Air Combat Strategies Generation of CGF Based on MADDPG and Reward Shaping

Abstract

The intelligence of the computer-generated force (CGF) is one of the important problems in air combat simulation. The air combat of CGF is modeled as a two-player zero-sum Markov game. An air combat strategies generation method of CGF is proposed to use a multi-agent deep deterministic policy gradient (MADDPG) algorithm. This paper proposes a potential-based reward shaping method to improve the efficiency of the air combat policy generation algorithm. Finally, the efficiency of the air combat policy generation algorithm and the intelligence level of the resulting policy is verified through simulation experiments. The simulation results show that this method has good convergence and better air combat performance with the strategy obtained by the DDPG algorithm.