Strategy Inference in Multi-Agent Multi-Team Scenarios

Creating simulations for multi-agent multi-team interactions is a daunting task. It is non-trivial to compose a situation where each individual agent maintains their own 'personality' while still following the assigned policy dictated by a team's central command. Further, the complexity is inflated by ensuring that each of these agent policies is coordinated into a cohesive team strategy. Finally, peaking the complexity, is evaluating the performance of the team's strategy against other teams' strategies in real-time. This is the work of this paper, proposing SIMAMT, the simulation space for multi-agent multi-team engagements, and testing it. We will first cover the system and how well it models the virtual environment for strategic interaction. Second, we will deliver results from a practical test of strategy inference within such an environment using the SIE (Strategy Inference Engine).