Kinematic Control of Redundant Robot Arms Using Neural Networks

10Zeroing Neural Networks for Robot Arm Motion Generation

10.1 Introduction

The model for solving nonlinear equations bears an essential similarity with the controller for controlling the plant: their residual errors are required to decrease to an acceptable small value as soon as possible. The exploitation of this similarity provides a possibility to investigate computational methods from the perspective of control system theory. It is worth mentioning that, in the field of numerical computation and control, there is always a great demand for robustness due to the existence of various noises or disturbances, such as round‐off errors and truncation errors. From the perspective of computation, many recurrent neural network models, e.g. zeroing neural network (ZNN), are analyzed and applied to the solution of various problems [128]. To improve the robustness for solving time‐varying problems, a noise‐tolerant zeroing neural network (NTZNN) design formula is proposed in [13], which can be used to design recurrent neural networks from the viewpoint of control. Then, such a NTZNN design method is explored to design a modified NTZNN in [1] for the online solution of quadratic programming with application to the repetitive motion planning of a redundant robot arm. As discussed in [7], nonlinear activation functions can be used to accelerate the convergence speed of original ZNN models. However, to the best of the authors' knowledge, there is no systematic solution on NTZNN with the ...