Inverse Kinematics Solution for Robot Manipulator based on Neural Network under Joint Subspace

Abstract

Neural networks with their inherent learning ability have been widely applied to solve the robot manipulator inverse kinematics problems. However, there are still two open problems: (1) without knowing inverse kinematic expressions, these solutions have the difficulty of how to collect training sets, and (2) the gradient-based learning algorithms can cause a very slow training process, especially for a complex configuration, or a large set of training data. Unlike these traditional implementations, the proposed metho trains neural network in joint subspace which can be easily calculated with electromagnetism-like method. The kinematics equation and its inverse are one-to-one mapping within the subspace. Thus the constrained training sets can be easily collected by forward kinematics relations. For issue 2, this paper uses a novel learning algorithm called extreme learning machine (ELM) which randomly choose the input weights and analytically determines the output weights of the single hidden layer feedforward neural networks (SLFNs). In theory, this algorithm tends to provide the best generalization performance at extremely fast learning speed. The results show that the proposed approach has not only greatly reduced the computation time but also improved the precision.