Abstract:
Increasingly complex space missions have motivated the development of autonomous command and control approaches which must handle high-dimensional, continuous observation and action spaces with hard-to-analyze behavior. Deep reinforcement learning (DRL) techniques are a rising area of research for dealing with such problems, but lack performance and safety guarantees which reduce their applicability for spacecraft operations. This work identifies and evaluates strategies for adapting DRL approaches to the spacecraft operations domain by synthesizing them with formal methods and adaptive verification approaches. Monte Carlo tree search (MCTS) methods are also explored, which can leverage shield agents as heuristic policies to find optimal and safe policies. Additionally, open-source spacecraft planning environments derived from the AVS Basilisk astrodynamics simulation package are presented and discussed. Results for operations sequences for a representative LEO Earth-observation mission and an Optical Navigation imaging campaign are presented to demonstrate the efficacy and promise of these techniques.
Speaker Bio:
Andrew Harris is a fifth year PhD student working in the Autonomous Vehicle Systems Laboratory at CU Boulder. From his early experiences developing low-cost small satellites for space debris monitoring at the University at Buffalo, Andrew's research interests have focused on enhancing small satellite operations and GN&C using AI and drag-assisted formation flight. His prior research experiences include NASA's Marshall Space Flight Center, the Air Force Research Laboratory, and the Jet Propulsion Laboratory.
Adam Herrmann is a second year PhD student and in the Autonomous Vehicle Systems Laboratory at CU Boulder. His research focuses on using Monte Carlo tree search methods for both Earth-observing and small body operations. Adam got his start in research at the University of Cincinnati, working on Earth-imaging CubeSat projects. He has also spent numerous semesters at the NASA Jet Propulsion Laboratory as a Mechanical Engineering Co-op supporting flight and research projects during his undergraduate degree.
Please let Marcel Llopis marcel.llopis@jpl.nasa.gov know if you have any questions about the talk.
-- Do not delete or change any of the following text. --
When it's time, join your WebEx meeting here.
Join meeting
More ways to join:
Join from the meeting link
https://jpl.webex.com/jpl/j.php?MTID=m4ab7e5a583b52562bd3f6efe187ed7e8
Join by meeting number
Meeting number (access code): 199 627 1432
Meeting password: ifPTRiNq465
Join by phone
+1-510-210-8882 USA Toll
Global call-in numbers
If you are a host, click here to view host information.
Need help? Go to https://help.webex.com
