Off-Policy Reinforcement Learning for Partially Unknown Nonzero-Sum Games

摘要

In this paper, the optimal control problem of nonzero-sum (NZS) games with partially unknown dynamics is investigated. The off-policy reinforcement learning (RL) method is proposed to approximate the solution of the coupled Hamilton-Jacobi (HJ) equations. A single critic network structure for each player is constructed using neural network (NN) technique. To improve the applicability of the off-policy RL method, the tuning laws of critic weights are designed based on the offline learning and online learning methods, respectively. The simulation study demonstrates the effectiveness of the proposed algorithms.