Intelligent Solution for Inverse Kinematic of Industrial Robotic Manipulator Based on RNN

Abstract

The joint angles required for the robotic manipulator to execute a task in a preset location should be calculated using inverse kinematic equations. Finding these equations is important but it requires hard effort and a large time. In this work an Artificial Neural Network, more specifically, Recurrent Neural Network (RNN) is designed and trained using MATLAB such that the inverse kinematics for a robotic manipulator could be calculated. First, the Denavit-Hartenberg approach is used to derive the forward kinematics of a 6 Revolute (6R) robotic manipulator. Then, a dataset of 100000 samples is produced using the calculated homogeneous transformation matrices to train the RNN. The results are outstanding with MSE of 0.0013 and RF of 0.99 when compared to other techniques that are mentioned in the literature.