Adaptive Predefined Performance Control Using Reinforcement Learning for a Single-DOF Manipulator

Le résumé fourni par la source

Predefined performance trajectory tracking control is widely recognized in both academic and engineering fields. However, the design of predefined performance controllers is a complex and model-dependent process. In this work, a predefined performance control with reinforcement learning adaptive law is presented for a one DoF manipulator system. The control law is constructed using a simplified backstepping approach that eliminates the need for virtual control variables, thereby enhancing implementation feasibility. A soft actor-critic reinforcement learning agent is employed to replace the nonlinear error transformation. The actor network generates the adaptive parameters based on system state errors, while the critic network evaluates actions using system outputs. Additionally, a sparse reward strategy is designed to improve the efficiency of policy learning. Simulation results demonstrate that the proposed method achieves accurate trajectory tracking while satisfying predefined performance specifications, showcasing its potential for real-world robotic applications.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.