Invited Talk: Making Agents Safe for Space
Speakers: Brad Clement, Daniel Tran, Steve Chien
Abstract:
NASA faces many challenges in making spacecraft flight (and ground)
software safe. A general approach is to thoroughly test when integrated
with other software and hardware in a realistic environment. Validating
autonomous software agents presents increased challenges partly because
of testing complexity and partly due to the kinds of more complex
environments these agents might be applied.
An example of such an agent is the Autonomous Sciencecraft Experiment that has been running on-board the EO-1 spacecraft since 2003. The agent recognizes science events, retargets the spacecraft to respond to the science events, and reduces data downlink to only the highest value science data. The autonomous science agent was designed using a layered architectural approach with specific redundant safeguards to reduce the risk of an agent malfunction to the EO-1 spacecraft. The agent was designed to be safe by first preventing anomalies, then by automatically detecting and responding to them when possible. This talk will also describe elements of the design that increase the safety of the agent, several of the anomalies that occurred during the experiment, and how the agent responded to these anomalies.
Validating agents for multiple spacecraft missions or interacting missions grows even more complex because the interactions add another factor to the number scenarios to test. Formalisms in multiagent systems are helpful for giving safety guarantees, but it does not necessarily reduce the testing needed.
Bios:
Dr. Brad Clement is a senior member of the Artificial Intelligence Group at the Jet Propulsion Laboratory in Pasadena, CA, where he is developing methods for coordinating planning and scheduling for single and multiple spacecraft/missions. He leads projects on distributed continual planning applied to simulated spacecraft and rovers for Mars, on scheduling resource allocation for the Deep Space Network, and on planning under uncertainty. He received a Ph.D. degree in computer science and engineering from the University of Michigan, Ann Arbor. His research interests include multiagent coordination, situated planning and execution, distributed systems, and AI in games.
Daniel Tran is a member of the technical staff in the Artificial Intelligence Group at the Jet Propulsion Laboratory, California Institute of Technology, where he is working on developing automated planning and scheduling systems for onboard spacecraft commanding. Daniel attended the University of Washington and received a B.S. in Computer Engineering, graduating with honors. He is currently the software lead for the Autonomous Sciencecraft Experiment flying onboard the Earth Observing-1 satellite.
Dr. Steve Chien is a Principal Computer Scientist at the Jet Propulsion Laboratory where he leads efforts in autonomous space systems and is the Principal Investigator for the Autonomous Sciencecraft Experiment. Dr. Chien is also an Adjunct Associate Professor with the Department of Computer Science of the University of Southern California. Dr. Chien was a recipient of the 1995 Lew Allen Award for Excellence. In 1997, he received the NASA Exceptional Achievement Medal for his work in research and development of planning and scheduling systems for NASA. In 2000, he received the NASA Exceptional Service Medal for service and leadership in research and deployment of planning and scheduling systems for NASA.