An artificial neural network strategy for the forward kinematics of robot control

Abstract

This paper is devoted to the development and implementation of the neural network technique to solve the forward kinematics problems of robot control, which are mainly singularities and non-linearities. In this paper, a network has been trained to learn the set of end effecter positions X, Y and Z from a given set of joint angle positions for a 6 D.O.F industrial robot. Training data sets were uniformly distributed over a particular region of the robot's working area so that the network can make good generalisation for the intermediate points. Experimental results have shown a good mapping over the working area for the robot. The proposed control technique does not require any prior knowledge of the kinematics model of the system to be controlled; the basic idea of this concept is to use a neural network to learn the characteristics of the robot system rather than having to specify explicit robot system model, which is a significant advantage of using neural network technology. Any modifications in the physical set-up of the system would involve only training the robot in a new path without the need for any major software modification.