Reinforcement learning-based boundary control for attitude tracking and vibration suppression in flexible satellites

Abstract

This study presents a boundary control strategy that addresses actuator constraints to achieve simultaneous attitude tracking and vibration suppression in a flexible satellite equipped with honeycomb-structured solar arrays. To enhance adaptability and autonomy, a reinforcement learning-based mechanism is employed to automatically tune the controller gains in real time. The approach leverages temporal-difference learning for online, model-free control, with a radial basis function (RBF) neural network organized in a critic–actor architecture to approximate the value function and control policy dynamically. The effectiveness of the proposed method is validated through numerical simulations across various scenarios, including nominal attitude tracking, response to external disturbances, and tracking a new desired trajectory under system uncertainties. Furthermore, the reinforcement learning algorithm is applied to a benchmark satellite dynamic model and control architecture adopted from earlier studies. Beyond facilitating autonomous gain tuning, the proposed approach demonstrates significant enhancements in control performance relative to the outcomes of the original framework. A key contribution of this work is the improvement of controller robustness against disturbances and uncertainties through intelligent, automatic gain adaptation. These findings highlight the potential of combining boundary control with reinforcement learning to enhance performance and resilience in flexible space structures. © The Author(s) 2025