Neural Network Approach to the Inverse Kinematic Control of Robotic Manipulator
Faculty
Department
Year of Publication
upload
Publication Type
Abstract
Robots are becoming more relevant across multiple industries; therefore, it is important for
engineers to innovate in all aspects of this technology. This Project explores the concept of inverse kinematic control in robotics, its significance, difficulties and explores a solution using Neural networks. It presents an analysis of the forward kinematics of various robot arms-using the Denavit-Hartenberg method- as a way to generate the data set for training and testing the Neural network, until a suitable performance benchmark was reached. The information on the robot arms that were analyzed was gotten from manufacturers publications, open to the public. Some of the robot arms considered includes; ABB IRB 1600; a 6dof robot arm, FANUC LR Mate 200iC; a 6dof robot arm, The Universal Robots UR10; a 6dof robot arm, among others. This project considers one of the simplest machine learning models for regression analysis, the Artificial Neural Network (ANN) The trained model displayed high accuracy in predicting the suitable joint angles, with all trained models having a R squared value above 0.70.
engineers to innovate in all aspects of this technology. This Project explores the concept of inverse kinematic control in robotics, its significance, difficulties and explores a solution using Neural networks. It presents an analysis of the forward kinematics of various robot arms-using the Denavit-Hartenberg method- as a way to generate the data set for training and testing the Neural network, until a suitable performance benchmark was reached. The information on the robot arms that were analyzed was gotten from manufacturers publications, open to the public. Some of the robot arms considered includes; ABB IRB 1600; a 6dof robot arm, FANUC LR Mate 200iC; a 6dof robot arm, The Universal Robots UR10; a 6dof robot arm, among others. This project considers one of the simplest machine learning models for regression analysis, the Artificial Neural Network (ANN) The trained model displayed high accuracy in predicting the suitable joint angles, with all trained models having a R squared value above 0.70.
Supervisor(s)
co-supervisor