Optimizing 7-DOF Robot Manipulator Path Using Deep Reinforcement Learning Techniques

摘要

This paper proposes three different Deep Rein-forcement Learning (DRL) techniques to find a free-obstacle path for a 7-DOF Robot Manipulator (RM). The robot is the Kinova Jaco Assistive Robot arm; its DH parameters and kinematics are presented. The suggested DRL methods are Deep Q-Network (DQN), Actor-Critic (AC), and Proximal Policy Optimization (PPO) algorithms. The environment, state space, action space, and reward function are defined to suit all proposed methods. The experiment is run to validate the performance of these methods in finding the path for the RM in two different environments. These environments differ in obstacles and complexity. The reward and average reward evaluation were used to evaluate the proposed methods. The results showed that the three models found the path in the suggested environments. The comparison between these methods is presented and discussed in this paper. The result discussion clarified the superiority of the PPO method in finding the free-obstacle path in a complex environment.