Nash Equilibrium Computation for Spacecraft Pursuit-evasion with Deep Reinforcement Learning Method

The Nash equilibrium of a pursuit-evasion game involving spacecraft is explored in this paper. Deep reinforcement learning (DRL) algorithm plays an important role in obtaining Nash equilibrium. A learning strategy utilizing the soft actor-critic (SAC) algorithm is proposed. Both the pursuer and evader are trained to improve their game strategies by adversarial learning. Firstly, the dynamic model and the process of pursuit-evasion are built to describe this game. Then, the reward function and structure of actor-critic (AC) network in SAC algorithm are designed, and the two-sided training framework is constructed. Simulation experiments verify the enhanced performance and efficacy of the learning method.