Powering a Path to
Autonomous Vehicles PATH Supporting California DMV in Development of Regulations Functional Safety Standards BMW, Intel and MobilEye Take Autonomous Driving to the Next Level
Lyft and the Third Transportation Revolution The California Connected Vehicle Test Bed —and More
INNOVATION Wind River and Intel drive the future of transportation
Everyone knows the automotive landscape is changing. Less clear is how to make the most of that change. Wind River is helping build the highway to our transportation future. Powered by deep expertise in industries demanding rigorous levels of safety and security, Wind River and Intel are actively working on automotive technologies that speed the development for tomorrowâ€™s connected and autonomous cars.
ÂŠ 2016 Intel Corporation. All rights reserved. Intel and the Intel logo are trademarks of Intel Corporation in the U.S. and/or other countries. *Other names and brands may be claimed as the property of others.
SURVIVING THE OPEN ROAD In the 1950’s storied highways like Route 66, the “Mother Road” beckoned weekend travelers and families would pile into station wagons and hit the highway. In our parent’s era, potholes and unskilled drivers were the worst a family would face on the open road. Today, it isn’t only drivers you have to deal with—it’s cars without drivers. In December, California’s DMV shut down Uber’s autonomous vehicle pilot program and revoked the registrations of their 16 autonomous vehicles because Uber failed to get the proper testing permits. During the trial, the car ran a red light. An investigation revealed that the human driver failed to react in time to correct the situation. In this issue, the second part of our article from the Michigan Department of Transportation addresses the Impact of Automated Vehicle Technologies on Driving Skills. The study concludes that human drivers need to be trained to take control of autonomous vehicles in situations like what occurred on the streets of San Francisco. The article on PATH Supporting the California DMV in Development of Regulations for Automated Driving Systems tackles the issue of balancing the need to protect the general public from unreasonable hazards posed by new systems that have not been well engineered, while encouraging the development and implementation of systems that can be demonstrated to improve traffic safety. The article also explains the new Federal and California state regulations that govern the testing and operation of automated driving systems. Lessons on when to take control of an autonomous vehicle can be learned from the exhaustive research conducted by Boeing and Airbus on pilots taking the controls from autopilot, and closer to earth, Nissan has written a fascinating article on how their engineers use experiences learned from designing and operating NASA’s Mars Explorer Robot to develop autonomous vehicles. Titled Putting Lessons from Mars to Work on Earth—Insights on Autonomous Vehicle Research, the article proves how public-private companies can work together effectively to develop real solutions. One of the best methods for proving the reliability of connected vehicles is the California Connected Vehicle Test Bed, which provides an ideal environment for companies to test and develop connected vehicle applications in a real-world setting. At Electronica and Embedded World in Germany, CES in Las Vegas, and other shows throughout the world, technology companies are showcasing their technology. In the future, working closely with organizations like the DMV, ITS and PATH and the Connected Auto Organization, these companies will have a head start on getting their cars out of the lab and onto the open road. The Connected Auto Organization, like its partner, California PATH, is built on the vision that safety can be systematically improved by adopting sensible approaches that address factors including infrastructure, vehicles, road users, and the interactions among them. So join the program, hit the highway—and let’s be careful out there. Glenn ImObersteg Thomas West Convergence Promotions PATH Co-Director
16 About Connected Auto Convergence Promotions LLC and California PATH (Partners for Advanced Transportation Technology) at the University of California, Berkeley, have partnered to advance the technologies that connect vehicles to the surrounding infrastructure and other vehicles. The goal of the partnership is to provide synergy and communications between academia, public institutions, automobile manufacturers and technology companies in the connected and autonomous vehicle industry.
Advisory Committee Alexandre M. Bayen Director, Institute of Transportation Studies Thomas West Co-Director, California PATH
Connected Auto Publisher Glenn ImObersteg Convergence Promotions LLC email@example.com Design and Production Carol Smiley, Graphic Designer Dave Ramos, DR Design
Sales Ruby Brower, Alan Hutton Copyrights and Credits The masthead, logo, and design of Connected Auto are copyright 2017, Convergence Promotions LLC. The contents of this publication are the property of the companies whose articles appear within. No portion of this publication may be reproduced in whole or in part without the express permission of the publisher in writing. All product names, descriptions, specifications, etc., are subject to change without notice. The University of California, California ITS, PATH and Convergence Promotions LLC take no responsibility for false or misleading information, errors or omissions. Any comments may be addressed to the Publisher, Glenn ImObersteg, at: firstname.lastname@example.org.
4 10 16 22 28 36
Web Traffic A compilation of the latest news on the web and in the press on connected and autonomous vehicles.
Powering a Path to Autonomous Vehicles TI evaluates the role power integrity plays in advancing high-bandwidth automotive systems like ADAS and AV.
PATH Supporting California DMV in Development of Regulations The Institute of Transportation Studies (ITS) assists in drafting regulations for automated driving systems.
Q100 Automotive Logic from Nexperia Secure and reliable automobile operation--even in the most demanding environments.
Putting Lessons from Mars to Work on Earth
The California Connected Vehicle Test Bed Caltrans and PATH have created an ideal environment for companies to test and develop vehicle applications in a real-world setting.
Impact of Automated Vehicle Technologies on Driving Skills Part 2 of a special in-depth research paper by the Center for Automotive Research and the Michigan DOT on how autonomous vehicles will affect driving behavior.
Taking Autonomous Driving to the Next Levelâ€”but Which Level? This article explores the collaboration between Intel, BMW and MobilEye in developing next generation AV.
The Third Transportation Revolution John Zimmer on Lyftâ€™s vision on transportation for the next ten years and beyond.
Nissan and NASA work together to provide insights on automotive research on earth.
Functional Safety Standards Complex neural networks and sophisticated deep learning techniques are employed to support autonomous driving.
WEB Traffic A Compilation of Connected Auto Articles Found Online
Daimler Advances Connected Car Technology through Open Source and Automotive Grade Linux Automotive Grade Linux (AGL), a collaborative open source project developing a Linux-based, open platform for the connected car, today announced at CES that Daimler is joining The Linux Foundation and Automotive Grade Linux. Daimler is the tenth automaker to join AGL and will actively contribute to developing the Unified Code Base (UCB), AGL’s connected car platform. Automotive Grade Linux is a collaborative open source project that is bringing together automakers, suppliers and technologies to accelerate the development and adoption of a fully open software stack for the connected car. With Linux at its core, AGL is developing an open platform from the ground up that can serve as the de facto industry standard to enable rapid development of new features and technologies. Although initially focused on In-VehicleInfotainment (IVI), AGL is the only organization planning to address all software in the vehicle, including instrument cluster, heads up display, telematics, advanced driver assistance systems (ADAS) and autonomous driving. The AGL platform is available to all, and anyone can participate in its development. Learn more here.
Does Harman’s Connected Car of the Future Foretell the End of Car Ownership? One of the underlying assumptions of Harman’s LIVS (Life-enhancing Intelligent Vehicle Solution) Platform is that we might be at the beginning of the end of the era of car ownership. Almost everything in LIVS is tied to a driver’s unique profile, which drivers and passengers load up when they enter the car. Conceptually, that means it doesn’t matter which car you get into — as long as it’s running Harman’s LIVS, you can log in and grab all of your personalized settings from Harman’s cloud services. Imagine Zipcar, if those cars had tons of connected tech inside — that’s where Harman is headed. Read more by chance Kinney at Chip Chick.
Ford Uses Hackathons to Develop Connected Car Technology Ford has been hosting hackathons in service to its Sync AppLink smartphone application connectivity technology, and they have already paid dividends in the form of experiments with brands such as IBM Watson. The Sync AppLink technology has done wonders for opening up Ford’s vehicles to applications for connected platforms, including mobile payments for fuel, navigation, wearables integration and other capabilities. Many of the varied usages of the Sync AppLink tech emerged as products of partnerships with high-profile brands, such as ExxonMobil, Samsung, DriverScore and Dash Radio.
“Hackathons showcase just how easy it is to use the Sync AppLink platform to integrate apps with our in-vehicle voice commands and displays,” said Doug VanDagens, director of connected service solutions at Ford. “It’s a great way to introduce the Ford Developer Program to app developers and give them access and tools to bring their apps to an entirely new market.” Read more at Mobile Marketer.
Connected Auto The connected vehicle is already a mainstream reality
Connected Vehicle to Everything of Tomorrow (ConVeX) Formed
Next in Connected Cars: Qualcomm Technologies, Inc. is a proven, License Plate Advertising trusted solution provider for automotive
AUDI AG , Ericsson, Qualcomm SWARCO Traffic Systems and the University of Kaiserslautern, announced the formation of Cellular penetration in new light Connected Vehicle to Everything of Tomorrow vehicles sales by 20211 (ConVeX) – a consortium to carry out the first announced Cellular-V2X (C-V2X) trial based Illustration upon the 3rd Generation Partnership Project’s (3GPP) courtesy Release 14, which includes Vehicle-to-Everything (V2X) Qualcomm communication. The trial efforts are expected to focus on Vehicle-to-Vehicle (V2V), Vehicle-to-Infrastructure (V2I) and Vehicle-to-Pedestrian (V2P) direct communication, as well as Vehicle-to-Network (V2N) wide area communications. ConVeX will be executed by a cross-industry consortium that brings diverse expertise to the trial. ConVeX will be cofunded by the participating organizations and the German Federal Ministry of Transportation and Digital Infrastructure (BMVI) within the scope of the “Automated and Connected Driving on Digital Test Fields in Germany” funding guidelines. Read more from Lynn Walford at Auto Connected Car News.
License plates are about to be turned into digital2 screens that have the ability to #1 in telematics display number information Decades of industrytraditional experience when the car is moving but convert to Broad portfolio of technologies digital advertising when the car is parked.
Combination of Strategy Analytics, Jan. ‘16 and LMC Automotive; 2 Qualcomm Technologies, Inc. company data;
340M+ ASICs shipped, serving 20+ OEMs globally3
In a sign that the connected car is going to involve more than just Internet connectivity, a company at the auto show in Detroit this week is introducing a digital license plate.
Includes SoC, Cellular, Bluetooth, Wi-Fi , GNSS and PLC
The rPlate and information platform comes from Reviver in partnership with motor vehicle departments. The smart license plate is an IoT platform that includes DMV registration automation, hyperlocal messaging and vehicle management. Read more from Chuck Martin, Media Post.
Google’s Waymo Invests in LIDAR Technology, Cuts Costs by 90 Percent Waymo is developing hardware and software to make the self-driving car a reality. Waymo, the Alphabet selfdriving car division that was recently spun off from Google, is working on getting that cost as low as possible. According to a recent article from Bloomberg, the company has spent the last 12 months working on “scalability.” The company’s efforts have led to a “90 percent” decrease in the cost of the LIDAR sensor, which is typically the most costly item in a self-driving car solution. On a self-driving car, the LIDAR sensor is a spinning cylinder that usually sits on the roof. By bouncing a laser off an object and measuring the time of flight, LIDAR can tell how far away something is. Thanks to the spinning, these sensors can “see” in 360 degrees. Most self-driving car solutions
use LIDAR as the major sensor, giving the car a “big picture” view of the world so it can see pedestrians and other vehicles. The first public Google Self-Driving Car prototype, built on a Toyota Prius, is a good example of how everything works. The biggest component was the Velodyne HDL-64E LIDAR sensor, which cost a whopping $75,000. The LIDAR sensor needed to be up high to see around the vehicle, so Google mounted it on a large riser. This 360 degree sensing was good for a distance view, but not great at detecting up-close objects, thanks to a dead zone around the LIDAR and obstructions from the car body. To fix this, Google augmented the LIDAR input with several black radar boxes stuck to the front and back of the vehicle. These boxes filled in the blanks for close objects. Read more from Ron Amadeo at ARS Technica.
AT&T Delphi and Ford Join V2X AT&T, Delphi and Ford are developing a new capability to enhance Vehicle-to-Anything (V2X) communications. The platform is designed to help vehicles “talk” with each other and smart cities infrastructure to improve safety and vehicle security, reduce traffic congestion, save money and protect the environment.
Why Uber’s Self-driving Program Failed in San Francisco California’s DMV just shut down Uber’s autonomous vehicle pilot In December, Uber’s semi-autonomous vehicles hit the streets of San Francisco as part of its latest pilot program. A week later, the California Department of Motor Vehicles revoked the registrations of Uber’s 16 vehicles, and the company says it’s taking its program to another city for a public trial. What went wrong? After a fairly seamless, high-profile launch in Pittsburgh, the rollout in San Francisco was bumpy right from the beginning. First, the DMV issued a warning to Uber that it had not obtained the proper testing permits for its pilot program. Then, a few hours after the trial began, The Verge reported that one of Uber’s cars ran a red light, nearly hitting a (human-driven) Lyft car. Uber reviewed the case and determined it was actually the fault of the human driver sitting in the car—remember, Uber still has human drivers who can “take over” from the self-driving system as needed. Then there was the bike lane problem. Uber’s vehicles had a nasty habit of driving into San Francisco’s bike lanes without warning. This was not the fault of humans but a software error, claimed Uber, noting that the problem had not come up in Pittsburgh, which also has a robust cycling network. Uber pledged to fix it. Read more by Alissa Walker, Curbed.
These companies are laying the foundation for the next generation of urban planning and safer driving. In the future, autonomous vehicles will interact with connected traffic lights, roadside monitors, signage, and almost anything surrounding them. The research developed jointly by AT&T, Delphi and Ford can monitor traffic conditions and notify drivers over the AT&T LTE network to approaching vehicles and events. Think events like airbag deployments, vehicle collisions, hazardous road conditions, bad weather and wrong-way driving. The result will be fewer accidents and safer driving. Through the V2X platform, the nationwide AT&T LTE network would extend the range of DSRC communications. The platform uses the network to send notifications and updates for security credential management to each vehicle. Read more at Connected Car News, Source: Byron Jonston, AUTO Connected Car News
NHTSA Notice of Proposed Rulemaking Proposes V2V for All Light Vehicles
Disengagement Rate for Google Cars Outshines Tesla’s
The US Department of Transportation’s (US DOT) National Highway Traffic Safety Administration (NHTSA) has published its Notice of Proposed Rulemaking (NPRM) for public comment by 12 April 2017.
Google’s Waymo unit, which deals with autonomous cars, reported that it had 60 selfdriving cars on California roads last year and that they traveled more than 600,000 miles. It also reported that disengagement—where a driver had to take control—occurred on average once every 5,000 miles, while Tesla reported that its four cars in the state disengaged once every 3.5 miles.
This document proposes to establish a new Federal Motor Vehicle Safety Standard (FMVSS) to require all new light vehicles to be capable of vehicle-to-vehicle (V2V) communications, such that they will send and receive basic safety messages to and from other vehicles. The proposal contains V2V communication performance requirements predicated on the use of on-board dedicated short range radio communication (DSRC) devices to transmit basic safety messages about a vehicle’s speed, heading, brake status and other vehicle information to surrounding vehicles and receive the same information from them. The agency believes that V2V has the potential to revolutionize motor vehicle safety. It aims to create an information environment in which vehicle and device manufacturers can create and implement applications to improve safety, mobility and the environment.
A Google self-driving car at the intersection of Junction Ave. and North Rengstorff Ave. in Mountain View, CA. Author: Grendelkhan, Wikipedia Commons
ITS America Developed Guide to Licensing Dedicated Short Range Communications for Road Side Units
Chrysler, Panasonic Team Up for Portal Concept Car Panasonic has partnered with Chrysler to create the Chrysler Portal concept vehicle. Standout tech features of the vehicle’s cabin include home automation controls, personal listening zones and a display that tracks passenger and driver information in addition to data from outside the vehicle.
On January 19, USDOT FHWA announced vehicleto-infrastructure guidance and published the documents. Among the published documents, the Guide to Licensing Dedicated Short Range Communications for Road Side Units was developed by ITS America. The goal of this document is to make licensing requirements transparent and best practices accessible to any organization, public or private, seeking to deploy “connected vehicle” Dedicated Short Range Communications (DSRC) Roadside Units (RSU) and services that support vehicle-to-infrastructure applications.
ADAS Solutions Past. Present. Future. TI is committed to the development of safety-enhancing applications that enable a safer driving experience for drivers, passengers and pedestrians.
Rear view camera A camera attached to the rear of a vehicle aids in back up, and to alleviate the rear blind spot.
Provides a 360 degree (or bird’s eye) view of the vehicle’s surrounding.
PRESENT Autonomous vehicles An autonomous car (or self-driving car) will be capable of sensing its environment and navigating without human input.
Drive with us: http://www.ti.com/drivewithus
The platform bar is a trademark of Texas Instruments. © 2017
Powering a Path to
Morgan Stanley estimates that self-driving vehicles could deliver $1.3 trillion in annual savings to the U.S. economy, but advancing a reliable autonomous vehicle (AV) will require a significant increase in computing horsepower . As designers advance electronic control modules (ECMs) that support teraflop processing, a deeper understanding of power integrity (PI) will be necessary. Just as signal integrity (SI) addresses interconnect impedance in high-speed digital circuits, PI addresses interconnect impedance of the power distribution network (PDN). When it comes to powering advanced driver assistance systems (ADAS) and AV systems, designers need to consider PDN for the radiated and conducted emissions they create, and interaction with circuit board parasitic impedances that can create unacceptable noise in high-speed digital imaging systems, wireless communication and precision analog circuitry. In this article, I examine the fundamental considerations for designing an automotive, off-battery switch-mode power supply (SMPS) and point-of-load (POL) regulator that considers power integrity. By John Rice, Senior Member Technical Staff, Texas Instruments
Vision for the future Electronic control modules equipped with multi-core, gigahertz processors already crunch a staggering number of lines-ofcode.  This number is expected to at least double as high-bandwidth imaging systems required for next generation ADAS and AV attempt to replace driver fallibility with splitsecond, decision-making algorithms [1,3]. Built on a fusion of imaging sensors including radar and LiDAR, these systems will process somewhere around 1 GB of information every second. “Super chip” processors delivering high processing speed need clean power, and designers advancing this technology will need to pay greater attention to the PDN.
PDN in brief
The goal of the PDN is to create “clean” power for high-bandwidth electronics.
The goal of the PDN is to create “clean” power for high-bandwidth electronics. In his book, “Signal and Power Integrity Simplified,” Eric Bogatin defines PDN as, “all those interconnects from the voltage regulator to the pads on the chip and even the metallization on the die that distributes power and return current.”  This includes the power supply itself, the circuit board, bulk decoupling capacitors, vias, traces, power plane, solder bumps and package bond wires. In short, clean power means that when all these interconnects are considered the composite impedance of the PDN is below a manufacturers specified limit, often referred to as the target impedance. It is not uncommon for this to be 10 mOhms or less from DC to 1 GHz.
The “off-battery” regulator The ECM does not typically run from the car battery, although it may need to be powered from the battery under a “key-off” condition. In key-off, the total quiescent operating current of the ECM is usually specified to be under 100 µA. In run-mode the ECM is powered by the alternator ignition system and that net is anything but clean. In fact, power distribution from the alternator to the ECM is polluted with a host of fast-acting and high-energy voltage and current transients defined by ISO7637.
The automotive ECM voltage typically varies from 9V and 18V, but must also survive highvoltage transients, double battery jumpstarts, and more recently a “warm crank” condition associated with engine “start-stop” operation that can result in a voltage dip below 5V. As such, the off-battery switch-mode power supply (SMPS) must buck, boost, and on occasion buck-boost the ignition voltage. Like the proliferation of electronics driving the Internet of Things (IoT), an AV will certainly result an increase in ECMs to support a robust and reliable artificial intelligence.
To buck, boost or buck-boost Designing an off-battery SMPS for an automotive ECM is complex. A typical ECM converts the ignition/battery input into intermediate voltages of 5V and 3.3V. These voltages are used to power everything from the controller area network bus (CAN) to the instrument cluster, gauge stepper motors and other downstream point-of-load (POL) regulators. Some processor cores are now operating below 1V at 10A, alongside lowvoltage differential system (LVDS) imaging
systems, high-speed DDR memory, microwave RF electronics, and other high bandwidth and precision electronics, all requiring clean power to function reliably. To support the PDN target impedance illustrated in Figure 1, code-hungry processors and FPGAs use integrated FETs, multiphase and multi-output power management ICs (PMICs) operating above 2 MHz. Even the off-battery regulators are starting to use 2 MHz converters to avoid AM-band interference, but this approach has its own challenges. Electromagnetic interference (EMI) is a complex topic as are the methods for eliminating it, but the best approach has always been to eliminate EMI at its source. Devices like the LM5140 are designed for off-battery, high-frequency operation using innovative gate driver circuitry to spread the EMI noise spectrum and to minimize switch-node dv/dt that, otherwise, would activate printed circuit board (PCB) parasitic impedances and cause EMI.
Figure 1. Typical FPGA/ processor interconnect impedance requirement
Integrated MOSFET technology with devices like the LM53635-Q1 use innovative package technology to eliminate parasitic switch-node ringing that otherwise can disrupt the reference plane and radiate noise. Figure 2 illustrates the improvements between conventional wirebonded, integrated FET regulator and this device. If the “battery” input was always greater than 6V, then a buck topology could be used for both the 5V and 3.3V rails, but as previously mentioned, the off-battery SMPS may be required to generate 5V when the input is below 5V. Cars with start-stop capability often include a centralized voltage stabilization module (VSM) to address this condition. This boost converter ensures that critical nets are not affected by a warm-crank condition, and are designed to deliver between 200 W and 1400 W for short periods of time, typically less than 100 ms. Although these systems are generally not thermally challenged, peak electrical and thermal stresses can be very high and must be carefully analyzed. A robust and efficient approach is to use a multi-phase, synchronous boost converter. The LM5122 was specifically designed to support this topology—this controller has high current integrated FET drivers, can operate down to 3V and can be
interleaved with other devices for a scalable solution. The device circuit behavior can be analyzed on the TI WEBENCH™. Independent voltage stabilization at the ECM can be implemented using a pre-boost architecture in front of a buck converter. Boost controllers like the 60V, 2 MHz LM5022 can be configured as either a boost or SEPIC buck-boost converter. Whereas the pre-boost architecture stabilizes the input on an “as needed bases”, the ECM would still require a downstream buck converter. Configured as a SEPIC, the downstream buck converter is eliminated. However, SEPIC converters typically have a lower control loop bandwidth. This is necessary to address its complex power stage transfer function. Consequently, SEPIC converters are not well-suited for processor POL regulation, but they do work well in creating a voltage-stabilized intermediate voltage.
Keep the electrons moving forward One of the many famous quotes by Albert Einstein goes something like this, “Every design should be as simple as possible, but no simpler.” Unfortunately, determining when good is good enough is not always easy; we keep pressing forward to make things “better.” When it comes
Figure 2. Switch node of the LM53635 (left) compared with a conventional wire-bonded regulator
to reverse battery protection, the PN junction diode is as simple as it gets. It is low cost, only has two terminals, is available from many suppliers in many packages, can stand off very high negative voltages, and fails in a predictable way. But the reverse battery blocking diode has at least one significant deficiency—its forward voltage drop. As ECM current increases, that deficiency becomes increasingly problematic.
off within 2 μs. Because the device floats on the supply, it requires no quiescent ground current. One limitation of the device is that although it reduces the forward voltage drop 99 percent of the time, it does not fully address dropout concerns since the MOSFET body diode is conducting one percent of the time with a higher forward drop.
To address this issue, ECM designers have used MOSFETs with the body diode blocking reverse current in the off-state. In the on-state, the MOSFET channel is enhanced to minimize the forward voltage drop. That works fairly well, but it has issues. For starters, getting a MOSFET that can stand off—400V ISO pulses is not cheap, so a transient voltage suppressor (TVS) is generally needed. Also, the quiescent current associated with fast-acting gate control can be prohibitive. To overcome these deficiencies, the LM74610-Q1 “smart diode” was developed. This innovative three-terminal device, shown in Figure 3, uses the body diode of the MOSFET as its power source boosting that voltage to keep the MOSFET gate on 99 percent of the time. The remaining 1 percent of the time, the IC refreshes the charge pump capacitor that keeps the MOSFET in saturation. When a reverse condition is detected, the IC actively turns the MOSFET
Power integrity will become increasingly important in advancing high bandwidth automotive systems like ADAS and AV. Whether you are advancing an “off-battery” regulator or lower voltage point-of-load regulator, the importance of designing for power integrity in high bandwidth applications cannot be overstated. Two excellent book references are provided that explain the subtleties of designing for, and measuring power integrity and point-of-load PDN [4,5].
“Every design should be as simple as possible, but no simpler.” — Albert Einstein
References 1. Intel, Self-Driving Car Technology and Computing Requirements http://www.intel.com/content/www/ us/en/automotive/driving-safetyadvanced-driver-assistance-systemsself-driving-technology-paper.html 2. Charette, Robert, N. This Car Runs on Code, IEEE Spectrum. http://spectrum.ieee.org/transportation/ systems/this-car-runs-on-code 3. Estl, Hannes. “Paving the way to self-driving cars with advanced driver assistance systems,“ Texas Instruments white paper, August 2015 4. Bogatin, Eric. “Signal and Power Integrity Simplified”, 2nd edition. ISBN13: 978-0132349796, Prentice Hall 5. Sandler, Steve. “Power Integrity” ISBN- 978-0-07-183099-7, McGraw Hill
Figure 3. “Smart diode” configured to minimize ECM power losses
PATH Supporting California DMV
in Development of Regulations for Automated Driving Systems By Steven E. Shladover, PATH
CALIFORNIA was one of the first states to pass legislation requiring the development of regulations to govern the testing and public operation of automated driving systems (systems that can take over enough of the dynamic driving task that they do not require the continuous supervision of a human operator). That legislation, SB 1298, became law in September 2012 following a lobbying effort by Google, whose lobbyists drafted the original version of the bill based on their preference for regulatory certainty over uncertainty. The legislation gave the responsibility for development of the regulations to the California Department of Motor Vehicles (DMV), and included a provision encouraging them to seek the technical support of the U.C. Berkeley Institute of Transportation Studies. This technical support work was done under a contract with PATH from July 2013 until the end of 2016.
The DMV developed the regulations for testing of automated driving systems on public roads by themselves, with a minimum of assistance from PATH (other than providing intermittent support to their work responding to questions about technical aspects of the testing regulations). The PATH efforts were focused on helping the DMV understand the technological and vehicle industry implications of different approaches to regulating the public operation of automated driving systems. Much of this work involved reviewing analogous regulations in other domains (aviation, air pollution, etc.) and in other countries for their potential applicability here. This is challenging because of the lack of specific technical standards or testing procedures that can be cited and because of the complexity and immaturity of the technology involved. We could not identify precedents for development of detailed safety regulations on a new technology at such an early stage in its development. The central challenge in developing regulations for automated driving systems is balancing the need to protect the general public from unreasonable hazards posed by new systems that have not been well engineered, while also encouraging the development and implementation of systems that can be demonstrated to improve traffic safety. After the minor technicalities are swept away, the central question that remains revolves around how to demonstrate that a new system is indeed safe enough that it should be permitted to share the public road space with other vehicles and vulnerable road users.
In our work to help DMV resolve this question, PATH focused on the concepts of functional safety and driving behavioral competency. Functional safety describes the process that the designer of a safety-critical system follows to ensure that the system avoids injuring or killing people even after it experiences failures and anomalous conditions of various kinds. The ISO 26262 international standard defines the procedures for applying functional safety to the development of safety-critical automotive systems. Driving behavioral competency is comprised of a set of driving behaviors that a system must be able to execute to demonstrate that it is minimally competent. This could be considered as the counterpart to the driver licensing exam that each novice driver needs to pass before he or she can be licensed to drive. It does not “prove” safety, but it at least shows that a system is not completely incompetent. The DMV defined three basic “areas of operation” to represent the different environments in which automated driving systems could be expected to operate – urban, rural and freeway, and PATH then specified a minimal set of 18 types of maneuvers that vehicles would need to execute to drive in these environments as the initial behavioral competency requirements that a manufacturer would have to meet in testing done by a third-party testing organization. These served as the starting point for an independent peer review by outside experts that the State of California requested before incorporating the behavioral competency concept into its proposed regulations.
We identified a list of relevant experts from industry, government agencies, interest groups and the research community and solicited their detailed review of the proposed set of 18 broad behavioral competencies. Fortunately, most of the invited experts responded favorably to the invitation to do the peer review, and provided their comments in writing, by conference call, in face-to-face meetings and in a by-invitation workshop. In the end, comments were received from 76 experts representing 41 organizations (9 established automotive OEMs, 8 other vehicle manufacturers, 4 Tier One automotive suppliers, 4 testing organizations, 10 research organizations and 5 interest groups). The peer review comments provided very valuable inputs regarding the basic approach to AV regulation as well as the specifics of the behavioral competencies. In most cases, there was a broad consensus on the appropriate direction to follow, with opinions divided on only a limited set of issues. The principal findings from the peer review were: • The reviewers identified some ambiguities in the draft regulatory language that had to be fixed, particularly in cases where there appeared to be internal contradictions or contradictions with the California Vehicle Code language specified by the Legislature. • It is necessary to be precise about the level of severity of the failures that are required to lead to a minimum risk maneuver or a safe stop response by the vehicle. • The regulations need to include provisions for remote supervision of automated vehicles by operators at a control center (especially for those vehicles that can operate without a driver onboard).
• The regulations need to provide for flexibility regarding adherence to provisions of the Vehicle Code, especially when traffic safety or traffic flow considerations require small violations of the code to avoid deadlock conflicts in practical traffic scenarios. • Behavioral competency and functional safety cannot be separated entirely from each other, but they need to be treated in an integrated fashion in specifying requirements on automated vehicle design and operations. • Satisfaction of the behavioral competency requirements only demonstrates a bare minimum of driving competency (to screen out the utterly incompetent), but it does not provide any assurance of safety. • The reviewers were virtually unanimous in disliking the concept of “Areas of Operation” (urban, rural or freeway) that was embedded in the original classification of behavioral competency requirements. They found this inapplicable and overly simplistic, leading to a strong recommendation that the behavioral competency requirements and test conditions be tied directly to the Operational Design Domain (ODD) of each specific AV system. The ODD is the specific combination of conditions in which the individual automated driving system has been designed to operate, based on the capabilities of its sensors and the sophistication of its software, as well as based on the condition of the roadway infrastructure and the extent to which it has been digitally mapped by the system developer.
The central challenge in developing regulations for automated driving systems is balancing the need to protect the general public from unreasonable hazards posed by new systems that have not been well engineered, while also encouraging the development and implementation of systems that can be demonstrated to improve traffic safety.
• The significance of the ODD was further enhanced with the strong recommendation that a key functional safety requirement for
any AV system should be that it precludes usage of any automation function or feature outside of its ODD. This means that if the vehicle moves outside the geographic boundaries of its ODD or if the weather changes to a condition that is not within the system’s ODD, the system must disengage its automation function.
The federal policy was released on September 20, 2016, followed closely by the updated draft California state policy, which is well harmonized with the federal policy.
• The regulations should define a three-stage approval process: 1. Functional safety process analysis 2. Closed-track testing of potentially hazardous conditions or rarelyencountered scenarios that need to be staged to demonstrate some behavioral competencies 3. Public-road testing of the remaining benign and commonly-encountered scenarios that are needed to show the remaining behavioral competencies. There is a need to protect against inconsistencies in the way third-party testers will implement the behavioral competency testing in order to ensure fairness, particularly in the absence of explicitly specified pass/ fail criteria. This could be done by having third-party organizations pre-qualified by the DMV or having the DMV hire one organization to do all of the testing.
• A few areas remained subject to some divisions of opinion among the reviewers: • How challenging should the minimum behavioral competency test requirements be? Should they represent novice-level or highly experienced driving performance? • Should the requirements include handling challenging “corner cases”? • What level of change to an AV system should require a re-certification? • Should behavioral competency testing be done by a third-party organization or should this be a self-certification by the system developer? After PATH delivered its report on the results of the peer review to the DMV in March 2016, the DMV worked closely with the American Association of Motor Vehicle Administrators (AAMVA) and NHTSA on the development of the model state policy section of the Federal Automated Vehicles Policy. The federal policy was released on September 20, 2016, followed closely by the updated draft California state policy, which is well harmonized with the federal policy. Many of the findings from our peer review were incorporated into the federal policy, including the emphasis on behavioral competency testing and use of the ODD as the framework for determining the requirements that a system needs to meet.
Automotive reference application for Linux and QNX Location Studio delivers a robust platform for automotive products, providing extensive support for embedded navigation applications and the evolving 'Connected Car' environment. Location Studio provides a superior in-car navigation solution, which includes automatic map updates, real-time traffic, and integrated content such as real-time gas prices, Doppler weather radar overlay, movie theater locations, showtimes, ratings and reviews. It also captures analytics data that enables automotive manufacturers to improve the driver experience and develop next generation vehicles. With a carrier grade platform and over 10 years of connected navigation solutions experience, Location Studio has the technology and experience to provide a superior connected car solution today. ÂŠ 2016 Comtech Telecommunications Corp. All Rights Reserved
Learn more! Visit www.location.studio
Q100 Logic By Amita Malakar, Nexperia
Secure and Reliable Automobile Operation— Even in the Most Demanding Environments The operating environment of automobile semiconductor components is much more hostile than that of semiconductors used in home or portable applications. A television set will generally spend its operating lifetime within an ambient temperature range of 0˚C to 40˚C. Due to internal heating, its semiconductor devices can be expected to operate between 20˚C and 60˚C. By comparison, an automobile is expected to start at temperatures lower than -20˚C and, in some cases, operate within the engine compartment at temperatures approaching 150˚C.
While an inoperable television is a minor inconvenience, a connected or autonomous vehicle traveling at top speeds on the highway needs to be able to operate safely— even in the harshest environments. To ensure the reliability of automotive electronics, the Automotive Electronics Council introduced its AEC-Q100 standard, which outlines procedures to be followed to ensure integrated circuits meet the quality and reliability levels required by automotive applications. As the global number one supplier, the introduction of its Q100 logic portfolio shows Nexperia continuing to lead the way in automotive logic. Nexperia offers the feature rich Low Voltage CMOS (LVC) logic portfolio to enable the migration of electronic solutions from 5.5 V to lower power mixed 5.5 V / 3.3 V and beyond. The LVC family includes Standard Logic functions with supply range 1.65 V to 3.3 V, as well as Mini Logic functions with supply range 1.65 V to 5.5 V. Operating at elevated temperatures reduces the lifetime of a semiconductor and temperature cycling has a negative impact on the stability of a package. In cases where there is no history of a product’s reliability within automotive applications, a series of stresses to simulate the life cycle within an automotive environment must be applied to guarantee conformance to the AEC-Q100 standard.
Q100 devices are: • manufactured in TS16949-certified and VDA-approved production facilities • flagged as automotive lots • subjected to additional process flow quality gates and stricter rules for lot dis-positioning and maverick lot handling This ensures that automotive products: • receive the highest priority • have greater traceability for improved quality analysis • that become outlier lots, passing a quality gate but outside of the acceptable distribution, are assigned to the non-Q100 type Six sigma design philosophy is applied to all Q100 devices. This ensures that an end user application designed to the datasheet limits can tolerate a shift as high as one and a half sigma in Nexperia’s manufacturing processes. As the process control limits are much tighter than one and a half sigma, this virtually guarantees trouble free end user applications. During electrical test process, average test limits or statistical test limits are applied to screen outliers within automotive lots. Figure 1 shows the distribution of devices passing a test and the calculated statistical test limits in yellow. Although the outliers are within the upper and lower specification limits they are not delivered as Q100 products. To ensure continued reliability, Nexperia logic maintains an extensive reliability monitoring program—the results of which are published half yearly. Nexperia’s first and second tier technical support teams give Q100 product design-in assistance their highest priority and upon request AEC-Q100 production part approval process (PPAP) qualification data will be made available. Due to the stricter qualification requirements of automotive end user applications, a 180-day process change notification (PCN)
Figure 1. Application of statistical test limit
There is a new name in Discretes, Logic and MOSFETs.
Nexperia is a dedicated global leader in Discretes, Logic and MOSFETs devices. Originally part of Philips and more recently NXP, we became independent at the beginning of 2017. Focused on efficiency, Nexperia produces consistently reliable semiconductor components at high volume: 85 billion annually. Our extensive portfolio meets the stringent standards set by the Automotive industry. And industry-leading small packages, produced in our own manufacturing facilities, combine power and thermal efficiency with best-in-class quality levels. Built on over half a century of expertise, Nexperia has 11,000 employees across Asia, Europe and the U.S. supporting customers globally. Introducing: Nexperia, the Efficiency Company.
To ensure continued reliability, Nexperia logic maintains an extensive reliability monitoring program—the results of which are published half yearly. approval cycle is applied for Q100 products instead of the 90-day PCN approval cycle for standard types. In the unlikely event of a quality issue, Nexperia logic guarantees a 10 day throughput time with initial verification within 24 hours for its Q100 portfolio. There are many examples of Nexperia Q100 logic automotive application areas. The Q100 can be applied in I/O expansion, interface logic, control logic, and display drivers. Control applications such as engine control units and body control modules change settings based upon a combination of input signals. Control logic consists of simple Boolean functions, such as AND or NAND, to facilitate changing settings in simple subsystems that don’t require a microcontroller. Large pin count controllers are expensive, so when possible to reduce the complexity and pin-count of control solutions, input/output expansion devices such as multiplexer/demultiplexer devices are used. Figure 2 shows an example of an 8:1 multiplexer used to sequentially switch analog sensor signals to a single analog to digital pin of a microcontroller. With high impedance inputs and low impedance outputs, interface logic such as registered or unregistered buffers and line drivers are used to interface between low drive outputs of a
Figure 2. 74HC4851 as a multiplexer in a remote sensing application
controller and higher loads of, for example, water pumps and window motors. Control applications such as engine control units and body control modules change settings based upon a combination of input signals. Control logic consists of simple Boolean functions, such as AND or NAND, to facilitate changing settings in simple subsystems that don’t require a microcontroller. Display drivers integrate serial-in, parallelout shift registers, which are common I/O expansion devices, with a number of MOSFET LED drivers. With 8-bit and 12-bit solutions, shift register based display drivers enable a controller to drive 8 or 12 LED’s using 3 output lines. Cascading devices as shown in figure 3 increases the number of LED’s controlled by the same 3 output lines. Display drivers reduce the size, complexity, pin count and ultimately cost of any microcontroller based solution. A summary of Nexperia logic’s Q100 portfolio including a search by function and a parametric search within each function can be found at www.nexperia.com, and unlike the standard types, each Q100 device has a dedicated datasheet confirming that it has been qualified in accordance with AEC-Q100 and is suitable for automotive applications.
Figure 3. NPIC6C596A in cascaded display driver application
Lessons from Mars to Work on Earth–
Insights on Autonomous Vehicle Research
By Nissan’s Dr. Maarten Sierhuis Shown: Nissan IDS Concept
he Nissan Research Center (NRC) attracts global talent as a center for open innovation by sharing its appealing qualities and providing information to the public. In the essay below, the NRC presents insights on autonomous vehicle research and artificial intelligence from Dr. Maarten Sierhuis, the director of the NRC in Silicon Valley, who is using his experience at NASA to work on autonomous drive technology for Nissan. Sierhuis came to the U.S. in 1989 from the Netherlands and worked for IBM and NYNEX Science and Technology until 1988. After earning his PhD in artificial intelligence at the University of Amsterdam, he worked for NASA and Xerox PARC. He spent 12 years with NASA where he developed a computer language underpinning intelligent systems for use in robots, spacesuits and NASA’s Mission Control Center. In 2013, he joined Nissan where he heads the Nissan Research Center in Silicon Valley and leads multiple teams of researchers working on autonomous vehicles, connected vehicles and human-machine interaction and interfaces.
A Career That Began in Space
Nissan’s Dr. Maarten Sierhuis
To accelerate the time it will take for autonomous vehicles to get on the road safely, at CES 2017 Carlos Ghosn announced a breakthrough technology called “Seamless Autonomous Mobility,” or SAM.
I started working at NASA Ames Research Center in Silicon Valley in 1998. After 12 years there, I went to Xerox PARC, where I served as director and ran research on multi-agent systems and human-machine interaction. It was at NASA, though, that I created much of what I’m putting to work for Nissan today. We started with the development of a simulation language that allowed us to model human behavior and multiple people working together. We were looking at how people might live on Mars and work with people back on Earth, as well as autonomous systems, including robots and smart habitats on Mars. We started first simulating this with our language, but once we started running this language in real time, it also became a programming language for autonomous systems in general. We built intelligent agents for robots, for Mission Control, and for the habitat. Then we added a speech-dialogue system to this—now astronauts could talk to their autonomous systems, including the robot and the habitat, as well as to systems in Mission Control. We put it to work in space; we actually built an intelligent agent in the space suit to monitor the astronaut’s health autonomously.
We helped design how robots would work on Mars with people on Earth. After this success, we were asked to automate flight controllers in NASA’s Mission Control Center for the International Space Station. The final project I worked on at NASA used my computer language to automate a flight controller for the ISS. This system went live in 2008 and it’s still in use, with all communication from the ISS taking place through it.
The Connection with Cars Building the autonomous system for a vehicle on Earth is really like building a robot that drives 80 miles an hour very close to other robots. That is very different from Mars, where
An artist’s concept portrays a NASA Mars Exploration Rover on the surface of Mars. Rovers Opportunity and Spirit were launched a few weeks apart in 2003 and landed in January 2004 at two sites on Mars. Each rover was built with the mobility and toolkit to function as a robotic geologist. Source: NASA/JPL/Cornell University, Maas Digital LLC (WikipediaCommons.com) NASA/JPL/Cornell University, Maas Digital LLC
Powered by Convergence Promotions
RoboDrive motors from TQ: When precision steering counts The RoboDrive technology provides the highest power density at maximum torque range and overload capability in a compact design. ■
RoboDrive motors are available in a variety of sizes, from small (25mm) to large (115mm) and cover a torque range from .025Nm to 11Nm
Order as rotor and stator kits for design-ins, or as geared motors
Select from solid or hollow shaft models
there are not that many people – at least not yet! Many issues come up when you think about humans interacting with each other and with robots, because the car needs to be on the road with pedestrians, bicyclists, and other cars. The idea of multi-agent modeling becomes key: knowing what everybody is doing, so that the car knows not only what it needs to do itself, but also its relationship with others on the road. In urban areas, we have to deal with pedestrians, bicyclists, motorcyclists, cars, animals—the whole spectrum of interaction becomes a very important study. The work that I did at NASA is very relevant in this context. In connection with these driving environment issues, I started a research project in North Holland, a Dutch province with a world-leading traffic management system. Every traffic light there is connected with all the rest, and they communicate dynamically based on how busy the roads are to decide when they will change. We used data from these traffic lights in intersections and built machine-learning algorithms to predict when a traffic light will go red or green, based on how far one is from the light.
We now use this algorithm with our autonomous vehicle software, so as the vehicle drives it gets information from the traffic system to predict how long the light will be red or green. Ideally, the autonomous vehicle doesn’t have to stop; it can automatically reroute and take lights that are green. As we optimize the autonomous system to avoid stopping for lights, we also send information from the vehicle back to the system so as to optimize traffic management on a larger scale.
Why Silicon Valley? In autonomous vehicles now, most of the technology is based on software and on artificial intelligence (AI). There’s no better place to do this work than in Silicon Valley. The NASA Ames Research Center was one of the first places where robots and autonomous systems were put together, and all the technology around autonomous vehicles was developed at universities and companies in the area. In the early 2000s, many AI researchers from around the world came to Silicon Valley to join the IT industry. Nissan realized this in the mid-2000s: If we wanted to be serious about building autonomous vehicle technology in house, we had to have a presence in Silicon Valley.
In urban areas, we have to deal with pedestrians, bicyclists, motorcyclists, cars, animals—the whole spectrum of interaction becomes a very important study.
SPONSORS May 15 â€“ 18, 2017 Santa Clara Convention Center, CA, USA
EDUCATE CONSUMERS, IMPROVE PUBLIC SAFETY, AND EXPLORE NEW BUSINESS MODELS FOR IMPLEMENTING TRANSPORTATION AND MOBILITY TECHNOLOGY OF THE FUTURE PRODUCT, STRATEGY, AND TECHNICAL TALKS
Get the latest insights from 100+ thought-leading speakers from inside and outside the automotive industry on topics like ADAS, fleet management, LIDAR & Sensor Technology, and much more.
NETWORK, NETWORK, NETWORK
Meet with the movers and shakers driving connected cars & autonomous vehicles forward through speed networking, hosted lunch table discussions, 1-to-1 scheduled meetings through our VIP team, and our evening networking party.
CCTA AUTONOMOUS TEST DAY
Take a day trip to the GoMentum Station Station, the nation's largest secure testing facility for autonomous and connected vehicle technology.
CONNECTED & AUTONOMOUS VEHICLES EXPO Co-located with the Internet of Things World, meet with 250+ exhibitors showcasing the latest technology & services to help you transform the industry.
#CONNECTEDCARS TO LEARN MORE GO TO TMT.KNECT365.COM/CONNECTED-VEHICLES
100+ SPEAKERS Co-located with:
The Nissan Research Center has relationships with people from Stanford and UC Berkeley —which provide a key talent pool for us, too— and we’ve set up very close, fruitful research collaboration with NASA, just down the road from us. We’re speaking with a number of Silicon Valley firms to see how we can work together. So it’s vital to be in this region.
Nissan Intelligent Mobility
Looking Towards the Future of Mobility
On January 5, 2017 during his CES keynote, Nissan Chairman of the Board and Chief Executive Officer Carlos Ghosn announced several advancements as part of the Nissan Intelligent Mobility blueprint for transforming how cars are driven, powered, and integrated into wider society.
At NASA, I researched how humans and robots would work together on Mars in the future. When Nissan asked me to do that for vehicles on Earth, it was very difficult to say no. I have a particular view on how humans and autonomous systems should work together, and I really appreciated that people at Nissan had a similar idea. Nissan believes that mobility is for the good of society; that’s one of the reasons I decided to come here. It’s exciting to think about a society where the right mobility system is used for the right purpose at the right time. We’re going to have trains interacting with shared vehicles that can seamlessly take me from work to home to school to pick up my children. Whatever I need to do in my life will be seamlessly integrated with mobility services at different places. I don’t believe that public transportation should go away. By integrating our vehicles —our trains, our planes, even our bicycles together into a society where we have more space for parks and beautiful landscapes —we’ll have more space for people. Nissan’s vision of virtually zero fatalities and zero emissions is really a great motivator for doing research to move in that direction. Autonomous technology can be applied not only inside the vehicle, but also in the cloud, in trains and in other transportation systems. We should always be efficient in the way we move around and interact with others. (Based on an interview carried out in September 2016.)
The world is facing serious challenges such as climate change, traffic congestion, road fatalities, and increasing air pollution. Nissan is committed to addressing these challenges by making transportation safer, smarter, and more enjoyable, with the ultimate goal of achieving zero-emissions and zero-fatalities on the road. Nissan Intelligent Mobility is the roadmap. Nissan Intelligent Mobility encompasses three core areas of innovation: Nissan Intelligent Driving helps give more confidence through enhanced safety, control, and comfort to everyone on board. The building blocks for autonomous driving are already built into Nissan cars with an extensive set of advanced safety features including Intelligent Around-View Monitor and Intelligent Lane Intervention. Autonomous drive technologies can already be found in certain Nissan vehicles today, including the Nissan Serena, the first model produced for the mass market to feature ProPILOT technology. Nissan has plans to extend this technology to more models in Europe, Japan, China, and the United States, with 10 models to be launched by 2020 by the Renault-Nissan Alliance. Nissan Intelligent Power makes driving more exciting for customers by continuing to reduce emissions and increase fuel economy. Nissan is committed to a holistic approach to achieving zero-emission mobility by making internal combustion engines more efficient and by putting more advanced technologies into their electric vehicles. They continue to
Nissan Intelligent Mobility is not about removing humans from the driving experience. Instead, it’s about building a better future for Nissan customers where cars are their partners, and where drivers are more confident and more connected.
advance a variety of powertrain technologies under Nissan Intelligent Power, which are most suitable to the different market segments and different regions across the world. They have a diverse range of EV-based technologies in their portfolio in addition to 100% electric vehicles, these technologies include e-Power (serieshybrid) and fuel cell electric vehicles. Each new technology supplements the portfolio, but does not supplant other technologies. Nissan Inelligent Integration keeps customers more connected by conveniently linking Nissan cars to the wider society. Nissan is helping to shape a sustainable ecosystem enabling cars to interact with people, other cars and road infrastructure. This approach will eventually lead to remote vehicle operation,
reduced traffic jams, more efficient carsharing, and improved energy management. Nissan Intelligent Mobility is not about removing humans from the driving experience. Instead, it’s about building a better future for Nissan customers where cars are their partners, and where drivers are more confident and more connected. “We invite others to join us, as well, from tech partners to e-commerce companies, ridehailing and car-sharing platforms, and social entrepreneurs who can help us to test and develop new vehicles and services, and make sure everyone has access to the latest technologies and services that bring value to their lives,” said Ghosn.
in the Machine Learning Era
By Robert Bates Chief Safety Officer, Embedded Software Division Mentor Graphics Corporation
ince its definition and establishment in 2011, the ISO 26262 international functional safety standard has rapidly emerged as the definitive guideline for automotive engineers looking to optimize the safety and reliability of electrical and/or electronic (E/E) automotive systems. The standardâ€™s adoption was initially spurred by the ongoing trend toward replacement of mechanical systems with electronic-based approaches in production automobiles.
More recently, the standardâ€™s footprint has further expanded, fueled by the era of smart connected vehicles, which has led to the incorporation of sophisticated advanced driver assistance systems (ADAS) in even the most austere of late-model economy class vehicles.1 Mid-range and luxury models have gone even further, increasingly employing complex neural networks and sophisticated deep learning techniques to support levels of autonomous driving.
ISO 26262 implies strict adherence to the automotive “V model” and its direct link between functional safety concepts and the specific requirements for how these concepts should be fulfilled. The shift toward machine learning for critical driving decisions would appear to break this link, thereby undermining ISO 26262’s assurance of safety. So, does the decoupling of safety objectives and their execution render ISO 26262 irrelevant moving forward? Mentor Automotive believes this is a conflict which can be solved, but like many other aspects of autonomous vehicles, the problem doesn’t fit neatly in the ISO 26262 framework, and there is no standardization for how this should be considered. Before we arrive at our answer and viable approach, let’s explore the link between ISO 26262 objectives and the kinds of complex implementations inherent in machine learning.
For example, an anti-lock braking system might have a technical safety requirement that states “prevent wheel lock to the rear wheels when the braking system is applied.”
ISO 26262 is a derivative of the more general IEC 61508 functional safety standard and focuses on the electrical and electronic (E/E) systems in road vehicles. It is applied throughout the design, development, and manufacturing cycles, as well as managing the relationships between an automotive company and their suppliers. ISO 26262 is targeted at vehicles up to 3500 Kg in weight and seeks to minimize the potential hazards caused by a malfunction in, or failure of the embedded electronic or electrical system. ISO 26262 requirements fall into two basic categories: those which refer to a high-quality developmental practice, and those which apply specifically to the management of the safety requirements.
High-Quality Developmental Practices Tied to Safety Requirement Management ISO 26262 provides guidelines for highquality design and development of hardware and software that are similar to guidelines specified in other safety standards for general electronic devices, medical software, and others. Following these guidelines ensures
that the resulting software is high-quality, although, to this point, it does not explicitly address how the requirements, designs, etc. actually fulfill the original safety concept. An example of the guidelines for high quality design and development of software might include ensuring that the testing of the software both verifies the original requirements and verifies every pertinent line of code. Another guideline could include creating a quality management system that documents how development will be managed and performed, and that it’s done in such a way that the results can be audited for conformance. As previously mentioned, ISO 26262 implies strict adherence to the automotive V model, where there is a direct mapping from the functional safety concept to the requirements to how they are fulfilled. In the traditional world of automotive functional safety, this is not a problem. For example, an anti-lock braking system might have a technical safety requirement that states “prevent wheel lock to the rear wheels when the braking system is applied.” As the system is designed, this will break into several decomposed requirements; one of which might be “When the wheel speed sensor reports a rapid decline in the wheel rotational speed, and if the braking system is applied, then command the breaking system to rapidly pump the brakes”, with the meaning of terms like “rapid decline” and “brake pumping frequency” to be determined by the technical safety concept. Much of the logic for managing this will be performed by the software in the Electronic Control Unit (ECU) for the braking system, which will inherit the ASIL level from that determined from the safety concept (almost certainly ASIL D). In this example, mapping the item’s technical safety requirements to the functional safety requirements is fairly straightforward. It is in these cases that the assumptions of ISO 26262 are fully realized; the safety requirements become realized as functional requirements for the item, so it’s straightforward to verify that these requirements were fulfilled.
So Where is the Problem? Consider a different example involving adaptive cruise control, which operates similarly to standard cruise control, but moderates the velocity to keep a safe distance to the car in front. In more modern systems, this might be tied into several driver-assist features, including lane monitoring and GPS assistance. So if the car in front is slowing down because it is about to take an off-ramp, the adaptive system performs differently than in the more standard case. Again, the technical safety requirements can be fairly well stated: the automobile should not impact the car in front of it as a result of using the adaptive cruise control feature. In these more complex systems, this might be broadened to also consider the effect of the operation of the system on cars behind and to the sides of the one using the adaptive system. This becomes a much more complex problem to solve from a technical perspective, and most all of the processing will be done in software.
Neural Networks and Autonomous Driving The example given above is actually a fairly simple subset of the functionality that must be available for autonomous driving. Autonomous driving still has an array of complex problems that are difficult, if not impossible, to solve using conventional embedded software development techniques. Fortunately, there are other ways to solve the problem. Predominately, the solution for these complex problems involves the use of neural networks or other machine learning
techniques. In essence, neural networks are taught the correct solutions to various, specific kinds of problems. Using neural networks to learn tasks has made it possible to automate many functions across all aspects of our lives; from handwriting recognition used by post offices and shipping companies, to industrial quality control systems that reject inferior goods, to other applications that impact our lives in numerous ways. As mentioned above, the network would first be taught to slow the car down to match the speed of the car in front of it. The network might then be taught that if the car in front occupies the far right lane as a highway exit approaches, and if that carâ€™s turn signal is on, then the autonomous vehicle should slow less aggressively under the assumption that the car in front might be taking the coming off-ramp. It is trained under many possible variables and under many different conditions. Once the software has achieved a basic level of training, it is then taken onto the road in closely supervised situations; on a test track, and then with a human driver ready to take over whenever something goes wrong. In short, it is trained in a way similar to the way we were all trained to drive; by learning situations, and then with practice with a competent driver along to support.
The software to support the neural network is complex, but is specifically designed to enable this kind of learning.
The software to support the neural network is complex, but is specifically designed to enable this kind of learning. The key point
for the purposes of this discussion is that implementation of the neural network software will usually show few signs about the task that it’s learning about. Instead, it understands the sensor inputs that it will receive (sensor data, visual data, etc.), and then learns based on this input what the appropriate outputs should be (actuator data, output signals that alarm the driver, annotated mapping data, etc.). The neural network software focuses on the learning and the application of the knowledge, not on the knowledge itself.
The Problem with Neural Networks
The problem boils down to a verification problem: how do you know the original safety requirements were met, and how safe is safe?
As previously described, ISO 26262 assumes that there is a clear line from the technical safety requirement to the implementation and verification of the item. In a system based on neural networks, this linkage is broken at the implementation level; in the case of the simple example given previously, the safety requirement might be “If the vehicle in front decelerates and is sufficiently close to the vehicle in question, decelerate and avoid the collision.” It would not be to “write a neural network and train it so that the collision is avoided”. This problem is not insurmountable. Otherwise, the future of autonomous driving will never be considered safe. However, no standardization exists regarding how this should be considered, nor does it fit neatly into the ISO 26262 framework. Part of this is simple. The software quality aspects outlined in ISO 26262 apply just as much to this kind of software as it does to any kind of software, and thereafter this software can be held to the same standards as automotive software is today. Understanding how the neural network software learns (i.e. understanding the requirements and design of the system), having the implementation be high quality (thoroughly reviewed including MISRA compliance), etc. is just as, or even more important for these systems as it is for the software controlling safety systems today.
The problem boils down to a verification problem: how do you know the original safety requirements were met, and how safe is safe? Obviously, these questions are tied. It’s possible to road test the vehicle under many conditions, and it’s possible to test the vehicles with more and more vehicles under different driving conditions; this kind of testing is common today. Automotive companies are creating increasingly larger pilot programs to show that these software systems are functioning as they expect and to rapidly deploy fixes as problems are found.
How Safe is Safe? With all of the testing that is already underway, a lot of data is being collected about how these autonomous systems behave in realworld conditions. However, how do we know that there are no latent faults that might cause failures in certain situations, even ones similar to ones already encountered? A well-reported case happened in mid-2016, where a driver using a prominent OEM’s autopilot system died when the system failed to detect a white trailer tractor against a brightly lit sky, causing the autopilot system to think the car was passing under a highway sign rather than running into the trailer. The OEM was clear to point out that its system is a driver assist feature rather than a true autonomous driving solution. We’re going to see examples of these kinds of accidents as we transition from human controlled to autonomous driving. While developers should not settle for any fatalities or serious injuries in autonomous driving, we also have to realize that the biggest
Michigan is delivering the future of transportation mobility today. Planet M is home to more vision, talent, research resources, and innovative collaboration than anywhere on Earth. Learn more by visiting us at www.planetm.com. #MichiganPlanetM
Autonomous cars at least hold the promise of faster response times, and they could even help your car communicate with the cars around you to avoid accidents. safety issue in automobiles today isn’t the car, it is the person behind the wheel. Even excluding distracted or impaired drivers, a completely sober, fully attentive driver cannot react as quickly to changing circumstances as an automated system, and likely will not always make the best decision in the split-second when something goes wrong. Autonomous cars at least hold the promise of faster response times, and they could even help your car communicate with the cars around you to avoid accidents. Some of this technology doesn’t exist today, and we don’t have enough data to say if it’s safe, but we should be thinking about what should be considered safe before we roll this out to wide usage.
A Probabilistic Approach At the end of the day, for an autonomous system to be considered safe, it has to prove itself to be safer than what is on the road today. According to the United States Department of Transportation’s Fatality Analysis Reporting System in 2014, there were 1.08 fatalities per 100 million miles driven; and it is estimated that 32% of these fatalities were alcohol related. 2 It would then make sense to consider that if a safety-related improvement were added to the automobile has the potential to decrease this rate by a significant amount, then it should be deployed even if it has potential or real known flaws. Of course, this can be difficult to clearly demonstrate, since the longstanding trend is clearly toward safer cars. The fatality numbers from just 10 years prior were 33% higher (1.44 deaths per 100 million miles in 1994)3; in fact, the number in 2014 was the lowest rate on record. So we must separate out this gradual, industry-wide improvement trend from the step-function improvement that increased automation will provide.
We must also be careful not to overstate the data that we do have. As an example, as of August 2016, a major autonomous vehicle technology player had traveled almost two million miles with their autonomous vehicle project, with no reported fatalities.4 Does this mean the technology is safe enough to be widely deployed? Consider the following: • The company is testing their technology in Mountain View, CA, Phoenix, AZ, Austin, TX and Kirkland WA.5 While this is impressive, there is a difference between the average driving conditions in Mountain View and those in Detroit (or most of the rest of the United States, especially in the winter). • Two million miles are impressive, but not having a fatality doesn’t mean much when compared to the average driver in the US (on average, you would have to drive 50 times that to see one fatality, based on the NTHSA data mentioned above). • And even that isn’t particularly significant; suppose this company’s project drove not two million miles, but 275 million miles without a fatality. According to the RAND Corporation, that would give us a 95% confidence level that the company’s system would have fatal accidents at the same rate (or lower) than we see today.6 Further, according to the same study, to verify that the company’s system has a failure rate statistically significantly lower than the human driver failure rate, the testing would have to go on for five billion miles.
For a system to be demonstrably safer than human driving, a trial would have to run for about five billion miles with a lower accident rate than we have today.
The Premier Conferences Devoted to Implementing IoT Technology in Embedded Systems
The Big Event: April 26-27, 2017 ❘ Santa Clara Convention Center IoT Device Security Summit ❘ September 14, 2017 ❘ Santa Clara, CA Connected Devices & Gateways Summit ❘ November 30, 2017 ❘ Santa Clara, CA
Register Now @
Of course, there are other metrics that will need to be considered beyond fatalities such as the numbers of accidents, and accidents with injuries. No question other metrics will have to be considered; which events that will be measured will be the responsibilities of the automotive original equipment manufacturers (OEMs).
Solving This Complex Riddle It is unfeasible for a single automotive supplier to run tests for the billions of miles necessary to prove that theyâ€™re safer than their manual counterparts, but itâ€™s also not necessary. Instead, several steps will have to be taken:
It is this kind of systematic focus on safety that has made all forms of aircraft travel safer year after year, to the point where airline travel is orders of magnitude safer than driving.
a. The existing OEM and Tier-1 suppliers must have a greater emphasis on simulation of autonomous driving systems before wide scale deployment of autonomous vehicles can occur. The number of simulations are limited only by compute horsepower, and itâ€™s realistic that simulations can run millions or even billions of scenarios. b. The existing OEM and Tier-1 supplier automated vehicle pilots must continue to gather the data required to ultimately show safety in practice. c. Once we have reasonable confidence that the data shows a significant improvement
over manual driving, then these systems should be allowed to be widely deployed. d. We must be able to improve over time. The industry is already learning from the aerospace industry on how to improve. In that industry, every accident regardless of scope, type of plane, etc. is reviewed and analyzed by the Federal Aviation Administration, often leveraging other industry groups such as the National Transportation Safety Board (NTSB) and the Air Safety Institute (ASI). These organizations look for why accidents occur, and in the case of system failure, make sure that the same kind of failure does not happen again by putting requirements back onto the aircraft manufacturers and maintenance procedures. It is this kind of systematic focus on safety that has made all forms of aircraft travel safer year after year, to the point where airline travel is orders of magnitude safer than driving. Each fatal accident associated with commercial planes makes headline news. If we can get to that point for automobiles, we will have accomplished something remarkable. There are several initiatives in different parts of the industry starting to focus on this aspect of continuous improvement. Before any large-scale roll-out of autonomous cars, an industry partnership must be created to support this activity, since it seems unlikely that the NTSB will do so.
ADAS and Autonomous vehicles have the promise to greatly improve the overall safety of vehicles on the road in the near future. However, as an industry, we must come together to quantify the improvement so that we can be certain that replacing the driver will not cause more harm than good. While the exact numbers are open to debate, a statistical approach will have to be used to determine when an algorithm should go from the test phase to the deployment phase.
Mentor Automotive, “Mentor Graphics Expands Comprehensive ISO 26262 Qualification Program,” https://www. mentor.com/embedded-software/news/ mentor-iso-26262, accessed, Jan. 2017.
National Highway Traffic Safety Administration Resource Website; “Fatality Analysis Reporting System (FARS) Encyclopedia,” https:// www-fars.nhtsa.dot.gov/Main/ index.aspx, accessed Sept. 2016.
At the same time, the industry must come together to share information when accidents do happen so that we can understand what things go wrong and improve to prevent recurrence. The complexity of autonomous vehicles controlled by machine learning technology requires us to take what we can from the existing ISO 26262 standards, and then adapt the way we think about verification of safety requirements when considering complex learning systems such as neural networks, which will ultimately help power systems that make driving more convenient, available to more people, and safer. All this complexity is simply too much for any one person, team or company to have all of the answers. However, by working together, we can realize the benefits of automation even in the most demanding and life-sensitive of industries.
Google, “Self Driving Car Monthly Report,” https://static. googleusercontent.com/media/ www.google.com/en//selfdrivingcar/ files/reports/report-0816. pdf, accessed August 2016.
5. ibid. 6.
Kalra, P. and Susan M. Paddock, “How Many Miles of Driving Would it Take to Demonstrate Autonomous Vehicle Reliability?” Rand Corporation paper RR1478, http://www.rand.org/content/ dam/rand/pubs/research_reports/ RR1400/RR1478/RAND_RR1478. pdf, accessed Sept. 2016.
Contact Information Robert Bates Email: Robert_bates@mentor.com
At the same time, the industry must come together to share information when accidents do happen so that we can understand what things go wrong and improve to prevent recurrence.
Automotive Interface Solutions
REAL TIME CLOCKS
California CONNECTED VEHICLE Test Bed
An Ideal Environment for Innovative Companies to Test and Develop Connected Vehicle Applications in a Real-World Setting By Benjamin McKeever, PATH
Map of California Connected Vehicle Test Bed RSU locations
Last fall’s inaugural issue of Connected Auto introduced the California Connected Vehicle (CV) Test Bed and announced it was “open for business”. In this article, we provide more information about the test bed’s capabilities, ongoing research efforts, future plans and how potential industry partners can access the facility.
HISTORY The California CV Test Bed was established in 2005 by Caltrans, the Metropolitan Transportation Commission (MTC), and the University of Californiaâ€™s PATH Program as the nationâ€™s first dedicated short-range communications (DSRC) test bed. It is conveniently located in Palo Alto, California near many of the automobile research labs, technology companies and industry start-ups in the Silicon Valley. The test bed provides an opportunity for partners to test and evaluate real-world implementations of connected vehicle technology and applications deployed in an operational setting. The original incarnation of the California CV Test Bed consisted of five intersections equipped with early-generation Denso 5.9 GHz DSRC devices that enabled vehicle-toinfrastructure (V2I) communications. At the time, it was called the Vehicle Infrastructure Integration (VII) Test Bed and was one of only two such test beds in the United States (the other being in Southeast Michigan). For many years, the Test Bed supported cutting edge research in connected vehicle safety, mobility and environmental related applications and services, but as standards changed and matured, it became out of date.
a 2-mile stretch of heavily travelled El Camino Real (State Route-82) in Palo Alto (see map on previous page). This busy arterial carries 60,000 vehicles per day and includes many transit routes. All of the intersections broadcast signal phase and timing (SPaT) and MAP messages over DSRC and are connected to a 4G LTE backhaul. With these consecutive signalized intersections all equipped, it becomes possible to test a number of V2I applications, such as signal priority and eco-driving, at a realistic scale. In carrying out the recent upgrade project, Caltrans District 4 Maintenance staff installed all of the new roadside equipment and established all the needed electrical power and signal connections in the field, with support from PATH Research and Development engineers. For each intersection, a new DSRC roadside unit (RSU) and antenna were installed on the existing traffic signal pole and a field-hardened processor, network switch, and 4G cellular backhaul connection were installed in the traffic signal cabinet. The image below shows a schematic of a typical RSU installation on El Camino Real.
TEST BED UPGRADE In 2014, with support from the United States Department of Transportation (USDOT), Caltrans and PATH upgraded the Test Bed to the most current version of DSRC and to ensure compliance with the latest national standards. The Test Bed was enhanced so that it now spans 11 consecutive intersections equipped with state-of-the-art 5.9 GHz DSRC devices along
Typical RSU Installation on El Camino Real
It is worth noting that the 2014 upgrade project involved more than swapping old radios for new radios and adding a few new intersections. The DSRC standards were changing significantly, as was the architecture of the entire CV system. New work at the national level on definition of messages such as SPaT and MAP and on the interfaces to traffic signal controllers had to be accommodated. Since Caltrans uses AB3418 rather than NTCIP protocols for communicating with its traffic signal controllers, the mapping between these protocols had to be developed as part of the project as well. In the end, a good deal of software development and implementation work was required. Also, a new processor was installed at each intersection to house CV application programs and to provide the interfaces among all the components. Finally, all the Model 170 traffic signal controllers were replaced with Model 2070 controllers along this corridor so that proper software interfaces were available for reading information such as SPaT messages.
Test Bed Corridorâ€”El Camino Real and Stanford Ave.
RECENT CV RESEARCH With a modern, state-of-the-art CV Test Bed in place, Caltrans has recently undertaken a number of research efforts that utilize the unique capabilities of the facility. In 2015, with support from USDOT and the Connected Vehicle Pooled Fund Study, Caltrans successfully demonstrated the development and field testing of a Multi-Modal Intelligent Traffic Signal System (MMITSS) on Californiaâ€™s CV Test Bed. This demonstration is noteworthy since it involved the integration of MMITSS software and DSRC technology into Caltrans-operated traffic signal controllers in a live traffic environment. Further, MMITSS demonstrated the potential to improve traffic operations along the corridor through better signal timing, reduced idling and offering safer conditions for pedestrians and cyclists.
Arada RSU Test vehicle and OBE
MMITSS demonstrated the potential to improve traffic operations along the corridor through better signal timing, reduced idling and offering safer conditions for pedestrians and cyclists.
In late 2016, Caltrans initiated two new projects with PATH and UC Riverside to further expand and enhance the existing CV Test Bed and applications and improve the Test Bed’s reliability for use by private sector partners. Under these efforts, the current CV Test Bed will again be expanded; this time from 11 to 17 intersections on El Camino Real utilizing new RSUs to be provided by USDOT. Each of these new RSUs will be equipped with the latest CV technology and security hardware, allowing them to be connected to the USDOT’s security credential management system (SCMS) to ensure trusted communications between vehicles, roadside devices and back offices. In addition to connecting their RSUs to the SCMS, Caltrans also plans to integrate them with existing or new Model 2070 traffic signal controllers. The RSUs will also be programmed to broadcast real-time position correction messages in addition to SPaT and MAP messages already being broadcast over DSRC to CVequipped vehicles. This latest expansion of the CV Test Bed is expected to be completed in 2017. Also, under these new Caltrans projects, PATH and UC Riverside will enhance previously developed applications such as MMITSS with new algorithms for transit signal priority (TSP) and freight signal priority (FSP) and develop and test new applications such as Eco-Approach and Departure (EAD) at Signalized Intersections using CV-equipped buses from partner transit agencies such as Valley Transit Authority (VTA).
SUPPORT FOR INDUSTRY PARTNERS The California CV Test Bed is currently operated and managed by PATH and is supported by Caltrans, MTC and USDOT’s Intelligent Transportation Systems (ITS) Joint Program Office (JPO). It is one of the USDOT’s network of Affiliated Test Beds and now conforms to the latest technology standards and architecture of USDOT’s CV research program. This designation makes the Test Bed eligible for ongoing technical support from USDOT’s CV Test Bed contractors and ensures that it is kept up to date with the latest national standards and that it is part of the national network of CV Test Beds. Later this year, data from the Test Bed will be transmitted to the National CV Data Warehouse, where it could be used by other interested parties for application development and research purposes. PATH is available to help guide new industry partners through the process of accessing and using the Test Bed. We recently developed a web-based remote monitoring tool for Test Bed users allowing them to remotely check on the operational status of each RSU and see snapshots of SPAT and MAP data coming from each intersection. In addition, the website will house a user’s manual to help industry partners better understand how to use the Test Bed if they are interested in application development. Furthermore, Caltrans fully supports bringing in new industry partners to test applications and technologies on the California CV Test Bed.
“The California CV Test Bed provides an ideal environment for innovative companies to test and develop their connected vehicle applications in a real-world setting, and we welcome their use of this facility”. —Greg Larson, Traffic Operations Chief in Caltrans Division of Research, Innovation and System Information
LOOKING FORWARD As the California CV Test Bed enters its 12th year of operation, there are many reasons to be optimistic about its future and the future of Connected Vehicles in the United States. The National Highway Traffic Safety Administration recently announced a Notice of Proposed Rulemaking that could make DSRC-based V2V communications required in new light vehicles as early as 2020, and many auto makers are starting to make plans for how to best deploy this technology. In addition, the rapid growth and advancement of the automated vehicle (AV) industry has created a great deal of excitement in transportation technology, especially in Silicon Valley. More and more, people are beginning to realize that CV and AV are complementary technologies and that they will inevitably merge at some point in the future. In recognition of this, Caltrans is already broadening its vision of the Test Bed to be one that supports AV testing as well and this has been well received by potential industry partners. Yes, indeed, the future of the California CV Test Bed is certainly bright.
Test Bed Corridorâ€”El Camino and California Ave.
Please contact Ben McKeever at California PATH (Ben.McKeever@berkeley.edu) for additional details on the California Connected Vehicle Test Bed.
DSRC Antenna at El Camino Real and California Ave. Savari RSU
More and more, people are beginning to realize that CV and AV are complementary technologies and that they will inevitably merge at some point in the future.
Impact of Automated Vehicle Technologies on Driving Skills (Part 2) By Center for Automotive Research (CAR) Michigan Department of Transportation (MDOT) (Please see final page for individual contributors)
In recent years, a great number of research teams turned their focus toward understanding the significance of human factors for automated vehicles (AV), specifically the impact of this technology on driving behavior, skills, and abilities. Some of the most important human factors to take into consideration for the deployment of AV are trust and reliance, situational awareness, behavior adaptation, and workload. In addition, there are also discussions regarding the impact of ADS on driving maneuvering skills.
Researchers have identified several issues with human operators’ use of automated vehicle features. Soft driving automation systems, which drivers can override, are particularly associated with problems caused by reduced driver mental workload (performance problems when the driver needed to reclaim control). On the other hand, hard automation applications, which have ultimate authority on the vehicle and can override the driver’s inputs, are associated with problems of trust, situational awareness and mental models. Finally, issues linked to behavioral adaptation are linked to both soft and hard automation systems.
Trust, Reliance Trust and reliance are very important human factors to consider when talking about the adoption and real-life use of AVs, especially because trust in technology takes a long time to build and an even longer time to repair when lost. In general, using automation can sometimes lead to incorrect levels of trust: •
Misuse: “users violate critical assumptions and rely on the automation inappropriately”;
Disuse: “users reject the automation’s capabilities and do not utilize the automation”;
Abuse: “designers introduce an inappropriate application of automation”.
Several studies have shown that, in the case of highly reliable systems, users tend to be complacent, over-relying on automation, and thus using it beyond its intended scope or failing to remain vigilant for potential malfunctions. Over-reliance on automation is believed to be responsible for loss of skill and mode confusion. Conversely, a system perceived as unreliable or not proficient will not be used, regardless of any potential benefits. In addition, research findings indicate that initial perceptions of reliability levels affect subsequent reliability estimates and trust ratings.
According to other studies, automated driving systems that provide information about the driving goals were more trustworthy and acceptable than systems that did not supply information. Moreover, informing drivers about situations in which the automation is uncertain improved operators’ trust in the AV technology and their reliance on the system. Studies have shown that system reliance varies with age. Younger drivers displayed less dependence on automation, and took less time to verify automation suggestions. Older drivers, on the other hand, reported greater trust in automation and experienced higher workloads. However, other research results seem to indicate that older drivers underutilize smart technology because they lack perceived need, lack knowledge of the devices, and perceive the cost to be prohibitive. One of the earlier research projects on the effectiveness of FCW focused on the impact of alarm timing on driver response and trust. This driving simulator test involved six driving situations that combined three driving speeds (40, 60 and 70 mph) and two time headways (1.7 and 2.2 seconds). The main finding was that alarm effectiveness varied in response to driving conditions. In addition, alarm promptness had a greater influence on how participants rated their trust rather than improvements in braking performance enabled by the alarm system. Most drivers expected alarms to activate before they initiated braking actions; when this did not happen, driver trust in the system decreased substantially, because the alarms were perceived as late. Headway times had a great influence on driver performance and perception of alarms. Specifically, when driving with a long time headway setting, drivers’ adaptation to late alarms induced a longer response to the brakes, compared to the no alarm condition, possibly resulting in impaired driver behavior.
SYSTEMS Human Factors Issue
Forward Collision Warning (FCW) Extended exposure (and knowledge of the system) leads to increased trust.
Distrust of the system primarily a result of false positives.
An inaccurate internal representation of ACC may cause the drivers to excessively trust the system.
Trust depends on road type (i.e., the system can be less reliable on rural roads where edge lines may not be conspicuous).
Cooperative Adaptive Cruise Control (CAAC) Trust is dependent on system reliability.
Drivers tend to learn relatively quickly when and where the system will work.
Drivers may have an inaccurate mental model of ACC, which may lead to inaccurate expectations of ACC performances and over-reliance on the device.
Reliance on the system is very limited.
Reliance on the system may occur with extended use.
Carryover effects could emerge, similarly to other technologies
Possible carryover effects with conventional cruise control.
A moderate amount detected (drivers used their turn signals more frequently after exposure to the system)
Behavioral adaptation to CACC may result in shorter gaps during manual control, which may be a safety risk.
Drivers can become distracted.
Over-reliance can lead to distraction since the system (especially those with autobraking) performs the maneuvers. Therefore, drivers are more likely to be engaged in secondary behaviors.
Can lead to minor distraction.
The increased automation of CAAC over ACC may lead to greater distraction.
Situational awareness (linked to distraction and over-reliance) can decrease, which makes drivers to be unprepared to intervene in a critical situation.
Situational awareness can increase, due to the LD alerts.
Can drastically reduce workload (when compared to driving without the system), especially in congestion.
This depends on the HMI (i.e., if the interface is visual and located below the windshield). Reliance on the system can lead drivers to engage in secondary tasks. Related to distraction.
Trust is dependent on experience with the system and knowledge of the system’s operation.
Lane Departure (LD)
Extended exposure can lead to overreliance on the system.
Adaptive Cruise Control (ACC)
If drivers overly rely on the system, they may lose situational awareness (and may not be ready to intervene in a critical situation). System use can lead to a decrease in workload if drivers rely on the system to warn them of events.
On certain roads, drivers may use the system to warn them of a lane departure while they are engaged in a secondary task.
In addition, older drivers tend to have faster reaction times due to greater situational awareness.
Reduced workload from CACC use may enable drivers to engage in non-drivingrelated tasks possibly leading to risks during system failures and emergencies. Possible decrease in workload which can lead to performance decrements due to “mind wandering”.
Another study comparing trust issues associated with a non-adaptive FCW and an adaptive FCW (that adjusted the timing of the warning to the reactions of each driver) indicated the safety benefit of both of these systems. This driving simulator investigation involved 45 experienced simulator drivers. When the FCW system was activated, brake reaction times were reduced and during the braking events, drivers maintained a greater distance from the lead vehicle. Results indicated a difference in trust related to FCW between aggressive (high sensation seeking, short followers) and non-aggressive (low sensation seeking, long followers) drivers. In spite of the safety benefits, the aggressive drivers rated each FCW more poorly than non-aggressive drivers did. The latter preferred the plain FCW to the adaptive FCW. Conversely, aggressive drivers, with their greater risk of involvement in rear-end collisions, preferred the adaptive system, as they found it less irritating and stress inducing. Concerning ACC, one survey of drivers who had this function on their vehicles showed that as experience with ACC increased, drivers became more aware of the functioning and, especially, the limitation of the system.
Carryover Effects, Behavioral Adaptation A carryover effect occurs when an effect “carries over” from one experimental condition to another. There is evidence that AV technology has carryover effects and influences driving performance, especially after drivers return to manual control. For example, drivers using CACC tend to become accustomed to the very close time gaps used by the automation. They will also tend to continue at similar gaps even after resuming manual control, thus creating potentially dangerous situations. The notion of carryover effects link to behavioral adaptation which, in the case of AVs, relates to the behavior changes that occur to drivers using this technology. If general behavioral adaptation increases a being’s chances of survival, when it comes to traffic safety, behavioral adaptations to automation sometimes has negative consequences. For example, several research efforts indicated that drivers tend to misuse the increased safety margins that ADAS features provide, by adapting their driving style (e.g., increasing their driving speed and paying less attention to the driving task than when driving without ADAS). Other research evidence suggested that factors like age, gender, degree of experience, personality traits, and driving style influence behavior changes. Nevertheless, some scientific reports also revealed positive changes in drivers’ behavior linked to automation. One study showed that drivers using ACC (and even conventional cruise control) had about 3 to 6 mph lower maximum speed compared to manual driving and spent less time at limit-violating speeds. It appears that having to consciously set the speed maintained by the ACC at discrete time-points contributes to better regulation compliance, than continuously adjusting the speed with the accelerator pedal. However, a research consensus for this point does not exist; an earlier study actually found that drivers went faster with ACC. A study focusing on behavioral adaptation examined to what extent the effect of FCW
Situational Awareness, Distraction
on response performance is moderated by repeated exposure to a critical lead vehiclebraking event. The trial, performed on a moving-base simulator, also studied whether these effects depended on how critical the events were (i.e. available time headway when the lead vehicle starts to break). The response times and accelerator release times became significantly shorter with repeated exposure for both the FCW and baseline groups (without FCW). The tests showed that the effect of FCW depended strongly on both repeated exposure and initial time headway. Drivers with FCW had faster braking response times than those without it. The effect of event repetition on response times was much larger for FCW drivers. These results provide an example of positive behavioral changes induced by driving automation. Another investigation focusing on lane departure warnings found that this feature made participant drivers less aggressive on average. Carryover effects identified as the warning triggered an overall nine percent increase in the rate of turn signal usage. Therefore, the LDW system made drivers more aware of the fact that they were not using turn signals as often as they should have. Risk compensation or risk homeostasis is another concern linked to carryover effects. According to this theory, drivers begin to accept more risk because they perceive the automation to be more competent. This could in turn lead to more distraction and reliance on the automated driving system.
Situational awareness (SA) is defined as the perception of environmental elements in time and space (operational dimension), the comprehension of their meaning (tactical dimension), and the projection of various changes in their status (strategic dimension). Researchers generally believe that, over time, engaging in secondary tasks while driving deteriorates the skills related to the three aspects of SA aforementioned, therefore lowering driving performance. This is an alarming prospect, because critical situations (e.g., automation malfunction, unexpected events) require quick reactions and a high level of SA. A relative consensus in the research world is that human attention represents a limited resource and that the brain needs a certain level of stimulus to maintain attention and performance levels high. Research results have shown that drivers tended to engage more in secondary tasks as the level of automation increased, thus becoming potentially more distracted. Another observation was that drivers were more likely to perform secondary tasks when the lateral control was automated, versus longitudinal control. In a relatively straightforward environment, performing secondary tasks was not found to be detrimental to driving performance. In fact, driversâ€™ attention on the roadway increased, especially when faced with a demanding task. A survey of drivers using the ACC on their vehicles showed that this technology is associated with lower situational awareness and mode confusion. Some participant drivers reported they had forgotten whether the ACC was activated or not. These drivers were therefore less able to determine whether a situation required their intervention or not. This research also revealed that because of a lack of feedback from the ACC (the system accelerates and decelerates without any indication it will do so), drivers could only react after the system had performed the action or when they realized the system was not taking action as expected.
Studies on brake response time (BRT) when using ACC clearly demonstrated the highly negative effects of distraction and reduced SA. Drivers using ACC had much higher BRTs than drivers with manual control did, even when the braking event was expected or easily anticipated. In addition, deceleration rates with ACC were twice as large as those with conventional cruise control, and substantially less safe than those during manual driving, therefore demonstrating a reduced SA. Finally, a similar study showed that drivers assisted by automation braked only after the collision alert sounded, significantly reducing the minimum time to contact. Drivers also performed the worst when they tried to regain manual operation from the ADS system. These conclusions are not unanimous among the research community and not all technologies have the same effect on SA. A recent study (driving simulator, 30 participants) analyzed driving and gaze behavior, as well as the engagement in a secondary task, when drivers used a forward collision warning (FCW) and braking system (FCW+). The analysis of the gaze behavior showed that driving with the FCW+ system did not lead to a stronger involvement in secondary tasks. Moreover, the FCW+ shifted the attention of the drivers toward the cockpit when the visual symbols of the HumanMachine-Interface (HMI) appeared. In this test, a substantial number of crashes occurred in the critical situations (e.g., a lead car braking unexpectedly) without FCW+. Conversely, using the FCW+ resulted in significantly fewer crashes,
because the automation reacted significantly earlier than the drivers could. Although the drivers were able to detect the deceleration on the lead car at the same moment as the system, they were not able to react fast enough, which made the fast autonomous intervention of the system necessary.
Workload Workload represents the overall level of attention that a task or group of tasks demand from a person. The Yerkes-Dodson Law stated that human performance is optimal when workload levels are in between the extremes. As the complexity of tasks increases, the workload increases, and the ability to handle supplementary tasks decreases. However, an excessively low workload can also be detrimental, resulting in fatigue or distraction. Experience performing a task tends to lower workload. Consequently, a novice driverâ€™s workload level may be very high even with basic vehicle control tasks, which leaves little attention resources for other tasks like traffic prediction or danger identification. Most research results showed that automation would reduce workload. In addition, these results show this effect was augmented as automation increased. For instance, automated steering induced greater workload decreases than ACC, compared to manual driving.
This can be a great advantage when drivers use this additional attention capacity to other driving tasks such as monitoring the environment to identify potential hazards or to predict changes in traffic. However, this attention transfer is not automatic, as drivers may also focus on non-driving tasks. There are several hypotheses on how AV technology will affect drivers’ workload, with no clear consensus on how the reduced workload of automation affects drivers in critical situations. •
Attention resource degradation hypothesis: As the human operator is no longer actively focused during automation control periods, attention resources may shrink in order to adapt to the reduction in demand. Therefore, when the driver resumes manual control, the attention demand increases very quickly and performance may be inadequate to ensure a safe transfer of control and overall driving.
Attention resource conservation hypothesis: As the attention demand is low during automated control, the driver can rest and replenish their cognitive resources. Thus, when needed (e.g., reacting in a critical situation), the driver will be able to deploy their cognitive resources.
Compensation hypothesis: The driver is able to recognize and compensate for a higher workload demand and increase their performance, possibly due to an increase in their general motivation level. However, this theory does not seem to apply to complex tasks.
Maneuvering Skills As it has been previously stated, prolonged use on AV technology may cause a loss of skill, especially when it comes to higher levels of automation, which may lead a substantial loss of skill over time. This is a tendency observed with many types of automation. Maneuvering skills that automation could negatively affect include: •
Maintaining longitudinal and lateral control
Respecting traffic signs, reacting to different traffic situations (e.g., speedway, inner city)
Handling weather conditions (e.g., rain, fog, snow, ice, nighttime)
Reacting to unexpected situations (e.g., vehicle failure, crash avoidance)
Interacting with other vehicles or participants in traffic.
Nonetheless, some of the most recent research showed that the need to retake manual control of the vehicle when attention is directed to a non-driving task could lead to dangerous and sudden changes in workload that can have a negative impact on driving safety. For example, performance deteriorated if the driver had to resume control to change lane due to an incident on the road.
List of abbreviations used in this article
Some evidence suggests that after using AV technology, drivers show poorer lane keeping performances, shorter headways, or delayed reaction times, compared to drivers that have not used AV technology. In addition, tests have proven that the type of automation support (longitudinal versus lateral) has a different impact on driversâ€™ engagement and performance. Other research teams did not expect that dual-mode vehicles would cause a loss of skill, because drivers would still partly use their vehicles manually, for automation levels one to three. Research on crash causation argues that the ability to detect and control traffic hazards improves uniformly as the amount of miles travelled increases. Therefore, the crash rate per unit of exposure will decline as the amount of exposure increases. Hence, as automation will decrease the number of vehicle miles actively driven, drivers will no longer possess the skills to avoid a dangerous situation that automation cannot handle.
Anti-Lock Braking Systems
Adaptive Cruise Control
Advanced Driver Assistance Systems
Automated Driving Systems
Automated Emergency Braking
Conventional Cruise Control
Center for Automotive Research
Collision Avoidance System
Connected and Automated Vehicle
Collision waning system
Electronic Stability Control
Forward Collision Warning
Intersection Movement Assist
Intelligent Transportation Systems
Michigan Department of Transportation
SAE ORAVS Society of Automotive Engineers (SAE) On-Road Automated Vehicle Safety (ORAVS)
Adela Spulber, Transportation Systems Analyst, CAR
Richard Wallace, M.S., Director, Transportation Systems Analysis, CAR Gary Golembiewski, Senior Research Associate, Leidos Matt Smith, ITS Program Manager, MDOT |Niles Annelin, Transportation Planner, Connected and Automated Vehicle Policy, MDOT
This document is a product of the Center for Automotive Research under a State Planning and Research Grant administered by the Michigan Department of Transportation.
Autonomous Driving to the Next Levelâ€“ But Which Level?
By Alan Hutton Connected Auto, Germany
While many drivers enjoy aspects of driving a car and being in control of every driving decision, others feel that this level of constant control is a burden. There are many others that would like to drive, but cannot drive due to physical ailments. Operational decisions and responsibility can come in many forms, particularly when it relates to the ultimate driving experience. On the other hand, when it comes down to an interface that could control the car and preserve the lives of the driver and other passengers in the event of an accident, this “unwanted” function begins to play it’s part in making driving safer. Indeed, many automotive manufacturers have already made it clear that “real-time decision making and security are non– negotiable necessities for the next generation of car technologies since a fraction of a second can mean the difference between life and death”. The migration towards autonomous driving is already in full production rollout, with the likes of Tesla already offering cars with a
significant degree of Advanced Driving Assisted Systems (ADAS) to improve driver/passenger safety and simplify the flow of traffic.
Can ADAS Really Transform a Society? Statistics show that 90% of all automotive related accidents are due to driver error. If driver error can be significantly reduced, or eliminated altogether, the number of traffic deaths will drop and insurance claims will go down. So can ADAS really make the monumental leap to autonomous driving a reality in the near future? Features in cars today including lane departure warning, adaptive cruise control, blind spot detection and emergency brake assist are key ingredients towards fully autonomous driving cars. According to Strategy Analytics, ADAS will be the fastest growing automotive segment and they believe that by 2025 we are likely to see 15 to 20% of all cars highly automated. ABI Research also states that the ADAS market will be worth $132B USD by 2026.
Driver continuously performs the
Driver must monitor
Driver does not need
performs the longitudinal or lateral dynamic
the system at all times.
to monitor the system at all times.
longitudinal and lateral dynamic driving task.
FULL AUTOMATION No driver required during entire journey.
driving task. AUTOMATION
Driver must be capable of resuming dynamic driving task.
No intervening vehicle system active.
The other driving task is performed by the system.
System performs longitudinal and lateral driving task in a defined use case*. Recognizes its limits and requests System performs driver to resume the longitudinal and dynamic driving task lateral driving task in with sufficient time a defined use case*. margin.
*Use cases refer to road types, speed ranges, and environmental conditions
Figure 1: The Six Levels of Autonomous Driving Source – German Association of the Automotive Industry (VDA)
Driver is not required during defined use
System performs the lateral and longitudinal dynamic driving task in all situations in a defined use case*.
System performs entire dynamic driving task on all road types, speed ranges and environmental conditions.
Autonomous Driving Classified
distance between vehicles in real-time and combined with collision avoidance and emergency braking systems.
Back in 2014, the Society of Automotive Engineers (SAE) issued six levels of autonomous driving and were based on the level of driver intervention (Figure 1). Self-driving cars require multiple connected technologies to work together, including GPS, radar, lidar, cameras, and sensors to name a few. To reach any one of the classification levels there are a number fundamental technologies utilized today to reach the various SAE classifications for both the driver and for the car: • There must be vision systems using camera technologies to enable lane detection/ departure warning, forward collision warning, pedestrian detection, traffic sign detection, rear view and surround view systems. • There must be automotive radar offering both short and long-range functionality to support
• There must be communication outside of the car (V2x) to allow vehicle-to-vehicle (V2V) communication and vehicle-toinfrastructure (V2I) connection with intelligent transportation systems. Combining the above functionality and technologies together represents a formidable amount of computing power requirements. Many chip suppliers have already stated that they generate over 1GB of data every second from in and around the car which is no surprise given the number of safety critical functions bring monitored all at the same time. Intel, for example, predicts that the computing requirements for self-driving vehicles will be in the region of 1M Dhrystone MIPS (Figure 2).
It Starts with Covenience and Ends with Safety
The Road to Autonomous Driving
Electromechanical Safety Air bags Electronic Stability Control ABS
Adaptive cruise control Emergency braking Lane keeping
Lane departure In-Vehicle Blind Spot Infotainment Parking assist Development Integration
Compute requirements increase with system functionality
10,000 Computer (DMIPS)
Numbers are for illustrative purposes only and do not represent actual measurements
Figure 2: Intel’s Processing Vision towards Autonomous Driving
It’s All about Safety and Security When a driver gets into a car, the one expectation is that they will be secure both in terms of safety and security. It is imperative that the data flowing inside and outside of the car is immune from malicious intent or individuals accidentally unplugging sensor nodes. Having a homogeneous computing ecosystem as opposed to an archaic decentralized computing approach is the way forward. These remote IoT nodes within the car are secure using cryptographic key locks built into protected hardware to manage multi-layer security. In addition to channel authentication, encryption and data integrity are established at the application layer to protect the data flowing through the pipe into the cloud-computing platform.
Moving Up the Levels with Autonomous Driving Earlier this year, BMW, Intel, and Mobileye announced their commitment to define an open platform for autonomous driving where levels 3 to 5 would be made available to multiple car vendors and other IoT players that could benefit from deep machine learning. The ultimate goal of this partnership is to move from the “conditional automation” (Level 3) towards taking the “full automation” (Level 5) and have a self-driving BMW fleet by 2021.
How could these three companies join forces? Some of the most obvious reasons include: • BMW are leaders in the automotive world and pride themselves on building the “ultimate driving machines”. They also realize that they cannot do this by themselves when it comes to fully autonomous driving. The BMW iNEXT model is expected to hit the market in 2021 and will be the foundation for BMW’s autonomous driving strategy.
“At the BMW Group, we always strive for technological leadership. This partnership underscores our strategy to shape the individual mobility of the future,” stated Harald Krüger, Chairman of the Board of Management of BMW AG. “Following our investment in high definition live map technology at HERE, the combined expertise of this partnership will deliver the next core building blocks to bring fully automated driving technology to the street.” • Mobileye offers leading edge technologies with their computer vision/machine learning technology (EyeQ5TM platform) and localization Road Experience Management (REM) technology. The integration of their fusion algorithm data in combination with the Intel® processing platform will be integrated into each autonomous vehicle.
“This is an important milestone for the automotive industry as we enter a world of new mobility,” says Mobileye Co-Founder and Chairman Professor Amnon Shashua. “This partnership is laying the groundwork for the technology of future mobility that enables fully autonomous driving to become a reality within the next few years.”
• Intel is the master of data processing. They have compute power that scales from Intel Atom™ to Intel Xeon™ processors delivering hundreds of teraflops of power-efficient performance.
“Highly autonomous cars and everything they connect to will require powerful and reliable electronic brains to make them smart enough to navigate traffic and avoid accidents,” says Intel CEO Brian Krzanich. “This partnership will help us to quickly deliver on our vision to reinvent the driving experience.” Conclusion Collaboration with end-to-end solutions will be key in the future development of autonomous driving. BMW, Intel, and Mobileye are just the tip of the iceberg when it comes to technology companies addressing the challenges from big data cloud computing to security and safer driving and saving lives. The combination of sophisticated algorithms and thousands of hours of trial periods on test roads being fed from a myriad of real-time data coming from an array of complex sensors is apparent. Autonomous driving will change societies and change lives for the better. The faster these technology companies embrace this disruptive approach, the better for safety and security features and functions during an enjoyable drive.
Mobileye is the global leader in the development of vision technology for Advanced Driver Assistance Systems (ADAS) and autonomous driving. Mobileye’s system-on-chip (SoC)—the EyeQ® family—provides the processing power to support a comprehensive suite of ADAS functions based on a single camera sensor. In its fourth and fifth generations, EyeQ® will further support semi and fully autonomous driving, having the bandwidth/ throughput to stream and process the full set of surround cameras, radars and LiDARs. Advanced Driver Assistance Systems (ADAS) systems range on the spectrum of passive/active. A passive system alerts the driver of a potentially dangerous situation so that the driver can take action to correct it. For example, Lane Departure Warning (LDW) alerts the driver of unintended/unindicated lane departure; Forward Collision Warning (FCW) indicates that under the current dynamics relative to the vehicle ahead, a collision is imminent. The driver then needs to brake in order to avoid the collision. In contrast, active safety systems take action. Automatic Emergency Braking (AEB) identifies the imminent collision and brakes without any driver intervention. Other examples of active functions are Adaptive Cruise Control (ACC), Lane Keeping Assist (LKA), Lane Centering (LC), and Traffic Jam Assist (TJA). ACC automatically adjusts the host vehicle speed from its pre-set value (as in standard cruise control) in case a slower vehicle is in its path. LKA and LC automatically steer the vehicle to stay within the lane boundaries. TJA is a combination of both ACC and LC under traffic jam conditions. It is these automated features which comprise the building blocks of semi/fully autonomous driving.
The Road Ahead...
Lyftâ€™s Vision for the Next Ten Years and Beyond By John Zimmer, Co-Founder, Lyft
A Country Built for Cars I remember when I first fell in love with cars. It started small with Hot Wheels when I was three and Micromachines when I was six. Everything about them was fast and exciting — even the commercials were narrated by the World’s Fastest Talker. I loved them. Then, when I turned 12, my dad and I began taking annual trips to see the real thing at the New York International Auto Show. I looked forward to going every year, because even at that young age, I felt a connection to cars and the freedom they represented. I think, in some ways, it was my love of cars that largely influenced how I saw the world. However, it wasn’t until I took a life-changing city planning course in college that I had an epiphany: cars weren’t just shaping my worldview, they were shaping the world itself.
Next time you walk outside, pay really close attention to the space around you.
In the class, we learned about the history of cities and the massive impact transportation had on their evolution — both on how they were built and how people lived in them. From then on, I couldn’t help thinking about the inextricable link between transportation and the design of the cities I lived in. And I started noticing a very basic problem everywhere, hiding in front of our eyes. Next time you walk outside, pay really close attention to the space around you. Look at how much land is devoted to cars and nothing else. Notice how much space parked cars take up lining both sides of the street, and how much of our cities go unused because they are covered by parking lots. It becomes obvious we’ve built our communities entirely around cars, and for the most part built them for cars that aren’t even moving. The average vehicle is used only 4% of the time and parked the other 96%. Most of us have grown up in cities built around the automobile, but imagine for a minute what our world could look like if we found a way to take most of these cars off the road. It would be a world with less traffic and less pollution. A world where we need less parking — where streets can be narrowed and sidewalks widened. It’s a world where we can construct new housing and small businesses on parking lots across the country — or turn them into green spaces and parks. That’s a world built around people, not cars. All of this is possible. In fact, as we continue into our new century, I believe we’re on the cusp of nothing short of a transportation revolution — one that will shape the future of our communities. It is within our collective responsibility to ensure this is done in a way that improves the quality of life for everyone. The coming revolution will be defined by three key shifts: 1. Autonomous vehicle fleets will quickly become widespread and will account for the majority of Lyft rides within five years.
Last January, Lyft announced a partnership with General Motors to launch an on-demand network of autonomous vehicles. If you live in San Francisco or Phoenix, you may already see these cars on the road. Within five years a fully autonomous fleet of cars will provide the majority of Lyft rides across the country. Tesla CEO Elon Musk believes the transition to autonomous vehicles will happen through a network of autonomous car owners renting their vehicles to others. Elon is right that a network of vehicles is critical, but the transition to an autonomous future will not occur primarily through individually owned cars. It will be more practical and appealing to access autonomous vehicles when they are part of Lyft’s networked fleet. Why? For starters, our fleet will provide significantly more consistency and availability than a patchwork of privately owned cars. That kind of program will have a hard time scaling because individual car owners will not want to rent their cars to strangers. Above all, passengers expect clean and well-maintained vehicles, which can be best achieved through Lyft’s fleet operations. Today, our business is dependent on being experts at maximizing utilization and managing peak hours, which allow us to provide the most affordable rides. This core competency translates when we
move to an autonomous network. In other words, Lyft will provide a better value and a superior experience to customers. I’ll have more to say on how the autonomous network will work a bit later in this piece. 2. By 2025, private car ownership will all but end in major U.S. cities. As a country, we’ve long celebrated cars as symbols of freedom and identity. But for many people — especially millennials — this doesn’t ring true. We see car ownership as a burden that is costing the average American $9,000 every year. The car has actually become more like a $9,000 ball and chain that gets dragged through our daily life. Owning a car means monthly car payments, searching for parking, buying fuel, and dealing with repairs. Ridesharing has already begun to empower many people to live without owning a car. The age of young people with driver’s licenses has steadily decreased since I was born. In 1983, 92% of 20 to 24-year-olds had driver’s licenses. In 2014 it was just 77%. In 1983, 46% of 16-yearolds had licenses. Today it’s just 24%. All told, a millennial today is 30% less likely to buy a car than someone from the previous generation. Every year, more and more people are concluding that it is simpler and more affordable to live without a car. When networked autonomous vehicles come onto the scene, below the cost of car ownership, most city-dwellers will stop using a personal car altogether. 3. As a result, cities’ physical environment will change more than we’ve ever experienced in our lifetimes. So why should you care about changes in transportation? Even if you don’t care about cars — even if you never step into a Lyft or an autonomous vehicle — these changes are going to transform your life. Transportation doesn’t just impact how we get from place to place, it shapes what those places look like, and the lives of the people who live there.
The end of private car ownership means we’ll have far fewer parked cars and the chance to redesign our entire urban fabric. Cities of the future must be built around people, not vehicles. They should be defined by communities and connections, not pavement and parking spots. They need common spaces where culture can thrive — and where new ideas can be shared in the very places where cars previously parked. Urban reimagination has the opportunity to deliver one of the most significant infrastructure shifts we have ever undertaken as a nation. The good news is that we have to make these investments anyway. The American Society of Civil Engineers recently gave U.S. infrastructure a D+, estimating that our country requires $3.6 trillion in infrastructure investment by 2020. If we have to rebuild and revitalize our roads and cities anyway, let’s do it in a way that puts people, not cars, at the center of our future. Before we continue looking forward, I want to take a moment to look back at how we got here. There’s something I haven’t mentioned yet: this will be America’s third transportation revolution.
How We Got Here: America’s First Two Transportation Revolutions America looked very different in the early days. At the turn of the nineteenth century, the U.S. was made up of loosely connected, largely agricultural communities. If you wanted to travel over long distances, the covered wagon was pretty much your best option. The United States, in other words, was still pretty divided.
Transportation doesn’t just impact how we get from place to place. It shapes what those places look like, and the lives of the people who live there.
That all changed over the next several decades, as America constructed a massive transportation network of canals and railroads. By 1860, the first revolution was in full swing as more than 30,000 miles of railroad track spread out across the U.S. — and as tracks linked together, so did communities, economies, and people. Wherever these transportation networks went, small outposts were transformed into thriving cities. Chicago, Baltimore, and Los Angeles exist as they do today because of transportation innovations that helped spark their growth.
Now fast-forward into the next century, when the assembly line automobile came onto the scene. For individuals, this brought almost unprecedented freedom. But for our cities, car ownership started a vicious cycle. More cars filled the streets, more roads had to be built to accommodate them. This second transportation revolution caused communities to spread farther and farther apart, which made having constant access to a car increasingly necessary — resulting in even more cars that needed even more space. In the process, our cities were dramatically reshaped to favor cars over communities. Across the country, city planners wanted to make it as easy as possible for drivers to access metropolitan areas. That often meant building highways straight through the centers of our most vibrant cities. Neighborhoods were literally split in half, and many never recovered. In some cases, neighborhoods were demolished to make room for cars. In Los Angeles, for instance, engineers built structures like the Four Level Interchange, which connects the 101 and 110 freeways and hosts 425,000 cars a day. The builders made room for it by knocking down 4,000 houses and apartment buildings. In addition to widespread demolition, there was also a more subtle way that cars began to reshape our cities. Streets themselves used to look very different than they do today. Most were more narrow, leaving room for sidewalks, front yards, and places where people could come together outside.
Back then, people used city streets as public spaces. Streets were where children could play. A place for shopping, where you could stop at a cart on the way home to pick up everything from dinner ingredients to shoes for your family. People spent a lot of time outside on the street, making friends, seeing neighbors, and living their lives within a true community. When streets began to be redesigned for more and more cars, all of these other benefits suffered. As time went on, streets became a place solely for cars. They encroached closer to homes so yards disappeared. People were left with narrower sidewalks or no sidewalks at all. That meant less foot traffic, which made it harder for small businesses, shops, and restaurants to flourish. Development patterns changed dramatically and the strip mall was born. With fewer people outside, neighborhoods became less safe because we lost the benefit of having “eyes on the street” most hours of the day. For the first time in history, cities were no longer centered on human social interaction. All of this made it harder for a community to thrive. And as changes like this played out across the country, the face of America’s cities was transformed for generations.
The Problem with Cars At this point, we should probably take another step back to answer a simple question: why is a company built around cars complaining about cars? The answer is that vehicles themselves aren’t the problem. The problem is how we use them — and just as importantly, how we don’t. I studied hospitality in college, so sometimes I can’t help looking at the world through the lens of a hotel. What’s the occupancy? Are you getting great service? It’s interesting to think this way about transportation — to imagine that our ground transportation is being run like a hotel.
Credit: California Historical Society Collection, USC Libraries
To measure the health of our transportation hotel, let’s start by looking into how much money we spend on car ownership and how often we actually use our cars. It may shock
you, but Americans spend more than $2 trillion every year on car ownership — more money than we spend on food. What’s even more staggering is that for all the money we spend on them, the 250 million cars in America are only occupied 4% of the time. That’s the equivalent of 240 million of the 250 million cars being parked at all times. For the most part, your car isn’t actually a driving machine at all. It’s a parking machine. Can you imagine a hotel where almost every room is empty? A hotel that spends an enormous amount of money maintaining those empty rooms, no matter how little they’re used? It would go out of business tomorrow. And if you think about occupancy of cars the same way, the observation is simple: America is running a failing transportation business. Plus, think about where all those unused cars sit while they’re idle. In 2011, researchers estimated that there are at least 700 million parking spaces in the U.S. That means our country has more than 6,000 square miles of parking —bigger than my home state of Connecticut. We can’t be this inefficient anymore, because we’re about to hit an inflection point that will strain our cities’ resources like never before. The U.S. already has ten cities with more than a million people and our urban population is growing fast. By 2050, almost 100 million more people will move to American cities. We don’t have enough space, housing, or public transit to accommodate this population influx, especially while keeping cities a desirable place to live. While fixing transportation won’t solve all these problems, it certainly doesn’t help to continue devoting so much of our space to unoccupied cars.
The Third Transportation Revolution The good news is we don’t have to keep building our country around car ownership. Technology has redefined entire industries around a simple reality: you no longer need to own a product to enjoy its benefits.
With Netflix and streaming services, DVD ownership became obsolete. Spotify has made it unnecessary to own CDs and MP3s. Eventually, we’ll look at owning a car in much the same way. A full shift to “Transportation as a Service” is finally possible, because for the first time in human history, we have the tools to create a perfectly efficient transportation network. We saw this potential in 2012 when Lyft became the first company to establish peer-to-peer, on-demand ridesharing, which is now what the world knows simply as ridesharing. What began as a way to unlock unused cars, create economic opportunities and reduce the cost of transportation, has today become the way millions of Americans get around. Ridesharing is just the first phase of the movement to end car ownership and reclaim our cities. As I mentioned before, the shift to autonomous cars will expand dramatically over the next ten years, transforming transportation into the ultimate subscription service. This service will be more flexible than owning a car, giving you access to all the transportation you need. Don’t drive very often? Use a pay-as-yougo plan for a few cents every mile you ride. Take a road trip every weekend? Buy the unlimited mileage plan. Going out every Saturday? Get the premium package with upgraded vehicles. The point is, you won’t be stuck with one car and limited options. Through a fleet of autonomous cars, you’ll have better transportation choices than ever before with a plan that works for you. Using the Lyft network will also save you money. We don’t often think about it, but owning a car and making monthly payments also means paying retail prices for every aspect of getting where you need to go : fuel, maintenance, parking, and insurance. In a future subscription model, the network will cover all of these costs across a large network of cars, passing the savings onto you. We cut the hassle and you get the one thing you really want: the true freedom to ride.
Ridesharing is just the first phase of the movement to end car ownership and reclaim our cities.
This piece originally ran on John Zimmer’s Medium. Read the full piece here.
Leading-Edge Automotive Research CAR forecasts industry trends, advises on public policy, and sponsors multi-stakeholder communication forums.
RECENT RESEARCH Impact of New Mobility Services on the Auto Industry Global Harmonization of Connected Vehicle Communication Standards Effects of the 2017-2025 EPA/NHTSA GHG/Fuel Economy Mandates Costs of Lack of Harmonization of Safety Regulations Impact of Automated Vehicle Technologies on Driver Skills
Smart & efficient
Logic for automotive applications
Nexperia brings over 50 years of proven quality, commitment, and efficiency to automotive Logic applications. Our product portfolio meets and even exceeds the AEC-Q100 standard. Nexperia -Q100 Logic is guaranteed for automotive applications by meeting OEM zero defect requirements, supporting higher visibility and traceability in automotive production. With over 1,000 automotive certified -Q100 types, Nexperia delivers the largest product portfolio in the industry. We offer a wide range of functional categories (gates, switches, shift registers, etc.), traditional HC(T), AHC(T), full feature LVC, AUP logic families, and PnT translating gates, in either Standard Logic (> 10 pin) or Mini Logic (â‰¤ 10 pin) types. With globally located sales and support teams, youâ€™re assured of priority technical support, priority design-in assistance, PPAP Q100 qualification data, and fast turn-around on support issues. To learn more, please visit www.nexperia.com/logic
Powering a Path to Autonomous Vehicles