Open-Space-Based Motion Planner of Mobile Robot for Multiple Obstacle Avoidance

Abstract

Motion planning for obstacle avoidance is an essential capability required for mobile robots. In general, there are multiple obstacles around a robot. However, it is difficult for the robot based on end-to-end motion planners to avoid the obstacles. This is because the previous motion planners have been trained for a single obstacle. In a framework of imitation learning, the instruction cost increases with the number of obstacles and positions. For this challenge, we focus on open space rather than obstacles. For the motion planner, open-space images are used as inputs. As well as the previous motion planners, the motion planner is trained for a single obstacle. Nevertheless, we show the robot based on the proposed motion planner is able to avoid multiple obstacles through the navigation experiments.