Observer-Based Adaptive Robust Actor–Critic Learning Saturated <i>PID</i> Controller for a Class of Euler–Lagrange Robotic Systems With Guaranteed Performance : Theory and Practice

Elhaki, Omid and Shojaei, Khoshnam and Chatraei, Abbas and Montazeri, Allahyar (2025) Observer-Based Adaptive Robust Actor–Critic Learning Saturated <i>PID</i> Controller for a Class of Euler–Lagrange Robotic Systems With Guaranteed Performance : Theory and Practice. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 55 (2). pp. 1400-1412. ISSN 2168-2216

Text (bare_jrnl)
bare_jrnl.pdf - Accepted Version
Available under License Creative Commons Attribution.
Download (2MB)

Abstract

This article addresses the output-feedback reinforcement learning (RL)-based saturated proportional-integral-derivative (PID) control design for fully actuated Euler–Lagrange (EL) systems which are uncertain subject to actuator saturation with prescribed performance. It is assumed that the actuator input nonlinearity, uncertain nonlinearities and unmeasurable external disturbances have a significant impact on the system. The presence of actuator saturation and complex uncertainties may inevitably give rise to the breakdown of the EL control system. The lack of prior knowledge of the system dynamics renders the presented technique to achieve a robust prescribed tracking performance without using velocity sensors. To conquer mentioned obstacles, a novel RL saturated PID controller, which is not dependent on the system’s dynamics and only requires measurable output signals is designed via actor–critic structure to deeply estimate and compensate complex unknowns. An adaptive robust controller is used to reduce external disturbances effects adaptively. The prescribed performance funnel control way is considered to guarantee predetermined output constraints. The high-gain observer (HGO) is used to estimate velocities and derivatives free of system dynamics, and generalized saturation functions are utilized to efficiently decrease actuator saturation danger. It is proved that suggested technique ensures a robust prescribed performance with input constraints in the absence of velocity sensors and the existence of considerable complicated model uncertainties. A semi-global uniform ultimate boundedness (SGUUB) stability for tracking deviation errors and state estimation deviation is ensured through a Lyapunov stability study. Finally, experimental results on a real robotic arm is carried out to further demonstrate the effectiveness of all theoretical findings.