Developing Combat Behavior through Reinforcement Learning in Wargames and Simulations

摘要

Progress in artificial intelligence (AI), particularly deep reinforcement learning (RL), has produced systems capable of performing at or above a professional-human level. This research explored the ability of RL to train AI agents to achieve optimal offensive behavior in small tactical engagements. Agents were trained in a simple, aggregate-level military constructive simulation with behaviors validated with the tactical principles of mass and economy of force. Results showed the combat model and RL algorithm applied had the largest impact on training performance. Additionally, specific training hyper-parameters also contributed to the quality and type of observed behaviors. Future work will seek to validate RL performance in larger and more complex combat scenarios.