Deep Reinforcement Learning (DRL) Enabled Warfighter Assistant

The U.S. missile defense system (MDS) includes various assets located at sea, on land, and in space that provide global coverage against missile threats to the U.S. and its allies. Elements of the MDS create a layered defense capability providing detection, tracking, discrimination, and intercept capabilities against missile threats. As the MDS and its elements continue to grow and evolve to keep pace with our adversaries, the complexity of the system has increased. Deep Reinforcement Learning (DRL) has demonstrated the ability to create agents capable of making decisions in these complicated scenarios however the motivation behind these decisions is difficult to discern, in this effort Toyon Research Corporation and the University of Colorado Boulder propose to develop techniques to understand and visualize these motivations through outcome visualization and counterfactuals. The resulting work will be integrated with analyst and/or warfighter interfaces using prescriptive and descriptive visual guidance. Approved for Public Release | 22-MDA-11339 (13 Dec 22)