A Method of Solution to an Inverse Kinematics Problem of a Redundant Manipulator Using Neural Networks

Abstract

An inverse kinematics problem of robot manipulators with redundancies has a bottleneck of computational time. In this paper, we proposed a method of solution to the above problem using a neural network with sufficient accuracy and short computational time. The learning algorithm proposed here automatically selects a set of joint angles, which maximizes a given cost function using redundant degrees of freedom, from a large number of solutions and forms the inverse kinematics model in the network. Application of the algorithm to the inverse kinematics problems of kinematic singularity and joint limit avoidance is shown as an example.