Neural Network-Enhanced Super-Twisting Control for Industrial Robotic Manipulators via Time-Delay Estimation

Abstract

Industrial robotic manipulators are often affected by uncertain nonlinear dynamics and limited measurable information. This paper proposes a time-delay-estimation-assisted neural super-twisting sliding mode control framework for uncertain robotic manipulators. The time-delay estimation mechanism reconstructs hard-to-measure physical quantities and provides training data for neural uncertainty estimation. Based on these data, a neural network is developed to approximate the unknown nonlinear dynamics, and its output is incorporated into a super-twisting sliding mode controller to compensate for residual errors and reduce chattering. Lyapunov analysis proves the asymptotic stability of the closed-loop system. Simulations and real-time experiments on an industrial robotic manipulator verify the effectiveness of the proposed method.