A model-free mapless navigation method for mobile robot using reinforcement learning

In this paper, a simple and efficient method is proposed using one of the basic reinforcement learning tool Q-learning. This method is to achieve the aim of mobile robot maples navigation with collision-free. Thus, an end-to-end navigation model is built, which uses lidar sensor massage as input and moving command of mobile robot as output, so as to simplify the process of environmental perception and decision making on action. Through simulation and experiments, its effectiveness is proved. After trained by this method, mobile robot can safely reach the navigation target in an unknown environment without any prior demonstration. In addition, an extensive quantitative and qualitative evaluation of this method is presented by the comparison with traditional path planning method based on foregone global environment map.