Collision-Free Motion Planning for Multiple Robot Arms by Combining Deep Q-Network and Graph Search Algorithm

In recent years, multiple industrial robots have been introduced into factories, and they are required to perform complex tasks. In order to improve work efficiency, it is also important to generate a collision-free trajectory in a limited space where the robots interfere with each other. However, in the actual collision avoidance motion, the robot may collide due to motion delay or disturbances. In particular, collisions between industrial robots and humans have caused fatal accidents in the past, and a collision avoidance motion planning method for multiple robot arms is required to avoid such accidents. Previous research has achieved a real time avoidance method for multiple robot arms using a graph search algorithm and Q-learning which is one of the reinforcement learning method. In this paper, we propose a collision avoidance motion planning method using DQN (Deep Q-Network) instead of Q-learning. By using DQN, it is possible to deal with the case where the input is an image and continuous state space. We compare the performance of the proposed method and the conventional method by simulation experiments. As a result, the proposed method shows its usefulness in the cases where the conventional method could not avoid collisions.