Using AI in Wargaming Simulation as a Multi-Domain Decision Support Tool

Command & Control (C2) planning presents an increasingly complex challenge, such as the growing availability of relevant data and how to process this in useable timeframes. By understanding the likely Red Force responses to a potential Blue Force Course of Action (CoA), planners are empowered to make better strategic decisions using insights from Artificial Intelligence (AI) wargaming. Modelling and Simulation (M&S) tools, in combination with AI, can rapidly predict Red Force CoA by consuming and processing observational data of the operational picture. Red Force Response (RFR) is presented as a decision support tool that exploits AI within a wargaming simulator to find potential Red Force CoAs. Using state-of-the-art Deep Neural Network (DNN) algorithms, including Proximal Policy Optimisation (PPO) and Curiosity Learning, integrated into a Multi-Agent Reinforcement Learning (MARL) environment, the RFR agent finds both high-performing and novel CoAs based on the reward and action selection diversity respectively. A 91% Red Force win probability was achieved in a tactical air scenario when trained for 17,587 episodes against a superior Blue Force. The concept demonstrates an effective use of AI for C2 planning, how cloud computing can be used to effectively train agents, and how the concept could scale to larger problems.