Reinforcement learning-based prescribed performance tracking control for robotic manipulators*

Abstract

This paper proposes a fixed-time trajectory tracking control strategy for robotic manipulators facing dynamic uncertainties and input saturation, combining prescribed performance control with adaptive reinforcement learning. A prescribed performance function is designed to strictly constrain tracking errors within predefined transient and steady-state boundaries throughout system operation. An adaptive actorcritic framework with dynamic parameter adjustment is developed, where the actor network generates control policies and the critic network evaluates action-value functions in real time based on system states. An anti-saturation compensation law, derived from transformed error signals, is constructed to mitigate actuator saturation effects. Furthermore, an adaptive nonsingular terminal sliding mode controller is incorporated to ensure fixed-time convergence of tracking errors independent of initial conditions. Lyapunov-based stability analysis rigorously establishes the boundedness and convergence properties of the closed-loop system. Comprehensive simulation results demonstrate the effectiveness and performance advantages of the proposed control scheme.