An Artificial Neural Network Approach in Solving Inverse Kinematics of a 6 DOF KUKA Industrial Robot

Abstract

Inverse kinematics is a mathematical method for computing the joint angles required to set the end effector of a robot in a particular position and orientation. The Inverse kinematics problem is challenging, especially for arm robots with several degrees of freedom. This paper provides two models of artificial neural networks for determining the Inverse kinematics solution of the KUKA KR/R900/SIXX 6 degrees of freedom manipulator. The Non-Linear Autoregressive Neural Network with Multiple Exogenous Variables Recurrent Model is used to create the first model. The second model, Adaptive Feedforward artificial neural networks, is used to investigate the impact of several training methods with one hidden layer, changing numbers of neurons in the hidden layer, and measuring the artificial neural networks learning performance on the Inverse kinematics model learning of a 6-joint redundant robotic manipulator. The obtained results showed that the "Bayesian Regularization" training algorithm achieved the lowest mean square error score of 0.005 with a neuron number of 250.