Reinforcement learning cooperative control of nonlinear multi-agent systems with quantitative prescribed performance

In this article, a distributed reinforcement learning-based adaptive fault-tolerant optimal consensus tracking control scheme is developed for uncertain multi-agent systems. These systems subject to quantised inputs and actuator faults, with the aim of achieving quantitative prescribed performance under directed communication topologies. First, a continuously differentiable quantizer is designed to effectively address the problems of quantisation gaps and reduce communication load. Then we propose a quantitative prescribed performance control (QPPC) strategy that embeds the quantised tracking error into the performance boundary. This enables the performance constraint boundary to be dynamically adjusted in response to the dynamic variations of the quantised tracking error. Furthermore, a fuzzy logic system (FLS)-based reinforcement learning framework is embedded into the backstepping control design. Within this framework, an actor-critic architecture is employed to learn the optimal virtual and actual control policies online by solving the Hamilton-Jacobi-Bellman equations. Meanwhile, by employing a bound estimation approach to estimate the fault parameter bounds, the proposed scheme maintains satisfactory control accuracy even in faulty scenarios. Finally, simulation results validated the applicability of the proposed scheme.