Optimal Control Design of Multi-Agent Pursuit-Evasion Game with Malicious Pursuer Environment

This paper proposes optimal control policies for multiagent pursuit-evasion game problem under a malicious environment based on reinforcement learning method. Firstly, the healthy pursuers and malicious pursuers are established by using subjectively malicious factor in the cost function and establish a game framework to study the relationship between healthy pursuers, malicious pursuers, and evaders. Then the Hamilton-Jacobi (HJ) equations are used to gain the optimal control strategys. In addition, Nash equilibrium analysis is given. Then the adaptive weight adjustment of the critic-actor neural networks(NNs) are used to achieve the approximation of the control strategys and value functions. Finally, numerical simulation verifies the effectiveness of the proposed strategy.