Fuzzy Reinforcement Learning-Based Safe Cooperative Control for Nonlinear Multiagent Systems

This article investigates safe cooperative control problem for nonlinear multiagent systems (MASs) with composite constraints and multiactuator faults. Specifically, the composite constraints primarily consist of state constraints and terminal time constraints. For the former, state-mapping functions and universal barrier functions are introduced to address asymmetric dynamic delayed constraints, which are further enhanced through state transformation to create new state variables. This approach eliminates the limitations imposed by initial system conditions and the influence of unknown terms, while accommodating both unconstrained and constrained scenarios. The latter introduces terminal time constraints on the basis of the former framework to rapidly satisfy system performance requirements. Furthermore, neural networks (NNs) are employed to approximate the reconstructed multiple fault information and system uncertainties. Then, an actor-critic-identifier architecture based on fuzzy reinforcement learning (RL) is constructed via an optimal backstepping (OB) method, enabling the solution of the Hamilton–Jacobi–Bellman equation within each subsystem without the need for persistent excitation. Finally, by integrating Lyapunov stability theory with graph theory, it is proven that all signals are bounded, and the followers ultimately achieve dynamic consensus with the leader. Simulation results are provided to demonstrate the effectiveness of this control strategy.