Multiagent Autonomous Reasoning System (MAARS) for Satellite Defense

In this document, we propose to carry out the research and development required to establish a distributed multi-agent system that will provide planning and execution capabilities that is aimed at protecting U.S. space capabilities against offensive space technologies. The uncertainty in detection and potential impact of offensives and their corresponding defenses, communication/time limitations and the highly sensitive nature of the decisions to be made are some of the major challenges of this application domain. To address these issues, our approach is based on two main components: Team’s Reshaping of Models for Rapid Execution (TREMOR) planning algorithm that is based on Distributed Partially Observable Markov Decision Processes (Dis-POMDPs) and also Adjustable Autonomy (AA). University of Southern California (USC)’s TeamCore research group has extensive experience with distributed multi-agent planning algorithms and Perceptronics has been developing systems that apply these algorithms and many other planning and execution approaches in a wide variety of domains with varying levels of autonomy and human intervention. We are confident that uncertainty handling in TREMOR and AA’s elegant handling of human intervention, along with the addition of team level plans and sensitivity analysis of potential defense actions, will be able to address all characteristics of our application domain.