Analyzing RL Agent Competency in Air Combat: A Tool for Comprehensive Performance Evaluation

Air combat represents a complex and dynamic domain in which human pilots face formidable challenges. Integration of artificial intelligence, particularly Reinforcement Learning (RL), has the potential to revolutionize the effectiveness of air combat operations. By leveraging RL techniques, autonomous agents can develop new tactics in response to changing battlefield conditions. In this study, air combat agents using advanced RL technique are trained, taking into account different initial combat geometries and relative positions. The results indicated a notable impact of the variation in air combat geometry on agent competency. To evaluate their competency and resilience, identical AI agents with symmetrical combat geometry were examined. Any deviations from the expected symmetrical results were detected, potentially signaling challenges encountered during the training exploration phase. When different agents were compared within this framework, their superiority was highlighted in specific air combat scenarios, providing valuable information to enhance the development of more intelligent agents.