Intelligent Hybrid Control for Free-Floating Space Robots: PSO–RL–SMC With Inverse-Dynamics Tracking

Free-floating space robots must capture and manipulate targets without attitude anchoring, which makes precise motion control difficult due to base–arm coupling, uncertain dynamics, and actuator limits. This work presents an integrated pipeline that addresses these challenges by combining particle swarm optimization to select a low-coupling initial posture, a reinforcement-learning policy to generate smooth approach motions, and a sliding-mode tracker mapped through free-floating inverse dynamics for robust execution. The method was evaluated on a seven-joint system with realistic inertial properties and viscous effects. Across complete capture runs, the controller achieved an overall root-mean-square tracking error of $5.9133\times 10^{-3}$ rad, which corresponds to approximately 0.3389°, an overall steady-state mean-squared error of $4.0954\times 10^{-6}$ rad2 (about 0.116° root-mean-square in the final window),and the torque norm bounded with a peak of approximately 40 N $\cdot $ m, demonstrating efficient and stable actuation. These results indicate consistent sub-degree accuracy, stable approach behavior, and effective suppress of base reactions, demonstrating that the proposed PSO–RL–SMC framework is a practical and high-precision solution for on-orbit target capture.