Kinematic Control of Redundant Robot Arms Using Neural Networks

4Neural Learning and Control Co‐Design for Robot Arm Control

4.1 Introduction

The rapid advances of mechanics, electronics, computer engineering and control theory in recent years have significantly pushed forward the research on manipulators and gained great success in various industrial applications. Redundant manipulators are a special type of manipulators: they have more control degrees of freedom (DOFs) than task DOFs. Redundant manipulators have received intensive research focus for dexterous manipulation of complicated tasks.

A redundant manipulator provides a nonlinear mapping from its joint space to the Cartesian workspace. For the kinematic control of manipulators, it is desirable to find a control action in the joint space such that a reference motion in the workspace can be obtained. The solution to such a problem usually is not unique due to the redundancy and an optimal solution can be reached in terms of certain objective functions and a set of constraints. However, the nonlinearity of manipulators makes it difficult to directly solve the problem in the angle level with a satisfactory accuracy. Instead, most work considers this problem in the velocity space or acceleration space, where the mapping is converted to an affine function, and seek solutions in the new space. Benefiting from the affine nature of the manipulator kinematics in the velocity or acceleration space, early work [25] uses the pseudo‐inversion of a manipulator's Jacobian matrix to address the ...