Intelligent Spacecraft Attitude Fault Recovery Using Deep Reinforcement Learning

Reliable satellite attitude fault recovery is essential for the success of space missions, particularly as satellites increasingly operate autonomously in dynamic and uncertain environments. Such uncertainties may arise from actuator degradations, subsystem anomalies, environmental disturbances, or other unpredictable operational conditions. Maintaining control resilience in these scenarios is critical to preserving mission objectives and ensuring system stability. However, traditional Proportional-Derivative (PD) controllers and existing deep reinforcement learning (DRL) algorithms such as TD3, PPO, and A2C often fall short in providing the real-time adaptability and robustness required for autonomous satellite operations. This study introduces a DRL-based fault recovery strategy designed for satellite resilience and adaptability under uncertainty. In contrast to previous DRL-based solutions that focused primarily on nominal operations or minimal noises, this work introduces a new method tailored for fault recovery. The proposed method, referred to as TD3-HD, employs Twin Delayed Deep Deterministic Policy Gradient (TD3) enhanced with Hindsight Experience Replay (HER) and Dimension-Wise Clipping (DWC) to accelerate learning under sparse post-fault rewards and stabilizes policy updates in degraded scenarios. This combination forms a distinct learning mechanism designed specifically for resilience and adaptability in the presence of actuator faults and structural variations. Experimental results demonstrate that TD3-HD achieves significantly lower attitude error, improved angular velocity regulation, and enhanced stability in the presence of uncertainty, highlighting its potential as a robust fault recovery framework for future autonomous satellite missions.