Reinforcement learning-based distributed cooperative output regulation for multi-agent systems

This article investigates distributed cooperative output regulation (COR) for nonlinear multi-agent systems (MASs) with unknown external disturbances. First, a fully distributed finite-time observer is designed to estimate the leader's state without global topology information. Then, a distributed COR strategy is developed by integrating a robust compensation mechanism with reinforcement learning (RL). Specifically, the compensation term handles nonlinear dynamics, while the RL-based optimal control term attenuates unknown disturbances. To implement the RL-based optimal control term, an actor-critic scheme is developed to approximate the optimal solution, avoiding the need to solve the Hamilton–Jacobi–Isaacs equation analytically. It is proven that all closed-loop signals are semi-globally uniformly ultimately bounded and that the regulated output errors enter a small neighbourhood of the origin within finite time. Finally, simulation results verify the effectiveness and advantages of the proposed method.