Fault-tolerant Control for Unmanned Aerial Vehicle Using Deep Reinforcement Learning

This paper develops a fault-tolerant control strategy for quadrotor unmanned aerial vehicles using deep reinforcement learning. To Address the issue of actuator failures, which can degrade trajectory tracking and system stability, we propose a novel fault-tolerant control framework. This framework integrates neural networks to approximate the dynamics of a quadrotor through offline data, generating residuals containing fault information for reinforcement learning training. An improved Deep Deterministic Policy Gradient algorithm is employed to accelerate the learning process and improve fault-tolerant capability, enhanced by Prioritized Experience Replay to assign training priorities based on residual values. The approach significantly improves the robustness and adaptability of the controller despite actuator faults. Finally, simulation results demonstrate that the proposed method maintains tracking performance effectively even under actuator faults.