Decentralized optimized finite-time backstepping control of large-scale high-order fully actuated strict-feedback nonlinear systems via reinforcement learning

Abstract

This paper studies the problem of finite-time optimized tracking control for large-scale high-order fully actuated strict-feedback nonlinear systems for the first time. Notably, reinforcement learning (RL)-based backstepping approach is employed for achieving the optimized control, such that the corresponding cost function is minimized and uncertain nonlinearities are allowed in the considered system. The control design incorporates a finite-time high-order Levant differentiator to estimate high-order derivatives, effectively solving the issue of “explosion of complexity”. Additionally, the finite-time error compensation signals are integrated to minimize filtering errors. The proposed scheme ensures that all signals of the closed-loop system are bounded and the tracking error can converge to a bounded neighborhood of the origin in a finite time.