Fault-Tolerant Optimal Control of Multi-UAV Formations based on Reinforcement Learning

This study proposes a fault-tolerant control strategy for multi-UAV formations to address actuator faults and external disturbances. The strategy integrates sliding mode (SM) control with neural networks (NN) to design an adaptive sliding mode controller that can compensate for lumped uncertainties, and the formation tracking control problem is further reformulated as an optimal control problem for the nominal system to achieve improved tracking performance. To solve this problem, an actor-critic reinforcement learning (RL) framework is employed to evaluate and optimize control performance efficiently. Compared to conventional RL methods, the proposed method simplifies the actor and critic RL updating laws, facilitates sufficient training of adaptive parameters to relax the persistence excitation condition. Lyapunov stability analysis confirms that the proposed control strategy can achieve the control objectives and ensure the desired performance. Simulation results verify the effectiveness and applicability of the strategy for multi-UAV formations.