A Deep Reinforcement Learning Method for Mobile Robot Path Planning in Unknown Environments

Abstract

In this paper, a path planning method without global maps for mobile robots is proposed. At present, the traditional path planning methods have become mature. However, most of these methods are based on map information, and complex models need to be defined. To address these issues, we propose an end-to-end path planning model based on deep reinforcement learning, which takes the lidar sequence and depth image as the inputs, and the robot's velocity and angular velocity as the outputs. With the proposed method, the mobile robot can plan the motion path from the initial position to the target position without colliding with the obstacles in the environment. Besides, we propose a neural network that has the ability of multi-sensor information fusion from different dimensions. We build a simulation environment for mobile robot path planning in the gazebo based on Robot Operating System(ROS). Simulation results show the effectiveness of the proposed method.