Developing an Adaptive Opponent for Tactical Training

摘要

This paper describes an effort to create adaptive opponents for simulation-based air combat, where the opponents behave realistically while at the same time fulfilling instructional objectives. Three different models are developed to control the behavior of red pilots against (simulated) blue trainees in a set of 2v2 scenarios. These models are then evaluated on their tactical and instructional performance, with the machine-learning model performing on par with the two hand-constructed models. The contribution of this paper is to investigate technology and infrastructure enhancements that could be made to existing systems used for simulation-based air combat training.