Motion Navigation Algorithm Based on Deep Reinforcement Learning for Manipulators

Abstract

In this research, we implement the deep reinforcement learning algorithm for handling the navigation problem of a 6-DOF robotic manipulator, which is a complex problem in robotics and requires real-time decision-making capabilities according to a particular strategy. A motion planning algorithm is developed based on combining the hindsight experience replay with the deep deterministic policy gradient algorithm (i.e., DDPG +HER), in which the robot is trained in the virtual physical environment. According to the experimental results, the developed algorithm can control the 6-DOF robot Nachi-MZ07 to perform a task autonomously and effectively.