Reinforcement Learning‐Based Consensus Control for Unknown Nonlinear Multi‐Agent Systems With Sensor Faults

ABSTRACT This paper introduces a fault‐tolerant control approach for a class of unknown affine nonlinear multi‐agent systems using a reinforcement learning (RL) algorithm. In this multi‐agent system, all agents aim to follow a leader and converge to the same position with unknown dynamics. However, sensor faults result in inaccurate measurement outputs, creating challenges in achieving an optimal consensus. To address these challenges, we first design an observer to estimate the states and mitigate the effects of sensor faults by using the error between the actual states and their estimates. Subsequently, we propose a learning‐based control approach to compensate for the effects of sensor faults in the presence of unknown dynamics. The learning approach employs an actor‐critic structure to handle these challenges and ensure consensus among all agents. Furthermore, stability is proved by a theorem that guarantees the consensus of agents in the presence of faults and unknown multi‐agent systems. The RL method presented in this paper demonstrates that system performance is maintained despite the presence of faults, as validated through numerical simulations.