基于强化学习方法的多智能体追逃博弈任务分配

This paper investigates the task assignment problem in multi-agent pursuit–evasion games under the influence of model nonlinear dynamics and external disturbances. An optimal task assignment value function is proposed that transforms the task allocation problem in games into a multiple pursuer trajectory tracking problem. An off-policy integral reinforcement learning method is proposed using input and output data from multi-agent systems. A neural network is introduced to fit the value function and derive the optimal strategy, and the nonlinear Hamilton–Jacobi–Bellman equation is solved iteratively. The optimal control policy and total cost of task execution are solved without the knowledge of the agent model. Simulation results verify the effectiveness of the proposed task assignment method.