Intelligent Robotic Arm Path Planning (IRAP2) Framework to Improve Work Safety in Human-Robot Collaboration (HRC) Workspace Using Deep Deterministic Policy Gradient (DDPG) Algorithm

Abstract

Abstract Industrial robots are widely used in manufacturing systems. The places that humans share with robots are called human-robot collaboration (HRC) workspaces. To ensure the safety in HRC workspaces, a collision-avoidance system is required. In this paper, we regard the collision-avoidance as a problem during the robot action trajectory design and propose an intelligent robotic arm path planning (IRAP 2 ) framework. The IRAP 2 framework is based on the deep deterministic policy gradient (DDPG) algorithm because the path planning is a typical continuous control problem in a dynamic environment, and DDPG is well suited for such problems. To test the IRAP 2 framework, we have studied a HRC workspace in which the robot size is larger than humans. At first, we have applied a physics engine to build a virtual HRC workspace including digital models of a robot and a human. Using this virtual HRC workspace as the environment model, we further trained an agent model using the DDPG algorithm. The trained model can optimize the motion path of the robot to avoid collision with the human.