Reinforcement Learning-Enhanced Fixed-Time Control with Disturbance Observer for Saturated Trajectory Tracking of Space Manipulators

This study proposes a reinforcement learningenhanced control strategy for saturated fixed-time trajectory tracking of space manipulators. The developed approach integrates a Critic-Actor neural network framework with a high-gain disturbance observer (DOB) within a collaborative RL-DOB architecture, enabling real-time lumped disturbance estimation and compensation. An integralaugmented fixed-time sliding surface is designed by incorporating a nonsingular terminal attractor and an antiwindup compensator, which collectively guarantee fixedtime convergence of tracking errors, suppress chattering phenomena, and reduce energy consumption. Numerical simulations conducted on a two-degree-of-freedom space manipulator demonstrate that the proposed methodology outperforms conventional RL-based saturated time trajectory tracking control in terms of tracking precision, torque smoothness, and energy efficiency. The closed-loop system stability is rigorously proven through Lyapunovbased analysis.