Optimal consensus control of nonlinear multi-agent systems by data-driven optimistic policy iteration

Abstract

This paper investigates the optimal consensus control of nonlinear multi-agent systems with completely unknown dynamics. To overcome the difficulties in solving the coupled Hamilton–Jacobi–Bellman equations for the optimal control, a data-driven optimistic policy iteration algorithm is proposed based on adaptive dynamic programming techniques, which generalizes the classical value and policy iteration algorithms. By approximating the performance function and control policy through critic and actor networks and considering the inevitable neural network approximation errors, we give the rigorous optimal consensus convergence analyses. The efficacy of our method is validated through two simulation examples, highlighting its superior convergence performance over the typical value iteration algorithm.