Enterprise is closely related to a system called Remote Agent, which I developed with NASA colleagues following the loss of Mars Observer, a spacecraft that, days before its 1993 scheduled insertion into Marsâ€™ orbit, permanently lost contact with NASA. While skilled operation teams were available to diagnose and repair Mars Observer, they were defeated by the lack of communication. Subsequently, NASA realized that it needed systems on board a spacecraft that could reason at a cognitive level, similar to its engineers. In May 1999, Remote Agent successfully controlled the NASA Deep Space One probe, two weeks prior to its asteroid Remote Agent successfully encounter, and demonstrated the ability to diagnose, repair, and re-plan its mission. controlled the NASA Deep
Space One probe for two weeks
The types of goal-directed methods demonstrated on Deep prior to its asteroid encounter. Space One have subsequently supported a range of missions, including Earth Observer One, the Spirit and Opportunity rovers and the recent Rosetta mission. Meanwhile, advances in the research community have made goal-directed systems easier to use, faster, and more robust. Enterprise leverages many of these advances. Returning to the second aforementioned finding in the KISS report, a significant barrier to adopting goal-directed execution widely is mission risk. The correctness of the actions generated by a goaldirected executive depends upon the correctness of the model used to generate them. Unless the executive can reason about model and environment uncertainty, and prove that mission risks are not excessive, deployment of goal-directed execution will be limited. Enterprise embodies the concept of a goal-directed executive that is risk aware. The operator specifies to Enterprise acceptable risk levels for different types of failures, such as missing a science observation, failing to reach an end-of-day rendezvous point within a specified time, depleting the battery below a specified level, or moving into an untrafficable area with soil slippage. The captain asks the navigator for routes between points of observation that avoid untrafficable areas and avoids battery depletion within the specified risk levels. The captain then selects an overall mission plan that maximizes science, while ensuring that the end-of-day rendezvous is reached on time, again to the specified risk level. If the captain cannot succeed, the communicator explains to the operations team why their goals are too risky, and proposes ways to relax the mission goals, so that risk becomes acceptable.
Goal-directed risk-aware autonomous explorers
Published on Nov 18, 2016
Annual magazine review of MIT Aeronautics and Astronautics Department research and educational initiatives.