Robust Policy Learning Control Design for Multiplayer Nonzero-Sum Games with Uncertainties
Ao Liu, Ding Wang, Menghua Li · 2023
This paper investigates robust control problems of nonlinear continuous-time multiplayer systems with infinite horizon by utilizing adaptive dynamic programming algorithms. Combined with pre-training, an improved policy iteration (PI) algorithm is developed to solve robust control issues of multiplayer nonzero-sum (NZS) games with actuator uncertainties. Pre-training of initial weights is added to the PI algorithm to relax the requirement of the initial admissible control policy. It implies that the admissible control pair can be obtained by pre-training of initial weights given randomly. Then, critic neural networks (NNs) are utilized to approximate the optimal control pair by applying the PI algorithm. Robust controllers can be obtained by modifying the optimal control pair. The algorithm accomplishes robust stabilization of multiplayer NZS games with uncertainties. Besides, initial weights of NNs can be set arbitrarily. Finally, a simulation example is given to demonstrate the effectiveness of the developed algorithm.