Building Adaptive Opponents for Pilot Training

This project is focused on using artificial intelligence and machine learning techniques to create adaptive agents for simulation-based air combat. This effort is part of a larger program at the Air Force Research Laboratory aimed at creating smart and agile opponents that respond in a realistic manner to unfolding events.  Our approach leverages two specific technologies to create the behavior models that drive the adaptive opponents. The first is SimBionic, an open-source tool for visually representing knowledge that supports quick, visual, and intuitive knowledge representation by drawing and configuring behavioral transition networks. The second is Dynamic Scripting, an online reinforcement learning algorithm. The developed agents have been evaluated on their tactical performance, controlling the behavior of opposing pilots in a variety of 2v2 air-to-air scenarios.