Flight Director In A Box: Using Learning to Develop Planning Agents For Exploration

TRACLabs has developed a procedure integrated development environment called PRIDE that is currently being used by NASA for ISS and Orion procedures. As these procedures execute, they log all the pertinent execution information. Automated planners, developed by NASA over the years will be used to sequence these procedures to achieve goals during exploration flights or for planetary base operations. TRAClabs has successfully integrated planning and procedure execution, but humans manually code the most likely sequence of activities for the planner to use, a practice that is sub optimal and fraught with error. As flight directors and their flight control teams train for exploration using PRIDE in simulated environments, the procedure sequences they develop for exploration can be captured along with all the execution data. TRACLabs proposes to use the resulting logs, along with archived telemetry, as a basis for applying apprenticeship learning to learn what procedures should be used in a given situation and in what order. Automated planners could then ingest these learned policies during explorations operations to yield plans as they would be generated from an experienced flight control team. Further, as the planner executes plans, we contend that the learner can continue to refine its policies during exploration missions.