Abstract:
Proposed missions to explore comets and moons will encounter environments that are hostile and unpredictable. A successful explorer must be able to adapt to a wide range of possible operating solutions to survive. Determining an appropriate operations strategy requires information about the environment, which is not available a priori for these missions. Instead, we propose an explorer that uses machine learning to continually adapt its behavior while limiting the cost of behavior exploration.

For example, a proposed mission to send a submersible to explore oceans beneath the ice caps of Europa, will need to explore a media of unknown depth, composition, and temperature. These environmental conditions will determine the power cost of movement and ability to stabilize, cost, reliability, and rate of communications. All these mean that a pre-designed mission strategy will need to be adapted in-situ. Furthermore, with space mission costing several hundred million dollars, safe learning is imperative. There must be some guarantees on the performance of the adapted agent for an expected range of contexts.

In this talk I will first describe several space mission concepts on which safe learning techniques could have high impact including the Europa Cryobot, Comet Lander, Mars Robotic Outpost, and Interstellar Missions. Correspondingly I will describe why learning, and safe learning are applicable. Then, I will describe efforts at JPL to use statistical reinforcement learning to enable adaptive planning agents. Finally, I will review some ongoing work to use machine learning in a less ambitious fashion to automatically recognize science events to trigger mission re-planning as part of the Autonomous Sciencecraft Flight on the EO1 Mission.

Acknowledgements:
The work described in this talk was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. The science response work described is part of the Autonomous Sciencecraft effort funded by the New Millennium and Intelligent Systems Programs. The reinforcement learning portions of this work were in collaboration with Barbara Engelhardt (currently at the University of California at Berkeley).

Biography:
Dr. Steve Chien is Technical Group Supervisor of the Artificial Intelligence Group and Principal Computer Scientist in the Exploration Systems Autonomy Section at the Jet Propulsion Laboratory, California Institute of Technology, where he leads efforts in automated planning and scheduling for space exploration. Dr. Chien is also an Adjunct Associate Professor with the Department of Computer Science of the University of Southern California and a Visiting Scholar at UCLA. He holds a B.S. with Highest Honors in Computer Science, with minors in Mathematics and Economics, M.S., and Ph.D. degrees in Computer Science, all from the University of Illinois.

Dr. Chien was a recipient of the 1995 Lew Allen Award for Excellence, JPLs highest award recognizing outstanding technical achievements by JPL personnel in the early years of their careers. In 1997, he received the NASA Exceptional Achievement Medal for his work in research and development of planning and scheduling systems for NASA. He is the Team Lead for the ASPEN Planning System, which received Honorable Mention in the 1999 Software of the Year Competition and was a contributor to the Remote Agent System which was a co-winner in the same 1999 competition. In 2000, he received the NASA Exceptional Service Medal for service and leadership in research and deployment of planning and scheduling systems for NASA.