The Research Of Aircraft Pursuit-Evasion Game Based on Improved DQN

Abstract

The intelligent combat of aircraft has become an important research area. This paper studies the 1Vs1 pursuit-evasion game confrontation scenario of the aircraft, aiming to enable the red agent to find the optimal strategy to avoid the blue interception. First, the aircraft is abstracted into an agent, the kinematics model of the aircraft is simplified, and the behavior model of the aircraft with the bait throwing behavior is established. Based on the improved DQN algorithm, with relative distance, line-of-sight angle and azimuth as the state; the reward function design comprehensively considers the time, number of bait throwing, maneuvering time, and miss distance, and proposes an aircraft pursuit-evasion game algorithm (RB-DQN) in a three-dimensional environment. Experimental results show that the algorithm can effectively control the aircraft to complete the escape mission, and has satisfactory generalization.