Continuous Control with Deep Reinforcement Learning for Mobile Robot Navigation

Abstract

Autonomous navigation is one of the focuses in the field of mobile robot research. The traditional method usually consists of two parts: building the map of environment, localization of mobile robot and path planning. However, these traditional methods usually rely on high-precision sensor information. At the same time, mobile robots have no intelligent understanding of autonomous navigation. In this article, a deep reinforcement learning method, i.e. soft actor critic, is used to navigate in a mapless environment. It takes laser scanning data and information of the target as input, outputs linear velocity and angular velocity in continuous space. The simulation shows that this learning-based end-to-end autonomous navigation method can accomplish tasks as well as traditional methods.