Reinforcement Learning for Predefined‐Time Optimal Tracking Control of Robot Manipulator

Abstract

This work presents an optimal predefined‐time tracking control framework for a nonlinear robot manipulator subject to exogenous disturbances. To guarantee that the tracking errors converge within a user‐specified time bound, regardless of the initial conditions, a predefined‐time sliding mode control (PT‐SMC) technique is devised. This feature guarantees strict temporal performance for safety‐ and mission‐critical applications. An actor‐critic reinforcement learning (RL) architecture is seamlessly integrated with the PT‐SMC technique to achieve optimal torque utilization. A rigorous Lyapunov‐based stability analysis is also provided to establish the predefined‐time convergence and the overall closed‐loop stability. Simulation studies conducted on a 2‐DOF robot manipulator demonstrate superior tracking accuracy, reduced control effort, actuator‐friendly control action, and deterministic convergence time compared with several state‐of‐the‐art controllers. Notably, the proposed controller improves the settling time by more than threefold. Overall, the proposed approach offers a cohesive framework that blends RL‐based optimal tracking control with predefined‐time robustness, making it well‐suited for Industry 5.0: human‐centric robotic systems.