Air Combat Stratigies for an Unmanned Fighter using Q-learning

Abstract

Autonomous air combat maneuvers for an unmanned combat aircraft have been accomplished using an intelligent learning approach. As interacting with an opponent, the unmanned fighter learns about the opponent action model by a reinforcement learning algorithm, Q-learning. The air combat problem is formulated using combat states and kinematic equations of motion of all vehicles. The bogy aircraft is assumed to follow a pure proportional navigation guidance (PPNG) to attack the unmanned fighter. One-step Q-learning algorithm is employed and states and action spaces are quantized. A reward function has been constructed for the unmanned fighter to learn wining strategies during the engagement. A numerical simulation has been performed to evaluate the proposed approach and the results showed that the unmanned fighter wins the combat with the optimized action value function.