6-DOF Manipulator Path Planning Based on Reinforcement Learning

• Main Authors: Tsai,Han-Hsien, 蔡瀚賢
• Other Authors: Chen,Ming-Fei
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/bx72t5

碩士 === 國立彰化師範大學 === 機電工程學系 === 107 === 　　The purpose of this paper is to develop a robotic autonomous learning system based on virtual modeling and reinforcement learning. Its main function is given by the mechanical arm model, the environmental obstacle, the initial coordinate of the robot arm and the target point, and then the system will be automatically generated a set of motion corners to enable the robot to smoothly avoid the obstacle and reach the target. The program development in this article will be divided into two parts. The first part includes arm simulation and collision detection. Firstly, the VTK visualization tool library is utilized, and the six-axis robotic arm files and obstacles are imported, and then the motion of the robot arm and the surrounding environment are visualized. Finally, it is judged whether or not the collision is caused by the directional bounding box algorithm. The second part includes the machine learning for the path planning. The DDPG model is established through Tensorflow package. Next, the reinforcement learning is used to interact with the environmental variables. The different reward functions are designed to test and discuss. Finally, the feasibility is verified by the actual machine operation. This experiment finally proves the feasibility of reinforcement learning applied to the path planning of the robot arm. Through this method, the robot arm can be autonomously planned, and the target point is within 10 mm of the error.