Reinforcement learning applications to combat identification

Abstract

Crucial to the safe and effective operation of U.S. Navy vessels is the quick and accurate identification of aircraft in the vicinity. Modern technology and computer-aided decision-making tools provide an alternative to dated methods of combat identification. By utilizing the Soar Cognitive Architecture's reinforcement learning capabilities in conjunction with combat identification techniques, this thesis explores the potential for the collaboration of two. After developing a basic interface between Soar and combat identification methods, this thesis analyzes the overall correctness of the developed Soar agent to established truths in an effort to ascertain the level of system learning. While the scope of this initial research is limited, the results are favorable to a dramatic modernization of combat identification. In addition to establishing proof of concept, these findings can aid future research to develop a robust system that can mimic and/or aid the decision-making abilities of a human operator. While this research does focus on a sea-based, naval, application, the findings can also be expanded to DOD-wide implementations.