Future space missions will rely heavily on automated multi-agent cyber-physical-human teams to perform a number of tasks, such as robotic servicing, habitat maintenance, health management, etc. In order for these multi-agents to become a reality, trust in them and uncertainty quantification will need to be factored into decision making policies by the system participants. Even if one assumes perfect information the problem is NP-Hard, thus timely optimal solutions are unachievable. The central objective of the proposed technology is to provide near optimal mission planning for autonomous multi-agents with uncertain and possibly untrustworthy data sources using a hybrid deep reinforcement learning-optimization approach. A new architecture will be developed under this optimization approach that will allow for consideration of uncertainty and trustworthiness in planning decisions. A use-case that incorporates realistic uncertainties will be employed to provide metrics on the proposed approach. The use-case involves multiple satellites working in coordination to provide vital information to ground agents, each with a task of resupplying outlying bases. Past results by the investigators provide a basis-for-optimism that the proposed approach is viable.
The Phase I effort will focus on extensive simulation studies and analyses. This will build a foundation to develop benchmark testing at the onset of Phase II, with the end of this work being a fully functional demonstration unit.
NASA’s Cyber-Physical Systems Modeling and Analysis Initiative was developed to support future space exploration missions. Autonomous multi-agent cyber-physical-human teams will be a vital aspect of this initiative. Past realizations to minimize the impact of costly validation and verification (V&V) processes resulted in minimally scoped autonomous operations. Current V&V processes for autonomous space operations will clearly require a much high level of trustworthiness than ever before, which will be provided by the proposed technology.
An obvious example involves the intelligent manufacturing sector, which integrates information technology and manufacturing technology. A large sector that has led to a disruptive impact is the driverless automobile sector. Both sectors will rely heavily on multi-agent cyber-physical-human teams. The proposed technology is easily extendable to these and other non-NASA sector applications.