Pareto-optimal synchronization control of nonlinear multi-agent systems via integral reinforcement learning

This paper investigates a class of cooperative synchronization control problem of networked multi-agent systems (MASs) with unknown nonlinear dynamics. It is shown that the synchronization control problem of MASs can be formulated as the cooperative differential graphical game, in which the objective of each agent is to minimize the total team cost in a cooperative manner. The theoretical derivation and established conditions are then provided to achieve Pareto optimality for the cooperative differential graphical games. Furthermore, an adaptive recursive multi-agent reinforcement learning algorithm is proposed to learn a generalized optimal team value function and the associated decentralized Pareto-optimal strategies, where a Nash bargaining strategy and a recursive Bayesian inference scheme with different forgetting factors are respectively developed. Finally, two simulation examples are presented to verify effectiveness of the proposed learning-based synchronization control approach.