Neural Learning-Based Adaptive Force Tracking Control for Robots With Finite-Time Prescribed Performance Under Varying Environments

Abstract

Endowing robots with the ability to maintain precise interaction force is critical for performing force control tasks in dynamic environments characterized by unknown and varying stiffness and geometry, such as aircraft wing skins and other thin, soft materials. This article presents an adaptive force-tracking admittance controller (AFTAC), ensuring tracking performance through the meticulous design of both the force-based outer loop and the position-based inner loop. First, a finite-time controller based on the adaptive neural networks (NNs) and prescribed performance function (PPF) is proposed to improve the transient convergence speed and steady-state accuracy of robot position tracking. The estimation of the robot dynamics model is based on the idea of single-parameter, which reduces the computational complexity of the entire process. Then, an adaptive force controller combining the admittance model and a disturbance observer (DOB) is designed to compensate for both structural and nonstructural uncertainties of the interaction environment, thereby enhancing force-tracking accuracy. Furthermore, the uniform convergence of the system is proved by Lyapunov’s method. Finally, a series of experiments are conducted on the ROKAE platform of a multidegree-of-freedom robot, and the results show the effectiveness of the proposed framework.