Robot Arm Movement Control by Model-based Reinforcement Learning using Machine Learning Regression Techniques and Particle Swarm Optimization

摘要

Robot arms are machines which are not only used in industrial technologies but also other applications such as medicine and agriculture. The robot arm movement control is important in the use of robot arms. This paper presents model-based reinforcement learning (MBRL) in robot arm movement control in case studies where targets are in random positions. The numerical studies of 2 and 3-DOF robots moving in planar motion are used as test problems. In 2-DOF robot control, 2 tasks - placing and reaching are employed. In 3-DOF robot control, noise is considered. Machine learning regression techniques - Gaussian process regression (GPR), artificial neural network (ANN), and support vector regression (SVR) - are used in environment modelling in MBRL while particle swarm optimization (PSO) is used as optimization tool in MBRL. The numerical studies show that MBRL with GPR has the highest performance with at least 95% success rate while MBRL with the other regression techniques such as ANN and SVR can succeed only 23-93% and 31-47%, respectively. Therefore, MBRL with GPR and PSO is suitable for robot arm movement control.