Space Vehicle Inspection - Safe Multiagent Coordination Using RL

This proposal targets the USSF Space Superiority, Space Mobility and Logistics, and Space Domain Awareness technology focus areas. The USSF needs to quickly react, plan, and decide on appropriate course of actions (COAs) in Rendezvous and Proximity Operations (RPO) for cooperative and non-cooperative targets in contested space. Every branch of the military and the intelligence community depend on USSF assets to provide data and secure global communications to accomplish their missions. If safety is not a consideration early on, an autonomous system could make a bad decision that cost mission success, the lives of humans on the ground that need timely data and communications, and billions of dollars in capability value. If Guardian preferences are not incorporated into the autonomy at both design time and run time, the operators may not trust autonomy to take control in time-critical situations Our primary focus in this project is on prototyping AI and Reinforcement Learning tools to cater to an asymmetric autonomous space fleet to augment the existing manned fleet through ongoing USSF programs such as the Infrastructure for Trusted Satellite Autonomy and Tactical Rapid Anomaly Protection (IT’S A TRAP) which have an emphasis on Next Generation Space Dominance to support our Guardians. Our key USSF goals includes the inspection problem for on-orbit servicing (refueling/fixing) assembly, manufacturing (OSAM) of a satellite and training it to work in "worst case scenarios" (i.e. a satellite that happens to be out of control in the worst possible way to make it nearly impossible to inspect).