Development of a learning system for robot control

Abstract

Robot Kinematics and Control has been a vital part in studying the motion of the robot manipulator. As technology advances, the use of machine learning algorithms in robot control is gaining popularity. As numerical calculations and derivation of the robot’s kinematics model could prove to be challenging, machine learning algorithms could be used to estimate the kinematic system of the robots. In the first part of the project, machine learning models would be developed using the Python programming language. The machine learning models include a densely-connected Neural Network and a new proposed learning algorithm, referred to as Data-driven learning. Afterwards, the models would be trained using the experimental data of a SCARA robot. The data consists of the position coordinates of the robot as well as the joint angle values. The parameters of the learning models would be varied to see the effects of said parameters. The second part of the project would involve several tests on the actual SCARA robot. The estimated system derived using the learning algorithms would be transferred to the robot. The robot would then be instructed to move to a certain setpoint, where its accuracy and overall path would be analysed. Additionally, several parameters on the robot interface will be varied to see the effect on the overall movement.