Reinforcement Learning-based Fault-tolerant Attitude Control of Spacecraft under Actuator Failures

A fault-tolerant control method based on the Deep Deterministic Policy Gradient (DDPG) algorithm is proposed for spacecraft attitude control systems in the presence of actuator faults and external disturbances. Initially, a spacecraft dynamics model under actuator faults is established, and a finite-time sliding mode controller is designed to ensure that the sliding surface converges to zero within a finite time. Subsequently, to address the sensitivity of the finite-time sliding mode controller's performance to its parameters, reinforcement learning is employed to optimize the controller parameters. Finally, simulation experiments are conducted. The results demonstrate that the proposed reinforcement learning-based controller can achieve convergence of both the attitude quaternion and the angular velocity within a finite time.