Behavior Switch for DRL-based Robot Navigation
Wěi Zhāng, Yunfeng Zhang · 2019
Recently, deep reinforcement learning (DRL) has shown great potential in mapless navigation of mobile robotics. However, without a map, most reported DRL-based controllers failed to drive robots out of local-minimum areas. In this paper, a hierarchical control framework is proposed to address this problem. In this framework, two low-level controllers, i.e., one for basic navigation and one for wall-following, are trained by DRL methods. The high-level switch is designed by some heuristic rules, which can adaptively choose the suitable low-level controller based on the current situation. Simulation-based testing results show that the proposed method is able to control the robot to navigate in challenging scenarios with local-minimum areas.