A Novel Model-Free DZNN-QN Algorithm for Redundant Manipulator Control

Redundant manipulators are widely used in industries like healthcare and manufacturing, but challenges such as manufacturing errors and mechanical wear hinder the accurate modeling required formodel-basedmotion control. To address this problem, a model-free motion control method is proposed, where the Jacobian matrix is estimated using a Quasi-Newton (QN) method (QN Jacobian Matrix Estimator, QNJME). A position error function is introduced to develop a continuous zeroing neural network (CZNN) solver, which is discretized using the Euler forward method to create the discrete zeroing neural network (DZNN) solver. Unlike traditional neural network methods, this article combines the DZNN solver and QNJME to propose the DZNN-QN algorithm, a high-precision and model-free motion control method. Its global asymptotic convergence and stability are proven through rigorous mathematical analysis. Simulative and physical experiments with the Kinova and UR5 manipulators show that the proposed method outperforms existing methods, particularly in terms of effectiveness, robustness, and practicality.