A Novel Variable-Parameter Variable-Activation-Function Finite-Time Neural Network for Solving Joint-Angle Drift Issues of Redundant-Robot Manipulators

Abstract

This article presents a novel variable-parameter variable-activation-function finite-time neural network (VPA-FTNN) to deal with joint-angle drift issues of redundant-robotic arms. Different from most existing recurrent neural networks, VPA-FTNN establishes an error-based finite-time-convergence neural dynamics equation with variable-parameter and variable-activation-functions features so that it can effectively deal with the joint-angle drift of redundant-robotic arms with higher convergence speed and accuracy. It should be noted that VPA-FTNN can achieve finite-time convergence without relying on special activation functions. In order to verify the advantages of the proposed VPA-FTNN, it is compared with the varying-parameter convergent-differential neural network and traditional Zhang neural network when dealing with the joint-angle drift. Simulation and physical experiment results demonstrate the effectiveness and practicality of the proposed VPA-FTNN in solving the redundant robot joint-angle drift problem.