A Novel Method Combining Leader-Following Control and Reinforcement Learning for Pursuit Evasion Games of Multi-Agent Systems

In the existing literature, many methods have been utilized in solving the pursuit evasion game. However, most of these methods have one thing in common: all pursuers have to know the evader's position. This paper presents a method combining the multi-agent leader-following control and reinforcement learning, which aims at addressing a pursuit evasion game only when partial pursuers have the knowledge of the position of the evader who moves randomly in two-dimensional grid space. Through decomposing the team task, pursuers realize the encirclement of the evader. Simulation experiment validates the effectiveness of our method.