Gaussian-disturbance-aware deep reinforcement learning for event-triggered MPC in satellite attitude reorientation

Satellite attitude reorientation is a critical maneuvering task in space missions. However, the presence of strong nonlinearity, high coupling, and external disturbances poses significant challenges for conventional control approaches. Traditional methods often struggle to balance control accuracy and computational efficiency without compromising robustness. To address these challenges in a systematic manner, this paper proposes a Gaussian-disturbance-aware deep reinforcement learning (DRL) framework for event-triggered model predictive control (ET-MPC) in satellite attitude reorientation, consisting of two key components. First, to handle system nonlinearity while improving resource efficiency, a DRL-based agent learns a MPC triggering policy. In particular, a structured state space model (SSM) is incorporated into the DRL architecture to enhance temporal dependency modeling and policy learning capability. Second, to cope with Gaussian disturbance encountered during on-orbit operations, a Kalman filter (KF) is embedded within the framework. By filtering Gaussian noise in the angular velocity measurements, the KF provides the DRL triggering policy and MPC optimizer with more reliable information, thereby improving closed-loop performance under the measurement-noise conditions considered in this study. Simulation results demonstrate that the proposed framework achieves accurate attitude reorientation while reducing unnecessary MPC updates compared with baseline methods. Moreover, the framework exhibits improved robustness in disturbed environments, indicating its potential for computationally efficient satellite attitude reorientation.