Reinforcement Learning-Based Active Disturbance Rejection Attitude Control for Flexible Spacecraft

To address attitude control challenges of flexible spacecraft equipped with piezoelectric patches, this paper proposes a reinforcement learning-enhanced linear active disturbance rejection control (LADRC) framework. The key innovation lies in a deep deterministic policy gradient (DDPG) algorithm being emloyed to adaptively tune LADRC parameters, eliminating the need for exact dynamic models while enabling simultaneous vibration suppression and precise attitude control. Firstly, a model of the flexible spacecraft with piezoelectric patches is developed and analyzed. The LADRC is then designed to manage the model's nonlinearity and uncertainty, enabling spacecraft attitude control. Secondly, the DDPG algorithm is employed to optimize the parameters of LADRC. An integral term is introduced into the reward function, and an early stopping criterion is added to enhance parameter optimization. Finally, simulation results demonstrate that the proposed framework achieves adaptive attitude control under dynamic disturbances without requiring prior system knowledge, and active vibration suppression of flexible appendages via real-time piezoelectric excitation. By integrating model-independent LADRC with data-driven deep reinforcement learning, this work provides a solution for coupled attitude-vibration control in flexible spacecraft without requiring precise dynamic models.