Autonomous Navigation of an AMR using Deep Reinforcement Learning in a Warehouse Environment

Abstract

Mobile robots have been used in warehouses worldwide as a means for distribution of goods and gained demand after the Covid19 labor issue. This paper proposes an Autonomous Mobile Robot (AMR) to navigate in a warehouse environment to its target location using LIDAR. The method used to solve this problem is a deep reinforcement learning algorithm called deep Q-network (DQN) to detect and avoid obstacles and reach the target location. DQN is used as it is desired for solving complex tasks. Training of the DQN algorithm is carried out in ROS Gazebo environment using LIDAR-based robot model. The LIDAR sensor detects the obstacles and the odometer sensor helps to find the distance between the target location are used as inputs for training the algorithm and optimal actions are taken based on the two inputs. A reward policy is awarded when an obstacle is avoided and reaches the target location. The results show that mobile robot can successfully navigate in an unknown environment through simulation and real life.