Reinforcement Learning control for Free-Floating Space Manipulator: Truncated Quantiles Critics

With the increasing complexity of space missions, the free-floating space manipulator (FFSM) systems have become indispensable for various space applications. This paper proposes a reinforcement learning (RL)-based control framework leveraging the truncated quantile critics (TQC) algorithm to simultaneously achieve: (i) precise end-effector (EE) position and orientation tracking, (ii) effective base spacecraft drift suppression, and (iii) collision-free operation maintenance. Additionally, we introduce a novel time differential reward function for controller training to enhance multi-objective learning efficiency. The proposed framework is validated through numerical simulations and Monte Carlo Experiments using a space station's manipulator system, with comparisons against two baseline RL approaches demonstrating its superior performance.