Solution of Complex Pursuit-Evasion Game Policy Based on Multidimensional Gradient Descent Adaptive Dynamic Programming

This paper proposes an adaptive dynamic programming (ADP) algorithm based on multidimensional gradient descent, which mainly solves the problem of pursuit-evasion games with complex objectives. In complex pursuit-evasion games, there is usually agents that complete both pursuit and evasion actions simultaneously. In order to describe the goal of each agent, the objectives are discussed separately, and the overall value function of the agents are reorganized to obtain the policy expression of each agent. Based on the actual model of the pursuit-evasion problem, we proposed a gradient-based online ADP method, which is used to approximate the term of value function. The necessary conditions for convergence of the algorithm are derived using the Lyapunov function method. Simulation experiments verify the feasibility of the algorithm in solving pursuit-evasion game problems of systems with complex objectives.