Invited SpeakerNicola Muscettola
Lead of the Autonomy and Robotics Area
Computational Sciences Division
NASA Ames Research Center, California
Title: Deploying Robust Autonomous Systems: Lessons from the Remote Agent
Abstract: The need to reduce cost and the desire to tackle ambitious exploration goals pushes NASA toward adopting ever more sophisticated and autonomous software systems for its missions. The techniques of Artificial Intelligence promise to solve several key problems in the design and implementation of such software. However, the increase in complexity produces legitimate concerns on system correctness, robustness and reliability. While AI has typically strived to demonstrate sophisticated capabilities, not much attention has been given until now to the problems of verification and validation. This talk discusses these issues from the vantage point of our experience in the Remote Agent Experiment. Remote Agent was the first autonomy architecture to operate in inter-planetary space, controlling the Deep Space One spacecraft for a total of 2 days in May 1999. Remote Agent demonstrated the ability of being commanded by high-level goals and generating plans that achieved them, and the ability to fix the plan after device faults while continuing to keep control of the spacecraft. We will discuss some of the lessons learned from the experiment on how to increase the acceptability of autonomous AI systems by the operational community.
Link: Autonomy and Robtics,