be applied to surfaces with a low curvature. However, the vertical elevation of the vertices in the Miura-Ori also results in certain structural disadvantages. Further research is necessary to determine if the structurally advantageous shape of the Yoshimu-

ra pattern outweighs the disadvantages in its fabrication, joining and assembly. Clear advantages of the joint configuration and assembly sequence include the direct connection of each plate to 8 adjacent plates, as well as the mutual blocking of the plates which only allows for a piecewise disassembly in the reverse order of assembly. Therefore, a traction resistance of the joints is not required and additional connectors such as screws, metal plates or adhesives are not necessary. The production of prototypes with 3D milling as well as 2D and 3D laser cutting systems has shown advantages and disadvantages of the individual solutions. The highest quality cuts on LVL plates can be achieved with saw blades, due to the large diameter and the large number of blades. However the production of the concave polygonal contours and slots of the through tenon joints is not possible with such tools. Instead, we have used milling bits with a radius of 6 mm, which allowed for the production of precise parts. The tight fit and precision of the joints was confirmed by a load-test of the arch prototype. An alternative solution was presented for the fabrication of small-scale prototypes using a geometric adaptation of the joints for 2D laser systems. The method allows for the rapid production of precise models, however the plates are only in contact along lines, not surfaces. Further research is required analyzing the influence of this method on load-bearing joints. Finally, the advantages of the 3D milling and the

2D laser cutting were combined using a 3D laser system for the production of through tenon joints on structural grade LVL panels.

The Nature of Robots

ACKNOWLEDGEMENT// The Authors would like to thank Franck Dal-zotto, Anders Holden Deleuran and TRUMPF Laser Technology. This research was supported by the Swiss National Competence Centre in Research (NCCR) Digital Fabrication.


Complex Curved Geometries with Robotically Fabricated Joints

conditions, structural requirements, toolpath development, and fabrication process. The research concludes with a discussion of a new module and joint hybrid informed by stereotomic and timber joint techniques, which takes advantage of the six axis robotic fabrication for a standardized multiple face joint between modules of varying sizes that enables a form and force fitting connection. KEYWORDS // Doubly Curved Geometry // Robotic milling // Joint Connections Project and Practical Application // Computational design to production

The Nature of Robots

Credits // Alexander Jung // Dagmar Reinhardt // Rod Watt


ABSTRACT // This research reports on the robotic fabrication for the complex architectural geometries of three intersecting domes. The project explores systems for modules through a tessellated skin of hexagonal tile modules that produce a macro geometry for a doubly curved, non-developable surface; and the smooth micro geometry of an interpolating structural rib that requires a customized manufacturing of modules and their integrated joints. It outlines the computational workflow between geometrical

01 – INTRODUCTION // Engineers and architects have for centuries shared a passion for relationships between form, force and structure. This is manifest in the complex curved geometries of arches, vaults, hypars, and spheres that determine the mathematical, structural, organizational and material rationale of shell roofs, bridges, cathedrals, or domes. In the legacy of precedents by Frei Otto (1990), Buckminster Fuller (1999), and Felix Candela (1967), this logic of complexity has been set as self-forming structures, and as rule-based geometries. These force active forms differ radically in structural performance and organization of components, and require a logic of parts for material processes that inform point, lines, surface planes and solids. Industrial fabrication and serialization, craftsmanship, details and intersections contribute to complexity here. The project discussed

that would stabilize overall geometry. 02 – COMPLEX CURVED GEOMETRIES:TOWARDS INFINITY // In contrast to vaulting systems, RBDM_ Robodome explores an exemplary series of three intersecting domes with different sphere diameters, skin tessellations and connective rib modules. Domes belong to a family of spherical truss systems, with a diversity of forces distributed as hoop force, meridian force, crown force, edge force, or radial force. Domes are sphere segments and as such contain levels of infinite geometrical symmetry in the pattern repetition of a surface module, and in the degree of sphere curvature for boundary arches that is always the same. The mathematical logic of a dome is thus simple but smart as complexity becomes affordable through repetition. In the following, the system geometry of spheres is introduced, and further evaluated for the affordances of robotic fabrication for a surface tessellation, and structural rib that organizes adjacent surface areas. 2.1 – GEODESIC DOME: ICOSAHEDRON TESSELATION // The project uses a Geodesic dome (Fuller 1999) as geometrical design model for the tessellation into producible segments. As design model, an icosahdron is constructed using three planes in a golden section, where the diagonal length of the planes equals the diameter of a dome. The vertices of the planes define the points for the triangles that will hit the sphere with their four corners, thereby creating twelve

The Nature of Robots

here develops a primary geometrical logic that extends computational modeling and scripting directly towards robotic fabrication of modules and joints. RBDM-Robodome uses the structurally and organizationally efficient geometry of three intersecting spheres to test systems of robotic fabrication for a tessellated tile skin intersected with a series of modular ribs. In a context of current robotic fabrication, complex curved surfaces of domes and vaults posit an interesting challenge for transfers from structural performance towards fabrication. New material production techniques of modular elements and the connection of parts through customized joint systems have been applied to structurally compelling, form-defined or force active constructions that explore robotic applications for complexity, namely, curved structures through robotic deposition of standard modular systems; structural vaults based on optimized segments of RDM Vault and discrete developable surface segments. Solutions for joint systems on the other hand have been designed as customized connections in wood, in a curved folded plate structures, or as interlocking modular joints. Yet two challenges remain; the robotic prototyping of joints in material volumes, and the problem of connectivity for such modules. This is interesting because an extended set of criteria needs to be considered: the material envelope, angle and length of toolbit, cutting path, and six axis defined surfaces have to be synchronized in order to calibrated, planar faces of modules


The Nature of Robots

equally sized triangles. A recursive projection of the midpoints of each triangle side towards the surface of the sphere creates the next frequency, resulting in smaller tiling of the generic hexagon module that constitutes the overall surface when repeated. Within this system, twelve pentagons appear on the tip of the planes. The frequency of triangle divisions can be increased infinitely, with the pentagons decreasing in size, remaining in their original position in the surface, not as part of the rib structure. This system is then extrapolated to the local intersection of two tessellation patterns arriving from two spheres.


2.2 – TWO APPROACHES for ROBOTIC FABRICATION: SKIN or STRUCTURE // The identical degree of curvature in domes allows the customization of segments for the robotic fabrication of a dome. Yet in this particular project, this homogeneity varies between domes. A further challenge came with differentiation of the base geometry into the skin and ribs, which were tested in two system series, and evaluated for affordances of robotic fabrication. The system develops the skin/shells based on tile sizes relative to sphere dimension and curvature, as a sheet that is efficiently formed over a milled plaster mold with varying radius. System 2 develops intersecting ribs that follow an intersecting curve between spheres, and bridge two tiles arriving from each side, and with a focus on segmentation of the rib into modules that can be robotically milled

from a volume. Both approaches were tested: several prototypical moulds of system 1 were robotically milled into plaster with a KUKA KR 60-3 industrial robot (radius R = 1700, R = 1950 and R = 2300 mm), using a 4KW milling spindle with 6 mm toolbit. Onto these moulds, surface tiles of 5 mm perspex sheets were air-suction formed, and assembled to matching faces into a compressive joint surface system. 2.3 – GEOMETRY RULES for STRUCTURAL RIBS // In contrast to the skin, the structural ribs had to extend the pattern synchronization between dome tessellations arriving from two sides, towards carrying a structural load at the intersection. This required an increase in complexity for geometry, and change in robotic fabrication method from sheet logic to subtractive process. Consequently, a number of different scripts were modeled in McNeel Rhino and Grasshopper (scripting plugin) to link criteria for tessellation pattern, rib curvature, modules and joints (Fig. a). The intersecting geometry of two spheres results in an inclined circle with a center point that anchors the geometry of the rib. Modules are segmented relative to material size (Fig. 4b). Each side of the rib follows the custom ‘mother geometry’ (center point) of its sphere, with a tiling run in Weaverbird (GH plugin for mesh associations) of 12 pentagons, and a variable of hexagon-shapes. All sides in the modules are planar,

03 – ROBOTIC FABRICATION of STRUCTURAL RIBS and JOINTS // RBDM then developed the computational modeling of geometry rules towards six-axis robotic milling, whereby the overall geometry is linked to a standardization of producible parts adequate to material properties and performance. The prototype paralleled aspects of serialization and customization; firstly, for a structurally effective rib system able to lock into the complex doubly shaped surface and its hexagonal tiling system; and sec-

ondly, a constructive detailing of compressive joints and a tension canal that are integrated for construction. 03.1 – WORKFLOW and ROBOTIC FABRICATION of RIB STRUCTURE // The structural system for the exemplary module series was further explored in a robotic simulation with KUKAlprc in order to adjust the size of the intersecting tiles by rotation, so that enough material remains. This data set contains a series of customized scripts for SRF rough and fine surface milling of top and bottom surfaces for each particular surface angle, facets and finishes. For a first material test, the robotic milling followed industry customs for volume milling (as in sandstone or wood) with support of added feet that allow steady positioning on the routing bed and precise turnover of the material sample. Modules were then robotically milled with a KUKA KR 60-3 industrial robot, using a standard flat headed 4KW milling spindle with 10 mm toolbit and 3 mm stepover, in a series of robotic protocols that require multiple manual turnovers but adequately present material behaviour of wood or stone, with fabrication axis angles relative to robotic axis deployment. A canal is drilled through each module at center of the joint to allow for insertion of a tension cable. Modular components are unique along each rib and respond to force-flow changes where the structure acts in compression, resulting in thicker sections (added to top of load distributed along center line of struc-

The Nature of Robots

while connecting surfaces, bottom and top surfaces are doubly curved. The resulting intersecting modules are used to generate the outline for a rib that combines shape information of two different patterns from each side. The rib is defined through the degree of surface curvature from two spheres, and through planar sides that connect to the skin. Divisions between the modules are generated through the orientation of intersecting faces of the edge conditions of valley and ridge. Into this section plane, a second mother geometry for the joint is inserted that is always the same, but varies position. As a result of this geometry definition, each modular component features a number of criteria that are continuous within the sphere; the sphere diameter; the continuation of interior arch curvature; the length of side edges that connect to skin patterns on each side; and the geometry of joint connection between a lower and upper module.


The Nature of Robots

ture) towards the ground.


03.2 – ROBOTIC FABRICATION of JOINTS // Accompanying the robotic fabrication of skin plates and the structural ribs, a third and novel part of robotic fabrication is finally discussed: each module also contains a three-dimensional joint that is inserted into the section plane between segments. These joints are multifaceted elements that are developed as a modification of traditional japanese wood connections, which go beyond simple finger joint systems, such as the three-faced halved rabbeted oblique scarf splice. RBDM uses a similar variation of a male-and-female joint, constructed here as a multiple of a 90° angle. This angle serves multiple purposes: the joint is embedded as geometric information into both the scripting and robotic process, and capitalizes on an industrial logic. Instead of many different connections, the same precise mother geometry is maintained while each particular module can differ in direction and orientation along the dividing surfaces. The triangulated connection also maximizes the surface contact between two modules; prevents horizontal movement; and provides structural efficiency. 03.3 – FROM GEOMETRY to STRUCTUTAL BEHAVIOUR // This allows the structure to work in compression as all vaults or domes, yet ensures precise construction as modules are connected as a compression unit that transfers loads to the ground, with the

triangulated joints counteracting lateral shifts. Through the combination of a geometric logic coupled with robotic fabrication, structures can be produced as sequences of equals (skin) and individual parts (ribs) that can be fabricated effectively, in series, to contribute to complexity. In continuation, the research will develop the robotic protocol to integrate the doubly curved geometries for an optimization in subtractive rough milling of stacked rib-modules, taking into account the shared surface degrees in one sphere that allows elements to be nested into each other. These butterfly modules can then be rough cut as stacked series in order to reduce material waste. Through the combination of a geometric logic coupled with robotic fabrication, structures can be produced as sequences of equals such as the tessellated skin, and individual parts such as the structural ribs that can be fabricated singularly, in series, and in true materials. 04 – CONCLUSION // This research has explored the robotic fabrication of three intersected domes: based on a hexagonal pattern structure that is differentiated into serial surface elements, a customized modular structure and integration of new joint system that combines aspect of timber construction with a stereotomic process. In doing so, the research project has demonstrated that a serial production of doubly curved surfaces both for surface and solid elements was achievable through robotic fab-

ACKNOWLEDGEMENT // This research is an ongoing initiative (2014) that has been informed by the SmartStructuresLab (2014-2015), and with pans of the research (Skin) developed in the robotic elective CodeToProduction (2015). The authors would like to express thanks to Gabriele Ulacco for the research elective, and the student team for their engagement. The research has been generously supported by The Faculty of Architecture, Design and planning, The university ofSydney, through a SEED Grant, and produced at DMaF. The authors would further like to express thanks to Marjo Niemelä for continued support.

The Nature of Robots

rication. Robotic fabrication was also applied to equip the material with a high level of joint detail, thereby seamlessly bridging between design process and fabrication, and furthermore incorporating construction and structural performance. This research project is currently developed for a 1:2 prototype. Knowledge about geometrical logic, material and fabrication process can then enable structures that are complex but geometrically smart, producible at affordable cost, with low material waste, and with close references to industrial cutting processes. In sum, the application of robotic processes allowed us to reconsider engineering precedents, and to reformulate this into a novel architectural system.



is established, in which optimization data is driving subsequent processes solving timber joint intersections, robotically controlling member prefabrication, and spatial robotic assembly of the optimized timber structures. The implication of this is studied through pilot fabrication and load-testing of a full scale prototype structure. KEYWORDS // Topology optimization // Digital fabrication // Architectural robotics // Advanced timber structures //

The Nature of Robots

Credits // Asbjørn Søndergaard // Oded Amir // Phillip Eversmarnn // Luka Piskorec // Florin Stan // Fabio Gramazio // Matthias Kohler //


ABSTRACT // This research presents a novel method for integrated topology optimization and fabrication of advanced timber space-frame structures. The method, developed in research collaboration between ETH Zürich, Aarhus School of Architecture and Israel Institute of technology, entails the coupling of truss-based topology optimization with digital procedures for rationalization and robotic assembly of bespoke timber members, through a procedural, cross-application workflow. Through this, a direct chaining of optimization and robotic fabrication

01 – INTRODUCTION // Topology optimization may broadly be defined as a family of procedures aimed at creating efficient structural layouts. In the form of continuum representations, this equates the redistribution of material within a Finite Element-discretized design space. In the form of truss-based representations, it equates the determining of the topological connections and cross section sizes from a pre-defined set of possible members. In a preceding research work, continuum optimization was explored for architectural concrete structures. While these studies successfully indicated significant potentials for design innovation and reduction of material consumptions compared to commonly found standard structures, the work also found an inherent complexity in translation from optimization result to construction design. Furthermore, current continuum procedures are not directly applicable to the majority of construction projects, which are realized through assembly of prefabricated semi-manufactures and components,

02 – STATE-of-the-ART // Recent developments within architectural robotics have presented novel procedures for digital fabrication of advanced timber structures. In the seminal experiments conducted at ETH Zürich (Gramazio and Kohler Research), integrated robotic fabrication and assembly have been demonstrated within layered assembly of timber structures. This process is currently being applied for large scale production of the 80 x 225 mt Arc-Tech-Lab roof

structure under construction at the ETH Hönggerberg University Campus. Furthermore, recent experiments at the ICD Stuttgart have demonstrated full scale fabrication of plated timber structures in combination with manual assembly through robotic CNC-milling of bespoke elements for the Research Pavilion 2011 while long-threaded developments at the EPFL Lausanne are investigating digital structural design and fabrication of new timber structures. Most recently, research was undertaken at ETH Zürich explore the potential of robotic assembly of single-joint spatial structures in combination with application of fast-curing, 2-component chemical binder. The collaborative work presented in this paper builds on this development, while process parameters have been extended to enable the realization of topology optimized structures. 03 – OPTIMIZATION of TIMBER STRUCTURE // The structure fabricated within the current study is essentially a rationalization of a result of a topology and sizing optimization procedure. We rely on well-established formulations from the field of structural optimization, where the purpose is to find the optimal structural layout of a truss (locations of existing members) as well as the optimal cross-section areas. Such layouts have been investigated since the early 20th century stemming from Michell’s classical work on least-weight gridlike continua.

The Nature of Robots

necessitating alternative modes of optimization. These limitations can be conceptually addressed through the application of truss-based topology optimization. This approach enables the optimization of pre-defined member-and connection types within predetermined ranges of cross-sections, hereby facilitating the generation of optimized designs, which align closer with current building and construction practice. However, the topological complexity of the optimization results derived from such processes necessitates digital new means of pre-fabrication and assembly to become practical to full scale building implementation. As of today, no method exists for direct realization of optimization result, in which the complex challenges in prefabrication and assembly arising from the complexity of the optimized topologies are handled in an integrated, digital process. The collaborative research presented in this paper addresses these challenges targeted at the special application area of digital manufacturing of timber structures.


The Nature of Robots 128

03.1 – IMPLEMENTATION of OPTIMIZATION USING OPTIMALITY CRITERIA APPROACH // The current implementation uses the so-called ‘ground structure’ approach, where for example the design domain is discretized using a fixed set of nodal points, which are then connected by a set of potential truss bars. The only requirement in setting the ground structure is that it should be able to transfer the loads to the points of supports without forming a mechanism. Then, the objective of topology and sizing optimization is to determine the optimal topology and cross-section areas of all potential bars, including eliminating unnecessary bars by assigning them a zero cross-section area. In its most basic form, the topology optimization procedure aims at “finding the stiffest truss”. This can be stated as follows: find the structural topology and cross-section areas, so that stiffness is maximized (i.e. external work is minimized), subject to an upper limit on the volume of material used, and provided that structural equilibrium can be satisfied. This implementation was applied in the development of prototype designs for the purpose of testing rationalization and fabrication methods. The prototype design was reached by optimization from 1711 possible connections in an irregular, trapezoid ground structure with 5x5x5 subdivisions in the XYZ directions. The configuration was fully supported on three nodes at points (1,1,1; 1,5,1 and 5,1,1) and eccentrically loaded with 5 kN single point load at (3,2,5).

Optimized for a minimization of compliance under a volume constraint of 0.14 m3, the processes resulted in a geometrically complex 34 bar structure. 04 – GEOMETRY RATIONALIZATION // The optimization procedure described in the previous sections accounts for the structural load-capacity of the topological configuration, and the dimensioning of member cross-sections. However, the output does not solve the geometrical intersection of members at node levels, which must be processed in a secondary step. To accommodate for this, a rationalization procedure is developed and implemented in GhPython. The objective of this procedure is to, given any topological structure, solve the necessary cutting sequence and orientation for bars in each node, based on the limitations of the robotically controlled sawing process. The outcome is a discrete list of revised bar geometries, which avoid in-node overlaps, while ensuring structural continuity from load-points to points of support, and minimizing the number of necessary intersections while maximizing the contact surface area at member joints. Members are discretized into ranges of pre-determined cross-section dimensions, and joints rationalized according to the member cross-section dimensions included in the joint joints containing members of only one dimension type are trimmed against a shared plane derived from the bi-sector of the center axis of the intersect-

05 – ASSEMBLY PROCEDURE // A predominant challenge for robotic fabrication of topology optimized space-frames is the auto-generation of valid assembly motion sequences, which must determine the chronological order of member insertion while avoiding collision with the structure under construction. Addressing this challenge an assembly processor is developed, which computes the assembly sequence and the respective trajectories directly from the node geometries, hereby avoiding simultaneously in-node collision at joint level and global collision at the structural level. This is conducted in an operation, in which members and nodes are sorted according to their distance to the robot base, and the bar with the smallest angle relative to the base-plane is selected. Once the first valid bar is found, insertion trajectories are computed by the sum vector of the normals of the contact faces of neighboring members within the same joint, defining collision avoidance as any trajectory which is >90° to any normals of the neighboring contact faces. If collision is found,

and cannot be solved through incremental search for alternative trajectories in the trajectory solution space, a combinatorial search is performed for the insertion sequence with least collisions; the obstacle member is retracted and validity is re-checked after every insertion of a new node. For every insertion operation, a connectivity check is performed at the end-node of each inserted bar member (opposite of the joint node). The connecting bar will be inserted ensuring, where possible, a build-up through triangulation, which help to ensure physical stability during assembly. 06 – ROBOTIC OPERATIONS // The fabrication setup at ETH Zürich consisted of a KR 150 L110-2 KUKA robot on a 7 m linear axis and a Mafell Erika 85 circular table saw. A custom positioning table for material feeding was added to the saw. The robot is equipped with a custom parallel gripper, which is capable of holding the beams stiff enough during the cutting process. Within the robotic process, the following steps were repeated for every beam: first, a wooden beam was gripped, and then positioned in 5 axes for cutting. Then, the positioning and cutting procedures were then repeated for all cutting planes. Finally, the robot could reach the final assembly position, where multiple beams can be prepared for gluing. Each step of the geometric constraints, the robotic movement and the assembly is explained in more detail in the following paragraphs.

The Nature of Robots

ing members. For joints containing members of several dimension types, largest dimension types are trimmed per bisector as previously described, while smaller dimension types are trimmed against the cross-section profiles of larger members. The result of this operation is that lower level bars will share a surface only with one higher level member, leading to simplified joining faces in this situation.


The Nature of Robots 130

06.1 – GEOMETRIC CONSTRAINTS // The idea of the cutting operations was to perform every cut with a vertical sawblade, while the robot could perform all complex geometric orientations. The orientations can be computed through a transformation matrix, which aligns the trimming plane vertically, while keeping on of the edges of the beam horizontal. This approach allows for a wide range of possible cutting angles. The positioning angle (in the XY plane) depends on the specific length of each beam, the distance to the tool and the saws maximal pulling distance. The cut angle (in the vertical plane) depends on the size of the gripper and parts of the end effector, which could cause collisions. A threshold angle of 60° was identified for the current setup. Building on previous research (NRp66 research projects, 2012-17), joints where connected using a fast-curing, two-component adhesive with curing times of 5-10 s. Each connecting face was perforated to allow for the adhesive to permeate deeper into the structure, hereby increasing resistance to tensile stress. Due to the high viscosity of the adhesive, gaps between connecting faces were sealed with tape during injection, to avoid leakage during the short-term curing. This approach allowed for accommodating tolerances between 2-10 mm, while ensuring strong connections.

06.2 – ROBOTIC MOVEMENT // A number of challenges were encountered in the programming of the toolpaths for robotic movements. The geometric operations and toolpath data was computed in GhPython. For the simulation and post-processing the software HAL was used. For angles beyond the threshold angle a regripping procedure wascreated. During this procedure, the robot places the beam on the table vertically and regrips it afterwards at a 90° angle. Due to safety reasons, beams were rotated horizontally in a safety-plane above the saw-blade. Therefore, the robots joint had to perform most of the movements. The rotational limits were easily reached. This issue was solved with using joint-movements to a custom unwinding position between each of the cuts, where the joint of the 6th axis can rotate in interpolated movements to a zero value. This created some additional movements, but provided a safe position for the motion-planning. It helped in avoiding collisions between very long beams and the robot during the cutting and positioning. Custom positioning of the robot base was used to allow the robotic rotations of the beams, which outer comers reached beyond the linear axis. For the positioning of the robot base three cases needed to be considered. Ideally, the base moves along a position normal to the current target plane. This works only for rotations along the maximal offset domain of the robot. Therefore repositioning of the base can be anticipated and performed

06.3 – ASSEMBLY TRAJECTORY // For the assembly motion planning a slightly curved reference surface was created, slightly hovering over the planned structure. The assembly direction planes derived from the assembly processor described in Chap. 6 were then pulled in the normal direction on the surface. This allowed for an easy control of the trajectories since the upper paths are projected above the structure, whereas the lower paths remain in ideal safe regions around the structure. 07 – ANALYSYS and LOAD TEST // The fabricated timber truss was finally tested under point loading in order to validate the effectiveness of the overall design and fabrication process. Since the performance of timber structures is determined by the capacity of the joints, a key question was the structural capacity of the glued connections. In the numerical simulations, both in MATLAB (within the optimization procedure) as well as in RSTAB, very small displacements were predicted under a load of 5 kN. Furthermore, the difference in displacement between the optimized design (with variable cross-sections) and the fabricated design (with three bar types only) was under 10%. At the

moment of writing, load-testing went to 13.6 kN, but failed to proceed to collapse due to rupture of the connecting metal braces. While this limits the measurement of the actual stiffness of the structure, it nevertheless indicates that joints - despite geometrical complexity and variability of gluing conditions-perform overall within expected range and that the prototype indeed is very stiff due to its optimized configuration. 07 – CONCLUSIONS and OUTLOOK // This paper has presented a process that facilitates integrated optimization, production rationalization, robotic fabrication and assembly of topology optimized space-frame structures. The method discussed presents a solution for the production of spatial structures with a high concentration of bars at individual nodes, the implication of this is demonstrated through optimization, fabrication and load-testing of a full scale structure. Analysis and tests show general consistency between predicted capacities in optimization, the analysis of the rationalized geometry and the performance during physical testing. While the presented work demonstrates the feasibility of the proposed process, a number of challenges were identified for further work. The high level of complexity of all steps of the described process necessitates either fully automated or highly automated construction processes to remain feasible in full scale architectural applications. This implies in particular that custom adaptive/feedback-based

The Nature of Robots

during a safe position as the unwinding position. Inward rotations can be performed with the maximum offset depending of the reach of the robot. Therefore a pattern for all base positions had to be calculated in advance for all toolpath targets.


processes are needed in the future for handling un-modeled material effects (in particular gravitational sagging of the structure during assembly and tolerances stemming from member warping during cutting) and therefore the development of novel, fully integrated design and fabrication workflows/tools are required. Finally, automation of the sealing process as presented is challenged by the high degree of joint complexity. While manually solvable, robotic automation would be key to improving the industrial applicability of the process.

The Nature of Robots



presented in this paper was performed within a research exchange between ETH ZĂźrich and Aarhus School of Architecture in collaboration with the NCCR Digital Fabrication MAS Programme and Israel Institute of Technology, Haifa. This research was supported by the NCCR Digital Fabrication, funded by the Swiss National Science Foundation (NCCR Digital Fabrication Agreement #51NF40-141853). The contribution of Aarhus School of Architecture was enabled through the generous financial support of the Danish Ministry of Higher Education and Science under the Elite Research Travel Grant Program. The presented research builds directly on the research findings and developments from the ongoing SNSF research project NRP-66, established in collaboration between ETH ZĂźrich, Bern University of Applied Science and Nolax AG. The primary constituent findings for the presented work are ongoing joining experiments involving two-component, super-curing adhesives and the principal process of robotic pre-sawing and spatial assembly of timber members. In particular, the authors would warmly like to thank: Dr. Volker Helm and Dr. Jan Willmann

for their helpful organizational support and discussion of research and paper content; NRP-66 collaborators Dr. Thomas Kohlhammer, Aleksandra Apolinarska and Peter Zock for fruitful discussions of analytical and structural approaches, knowledge transfer and help regarding the adhesive process; Student assistants Micha Ringer and Lazlo Blaser for their involvement in the fabrication of the prototype structure; Michael Lyrenmann for excellent photographic documentation; and Dominik Werne and the ETH HIF-Halle staff for their tireless involvement and support in the load-testing of the structure.

Automation of a Discrete Robotic Fabrication Process Using an Autonomous Mobile Robot

Credits // Kathrin DĂśrfler // Timothy Sandy // Markus Giftthaler // Fabio Gramazio // Matthias Kohler // Jonas Buchli // ABSTRACT // This paper describes the implementation of a discrete in situ construction process using a location-aware mobile robot. An undulating dry brick wall is semi-autonomously fabricated in a laboratory environment set up to mimic a con-

struction site. On the basis of this experiment, the following generic functionalities of the mobile robot and its developed software for mobile in situ robotic construction are presented: (1) its localization capabilities using solely on-board sensor equipment and computing, (2) its capability to assemble building components accurately in space, including the ability to align the structure with existing components on site, and (3) the adaptability of computational models to dimensional tolerances as well as to process-related uncertainties during construction. As such, this research advances additive non-standard fabrication technology and fosters new forms of flexible, adaptable and robust building strategies for the final assembly of building components directly on construction sites. While this paper highlights the challenges of the current state of research and experimentation, it also provides an outlook to the implications for future robotic construction and the new possibilities the proposed approaches open up: the high-accuracy fabrication of large-scale building structures outside of structured factory settings, which could radically expand the application space of automated building construction in architecture. KEYWORDS // In Situ Robotic Construction // Mobile Robotic Fabrication // Adaptive Fabrication // Robot Localization // 01 – INTRODUCTION // The degree of automation in the construction industry is constantly rising, particular-

The Nature of Robots



The Nature of Robots 134

ly in the area of pre-fabrication. On construction sites, however, the level of automation is rather low, and final assembly tasks of building components predominantly imply the use of manual labor. This is a fundamental difference to other industries (e.g. the automotive industry), where the entire process from production of single parts to final assembly is often fully automated. Therefore, robotic in situ fabrication – performed directly on the construction site -holds the potential to finally close the digital process chain between design and making and to leverage novel aesthetic and functional potentials in the field of non-standard architectural construction. However, the inherent characteristics of construction sites substantially differ from those in factory environments, which makes the implementation of in situ fabrication tasks significantly more difficult. Building sites are generally considered unstructured because they are gradually evolving and continuously changing shape during construction, floors are not necessarily flat and there is no guarantee for regular structures in the surroundings, as opposed to prevalent constant conditions in industrial production. Additionally, robots for pre-fabrication are commonly employed at an anchored position within a work cell and work pieces are brought to the stationary unit. Yet, to enable the fabrication of large-scale building structure that exceeds the workspace of a fixed robot, the employment of robots on constructions sites requires them to be mobile. Robots need to be able

to travel to the place of production and to move during construction, while still being able to localize themselves with respect to the working environment and fabricate structures accurately in space. To take on these challenges, the two ETH Zürich groups Gramazio Kohler, Research2 and the Agile & Dexterous Robotics Lab3 are developing an autonomous area-aware mobile robot, called the ‘In situ Fabricator’ (IF). Following its predecessor ‘dimRob’, described in the next section, IF consists of an industrial robotic arm mounted on a base driven by hydraulic crawler tracks. It is intended as a generic mobile fabrication robot for the future employment on construction sites. This paper presents a first physical construction experiment using IF: the fabrication of an undulated dry-stacked brick wall, made up of discrete production steps, in a laboratory environment set up to mimic a construction site. The experiment serves to demonstrate the robot’s generic functionalities and system architecture, as well as its integrated digital design and control software framework. In this context, objects of detailed investigation are (a) the automated fitting of the geometric description of key features of building site components (e.g. floor, walls, pillars) to captured laser range measurements made by the robot, (b) the precise robot localization using point cloud registration, and (c) the adaptability of a parameterized brick wall’s geometric description and its corresponding assembly sequences to

02 – CONTEXT // Concepts and exploratory setups to employ industrial robotic units for automated in situ fabrication tasks have been explored since the 1980s and 1990s, the most advanced of them being the mobile bricklaying robots ROCCO and BRONCO. These early concepts, however, are characterized by heavy duty machinery and rigidly planned production routines. As a result, assembly procedures largely depend upon uniform, standardized building elements, standardized connections, strictly organized fabrication routines and well controlled environments. In the last decade, however, robots have evolved through new developments in sensing, real-time computation and communication, within which inflexible top-down organization principles are replaced by flexible and adaptive bottom-up approaches. These advancements allow also for their customization as advanced design and construction tools. In 2010, the Gramazio Kohler Research group, together with the industrial partner Bachmann Engineering AG, developed and built IF’s predecessor, the mobile platform dimRob. It consisted of an ABB IRB 4600 industrial robot arm mounted on a tracked mobile base. Its hydraulic drive system was powered by a diesel engine and the system was steered manually using hydraulic levers. While dimRob already successfully demonstrated core concepts for in situ fabrication on

the basis of a variety of experiments its applicability was limited by a few key aspects. The original design of dimRob lacked the sensing required to allow the robot to build with high accuracy without being anchored to the ground using fold-out legs. This made it infeasible to build structures that would require the robot to move many times during construction. Also, dimRob had to be repositioned manually. This not only required substantial human intervention, but also placed a limit on the precision with which the robot could be repositioned. Finally, the robot arm was powered and controlled by a control box, which was not integrated into the robotic system, which significantly limited the autonomous capabilities of the overall setup. This motivated a major revision to drastically extend its capabilities. The result is IF, the ‘In Situ Fabricator’, whose main features are described in the following sections. 03 – IF SETUP // 03.1 – IN SITU FABRICATOR SYSTEM ARCHITECTURE // IF is designed such that it can autonomously complete building tasks directly on a construction site. The level of autonomy intended for the robot is defined to contain all of the facilities required for precise manipulation of building materials. In this way, human interaction with the robot is narrowed down to the specification of building tasks through high-level planning environments and dedicated interfaces. In order to achieve this, the robot is designed to be self-contained, with

The Nature of Robots

process-related parameters during construction.


The Nature of Robots

all components needed for construction on-board: mainly sensing, control hardware, and computing systems. A dependence on excessive setup of the construction site for building is also avoided. For this reason, the robot is designed such that it should not depend on external referencing systems.


03.2 – HARDWARE OVERVIEW // IF features the same robot arm as its predecessor, but additionally it carries a complete, retro-fitted ABB IRC5 industrial controller. The whole system is electrically powered by lithium-ion batteries, which enable it to operate for 34 h without being plugged in. The robot’s hydraulically driven tracks can still be controlled manually, but are predominantly operated in automatic mode where the tracks are steered precisely using an on-board control system. The drive system can achieve a maximum speed of 5 km/h on flat terrain at a total robot weight of 1.4 tons. IF’s on board computer runs a real-time enabled version of Linux (Xenomai), which allows for hard real-time data acquisition and processing, along with the robot operating system (Ros). For the experiment described in this research, IF was equipped with a Hokuyo UTM-30LXEW laser range finder mounted on the arm’s end-effector and an Xsens inertial measurement unit, attached to the robot’s base frame. Additionally, the robot was equipped with a vacuum gripper to pick and place bricks, and a brick feeder on its back, which can carry 6 bricks at a time and has to be

manually filled. 03.3 – COMPUTER ARCHITECTURE and COMMUNICATION // The high level planning of fabrication tasks, such as the sequencing of the robot’s positions and brick laying procedures, and computing the arm and gripper commands, is implemented within the architectural planning tool Grasshopper Rhinoceros. A custom TCP/IP implementation allows the online control of the robot’s arm and base. Commands are sent through a GhPython interface within Grasshopper to the robot’s ROS nodes for base movement, as well as to the ABB Robot Control Software for arm manipulation procedures. In return, all state and sensor data needed within the high level planning tool before and during construction is received within Grasshopper. Generally speaking, the robot’s setup allows for feedback loops at multiple levels of the system. All time-sensitive tasks are executed by control loops running on the robot’s low-level computer and the ABB controller, to control base and arm motion, respectively. The control of the overall building process, which is much less time-sensitive, is closed via the architectural planning tool. 04 – EXPERIMENT // This section details an initial experiment performed with IF, in which a dry stacked double-leaf brick wall is constructed in between two pillars. The material system consisting of discrete building elements and simple assembly logics-is specifically chosen in order

04.1 – ADAPTIVE BUILDING PROCESS // The building process begins once IF is moved to the construction site. (Note that in this initial experiment, the robot was positioned manually via a remote controller. While the robot has all of the sensing and computing capabilities for autonomous navigation on-board, the development of the autonomous navigation capabilities required is left as future work). At this time, it takes a 3D scan of its surroundings, which serves as a reference scan for the robot’s localization in space. Additionally, this scan is used to locate the true positions of key features of the working environment. These key features identify the interfaces to which the structure being built must attach (Fig. 3). This information is then fed back to the architectural planning tool (Grasshopper), and is integrated as a parameter into the generation of the wall’s geometric description. Since the true dimensions of the construction site generally deviate significantly

from the ideal dimensions of building plans, it is important to consider these inaccuracies before starting the construction. 04.2 – ON-BOARD POSE ESTIMATION // In order to build with high accuracy on the construction site, the robot needs to be aware of its position with respect to its work-piece. Because one goal of IF is to avoid dependence on external sensing systems, this means that the robot must be able to localize itself in its surroundings using on-board sensing and computing. For this experiment, the primary sensor used for localization is a laser-range-finder, mounted on the end-effector of the robot’s arm. By executing sweeping motions with the arm, 3D scans of the robot’s environment are generated. Point cloud registration is then used to find the relative transformation from the current robot position to the reference robot position. Non-linear least squares optimization performed using Google’s Ceres Solver (Sameer 2015) is used to find the relative transformation required to minimize a measure of point cloud quality between the measurement and reference point clouds. This registration method requires no reference markers to be placed on the construction site a priori, makes no assumptions about the structure of the robot’s surroundings, and is not severely impacted by objects that move within the site during building. For these reasons, this method should generalize to a wide variety of construction environments. Initial ex-

The Nature of Robots

to be able to solve basic problems of adaptive control strategies, construction sequencing and repositioning operations of IF, while still being able to subdivide the sequential building process into discrete production steps. (Note that while a more elaborate hardware setup could have been employed to use adhesive in between the bricks, or also to avoid the manual placement of bricks in the feeder on the robot, this was not done because these tasks did not fit the main goals of the experiment.)


The Nature of Robots

periments showed that, with only static loading, the robot’s base tilts up to 2° when the arm reaches far from the base. This can result in end effector positioning errors of up to 70 mm. To compensate for this base tilting, an inertial measurement unit is used to continuously measure the orientation of the base. This information is then used to adjust the target end effector position while the arm is reaching to place a brick.


04.3 – EXPERIMENTAL RESULTS and VALIDATION // The designed double-leaf brick wall was successfully constructed using the process described herein, requiring the robot to be repositioned 14 times. An average duration of 40 s for the picking and placing of a single brick was observed. While in the scope of the experiment it was not possible to directly measure the localization accuracy and the consistency between the final built structure and the initial design, the accuracy of the system was evaluated by measuring the position of placed bricks relative to previously placed neighboring bricks. These relative measurements were consistently within 3 mm of the value expected from the CAD model. It was also observed that the bricks placed closest to the pillar opposite from where the robot started building were within 7 mm of their expected position relative to that pillar. This indicated that global localization and brick placement errors did not accumulate over the course of the building process, since first, the pillar locations were

only determined and considered before any bricks were placed, and second, every point cloud captured from a new location was always registered against the same initial reference scan. IF was therefore successful in building a structure which was aligned to existing features of the construction site using solely on-board sensing and computation. It is important to note that, in the experiment presented, a specific production sequence was not defined beforehand, but derived from the resulting locations of the robot during construction: while the human operator navigated the robot to an arbitrary location, the machine then identified and reacted to the resulting location and continued building with no further human interaction. While the automated navigation and optimized production sequencing of the robot is left for future development, the chosen strategy demonstrated a successful integration of human intervention and automated construction. 05 – CONCLUSION and FUTURE CHALLENGES // The experiment presented in this research demonstrates a significant step towards enabling the robotic construction of complex structures directly on the construction site with minimal human intervention. As mentioned in the previous sections, the continuous exchange of information between true measurements and the underlying computational model allows for the compensation of material and process related inaccuracies during fabrication. With respect to the mobility of the machinery, production

by using a mobile robot for fabrication need to be investigated-not only in the context of their functional, but also in their aesthetic capacities.

ACKNOWLEDGEMENT // This research was supported by Swiss National Science Foundation through the NCCR Digital Fabrication (NCCR Digital Fabrication Agreement #5lNF40-141853) and a Professorship Award to Jonas Buchli (Agreement #PP00P2_138920). The building material was sponsored by Keller AG Ziegeleien. Special thanks also go to the lead technician of NCCR Digital Fabrication and photographer Michael Lyrenmann, as well as the project leader of IF’s predecessor dimRob, Dr. Volker Helm.

The Nature of Robots

sequences can radically be redefined, which allow for the construction of continuous structures. These structures don’t have to be discretized into separate building components due to constraints prevalent in pre-fabrication, but rather have to be redefined in accordance with the fabrication logics of the chosen material system, the mobile machinery and conditions on site. Eventually, this will demand novel mobile robotic building strategies, not only to realize complex design propositions directly on construction sites, but also to enable design processes, whose formal language and constructive details comply with the fabrication logic of the respective machinery used. Future research into in situ construction methodologies using IF will be focused on moving towards a more fully integrated and continuous construction process, aiming at simultaneous arm and track maneuvers and continuous location-aware manipulation procedures. In this experiment, the robot base was driven and repositioned manually while the industrial robot arm was controlled from within Grasshopper. In a next step, these separate processes need to be unified in a whole-body control framework that allows to plan optimal, simultaneous base- and arm motions. This will then open up the possibility to address open questions like optimal building sequencing in terms of required energy or overall building time. Finally, formal influences in the design vocabulary through structural and process-related boundary conditions



The Nature of Robots

Credits // Roland Snooks // Gwyllim Jahn //


ABSTRACT // This research demonstrates the effect of feedback between algorithmic, robotic and material behaviors on the emergent formal character of several recent design projects. These projects demonstrate a progression from single step linear feedback between fabrication and simulation constraints to the attribution of new material agency through real-time and recursive feedback between multi-agent behaviors and physical material. We present a prototype robotic control system and methodology that allows design to take place in and on an object rather than in its anticipation, and we speculate on the implications for generative

design and robotic fabrication. KEYWORDS // Robotic Fabrication // Multi-agent algorithm // Autonomus Robotics // Stigmergic fabrication // 01 – INTRODUCTION // The development of robotic fabrication strategies within architecture has matured sufficiently to enable the discourse to shift from predetermined operations and tool paths to an emerging interest in real-time feedback and rule-based autonomous operations. While a significant body of literature documents the value and role of autonomy in robotic machining processes, the design agency of robotic behaviors has experienced relatively limited practical investigation. We posit a strategy for encoding architectural design intention within robotic behaviors as an extension of multi-agent generative design processes. Intrinsic to this position is an argument for an ontological closeness between physical and digital material, robot and computational agent, design and fabrication. Our research explores the emergent characteristics of form and articulation generated through varying degrees of feedback between robotic fabrication and multi-agent generative algorithms in several recently completed projects. The conceptual domain of autonomous fabrication and distribution of design authorship has perhaps been most vividly established by Francois Roche through projects such as I’ve Heard About or the FRAC Orleans proposal. In these speculative projects

02 – MATERIAL AGENCY and FEEDBACK // The shift away from top-down hierarchical approaches in favor of attributing design agency to feedback between material behavior and robotic operations is embodied in the recent work of architects such as Del Campo (2014). Del Campo describes his research as moving away “from optimization and efficiency as the primary drivers of digital fabrication in pursuit of a model where materials assume maximum agency in the fabrication process”. Menges (2011) has argued that embedding material characteristics, manufacturing constraints and assembly logics allows a design to be driven through intrinsic performative capacities rather than through hierarchical relationships that prioritize form over materialization. While the discussion of the agency of human, material, digital or robotic design behaviors is perhaps becoming increasingly common within this broader context, our concern for agency and the nature in which it is defined in this paper has developed out of the computational design processes that draw on the

logic of swarm intelligence and operate through multi-agent algorithms. The approaches posited in the following three projects range from single step linear feedback between fabrication and simulation constraints, to recursive feedback between goals and behaviors within multi-agent systems and material phenomena. The pattern of silicon inlay in the Composite Wing project is generated through a multi-agent algorithm that is conditioned by structural performance prior to being robotically extruded. The limitations of robotic rod bending are encoded directly as agent behaviors within the Brass Swarm’s generative algorithm. Through the development of real-time computer vision and robotic control systems, the Feedback Deposition Studies explore the notion of an ontological closeness between robotic and multi-agent behaviors. 03 – AGENT-BODIES // Multi-agent algorithms such as Craig Reynold’s Boid algorithm consider the agent to be a point within Cartesian space. The multi-agent algorithms developed for the composite wing and Brass Swarm projects draw on the self-organizing logic of swarm intelligence and embed the agents within hierarchical structures described as Agent-Bodies. The conceptualization of the Agent-Body resembles the logic of ant-bridges, where it is the interconnected geometry of the ants, bodies that forms structural or architectural matter. Such connections emerge through feedback

The Nature of Robots

a simulated feedback between robotic behaviors and material results in the compression of design and fabrication decisions into a single process without an a priori model. Within industry focused or more applied domains, pragmatic real-time sensing and feedback is enabling the necessary control and accuracy to deal with material and fabrication tolerances exemplified in the Stratifications project of Gramazio and Kohler (2014).


The Nature of Robots

between the behaviors of neighboring Agent-Bodies within the generative algorithm. When these behaviors are conditioned by structural forces, such as deflection and bending moments on a fiber-composite surface, the patterns generated by the Agent-Bodies are a negotiation between behaviors designed to generate emergent patterns and those resisting structural load.


04 – PROJECT 1: COMPOSITE WING // Composite Wing is part of an ongoing series of prototypes that explore the design of fiber-composite surfaces through the use of multi-agent algorithms. Composite wing’s translucent fiber-composite structure is embedded with an intricate inlay of vein-like bodies that grasp, intertwine and disperse in a complex interplay of structural and expressive concerns. Each Agent-Body is a unique shape that emerges from repeated interaction with other Agent-Bodies and local structural conditions, and as a result the bodies inlayed into the fiber-composite were fabricated at two scales. Large-scale bodies are robotically milled from high-density foam and provide the primary structure. Small-scale bodies are robotically extruded from silicon and provide local stiffening by increasing the structural depth and imparting a corrugation, or micro-beam, in the surface. We have explored two approaches to robotically extruding inlay within composite surfaces: extruding rigid thermoplastic onto molds, and extruding flexible silicon onto malleable meshes.

The thermoplastic is extruded using a plastic extruder originally developed for welding plastics. The proprietary motorized screw based end-effector enables a consistent and reliable inlay, however, the rigid nature of the plastic requires that it be extruded directly on a mold. This was prohibitive for the Composite Wing project as the inlay was being extruded concurrently with the mold being used in a separate factory for laminating the fiberglass. Consequently a pneumatic silicon extruder was developed to extrude onto a flat mesh, which could then be laid into the curved mold. The extruder was mounted to a KUKA KRl50 robot on a 4 m linear track enabling large surfaces of the Agent-Body pattern to be extruded. The speed and precision of the robotic extrusion process allowed the highly differentiated pattern of the Agent-Bodies to be realized. However the relationship between these two operations is linear - one enables the other. Without feedback between the computational and robotic these two realms do little to expand the space of possibility of the other. 05 – PROJECT 2: BRASS SWARM // As opposed to the linear relationship established in the Composite Wing, Brass Swarm encodes constraints developed through collaborative robotic rod-bending techniques to construct the complex and highly volatile geometries of a multi-agent system - thus establishing a feedback between fabrication limitations or behaviors and generative design procedures. The rod-bending technique, which

interaction of the control-point agents. 06 – COLLABORATIVE ROD BENDING // Several methods for rod bending were developed using two KUKAAgilus KR10 R1100 SIXX robots. The interaction of the robots was programmed in KUKAlprc with the use of KUKA RoboTeam synchronization functions linking the master and slave robots. The first method rotates the rod to a consistent bending axis enabling it to bend in any plane. The second method, which we describe as shear-bending, creates two bends in a single action. These bends can be non-planar without a separate rotation operation This is an efficient and fast method avoiding the constant rotation of the rod for each bend, typical of many bending procedures. This technique, however, is limited to making pairs of bends in opposite directions. Thus a Z shape can be bent, but not a C shape. The process, is fully-automated, with one of the robots picking up a rod from a pre-labeled array of rods. The constrains on the geometry imposed by the limitations of each robotic bending technique were incorporated within the design of the Agent-Body as well as its algorithmic behavior – creating a single step feedback between robotic operations and the generative algorithm. While fabricating all of the rods with unique bends is efficient (all 300 rods were bent in one day), the manual assembly of these parts is labor intensive and difficult. The advantage of using industrial ro-

The Nature of Robots

utilizes two KUKA industrial robotic arms, is constrained by factors such as a minimum length between bends, a maximum bend angle and in some cases the relative direction of subsequent bends. By encoding these limitations in an existing set of multi-agent behaviors, robotic fabrication constraints interacted with a broader set of design concerns intended to drive the formation of pattern and geometry within the project. The generative design behaviors of the Brass Swarm are primarily concerned with the formation of a coherent manifold surface topology, the interaction of the limbs of Agent-Bodies, and the associated emergent characteristics of their form. The more pragmatic constraints of the rod-bending process are intended to condition this emergent outcome rather than drive its formation. These constraints can be classified as those derived from the empirically tested constraints of the robotic bending process, and those, which are informed by the topology of the assemblage. The robotic constraints relate to the entire rod, but operate computationally on each agent-based control-point of the rod. The brass rods all have a standard length, so while bodies digitally stretch and deform to interact with their neighbors, a pragmatic behavior operates to maintain the length of individual rods. The size and shape of the robot grippers determine the minimum distance between bends as well as maximum angle of any given bend. These constraints are translated into behaviors, which influence the


The Nature of Robots

bots for rod bending (as opposed to standard CNC rod bending machines) is not in the bending operations themselves, but in the potential to combine fabrication and assembly. The robotic positioning and welding of rods aided by sensor feedback has been demonstrated by Dave Pigram and Wes Mcgee (2014), however the application of this to highly complex assemblies such as the Brass Swarm will require a sophisticated vision system and complex assembly planning and approach paths.


07 – CONSTRANITS and CHARACTER // The topology or connectivity of the network of agents is important in understanding their stability and structural integrity. Topological descriptions of the network provide a tool for global analysis of what is an inherently globally ignorant generative system. A graph is used to describe and interrogate the connectivity of the bodies, which enables structural analysis as well as an understanding of connectivity (which influences stability) to be passed back to the individual bodies and influence their behaviors through a set of structural heuristic behaviors. Thus the character of the Brass Swarm is conditioned by both fabrication constraints that are localized to a single Agent-Body, and global constraints that are an emergent property of the form. By exploring alternative algorithmic behaviors capable of satisfying these local and global constraints the emergent character of the resulting designs can exhibit radical

and unexpected formal novelty. The drawings demonstrate the effect of an alternative set of robotic and material constraints upon a similar self-organizing agent system. An arbitrary initial distribution of square-profiled rods tends towards woven and knotted configurations when rods are allowed to twist and are conditioned to form pinned in-plane joints. This conditioning of generative behavior is a form of design agency. As a consequence the robot influences the formation and contributes to the character of the project. This difference is evident through a comparison of the Brass Swarm, which privileges alternating bends (capable of being shear-bent), and the rod-twisting experiments. Brass Swarm maintains a strong directionality and relatively even field, while the rod-twisting assemblies rotate and knot to enable planar connections. So while these behaviors may be intended as pragmatic conditions, they are instrumental in establishing the highly expressive and often unanticipated characteristics of the project. 08 – AGENT BASED ROBOTIC CONTROL PARADIGMS // The third approach posited in this research to the relationship of the agent and the robot is explored through a workflow that integrates feedback between material, sensor, agent and robot. This workflow is tested through a series of short experiments. Feedback collapses the sequential relationship between these processes and establishes a closeness of behaviors. What is significant is that the robotic, materi-

09 – PROJECT 3: FEEDBACK DEPOSITION STUDIES // To demonstrate a generative approach to real-time feedback between material, robot and computational agents, we have undertaken an initial pilot project. This workflow links a vision system (Microsoft Kinect), agent design behaviors (Java/Processing), real-time robotic control (RSI/RSI Server), and a volatile material deposition end-effector (polyurethane foam). A computational agent issued to navigate a scanned point-cloud of the current state of the deposited foam. A series of design behaviors influence the path of the agent, which is referenced as the target for the robot and its further depo-

sition of foam. To explore feedback between agent behaviors and robotic material deposition several algorithms were developed to find peaks or valleys within a snapshot of the point cloud scan. This iterative re-forming of the surface of the foam generated a stigmergic interaction in which the volatility and behavior of both the polyurethane foam and computational agent created a negotiated form. The closeness of the behavior of an agent within a digital environment, and the behavior of the robot in a physical environment frees the digital from any concern for modeling the physical. The computational agent simply responds to the physical and doesn’t require any encoded knowledge of material. Inherent within this process is a principle of anti-simulation, whereby physical and material phenomena are observed rather than anticipated by the digital model. 10 – CONCLUSION // This series of projects and experiments trace what began as an attempt to use robotics to construct geometry generated through multi-agent algorithms, through to positing the robot as the agent in a compression of design and fabrication. This progression parallels a shift in emergent character from that of the algorithm expressed within the Composite Wing - to a negotiated character of the material and computational agent-evidenced by the Feedback Deposition Studies. The experiments with stigmergic depositions embody an ontological shift from a

The Nature of Robots

al, and computational processes can now run concurrently, enabling feedback to become intrinsic to the ontology of the computational model-the robot and the agent are polymorphic, as are the physical and digital models of materiality. A typical workflow for programming KUKA robots is a linear sequence of translations from a desired 3D model, to toolpaths, to robotic instructions written as a linear sequence of KRL commands. To establish real-time feedback we have developed a workflow in which the KUKA robots send commands at 4 ms intervals through a custom server that operates between the design software (Processing/Java) and the robot. This server handles path-planning through KUKA RSI based on design responses from Processing and relays the robot’s position and orientation back to Processing.


The Nature of Robots 146

closeness of agent behaviors operating on simulated material to the anti-simulationof robotic behaviors operating on physical material. Within this anti-simulation, design takes place in and on the object rather than in anticipation of the object and conceptually attributes algorithms, robots and material with equal design agency. Explicitly engaging with material performance and behavior as design drivers is a prevalent theme within the discourse of generative design. By contrast, anti-simulation is not an attempt to digitally model material behavior in order to anticipate known structural, material or formal constraints, but is instead predicated on the attribution of new agency to material through real-time feedback between digital and material agents. We demonstrate that such feedback gives rise to material behavior within digital models, and emergent character within stigmergic material depositions, without encoding the epistemology of these behaviors by a human designer. The ambition of continuing this research in the future, as an extension of the trajectory of feedback approaches outlined above, is to explore further design implications arising from the closeness of computation and material agency. We speculate that this will open a space of experimentation around an engagement with error, inaccuracy and unpredictability within the design. A situation where fabrication precision becomes irrelevant and instead the precision of sensing is critical to the closeness of the digital and physical. This approach requires re-

placing empirical testing and calibration with feedback loops that self-correct and self-stabilize over time.

The SPIDERobot: A Cable-Robot System for On-Site Construction in Architecture

their advantages and limitations, this paper presents an alternative strategy to automate the building construction processes in on-site scenarios. The SPIDERobot is a cable-robot system developed to perform assembly operations, which is driven by a specific Feedback Dynamic Control System (FDCS) based on a vision system. By describing and illustrating this research work, the authors argue about the advantages of this cable robot system to deal with the complexity and the scale of building construction in architecture.

Credits // José Pedro Sousa // Cristina

Gassó Palop // Eduardo Moreira // José Lima and Pedro Costa //

ABSTRACT // The use of robots in architectural construction has been a research field since the 1980s. Driven by both productive and creative concerns, different systems have been devised based on large-scale robotic structures, mobile robotic units or flying robotic vehicles. By analyzing these approaches and discussing

01 – INTRODUCTION // The use of robots in architectural construction can be traced back to the 1980s. By then, robotic technologies were employed in Japan to introduce a high level of automation not only in the factory but also in the construction site. However, the efficiency of such approaches still required a lot of manual work and design standardization. As a consequence, such on-site robotic systems didn’t prove to be satisfactory and lost importance overtime. In 2005 at the ETHZ, Gramazio and Kohler (2008) recovered the interest in robotics with a stronger focus in enhancing creativity. When digital tools assisted an unprecedented design freedom, it was crucial to find the appropriate flexible manufacturing technologies to materialize novel tectonic

The Nature of Robots

KEYWORDS // Cable-Robot // Spiderobot // Automated construction // Digital fabrication // Non-standard architecture //


The Nature of Robots

strategies. The architectural interest in robotics has then spread to other schools and research groups, and motivated the creation in 2010 of the Association for Robots in Architecture. Despite its success in other industries, the use of industrial robots still presents some limitations when facing the scale and complexity of the building construction industry. Its limited range of action and movement makes its application more suited to prefabrication than on-site construction. Furthermore, while the factory space provides a controlled and safe environment to work with robots, the accidental and weather-exposed conditions in the construction site sets a highly unstable scenario to work with such machines. The adaptation of the industrial robot for on-site construction is thus a complex challenge, so other robotic approaches may be explored. By considering the move from the fabrication of components to the construction of buildings, the next chapter surveys some of the current trends facing automation and robotics in construction.


02 – ROBOTIC SYSTEM for ONSITE CONSTRUCTION // Looking to introduce robotic technologies into the construction site, architects and engineers have explored several strategies. This paper proposes to resume them according to the following categories: large scale robotic structures; mobile robotic units; or flying robots.

02.1 – LARGE SCALE ROBOTIC STRUCTURES // This section comprises those approaches looking for augmenting the scale of conventional structures and robotic setups to fit them to the scale of the architectural buildings. The robotic construction initiatives in the 1980s and 1990s in Japan were similar to a big scaffolding structure, integrating robotic systems to perform different operations. Bock and Langerberg refer to this concept as Integrated Automated Construction Sites and describe them as ‘partly automated, vertically moving on-site factories providing shelter for an on-site assembly’. The WASCOR (WASeda COnstruction Robot) group and the Shimizu Corporation were among the first initiatives to promote this trend. In a different way, Behrokh Koshnevis devised a large-scale gantry bridge structure moving horizontally along two parallel lanes to support his contour crafting technology. At the University of Southern California, the team conceived an automated system for carrying a material deposition nozzle to 3D print architectural buildings in a single-run. Resembling a big CNC router, this type of structure concept was followed in other similar strategies, like the D-Shape technology developed by Enrico Dini. Despite the robustness and high operational capacities proposed by these systems, their real application in the construction sites is problematic. According to “these big and heavy robots are difficult to transport to the construction site, have some unsolved scientific

02.2 – MOBILE ROBOTIC UNITS // To overcome the stationary condition of industrial robots and cope with the large size of building constructions, the placement of robots over mobile platforms has been another research avenue. Started in 1992, the Rocco (Robot Assembly System for computer Integrated construction) project departed from the understanding that an “articulated robot placed over a mobile platform (a lorry, a towable platform or an autonomous mobile robot) results very appropriate for the assembly tasks on a construction site”. Besides introducing this mobile condition, two large robotic arms with a higher payload and range of action than conventional ones were developed to allow using larger and heavier material blocks. The focus in automating construction tasks based in repetitive operations had led to other similar initiatives, like the robotic bricklayer S.A.M. (Semi-Automated Mason) developed by Construction Robotics. With a deeper interest in addressing creative issues, Gramazio and Kohler initiated in 2011 a research line on In Situ Robotic Fabrication at the ETHZ. To assure the adaptation to the “continuous changing conditions, unpredictable events, obstacles, and the activities and movements of people working on-site”, they devised a robotic arm mounted on a mobile unit integrating additional systems, like sensor and scanning technologies and differ-

ent end-effectors. This mobile robotic strategy is interesting to avoid the complex setup of heavy large-scale structures. However, ground mobility still has to solve some technological challenges to overcome the unstructured constraints of the construction site environments. 02.3 – FLYING ROBOTIC VEHICLES // The exploration of aerial modes of robotic construction is a recent research avenue in the field. Launched by Gramazio and Kohler in collaboration with Raffaello D’Andrea at the ETHZ in 2011, this strategy employs flying vehicles to manipulate building components in the air, thus avoiding the problems of ground-based mobility and the need for scaffolding or cranes. This approach also considers the cooperation of several aerial robotic units to allow the execution of different and synchronized building construction tasks. This team first demonstrated this approached in the Flight Assembled Architecture installation at the FRAC Centre in Orleans (France). A set of four quadcopters lifted, transported and assembled a tower structure made out of 1500 lightweight foam modules. The research on Aerial Constructions continued in other experiments, by testing the assembly of space frame structures and also the erection of tensile structures. The advantages of the aerial robotic construction applications promise an unprecedented freedom in building construction, which can stimulate new

The Nature of Robots

and technical problems, and need a very high investment”.


The Nature of Robots

ways of thinking and designing architecture. However, this research field is in an early stage of development. The automation and cooperative control technologies, the energy autonomy or the payload capabilities are some of the technology challenges to face in the near future.


03 – The CABLE-ROBOT SYSTEM - SPIDERobot // Facing this trend of developing automated and flexible modes of on-site construction, this paper presents the research of an alternative technology based on a cable-driven robot system (i.e., referred in this paper as cable-robots)-the SPIDERobot. Under development since 2013 at the INESC TEC and the FAUP (Faculty of Architecture of the University of Porto), this approach is based on similar systems developed in other fields, like the sports and entertainment industries, like the Skycam or the Cablecam. The transfer of such systems to a building construction field has the potential to overcome some of the problems found in the examples discussed in the previous chapter. The next sections describe the principles of the SPIDERobot, as well as the first functional prototype that was built and tested to evaluate the concepts. This research is still in an early stage, with first results being discussed in the conclusion. 03.1 SYSTEM DESCRIPTION // Cable-driven robots are automated systems where multiple cables are attached to a mobile platform or end-effector. A positioning system

controls the cables by actuating the motors for extending or retracting the cables. The overall system is thus relatively simple, which opens some interesting advantages to use cable-robots for on-site construction applications. When compared with other robotic construction systems like those presented in Sect. 2, cable-robots are easy and inexpensive to transport, assemble and dissemble on-site, due to the lightness of the cable-based system. Furthermore, the configuration of the cable-based structure allows the definition of larger translational working spaces, which is decisive to face the scale of architectural constructions. Unlike flying robotic vehicles, cable-robots can have much higher payloads and work continuously by means of constant energy supply (i.e., avoiding the use of batteries), while complying with the safety requirements. Despite these advantages, cable-robots also present some critical features. The number and movement of the cables can cause interference within the working space, and their force in the downward direction is limited. The cable system also faces specific technological challenges regarding the control of the precision due to the tension forces and some elasticity of the cables. In this context, the SPIDERobot is a low-cost prototype of a cable-driven robot developed to perform assembly operations in on-site construction scenarios. Its structure consists in 4 actuated cables, which are fixed on the top

03.2 PRACTICAL EXPERIMENT // The SPIDERobot prototype was tested in the assembly of an irregular structure made out of 18 foam blocks

with 120 x 60 x 30 cm. The design of the structure was modeled in Rhinoceros with the goal of defining a geometry that could challenge conventional modes of construction. Then, the different spatial coordinates and orientation of each block (i.e., defined by the coordinates of two points) were listed in an Excel file with the help of Grasshopper. This information was used to inform the SPIDERobot about the position of the blocks in the structure. For picking them from the feeder site, the SPIDERobot took advantage of its FDCS based on a vision system to automatically detect and recognize them in the working space. With this feedback, the robot adjusted its height and orientation to pick the blocks correctly. With this kind of intelligent behavior, the placement of the blocks in the construction feeder site does not have to be rigorous. In the experiment, the blocks were placed in the feeder site in stacks up to 5 units. The assembly of the 18 blocks was completed in around 16 min. The whole process was slow, but revealed to be accurate. 10 – CONCLUSION // This research presented a cable-robot system as an alternative strategy for automating the on-site construction in architecture. Moved by design creativity concerns, the authors tried to overcome some of the technological, physical and economical limitations presented in other research approaches based on large-scale robotic structures, mobile robotic units or flying robotic vehicles.

The Nature of Robots

comers of a frame with 120 x 60 x 135 cm, and connected to a central mobile platform equipped with a rotating gripper. The system configuration presents 4 degrees of freedom (DOF), which comprise the xyz movements and the rotation angle around z-axis, Regarding other cable-robot systems, the SPIDERobot presents some combined distinct features. By using only 4 cables, it reduces the possibility for cable interference with obstacles and leaves more useful working space than systems with more cables. However, because this option leaves the kinematics of the robot under-constrained, the gravity force affects the cables tension and consequently the precision of the whole system. To deal with this situation, the large majority of cable robots use tensor-feasible controlling systems for positioning the mobile robot in the workspace. In a different way, the SPIDERobot presents a specific Feedback Dynamic Control System (FDCS) that does not require sensors for measuring the cables tension. Instead, the proposed FDCS control is based on a vision-based system, which can be something similar to a differential GPS or laser measurement system on the construction site. By using the information available in the environment, the FDCS controls the positioning of the robot while assuring that the length of the cables is always within safe values.


The Nature of Robots 152

The current stage of the research work converged in the production of the SPIDERobot prototype, and it can open the discussion both on a technological and on an architectural level. On the one hand, in analytical studies conducted by the authors the results demonstrated that the topology of the FDCS implemented in the SPIDERobot revealed to be more accurate than the traditional force-feasible approach to the kinematics of cable-robots when performing pick-and-place operations. Therefore, the FDCS proved to be a promising system to be scaled to the size of real construction environments. Currently, the research work is already centered in further exploring the vision system of the FDCS to expand the autonomous capabilities of the system, like in the automatic detection and avoidance of obstacles. Future research directions will be focused in scaling-up the prototype, refining the vision-system and improve the robustness and speed of the motors. On the other hand, the SPIDERobot has the potential to challenge the traditional concepts of designing and building in architecture. By considering its 4 DOF and the geometric configuration of the 4 cables, architects can incorporate such parameters in the creative process to drive design customization possibilities towards aesthetically convincing and functionally efficient buildings. Indeed, unlike other robotic approaches, the cable-robot system is not limited to assist the prefabrication of building parts

(e.g., brick walls). One of its great promises is in the on-site construction of buildings through the assembly of pre-fabricated building parts or the stacking of building units (e.g., prefabricated housing modules), like in high-rise building solutions developed in the FCL design studio in Singapore. Furthermore, the simplicity and flexibility of the system also facilitate its integration in both empty construction sites (e.g., with the help of cranes) and in highly dense urban scenarios (e.g., by taking advantage of existing buildings to set up the cable system). In conclusion, the exploration of cable-driven robots can be an effective solution for stimulating design creativity and expanding digital fabrication processes to the realm of digital construction in architecture. Its application in practice can also foster the vision of different and complementary robotic construction technologies cooperating in the on-site construction of architectural buildings. ACKNOWLEDGMENTS // This work was developed in the scope of the Research Project with the reference PTDC/ATPAQI/5124/2012, funded by FEDER funds through the operational Competitiveness Programme - COMPETE and by national funds through the FCT (Foundation for the Science and Technology).


botic arm to pick, cut and subsequently assemble the components of the structure. To reduce waste, a sensing procedure was developed to generate component based on the form of the found material piece and fit it in the existing structure, similarly to how the Caddisfly Larvae builds its cocoon exclusively with found material. We aim to investigate how the sensor enabled waste control can potentially adjust the form of the assembled structure.

Credits // Kaicong Wu // Axel Kilian // ABSTRACT // Advances in robotic fabrication and computational geometry have opened up new possibilities for including robotic assembly and material selection into the loop. We introduce a method for computing and constructing architectural geometry through the negotiation between the design intention and the constraints of assembly and materials. A small scale experimental structure has been modeled and partially built from EPS foam sheets, using an industrial ro-

01 – INTRODUCTION // one of the major trends of contemporary architecture is about free forms, which triggers many geometric problems that are collectively called Architectural Geometry, the discussion of the related problems focuses on two main areas: rationalization – and fabrication-aware design, which are also referred to as post-rationalization and pre-rationalization. Fabrication-aware design as digital modeling, which automatically generates buildable formal solutions, poses more unsolved problems. Robotic fabrication as one of the advanced prototyping methods provides potentials for finding formal solutions in this research area. It has been used to demonstrate the advances in performing custom fabrication such as wire-cutting, milling or incremental-forming. Recently, increasing attention is being paid to robotic assembly research. For

The Nature of Robots

KEYWORDS // Robotic Assembling prototyping // Robotic Fabrication // Material Sensing // Waste control // Architectural Geometry //


The Nature of Robots

instance, designers have used robotic arms to assemble custom brick walls and assemble on-site constructions or prototype tower models. More assembly based research projects have automated the construction of complex timber structures and roofs. This research presents a robotic assembling prototyping method, in which fabricating and assembling irregular components are controlled by sensor enabled material selection. The form of the structure is modeled and constructed by iteratively computing feedback from the negotiation between the design intention and the constraints of robotic assembly and found materials.


02 – FORM, ASSEMBLY and MATERIAL // The form of a constructed structure is intrinsically linked with the assembling process and the material it is built with. Construction by a Caddisfly larvae serves as a precedent where a cocoon is assembled on shape recognition in found material elements, and fit into local context of the cocoon. A study of human assembly of laser cut parts varies this to precomputed and custom cut pieces instead of found ones. Whereas constraints of assembly and materials are critical to design and construction of architecture, here, a Caddisfly larvae cannot customize found materials but instead develop an ability to assemble the fragments by finding a fitting position in context. In a context of robotic fabrication, this approach can act as a valuable framework for formal feedback

in robotic construction and assembly. 03 – METHODOLOGY: ROBOTIC ASSEMBLY PROTOTYPING and SENSOR-ENABLED MATERIAL SELECTION // This methodology is used for a project with assembly based robotic fabrication setup using EPS foam sheets with sensor-enabled material selection. The core challenge is to achieve the integration of picking, cutting, and fitting the components of a structure. In this research, a 6-axis ABB IRB 7600 industrial robot arm was used to run the prototyping. The components are cut by hot wire from 24 x 48 x 1 in. Expanded Polystyrene (EPS) Foam Sheets. Special end-effectors with small diameters were designed to avoid collisions with the cutting tool and to hold components of different sizes. Hot glue is used to quickly attach the components into fitting positions. 03.1 – GEOMETRY and MATERIAL CONSTRAINTS // Based on the constraints of the chosen EPS foam sheets, all the component geometries need to be flat. The Tangent Plane Intersection Mesh developed by Troche is used to generate a planar hexagonal mesh from a double curved surface. To develop the thickness of the components, all the components are offset outwards with the thickness of the foam sheets. The resulting offset mesh of constant thickness cannot be a polygon mesh of equal valence. To maintain the valence of the offset polymesh, the edge surfaces of each component become twisted.

03.2 – SENSOR ENABLIED PICKING // Without sensing, the EPS foam sheet has to be prepared carefully and positioned in the modeled orientation. As a result, scraps are created by cutting off undesired foam. Sometimes the placement inaccuracies result in partial cuts. The total waste consists of the scraps and the partial cut sheets. Reusing these oddly shaped pieces makes positioning them manually much harder. Thus, Kinect sensing was used to identify component orientation in the found material piece. So a procedure was developed using computer vision to detect the shape of a scrap piece and automatically determine the correct pickup position for a to-be-cut component. For this, a Kinect sensor delivered data into Processing as two separated point clouds using the “SimpleOpenNI” library and coding reference. For the end effector, the center of the point cloud is averaged to be at the end effector center. For the foam, the point cloud of the central area of the top surface is averaged (to reduce noise) to be the height of the pickup surface. Both the center of the end effector and the projected point cloud

of the foam are read by Grasshopper (GH) with the add-on “gHowl” into a digital model with their coordinate system origin being the Kinect position. Given the absolute coordinates of the end effector center in the digital model, the material point cloud can be calibrated into the model space. From the point cloud, a Laplacian Mesh is created by GH as the geometric domain of the found piece. Finally, the component is parametrically oriented into the piece by aligning the longest edges. 03.3 – HOT-WIRE CUTTING // For robotic fabrication, the edge surfaces of a component are sorted by segment order of the inner polyline. Each edge surface is divided into several section planes and the surface is oriented to align with the cutting tool by the planes to cut off the unwanted part. The project adopts here The Mussel add-on for Grasshopper developed by Johns to generate Rapid Code to control the tool path of robot arm. 03.4 – FITTING to EXISTING STRUCTURE // The component remains attached to the end effector and is fitted into the assembly position until human operator fills hot glue into the gaps to attach it to the neighbors. A non-trivial problem is determining the collision free assembly sequence of parts. The components are sorted by the height of their area centers to establish the assembly sequence and to ensure that the arm will always approach the already installed components safely from above. The cutting and assem-

The Nature of Robots

Component sizes are based on the curvature of the design surface and assembly, so that curvature details are maintained while the total amount of cutting lines is optimized. In addition, size was further evaluated for pick-up limits of the end-effector, and the fabrication constraints of the foam sheets.


The Nature of Robots

bly tool paths are compiled into a program simulated in ABB Robot studio. The robotic arm effectively acts as a temporary “scaffold” to secure the new component in its correct position. Once the new component is glued to the existing ones, the new component becomes part of the structure and the arm can be removed.


03.4 – FORMAL FEEDBACKS from WASTE CONTROL // Fitting a found material piece least altered into the existing assembly is achieved by orienting the geometry measured by Kinect in the parametric model. The current top edge of the structure is modeled by isolating the boundary of the inner polygon of the structure model. The lowest point on the top edge is located and the predicted material is oriented by referencing to the two attached edges of the lowest point. The oriented material is shifted to cover the lowest segment of the edge. It is reoriented to be tangent to the design surface and is trimmed off by the top edge. The left over geometry is extruded to the material thickness by referring to the vertex normal of the polygon of the existing structure. The generated component defines the new tool paths and updates the existing model as the input of next generation. 04.1 – LINKING FORMAL ADJUSTMENT and WASTE CONTROL // Fitting the found material piece determines the generation of components and how they are cut and fit into the existing structure. By changing the fit-

ting strategy and the form of the found piece, the structure is remodeled by the assembling of components that are made within the material constraints. A relationship between the formal adjustment and the sensor enabled waste control can be roughly approximated. 05 – CONCLUSION and FUTURE WORK // This project has conceptualized and prototyped a robotic assembly method based on the combination of design intention, the constraints of assembly and the constraints of found materials, with several prototype studies. By robotically fabricating and assembling irregular components from EPS foam sheets; the project built a link between computing programs, fabricating tools and sensors with formal feedbacks. By detecting found materials, fabrication was measured and modeled within the material constraints in digital space to reduce fabrication waste and control assembling tool paths. Future research includes several new steps: Firstly, the sensing technique can be developed to track differences between the physical and the digital models, and allowing for compensation. Secondly, assembly has been restricted to manual fixing components, which should be replaced by robotic fixing. Thirdly, the sensing procedure could check potential structural failures while the object is assembled. Finally, a more robust and optimized computation is required to model the relationship between waste control and adjustment. Similar to the cocoon

The Nature of Robots

of Caddisfly larvae, the assembly outcome will be different based on the material context in which it is built. Further material prototyping will deliver feedback for adjusting the form of a structure. Yet the first results discussed here on geometric detection of found material piece, robotic assembly and minimizing waste are contributing to the design of new methods for freeform architecture.



ject, while also documenting an interaction design studio that prototyped sensor-based integrations with the BotBar. KEYWORDS // Robotic Education // Multi-disciplinary Design // Smart Technologies // Creative Robotics //

The Nature of Robots

Credits // Marjo Niemelä // Samantha Horlyck // Susana Alarcon-Licona // Dylan wozniak-O’Connor// Gabriele Ulacco, Rodney watt // Rob Saunders //


ABSTRACT // The BotBar has been developed to respond to the significant challenge of integrating smart technologies and sensor loops with industrial robot arms. The process has focused on the robot as an open design platform, utilized as a nexus for education and collaboration between the disciplines of Architecture and Interaction Design. This paper discusses the success and challenges that have emerged from this pro-

01 – INTRODUCTION // The BotBar is a flexible design project that aims at exploring the use of robots, human-robot interaction, sensors and supplementary technologies by employing a robotic bartender as a versatile and engaging teaching and training tool. Robotic bartenders are not a novel concept nor is the use of technology, and this project does not situate the creation of the BotBar itself as the end result. Instead, it positions the BotBar as a useful platform with which to foster collaboration and skill sharing between disciplines with a mutual interest in robotics. There are diverse examples of the development and use of creative robotics throughout industry and education; however, the introduction of industrial robot arms into an education setting presents challenges regarding accessibility, integration and knowledge propagation. The BotBar was developed to provide students and researchers with the training and tooling to successfully overcome the challenge of incorporating industrial robots in their projects. However, the BotBar is not only a facilitation tool but a boundary object for design disciplines that can be repurposed and expanded in a variety of directions.

02 – FUNCTION as BOUNDARY OBJECT in a SOCIAL CONTEXT // Robots are a prime example of boundary objects capable of connecting groups with different backgrounds in the construction of new knowledge and new approaches. As Pickering (1992) states, boundary objects aim to local needs, to promote shared coherent actions and knowledge and to allow parties to resist evident translation and reconstruct methods. The concept of the “bar”, as a complementary boundary object makes this a successful exercise with potential to be expanded but is not without its limitations. This project began as a training platform and method to understand how to foster and manage shared work across diverse skill sets, which exist within many design and education contexts. Since then it has been developed into a platform able to engage people from diverse technical backgrounds collaborating on a singular, approachable output which serves as a functioning interactive system to bring robotics to the wider community and highlight human-robot interaction. As a boundary object, the BotBar engages distinctive groups such as designers, programmers, and architects with each other’s disciplines and skill sets while also defining important reference points and tasks between each group. At the boundary of these fields the robotic arm can function as a point of crossover between robotic fabrication and the many potential uses within Interaction Design. The creation of the BotBar has involved tasks ranging

The Nature of Robots

Within the broad disciplines of architecture and interaction design, robots are increasingly being used in design education. The BotBar aims to advance our understanding of the bridge between sensor loops and the possibilities these offer to robotic fabrication and human-robot interaction. Additionally, the BotBar aims to alter the customary view of an industrial robot arm from that of a labour machine to that of an autonomous social robot, which responds to dynamic inputs, gives feedback, and interacts with humans. In their paper The Framed Pavilion, Dank and Frieissling (2012) explore some of the challenges of working on non-standard architecture and design projects involving robotics, caused in part by the wide range of collaborators-often-involved. The BotBar posits an alternative method of collaborating whereby all information is not necessarily input to a parametric model. Instead, it aims to embrace the diversity of collaboration and draws on varied inputs to create a series of intertwined feedback loops: from loops between sensor input and real-time robot adaptive technology to the larger conceptual loops of enabling creative collaboration and skill sharing. The skills of those designers who work on a project such as the BotBar act as catalysts, which feed back into the project as improvements or variations upon the design and function of the BotBar.


The Nature of Robots

from design, fabrication, parametric modeling, motion programming and hardware integration; a critical part of the process involves the input of skills and knowledge sharing from each discipline involved, as well as negotiating the best approach from such a vast skill base. Advances in technology, such as the use of industrial robot arms, present changes in the nature of contemporary societies and consequently it is relevant for students and researchers working in areas of designing, improving and constructing environments to experiment with and generate meaning from these technologies.


03 – BOTBAR 1.0 and 2.0 // This project has enabled students, academics and researchers from multiple disciplines to work hands-on in a safe environment with a KUKA KR6. The first iteration of the BotBar engaged architectural and interaction design researchers through a division of tasks whereby the physical design and construction of the bar was largely handled by the architecture discipline while the control interface (utilizing GhPython scripts to activate programs on the robot controller) was developed by the interaction design discipline. However, motion programming with Grasshopper and KUKAIPrc - an integral part of the BotBar 1.0 - was a task both disciplines could collaborate on. The first BotBar iteration followed a pre-programmed toolpath, the second iteration has focused on exploring robot-human interaction, sensor input,

and social experimental interfaces for real-time and pre-programmed processes in addition to the use of industrial robot arms in challenging environments such as an unstructured and crowded room. This social or public environment can present difficulties in using standard, accessible input sources such as Kinect motion sensors, which may struggle with detecting multiple users. It also raises issues in regards to safe operation where the robot may be more accessible than in a factory setting. Particular attention was paid to designing the bar for safety without needing to include barriers, and as such the second iteration of the bar ensures a safe distance from the robot to the user by separating them with the structure of the bar surface, which is greater than the reach of the robotic arm. It also incorporates rotating bar sections controlled by Arduino to further the physical separation between the work envelope of the robot and the users. The flexibility of the BotBar platform is in part that each of the areas explored in each version can be developed, reiterated, and rethought by a design team who can use their existing skill sets (such as parametric modelling, Arduino, Python, and traditional fabrication) in an environment which fosters the cross-pollination of a broad range of skills, ideas, and knowledge between different disciplines. In addition, each area can scale from relatively simple designs to a level of complexity that incorporates multi-process, multi-robot interactions. The BotBar is intended to be used

04 – CHALLENGES // Robotic arms, while increasingly commonplace in architecture and design faculties, still present challenges for design educators as their inherent complexity of operation often renders them inaccessible to researchers and students who tend to work within short project cycles. Following a social constructivist philosophy, where technology is seen as an integral part of society and “properties and effects are usually defined in a particular social context”, the training and teaching model was shaped to accommodate the needs and skills of our community and explore three significant challenges. 04.1 – ACCESSIBILITY // Students and researchers have the opportunity to work in the further development of a real application with areas of interest

ranging from motion path programming to the exploration of different interaction paradigms. The development of flexible training platforms is one approach that has enabled students and researchers to engage with robots on research-relevant projects by providing them with enough knowledge to start using robotics as soon as possible in a safe and creative environment. 04.2 – KNOWLEDGE // Experimental robot projects, such as the BotBar, aim to create, promote, and deploy the training modules, teaching skills, and specialist knowledge required to successfully use industrial robotic arms. They do so by providing a space to explore different approaches to hardware configuration, motion programming, programing languages, controller familiarity, and other complex tasks that can create opportunities for innovation such as the implementation of custom end-effectors or of communication between robots and external devices. 04.3 – INTEGRATION // Different modes of operation are being explored to allow students to work on projects without requiring a comprehensive understanding of the complete system. We have implemented the communication with different programming environments (i.e., Python and Processing) as gateways to integrating supplementary tools that support the background and skills (e.g. Arduino and Grasshopper) of the students.

The Nature of Robots

as an accessible system; by utilizing well documented and easily controllable components, such as Arduino powered motors and Raspberry Pi controlled pumps, different fields can provide input to the project. Instead of controlling these various components and the robot with a relatively unfamiliar language for many users, such as KUKA Robot Language, control of the BotBar is handled through the computer with the simplest tools available. This allows motion programming with Grasshopper and KUKAIPrc to be used side by side with interfaces built with Processing, and allows other disciplines or fields to contribute to the BotBar using familiar computer or hardware based processes.


The Nature of Robots 162

05 – TRAINING and TEACHING PROGRAMS UTILIZING the BOTBAR // Utilizing the BotBar initially for a training program and then a teaching program has seen the creation of several successful projects by research teams. The program is structured in a way that introduces the fundamentals of working with a six axis industrial robotic arm but quickly focuses on integrating individual research interests, aiming to the industrial robots as open design interfaces with adaptable tools to achieve a goal. This program begins with an interactive hands-on tutorial with the robot that focuses on safe operation, basic movement, six-axis familiarization, and an awareness of physical restrictions such as axis limits. The program then moves to manual path programming, such as a simple drawing program, before introducing software tools, which enable complex parametric control of the robot. An understanding of the fundamental concepts of robot motion and control empowers students and researchers to explore and partake in a wide range of robotics projects. In order to encourage collaboration, propagate knowledge, and share skill-sets, robotic researchers have formed three broad and overlapping research clusters:Subtractive, Additive, and Smart Technologies. These clusters include robotics projects and units of study across multiple disciplines and provide the opportunity for students and researchers to interact with robots in a creative environment.

06 –BOTBAR as a CONCEPT to EXPLORE INTERACTION DESIGN // BotBar 1.0 raised attention and highlighted opportunities among different disciplines including design computing and interaction design. The second version of the project was envisioned as a platform for student participation as Robots provide an embodiment and the ability to add social interaction to the learning context. By working with a standard robotic arm, interaction design students were able to gain knowledge and inspiration from previous work in education, industry, and society. With the aim of exploring human-to-robot interaction in a specific social setting, students from the Master in Interaction Design and Electronic Arts were immersed into a studio-based unit around the BotBar project. This platform aimed to develop possible scenarios where robots and humans interact in a harmonic and collaborative way. These ideas were nourished by lectures encouraging user research and context analysis, as well as tutorials adapted from the training programs. On the one hand, modules on Rhinoceros, Grasshopper, and KUKA|Prc determine the basics of robot control and set up, as well as 3D modeling and fabrication methods. On the other hand, sessions on Arduino and Processing demonstrate the opportunities of external hardware and software integration. In this study, where the initial objective was adapting the resources for interaction design students, a custom XML

• Team 1 // Pubpop implemented a webcam mounted to the robot arm as part of the end-effector. Custom software and hardware were developed to allow QR codes to be used to select drinks and to construct a tactile skill game that brings together multiple participants. The focus of this project was on creating an amusing social experience, where the robot acts as a facilitator and provides feedback to the participants through its movements based on both the scanned QR codes and the output of the tactile skill game. • Team 2 // Faces developed a tangible tabletop interface as an integral functional and aesthetic element of the BotBar design. The interactive menu uses tokens of different colors to represent different ingredients and triggers corresponding programs from the robot controller. In this case, the group’s attention was directed towards interaction innovation within the specific context. • Team 3 // Bounce aimed to engage users in an interaction with the

robot through a dance battle. A Kinect sensor was used to facilitate the interaction, to capture users’ movements and to provide information that generated feedback from the robot. An important part of this exploration was the analysis of human to robot movements and the implications of robot motion as a creative communicative method. The design of the environment, including the lights and designated interaction spaces were also incorporated into the project. From a tactile interface triggering pre-programmed movement routines to a Kinect sensor based interaction and custom robot motion study; the flexibility of the BotBar has been as an open platform for exploration and knowledge acquisition in different areas which complement the students’ education. The direct result has been to highlight several areas for further development. In particular, the complexity of sensor integration in the BotBar will to continue to expand as it is integrated into a wider range of future units of study. 07 – SUCCESS and LIMITATIONS // The aim of this project was not merely to create a robotic bartender: this incarnation of the project has been versatile enough to transition from a training program into a teaching platform. However, increasing complexity will begin to reveal limitations. The BotBar has been particularly successful in familiarizing fabrication staff, students, and researchers with KUKAIPrc, end-effector design, and

The Nature of Robots

interface was implemented. This Processing interface bridges the communication between robot and computer and allows users to activate movements without the need of using the robot teach pendant. Students were encouraged to explore their concepts through experimentation with a broad range of technology and processes, generating interesting and diverse results:


The Nature of Robots 164

knowledge such as path planning and singularity identification. In one semester the BotBar has enabled the above documented studio of Master of Interaction Design students to work with robotic motion programming and sensor technologies as applied to the robot arm. It is anticipated that with more open tools to interact with the robot and generate results quickly the BotBar will be used within undergraduate studios next. The success of this program has also been in cultivating interest in robotics with a human-centered scenario that is broad and relatable while still able to showcase real outputs from multiple disciplines collaborating around a common object of interest. The limitations of the project in its current form as a bartender are limitations of scale, which affect architectural output. A future incarnation has been proposed which would use the robot to self-assemble on-site, however with a 6 kg payload and limited reach this may not be a feasible scenario. It may be necessary to move away from the experience-focused output of a cocktail to a more formal exploration of spatial interaction. The BotBar has proved itself to be an ideal opening to this conversation, but in order to have professionals and students working together on integrated projects that are of equal interest to architects and interaction designers another evolution of the project is likely needed.

08 – CONCLUSION // The breadth of engagement with a robotics program that used the BotBar as a catalyst is encouraging. However, the future of the platform will be to develop an interface to increase the accessibility of robot arms for students and researchers who lack programming skills. An API, which uses a library of sensor inputs alongside a range of end-effectors will foster further cross discipline engagement while allowing more advanced interactions and programs to be explored intuitively and prototyped quickly. This multi-disciplinary engagement has the potential to evolve further with the robot continuing to act as a boundary object, located at the intersections between the disciplines of interaction design and architecture, that can foster collaboration and the intermingling of skill sets-demonstrating that since meanings are not embodied in boundary objects, divergent uses, interpretations, and reconstructions are likely. ln BotBar 1.0 and 2.0 the work was divided so that the architecture discipline handled the design of the environment, the physical bar fabrication, the end-effector fabrication, and the automation processes while the interaction design discipline developed the interfaces and software bridges. In future iterations of the BotBar, this divide needs to become more apparent and thus more efficient or to be dissolved so that the tasks overlap and better foster research, training, and education in creative robotics and fabrication in architecture and design.

The Nature of Robots

ACKNOWLEDGEMENT // This research has been supported by the Faculty of Architecture, Design and Planning, and has been developed at DMaF Lab. The authors would like to thank Interaction Design students Duane Allam, Angela Graf and Guilherme de Macedo from “Faces” project; Abhiruchi Chhikara, Qingwei Kong and Yonghan Ji from “Bounce”; as well as Yu Guan, Yan Song and Dan Zhang from “Pubpop”, for their inspiring projects and their participation in this research.



quired in robotic fabrication processes. This paper discusses a series of tests of scanning techniques on three example substrates typical to wood construction, including: lath for plastering and stucco, spaced sheathing for cedar shingles, and traditional stick framing. Scanning substrates accounts for the gaps in tolerance that emerge from rough to finish construction such as variation in as-built dimensions, misalignment of members, and the adaptive behavior of materials as they adjust to new conditions. From a comparison of scanning techniques, a cost benefit matrix is developed to aid in evaluating the appropriate application of scanning techniques for various robotic applications. KEYWORDS // Reality Capture // Motion Capture // Robotic Sensing // Robotic Fabrication // Photogrammetry //

The Nature of Robots

Credits // Joshua Bard // Richard Tursky // Michael Jeffers


ABSTRACT // While there has been substantial development in the use of industrial robots for the tool pathing and assembly of fabrication components for architecture, there exists a scope for improving a methodology for the mapping of material substrate in architectural construction settings. Construction tolerances posit a problem since they vary widely from rough to finish applications and are often at odds with the demanding precision re-

01 – MOTIVATION // Robotic technologies have been incorporated in nearly every aspect of building construction from excavating sites to erecting structural frames and applying material finishes. While the application domains of robotic technologies in architecture may span the entire construction process, the typical implementation of industrial robots on construction sites is not seamless. Instead, robots are typically deployed for discrete tasks in the construction process with varying degrees of autonomy, often working in concert with a complex array of human activities. This stands in contrast to factory settings where automation and strict control of manufacturing tol-

1. 2. 3.

Lath-for plastering and stucco (Lath) Rib Assembly-for cedar shin gles (Digitally Fabricated Sub strates, DFS) Stick Framing-for interior and exterior sheathing (Stud Wall)

Surface descriptions generated by these scanning techniques were benchmarked against robotically probing each substrate with an end-of-armtool (EOAT). From this comparison, a cost benefit matrix was developed to aid in evaluating the appropriate application of scanning techniques for various robotic applications. 02 – DEVELOPMENT of COST BENEFIT MATRIX // The technology space of real-time sensing and reality capture is rapidly expanding, thus providing designers with ready access to autonomous workflows, adaptive feedback, and environmental awareness in robotic fabrication. Understanding the breadth of sensory devices and techniques is the first step to isolating what approaches are best for specific applications. This paper reviews three common scanning approaches and develops a matrix that describes the various costs and benefits of each system. These will be discussed in the following. The factors evaluated cover issues or advantages in use, integration and output. xx Scan Time describes how much information can be acquired, interpreted, and brought back to the end-user in a meaningful format on which to base next actions. This may be a single ‘scan cycle’ or many scan-cycles arrayed over a given search-space. Information processing time also weighs heavily on some strategies. Scalability refers to the ability for

The Nature of Robots

erances can currently be coordinated across the entirety of complex product workflows. On-site building construction poses particular challenges to industrial robotic applications. In addition to issues of mobility in changing environments, the need to interface with multiple material systems installed with varying acceptable tolerances by the building trades calls for improved methodologies for the mapping of material substrates in construction settings. Construction tolerances vary widely from rough to finish applications and are often at odds with the precision required in robotic fabrication processes. For example ASTM C926 stipulates a plane tolerance for veneer plaster of ± ¼ in (6 mm) in 10 ft. (3050 mm). In order to generate safe and accurate motion planning for fabrication processes, the translation from planned virtual models to as-built realities must account for variation in dimensions, misalignment of members, and the adaptive behavior of materials as they adjust to new conditions. This paper discusses a series of tests with common scanning techniques by using three typical wood construction systems where a finish material can be applied to an architectural substrate:


The Nature of Robots

the same device(s) to jump scale of search-spaces. This is largely tied to workable ranges. Workable Range describes the device’s optimal or allowable range of (most) accurately detecting or acquiring real-space data.


02.2 – INTEGRATION CONSTRAINTS // Cost of System include all associated (financial) expenses of the device, peripherals, and software. Required Software discusses all necessary interfacing technology to communicate with, extract, and interpret data from sensor device. Lower-end devices may involve user-developed communication strategies and communication techniques to successfully integrate data to an applied robotic workflow. EOA-Integration may also be referred to as ‘Eye-in-Hand’. Devices that can be hosted EOA have the advantage of a mobile frame in space, and constant data of the location of that frame (robotic kinematic model), thus allowing a given sensor to be in multiple locations, orientations, and distances in relatively short periods of time, and tied to a continuous process. However, spatial constraints of the end-effector, tooling, task-conditions, data transmission, and power requirements must be successfully integrated or solvable for EOA. 02.3 – OUTPUT and COMMUNICATION // Accuracy refers to the margin of error, or fidelity of resolutions available on device. Latency is the factor of lag time present within

a single scan cycle and communication. This does not include processing time to compute higher levels of information and data types (this is a factor of Job Time). Data Type is the end-output data type(s) available to the end-user. If user-developed software, often this will be primitive data types. Proprietary software will likely provide multiple options, formats, files, for output. Provided SDK’s for such software may further extend this category. 02.4 – BENCHMARK // Manually jogging a robot fitted with a probe is the most effective way to directly acquire, with very tight tolerance, a point in space with respect to the robot’s coordinate system. For the purposes of establishing a benchmark, and demonstrating a high-accuracy method of obtaining substrate data, this technique was applied on all three substrates with upwards of 100 sampling points each. Note that manually operating the robot in direct contact with work surfaces requires human proximity, which increases risk of damage, human injury, and increases scan time. Probing is best utilized for finding key, or defining, elements of a known shape in an unknown location to ‘register’ digital model data to real world probed data for subsequent robotic actions. Both human operated and automated probing can be performed with high levels of accuracy, at the cost of invested time. All probe data is immediately understood relative to the robot-space model without the need

03 – THREE SCANNING APPROACHES // 03.1 – PHOTOGRAMMETRY // Using an EOA-mounted camera (Nikon D3100 with AF-S Nikkon Zoom Lens set at 24 mm, f/18 aperture priority), planar toolpaths were prepared to quickly guide and aim the camera at many sections of the target surface area. The advantage of utilizing the robot arm was primarily an issue of reach. A significant factor is a high degree of coverage, since missing data cannot be translated into the output mesh. This allowed issues of occlusion, or hard to reach areas of the surfaces to be shot from a number of angles, producing the best results. Images were processed using Autodesk’s Memento Beta. Since this software is cloudbased, there is significant lag time (test have shown typically 8-24 h) between uploading the images and getting a workable mesh file. The input data required are simple JPEG images that have framed significant shared features among the other images. These shared features are then used to correlate and compute a change in distance relative to other features in

the image plane. This software is designed to deliver 3D depth information as a factor of correlating the image data across all other images in the set. High face-count meshes were generated for each substrate, with very little ‘noise’ in the target areas. 03.2 – INFRARED MOTION CAPTURE // A six camera OptiTrack Flex 13 camera array was ceiling mounted above the test substrates. Each camera had a resolution of 1280 x 1024 and a frame rate of 120 fps. OptiTrack’s proprietary software, Motive Tracker, was used for internal calibration of cameras and streaming camera output. A custom streaming component in Grasshopper (plug-in for McNeel Rhino and visual scripting environment) brings rigid body tracking information directly into the CAD environment. A custom calibration script automatically aligns all tracking with the lab robot’s base coordinate system. Once calibrated, the location and orientation of custom tools can be tracked relative to the robot’s coordinate system. A custom probe was created with interchangeable tips for various types of scanning. Users positioned the handheld probe at each of the benchmark probe points for comparison and also raced along key features of each substrate for direct surface generation from scanned guide curves. There exist benefits and drawbacks within this approach: Benefits include the user directed flexibility in determining how to scan specific features in an open-ended environment; a high level of accuracy and precision;

The Nature of Robots

for further transformation from another coordinate system. Drawbacks from probing include the fact that input is still point-based, a pixel in an image, and therefore slow to enumerate an array of information. The act itself requires slow and careful manual motion, especially with probes of higher precision and sensitivity.


The Nature of Robots

and the capacity to track plane orientation in addition to point location. Drawbacks to this scanning approach include the need for human input during the scanning process; high monetary cost of system, size limitations; and potential occlusions of capture volume.


03.3 – 1D DEPTH SENSOR // A simple proximity sensor (Sharp GP2Y0A02YK0F) wired into a microcontroller with a digital radio platform allows one to point-and-shoot for depth measurement. Using the tcp normal vector and origin can generate the resultant pinged location as a factor of its distance along the normal vector: In this case study, three separate toolpaths were prepared based on the design-documentation of three surface geometries and rough locations. Scan-lines were estimated as valuable areas to search in hopes of capturing an edge condition to then justify the 3D geometry against. However, with 200 m between a read of the sensor and overall scope of toolpath distance, resolution was sacrificed for the sake of speed. Scan paths operated with <20 mm steps along lines that swept the extent of the 4’ x 8’, substrates. This then generated upwards of 300 data points per surface. Many readings also fell off, or through, openings in the more sparse substructures, introducing clear biases in the data. Using design-geometry data, most readings could be culled as invalid readings. However, those readings that were not culled still exhibited ambiguity

from sensor noise or registered points on parts of the structure that were not strictly the desired top surface. 04 – COMPARISON of RESULTS // Each of the following points explains the performance of a given scanning process against the three different substrates. These histograms show the frequency of a given scan point being within some millimeter tolerance range of the benchmark data model. MOCAP comparisons could be done point-to-point, whereas Momento and 1D comparisons had to perform nearest-point to a mesh constructed from the benchmark data. Therefore, error takes into account our closest approximation to the probe data, and not actual surface geometry, as is discussed here: 1. The 1D sensing has the worst range, as the scan line technique was designed to capture the difference in depths that could be understood as edge. Even culling complete misses does not take into account scan points that hit near, or just past the top surface, where the probing was targeted; 2. Photogrammetry delivers a mesh object and the most error is introduced in simple transformations to calibrate it back against the real object, as the mesh output has no dimensional register to real-space. A best fit genetic algorithm was used to obtain the best transform for this output to reposition the mesh in the working-model;

05 â&#x20AC;&#x201C; CONCLUSION // After test scanning three material substrates using three common reality capture techniques a significant amount of information can be leveraged to better select the appropriate scanning technique for a desired operation. Significant criteria to be considered include the price point, ease of use, accuracy, and robust performance under onsite environment constraints. For overall accuracy, hand-held MOCAP probing provided the most reliable results. This technique also favors flexible definition of environment features based on user input. If autonomous scanning using EOAT is desired, then the photogrammetry approach provided accurate results with automatic mesh generation. If evenly distributed, dense information is desired across the scene then Photogrammetry provides the most expedient approach, excluding cloud based post-processing of meshes. If select features are of interest, MOCAP provides the best flexibility while scanning. If target-


// Special thanks to research assistants Brian Smith, David Blackwood, Nidhi Sekhar. Autodeskâ&#x20AC;&#x2122;s beta version of Memento was used to for the photogrammetry section of this research. Optitrack Cameras were used for all motion capture. HAL Robot Programming and Control was used to stream MOCAP sessions into Grasshopper.

The Nature of Robots

3. IR Tracking was used with a hand-held stylus that could be placed on any surface given the tool was not occluded from the cameras. For this data, the points were directly obtained on top of the actual physical locations of the probing benchmark data. This represents a clear depiction of device-to-device tolerances, whereas the other techniques required additional geometric relationships to be established to build an equitable metric;

ed reference points are of interest, 1D depth sensing can be quickly deployed using motion planning. Future research building up on these findings aims to refine each scanning technique. It is suggested that for more robust 1D depth sensing, a custom triangulated laser range finder can be developed. For MOCAP scanning, better automatic calibration scripts are being developed to align live streaming with robot coordinate systems. Through the development of an on-robot MOCAP camera mount issues of limited capture volumes with fixed, ceiling mount systems could be overcome. The survey of scanning techniques for surface description of architectural substrates, as has been discussed here, offers a scope for further investigations in applied fabrication techniques for each substrate system, and can be extended towards robotic plastering, robotic shingling, and autonomous wood frame assembly.


ROBOTICS-BASED PREFABRICATION in ARCHITECTURE Prefabrication the Shinsa-Town Free-Form Ceiling Structures

lected expandable polystyrene (EPS) as the material of the ceiling structure, and we developed and utilized BAT (a Grasshopper plug-in), to process the work as a free-form production method. We also invented a new cutting method to implement the specific types of components that were otherwise unlikely to be implemented due to the limitation of the straight hotwire. This research describes a transport system for the components of the framework, and a robotics-based on-site installation method that is required for the utilization of a robot in the fabrication of these structures.

The Nature of Robots

KEYWORDS // Curved Ceiling Shape // Hot-wire cutting // Robotic Fabrication // Industrial Robot //


Credits // Xun Li // DongHan Shin // JinHo Park // HyungUk Ahn // ABSTRACT // Existing methods for the production and installation of freeform ceiling structures were not suitable with respect to construction cost and time period that were assigned to a Shinsa-town project. Hence, we selected the Robotic-based Digital Fabrication Method that was being tested at that time. Considering the construction cost and period, we se-

01 â&#x20AC;&#x201C; INTRODUCTION // Industrial robots can be widely applied in various areas such as constructions, interiors, and industrial design because of their accuracy, flexibility, and immediacy. Particularly, they can bring many advantages in terms of time and cost, especially for producing curved surfaces by processing the expandable polystyrene (EPS) selected in the Shinsa-town project. Indeed, the reduction in processing time of the robot-based hot-wire foam cutting method is outstanding when compared with the conventional CNC machining or manual laboring that is normally used. Consequently, the reduction in processing time implies a reduction in the construction period and in construction cost. Because the cost and time available in

method. In addition, by classifying the phases of the Shinsa-town project into production and installation phases, and establishing detailed sub-phases, the problems that occurred in each phase and the solutions for improvement are discussed. 02 – METHODOLOGY // 02.1 – CONTEXT // To implement this design after receiving a request for this project, conventional production methods were investigated and classified into three types: Galvalume Pipe Bending Method (GPBM), CNC Machining. Next, we investigated the robotics-based digital fabrication method from a production time, materials, and cost perspective, based on the experience we gained previously by using the ABB robot in the HOT-WIRE CUTTING RESEARCH PROJECT (HCRP). Through this project, we collected data for estimating the time and cost required for cutting the foam by hotwires. Based on the results presented, it was concluded that the robotic-based digital fabrication method is the best choice in terms of efficiency and economic feasibility; and based on this finding, production was begun. 02.1 – CONTROLLING MATERIAL DISTRIBUTION // First, the curved surface was divided into a regular grid. The size of the grid was determined by considering robot size, module weight, and minimization of material loss. Next, a volume mass that

The Nature of Robots

the case of small-scale constructions such as interiors and remodeling of outer walls are limited, both time and finance are insufficient for performing research on the methods and materials to implement free-form shapes. If a designer or a building owner requests to re-work due to errors pertaining to accuracy, the responsibility shifts completely to the manager of the construction. In order to prevent such problems in advance, commonly manual labor of highly skilled workers is required for free-form architecture, resulting in high production costs and extended construction time. The Shinsa-town project is the first project where an industrial robot is utilized on a Korean construction site. For this reason, there is no precedent with regard to the system required to establish the robot-based digital fabrication method. Thus, thorough recording and investigation of the details of Shinsa-Town project should be considered as the most crucial tasks that enable the robot to be more widely used in the Korean constructions. This research presents all the issues that occurred during a series of processes from the selection of free-form production method to the production of components and the installation at the site. Based on the analytical results of the types proposed by Chohelo A+U, an autonomous analysis was conducted by comparing the data of traditional manufacturing methods with the data obtained through research on the robotics-based digital fabrication method to identify the advantages and disadvantages of each


The Nature of Robots

is intended according to a standard plane was determined; it was divided into 3 sections, with thicknesses of 600, 900, and 1560 mm, such that the curved surface is included in the volume occupied by these three sections. Based on the grid set above, the entire volume was divided into approximately 200 components, and each component was labeled. The basic material processing method is solid-based cutting, which produces a basic component by cutting the EPS (Expandable Polystyrene) into a fixed size, and then cuts them as per the desired shape. However, the three types of BC for the framework were configured with consideration for minimizing material loss, since the construction period would have been extended indefinitely if the production of each of the 200 BCs was optimized to size. In order to simply minimize the loss of material, each component was categorized as one of three basic types according to size.


02.1 â&#x20AC;&#x201C; CONTROLLING MATERIAL DISTRIBUTION // For hot-wire applications, two curves or guidelines are required to create a path. The tool path along which a robot moves is made based on these guidelines. The tool-path of hot-wire is a set of coordinate planes that continuously moves between two guidelines, which are determined by a series of straight lines and a given point. The x-axis is located on the straight line and the z-axis is determined by the center of the straight line and the given point. Thus, the hot-wire is always located in

a series of coordinates and the center of hot-wires always corresponds to the origin of the coordinate plane. The process of generating the tool-path is as follows. First, two curves that serve as guidelines are extracted relative to the shape of the component. Once the curves are defined, they are divided into a number of points. A straight line connects the uniformly divided points in a sequence, and the coordinate system is based on the straight line and a given point. As the number of points increases, the difference between the designed and produced curve surfaces decreases, but the time for processing increases. The time for processing affects the material loss because it is related to the time that hot-wires stay on the EPS. In addition, the optimized dividing of guidelines is crucial because of the processing time affecting the construction time as well. Thus, the guidelines were divided into 30 mm gaps in this project. In the â&#x20AC;&#x153;Xun Cutting Methodâ&#x20AC;? only the ruled surface can be cut by hot-wires. In the case of most components, a method where hot-wire cut the component along with the guideline considering two separated curves as the guide curve is applied, however the shapes shown below cannot be cut by designating the two guide curves. In order to solve this problem, several experiments were performed through computational modeling, with a first set that tested methods for cutting a simple plane, followed by tests for curved surface cutting. None matched the original curved surface precisely.

03 – ASSEMBLY and ONSITE CONSTRUCTION // After cutting all components, these were assembled in groups of three and cut one more time after fitting them in a 900 mm x 900 mm frame to minimize inconvenience

of additionally cutting the components due to difference in their sizes. A special EPS suitable adhesive was used here, which combined foam bond and cement. Subsequently, the module was moved to the site, where further assembly into larger areas was performed. Most shapes can be implemented by a general cutting method that sets two guidelines, but this method is not suitable for some shapes. In order to solve this problem, we intended to use the ‘Xun Cutting Method” previously introduced. This method implements the shape required for the second processing at once within acceptable error, by giving guidelines for changes in configuration. The components that were delivered in accordance with the grid marked at the site were fixed with bolts at the precise location based on the name marked on the component. Then, the finishing work was additionally done, followed by flame resistant treatment and plaster coating. 04 – DISCUSSION // The research faced a number of issues during production and installation processes. First, additional types besides the basic type were necessary, while making the unit box because of irregular grid gaps. Thus, we should make the tool-path that cuts the box. If we had considered production issues more deeply in the design process, those problems would have been solved. Secondly, we used the BAT plug-in (which is still under development) to design a tool-path, convert it into a code, and deliver it to the robot; and

The Nature of Robots

A new method was developed to cut curved shapes. While selecting the two guide curves, these were not selected both as curves, but select one side for curve and the other for multi-curves. The number of separation between the curves is important in order to achieve the shape closest to the original shape during the guide curve configuration process. If a curve is exactly divided at the point where two curves meet, the boundary of the side in the component that was cut appears clearly, and four lines will be recognized as one. This results in a smoother shape when dividing overall lines, and yields a result close to the actual shape. Furthermore, since one guiding line is much shorter than the other, the amount of foam loss increases because the time that hot-wires stay at the shorter curve is relatively long. We reduced the required strength of hot-wires, as well as the moving speed of the robot by considering this point. While delivering a code to the robot, if two guide curves coincide, the starting or ending point is not recognized. Thus, the closed curve within the error rate should be cut, so that it can be recognized as two independent curves. These tests were further evaluated for accuracy between 3D modeling and robotic fabrication.


The Nature of Robots

because the real-time upload module was not developed, it was delivered via a portable device. We spent a lot of time in this process. Especially, it was inconvenient to proceed when modifying the tool-path was required. Consequently, a real-time communication will be necessary to increase the efficiency in robotics-based fabrication. Thirdly, although the robotics-based fabrication is a highly accurate method, errors still occur. The small errors combined together result in a large error in the relatively large-scale construction. In this project, if there is a small error on a small component, then a major problem can happen when 200 small components are assembled together. In order to minimize this error, we cut the unit box to a larger size and expanded the starting and ending point a little more while generating the tool-path that cuts the component.


05 â&#x20AC;&#x201C; CONCLUSION // This paper reported on the process of production and installation of a curved surface ceiling structure by a robotics-based digital fabrication method. We successfully implemented the proposed design via this method with less cost and in shorter time than other conventional methods. We experienced the efficient, economic, and flexible features of robotic fabrication and could verify the vision as actual interior constructions in this process. It is also expected that the robotic fabrication will be widely applied in a shorter time, if several problems can be solved.

Firstly, the robotics-based fabrication has the advantage that proposed designs could be executed effectively. In order to apply it more efficiently in the design phase, a concept of robotic fabrication needs to be integrated at the beginning phase of the design process. Additionally, we should identify the movement of the robot and get information about the tool installed on the robotic arm. Secondly, we should investigate the specification and performance of the materials that are used for production. In this project, we could have reduced the time for production if we had designed with such knowledge. If we had considered the hot-wire cutting method in designing the curved surface of this range, we would have predicted the situation and invented a new production method such as the â&#x20AC;&#x153;Xun Cutting Methodâ&#x20AC;? and there would not be the need for an experiment during the production. Thirdly, further studies of on-site installations using the robot should be performed. If our study proceeds to the phase where the robot has information about the location of each component and installs it in the exact location, it can eliminate errors that can occur while installing manually, and thus further reduce the time for installation. ACKNOWLEDGEMENT // The authors would like to thank Daeyeon Kim (Robotics departmenQ, ABB Korea and Youngmin Park, and KUKA Robotics Korea for their continuous and generous support.

STEREOTOMY of WAVE JOINED BLOCKS Towards a Wave-Joined Stone Construction Using Wire Cutter Toolpath Generation

object. This leads to the design of a wave jointed block capable of an extended structural ability, concealing the majority of the cutting effort inside the joined blocks. The proposed fabrication system uses a wire cutter end effector following a toolpath generated from quad based mesh topologies. This single tool cutting system maximizes the efficiency of the cutting process and returns the once technical aspects of robotic construction back to the designer.

Credits // Simon Weir // Dion Moult // Shayani Fernando // ABSTRACT // This research focuses on developing a system for using 6-axis robotic arms to cut interlocking blocks with wire. Tracing the trajectory of stereotomy through millennia of practice, an extrapolation is presented that stereotomy will serve increased formal and structural complexity. The addition of robotic carving to stereotomy also removes the ethical-aesthetic connection between the carver’s effort and the visual attention given to the

01 – ORIGINAL VALUE // The art of stereotomy has its origins in the prehistory of architecture. The origin of the English word architect, are Herodotus’ two and half thousand year old accounts of architects directing the careful cutting and assembly of natural materials into massive objects. Whether using timber or stone, the architect was a type of sculptor, taking the earth’s materials and rearranging them; first hewing them from their origin in nature, fashioning them in a precise manner, and finally joining these pieces together. The deliberate stepby-step process of taking irregularly shaped, roughly hewn materials-typically stone-and fashioning them into precisely predetermined forms is the art of stereotomy. Procuring very large slabs of stone was a difficult task. This was recognized by the ancient communities that encountered them. Herodotus de-

The Nature of Robots

KEYWORDS // Stereotomy // Toolpath Automation // Digital Fabrication // Vaults // Wave Joint //


The Nature of Robots 178

scribed an impressively large shrine carved from a single piece of stone that was dragged whole from the quarry by two and half thousand men and their beasts for three years. Michelangelo’s over-life-sized figure of David was renowned as a masterwork of stereotomy and comprised of an extraordinarily large block of Carrara marble that was quarried in the 1460s. Sculptor Agostino di Duccio roughed out the figure in the quarry-to reduce the block’s weight-before moving it to Florence. Apparently overwhelmed by the responsibility of the block, Duccio abandoned the block and it remained in Florence until Michelangelo took on the task in 1501. To reduce his labour and to honour this rare behemoth slab, Michelangelo removed as little as possible. It was then the size of the figure cut from this block that impressed his peers. A second principle for considering stereotomy is displaying the finesse of the carver. In sculpture, as in architecture, eyes are drawn toward some areas more than others. The wise architect typically directs the most energy to these details: whether they be a figure’s eyes or a column’s capital. The joints within the walls, never seen, are cursorily cut. Directing the carver’s time to the visible details, the architect enacts an ethic of aesthetics by allowing the carver to see their labour and knowing that others can too. This second principle dissolves when a robotic arm wields the saws and files. Robotic masons change this fundamental aspect of stereotomy. The third principle recognizes that

across the history of stereotomy, increasing structural efficiency aligned with new developments such as vaults evolved from corbeled surfaces smoothed in situ; pre-cut voussoir vaults with joint surfaces at normals to geometrically estimated thrust lines; and Antonio Gaudi’s hanging catenary models tracing asymmetrical thrusts. With the arrival of steel and concrete Modernism, stereotomy diminished into an academic focus on descriptive mathematics. This research ultimately led to the CAD and CAM systems that are accelerating structural analysis options such as Philippe Block’s Thrust Network Analysis tool. 02 – WAVE JOINED BLOCKS // Extrapolating the trajectory of stereotomy points towards two conditions. First is that continuing to increase cutting complexity for greater structural capacity by using the tensile strength of typically compression-only vaulting materials. While adding a second material to provide tensile strength can produce remarkable results, as demonstrated by Fallacara (2012), the different thermal expansion rates ultimately cause damage, reducing lifespan; consequently we seek single material construction. The second observation is that with robotic fabrication, the ethical relationship between effort and effect is attenuated. This opens the possibility that more effort might be used on an unseen surface. These trajectories led to the development of the wave jointed block, whose edges interlock to transfer bending loads.

03 – TOOLPATH GENERATION USING QUAD MESHES // The cutting process for the wave joint blocks has been prototyped using a hot wire to

cut through foam blocks and an innovative approach to toolpath generation. Descriptions of 3D objects in a computational environment are usually done with NURBS based formats or mesh based formats. NURBS are an equation-based format, whereas meshes construct shapes using vertices, edges and faces. Each face is made out of 3 or more edges. A 3 edged face is known as a tri, a 4 edged face is known as a quad, and an n-edged face, where n > 4 is known as an n-gon. Despite the typical usage of NURBS for digital fabrication, our approach towards designing, prototyping, and digital fabrication employs a mesh based topology using quads. For organic, sculptural forms, mesh modelling is generally much faster than modelling with NURBS. This speed increase gives the opportunity for the designer to prototype and experiment with iterative forms more efficiently. Once the design has been finalized, the mesh retains valuable information about the relative resolution of different parts of the mesh that has further impacts on simulation and fabrication processes. For example, during extruded print prototyping, most extrusion toolpaths are generated from a triangulated meshed STL format. The resolution differences allow the designer to specify different print resolutions on different parts of the mesh. In the wave jointed block, a higher resolution was applied to interlocking surfaces than to external surfaces. This allows the interlock to be accurately studied

The Nature of Robots

Construction blocks with interlocking sinusoidal edges have appeared several times, e.g. patented by Robson (1976) and developed by Estrin (2014). These blocks assist the construction process since they slip into place. Consequently, Estrin and Dykin et al. (2005) proposed these for robotic construction in extra-terrestrial environments. These blocks use a single wavelength along their long sides and half a wavelength on the short side; and have been tested for axial compression strength. Differing from these precedents, this project uses blocks that are designed with multiple wavelengths of relatively high amplitude that transfer bending loads between the blocks. The amplitude of the wave can be altered for different amounts of force-shallow waves transfer little force, while deeper waves can transfer more. Secondarily, while Robson’s and Estrin’s blocks appear to be a cast material with sinusoidal curves on both the long and short edges of the blocks, these blocks are designed with flat edges on the short sides so the whole block can be cut using only a wire cutter. The design of the wave joint includes a repeated horizontal section, which adds robustness to the otherwise narrow and fragile edge. Then, in order for the blocks to interlock, the horizontal tab is also included in the middle of the block.


The Nature of Robots 180

while the external non-interlocking surfaces are less precisely and more quickly extruded. The meshes of the blocks have been built using quads, as they can produce organic or natural contours around the mesh’s surface, which may be manipulated by the designer. These contours may then be converted into a cutting toolpath that can be adapted for different end-effectors. In the wave jointed block, it is envisaged as being cut out of stone with a diamond wire cutter. To prototype this process, a hot-wire cutter was used to cut though foam. The wire cutter end-effector implies that the geometry follows a ruled surface. There are also some potential benefits to using this quad-meshed approach for CNC milling. Typically, mills inevitably leave striations on the object. These striations are then sanded down into the final product. Using quad meshes allows the designer to determine (or influence) the directions of the toolpaths, and hence the directions of these striations. These striations can then be positioned as a desirable effect, rather than an unwanted after-effect of the cutting operation. 04 – EDGE LOOPS // Meshes built with quads are able to describe the directionality of contours throughout the surface of the mesh using the concept of edge loops. An edge loop is determined using the algorithm below: • Selects or deselects edges that:

• If edges has 2 faces: - Has vertices with valence of 4; - Not shares face with previous edge; •

If edge has 1 face: - Has vertices with valence 4; - Not shares with previous edge; - But also only 1 face;

If edge has no face: - Has vertices with valence 2;

As shown, the directional path defined by an edge loop is terminated when a vertex with 3 or more than 4 valence is reached. For this reason, meshes built with tris or n-gons are inappropriate for deriving natural toolpaths from edge loops. The two directions of a quad mesh’s edge loops are used to determine the initial cutting path, as well as which neighboring edge loop to cut next. By moving from one edge loop to another, the shape can be completely described. The KUKA robot’s KRL instructions determine the robot’s kinematics using 6 parameters: X, Y, Z, A, B, and C, X, Y, and Z refer to the translational coordinates of the tool’s centre point while A, B, and C refer to the rotations around the XYZ axes respectively for the tool. To determine the translational coordinates of the toolpath the midpoints of each edge in the edge ring are used to provide a base solution. Depending on the length of the wire, and the available buffer space along the length of the wire, the midpoint may be moved along the edge. To determine rotational values the wire cutter can be

05 – PROTOTYPES for WAVE JOINT STEREOTOMY // The toolpaths for the wave jointed block were staged as a 2-pass sequence beginning with a rectangular block. First, the block is trimmed into an arced block similar to a traditional voussoir block using a low resolution mesh. Next, the wave joints are cut from both ends. Through the wave joints, the resolution of the mesh is increased, as indicated by

abundance of the orange lines indicating surface normal around the wave joint. This process was prototyped in foam using a KUKA robotic arm mounted with a hot wire end effector to produce approximately 1:2 scale blocks, 60 x 30 x 20 cm. when identical prototype blocks are stacked into a tower 5 blocks high the bending load is easily transferred to the base producing a rigid ‘bent column’. The regular alignment of the waves in this design also allows for the rearrangement of the blocks into the surface of a vault. Further analysis of the structural loads carried across the joints will later determine the size of the joints needed to bear the weight of a stone material used in a similar manner. 06 – CONCLUSION // These models demonstrate the practicality of using wire cutters to produce wave jointed blocks that can be easily assembled into arches and vaults. Although the arch and vault surfaces are presented here as prototypes, the system can be adapted into an unlimited number of architecture elements and structural solutions. In a complete vault the joints do not need to carry their bending load between blocks, and the wave joint will weaken the blocks compressively. Therefore a design for an arch using this joint would need to be more massive relative to its structural load. The bent column, like the arch surface, is not a form that is impossible from solid stone. It is only impossible to produce with 5 traditionally joined columns sections. The possibility that this joint

The Nature of Robots

assumed to approach co-planar to the quads of the edge ring, or to approach normal to the quads of the edge ring. Of the three rotational axes, one axis always runs along each edge of the edge ring whereas the other two depend on the assumption made. The other two axes may be derived from the cross-product of the quad’s vertex or face normals and the cutting edge. Using these axes produces workable toolpaths, but they may be altered if collisions are detected between the object and the end-effector. As the KRL’s rotational values are intrinsically rotated - i.e. the rotational axes are performed in sequence and change after each rotation - a mapping needs to be done to covert the global 3D rotational axes to the values required by the KRL. Once the mandatory parameters are determined these can be converted into different types of continuous paths, such as linear or spline paths (if interpolation is desired due to a lower resolution mesh). Further manual adjustments of the robot’s kinematic calculations can be introduced but is generally unnecessary.


provides would be best expressed in very large structures where the fabrication options, transport logistics and lifting capacities limit the maximum size of pieces assembled. The wave joint design can take many variations. Constructed as a ruled surface between two waves, there are two frequencies and amplitudes that can be modified, as well as their relative positions. These variations constitute a range of structural capacities. After millennia of stereotomic innovation the interactions between descriptive geometry and construction robotics are enabling new opportunities in stereotomy to emerge. The progress toward greater structural effect has always been produced by the judicious joining of blocks. The mass production capabilities of construction robotics in stereotomy can be directed toward precision joint cutting that can achieve new forms in architecture.

The Nature of Robots

ACKNOWLEDGEMENT // This research has been supported by The Faculty of Architecture, Design and Planning, The University of Sydney, and was produced at DMaF.


Credits // Nicholas Williams // John Cherrey // KEYWORDS // Robotic Fabrication // Digital Fabrication // Mass-customization // Robust Manufacturing // ABSTRACT // This research reports on the extension of a simple design concept into a technique for the rapid fabrication of customized components of acoustic panels with ruled surfaces. Recent proposals for the robotic fabrication of construction components include examples of techniques for cutting ruled surface geometries through the pairing of an industrial robot arm with a linear blade. While these demonstrate the fabrication of curved and complex geometry, they do not resolve many technical issues around

speed, accuracy and material finish, critical to a robust process demanded by the manufacturing industry. To address these, the research presented here pursued a detailed investigation into of the history of band saw cutting technology. Key knowledge of material crafts and obsolete applications of ruled geometries both offer significant insights. Using these in an iterative development, a rapidly improved robotic design and fabrication process is demonstrated here. 01 â&#x20AC;&#x201C; MOTIVATION // The potential integration of robotics as an active element in the design and fabrication of customized, non-standard architecture has been intensely explored in recent years. Many research projects have demonstrated that through the consideration of fabrication constraints in early stage design and the iterative production of material prototypes, significant opportunities exist to enhance design exploration. Enhancements have been described and documented for a potential to conceptualize novel material effects, and process efficacy, closely aligning design proposals with means of fabrication. The research project discussed is equally interested in robotic fabrication, applied here towards the design context of a music teaching space. The project brief demands the retrofitting of an existing classroom - a demountable room within an existing space. This constrained the design to a rectilinear space of approximately 12

The Nature of Robots



sqm. Within these constrains, the design seeks a novel solution for a panel system that lines the interior walls as the critical element for aesthetic and acoustic room criteria. An iterative research process was undertaken to develop a design system, and to develop a robust outcome, which could be applied at larger scales of production. Materials demanded durability and a high quality of finish appropriate to the classroom environment. The robotic fabrication process warranted a rapid and reliable solution that could be comparable in efficiency to existing processes of product manufacture.

The Nature of Robots



02.1 – ACOUSTIC IMPERATIVES and STANDARD SOLUTIONS // This design of a music practice room presents challenges in terms of architectural acoustics, and specifically sound attenuation, which centers on the minimization of sound transfer between the proposed space and those surrounding it. A Sound Transmission Class (STC) of 60 rating was identified as a target performance level. To achieve this, a modular system of massive sheet materials mounted on timber frames was designed and constructed. This was calculated through variations on guides for acoustic attenuation with commercial wall systems. Key issues and strategies for the acoustic design of small music spaces have been outlined by Osman, and include diffusivity to reduce occurrences of flutter echoes and specular

reflections. Standard solutions are the application of absorptive materials and the use of quadratic diffusers. Furthermore, previous research into architectural acoustics includes the shapes of wall surface articulation applied to the design of acoustically diffuse spaces by Peters (2010); or acoustic behavior of ruled-surfaces of hyperboloids (Burry et al. 2011); or shaping spatial acoustics through robotic fabrication (Reinhardt et al.2014). These studies attest to a rich field of acoustic research through empirical studies on complex forms. 02.2 – FABRICATING RULED SURFACED // Ruled surfaces have a long tradition in architecture, and have been manufactured through both additive and subtractive processes. More recently, studies have been produced through exploring the flexibility offered by industrial robotic arms to cut volumetric materials using a range of linear blades. Other applications of robotic end effectors include the use of hot wire to cut foam (Feringa 2012; McGee et al. 2012; Brell-Cokcan and Braumann 2013); wire diamond blades for stone cutting (Feringa 2014); or bandsaw blades for cutting timber (Johns and Specifically Foley 2014). These latter studies explore variations of blade or material mounting, from a blade mounted on the end of the robot arm, to alternately a fixed position of the blade and the material held by the robot. Here, a robot is programmed to move the tool in a trajectory that causes the brank to move through the blade in a given sequence

04 – DEVELOPING a DESIGN SYSTEM // 04.1 – DESIGN CONCEPT // Engaging these criteria, this research centers on robotic fabrication through a combination of a robot arm that feeds material to a floor-standing bandsaw in order to cut laminated blanks. It was proposed that long, rectilinear blanks be cut in a single pass with each of the resulting pieces on either side of the cut then applied as wall panels. The application of both pieces produces minimal material waste. A number of considerations are embedded within this approach. Firstly, the resulting curved, non-parallel wall surfaces are good for the acoustic diffusivity. The blanks are a custom lamination of differing materials - primarily timbers but also foam and acrylic. These are cut with a surface shape defined by two planar curves running along either side of the piece. The cutting process reveals these layers, cut at acute angles across their boundaries. By varying the shape of cutting the surface not only is the form of panels varied but also the profile shapes of the cut layers of laminated materials. This has implications for acoustics, with the differing materials effecting reverberation time. There are also aesthetic implications through the patterns created. For the design of the room, a generative process was developed to explore patterns. This centered on the translation of amplitude curves of sound waves to create cutting geometry for

The Nature of Robots

in order to achieve a desired shape. 03 – A MEANS or an END? // While some of these examples are explicitly experimental, others take current industry processes and increase flexibility of geometry through robotics. Of particular relevance to this research is Johns and Foley’s timber cutting (John 2014). But as this research argues, while this suggests exciting industrial opportunities, many barriers remain in scaling up the process. This research contents that robustness is at the heart of scaling both physical size and volume. While the term is commonly used by designers to describe material and construction durability, in manufacturing it relates more commonly to the resilience of processes. Here, robustness describes a high volume manufacturing process, in which speed and reliability are vital. In a contemporary manufacturing context, flexibility is added through robotic fabrication to enable customization of parts. Furthermore, the need for processes to remain robust - through minimization of complexity, and uncertainty in other aspects of fabrication - has been noted. While current research has prioritized the design of novel formal and aesthetic effects, the research presented here privileges consideration, of process, robustness, and aims at a fabrication process that is reliable and rapid. In doing so, constraints to formal outcomes may become apparent, challenging, researchers to identify productive trade-offs in negotiating the constraints.


The Nature of Robots

blanks. The selected pattern applied to the room is based on a set of tuning frequencies, with time steps mapped across the surface.


04.2 â&#x20AC;&#x201C; FAILING FAST // A first mockup panel was rapidly and successfully produced by cutting a 600 mm foam blank with a hot wire. The blank was held by vacuum on the end of the robot arm and moved through the wire. This mimicked precedent research, albeit with a part composed of several foam layers glued together. A wire bandsaw blade was sourced to enable a larger scope of material cutting; a spiral blade of a compact 1.5 mm diameter typically used to cut metals. A second material palette was prepared with 600 mm long blanks that were fabricated including a mixture of timbers, Dupont Corian and Echopanel (a felted PET sheet with acoustic absorptive properties). These comprised the desirable palette of materials for panels. First cuts proved to be highly problematic in terms of fabrication. The materials cut with only varying levels of success: the Corian was clean; the timbers, both hardwood and softwood, tore and burnt; while the Echopanel melted producing bad surface smears. Both the burning and melting was the result of sawdust not clearing from the blade and hence becoming clogged. Substitute materials were briefly considered but the larger fabrication problem remained; the easy deflection of the wire blade restricted cutting to very low speeds. The first 600 mm

blanks required 20 min to be cut without blade deflection. Such a slow process was not only laborious but also contradicts aspirations for a robust process. It was decided that the use of a wire blade presented insurmountable barriers and so this approach was abandoned. 04.3 â&#x20AC;&#x201C; SEEKING OLDER PRECEDENTS for PROGRESS // A further process of investigating craft precedents was undertaken to better understand historic applications of band saws. Two discoveries proved important. The first was the application of a similar technique to that used to fabricate curved beams for boats. Through adding a rotary axis to the bandsaw, the blade could be tilted as a timber member is passed through. Ruled surfaces could be accurately robotically fabricated through an indexed relationship between the linear movement of a work piece and the rotation of the bandsaw structure itself. In parallel to identifying this potential, the design of bandsaw blades was also investigated. Blade manufacturers commonly specify a minimum radius of cut that can be achieved in a plane perpendicular to blade. This is proportional to the thickness of the blade. There is, however, scant information published as to the rotation in a perpendicular plane, twisting against the plane of the blade. The planar depth of a standard blade means that twisting of a work piece commonly causes distortion of the blade. While previous research from Johns (2014) has mentioned the need for such twist, rotation

04.4 – MATERIAL PROTOTYPING // With the adoption of a planar bandsaw blade with modified teeth and profile, a sequence of example cuts were undertaken stepping through radius and twist in the two planes. This blade was iteratively modified in order to increase kerf width. A series of short 300 mm blanks of cheap timbers served as generic blocks to test geometric limits. Further tests with Echopanel and other selected hardwoods was also undertaken to identify limits at which a good cut finish could be demonstrated. The technique was also applied to the production of further full-scale prototypes to address a combination of shape, custom laminated blanks, and the control of pattern across multiple panels. Large prototypes include those fabricated with geometries taken from small section of the larger wall design. They confirmed the geometric limits identified through test cuts to

shorter blanks. A level of control was achieved to demonstrate continuous shapes and selective wash boarding in key areas of the designed surface. Final fabrication of the Music Room is currently being undertaken with cutting of 50 blanks scheduled to be completed within one day. In exploring the geometric limits of the system, a ‘wash boarding’ effect was also identified which occurs where the blade is periodically but minimally deflected. The geometric conditions at which this occurs are readily identified and this effect can be introduced to accentuate key parts of the larger surface pattern. 05 – RESULTS // Through close attention to craft precedents, the technique presented has developed to be more robust in terms of cutting speed and reliability of finish. A conventional bandsaw blade has been adopted allowing a material feed rate of over 500 mm/min across material up to 200 mm wide to be cut. With two parts produced simultaneously, this equates to a rate of 0.2 sqm/min. Through modification of this blade, good rates of curvature in the two key planes of rotation are accommodated. Using a 12 mm blade, tums of as little as 50 mm radius can be achieved. Twisting is also accommodated at up to 24” per 100 mm linear travel. Most importantly, these rates of curvature can be achieved without deformation of the blade or reducing speed. This results in a finish, which is consistent with existing saw-cut materials on the market. Compared

The Nature of Robots

has been constrained by conventions of the blade. To compensate for this, cutting speeds and path have been compromised. Through understanding blade manufacture, it was identified that a significant twist could indeed be achieved through modifying the blade to increase the kerf. With a wider channel around the cut, the blade misses the twisting work piece. With a precise control of twist enabled through a robot arm, a detailed process of blade modification and testing was undertaken to establish the geometric limits of such an approach.


The Nature of Robots

to early tests with wire blades with spiral cutting surfaces this is a significant improvement in speed (close to 20 times faster) and reliability of finish (no burning or tearing of the material). Despite modification, the conventional blade does not provide the geometric freedom of a wire blade. Rather, tradeoffs between geometric freedom and the robustness of process have been carefully considered. We have been able to extend the geometric possibilities of a conventional bandsaw blade to a point where a 180° twist can be achieved over the relatively short length of 750 mm. This is sufficiently flexible for application in the Music Room project. It would also be geometrically applicable to other freeform timber components such as curved glue-lam beams. Importantly, cutting speeds and surface finish have not been compromised through this.


06 â&#x20AC;&#x201C; DISCUSSION and CONCLUSION // This research has presented key iterations in developing a technique for robotic fabricating of customized panels with ruled surfaces. With the improvements through integrated knowledge from material craft and manufacturing, a robust technique was developed and applied to the production of a wall panel system for a music classroom. This project demonstrates improvements to process achieved by extending a known technology and integrating the deep knowledge around this. This is an important shift from early

research focus. Early attempts to improve a novel but untested blade and technique were abandoned in favor of maximizing the geometric freedom offered by a more conventional tool. This has implications across much research engaging digital fabrication, through demonstrating benefits of close interactions between robotics and practices of material craft. Future research will explore the application of this technique to other design applications, next targeting the prototyping of other building components and products. Many of these components are currently milled and this technique offers potential reductions in volumes of dust as well as machine time. Furthermore, the research will be further expanded through the application of additional end effectors such as the Tungsten Carbide and bimetal bandsaw blades, so as to enable cutting a wider range of materials, and to enhance refinements of surface finish. ACKNOWLEDGEMENT // The Music

Room project was supported through awarded by an Innovation voucher the Business R&D Voucher Program Program trough the department of Business and Innovation, the State Government of Victoria awarded in partnership with Deutsche Shule Melbourne. Professor Xiaojun Qui from RMIT University contributed expertise and research in acoustic. Fabrication was undertaken in the Architecture and Design workshops at RMIT University.


feed forward material formation could be combined or even replaced by feedback-based production processes. The different time points of material analysis not only allow for greater control but also enable completely new production methods. KEYWORDS // Robotic Fabrication // Sensory Feedback // Formation Strategies // Material Analysis // Adaptive Formwork //

Credits // Felix Amtsberg // Gernot Parmann // Andreas Trummer // Stefan Peters // ABSTRACT // The submitted paper presents the results of a 3-year research project in the field of adaptive forming technologies for freeform structures made of UHPC (Ultra-High Performance Concrete). The focus of the research is found in the analysis and comparison of the developed robotic-driven formwork. During the research it was observed that the typical concept of process creation of a direct

01.1 – RELEVANCE in ARCHITECTURE // The efficient production of freeform building elements is still of essential importance to architecture. Concrete has been used as a construction material to realize inspiring double curved structures during the 20th century by Nervi, Saarinen, Isler and others, but has become less important to this kind of architecture in the present day. Its castability in combination with reinforcement materials like steel, glass, and carbon fibre, and its further development to high and ultra-high performance concrete can result in a material, which is ideal for the fabrication of freeform structures. But double curved concrete structures remain at the fringes of the building industry even as the development of new digital design tools has enabled more freeform designs. This is due to the relatively high expenses of the form works necessary for production, as noted by Jörg Schlaich (Sobek 1987) in the fore-

The Nature of Robots



The Nature of Robots

word of Werner Sobek’s dissertation, which aimed to reduce formwork expenses with innovative approaches to moulding systems. Time is still the essential factor in production costs and waste reduction is an additional factor in each production cycle with new production techniques analyzed according to machining time or material waste.


01.2 – RESEARCH MOTIVATION (Adaptive Formation) // Over the past decades many research projects have focussed on automated, digitally-controlled processes for generating form. Several researches have investigated the realization of reconfigurable tools for a sheet metal formation process from within the building industry, as well was within the aerospace industry. But as of 2003 hardly any of the solutions patented during the last 140 years have been applied commercially. Especially during the last ten years, digital fabrication has started to step into the environment of architecture, art and design. Newly-developed digital tools have enabled a direct link between design and fabrication; developed new machines and systems; and deeply changed the way architects, artists, and designers work. The concept of adaptive formworks was rethought by several research groups in the environment of architecture: ADAPTIVE MOULDS developed by ADAPA and the FLEXIBLE MOULD PROJECT are just two examples in this research field. However, big innovative construction companies and

specialized companies still predominantly use subtractive CNC milling of foam to produce double curved structures. Concluding, it can be said that the adaptive form generation in the field of double curved concrete structures is still not state of the art. 02 – RESEARCH PROJECT // In response to this the research project ‘Shell Structures made of Ultra High performance Concrete (UHPC) - Thin Walled Double Curved Construction Elements Made of High-Performance Concrete for Innovative Shell Fabrication’ was started in 2011 to develop a resource efficient production process for double curved thin-walled concrete elements using just one industrial robot as the central unit for all essential production steps. The implementation of industrial robots was the logical choice for production as they combine versatility, accuracy, and controllability while also allowing for a direct transfer between digital design and fabrication. 02.1 – DIGITAL PROCESS CHAIN / NEW DEVELOPED FABRICATION PROCESSES // The designed production process starts with the creation of a digital form. This form is analyzed and subdivided into elements that can be prefabricated. This geometrical data is then used for the following production steps: adjusting the robot driven adaptive formwork; generating toolpaths for different casting concepts; and for formatting and sanding down the contact faces using wet machining.

03.1 – ADAPTIVE FORMWORK GENERAL CONCEPTS and REQUIREMENTS // The geometric analysis of a pavilion designed for the research project define the parameters for the construction of the formwork system. The project had the following requirements: • Realizable bending radius of > 1.5 m as derived from research about the curvature of shell structures and designed freeform geometries. • Form accuracy of < ±1.0 mm as defined by the pre-stressed dry joint technology. • Robustness and simplicity as defined by the adjustment as done by an industrial robot where complex machined elements should be avoided within the formwork. Three different formwork concepts where designed and chosen during the research: the PIXELFIELD, the PINFIELD, and the CLAY MOLD. 03.2 – PIXELFIELD // The concept of the PIXELFIELD is based on a dense high-resolution package of rectangular plastic bars with a spherical head and a detached membrane. The first conceptual model was planned and tested with an arrangement of 15 x 30 pixels at 10 x 10 x 60 mm. For the real scale prototype, the first and last pixel

03.3 – PINFIELD // The concept of the PINFIELD consists of a uniformly spaced field of pins with ball joint heads that attach to the mold surface. The first conceptual model was planned and tested with an arrangement of 7 x 7 pins with a distance of 40 mm while the full-scale model was designed with 5 x 5 pins and a distance of 250 mm. The formation process is done by a serial screwing of every pin, which requires cyclic adjustment of the pins to avoid local overstretching of the mold surface. 03.4 – CLAY MOLD // The concept of the clay Mold differs in terms of production logic. Instead of developing a formwork using a complex adjustable mechanism, the intention here is to plastically deform a raw mold material using different actuation techniques. The tool affects the material directly, and tool development for specific tasks is relatively easy. The material used for the research is an oil-based clay that doesn’t dry and can be reused. The displacement of the material is done by hammering and pushing

The Nature of Robots


were halved every second line to create a bracing. The resolution is 30/31 x 30 pixels; the pixel size 30 x 30 x 400 mm and the mold area 900 x 900 mm. The formation process is done serially by pushing each pin separately. Earlier tests used a wheel adjustment. When a row is adjusted, the position is fixed by clamping. A membrane surface is laid on the actuation pins after adjusting the whole area, and the form is then imprinted with a filling material.


with a simple end-effector. The formation process can be described as material-neutral because there is no subtraction or addition of material during the production process. 04 – INFORMATION and FEEDBACK / DEVELOPMENT and PRODUCTION // An important distinction in this research is between INFORMATION and FEEDBACK. The relation of INFORMATION, analysis, and approximation is immanent in the development of fabrication processes. Work steps, actuation, and the response of the effected material are observed and adapted to the specific requirements.

The Nature of Robots

04.1 – OPTIMIZATION // The optimization is based on design criteria, material information and production methodology. The process itself is redeveloped, optimized and adjusted, until the requested parameters can be achieved. The production itself is not under observation anymore. The process is planned, production is started, and the final result is expected.


04.2 – INFORMATION as FEEDBACK // Information is defined as feedback if it takes a direct influence on the fabrication process. This characterization is difficult, as the determination of a stock predefines the milling process, but does not affect the final result, just the fabrication process that in this case is the toolpath. It is the acquisition of specific information during the creation of a process, and its use for analysis that allows the im-

plementation of the information before and/or while the fabrication process is running. The simplest example of information-based process is in the analysis of different scrap pieces and their use for the production of a desired shape or in the analysis of a specific log and the use of geometrical information for resource-efficient production according the analyzed shape. The next category is the ‘multiple cycle feedback’ - the continuous gain of information during the production process. Between two actuation cycles the fabrication machine stops, the current state of the work piece is then scanned, analyzed, and the resulting the information used for the next fabrication step. The extraction of sensory information and actuation are separate processes occurring at different times. This strategy enables reaction on appearing indeterminablenesses, and thus enables iterative production cycles. An example is the iterative formation of clay as a mold material, as presented in this paper. “Real time feedback” is the continuous gain of information while the production process is running. The production process is permanently under observation. It enables reaction while actuating. This process requires a direct link between actuation machine and sensors. An example is the sensory controlled bending of timber elements. 05 – SENSORY INFORMATION and FEEDBACK in DEVELOPMENT and PROCESS GENERATION of ADAP-

05.1 – PIXELFIELD: UNCERTAINTY // While the adaptive mold was planned meticulously and constructed precisely, the process still left space for uncertainty. The clamp and release or every line resulted in slightly different bearing situations, where deviations of the pixel size in the range of less than a 1/10 of a mm could add up. The increased dead load in every line and varying friction between the pixels were hard to predict and impossible to simulate. Almost all of the adjusted pixels remain within the requested tolerances. Close to 70% are within an accuracy of ± 0.25 mm; 20% within ± 0.50 mm;

10% accuracy of ± 0.50 mm; and the number of pixels out of range below 1%. After adjustrnent a scanning cycle was implemented to detect the pixel out of range, and they were then readjusted. The next control scan shows the successful replacing and the continuance in position of the neighbors. 05.2 – PINLFIELD: INDIRECTNESS // The PINFIELD is the first concept that was tested during the research. It was simple, fa1’ and using screwing as a formation process proved reliable. The actuation influences only singular points on the surface. Most of the area is deformed indirectly by local deflection of the adjacent eccentric pins, or generated under the increasing tensile force that appears when the mold surface is stretched. The same adjustment results in different geometrical output curvature, depending on different aspects such as the position or orientation of the desired shape on the mold surface or mold material changes. The high number of varying influences, and the fact that a readjustment of a single pin effects on a whole area invites further research on iterative adjustment process with continuous visual information. 05.3 – CLAY MOLD: ITERATION // The Clay Mold uses a process of plastic material formation where the amount of material is not subtracted during the production process, but rather displaced with any deformation affecting the surrounding area. This reaction to the actuation changes from centre to side, from ridge to

The Nature of Robots

TIVE FORMWORKS // The implementation of sensors has catalyzed recent discourse on robotic in fabrication architectural design and introduced a new emphasis on the value of information in the production process. Results within production cycles so information describing current production states can be used to improve the process. An three formworks were designed independently and with strong differing actuation strategies, material use and expected deformation behavior yet each process was remarkably simple. While moving, screwing, and are processes that are easy to realize by using industrial robots, their detailed effects are hard to control and hard to predict. The successful use of each formwork depended on the implementation of visual feedback during the fabrication process to ensure e the generated form satisfied the defined requirements.


The Nature of Robots

recess, while the implementation of visual sensors enables an iterative production process. Cycle-by-Cycle the stock is scanned, analyzed and the desired form is approximated.


06 – CONCLUSION // At the start of the research project, there was no intention to employ sensory information as feedback within the production process. The production methods were developed at different stages of the research project and seemed to be reasonable concepts, easy to generate and to control. SENSORY INFORMATION was an essential tool for the development and improvement of the concepts, but as FEEDBACK it has shown new possibilities within digital controlled fabrication. All of the formworks presented need not just INFORMATION to generate the production process, but also FEEDBACK on their current state in order to control, readjust or approximate the form in order to achieve the desired shape. The implementation of feedback enables new perspectives on project-relevant production techniques whereby continuous control of formation is possible and feed-forward processes are improved across multiple cycles or even in real-time.


sented project is part of the FFG-Bridge Project 836524” “Shell structures made of Ultra High Performance Concrete (UHPC)”. The concept of sensory information was developed in collaboration with Felix Raspall and Martin Bechthold and part of the workshop “Digital Material Formation” at the Rob|Arch 2014.The adaptive clay mold is part of Florian Landsteiners Master Thesis written at the Institute of Structural Design in 2014-15, Graz

FREE FORM CLAY DEPOSITION in CUSTOM GENERATED MOLDS Producing Sustainable Fabrication Processes

on a robotic tooling path enables a continuous and sustainable adaptation process due to the fact that clay is reusable, can mimic other materials in viscosity and is compatible with a range of sustainable aggregates through to firing stages. This paper describes ongoing research into a two-step robotic fabrication of free form clay printing; namely, as (a) the robotic milling of a sustainable formwork; and (b) as controlled deposition of liquid clay into a form or mold.

Credits // Kate Dunn // Dylan Wozniak O’Connor // Marjo Niemelä // Gabriele Ulacco // ABSTRACT // In a context of free fab printing, this research explores a series of investigations into the potential of 3D printing with clay that address the problems of viscosity, tool paths and setting times. The material of clay is explored here in order to simulate architectural building processes that use both subtractive and additive methods of construction that cannot be performed by a gantry style model of robotics. The use of clay deposition

01 – INTRODUCTION // 3D printing processes are an area of robotic fabrication that enables architectural building processes through additive manufacturing techniques. Free Form Fabrication (FFF) or Extrusion 3D printing processes rely on the extrusion of a material in a pattern determined by an STL file. In FFF, material is extruded successively onto a work bed or plate, whereas in 3D printed robotic processes. Only the robotic reach, and the work bed limitations limit the object size. While studies have been conducted in a diverse range of material from polymer to plastics and clay, specifically the area of robotic clay printing has significance since the material characteristics of clay such as viscosity, tool paths and setting times present a challenge to robotic manufacturing, but also offer the benefit of sustainable adaptation processes.

The Nature of Robots

KEYWORDS // Robotic Free Form // 3D Printing Techniques // Sustainable Tool Path Processes //


The Nature of Robots 196

The research builds on ceramic material and delivery methods as an ideal approach for large-scale 3D printing, due to the properties and material specifications. The unique property held by clay is that it combines the strength of a solid with the fluidity of a liquid, with plasticity in terms of slide or â&#x20AC;&#x2DC;shearâ&#x20AC;&#x2122; as initial response to pressure, and strength attained by a mix of particle sizes and thixotropy, which holds fine particles in a strong network (Hamer 2004). Ceramics or clay in a raw state change consistency, and can mimic other materials through the addition or reduction of water, defloculant, sodium silicate, alcohol and a range of up-cycled industry waste aggregates (crushed materials added to a body to deliver different properties of strength, moisture absorption and refractory qualities). Aspects of dissolvability allow for the use of clay specifically in robotic manufacturing where machinery and equipment are sensitive to material blockage. Furthermore, robotic 3D clay printing in a customizable mold support new work methods that are highly sustainable because the subtractive continued shaping of molds enables reuse and refinement. This paper describes preliminary results from an ongoing research into a two-step robotic fabrication of free form clay printing; namely, as (a) the robotic milling of a sustainable formwork; and (b) as controlled deposition of clay composites into a mold. The research investigates a series of robotic precedents and commercialized 3D printing techniques in order to de-

velop a framework for material and fabrication techniques. It focuses specifically on multi-functional processes including robotic arms while also addressing material sustainability in the 3D printing process, and the potential of up scaling for free form printing. In the following, a background overview is presented. A second part describes the present framework for the research (including technical components of pump, setup of robotic arm and work envelope, limitations of aggregate clay body, and customized end effectors). The paper concludes with a report on ongoing exemplary studies into the custom manufacturing of a mold and slip cast in clay, free form deposition of clay, and robotic deposition of clay within a mold. 02 â&#x20AC;&#x201C; From COMMECIALIZED 3D PRINTING to ROBOTIC APPLICATIONS // The medium of ceramics has a long tradition of forming techniques such as wheel throwing, hand building or casting, but has seen some recent research into robotically controlled investigations into free from deposition and scale adaptations that test the material limits. Ceramics is used in industry and design in a range of modes from slip casting injection molding and wheel forming to 3D printing, which differ in terms of techniques and strategies, use of material body, and deposition techniques. 3D ceramic printing is currently commercialized on a small scale by companies such as Figulos and Shapeways, production costs are high, and there exist quality and stability issues with material shrinkage or

03 – FRAMEWORK for ROBOTIC MILLING and 3D PRINTING // Using a robot to deposit material selectively in to the mold can be significant because this reduces several steps from a traditional ceramic fabrication workflow, and advances the potential of free form ceramic deposition by utilizing the full range of movement available with a 6 axis robot. A continuous trajectory from the subtractive cutting of molds towards the additive deposition of material allows for a simplified workflow, and more importantly demonstrates a two-fold robotic process where one single machine is deployed for multiple fabrication steps (Keating and Oxman 2013). The research developed these considerations into a series of material and robotic path studies into free form clay deposit. Several aspects were included: designing with time; viscosity of material, placement, and hardening; three-dimensional deposit of material (in space); velocity as aspect between tool path and material sedimentation; 3D printing as gestural tracing in material, equivalent to drawing, a comparison of different 3D printing materials and techniques; studying differences between material bodies using clay as a composite material combined with different aggregates. These considerations were tested against the project brief that explores robotic fabrication of a mold, and consecutive clay deposition as two-step processes that combines additive and subtractive techniques. In order to derive a framework, the research tested these criteria through a series of prototyping experiments, in-

The Nature of Robots

cracking that is typical for ceramic or porcelaineous materials. In a context of free form robotic fabrication, research has been undertaken into three-dimensional depositing of plaster within the work envelope of an industrial robot of ‘Morpheaux’ (Bard et al. 2012).’Robosculpt’ (Schwartz and Prasad 2012) shapes molds for fibre-glass chairs by robotically subtracting from a manual packed clay solid. ‘Objects of Rotation’ (Dickey et al. 2014) uses different clay shaping tools in a collet chuck attached to a robot arm that are used to mark or shape columns of clay secured on a clay throwing wheel, and which enables digital automation and up-scaling for clay modeling techniques. A significant research for combination of subtractive and additive combinations of formwork has been undertaken by ‘Woven Clay’ (Friedman 2014). Here, a styrene formwork is created using a router subtracting material, and additive robotic fabrication is applied. The porcelainous clay is deposited by robotic extrusion of a paste style of clay into a woven pattern onto an undulating foam bed, so that the material is set to dry within the formwork. This set an interesting precedent for the combination of different sets of robotic applications within one work process. Furthermore, material processes of ceramics can be exploited through robotic manufacturing for serial tests of surrogate or compounds, such as waste or up-cycled aggregates that address sustainability in large scale 3D printing.


cluding: 1. Subtracting material to shape details by testing the fabrication of molds for slip casting that are fabricated using a spindle router from a block of pottery plaster. While the creation of a mold traditionally requires a master object to cast from, here, sequences of subtractive cutting of forms of the mold can be pre-determined digitally, and continuously removed with completed stages of fabrication. 2. Exploring different aggregate conditions of a material body in order to understand requirements for robotic deposition process of liquid material bodies.

The Nature of Robots

3. Using a robotically controlled extruder end effector to lay material first in a repeating, incrementally stepped up 2D pattern, then at angles, speed and precision unachievable by hand slip trailing or by a simpler gantry frame setup.


04 – WORKFLOW and ROBOTIC FREE FORM CLAY PRINTING // Several technical elements further contributed to the framework, such as the fine-tuning of the pump setup, hoses and deposition bed; the limitations of aggregate body of the clay material; and the development of a customized end-effector, which are discussed in the following. 04.1 – SETUP and WORK ENVELOPE of ROBOTIC ARM // The 3D clay depositing process is prototyped

here on a KUKA KR10-RI100 with a working envelope of approximately 1m3, using the robot as a 3D clay printer. A remote pump draws material from a reservoir, attached to the robot arm with a 6 m hose to utilize the full range of robot reach and axis movement. The extrusion system is configured to work with a volume of material vastly beyond a traditional robotic payload. The KUKA KR10-R1100 is stationed at a standard workbench height of 900 mm on a portable, custom fabricated steel base. Initially a series of intuitive tech pendant movements to test appropriate speed and height for tool-paths was conducted, followed by extrusions for various material clay aggregates with high control in KUKA l prc. 04.2 – TECHNICAL COMPONENTS of PUMP, HOSE and DEPOSITING BED // The research used a Moineau or Progressive cavity pump with a marine grade stainless steel core or driver deposition. It draws from a plastic reservoir feeds the printing material into the chamber of the pump, and into a hose (diameter 25 mm) attached to the robot end effector for extrusion. The extruder setup utilizes an aluminum coupling which holds several “off the shelf” irrigation components to act as the extrusion nozzle during the material testing phase. The current configuration enables researchers to experiment with a range of materials of various viscosities as well as aggregates in an accurate, repeatable and documentable way. In addition, there are a series of control valves to ensure

04.3 â&#x20AC;&#x201C; LIMITATIONS of AGGREGATE BODY of the CLAY MATERIAL-SCALING UP // As a specific focus of the research, different materials for robotic extrusion were tested based on white earthenware casting slip (commonly used for casting) composed of clay particles suspended in a liquid composite body, and with additives including kaolinite, crystalline silica, water, sodium silicate and polyacrylate dispersant (dispex). Furthermore, waste materials were included as aggregates to address sustainability, including recycled paper pulp and softwood sawdust. Maltodextrin and fine sugar were added to the mix to aid adhesion of the layers during deposition, while cellulose fibre and bentonite are added to provide structural integrity. Alcohol (methylated spirit) was added to decrease setting times through evaporation. Initial tests were successfully prototyped by extrusion in the closed form mold. This is significant because it presents an alternative to common slip casting. In slip casting, liquid porcelain or clay is filled into a mold, and water is absorbed by the mold so that the material solidifies, and excess liquid can be emptied. while slipcasting

can include negative aspects such as overspill, material wastage or mistimed absorption, the controlled robotic deposition can improve these aspects by selective delivery of material. 04.4 â&#x20AC;&#x201C; DEVELOPMENT of CUSTIMIZED END EFFECTOR // In conjunction with the material composite studies, several iterations of end-effectors were produced to explore strategies for clay deposition. An electrically triggered solenoid valve with a roll plate coupling resulted in limitations of working with multiple viscosities, because maintaining a consistent pressure with water to correctly operate the valve during movement proofed difficult. Another aspect included the need for different off-theshelf irrigation fittings to enable trials of nozzles with varying diameter. As a result, a modular end effector was developed so that individual components can be swapped for different material experiments, programs, or in the event of passages becoming clogged or damaged. In addition, once experiments moved from 2D toolpaths that either spiraled or stepped up in the Z direction to depositions within 3D forms, a nozzle 150 mm in length was added to enable placement of materials with less movement of the robotic arm, and less danger of collision with the work-piece, allowing increased velocity of movement and a more versatile placement angle.

The Nature of Robots

safety of operation, maintenance and cleaning of the equipment, including one valve at reservoir outpoint, one in the pump cavity, and a third attached to the nozzle or end effector. At a later stage, this setup will be reconfigured to incorporate an air powered piston valve capable of being synchronized with remote control of pump velocity.


The Nature of Robots 200

06 â&#x20AC;&#x201C; CONCLUSION // This paper has discussed the initial stages of an ongoing research in the context of free fab printing as a series of investigations of the potential of 3D printing with clay. These initial tests that have been undertaken enable us to better understand the problem of relationships for robotic printing of clay in terms of mold, viscosity, tool path and multiple times (depositing times, liquid run times, setting times). The research discusses a robotic tooling for a two-step robotic fabrication of free form clay printing; as the robotic milling of a sustainable formwork; and as controlled deposition of liquid clay into a form or mold. The research successfully prototyped first results with the robotic milling of form molds that have minimal waste in the production, and the consecutive deposition with different clay mixtures. The robotic deposition demonstrated an improvement of common slip cast processes and reduction of spillage and material waste, which is part of traditional practices. Furthermore, the use of waste substances from industry as a component of the deposition material was successfully tested, and thus aspects of sustainability could be associated within the context of 3D printing. As a future research trajectory, the process could be adapted to specific site conditions by the including locally sourced materials as aggregates, and furthermore be upscaled towards architecture applications.

3D Printed Interlocking Modules

Credits // Brian Peters // ABSTRACT // The solar Bytes Pavilion is a temporary structure that highlights a potential for architecture, where buildings are fabricated using new techniques (robot arm, 3D printing), incorporate smart technologies (light sensors) and are powered by renewable energy sources (solar power). Taking advantage of a robot armâ&#x20AC;&#x2122;s strength and range of movement, the pavilion was 3D printed with an experimental extruder and the result is a structure comprised of ninety four unique modules that charge during the day and glow at night. KEYWORDS // Robotic Fabrication // Parametric Design // Digital Fabrication // Robotic Manipulation // Renewable Energy // 3D Printing // 01 â&#x20AC;&#x201C; BACKGROUND: 3D PRINTED ARCHITECTURE // Currently, several architects and engineers (Khoshnevis et al. 2006; Buswell et al. 2006; Fixsen 2015) around the world are creating large-scale 3D printers in order to

produce large-scale structures. They are utilizing various 3D printing techniques and exploring different materials, such as concrete, adobe, artificial stone and plastics. At the same time, there is an emerging, alternative approach to 3D printing architecture that is inspired by an ancient building component; brick. Smaller machines are utilized to fabricate smaller building blocks that assemble to form larger structures (Peters 2014), which was a technique that helped drive this research project. There are two approaches currently being utilized to 3D print plastic with a robot arm. The first is printing with a standard fused deposition modeling (FDM) technique, where successive layers are printed on top of each other and rely on the layer below for structural stability. Dirk Vander Kooij was one of the pioneers of the FDM style technique, printing functional furniture in recycled plastic. The second approach relies on the plastic quickly drying and hardening in midair during the printing process, allowing one to print without support below. This technique was first introduced with the use of resin in the Mataerial project (Laarman et al.2014) developed in collaboration between Joris Laarman Studio and the Institute for Advanced Architecture of Catalonia (IAAC). Subsequently, there have been many projects investigating this technique, such as the Mesh-Mold by Gramazio and Kohler, as well as many others. While these recent projects highlight the geometric possibilities of utilizing plastics and 3D printing

The Nature of Robots



The Nature of Robots

with a robot arm, they are not creating structural enclosures.


02 â&#x20AC;&#x201C; PROJECT GOALS // The Solar Bytes Pavilion is the result of research with multiple goals and is the first of two iterations of a full scale, 3D printed pavilion. The first overarching goal for these pavilions is to 3D print a fullscale structure using a relatively small machine. This approach requires individual modules to be printed and joined together to create the full form. It was decided early in the project that interlocking joints would be developed to eliminate the need for mechanical fasteners between the modules and allow the structure to be self-supporting. A second goal for this project was for it to serve as a test of the structural capability of the 3D printed plastic, as there would not be a separate structural framework. Finally, the intent was for the pavilion to glow at night, while not being connected to the grid, and therefore solar powered lights needed to be integrated into the design. The Solar Bytes Pavilion serves as a prototype for a second structure and was analyzed both during and after its fabrication to determine its performance and consider improvements to the system before embarking on the design of the second structure. The overall form of the pavilion follows the form of a catenary arch that is then extruded to create a barrel vault. The vault is then slightly skewed from north to south to help with the overall stability of the pavilion. The vault was then positioned to follow the path of

the sun, spanning from east to west to maximize solar exposure for the solar powered lights. Once the form was established, the structure was subdivided into modules. Several subdivision patterns were experimented with, such as triangular, Voronoi, and hexagonal. The hexagon was determined to be the best option for two reasons: (1) the smaller surfaces of the sides/ faces of the modules limited warping during printing, and (2) the shape offered efficiencies in both material and fabrication time. For example, initial tests were performed with triangular patterns, however the prints often warped because the three faces of the module had large, flat surface areas. While the hexagon proved the most reliable to print, it is less resistant to compression loads. Therefore an additional interior wall was incorporated into the design of the base to resist those loads and produce a very stiff and resilient module. The final pavilion is constructed out of 94 of these hexagonal modules that each has an integrated solar powered LED at their peak. All of the modules are unique, ranging in size from 35 to 48 cm in both width and length, and with a median height of approximately 300 cm. The modules were 3D printed with translucent plastic, allowing the structure to filter sunlight during the day and create a uniform glow at night. The light effect is enhanced by the use of interlocking, dove-tail joints that reduce the visual division between each module, which ultimately creates a light-weight structure, both

03 â&#x20AC;&#x201C; PARAMETRIC MODEL // A parametric model (Grasshopper definition) was developed for this project, which was essential from design to fabrication. There were several key design parameters that were controlled by the definition, such as the subdivision pattern, size and shape of the interlocking joint, and the height of the peaks that contain the solar powered lights. Fabrication parameters could also be manipulated in the parametric model, including the printing speed,

layer height, and fabrication code. The prototyping process was expedited by creating a direct link between the parametric model and physical tests, which minimized the time, spent redesigning and modifying the module. This was essential for the research, since the 3D printing material and technique were highly experimental. The final part of the definition was a custom script to contour each module into a series of polylines stacked in the Z-coordinate, and then subdivide those contour lines into XYZ-coordinate that were then used as the G-code for the robot arm. The KUKA l prc plugin was used to translate those XYZ-points into a programming language specific for KUKA robots. 04 â&#x20AC;&#x201C; FABRICATION: MATERIAL // One of the main design objectives of the project was to create a glowing pavilion at night, which required the use of a translucent material that can be 3D printed. There are only two materials that fit these criteria: glass and plastic. Since 3D printing with glass is still difficult at this scale, plastic was used. There were several obstacles that needed to be understood about printing with plastic that significantly influenced the design of the project. The first was which material could be used with the extruder. At the outset, the project aimed to use either polylactic acid (PLA) or recycled plastic. There would have been two main advantages to using PLA versus polypropylene (PP), polyethylene (PE), or polyvinyl chlorine (PVC). First, PLA is a bio-based plastic, meaning that it can

The Nature of Robots

visually and structurally. The arch form of the pavilion highlights the structural capabilities of the modules, which were designed to carry the load of the structure in compression with minimal reinforcements. Two thin-gauge steel arcs on each end of the pavilion help resist wind loads and prevent the pavilion from moving laterally. The pavilion is secured to the ground with a series of stakes that attach to the steel arcs and assist in both resisting tipping and sustaining the compression vault. Compression strength testing was performed on the modules to determine the weight a module could withstand before structural failure. The tests revealed that a single module could withstand approximately 450 kg of pressure. Since the modules were printed in plastic, they were susceptible to bending before the failure, but compared to more brittle building materials, such as ceramics or concrete, plastic offers a great strength to weight ratio, as the each module weighs approximately 1.3 kg.


The Nature of Robots

be created from plant-based starches. And second, PLA warps less because the rate of shrinkage is minimal (www. However, at the time of final production it was not viable because it could not be consistently extruded during the prototyping phase. A translucent “natural” PP filament, with no added color, was ultimately used to achieve the design goal of light transmission at night and warped far less than polyethylene.


04.1 – FABRICATION: 3D PRINTING EXTRUSION // The modules were 3D printed using a DOHLE hand welding extruder, the Mini CS that was attached to a KUKA Agilus and utilized a FDM style printing process. The extruder was capable of extruding half a kilogram of plastic per hour by accepting 4 mm thick PP plastic filament that was fed from a spool suspended above the robot arm. The modules were printed with a continuous extrusion; material flowed from the print head without stopping and starting, following a series of vertically stacked printing paths composed of polylines. The goal for each polyline was to create an unbroken path that was both structurally stable and time efficient. The size of the extrusion resulted in rigid and strong print walls that were roughly 1 cm thick. The fabrication time for each module ranged from 3 to 4 h depending on their size. Since the modules required absolute precision during the fabrication process to allow for the interlocking joints to work, the robot arm provided an extremely pre-

cise tool, as well as a very stable base to mount the extruder. 04.2 – FABRICATION: DELAMINATION // One of the biggest challenges with FDM style plastic 3D printing is the possibility of delamination between layers either during or after printing. In this printing process, material is extruded onto the layer below and therefore relies on the bond between those layers to produce a structurally sound object. Delamination is often the reason for an object’s failure and can be the result of multiple variables. First, each type of plastic has a unique rate of shrinkage when undergoing the phase change from liquid to solid, which can affect the surface area between layers. Second, the temperature of the printed material, printing environment, and the cooling method are all factors. If the printing temperature is too low, then the material will not be hot enough to bond with the layer below, and if the temperature is too high the material will burn. Cooling the printed plastic effectively is also one of the subtle intricacies of the printing system, since if the material is cooled too quickly it will delaminate, and if it is cooled too slowly the print will begin to warp significantly. The initial prototypes often failed due to delamination, so the layer height and 3D printing speed were both reduced, while the printing temperature was increased. The initial layer height was set at 2.5 cm, but was ultimately adjusted to 1.8 cm, while the printing speed was reduced from 20 - 10 mm/s. The printing temperature, however,

05 – INTERLOCKING JOINT CONNECTION // A primary design feature of the pavilion was the development of interlocking joints between the modules. It was inspired by a traditional method of construction, stereotomy, which has experienced a rebirth with the rise of multi-axis digital fabrication tools. However, it utilizes an additive, rather than subtractive, process where complex interlocking blocks are 3D printed rather than CNC milled out of a solid material. Since the modules are unique, each one was designed and fabricated for a specific location in the assembly. The interlocking connection is based on a sliding joint, where each of the six sides of the hexagonal modules has either a male or female connec-

tion. The unique ails and pins of the isosceles trapezoid shape are reminiscent of a dovetail joint, which works to prevent the modules from moving and is critical to the pavilion holding its form. Since the underside of the pavilion is exposed to reveal the structural geometry, the design of the interlocking joint became a design feature rather than merely a structural detail. The joints also allow for easy assembly and disassembly of the pavilion, which is aided by unique numbers embedded into the base of each module that indicate its position within the overall form. Several rounds of prototyping were needed to perfect the interlocking connection. The joints needed to be loose enough so that the modules can easily snap together, but not so loose that they ruin the structural integrity of the pavilion. 06 – SOLAR POWER // The module design includes an integrated solar powered light that comprises a small photovoltaic panel, a rechargeable battery, an LED light, and a light sensor. Each solar cell acts independently, capturing and storing energy as well as sensing light levels for each individual LED, leading to an autonomous system that reacts to its isolated solar conditions. This creates a recording of the sun’s exposure over the course of the day that is then reflected at night as the LEDs light up. For example, if it is cloudy in the morning and clear in the afternoon, the pavilion’s east side will be illuminated for a shorter period of time than the west side, reflecting

The Nature of Robots

was increased from 104° to 121°. To cool the plastic during the printing process, an air nozzle was integrated into the design of the extruder. During the prototyping stage, several variables were tested, such as temperature and direction of airflow, and it was determined that the nozzle should be directed directly after the point of extrusion and set at a temperature of 21°C. To better understand the lamination process, an infrared camera was used to examine the thermal properties of the plastic and study the bonding strength between layers. The thermal imaging camera was able to visualize the complex thermal properties of the 3D print during the printing process, and was integral to identifying which settings red to a stronger bond between layers.


The Nature of Robots

those environmental conditions. The solar powered LEDs are installed at the central peak of each module to evenly disperse light within the interior of the module and add to the glowing effect of the pavilion.


07 â&#x20AC;&#x201C; CONCLUSION // This pavilion is an achievement in 3D printing a fullscale structure; nevertheless there are multiple design improvements that will inform future fabrication. For example, a subdivision pattern that better highlights that each module is unique should be utilized, since it was hard to perceive that each module is different. Additionally, the depth of the structure could vary to use less material on the top than bottom and to improve material efficiency and structural integrity. Thirdly, a double curved surface could aid structural stability and perhaps further eliminate secondary reinforcements. During the fabrication process, connecting and con trolling the extruder through the robot arm programming language could permit stopping and starting during printing without stringing between the openings. Another fabrication improvement we are currently experimenting with is using an extruder that can accept plastic

pellets, which could lead to the utilization of material mixtures that include recycled plastics, bio-based materials and composites. Finally, a larger extruder could decrease the relatively slow fabrication time. As a future research, the Solar Bytes Pavilion will proceed to take full advantage of the geometric freedom that a robot arm offers. In a second iteration, the pavilion will be recycled by 3D printing a completely different structure using the same plastic. Modules will be shredded into small pellets and directly feed into an extruder that prints with recycled plastic. We plan on fabricating a larger structure with a similar set of goals, but apply knowledge gained during the design and fabrication of the Solar Bytes Pavilion.


ality has been approached from both digital and physical perspectives. At a digital materiality level, a customized computational design framework has been implemented for form finding of compression only structures combined with a material distribution optimization method. Moreover, the chained connection between the parametric design model and the robotic production setup has enabled a systematic study of specific aspects of physicality that cannot be fully simulated in the digital medium. This established a feedback loop not only for understanding material behaviors and properties but also for robotically depositing material in order to create an informed material architecture.

Credits // Sina Mostafavi // Henriette Bier // ABSTRACT // This research presents and discusses the development of a materially informed Design-to-Robotic-Production (D2RP) process for additive manufacturing aiming to achieve performative porosity in architecture at various scales. An extended series of experiments on materiality employing robotic fabrication techniques were implemented in order to finally produce a prototype on one-to-one scale. In this context, design materi-

01 â&#x20AC;&#x201C; BACKGROUND: 3D PRINTED ARCHITECTURE // Materially informed Design-to-Robotic-Production (D2RP) systems explore the extents to which rapid and flexible robotic fabrication methods can inform and enhance established generative design to materialization and production practices. In the case study of this paper, the focus is on experimentation with optimized material deposition for a compression-only computationally derived topology. The study has explored the possibilities of designing and fabricating material architectures with various levels of porosities, ranging from architectural (macro) to material (micro) scales. By employing

The Nature of Robots

KEYWORDS // Informed Design // Robotic 3D Printing // Porosity // Material Architecture // Material behavior //


The Nature of Robots 208

performative and generative design methods, industrial robotic production techniques and material science experiments, the D2RP aims to close the loop from design to 1:1 scale fabrication. With this goal, the main research components of the presented case study deal with specific aspects of materiality in relation to design computation and robotic 3D printing. In this context, the chosen fragment of a computationally designed pavilion required translation of the optimization results from a finite geometry into a continuous robotic path for material deposition in order to create an applicable material architecture. The integration of physical material properties into design by means of digital design interfaces and computational design methods has been explored in both practice and academia. The historical survey with respective related cases and paradigms is not within the scope of this paper but relevant to the goals of the presented case. In order to position this project in this larger field of research two major types of approaches have been identified. One presents cases in which, in order to study design materiality, the design system relies only on virtual modeling, simulation, analysis, and abstraction of physicality through implementation of certain computation methods such as Finite Element Method (FEM), Computational Fluid Dynamics (CFD), Particle Systems, etc. The other one presents material experimentations and the design system focuses mostly on constraints and potentialities of certain material

and/or fabrication method that is integrated into digital modeling platforms, i.e. parametric design models. The proposed D2RP system establishes a feedback loop between the two approaches. In order to achieve this goal, at digital materiality level, a systematic and chained strategy for design information exchange is established by designing and implementing a customized parametric form finding system for compression-only structures combined with topology optimization. At physical level, the direct connection to the robotic production system, in addition to improving the production method has led to the direct study of certain aspects of physicality that cannot be fully modeled inside the digital design platform. Therefore, the production system becomes not only a means of fabrication but also simulation. Recent advances in both robotics and 3D printing have introduced new approaches towards architectural materialization and production. Considering materiality and architecture at multiple scales, there are a few projects that successfully bring the two together. In some examples a scaled up printing machine is employed to horizontally deposit layer-by-layer building material. The explored and presented robotic 3D printing project proposes an alternative method of material deposition aiming to create a multi-dimensional material architecture. This is achieved while taking the behaviors and properties of the implemented material into consideration, which in this case is ceramics, as well

02 â&#x20AC;&#x201C; D2RP DEVELOPMENT // The D2RP proposes a roadmap for development and improvement of a robotic 3D Printing technology for fabricating 1:1 building components. The roadmap includes three initial case studies, concluding with creating a direct link between design and production: multi-colored light robotic 3D printing, robotic pattern studies, and ceramic robotic printing. Multi-colored light robotic 3D printing involves mounting a color changing light source on the robotic arm. This project addresses the connection established between motion and information extracted from the virtual 3D model. Being able to study the three-dimensionality of robotic motion contributed to developing a new approach to 3D printing, different from the slicing-in-layers printing technique. This provided possible directions for defining a 3D printing method, in tune with the structural

characteristics of the final prototype. The study of robotic motion defines the boundaries of the digital design-space in relation to the physical solution-space. This informs the parametric setup with ranges of reachability and optimized orientations. It also contributes to being able to maximize the overall space used. In addition, by numerically controlling the on-off light pattern and light colors by means of an Arduino microcontroller, the team reached the goal of further extending design possibilities in such a way that multiple materials can be deposited at certain coordination based on the information extracted from the virtual 3D geometry. As the first step, any given curve, in digital, is reproduced, in physical, with multi-colored light curves captured by means of long exposure-time photography. Later this approach is tested on the compression-only designed pavilion represented by a network of curves. As part of the second set of preliminary studies, the robotic pattern project focuses on drawing geometric patterns that explore variation in densities and resolutions to reach the desired porosity. This informed the design of robotically controlled routines for material deposition to reach a functionally graded structure. The established parametric system, derived from these experiments, involved size of the overall shape, thickness of nozzle for material deposition, number of targets to describe the robotic motion and the method of approaching defined targets. As a consequence the team formulated two categories of

The Nature of Robots

as by integrating material optimization routines in the D2RP system. D2RP consists of four main research components: Design computation, tooling/production set-up, robotics, and materiality. Each set experiments design and exercises presented in the following section, explores possibilities, of integration by establishing feedback loops between the four components. Parallel to the lab-based explorations for the development of the D2RP a studio design project was considered as a pilot case study. In this project architectural and Material porosity at various scales is considered as the main design driver.


The Nature of Robots

material deposition: Continuous flow and on/off numerically controlled flow patterns. The ceramic robotic printing explores possibilities of production of 3D printed building parts and establishes a production method where all parameters are calibrated for the developed physical set-up. The team designed an extruder connected to an end-effector mounted on the head of a robotic arm, where the material source was exterior to the robotic arm in order to maximize the freedom of movement, in order to achieve an optimum multi-dimensional material-architecture. Initial experiments ranges from simple layer-by-layer material deposition to study material flow to 3D dimensional printing on doubly curved surfaces. Considering the fact that natural materials are not fully predictable several material properties like plasticity, viscosity, flow rate and short-term material behavior were investigated and documented at different robot-motion speeds in order to provide complete information sets for the next prototyping phase.


03 â&#x20AC;&#x201C; DESIGN and PROTOTYPE // In order to develop a coherent computational design system specific to this project, the first step was to implement methods for form finding of compression-only structures, derived from the innate characteristics of the material. In addition to eliminating tension forces in the derived topology, this part of the design system was implemented as a parametric strategy to define the porosity at macro or architectural

scale to fulfill certain functional and locational requirements. Furthermore, in order to achieve the micro porosity level, a finite element method for material distribution optimization was implemented on a part of the designed pavilion. The optimization also considered local and global load and support conditions. To implement a generic and repeatable method on other parts of the topology, the challenge was to be able to parametrically change the method of finite-mode geometric representation like point cloud and mesh to a vector based or NURBs (non-uniform rational B-spline) geometry. This was achieved by applying a segmental system in the very initial topology, retrievable at different stages of form finding and parametric geometric transformation. By applying the computational design system several configurations were generated, in each distributing the compression only material where needed and as needed, while taking the structural performance at both macro and micro scales into consideration. The challenge of the next step was to materialize these differentiated densities by creating unified topologies that express structural loads consistent to the design approach and robotic fabrication potentialities and constraints. At this stage, various algorithmic form finding and optimization techniques, mostly in the Rhino-Grasshopper platform and Python scripting language, were applied. This allowed the systematic exploration and evaluation of design alternatives in the design-solution space, eventually providing the

through stability due to significant mass and specific geometry. What the study aimed to prove was that by controlling the geometry and the material deposition, compression structure could become lighter, and significantly improve their material cost and their thermal insulation performance. A way or achieving material deposition optimization is by controlling the parameters of the production setup. This is briefly described as follows: the extruder system designed and built by the D2RP team manages a plunger-based mechanical extruder of a paste of ceramic-clay, water and specific color-pigment that increases gluiness. The numeric control of clay extrusion was experimented and valuable results for dynamic extrusion were recorded, while implementing a discontinuous porous pattern. But due to shifting research objectives, only continuous clay extrusion was used for the fabricated prototype. Therefore, a custom design routine was developed in order to extract a continuous motion path to generate the designed material architecture. To achieve continuous material deposition, similar to the challenge of translating mesh to NURBs in macro scale, in micro lever a generic parametric system is developed to translate the discrete result of optimization to continuous curves. In brief, the method involved picking a starting point and recursively searching in the extracted point cloud to generate a continuous spline. From topological and computational point of view, this helped the system to directly and efficiently pro-

The Nature of Robots

required information for production, path generation and kinematics simulation with the ABB-Robotstudio. Simultaneously, the initial material experiments and information sets informed the design process, design materiality and robotics. This was achieved through step-by-step documentation of a series of purposeful design-to-robotic production experiments with fixed and variable parameters. Specific to this project, the resulting dataset provides information on the possibilities of the developed D2RP system for robotic ceramic 3D printing, such as maximum angle of cantilevering, maximum length of bridging material without supports, minimum and maximum size of the nozzle, material flow, motion speed, etc.For production purposes, the topology of the pavilion was sub-divided into unique components. As the research progressed it became apparent that due to the significant variety custom building components featured in the design, the robot manufacturerâ&#x20AC;&#x2122;s software functionality needed further customization. For this purpose a link between the design and the simulation environments (Rhinoceros platform and its add-ons) and the rapid code interpreter of Robotstudio has been implemented. This direct link between the design model and robot controller enabled the implementation of a greater range of unique, longer, continuous tool paths. As a construction material, clay-ceramics is commonly used for compression-only structures. The structures based on compression perform


The Nature of Robots

vide an applicable tool path, considering material properties, behaviors and hardware-software specifications of the developed D2RP system. Throughout the process the extrusion speed was adjusted empirically according to observed structural and aesthetic considerations. Extrusion parameters were controlled through line-size and nozzle customization at the tip of the robot end-effector. We experimented with nozzles of various profile sizes and shapes. For the fabricated prototype, a nozzle featuring a square 1 cm2 aperture was used. Finally, within the studyâ&#x20AC;&#x2122;s agenda of 1:1 fabrication and architectural performance aims, it can be concluded that the prototype achieves both improved 3D printing speed and reliability.


04 â&#x20AC;&#x201C; PROTOTYPE // According to the design brief, the architectural object was relating to the surrounding environment via pores of varying in size according to functional, and structural requirements. The fabricated fragment explores these connections, materializing a piece of structurally optimized compression only urban furniture at 1:1 scale. While developing a customized design-to-production setup, the team achieved optimization in motion path generation. Common 3D printing techniques employ non-differentiated routines for slicing and ordering material layers into motion paths. The prototype was produced embedding fabrication potentialities and constraints into the design. It must be noted that, although the computational 3D model comes close to the actual prototype,

the two entities remain different mainly due to emergent material properties. Differences between virtual and material exemplify emergent aesthetics inherent to the material behavior. The emergent aesthetics inherent to the prototype are as much due to the 3D layering technique as to how material extrusion varies along the path. 05 â&#x20AC;&#x201C; CONCLUSION and DISCUSSION // Advancements in robotic building as presented in this paper indicate that future building systems are customizable and increasingly robotically produced and operated. The presented D2RP system demonstrates that informed porosity in additive manufacturing is relevant for the development of materially informed architecture. Porosity at macro (building), meso (skin), and micro (material) scales implies optimization of spatial configurations and material distribution. Using this approach we strive not only to control mass-void ratios but also to achieve an integrated design, from overall building configuration to the architectural material. In the context of the third and fourth industrial revolutions, the flexibility of such D2RP system can be understood with respect to the interaction between designers, users, and NC systems aiming to produce highly customizable and on-demand building components. Robotic Building (RB) eliminates the current problem of missed optimization opportunities due to a fragmented and sequential process of architecture engineering manufacturing. In a larger

ACKNOWLEDGENTS // This paper has

profited from the contribution of the Robotic Building team (authors, Ana Anton and Serban Bodea) and Hyperbody MSc 3 students (Fall semester 2O14). The project presented has been sponsored by 3TU. Bouw Center of Excellence for the Built Environment, Delft Robotic Institute, 100 % Research and ABB Benelux.

The Nature of Robots

context, the additive D2RP approach presented in this paper is part of the Robotic Building (RB) project, which focuses on linking design to materialization by integrating multiple functionalities (from functional requirements to structural strength, thermal insulation, and climate control) in the design of building components. Scaling up the technology of 3D printing from object to building was the specific goal of the presented case study. This was achieved by integrating the technology in an informed, chained design-to-production system, in which the 3D printing and robotics are not only ways of manufacturing but also methods and tools for simulation and testing of certain aspects of materiality, which lead to new opportunities for design exploration and creation. For the authors, it was important to develop the technology not as an isolated node but as an integrated working-operating module connected to a real-life design problem. The main consideration in architecture and building construction is that the factory of the future will employ building materials and components that can be robotically processed and assembled. This requires the development of multi-materials, tools, and robots, for D2RP processes that will be implemented incrementally in the next phases of this research.



stress lines are curves that indicate the optimal paths of material continuity for a given design boundary, the proposed stress-line based oriented material deposition opens new possibilities for structurally-performative and geometrically-complex AM, which is supported here by fabrication and structural load testing results. Called stress line additive manufacturing (SLAM), the proposed method achieves an integrated workflow that synthesizes parametric design, structural optimization, robotic computation, and fabrication. KEYWORDS // Robotic Fabrication // Additive manufacturing // Principal Stress Line // Oriented Material Deposition // Fused Deposition Modeling // Topology optimization //

The Nature of Robots

Credits // Kam-Ming Mark Tam // James R. Coleman// Nicholas W. Fine // Caitlin T. Mueller //


ABSTRACT // This research presents a new robotic additive manufacturing (AM) framework for fabricating 2.5D surface designs to add material explicitly along principal stress trajectories. AM technologies, such as fused deposition modeling (FDM), are typically based on processes that lead to anisotropic products with strength behavior that varies according to filament orientation; this limits their application in both design prototypes and end-use pans and products. Since

01 â&#x20AC;&#x201C; INTRODUCTION // contemporary additive manufacturing (AM) technologies, such as fused deposition modeling (FDM), compliment earlier CAD advances to enable complex geometric exploration. Current 3D-printing platforms, however, conform to a traditional CNC-based paradigm, in which machine operation is isolated from design. With a limited interface between conception and materialization, designers have few opportunities to affect the qualities that the fabrication embeds in the final artefacts. This research presents a new robotic-enabled FDM technique that is structure-based and material-centric, to enable the systematic reproduction of high quality, and performance based printed structures for

strength. Recognizing the directionally dependent performance characteristics of FDM processes, filaments are aligned to major axial networks of curvatures called principal stress lines. 02 – STRESS LINES: THEORY, SUITABILITY and ADDITIVE MANUFACTURING // The emergence of structural analysis tools within common design platforms, such as Karamba 3D3 and other similar plugins, has created an environment favorable to stress-line-inspired fabrication, yet often structural patterns are employed without being substantiated by structural logic. Precedents combining robotic fabrication with stress lines include the ICD’s Leichtbau BW Installation, and the GSD’s Robotic Beat Rolling. In contrast, this research focuses on the production of stressline-based surfaces with enhanced structural performance. Stress lines are numeric integrations of principal stress directions over each infinitesimal element that comprises an investigated structural body. Designers are interested in principal stress lines because they provide a visualization of the natural force flow in a structure, which shows the lines of desirable material continuity for a given design domain. This characteristic is evidenced by the typical convergence of results obtained from analytical and numerical optimization procedures and from the principal stress lines generated for the same design boundary. Theoretical properties of stress lines support their application in FDM: first-

The Nature of Robots

a given 2.5D (surface) design boundary using common software interfaces and electronic components. Called Stress Line Additive Manufacturing (SLAM), the proposed method is contextualized within a broader conceptual structural design methodology; the integrated software-hardware framework addresses FDM-today’s most common AM method and seeks to facilitate flexible design-space exploration and fabrication standardization. From standardized 3D printing platforms, such as the hobbyist and professional options developed by MakerBot Industries and Stratasys, to full scale AM systems like D-Shape and ContourCrafting, layer-based conventions reduce aesthetic quality, material efficiency, and geometrical accuracy. Particularly problematic for FDM-based techniques, layer-based conventions produce anisotropic material behavior with strength and ductility properties that vary significantly depending on the filament orientation: the tensile capacity of specimens loaded perpendicular to filament orientation can be up to 50% weaker than specimens loaded in parallel, because the weak fusion between horizontal layers provides a natural weak point for breakage. These problems limit both the durability of the printed specimen, and the end-use application of AM. This paper research on FDM’s most fundamental mechanism: robotic flexibility is used to directly achieve filament depositions that conform to desired material behavior-printed along trajectories revealed computationally to enhance assembly


The Nature of Robots

ly, conventional numerical optimization methods, such as ground structure and homogenization methods, tend to produce complex results that are computationally exhaustive to manipulate, and difficult to summarize as line-based paths. Stress lines offer computational ease for fabrication and optimization information. Secondly, design abstraction is enabled as stress line methods are highly suitable for geometric exploration, so that results can reflect design boundaries, regardless of the scaling of material properties, applied force, or the objects dimensions. Thirdly, geometric compatibility is significant. Since stress distribution for elastic continuum bodies are also continuous, stress lines create contour-like fields with curvatures that typically traverse from one design boundary to another: these properties lend to their consistently-clean robotic deposition and efficient linkages as printing paths. Thus, a framework becomes available from stress line computation to load testing to robotic fabrication.


03 â&#x20AC;&#x201C; METHODOLOGY // To capitalize on the critical relationship between architectural geometry and structural behavior in order to generate innovation in each, the project combines structural design space exploration, topology optimization, and robotic fabrication, and introduces a physical testing protocol in order to deliver structurally informed geometric feedback. It builds on previous research into stress lines where computation is broken down into initialization, gen-

eration, and processing. The design scope is here a 4-support grid shell with application for design space exploration with 2.5D membrane structures. The fabrication environment contains a custom extrusion module that is mounted to a KUKA KR6 R900 small robot that is located centrally inside a contained envelope. For the prototyping, the project used 3D modeling software McNeel Rhinoceros and the scripting plugin Grasshopper, with structural analysis conducted using the plug-in Karamba on a surface design form-found using Kangaroo Physics. Robot programming followed a precise set of criteria for the KRL code, with robot instructions generated in RobotMaster, a plugin to MasteCAM.6 The print surface (milled from a laminated MDF block) is positioned at an eccentricity from the robot to minimize joint issues. PLA plastic is used as filament material, for its adhesion capacity to MDF and printability on an unheated surface without warping. Adapted to SLAM, a multi-objective processing method was developed to balance aesthetic, fabrication, and performance objectives. Essentially, the optimization procedure assembles a new 3D-frame structure in each iteration with stress lines that are heuristically selected by a genetic algorithm. A FEA is conducted to determine the strain energy, normalized by material volume, of the new frame structure with each iteration; the combination of stress lines minimizing total strain energy is selected for materialization. Specifically, each stress line within the

04 – ROBOTIC TOOL and WORKFLOW DESIGN // 04.1 – END EFFECTOR DESIGN // Referencing current extrusion devices in consumer-grade applications, the custom extruder is composed of a waterjet-cut aluminium frame that is mechanically coupled to the robot via a pneumatic tool changer. The aluminium frame holds a commercially available extruder and control electronics. The Signstek extruder accepts 1.75 mm PLA (polyactic acid) plastic and is composed of a 1.8 degree stepper motor, heating element, thermistor, and cooling fan. Using an Arduino Uno microcontroller and a N-Type MOSEF, the closed loop temperature control (PWM) of the extruder nozzle was utilized. The KUKA’s 24 V signal outputs were monitored by the on-board Arduino and were used to start and stop the stepper motor/extrusion. Control of the stepper motor was achieved with an EasyDriver board that uses an Allegro A3967 motor driver chip. 04.2 – END EFFECTOR DESIGN // The typical workflow consists of the following iterative and trial-and-error procedure: 1. Import of print surface and stress line data from CAD into CAM software; 2. Specimen positioning in the workspace; 3. Print surface calibration in the CAM model space; 4. Clustering of paths based on robot work volume;

The Nature of Robots

base field is assigned a binary value that determines its inclusion or exclusion in or from the assembled frame structure for each iteration. The list of binary values for each of the two principal planes is in turn generated according to a ratio of on- to off-values — a ratio that acts as proxy — for stress line spacing corresponding to various global stress line densities. Thus, an infinite design space is characterized only by two design variables in the procedure. With the density of the stress lines optimally calibrated, a series of rules-based corrections are iteratively applied to the resulting stress line topology in order to achieve additional improvements in the results. A rule-base system can codify and implement existing rules, and can expand to accommodate additional rules as the experimentations continue to develop new knowledge. Rules that were designed specifically for the SLAM framework to improve filament extrusion ease include the removal of line segments in areas with significant stress line overlap, and the realignment of otherwise converging stress lines in highly stressed areas. General structural rules include modification of stress line curvatures to facilitate force transfers at intersection nodes, and the insertion of bracing members. Although the initial implementation of the rule-based system was largely dependent on manual operations, the process is to be automated in future developments.


5. Assigning stress-line-based geometry for toolpath genera tion; 6. Iteration of possible configura tions in CAM and RobotMaster interface; 7. Optimization and simulation of robot trajectory; 8. Export KRL code and run pro gram in KUKA.

The Nature of Robots

Due to robot work volume and reach limitations, stress-line-based paths are clustered into a series of separate print programs. Reasonable estimation of the KUKA arm’s limitation guides the global clustering of stress lines corresponding to different print surface orientation, whereas geometric similarities guide the clustering of stress-line-based path internal to each surface orientation. Next, the stress lines assigned to each cluster are linked, and sequenced in KUKA’s a way that minimizes the total travel distance. Joint configurations and toolpath generation parameters may be altered to affect the filament quality. These quality determinants include:


1. Offset // To ensure a consistent quality in the printed filament and improved adhesion between filament and print surface, the offset between the nozzle and the printed surface is set at a level to allow the tip of the nozzle drag along the top of the extruded filament uniformly across a stress curvature.

2. Extrusion activation timing // To mitigate the loss of filament due to a residual pressure gradient across the nozzle and to ensure the production of normalized flow at the start of new stress-line-based paths, the filament is retracted immediately after the end of a tool path to break connection with the previous filament, and extruded again just prior to recommencing print for the stress line. 3. Travel and move rate // The robot’s movement is measured by its move rate, which differs from the actual travel rate of the extrusion tip over the print surface to rotational requirement of the joints that are specific to the geometries of each stress-linebased path. Typically, decrease in the travel rate corresponds to the thickening of the deposited material, whereas increases may cause poor adhesion. 4. Tool axis orientation // 5-axis settings achieve the greatest consistency in the filament’s cross section profile, thus leading to the best aesthetic and structural performance. However, 4-axis settings can be used when joint rotational and collusion issues prohibit stress-line-based paths to be printed at complete normalcy. 05 – RESULTS: QUALITY and PERFORMANCE // Several surface topologies corresponding to common loading cases were produced for the 4-support grid shell case geometry. The superimposition of several layers of stress lines based on different loading conditions to induce addi-

failures occurred in the shell where tensile stresses were predicted to occur. More significant advantages for the SLAM method are expected in complex geometries and loadings that induce more tension. 06 â&#x20AC;&#x201C; DISCUSSION // There are several important directions for future work in SLAM that can be broadly categorized into four objectives: 1. 2. 3. 4.

Continue to investigate the ma terial and strength behaviour of artefacts printed using the SLAM method; Standardize and improve the SLAM procedures to achieve better precision; Develop a computational framework to automate the generating stress-line-based structures and robot paths; Continue to explore and ex pand the design applications of SLAM.

Considering that SLAMâ&#x20AC;&#x2122;s development was motivated by the recognition of anisotropy in artefacts printed using conventional FDM methods, future SLAM research will continue to seek validation on the merits of oriented-filament deposition. Particularly, research is underway to confirm the tension capacity of prints using the SLAM method, since the form-found 4-support grid shell geometry used in the initial implementation is not expected to experience significant tensile stresses. Modification on the boundary conditions of the grid shell, such

The Nature of Robots

tional structural resiliency was also explored. The SLAM methodology s relatively successful at achieving complex, structurally-meaningful geometries and topologies. To validate that SLAM-produced specimens perform better than conventionally 3D-printed parts, a comparative load test was completed on a number of specimens: three printed using the SLAM method, and four printed using a conventional layer-based 3D printer. The load test consists of a single centralized vertical point load that was applied until a peak load was reached. MakerBot was selected as the technology to be compared to the SLAM method, as it is one of the most popular consumer-grade 3D printing platform available to designers that uses PLA plastic. An effort was made to ensure that all specimens have similar total material volume. The MakerBot MB prints included both a solid constant-thickness shell (labelled M-C) and three variously-discretized shells, which include a random-generated (M-D.R), a stress-line-inspired (M-D.DL), and a grid-based topology (M-D.G). While the number of tests conducted is not high enough to be statistically conclusive, these preliminary results suggest that the SLAM method does lead to improved structural performance, as indicated by the increase in ultimate load and improved ductility after initial failure, as shown in the normalized Ioad-displacement plot. The potential gain is particularly MB evident in the specimen with uniform thickness, where


The Nature of Robots

as its supports and loading positions can significantly induce greater tension in the system. The exploration of more complex surface geometries that are expected to experience both tension and compression can also provide a better understanding on the tensile strength properties of artefact printed using the SLAM method. Beyond the immediate time frame, significant future milestones include the computational development of 3D-solid stress line computation, the elimination of the support structure, and the expansion of the extrusion module’s hardware capabilities, such as the incorporation of sensors to allow extrusion parameters to vary intelligently according to emergent conditions of the printed surface. These advances will open possibilities for free-form and real-time additive manufacturing. Thus allowing technique to be implemented for more complexly curve 3D-surface design, and for full-scale construction.


07 – CONCLUSION // The research pursued in this paper constitutes a promising first step in validating amnew approach to AM that synthesizes multi-axis machining capability and structural-led computation to enable the production of structurally-performative and geometrically-compelling 2.5D surface designs. Significantly, structural load testing provided initial verification that the proposed method outperforms methods using the conventional layer-based paradigm. The

research also presented new strategies that specifically addressed the challenges of robotic manipulation at the prototyping scale. The most important contribution, however, is the demonstration of a new consolidated methodology encompassing parametric design, form-finding, structural optimization, robotic computation and digital fabrication, which uses robotic-integration to achieve a structurally-informed method of fabrication that provides designers with an opportunity to explore a fuller design space that considers both geometry and performance. ACKNOWLEDGENTS // The authors wish to thank the following students who assisted with various aspects of this research: Jonathan Mackaman, Akshat Bubna, Elizabeth Bianchini, Katie Gertz, Colin Poler, Xinyi Ma. Additionally, the authors acknowledge MIT fabrication lab coordinators Justin Lavallee, Chris Dewart, Jen O’Brien, shop monitors Inés Ariza and James Addison, and testing lab technician Stephen Rudolph.

BUILD-ING the MASS LO-FAB PAVILLION Dynamo-Driven Collaborative Robotic Fabrication Workflows for the Construction of Spatial Structures

collaborative robotic fabrication techniques and a combination of traditional craftsmanship and computationally driven manufacturing processes. In order to move from the computational design environment to one of material, the team worked in collaboration with Autodesk to develop a novel design-to-robotic fabrication workflow using the emerging visual scripting interface Dynamo. A custom robotically assisted welding process was developed to assemble 1880 steel parts making up 376 nodes that saved over 3 weeks of labor when compared to traditional processes.

Credits // Nathan King // Nathan Melenbrink // Nick Cote // Gustav FagerstrĂśm //

ABSTRACT // This project-based paper describes the iterative design, structural optimization, and fabrication of the experimental grid shell structure developed for the MASS Lo-Fab pavilion. In this case, formal complexity is resolved through functional complexity that emerges in both elements of the structural system that each maintains a level of simplicity appropriate to respective manufacturing processes and material properties. The structure was fabricated using state-of-the art

01 â&#x20AC;&#x201C; INTRODUCTION // Formal complexity often has implications for structural typology and conversely structural typology impacts formal complexity. Grid shell structures offer the potential for large unsupported spans; can achieve structural stability through double curvature; are typically composed of nodes and struts; and have historically had particularly stringent construction criteria that aim for the formation of identical elements. Today, computational design and analysis tools have eliminated many of the design restrictions that governed previous examples. In many cases formal complexity results in custom components that exhibit a high level of functional complexity. In this particular case, individualized

The Nature of Robots

KEYWORDS // Robotic Fabrication // collaborative Robotics // Digital Fabrication // Robotic Welding // Dynamo // Grid Shell // Design Robotics //


The Nature of Robots 222

steel nodes were fabricated, each having four custom tabs and a single uniform element – a central cylinder — that provided registration for robotic tooling and welded assembly. Individualized struts were uniform in section but varied in length. Each strut had a custom dado in either end to receive the steel tabs from the welded assembly. On-site, a uniform bolted assembly process was utilized thus segregating the highly custom to the automated or high craft environments of the shop and simplifying onsite assembly such that unskilled labor could participate in the construction with limited training using a codified brand on each strut for guidance. The MASS Lo-Fab Pavilion was developed during a series of collaborative charrettes held between MASS Design Group, The Virginia Tech Center for Design Research (CDR), students from VT Architecture and Industrial Design programs, and the computational design consultant, The United Nathans. Over a period of several weeks students from the School of Architecture and Design produced building components at the Virginia Tech Research and Demonstration facility in Blacksburg, Virginia. Working with designers from MASS Design Group, a Boston-based architecture firm, and Rudabega, a Blacksburg-based furniture design and construction firm, the students deployed the experimental structure on the Rose Kennedy Greenway in Boston Massachusetts as part

of the Design Boston Biennial in July of 2015. This research describes the project-based development of the grid shell structure from conception to fabrication with a particular focus on the newly developed collaborative robotic fabrication environment and associated Dynamo-based robot motion control plugin that were developed as part of this research. 02 – STRUCTURAL FORM FINDING, OPTIMIZATION and ANALYSIS // The initial toroid form was developed by MASS Design Group, the project architects, and the project’s student collaborators based on programmatic desires. The guidelines for the global form stipulated that it be a single surface structure in order to simplify construction and to unify structural integrity as a shell. Further design constraints included site-specific pedestrian axes to which the form should be oriented as well as a height restriction of 12 feet. The primary tools for form-finding and surface discretization were Grasshopper, Kangaroo, and Millipede. Rather than following a conventional workflow of first generating a NURBS surface and then discretizing it, a low-polygon mesh approximating the topology was first generated and then subdivided into a diagrid. This diagrid structure was then deformed according to specific forces using the Kangaroo physics solver for Grasshopper. While the forces entered into the Kangaroo simulation were guided by an expected global output, the final curvatures and discretization patterns were unknown

02.1 â&#x20AC;&#x201C; STRUCTURAL ANALYSIS // 3D non-linear analysis was carried out using centerline geometry of the grid shell. The base conditions of this unique geometry, with inner ring and outer ring, provide efficiency in

structural performance. Both inner and outer rings would be anchored to the ground using 3/16 in. rebar irons, which were placed adjacent to joints to minimize bending in the ring beam steel plates. No further foundation work was deemed necessary, as there was minimal risk of uplift or overturning due to the high ratio of self-weight to wind load (this structure is over 70% permeable). Due to the hyperstatic nature of the structure, a decision was made to size all members and joint steel according to the highest load occurring throughout the structure. This decision was further reinforced by the high probability of visitors climbing on the structure. Because of the high efficiency of the gridshell layout, where there was very little bending or shear, joint fastening could be designed based predominantly on keeping members in tension and compression. For this reason, two bolts per connection, placed perpendicularly to the member direction, proved sufficient. Bolts were fitted with washers on both sides to alleviate stress in the wood immediately surrounding the bolt holes. Members were oriented normal to the notional design surface throughout to ensure maximum cross sectional efficiency in the plane where any bending would occur. This layout of the members also brings the additional benefit of minimizing the rotation of the dado slot receiving each joint plate. 03 â&#x20AC;&#x201C; DYNAMO-DRIVEN ROBOTIC MOTION CONTROL // Recent developments in the use of graphical editors

The Nature of Robots

and left to the described form-finding procedure to determine. The Kangaroo physics solver was set to include a force of upward thrust, forming the mesh into a catenoid shell, and as an objective to seek planarity between adjacent struts, thus leading to higher density and shorter strut lengths at areas of greater curvature, specifically the top of the structure. The workflow was arranged such that the designer could edit the control points of the input mesh with a corresponding real-time visualization of the physics simulation, complete with diagrid surface discretization and a Finite Element model generated with Millipede. It should be noted that the metric from Millipede (in this case a visualization of deflection) was used as visual feedback, but was not explicitly entered into any kind of topological optimization. In other words, it afforded the designer a real-time visual evaluation of the structural behavior in order to better inform intuitive decisions involved in manipulating the global form of the pavilion. Once a final form was proposed, further analyses for deflection, bending and torsion were also conducted to ensure the design passed baseline standards. The final form and construction detailing was evaluated though non-linear analysis based on centerline geometry.


The Nature of Robots 224

for the development of robot control have involved closed, unidirectional processes and only a few rigid, albeit simple, tools. In the past 5 years a number of integrated design-to-robotic fabrication workflows have provided many designers and researchers with ready access to robotic control and reduced the need for programming to the point where it is no longer required for machine access. These tools (HAL,, DRG-PRG, and others) — almost entirely relegated to a Rhino-based design environment — have begun to exhibit complications that rival even the proprietary motion control packages provided by the robot manufacturers. As tool developers, we are left searching for a middle ground-maximum functionality with minimum complexity. Dynamo is an open source, graphical algorithm editor that enables users to construct generative tools according to their own specifications. Dynamo offers a robust selection of “nodes” that may be arranged into custom “graphs.” These terms respectively refer to elemental (low-order) and composite (high-order) operations in a Dynamo workspace with the graph being an arrangement of nodes connected by wires relating input and output. Nodes tend to be functionally agnostic; they are specific, basic, and versatile such that their combination avoids predefined outcomes. While these notes would suggest that a Dynamo workspace demands somewhat greater interaction than comparable software - such as Grasshopper – it exposes advanced functionality while

avoiding totalizing simplicity. To take advantage of this functionality the robotic control workflow integrates with the advanced functionality of the Dynamo core in order to control the production of Rapid code: the language used for the ABB robotic manipulators (or other robot-specific syntax). The workflow uses a custom library of “zero touch” nodes developed in C# as a plug-in for Dynamo whose names, inputs, outputs, and methods correspond with Rapid methods. The nodes are organized into four classes – target data, program data, constants, and instructions that, when wired properly as a graph, output PRG files to a workstation, virtual controller, or external drive. The workflow allows users to: 1. Arrange data types, instructions, and functions as in Rapid using the Dynamo core, utilities, and file-writers; 2. Interface with Dynamo, the virtual controller, and workstation to communicate, edit, and enter the following: target data, position data, tool data, work-object data, and program files; 3. Manage multiples of this data per program file; 4. Conduct Run Modes, Number Sliders, and While Loops in the Dynamoworkspace to populate a destination with PRG files for production; 5. Parametrically generate program files in Rapid syntax using Dynamo; 6. Interface with the virtual controller in RobotStudio for simulation and dataentry/retrieval.

04 – FABRICATION // 04.1 – CODIFICATION and PART ORGANIZATION // An organizational system was developed that provided all construction data within the individual parts so that much of the fabrication and assembly could be conducted without any shop drawings. For each node a series of four tabs were custom laser-cut and etched with an index number denoting (node#_flange#) and the process was executed from flange zero (node# - 0) to thee (node# - 3) in order to maintain consistent organization of the nearly 2,000 steel parts throughout fabrication and assembly. A uniform brand was burned into each rough-cut wood member to organize strut data and to communicate relational data corresponding with each steel flange. The “H” brand organized the following data that drove the final stages of the wood fabrication and ultimately the assembly on site. All fabrication data was gathered and organized through a model-linked CSV file that fed a cloud-based spreadsheet. First, the strut number was notated,

The Nature of Robots

03.1 – ROBOTIC PROGRAMMING PROCEDURE for COLLABORATIVE WELDING // The purpose of each tool path was to weld four steel flanges onto a steel cylinder assuming that each must rotate in the same direction; to clear the table while rotating; to not over-rotate; and to be time efficient. Numeric data, which specified four flange angles for each of the 376 steel cylinders, was imported from a CSV file and referenced in Dynamo. With each of the four values the sixth variable of a known joint configuration was replaced indicating the position of the sixth axis of the robot. At each of these joint configurations, the cylinder attached to the end effector would be in position to weld a flange. Before and after each of these positions was another joint configuration assigned such that each flange would correspond with three key operations: raising the assembly above the work surface rotating the sixth axis into position, and then lowering the assembly to the work surface. The rotations were cumulative and totaled to 270% per assembly. A joint configuration which positioned the sixth axis at 0° and the initial assembly beneath the table was inserted at the beginning of the list to avoid cumulative over rotation.These configurations were wired to a joint target node and subsequently to a movement instruction node. The movement instruction node also received speed-data, zone-data, set name, tool name, and work-object name. Two nodes for program data (ToolAtVals and WobjAtvals) were de-

fined with tool and work-object configuration data gathered from the workstation. The outputs of these nodes all connected to a PRG writing node (CreateRapid) which, when the graph ran, produced a file for production. Each flange group was indexed to a value on a number slider that interacted with the automatic run-mode in Dynamo to populate the destination drive with all 1,504 unique positions in a matter of seconds.


followed by the strut length rounded to the third decimal place, and on either side of the marking, a flange association and dado angle was placed. Once codified, the rough-cut lengths contained all the manufacturing and assembly data needed to complete the structure.

The Nature of Robots

04.2 – COMPUTER ASSISTED CRAFT-BASED STRUT PRODUCTION // Each strut had a uniform 2.5” x 2.5” section but varied in length and end condition. Drawing from the part details embedded in each brand, an automated compound miter saw was used to cross-cut precise lengths and the dimensioned members were then drilled uniformly to accommodate the bolted assembly. Using the data embedded in the part, each 3/8” dado was located and cut to 3-1/8” depth using a single table saw pass. These parts were then finished and organized into groups for shipping and assembly.


04.3 – COLLABORATIVE ROBOTIC NODE FABRICATION // To facilitate the robotically assisted welding operation, a multi-robot human-machine collaborative work cell was developed consisting of two inverted ABB IRB 120 s mounted beneath the work surface. A custom end effector was developed consisting of a 3-jaw chuck capable of gripping the inner or outer wall of the 2” steel pipe that remained consistent throughout all nodes. The end effector was isolated from the robot using a phenolic extension to ensure no electrical interference from the grounded

welding table. In this case the gripper was manually engaged to account for deviation in pipe dimension but in other applications, precision materials or force sensing could adapt to a fully automated production strategy. The end effector was positioned to pass through the work-plane to facilitate welding while protecting the inverted industrial robots mounted beneath. Here, the robotic arm is completely separated from the user, enabling safe collaborative interaction. To amplify safety and to achieve enhanced quality control a machine operator, in addition to the welder, managed individual node files and maintained manual enabling operation. Each team of two fabricators-welder and operator developed a call and response cadence to verify that the program pointer, the manually placed steel tab, and the robot position all corresponded-a strategy that eliminated all but a few placement errors. In a comparative analysis it was determined that the collaborative robotic fabrication of the steel nodes reduced manufacturing time from 39 to 3 min per node, a total savings of 225 labor hours at a sustained production rate. This savings ultimately accounts for an overall reduced labor time of approximately 3 weeks when accounting for teams of two fabricators. 04.4 – ASSEMBLY LOGIC // Once fabricated, all parts were packaged in groupings based on their respective position on the structure. Onsite, these packages were catalogued and organized about the perimeter

05 – CONCLUSION // The MASS LoFab pavilion enabled close collaboration between design practice (MASS), academia (Virginia Tech), and industry (Autodesk and Rudabega). This type of collaboration is increasingly important to realize innovation in the Architecture, Engineering, and Construction (AEC) industries. From a practical perspective the grid shell represents a structural typology that can be used in resource-limited settings where long span members are not available. By positioning a need that emerges from design practice in the context of applied academic research new opportunities emerge. This project served as the platform for advanced research in robotic fabrication, the development of novel computational workflows, and established new partnerships that will help bring emerging Design Robotic automation strategies closer to application in AEC industries. The construction resulted in the fab-

rication of 376 custom nodes consisting of 1880 parts and 720 individual struts consisting of 1440 individual end conditions through a combination of high-craft and high tech fabrication strategies. On-site, strategic assembly instruction embedded in each component allowed for construction by volunteers, students, and staff having limited to no construction experience. During the project two new software tools were developed including the discussed Dynamo-to-robotic fabrication workflow and an emerging Autodesk Fusion-based kinematic simulation environment. These tools will be further developed and tested during the Dynamo-BUILD workshop at the 2016 International conference on Robotic Fabrication in Architecture, Art, and Design conference in Sydney, Australia.

ACKNOWLEDGENTS // This project-based research was conducted with the support of Autodesk, the Autodesk BUILD Space; Virginia Tech College of Architecture and Urban Studies, Virginia Tech School of Architecture + Design, Center for Design Research and The Institute for Creativity. Arts, and Technology; the Rose Fitzgerald Kennedy Greenway Conservancy; The United Nathans: and Rudabega.

The Nature of Robots

of the structure’s base for use during assembly. The assembly sequence was established to maintain a geometry that was continuously self-supporting, moving from the outer ring of the structure inward. This strategy allowed for concentric assembly without the need for sintering or formwork-only limited shoring was used for added safety. The bolted connection allowed zero tolerance between the bolt (pin) and the wooden strut, but holes in the steel allowed for 1/16” tolerance to resolve any site variation or fit issues between parts.


The Nature of Robots

Part III - Real-Time Control


User Interface for Live Robotic Control Credits // Curime Batliner // Michael Jake Newsum // M.Casey Rehm // ABSTRACT // Through the development of user interfaces that leverage real-time control, the robot emerges as a design platform where programming, simulation and execution collapse into a singular act in time. This reduction of the typical robot workflow allows design processes to continuously engage with adaptive contexts whether they are deformations of material, nuanced data or instantaneous design input. The case studies, presented in this paper, demonstrate design potentials for developing new interfaces, where the digital and physical are mutable, letting designers intuitively engage with matter and representations in flux through robotic interactivity and autonomous agency. KEYWORDS // Real Time // User Interface // Design Research // Interactivity // Autonomous Agency // SCI-Arc // Robot House // 01 â&#x20AC;&#x201C; CONTEXT // 01.1 â&#x20AC;&#x201C; PHYSICAL INTERFACE // SCI-Arcâ&#x20AC;&#x2122;s conceptual approach to design cannot be separated from the development of software. Since the beginning of the Robot House, students and faculty have been working on custom design interfaces by integrating robotic motion control into ex-

isting design tools that architects and designers use every day in school and practice. Bespoke plugins for Maya and Grasshopper have reduced the learning curve of the robot as a technology and have allowed designers to integrate the robot into their design process. Treating the robot as a physical design interface, where real world feedback is introduced into this digital process, the digital model serves as a point of departure rather than an idealistic goal. Eventually, the model as a design tool can be removed altogether. As a consequence, the limits of where design stops and production starts are becoming increasingly intertwined. While significant efforts have been made for narrowing the gap between digital and physical, the research is inherently limited by the hierarchical structure of offline programming where the robot is being constrained to a predefined timeline or sequence. With this model, synchronization with other machines and digital devices happens at predetermined positions in the sequence. Feedback is limited to observation of the particular process. Errors in the underlying logic, the sequencing of the robot, are fatal as once programmed it cannot be altered while the program is running. Since motion planning and runtime are separated in time, many experiments need to be reset resulting in a linear way of working. While in this scenario, the design process can be iteratively refined and calibrated, yet design processes ultimately remain constrained by the sequence.

The Nature of Robots



The Nature of Robots

The ability to alter a sequence, during runtime, narrows the gap between digital and physical workflows, which significantly increases speed in project development. More important than speed is the potential of opening up new research trajectories, which require alternative modes of control that break away from a static sequence. The ability to alter the robot’s logic in runtime can change how we design, since it establishes an intuitive interactive bridge between the digital and physical. This platform, where information can be transferred at any time in both directions, allows the user to reshuffle, reverse, replay, layer and combine sequences. The digital environment gets access to the physical as the physical gets access to the digital, and new design process, which requires simultaneous data flow in both direction.


01.2 – SOFTWARE MEDIATION // Intelligent real-time software has saturated both contemporary design and, more importantly, society. Our methods of interfacing our world are increasingly mediated by software, which interpret, analyze and sometimes adapt to our intentions. The sophistication of software, along with its integration with hardware, allows it to slip seamlessly into our daily tasks and produce an intuitive extension of intentionality into ever expanding territory. The mouse, joystick and keyboard have dominated how we convey intentionality to computers. For designers, especially digital artists, this is not

always intuitive. Interfaces, such as the Wacom tablet and touch screens, include drivers that have been tuned to effectively translate perceived human motion into useful movements within digital environments for vector and pixel graphics. Through additional software packages, the user can further define the translation of these moves. This model of the interface extends beyond purely simplifying the translation of human inputs into effective outputs. It allows the user to manipulate these translations towards amplified effects. Without physical registers, touch inputs are imprecise, thus technology must be forgiving to create a pleasing user experience. Software, such as Swype, adapts and learns the user’s tendencies by creating bespoke gesture libraries for the user. This significantly increases speed and accuracy while enhancing user experience. The speed with which it adapts and performs sophisticated tasks allows for it to effectively disappear in the user experience. Mediating software applications have recently gained more popular attention through the release of the Google Deep Dream to the public. The underlying software was originally developed to explore the use of “deep learning”, neural network architectures to identify specific elements in images to improve image-based search. Google Deep Dream exposes the generative potentials of misappropriation of these tools. Digitally generated images, similar to experiences of pareidolia, are made when the image search is applied to images that lack

02 – LIVE // Live, a real time motion control platform developed for industrial robots paired with a series of novel interfaces, aims to open up the robot for nonlinear processes, by coupling sensing devices and advanced programing with robotic technologies. The Live platform is an expedient for the exploration of interfaces that intuitively translate human inputs into robotic motion, the interplay between human inputs and analysis driven robotic interpretations, and interfaces for developing robotic agency. Live utilizes the robot’s ability to send and receive information, through timed synchronous tasks, which gives the robot the capacity to continuously engage with current context available to the designer’s programming logic. Incoming information is processed into motion, tooling and configuration commands. The designer is given the ability to add information for later use as well to intervene or alter the robotic operations at any time. This programming method has alternative goals where the robot is no longer programmed to be optimized for repetition and precision. Instead, the robot is programmed to be versatile, nuanced and interactive. This removes the constraints of offline-programmed operations that need to anticipate the contextual information through simulation and contextual restrictions. Online Program-

ming encourages design processes to move away from the idealized static CAD model to a context aware model where a shifting environment is anticipated. Inputs can be dynamically adapted through external sensors, and events. At any time the design processes are described through a model of adaptable reconfigurable relationships. 03 – CASE STUDIES // 03.1 – TANGO // Tango is a dance between two performers, a human and a robot. Using a Microsoft Kinect, the human’s movements are tracked and translated. Whenever the human moves, the information is sent to the robot in real-time. The robot’s movement is instantly activated by the control platform Live. While the program runs real-time, the performance itself gets out of sync. This gap never fades completely. With some training from the human and the possibility to adjust to how the system processes and translates information in real-time, the gap becomes an intuitive part of the interaction. Human movement varies between individuals, therefore the user interface needs to be easy to learn, while being reconfigurable, on the fly. While the robot gives feedback to its human partner with its body, a visual interface mapped onto the walls gives the human additional information about the internal workings. The visualization includes the real-time representations of the human and robot bodies, so the participant can understand how each step will

The Nature of Robots

the specified elements. The software probabilistically transforms the image by amplifying what is most similar to those absent elements.


The Nature of Robots

impact on the performance. Additional information, such as speeds, position and filtering, communicate the robot’s limitations to the human, helping them understand if they need to slow down or if the robot responses can be adjusted. All filters and damping functions are adjustable on the fly by the operator, and some variables, such as speed and pausing, are also made available to the performer through body gestures. The choreography, from a programing perspective, is set up as a master-slave condition. Yet, the performance appears as a real dance between equal partners, since the robot’s body influences the human’s behavior, which closes the interactive loop. Additionally, the robot’s motion can be recorded, replayed and modified letting the user choreograph the robot in real-time for later use. The flexible interface allows the human to tune the relationship granting them mediating agency in this spatial interface.


03.2 – AUTOMATON // Utilizing Henri in conjunction with Live, an application was developed in which the platform attempts to perceive potential features in a blank mass and then iteratively amplify those features. In this project, the human intervention occurred at the level of designing the intelligence of the platform to analyze its context and then effectively manipulate it. This type of application has uses in creating production methods for dynamic material or for complex contexts in which direct human inter-

vention or analysis is prohibitively difficult. Unlike in Augmented Materiality, a project which also establishes a live link between sensor material and robot with a totalized goal of a rationalized physical output in the above example the robot analyzes recursively the wax monolith for potentially conical regions, and then creates tool paths to amplify these features until a maximum limit on material porosity is achieved to avoid structural collapse. In these conditions, it is necessary to generate design through the codifying of intention at the level of local behaviors. 03.3 – DRAUGHTSMAN // The drawing series executed with Henri and Live synthesizes the approaches explored in Tango and Automaton. Multiple levels of interactivity, between interaction and production, are quickly explored in this 2D setting. Three basic versions of the application have been developed that parallel ranges of involvement between the human operator and the intelligence of the interface discussed in the Real-Time section of the text. The first iteration of the application utilizes a tablet interface to translate human pen marks into robot painting behaviors. In this version, the underlying algorithms for translating inputs into motion are tuned to make the real-time control of the robot, through the pen, completely seamless and intuitive. Filters and analytic algorithms are used to interpret and refine potentially noisy inputs into velocities and paths that have a

04 – CONCLUSIONS // Real-Time robotic interfaces as design methodologies are still new territories at SCIArc’s Robot House. The use and appropriation of input sensors and output tooling are still in their infancy. Live interfaces continue to be prototyped for a wide range of working modes where designers can approach robotics as an intuitive extension of their own design processes. Using the Live platform, digital and analog methods converge through real-time inputs and outputs between both worlds where the robot is the mediating device. These new interfaces allow human skills to be transferred directly into informed robotic motion controls. This workflow moves designers and craftspeople away from offline programming through the in-

tegration of gestures and sensor embedded smart tools that interact with robot collaborators in high skill applications. Moving to trajectory based models, hands can drive the robot intuitively, through vision inputs like low fidelity devices such as the Leap Motion or the Xbox Kinect. These digital input devices can be tuned through the interfaces to perform as physical emulations of brushes that are found in design software such as Photoshop or ZBrush. The robot can then be situated as a physical modeling program where the designer can work directly with real material properties, so design and fabrication emerge as one continuous act. When the designer wants to carve or draw with their hands, they can engage the material directly. Conversely, when the designer needs parametric patterns or machine precision, the designer can leverage parametric controls. In this context, digital designers do not need to simulate material affects to try to anticipate the fabricated results. Instead, their assumptions can be tested and calibrated simultaneously. The time distance between output, evaluation and modification are reduced to milliseconds, so that design intentions can be expressed without compromise. Vision inputs can be used to monitor material in real-time, thus feeding information into algorithms that analyze the material properties to output updated paths according to codified design intentions. This opens the possibility of correcting processes that the robot can layer on top of

The Nature of Robots

clear scalar relationship and linearity towards the designer’s intention. The second iteration of the application begins to push the agency of the software’s intelligence forward hierarchically. Generative algorithms create a collaborative relationship between the software and the designer. In this case, the operator makes a mark on the page. The robot platform utilizes a web camera to analyze the marks for brightness levels. Mapping the pixel information of the camera to a grid, the robot manipulates the captured mark using a growth algorithm. It then adds its interpretive marks to the painting. In this way, the final composition becomes a negotiation between the user as artist and the designed machine intelligence.


The Nature of Robots

an existing operation, either autonomously or after notifying the user. This approach lets digital designers engage with less deterministic processes. These novel interfaces aim to layer intelligence, mediated through software, with manual intuition. Equally, it invites skilled craftsmen to digital making, through sophisticated human-robot collaboration models based on sensory inputs. This builds a common base for exchange and collaboration amongst designers of different skill sets and machines. Here, design and making is not a one directional path with an initially determined end. It becomes a multidirectional open-ended process.


TOWARDS ON-SITE COLLABORATIVE ROBOTICS Voice Control, Co-Speech Gesture and Context Specific Object Recognition via Ad-hoc Communication

any device using TCP/P to share variables with the full abstraction of the original machine software platform and can therefore be used synchronously by a vast array of equipment including CNC machines, industrial robots, construction equipment, mobile devices and PLCs. KEYWORDS // On-Site Worker Assistance // Collaborative Robotics // Multimodal Communication // Advanced Optical Sensor //

Credits // Thibaild Schwartz // Sebastian Andraos // Jonathan Nelson // Chris Knapp // Bertrand Arnold // ABSTRACT // This work presents a novel set of accessible and unified hardware and software solutions that facilitate the implementation of natural human-machine interactions, as required by collaborative robotics in both indoor and outdoor environments. This extensible framework supports vocal control, co-speech gestures, and object recognition with feature tracking and adaptive resolution. The interactions are based on a new network messaging protocol that allows

01.1 â&#x20AC;&#x201D; COLLABORATION CONTEXT // The present prevalence of new tools and techniques provides an opportunity for the paradigm of the architect to evolve. At Bond University the robotic curriculum is aimed at exploring and nurturing this possibility, with particular focus on collaborative robotics as an evaluative territory in which to conceive new ways of working and thinking. HAL Robotics, with their HAL software suite, are aiming to create an accessible vendor-agnostic solution for the simulation, programming, and real-time control of robots and peripheral devices, while Soundisplay develops high-performance, acoustic and optical sensing technologies for entertainment and various industrial applications, including robotics. It is our shared belief that collaborative robotics represents the next phase of the intersection between design and construction-an undoubtedly fertile space where many architects

The Nature of Robots

01 â&#x20AC;&#x201D; INTRODUCTION //


The Nature of Robots

and designers have the distinct opportunity to take more control over not only their work but also the methods and processes with which it is made. Exercising this control is seen to facilitate a fuller understanding of design processes and the decisions that help drive these processes from both ends: how construction works backwards to affect design and how design moves forward to affect construction.


01.2 — TECHNICAL CONTEXT // In-situ applications of robotics-whether applied to fully automated construction scenarios, or on the contrary envisioned as intelligent, service-oriented, distributed manufacturing systems-have remained an ongoing research topic since the early 1980s. Beyond the various sociological and economic challenges surrounding the application of such innovative processes, major technical issues still need to be addressed in order to ensure the compatibility of past and future research results with the requirements of the construction industry. The authors have identified three of these bottlenecks, which amongst the large scope covered by this research field, appear to directly impact the progress of efforts led by actors from the Architecture, Engineering & Construction (AEC) industry: • The absence of generic communication protocols allowing the exchange of structured data between industrial machines and the highly limited abstraction of the alternatives currently proposed by research con-

sortiums; • The inability (and lack of interest) of robot manufacturers to provide intuitive control interfaces for the full range of their products even though the quickly growing market of new generation, low-payload robots has clearly proved customers’ enthusiasm for such improvements; • The incompatibility of most sensing technologies with intensive outdoor applications such as construction sites, which are often radically different to the traditional industrial environments which these technologies have been primarily developed for; Instead of attempting to solve these problems individually, the authors decided to tackle them simultaneously, thus enabling various levels of interaction to be natively linked to each other and minimizing the risk of subsequent technical limitations resulting from design omissions. This work presents a novel set of accessible, unified hardware, and software solutions facilitating the implementation of natural human-machine interactions, as required by collaborative robotics in both indoor and outdoor environments. This extensible framework supports vocal control, cospeech gestures, and object recognition with feature tracking and adaptive resolution. The interactions are based on a new network messaging protocol allowing any device using TCP/ IP to share variables with the full abstraction of the original machine software platform and can therefore be used synchronously by a vast array of

02 — GENERIC ROBOT COMMUNICATION PROTOCOL // In order to harmoniously account for context and operator information, robots need the ability to manage and exchange large amounts of data with additional devices. Traditionally, such a machine is often linked to a cell equipped with a few sensors to take simple events into account, such as the intrusion of an operator into the robot workspace during the execution of a task triggering an alternative behaviour (reduced motion speed, immediate halting of program execution etc.). The content shared by systemic components in these situations is extremely basic: switches return digital, electric signals while other sensors return digits in the form of an electrical current. Ultimately this means that any data is tightly linked to its electrical reproduction. Collaborative robotics requires much higher abstraction and density of data: the most basic vision application will already deal with messages 3 orders of magnitude larger but smart sensors can necessitate even bigger amounts of data to be processed. It is obvious that there is a need for radical change in industrial machine-to-machine communication systems that can move beyond simple signal sharing via PLC towards

peer-to-peer communication on robust fieldbuses. Such scenarios have been anticipated since the mid-1980s, but the very late and restrictive standardization of these protocols has left industrial networks a jungle of low-level, low-bandwidth networks. A few modern solutions, mostly based on the Ethernet technology stack, are now slowly replacing these old infrastructures and are a very interesting vector of modernization of both networks and endpoints as they support various protocols already in use for internet communications such as TCP/ IP. Allowing a virtually unlimited number of connected devices (millions, in comparisons to dozens or hundreds at best for older systems in use), they enable the development of technologies required by specialized machine networks, such as those that could animate building sites in a not-so-distant future. Considering this evolution, the toolset presented in this document is based on an application layer protocol with interesting properties: • It allows peer-to-peer communication of structure data, based on an easily modifiable object dictionary; • It minimizes network load, by using extremely lightweight messages; • It is very easy to use with existing binary serialization utilities provided by general-purpose programming languages; • It is platform and vendor agnostic; In the context of this study, the afore-

The Nature of Robots

equipment including CNC machines, industrial robots, construction equipment, mobile devices and PLCs. We conclude with the description of a testing scenario to be deployed during the conference workshop.


mentioned protocol is used to normalize the communication between smart sensors and industrial robot controllers in order to deploy vocal control and three-dimensional machine vision to such machines via phones, computers and other Human-Machine Interfaces (HMIs).

The Nature of Robots



03.1 â&#x20AC;&#x201D; INTELLIGENT CONTEXTUAL MULTI-ENVIRONMENT SENSIN // 3D sensing appears to be the ideal platform to offer real-time control, feedback, or assisted operation in any environment. However, many challenges make current 3D technologies impracticable in most situations. The technical limitations of 3D sensing, mostly camera-based systems are numerous: from random behavior under light varying conditions; unacceptable resolution flag compromise; unrealistic post-processing computation needs; non-deterministic shape recognition success rate; non-deterministic timing; or simply a slow sensing rate. Additionally, a robot cell needs much more than 3D sensing-ideally it needs security sensors, calibration, telemetry, motion tracking, feedback control, and machine vision with shape recognition or human machine interfaces. This results in unrealistic investments and complexity to setup multiple sensing devices that should work seamlessly. Techisplay 3D-Itechnology is an industrial grade and affordable sensing and control platform that over-

comes most of these problems. This new type of optical sensors benefits from exceptional light radiations sturdiness and can work from pitch black to outdoors lighting (even working under direct varying sunlight exposure) with steady specifications. The sensing is operated at speeds exceeding 1000 frames per second, which permits sensing of objects in motion and even vibrations. The new type of sensing data generated allows a different algorithmic treatment, resulting in ultra-low-latency three-dimensional shape recognition with a success rate approaching the theoretical 100%. The sub-millimetre resolution can be even finer when varying sensing timing for critical operations such as calibration or precision assembly. The ability to use a generic 3D sensing platform allows the fusion of calibration and telemetry (real spatial coordinates), human machine interfacing (joystick simulation, gestural interface), machine vision (shape recognition and feature tracking), motion tracking (trajectories tracking and prediction), vibration sensing, impact sensing, and security (both using shape recognition and context). The parameterization of 3D sensing and 3D control as opposed to point cloud, pixel based acquisition, statistical analysis or post-processing, is to simplify the use of sensors. In a professional environment it transforms a flow of raw information into a legible stream of controls and parameters, usable by both machine and human users.

proposed by the authors to the problem of inter-comprehension is a common language, both simple enough to be parsed in real-time, yet complete enough to retain a high level of abstraction for the user. Dear Mister Robot is just such a language. DMR is a restructuring and simplification of the English needed to control robots in an extensible, intuitive and robust way that is ideal for both human-machine collaboration and teaching. At the heart of DMR is a concise list of keywords and a set of new parts of speech; presented in full in the DMR dictionary along with definitions tied to the WordNet semantic dictionary and usage examples. An exemplary DMR phrase could be something as simple as “Move Left 12 cm”. This sentence is so easily interpreted by a human lector that to describe its intended effect is almost redundant. Yet the same cannot be said of a numeric parser. To align a human understanding of the language with that of a machine all DMR keywords are stored within a limited set of command types: movement, reorientation, tool action, wait, stop, and conditional. Each command type has predefined parameters, some of which are required to validate a command, while others are optional but increase the users’ control of a given action. For example, in a movement command, such as the one previously shown, we require only a direction (left) and a distance (12). In fact, in this case, the addition of units (cm) is optional but its omission is only advised with an a priori knowledge of the machine’s settings or when the

The Nature of Robots

03.2 — NOISE-FREE VOCAL CONTROL BASED on NATURAL LANGUAGE // Vocal recognition, as supported by advances in artificial intelligence, is becoming an increasingly viable mode of communication between man and machine. In most cases the machines in question are our phones, tablets, or computers but the same technology can easily be extended to robotic manipulators-the trick is in the digital interpretation of our instructions. While audio feature and voice recognition algorithms are extremely promising, they suffer from surrounding noise problems when used in real conditions. That affects recognition performance and success rate drastically. Work in this direction has been significantly helped by successful research and drastic improvements in noise-reduction technologies, including Soundisplay’s O.D.O. O.D.O sensor is primarily a novel ultra-sensitive acoustic contact sensor. It potentially transforms any surface or material it is connected to into an acoustic sensor. The acoustic sensor has the useful property of ignoring airborne acoustic waves except when directly applied at very close distance, leading it to behave as a very unique ultra-proximity microphone that is ideal for audio analysis, vibration analysis and speech recognition as it allows audio recognition algorithms to approach their ideal performance. Beyond these technical sensing aspects, the content of the communication between humans and machines is open to experimentation. The solution


The Nature of Robots

units are implied by the command’s context e.g. “Move Left 12 cm then Down 4,” whereby the 4 infers the cm from the previous command. Raw DMR sentences are sent to a proprietary parsing algorithm to extract all the useable information and execute commands in real time. While this command-by-command vocal control is useable, collaboration in this manner is exceedingly tedious and does not particularly improve upon the current scripting for control paradigm. DMR, therefore, introduces tasks as a means of containing, combining, and sequencing commands. Tasks are the centre point of DMR’s extensibility allowing users to create and name command sequences on the fly with strongly typed variables, UO based conditionals, and a plethora of other subtle extensions. On its most basic level a task is just a list of commands, but by having the ability to recursively nest tasks within one another, and within conditional commands, we can start to produce increasingly complex robotic behaviors.


03.3 — SENSOR FUSION for NATURAL INTERACTION: Co – SPEECH GESTURE and OBJECT IDENTIFICATION // When we interact with one another an enormous amount information is inferred based on what we can see, feel, smell, taste, or hear. This contextualization of our speech is found in the physical environment, gesticulation, or even body language. Take for example the simple request “pass me that brick”; unless we can

see the speaker pointing, or the brick in question is otherwise clearly defined, the task is difficultly completed. Similarly, “come here” is more often than not associated with a co-speech gesture such as a waving of the hand. Both of these can be considered context for the speech and are used to remove ambiguity or add extra information to a phrase. Both gestural and vocal control give a user intuitive, real-time control of a manipulator but, as they have been presented thus far, require the user to be in continuous communication with their machine and have either a comprehensive knowledge of the environment in which they are working or a means of accessing contextual data e.g. through vision. The step that we would like to take here is one towards adaptive automation; a means of giving our machines knowledge of their own environments and capabilities to react to what they are sensing. By increasing the robot’s comprehension of its environment we can quickly start adding levels of abstraction to our communication methodologies. The first tests done to push our interfacing capabilities with DMR were through a 3D camera mounted on a 6-axis robot’s tool and a small piece of code that enabled the robot to identify bricks, their colours, and their locations in space relative to the end-effector. Exclusively using DMR and the accompanying DMR Interface we were then able to instruct the arm to “pass me a green brick.” The process of robotic control is then incredibly similar to the way in which our

04 â&#x20AC;&#x201D; CONCLUSION // This paper presents a set of unified real-time software utilities and corresponding smart sensors aiming to ease the integration of natural human-machine interaction for on-site collaborative robotics. With an adequate communication protocol and linguistics framework used in conjunction with context-aware sensing, we expose how robots can be turned into intelligent assistive devices that can adapt to human complex operations, changing environment parameters, and even unforeseen situations. The fluid operation of such a system aims to enhance human perception speeds and offer a real sense of interaction where technology disappears as a process and appears as continuous assistance.

The Nature of Robots

subconscious directs our own bodies. Both entities move towards the object, orientate themselves appropriately to be able to pick it up, and upon detection of sufficient proximity, close a hand or gripper to gain possession of the object in question. In these first experiments the robot is exclusively using vision, so has a much less fine motor control than we have with a combination of sight and touch, but the vagueness of our commands remain similar and that is where our new modes of interaction shine. The combination of these systems permits two very distinct benefits compared to a traditional programming model. The first is the ability to communicate in a contextually enriched language; allowing us to push towards task-based robotics, higher levels of abstraction, and â&#x20AC;&#x153;avoid having to program every little movement or actionâ&#x20AC;?. The second advantage of contextual awareness, especially that in real-time, is the ability to adapt to changing environments. This affords us the possibility of breaking robots out of their impeccably calibrated factories and into the real world. The continued addition of these robotic senses allow us to leave the robot in a state of autonomous work until a task is achieved an essential property of any collaborator.



The Nature of Robots

Credits // Roland Snooks // Gwyllim Jahn //


ABSTRACT // This paper posits a model of generative fabrication in which agent-based models imbue physical material with digital agency. We demonstrate a process in which real-time feedback is developed between industrial robots and multi-agent algorithms to explore the generative potential of the interaction of computational and material agency. This design research represents an inversion of material agency, from which two key concepts have emerged: parallelism, and stigmergic robotics. Rather than encoding material behavior within digital models, physical material takes on digital behaviors through an inversion of material agency. Parallelism describes a hybrid of digital and material behaviors through the closeness of their interaction. Stigmergic robotics collapses design and fabrication processes into a single operation where the robot operates as an extension of the digital agent

generating form through a series of design behaviors operating directly on physical material. KEYWORDS // Robotic Fabrication // Multi-Agent Algorithm // Stigmergic Robotics // Material Agency // 01 — INTRODUCTION // The relationship between the behavior of physical material and digital algorithms is increasingly being explored within experimental architectural practice. Within this context we demonstrate a process in which real-time feedback is created between industrial robots and multi-agent algorithms that hybridises their behaviors and imbues physical material with a ‘wild’ digital agency. This process inverts the prevailing approach to material computation (Menges 2012) in which material behaviors are encoded within digital algorithms for the purposes of form finding, optimization or simulation. Instead we posit a model in which the digital knows nothing of the physical, but instead operates at such a close-

01.1 — VOLATILE CHARACTERISTICS of COMPUTATION // Volatile self-organizational processes of computation - material or digital - generate expressions and characteristics specific to their emergent processes. The generative capacity of these processes have been exploited by artists as diverse as Perry Hall, Casey Reas, Anish Kapoor and Roxy Paine. Hall (2015) is concerned with the material behavior of paint over time through thermodynamic, magnetic and chemical interactions. The reactions he provokes are intended to tease out the generative capacity of the material - not as a pure expression of material, but one that is curated. Reas is concerned with the generative capacity of digital computation, programming the underlying behaviors for the generation of pattern and form. The ambition of our research is to draw out the emergent characteristics of a hybrid generative capacity between the “inherent shape-generating capabilities of matter” and abstract digital behaviors - characteristics that are neither a retreat to nature nor derived from the digital. Robotics becomes the interface between these two realms of behavior. While the relationship between machine behavior and that of material is illustrated in Kapoor’s Between Shit and Architecture and Roxy Paine’s Scumak pieces, this relationship is linear-the machine constrains mate-

rial. Our concern instead is the continuous feedback, facilitated by vision systems, between the behavior of material and the algorithmic behavior executed by the robot. Within architecture, this relationship between volatile material behavior and autonomous robotic fabrication has been conceptually established by Francois Roche. 01.2 — DEISGNING in MATERIAL // The resolution of scanning technology and simulations through General-Purpose computing on Graphics Processing Units (GPGPU) are increasingly acting to blur the distinction between physical and digital reality, with implications for robotic fabrication, simulation and design. Alisa Andrasek speculates that “the found physics of complex material processes, can now be encapsulated through computational physics and large data sets, embedded into simulations and, consequently, physical constructions” (Andrasek 2012). Similarly, the accessibility of microcontroller platforms and consumer sensor hardware brings designers closer to the machines and processes that allow them to act in material. Systems responsible for materializing or sensing a design are beginning to come into contact with those responsible for design decisions governing the form, pattern or structure of the designed object, resulting in new paradigms for robotic control and design.

The Nature of Robots

ness of feedback that the digital and material are caught in a mutual process of volatile reaction.


The Nature of Robots 244

01.3 — ROBOTIC DESIGN PARADIGMS // A convergence of control and design systems can be witnessed in the recent proliferation of bespoke robotic control plugins and hardware. The vast majority of these systems extend the functionality of computational design software (such as Rhinoceros, Grasshopper or Maya) by generating robotic instructions from existing digital models (that Menges (2015) would describe as an ‘instructional’ approach). In contrast, ‘behavioral’ approaches typically incorporate feedback to respond to constraints and design requirements during fabrication, demonstrated in the recent work of Casey Rehm or Adam Fure. Designers face significant challenges working exclusively within either of these domains. Instructional approaches tend to be limited by the capacity to model and anticipate material constraints, behavior and tolerances, and the accessibility of particular operations and their associated patterns of use can result in hegemonous design practice (Leach 2015). Behavioral approaches tend to be equally limited by the challenge of encoding high level design decisions within rudimentary sensory systems, microcontroller hardware, and robotic communication platforms. Real time control of KUKA industrial robots, for instance, currently requires proprietary software and the development of third party protocol to communicate with the robot. In response to these challenges, we have developed a distributed, real-time, object oriented

control platform in which responsibility for allocating and updating robotic tasks and behaviors can be distributed across multiple design agents (software packages, sensors, simulations or humans) in parallel. 02 — PARALELLISM // Rather than conditioning simulated models with material behavior or crudely approximating digital complexity with robotic instructions, we posit the parallel operation of computation and material that creates a hybrid or infusion of digital and material behaviors through the closeness of these behaviors enabled by real-time operation. Robotic parallelism benefits from the unconstrained behavior of both the computational agent and physical material. Encoding material constraints and behaviors into the algorithmic model either renders material behavior as a simplistic caricature that confines the design process to quantifiable behaviors, or over constrains the algorithm to the empirical model. Parallelism exploits contingencies and complexities of physical material, enabling the most volatile behavior of material to participate in the generative process. The ambition of this is to match the speed and sophistication of material processes through digital computation, where each responds to the other at a fine-grain scale of operation and material manipulation (deposition and subtraction). This establishes a stigmergic relationship between material and computation, where the robotic agent scans and responds to the deposition and removal of material.

03.1â&#x20AC;&#x201D; CONTROL // The technical workflow for stigmergic robotics involves a feedback between the robot, deposition and subtractive tools, material behavior, vision system and multi-agent algorithm. Our series of design experiments use two KUKA Agilus KR10 R1100 SIXX robots controlled by a real-time server through KUKA RSI, expanding foam and plastic extruded from end-effectors, Microsoft Kinect 2 and Structure IO sensors and a Java and Processing based algorithmic environment. 03.2 â&#x20AC;&#x201D; KUKA RSI // KUKA s proprietary Robotic Sensor Interface (RSI) software allows an external computer to communicate directly with the robot controller via XML strings sent using a User Datagram Protocol (UDP) server

The Nature of Robots

03 â&#x20AC;&#x201D; STIGMERGIC ROBOTICS // This approach draws from the logic of multi-agent systems, and in particular stigmergic behavior that generates complex order through the accretion and re-organization of matter. Stigmergic systems such as those that underlie the formation of termite mounds or the self-organization of ant trails, operate through the indirect communication of agents. In these systems the agent (termite/ant) interacts directly with its environment (pheromone deposition). It is through the feedback of the agent altering the environment and the environment affecting the agent that complex order emerges. This self-organizing logic provides the basis for a stigmergic fabrication approach (Snooks 2012). While these systems typically involve many simple agents, there is no need for direct interaction between agents as the environment becomes the substrate of communication. Consequently the same level of complexity can be achieved through the interaction of the environment and either a single agent or a large population of agents-assuming the number of operations is equivalent. The capacity of industrial robots to run autonomously for extended periods of time allows for either a single, or low population of robots, to form complex stigmergic assemblages. In this paper we describe two strategies: a robot depositing and interacting with a volatile material, and secondly two robots interacting with a volatile material---one depositing material while another

removes material. This approach to stigmergic robotics has arisen from our long-standing design research involving multi-agent algorithms and behavioral approaches to formation, and has further developed within what is an emerging discourse on real-time robotics and sensor feedback systems. This emerging domain encompasses the parametric updating of models based on sensor feedback to the generation of tool operations based on scanned matter. The original contribution our research offers is the real-time interaction of material and computational agency, where the formation of architecture is directly driven by their behavior, rather than an a priori parametric model or generative algorithm.


The Nature of Robots 246

over a local ethernet connection. RSI provides functions for making corrections to the robot position and digital IO signals, and sends information to the control PC detailing robot joint angles, global position, velocity and IO values in 4 ms intervals. Any correction made to the position of the robot is executed within a single interval, limiting the ability of RSI to make substantial modifications to the existing position or motion of the robot without exceeding the limits of the joint motors. However, the pilot project below demonstrates that provided a stable connection can be maintained between the robot controller and the external PC, most of the functionality that would typically be found within a simple KRL toolpath program can be reliably executed through the single RSI_MOVECORR command by handling path planning and network communications on an external server application. This allows for non-linear design workflows whereby adjustment to robotic behaviors can be made by designers on the fly, in addition to facilitating the integration of KUKA industrial robots within more complex robotic systems that may incorporate dedicated computer vision processors, large sensor networks, or remote operations. 03.3 — SERVER APPLICATION // A C# application was developed that runs two UDP servers concurrently to handle communication with external machines (Extemal Server) and the robot (RSI Server). The RSI Server communicates with the KUKA control-

ler over a local ethernet connection and maintains data transfer using RSI XML formatting. The External Server handles sporadic communication with other control systems over an internet connection. Communication between the two server instances occurs through reference to thread safe dictionaries, allowing updates to the position, speed or Io signals of the robot to be passed to the KUKA controller by any program capable of sending XML strings over UDP. The RSI Server calculates acceleration and interpolated trajectories to a given target position and stores them as a list of correction commands that can be executed at 4 ms intervals and independently of an external connection. Any update to the target position provided by the External Server will trigger a recalculation of the RSI server path. A consequence of this server architecture is that responsibility for issuing robot commands is shifted from the KRL program to an external control system that is wholly independent of the procedural limitations of KRL. Although we demonstrate an approach that dynamically assigns robot tasks and behaviors within a multi agent simulation in Java, the C# application could be integrated with other platforms such as Grasshopper/Firefly or Python over UDP. 04 — POLIMORPHIC BEHAVIORS (AGENT — ROBOT ) // Our strategy of stigmergic robotics explores computational behaviors that each describes a simple design procedure. It is the relative weighting and negoti-

Hashed Scanning: The digital representation (“environment”) of the physical workspace of the robot and multi-agent system is represented as a hashed collection of points that map a color value to a discretized coordinate in space. These values can be selectively updated at up to 60 frames per second by querying the buffer of a Kinect 2 scanner positioned above the workspace, or a Structure IO scanner

fixed to the end effector of the robot. The combination of scanners allows the robot to build up a 3 dimensional representation of its environment and avoid areas of occlusion. Behaviors allow the hashed values of the environment to be selectively updated by masking scanned points using a color range, or by cropping scanned points to a bounded region of space. Feature Analysis: The robot-agent is made aware of its local environment through feature analysis and detection behaviors. These behaviors iterate over collections of points within a bounded region of space to identify local geometric features (high or low points, plane approximation, normal approximation, curvature analysis) or find suitable conditions (closest point, closest value). Feature Exaggeration: Features within the environment are manipulated and exaggerated by associating agent behaviors with the actuation of end effectors that control material in the physical workspace of the robot. Because scanning of changes in the physical environment occur at the same rate as behaviors responding to changes in the digital environment, feedback between agent and environment takes place through the medium of physical material. The digital behaviors of the agent are encoded with an awareness of how the manipulation of physical material will effect the digital environment. Local features can be exaggerated (such as adding to high points or removing from low

The Nature of Robots

ation between these often competing procedures that creates the behavior of the robot and the emergent characteristics produced in this process of formation. De Landa (2000) argues that complex formations emerge when material negotiates differences in intensity. A difference of intensity exists at both the level of the competing computational behaviors as well as competing robots. Our experiments leverage the capacity of multiple robots to interact simultaneously on the formation of physical material — each embodying different intentions. Examples of which include robots competing to deposit or remove material, or robots that operate in a volatile fashion versus those which attempt pragmatic resolution. The robot is conceptually and syntactically an extension of the agent. The robot’s repertoire of behaviors expands on that of the generative design agent extending their capability into the physical realm through deposition and subtractive behaviors in response to real-time sensor feedback. The technical implementations of these behaviors are outlined as follows:


The Nature of Robots

points), material behavior can be exaggerated (for example by combining pouring material with computational behaviors that diagrammatically model fluid dynamics), and fabrication logic can be encoded (by approximating contoured paths, or by detecting and filling gaps).


Patient Agency: Multi agent systems are typically modeled with continuous motion, because it is the differentiation of the trajectory of the agents as they negotiate multiple constraints and conditions that tend to reveal higher orders of emergent formal possibility. However, when the agent becomes polymorphic with a single robot, the necessity arises to condition the behavior of the agent to the speed of the robotic fabrication process. We term these behaviors as “patient agency”, in which the digital behavior of the agent is made both semantically and conceptually to ‘wait’ for the completion of a physical task by the robot. This enables the robot and agent to operate simultaneously with the inclusion of specific tasks (such as deposition or pick and place), without resorting to a sequential process of digital simulation prior to robotic execution. 05 — MATERIALS // Expanding polyurethane foam facilitates experimentation across a spectrum of material behaviors. By controlling the expansion rate of the foam, the material can behave either in a highly predictable or highly volatile fashion. Because it is lightweight it can easily be used in an additive or subtractive fabrication

process. Relatively rapid cure times facilitate the construction of novel assemblages over time. Surface texture and opacity makes it easy to mask as the workpiece in scans. The extrusion of thermoplastic (HDPE-High Density Polyethylene) is comparatively stable. While the plastic shrinks considerably as it cools, it is sufficiently stable to build several consecutive layers. The limited tensile strength of the plastic during its extrusion enables bridges or catenary arches to be strung between fixed points. This capacity enables an importantly different array of formal possibilities to the expanded polyurethane. 06 — END - EFFECTORS // The actuation of end effectors takes place through UDP over a wi-fi network and is operated through the same control architecture that communicates with the robot. An Arduino running a client program listens for string commands that can then operate motors connected to a motor driver. Four basic tools will be used in the workshop that facilitate the primary operations of this research: a. Foam deposition gun: This off the shelf tool has been modified to be controlled by a stepper motor through a calibrated gear rack (Arduino controlled). A low feed rate allows foam to be extruded with a relatively high degree of control. Controlled bursts allow material to be placed, while a high feed rate results in highly volatility deposition.

b. Air turbine motor: This pneumatic spindle is used for subtractive operations including milling, brushing and swarf cutting; c. Plastic welder extruder: This thermoplastic deposition tool enables pelletized plastic to be extruded with consistent velocity based on a screw drive mechanism. The velocity of the extruder is controlled directly through the digital IO ports of the robot;

straints) it facilitates the exploration of broader design spaces and thus leads to greater formal diversity and novelty. Future advancements of this work will extend competitive behavioral models and explore self-organization of additive and subtractive behaviors driven by limitations on the total amount of material within the autonomous fabrication environment.

07 â&#x20AC;&#x201D; CONCLUSIONS and SPECULATIONS // The intention of this research is to explore a hybrid space of formal characteristics and expressions that emerge from the closeness and parallelism of material and robotic-agent behavior. This is primarily posited as a design approach that is potentially independent of fabrication. The robot is considered an extension of the agent, one that allows the computational agent to directly negotiate with the full, and perhaps unknowable, generative capacity of material. While the space of algorithmic and material expression has been heavily explored in the previous few decades, our concern is to imbue physical material with a digital agency. We speculate that as this digital agency has the capacity to negotiate competing conditions (such as design goals, material behavior, local conditions or fabrication con-

The Nature of Robots

d. Heat gun: A heat gun is used to melt previously deposited plastic. The behaviors developed for this equate to feature smoothing operations.



The Nature of Robots

Credits // Alexandre Dubor // Guillem Camprodom // Gabriel Bello Diaz // Dagmar Reinhardt // Rob Saunders // Kate Dunn // Marjo Niemelä // Samantha Horlyck // Susanna Alarcon-Licona // Dylan Wozniak-O’Connor // Rod watt //


ABSTRACT // This research examines the potential for creative practitioners to adopt robotic fabrication processes augmented with the introduction of sensors. Typically, the outcomes of a fabrication process are predetermined, however, with the introduction of sensors, design and fabrication process may be interrupted by real-time feedback. In such a system, design roles and authorship become secondary to the process of manipulating data, such that new rules of de-

sign can be introduced and developed in response to materials. Hardware and software such as Arduino, Grasshopper, Rhinoceros and Processing have opened up new strategies of hacking, coding and robotic manipulation that can be embedded in robotic fabrication processes. The addition of sensors provides feedback about material location and characteristics, work environment and co-workers, so as to support architectural dialogue. This paper proposes a framework for designing new protocols for human interaction and machine response in robotic fabrication systems. KEYWORDS // Human-Machine Interaction (HMI) // Collaborative Processes // Robotically-Assisted Design Creativity //generative Fabrication // Material Feedback // Robotic design Workflow // 01 — INTRODUCTION and MOTIVATION // Progress in robotic fabrication and manufacturing has accelerated in recent years through research in industry, practice, construction and manufacturing (Gramazio and Kohler 2014). Robotic fabrication labs are now embedded in professional practices, educational institutions and research centers across architecture, art and design. While robotic fabrication has extended previous automation processes of the automotive industry towards complex and singular fabrication solutions, the challenge is now to expand the negotiation of robotic processes-to influence toolpath options and define new material pro-

ing new protocols for human interaction and machine response. 02 — EVOLUTION of DIGITAL FABRICATION WORKFLOWS // 02.1 — FILE TO FACTORY // “File to Factory” has become more common as the availability of digital tools and digital fabrication has increased. Designers and artists have used these workflows as a way to materialize digital objects, allowing them to bridge the gap between digital and material worlds with an expectation that the machine will materialize their designed object as it appears on the screen. CAD/CAM (computer-aided design and computer-aided manufacturing) software has also become increasingly accessible, making the process of materialization easier. The gap between digital and material worlds is not a barrier to be overcome but can also be seen as a place for exploration and experimentation. While materialization is the focus of many practitioners, the classic “File to Factory” approach lacks flexibility and the opportunity for feedback as part of an exploratory process. 02.2 — PARAMETRIC PROCESS // Public interest in digital fabrication and the rising availability of 3D printers has allowed an increasing number of non-specialists to understand and adapt the logic and mechanisms behind the materialization process. It is now becoming common for users to change parameters of a digital fabrication process, feed-rates and the

The Nature of Robots

cesses-in short to introduce a form of design thinking for robotics with the goal of enhancing creativity and the evolution of design processes, models, and techniques. In this paper we ask: How do robots and humans work together to explore material agency? How does the application of robotics expand design affordances or intuition? Robotic fabrication processes enable designers and architects to explore the boundaries between digital and material worlds. Beyond optimization criteria or parametric design, new design strategies such as generative design and collaborative design are enabling new ways of approaching material exploration through robotics. Open source software and hardware enable new forms of design, yet these new tools also demand design frameworks dealing with robots, data, sensor technologies and material contingencies. Like computational composites, robotic composites posit a challenge: How do we think about hybrid processes that bridge different ‘hardware’ (robot, human, end-effector, material) and ‘software’ (data, programs, toolpaths, workflows)? This research proposes a framework for robotic fabrication, which links data, workflow, interaction, feedback, material behavior, protocols and time as major project constraints. This paper provides an overview of different creative practices using robotic fabrication augmented by sensor feedback. It examines the feedback loops involved in these practices and concludes with a proposal for a framework for design-


The Nature of Robots 252

nozzle temperatures, to adapt it to their specific requirements. In addition, the spread of open source hardware and software has empowered hobbyists, artists and designers to build their own machines, permitting the rise to new types of machines and fabrication processes. In architecture, industrial robots have proven to be a robust and flexible research platform, allowing the precise placement of many types of tool within a large working envelope, allowing the designer to focus on the design of novel end-effectors and processes. In parallel, parametric design has gained momentum in many design disciplines as a way to explore a space of possible designs when the final outcome is not precisely defined. This has resulted in a shift from shape design to process design by empowering designers to manipulate the fabrication parameters as well design parameters. Recent initiatives have brought computer aided manufacturing (CAM) into parametric software, giving designers access to tools and languages to manipulate both design and fabrication, considerably decreasing the learning curve as well as the speed of exploration. Introductory digital and robotic fabrication workshops at Institute for Advanced Architecture of Catalonia (IAAC) and The University of Sydney combined parametric tools, Rhinoceros3D + Grasshopper3D, with CAM plugins, KUKAlprc, to allow students to explore the potential and limitations of robotic fabrication processes, 3D printing. By varying parameters exposed within a

predefined process, students are able to learn from materialized results and move quickly through iterations. Within this parametric workflow, teachers and students analyze the results of iterations and provide the feedback for material exploration. Consequently, students are able to achieve significant results within a day. 02.3 â&#x20AC;&#x201D; LIMITATION AND CHALLENGES // The division between design and fabrication process is slowly disappearing in favor of a continuous form of design, which includes fabrication as an essential element. While providing a great framework for fast iteration and exploration, linear approaches reach their limit when fabrication becomes more complicated, requiring lengthy iterations. In addition, complex fabrication processes that use non-static materials, e.g., clay or polymer, require more precise and sensor feedback to enable tracking, fine-tuning and synchronization between material, machine and design. Sensors thus enable real-time feedback loops that have the potential to radically change the design process. 03 â&#x20AC;&#x201D; (IM)MATERIAL RESPONSE // When material is understood as relative to time and protocol, material transformations can be considered as a series of actions influenced by a range of variables or agencies. These include immaterial factors such as velocity, density, mix ratios, temperature and evaporation. As such, an indeterminate, unpredictable material self-formation can be considered a

03.1 â&#x20AC;&#x201D; MATERIAL as PROCESS // In deposition processes, Free Form Fabrication (FFF) or extrusion-based 3D printing, materials are processed by the extrusion of a liquid, or viscous, material, e.g., clay, wax, concrete, polymer. The success of the extrusion is highly dependent on the material properties being adequately linked to the fabrication protocol, e.g., feed rates and toolpaths. As 3D printing, parametric design and CAD/ CAM technologies advance so does the need for control, manipulation and development of suitable materials. A coupling of material protocol to sensors can enable new design approaches. For example, predicting the final deposition location of a clay extrusion implies calculating the shear viscosity of clay at the extrusion point. This depends on environmental conditions (e.g. air temperature, relative humidity) and the time the material undergoes shear. By obtaining the fluid speed, based on the pressure exercised on the material, it is possible to calculate the vector that the extrusion will follow. Finally, by obtaining the deposition plane position we can calculate the final extrusion location and adjust the fluid speed to match a deposition target. A model provided with real-time data

from direct (extrusion cylinder pressure) and environmental (air temperature, relative humidity) sensors can deal with complex material behaviors. These behaviors are difficult to predict within digital simulation and make the use of predetermined tool-paths obsolete, as they have too little tolerance to guarantee a successful outcome. The use of sensors allows for the bridging of the gap between the expected outcome and reality. The use of sensors is critical for understanding complex material behavior. Digital sensors are devices capable of turning physical properties into data. Traditionally industrial sensing equipment has been tied to specific industry sectors, making them expensive and difficult to operate. The rise of consumer electronics such as digital cameras and smartphones has made available low-cost digital sensors for a wide range of physical properties, e.g., temperature, proximity, pressure. 03.2 â&#x20AC;&#x201D; MATERIAL FEEDBACK SENSOR TOOLKIT // The availability of digital sensors makes it possible to assemble an inexpensive toolbox of sensors useful for digital fabrication. Multiple approaches to sensing can be quickly tested in order to understand how a material behaves before moving to more specific, industrial grade solutions. Furthermore, the development of open source hardware and microcontroller platforms, such as Arduino, has democratized access to electronics by providing tools and documentation. At the same time digital fabrication tools allow for the cus-

The Nature of Robots

material computation. In some cases time may affect material conditions, e.g., velocity may be a factor affecting toolpaths when working with semi-liquid materials that exhibit sedimentation.


The Nature of Robots

tomization of sensors. The Material Feedback Sensor Toolkit is a first attempt at establishing a collection of sensors and tools for sensing material behavior. None of the sensors listed are industrial-grade, instead they were developed for consumer electronics. The use of consumer-grade sensors can require more work than industrial sensors but this is compensated by the low cost and extensive range of the sensors available. The use of these sensors has been made possible due to the work done by the open hardware community in documenting and exploring the use of these devices. The sensor selection prioritizes low cost, open source and the existence of good documentation. Most sensors are compatible with the Arduino electronics platform. The total cost of the toolkit, including wiring and the Arduino development board, is less than 1.000 USD.


04 â&#x20AC;&#x201D; CODING INTUITION: EMBEDDING SENSOR and LOGIC in DESIGN // Gathering the right data is only half of the process in a feedback workflow. The data must be turned into decisions and finally actions. Embedding sensors in fabrication processes is not new. Closed loop control systems such as the Watt (or centrifugal) governor date back to the origins of the industrial revolution and have been extensively used in industry since. Closed loop control systems are based on the idea that an error in a system can be corrected by continuously measuring the output

with a sensor in order to adjust the input based on a threshold. Originally, control systems were designed in the form of analogue devices tied to their own mechanics. Digital sensors and microcontrollers allowed industrial control systems to become cheaper, smaller and more easily programmable. Despite the importance in industry, however, traditional control theory is focused on process efficiency, optimization and safety. Approaching feedback from an experimental point of view requires a different approach. When control systems in fabrication are seen from a material instead of a machine perspective, the design of the controller becomes part of the design process itself. Consequently, the focus becomes exploring the material by connecting its behavior to the machine control system using relatively simple logic. This is critical when we look at how the complexity of modeling certain fabrication processes using tools like Grasshopper can result in significantly less experimentation. 04.1 â&#x20AC;&#x201D; INTEGRATING MATERIAL FEEDBACK into DESIGN SOFTWARE // All the sensors in the Material Feedback Sensor Toolkit can be integrated into parametric design software such as Grasshopper to allow designers to integrate material feedback into their digital design process. For example, in the Magnetic Architecture, data from a camera informed the decision-making process for each step according to the materialization of the previous tool path. The experi-

04.2 — ENCODONG the LOGIC into the MACHINE // When working with continuous fabrication processes, e.g., material extrusion in additive manufacturing, real-time feedback is required. Industrial robots can be connected in “near real-time” with parametric design software, e.g., Hal Robotics streaming supports communication speeds up to 5 Hz, allowing the feedback loop to be significantly shortened. Continuous path adjustments, however, need even faster reaction times requiring the logic to reside within the robot controller. The threshold values for sensors were defined in Arduino code and used to rigger digital or analog inputs on a KUKA Robot Controller (KRC). KUKA IO was developed for this experiment to facilitate communication of the Arduino with the KRC. Using this framework, a simple KRL script (approx. 20 lines) produces a rapid feedback loop (<20 ms) encoding the desired logic. The material process used, plastic extrusion, is difficult to predict but could

be tracked in real-time using distance and temperature sensors. This framework and the specific workflow have some significant limitations. The single byte that was exchanged through the input/output port of the robot controller restricted the control that the Arduino could have on a running process. In addition, microcontrollers, such as those used on and Arduino Uno, have limited processing ability, restricting the types of sensors that could be used. Finally, traditional robot control languages, such as KRL, are restrictive when compared to modern scripting languages, which limited the possibilities available to experienced coders and made it difficult for inexperienced users to code logic to produce desired behaviors. The importance of manual experiments to understand a material’s behavior became apparent. Manual tests were conducted to simulate the sensor-robot logic and understand what needed to be scripted. The need for manual experiments may have been avoided with better support for rapid development of control software for the purposes of material experimentation. The most common solution to these limitations is to externalize the controller on a remote computer giving users the possibility to code the robot motion and behavior in another language, Java or C/C+, and communicate through a faster, machine-specific protocol. An example of such framework is OpenKC, which is an open source, real-time control software specifically designed for the KUKA Light Weight

The Nature of Robots

ment used the Firefly plugin to feed the data from the camera into a Rhinoceros3D + Grasshopper3D script, which produced code using KUKAlprc. In this setup an iterative logic is encoded to compute each successive toolpath one step at a time. In this experiment, sensors permitted the integration of a self-organizing process as part of the fabrication and design. The feedback loop in this fabrication process took approximately a minute, limiting opportunities for experimentation.


Robot (LWR), coded in C/C+, and based on the KUKA RSI-XML interface. Robot manufacturers are also starting to make their controllers more accessible to researchers and designers. Universal Robots, and more recently KUKA, offer APIs to control the motion and get information on a robot’s state.

The Nature of Robots

05 — DESIGNING PROTOCOLS for HUMAN-MACHINE INTERACTION // The discrepancy between material contingency, digital control, technological limitations and designer’s reativity reveals the difficulties in defining a suitable interface to interact with in this context. We envision an ideal framework to facilitate interaction without the technological issues mentioned previously while providing space for creativity through craftsmanship, e.g., manual experiments, and generative fabrication, e.g., fabrication responsive to material behavior.


05.1 — CRAFTMANSHIP and DIGITAL FABRICATION // Investigating ways of depositing materials that have been traditionally formed either by hand, such as clay, opens up the possibility of investigating the place of the handmade and the concept of analogue authorship in digital fabrication. For example, can the author, designer and creative practitioner alter a program that has been set in motion by interacting with sensors? L’Artisan éctronique by UNFOLD addresses the manipulation of a printing process through human intervention

using sensors. A separation of the human hand and the material process, however, caused a delay and a disconnection in the creative process between a user’s input and the material feedback. Objects of Rotation was a project undertaken at the Harvard Graduate Design School allowed the use of mark-making processes on rotating clay. The clay is unresponsive, however, and there is no place for the human hand. Both of these projects go some way to addressing the place of the handmade in digital fabrication and how creative practitioners may utilize robotics. But there remains a need to investigate the smooth exchange of design intention between analogue and digital processes-an exchange that opens a space for spontaneous, reactive authorship in digital fabrication. We propose the exploration of a new framework where a craftsman’s intuition and sensibility can be combined with the power of digital analysis and the precision of robotic fabrication. We envision this framework being particularly useful for fabrication involving complex material behaviors such that it remains open-ended for creative exploration. To test the proposed framework, we are exploring clay-modeling processes using additive manufacturing. 05.2 — EXPERIMENT 1: ALTERNATING MANUAL and DIGITAL MANIPULATION // Our first experiment will introduce 3D scanning and data from other sensors, room temperature or humidity, within a manual fab-

05.3 — EXPERIMENT 2 : HUMAN FEEDBACK within REAL TIME PROCESS // While additive manufacturing with clay has been used since ancient times, however, 3D printing layer-by-layer is quite different from traditional crafts, requiring a level of precision that is almost impossible for craftsmen, especially when trying to have homogeneous material deposition. The introduction of cooperative robots able to safely share a workspace with humans opens up the possibility of a robot and human working simultaneously on an object, possibly with the same tool. We propose using the force feedback sensors of a KUKA LWR iiwa robot to feel the indication of a user manipulating a tool attached to the robot. The tool and the robot would be free to move until it reaches one of the constraints dictated by a model. In the case of 3D printing with clay, a robot might maintain a constant speed in the XY plane in response to human input. In a similar fashion, movement can be constrained to a specific height from existing object using data from a distance sensor attached to an end effector. This would allow a user to move freely along an extrusion path while maintaining the specific constraints of the fabrication method, extrusion speed and layer height. Such real-time feedback needs to be programed with a fast response rate and therefore requires coding in the robot language to achieve a feedback cycle of less than 50 ms. An external link will be used to connect the robot controller to a separate

The Nature of Robots

rication process allowing information about the manual process to be captured. Computational analysis of the process may allow improvements in the iteration of a design by providing the designer with specific information, geometrical, topological or structural analysis. In addition, the data may also be used to elicit feedback from remotely located co-designers or clients. Having digitized a manual fabrication process the reverse of the process would be to “materialize” the data captured. An additive manufacturing process will be used to reproduce the previously scanned object. This materialization will allow a network of collaborators to get physical copies of the object and the possibility of manipulating the object, by modifying the shape. Using integrated CAM software, such as KUKAlprc, we can close the loop of digital iteration using a common platform with a feedback loop of minutes or hours. This experimental setup will allow the reproduction of a manual design task and generate an “augmented” fabrication but a clear difference will still exist between the human produced and the 3D printed copies. These differences will be evident at the multiple scales, material continuity, physical behavior, and texture.


The Nature of Robots

computer where each robot position is recorded. This data will serve to make a session reproducible without additional human input but also provide feedback from digital analysis of the object being produced. This analysis can then be projected back on the workspace or object to provide a non-invasive feedback with a slower response rate (>1s) to complement the real-time force feedback. In such a setup, the user is not only exploring the toolpath by moving the tool in space but also the different parameters of the fabrication process, by changing the rules that the robot follows. These parameters and logic become core information in the design research-information that can be shared with a community and continuously adapted.


This paper has provided an overview of an evolution of creative processes supported by computational design and fabrication and the potential for future changes supported by data feedback. The paper has discussed this via a series of case studies examining different feedback loops and a proposal for a framework for designing new protocols for human interaction and machine response. The act of giving a machine freedom to assist the creative process leads to unexpected and useful information both from the machine and material perspective. By coupling Human-Machine Interface with robotic fabrication, sensor feedback and digital computation, new possibilities for creative collabo-

ration is appearing. Collaboration between robots and human can enhance creativity and innovation by supporting designer and researcher while exploring complex material system. Such material exploration through robotic fabrication can gain precision and in depth information from sensor analysis of the material, the context and the userâ&#x20AC;&#x2122;s movements. The advantages associated with an open-source framework and low cost sensors may permit widespread adoption of this approach and enhance new collaboration between researchers and designers. The creation of a flexible framework for Instant Robotic Toolpath Revision intends to make such practice more accessible to a wider range of designer and researcher and aims to extend its applications to other fields and industries.


the typically linear and compartmentalized nature of the processes linking design with construction and therefore open up new ecologies of design practice and opportunities for innovation.

Credits // Dave Pigram // Iain Maxwell // Wes McGee ABSTRACT // This paper identifies the disciplinary potential latent in the combination of algorithmic design and sensor-enabled robotic fabrication to achieve multiple channels of feedback between design, fabrication and assembly. Three key methodological shifts are identified. The first is a shift to fabrication-aware-form finding. In comparing analogue form finding to digital form finding practices, it is clear that a greater number and diversity of constraints can be negotiate within an information-based digital environment. The second methodological shift is to bi-directional negotiation between design and production limits. Robotic fabrication is highly customizable-initial constraints do not need to be seen as fixed. The final shift is the introduction of sensor feedback and near real-time control. This permits the continual redefinition of parts during fabrication in response to material, dimensional and assembly-volatility. Taken together, these shifts challenge

01 â&#x20AC;&#x201D; GENERAL // This research posits the possibility of an expanded definition of architectural form finding by embedding algorithmic design methodologies and robotic fabrication strategies in the form finding process. Three emerging research trajectories are presented that when seen collectively point towards a highly flexible and integrated approach to design: Fabrication Aware Form Finding // Robotic Sensor Control // Adaptive Part Variation. These interrelated fields of research have led to the realization of highly productive feedback loops between the once separated domains of design, part production and assembly. This framework not only consolidates and embeds the logics, constraints and viability of part assemblies (and production) within the generative cycles of design but, critically, affords increased opportunities for innovation throughout the entire production chain. Each of the concepts introduced will be situated through the description of a case study.

The Nature of Robots

KEYWORDS // Robotic Fabrication // Algorithmic Design // Fabrication-Aware form finding // Digital Form Finding // Analogue Form Finding // Sensory robotic control // Adaptive Part Variation //


The Nature of Robots

02 — ANALOGUE FORM FINDING (AFF) // Form finding seeks the discovery of optimal shapes against a known set of design constraints-typically structural. Exemplary methods include: Antoni Gaudi’s nested hanging-chain models to resolve highly complex load paths within compression-only masonry structures and Frei Otto’s study of minimal surfaces such as soap-film bubbles to realize complex membrane structures. Typically, Analogue Form Finding (AFF) methods attain a single design solution, via equilibrium, for a given staffing state: material substrate, boundary, support conditions and assumed loading scenario.


03 — DIGITAL FORM FINDING (DFF) // The analogue methods above can and have been widely reproduced digitally through the implementation of dynamic relaxation. “The basis of the method is to trace stepby-step for small time increments, Dt, the motion of each node of a structure until, due to artificial damping, the structure comes to rest in static equilibrium” (Barnes 1999). Kilian and Ochsendorf (2005) describe a method for the re-conception of architectural geometry (surfaces, volumes etc.) as a network of weighted particles (nodes) tethered to one another by variable springs (members). Within such a computational approach, each constituent particle negotiates its immediate neighborhood of connecting springs (topology) towards a state of equilibrium or equal residual force (towards a given spring length).

Dynamic relaxation (DR) is therefore the iterative application of Hooke’s law of elasticity: for elastic deformations of an object, the magnitude of its deformation (extension or compression) is directly proportional to the deforming force or load. The advantage of implementing form finding methods digitally is that the initial design constraints, topological configuration and constituent variables of the system (i.e. node masses, spring lengths and elasticity) can be managed, modified or discontinued during the form finding process. Additionally, embedding variegated materials (weights, strengths, thickness and densities) is a simple numerical operation applied to either the nodes or members allowing for an accelerated exploration of numerous design solutions. Clifford and McGee’s La voûte de LeFevre project varies element thickness to reach a target vault underside geometry, serving as an example of a form finding inversion virtually impossible with analogue methods. The digital allows for both an engagement with and a convenient departure from the pragmatic constraints of the physical world. 04 — FABRICATION-AWARE FORM FINDING (FAFF) // 04.1 — DEFINITION // Fabric-Aware Form Finding (FAFF) seeks to embed the constraints of manufacture – material handling and processing-within the iterative cycles of a generative computational design framework and should be viewed as a consequential

04.2 — FAFF CASE STUDY: UTZON | 40 PAVILION // The Utzon | 40 pavilion serves as a purposeful demonstration of the possibility of a FAFF design approach. The pavilion completed for the 40th anniversary of the Sydney Opera House draws inspiration from the material palette and innovations of Jorn Utzon’s design

while positing the capacity of algorithmic design methodologies coupled to CNC production chains to overcome the challenges of standardization that so problematically impacted the original.To do so, an expanded implementation of dynamic relaxation (DR) was developed that allowed for the simultaneous negotiation of global shape and the constraints of part manufacture. All critical aspects related to the 5-axis CNC machining of parts were re-conceived as dimensional and geometric limits (minima and maxima) influencing the spring-member network that defined the DR model. Essential fabrication constraints and their translation included: • Minimum (450 mm) and maximum part length (1200 mm). These dimensions were governed by the vacuum pods (150 mm x 150 mm) used to fix the parts during machining and acknowledged the very practical concerns of minimizing the repositioning of parts, maintaining five-axis clearances and restricting the cantilevering of parts within an acceptable range. The slack length is decreased during the dynamic relaxation process to counter an exceeding of maximum length and vice versa: • Maximum angular deviation between node normals (15°). This parameter results from the design detail where each member’s top and bottom edges are cut as twisted ruled surfaces to eliminate stepping at the joint. Greater node deviation resulted in very sharp and vulnerable edg-

The Nature of Robots

extension of the form finding traditions of architecture. Such approaches demand the development of specific parametric or algorithmic protocols, implemented via bespoke software or scripts that extend the functionality of existing software. Typically, such models mix opportunities of user interaction with underlying orders of geometric control and/or optimization routines to ensure only valid outputs are derived. Examples of criteria for validity may include topological limits (allowable connections), panel shapes (triangles, quads), curvature (planarity, single, double, syn- or anti-clastic) and maximum and minimum part or angle sizes. As such, FAFF positions itself in opposition to the typical events of rationalization that traditionally occur after “designing” has finished. Within a FAFF paradigm, the transformations that a design undergoes in order to be built happen within the primary design process and thus the outcomes of these steps (positive or negative) can feedback upstream for negotiation with other factors. In this way, the viability of a project can be embraced by the designer, avoiding later degradation of design intent during building.


The Nature of Robots

es. When this value is exceeded, the slack length of the member between the nodes in question is contracted to pull the nodes closer together, thus reducing normal variance;


05 — ROBOTICS and FABRICATION AWARE FORM FINDING // Fabrication with industrial robots differs from the use of dedicated or single-purpose machines (NC or other) in the obvious and here extremely significant aspect that they have been designed specifically to support many styles of customization. The use of custom end-effectors and control interfaces for example can significantly broaden the range of possible outcomes. It is this aspect that is an amplifier of the potential of more typical implementations of fabricationaware-form-finding where it is often assumed that the constraints are fixed and so feedback takes the form of iterative design changes assessed for viability against those fixed constraints. With robotic fabrication, a system that permits evolution, the feedback can occur in both directions. Negotiated changes could be demanded of the fabrication system and not only of the design. The assessment of a proposed design (or component, process etc.) would not only include an indication of viability but a measure of what changes to the constraints would lead to that design achieving viability. In this way design loops that understand limitations, can be extended to also analyze the payoffs from reducing those limitations. There are many ways that changes to the fabrication system could be made

to increase capability. They may be physical or information based. Examples include: • End-effector redesign; • Implementation of sensors to respond to various forms of volatility; • Additional empirical tests to better test specific edge cases and verify associated limits; • Algorithm optimization for con ditions prevalent in a given de sign; • Interface changes for example to change the calculation of ori entations to relieve pose is sues; • Fixture redesign, or allowance for the production of multiple fixtures; In the case of architecture, where the cost of many building packages – façade panels - far surpasses the cost of most industrial robotic equipment, and certainly the cost of modifications to existing equipment, it is entirely likely that a cost-benefit analysis would reveal that enhancements to a fabrication system would pay for themselves by absolving the need for changes to a project. Clearly many other non-financial barriers remain-the frequent demand for competitive tenders for example-though some of these may also be negotiable via feedback’ One might imagine a tender process where companies with increased production capabilities can bid based on the enhanced project that they could deliver, clearly under-

06 — SENSORY ROBOTIC CONTROL (SRC) // 06.1 — DEFINITION // Sensory Robotic Control (SRC) introduces real-time sensor data as a dynamic feedback during machining operations. The concept massively challenges the explicit nature of machine instruction code, namely pre-ordained blind motion, and mechanisms of visual, force or other sensing activities. 06.2 — CASE STUDY – SENSOR ENABLED TOOLING // Over the past two years the robotics lab at the Taubman College of Architecture has been gradually implementing an external Programmable Logic Controller (PLC) system to expand the capabilities of its 3 dual Kuka Robot workcells. Numerous manufacturers produce compatible PLC systems, in this case the external PLC is implemented using Beckhoff’s TwinCAT XAE 3.1 software and Beckhoff hardware, and communicates via the EtherCAT protocol with Kuka’s KRC4 controller. One example of sensory tooling is a stepper controlled automated fiber placement head. In this tool a NEMA 17 size stepper is used to feed composite tow through the head, both as “restart” after cutting at the end of path, and during placement in “velocity synchronous mode”. Start and stop commands are issued via subroutine commands within the KRL (Kuka Robot Lan-

guage) program, and the actual motor control is implemented via PLCOpen function blocks under TwinCAT 3.1. The second example is the ongoing development of extrusion heads for additive fabrication processes. In this case a 2 kw servo is controlled via the external PLC to synchronize with the robot motion. As the exact deposition rate is somewhat non-linear with the velocity of the motor, a “cam table” is implemented (also using PLCOpen function blocks) to adapt the motor velocity to the robot velocity. This cam table is empirically generated, and corresponds to specific polymer viscosities and temperatures. While not impossible, such a control framework would be relatively difficult to implement natively in the robot controller. A smaller stepper motor driven extruder has also been implemented using the same framework. 07 — ADAPTIVE PART VARIATION (APV) // All assembly processes demand attention be paid to dimensional variation and the provision of tolerances. This is particularly true within architecture where: parts are either large, come in high populations, or both; cumulative errors are significant; and site irregularities are common. Adaptive Part Variation is a strategy where the real-time redefinition and fabrication of parts occurs during the serial process of assembly thereby allowing detected errors to trigger a conditional design response. Necessary preconditions to establishing APV workflows are: the ability to progressively measure an assembly

The Nature of Robots

stood against the alternate project deliverable by a company with truncated production capabilities.


The Nature of Robots

in process; to recognize key features within that assembly; and to update the digital model to register variance. Adaptive Part Variation demands a radical repositioning of design intent away from static geometric descriptions towards relational frameworks and the definition of contingent responses to error. One immediate consequence of this shift is a radical reappraisal of prevailing approaches to the provision of construction tolerances. Typically, architectural details must consistently cater to a predicted worst-case, APV however, affords the definition of an enriched set of design responses that can directly address each actually occurring situation.


07 â&#x20AC;&#x201D; VOLATILITY // The conventional procurement of architecture operates on an assumed resilience of the design against the encroaching scales of economic production and the reductive and expedient cultures of building delivery. Scenarios generally characterized by linear and highly compartmentalized workflows. Each of the thematic introduced here, in one form or another, seeks to embrace volatility within the ecology of architectural production. Fabrication-Aware Form Finding (FAFF) seeks the realization of an increasingly flexible design model capable of negotiating the often-competing imperatives for design within its earliest phases. Critically, it seeks the establishment of multiple pathways through which such transactions can be conducted, allowing the designer to elect the trade-offs of most benefit to the greater project.

Sensory Robotic Control (SRC) challenges the explicit command of machines through the introduction of either on-board or design model feedback mechanisms of adaptation and response. Here the pre-programmed point-to-point and linear moves of G-Code, Kuka Robot Language (KRL) or similar are toppled by increasing environmental, material or task awareness. Finally, Adaptive Part Variation (APV) challenges the static and predetermined nature of conventional construction detailing that must consistently cater to the worst-case via generic construction tolerances. Instead, APV opens the possibility of deploying distinct conditional responses to local moments of dimensional variance. Observed collectively, the acceptance of volatility suggests that the pursuit of automation could be reframed as the search for a new form of design intelligence, where the logics of design and making intermingle within a reflective design-to-construction model. 08 â&#x20AC;&#x201D; CONCLUSION // The traditions of form finding have provided fertile territory for architects to produce sophisticated forms with rational underpinnings that contribute to their viability as well as their highly specific character. The digital implementation of form finding has allowed an increase in the number and nature of negotiated inputs and the addition of fabrication constraints has provided both pragmatic and inventive influence as well as allowing for exploration of design and material territories

The Nature of Robots

of high volatility. Robotic fabrication in combination with algorithmic design methodologies and file-to-factory workflows affords new pathways of reciprocity between design and fabrication. Scenarios where discovery and invention may initially occur on either side of the design and making ledger but must ultimately permeate the entire design process. The addition of computer vision sensors extends this paradigm, adding a further feedback loop to include responses to real-time events and dimensional errors prevalent during construction. The series of shifts outlined in this paper collectively challenge the dominance of linear and compartmentalized models of the production chain linking design with construction. Taken together they set the context for new ecologies of design practice premised on increased reciprocity and provide many opportunities for innovation.


The Nature of Robots

Temporary Conclusion


It is tempting, indeed, to view the introduction of robots in architecture as a reformulation of modernist efforts to transform the field of architectural production into a fully automated and thoroughly rationalized industry. This conclusion falls short, however. What is under way is a comprehensive development in the entire practice of architecture: the unification of design and production that – in conjunction with digital, processes - opens up entirely new opportunities for architectural materialization. The robot thereby revises the architectural notion of machinic processes and calls into question the previously clear separations between design and construction, information and mechanics, and technology and building culture. More than one million multifunctional industrial robots are in use around the world today, predominantly articulated arm robots. Their number has risen steadily in the past decades. The fundamental condition for automation in industry and, in turn, for the dissemination of robots is information technology, which in the early 1980s began to allow for machines with digital controls. While this first wave of “digital automation” radically penetrated many industry sectors such as automotive manufacturing and put into place entirely new, previously unthinkable standards of productivity and quality, the dissemination of digital production machines in the building industry remained a marginal phe-

nomenon. At this point, we distinguish construction automation and general robotization from the use of robots in architecture discussed in this book. Against this background we are striving for a definition of robots that is targeted to architecture and is therefore distinct from that used in other disciplines or industries, and which is at the same time more precise than the ideas that are commonly associated with the term robot. The exploration of the robot has two delimiting factors: the first is that we are not aiming at the technical development of robots as such; central to our research, rather, are robot- supported materialization processes viewed exclusively from an architectural perspective, This harking back to a proven and inexpensive (because mass-produced) fabrication machine, that is at once robust and flexible, distinguishes our approach fundamentally from the efforts of the “early days of robot-assisted construction”. Or to put it another way: We perceive the industrial robot as the quintessential, generic production tool. This second delimiting factor arises, then, when we ask ourselves whether the potential of robots has already been exhausted in traditional industrial applications; that is, whether the paradigm of repetition presents a sine qua non. The true potential is to correlate explicitly the robot’s flexible and generic nature with architecture’s specific production conditions, and not the other way around.

The Nature of Robots



The Nature of Robots 268

Yet with the new millennium, after postmodernism had become mired in philosophical discourse during the 1990s, digital technologies made a breakthrough on a broad cultural level and had an increasing impact on the architectural understanding of design. With the new “virtuality of the digital world”, the design disciplines were witness to a sudden proliferation of formally complex, yet materially indeterminate architectural visions. Completely without gravitas, architecture, with its virtual environments, momentarily stood at the pinnacle of autonomy, fixed on purely superficial properties of the abstract virtual world. Yet at the same time the “digital, project” increasingly gained material content through the introduction of computer-controlled processing machines, such as milling and laser cutting, adapted from other industries. Focusing on their conceptual openness and versatility, the de bate over robots can be decisively expanded. Precisely these characteristics, which open up an unsuspected degree of freedom in the reciprocal spatial relation of the machine to the object, allow a constructive Liberty, which distinguishes the industrial robot from specialized digital fabrication machines. In other words, design and building do not take place outside, rather inside, a spatial-machinic logic of materialization. This not only implies a profound technical confrontation with the medium of production itself but also enables a direct, conceptual link between the spatial characteristics of design and the spatial-constructive material-

ization possibilities of the robot, and thereby considerably, y expands the architectural design concept. The use of the robot in architecture represents not so much the continuation of unfettered industrialisation as the expression of a process more fundamental for the information age: the emergence of a weaving of diverse references that increasingly blurs the distinction between the author and the actual producer. This new “mentality” invalidates the historically shaped ideas of the machine, the concepts of modernist division of labour and work sequences, as well as the related spatial and temporal surveying of rigid human-machine systems. Quite clearly, the construction site of the future will be a complex “ecosystem” consisting of human beings and machines, in which the cooperation of robots with humans will play a decisive role. Wherever human abilities are superior, manual intervention can be selectively applied to complement the abilities of machines. It astonishes that current debate about the relationship between human beings and machines voice, almost exclusively, reservations whose content can be traced back to the early years of industrial automation. These reservations pertain to the potential destruction of jobs, craft traditions and social values. In1995, when the majority of people did not know what the Internet was, the American Jeremy Rifkin, until, then an outsider economist, prophesied the “end of labour”. However, the thesis of the end of labour through

The Nature of Robots

automation - regardless of what kind has been refuted over and over since the beginning of industriatisation. Alhough it is true that machines have repeatedly forced painful structural changes on society, in the end increased productivity not only resulted in a higher standard of living but also led to new professions and jobs in other places. Clearly, it can be expected that architecture will not be immune to such structural changes induced by robotic technology. The answer to this profound challenge is to establish a reciprocal think between robotic technology and the materialist reality of architecture. The synthesis of these previously separate domains with, dissolve previous dichotomic principles such as machinic/manual, digital/ physical, dynamic/static. This synthesis calls for a positive, but in no way naive handling of these new technologies to confront them with the architectural substance and therefore to inscribe them culturally into the discipline. Therefore, to look at architecture from the side of robots, and all that becomes possible through them, is only the other - still largely uninvestigated - side of the coin.



The Nature of Robots

The Nature of Robots