Time-Synchronized Control for Spacecraft Rendezvous and Docking With Reinforcement Learning-Based Optimization

This article presents a reinforcement learning (RL)-based time-synchronized optimal control approach for spacecraft rendezvous and docking (RVD) operations. The proposed method builds on a fixed-time-synchronized backstepping structure to ensure simultaneous convergence of translational and rotational errors, while an actor–critic RL mechanism is embedded to optimize the virtual and actual control inputs online. Here, “time-synchronized convergence” specifically refers to the simultaneous convergence of the six-degree-of-freedom state errors, including position and attitude errors, ensuring accurate coordination required in the RVD process. The algorithm comprises two phases. The first phase uses an RL algorithm to generate optimal virtual controls under prescribed performance constraints, ensuring collision avoidance and operational safety. The second phase computes optimal control inputs to track the virtual controls, thereby ensuring energy-efficient and precise docking. Rigorous proofs demonstrate that the control strategy guarantees synchronized convergence of system state errors under Bellman optimality. Compared with traditional methods, the proposed approach improves control performance and energy efficiency, offering a robust and effective solution for autonomous RVD operations.