IRL Method for Time-Continuous Two-Player Nonzero Sum Game of Unknown System with Constrained-Input

Abstract

This paper addresses an off-policy method for solving the optimal control problem of the two-player nonzero sum (NZS) game when the inputs are constrained. Benefit from integral reinforcement learning (IRL) technology, proposed method can achieve the optimal solutions when the system dynamics can hardly obtained in practical applications. The equivalent convergency proof of the proposed method is also figured out. Actor-critic framework is constructed by utilizing the neural networks (NNs), in detailed the actor NNs are employed to approach the optimal control meanwhile the critic NNs are employed to approach the value functions. At last, a simulation is supplied to show the accuracy and effectiveness of the method we proposed.