An Unsupervised Neural Network Approach for Inverse Kinematics Solution of Manipulator following Kalman Filter based Trajectory

Abstract

A novel unsupervised approach for inverse kinematics solution of a manipulator using artificial neural network is presented. Forward kinematics equations determine the motion of manipulator's arm and have a unique solution. But there is not a unique solution for inverse kinematics as manipulator may have more than one configurations to reach a particular point. Here in this paper, we have taken a PUMA 560 robot with six degrees of freedom with aim to grab an object moving in circular path in XY plane with a known constant height and kalman filter has been used to determine accurate position of that object. Contrary to supervised learning approach, which needs a huge amount of data to train the system, we have used a real time unsupervised approach to solve inverse kinematics problem which is more efficient. Joint angles of the robot are determined in real time using unsupervised feed forward neural network with backpropagation training algorithm.