Mobile Robot Obstacle Avoidance Based on Optical Flow Images in Motion Planning

Abstract

For autonomous navigation, mobile robots must avoid collisions with dynamic obstacles such as pedestrians. In previous work, we proposed a motion planner based on a CNN with RGB and depth image inputs. To enable the robot to plan avoidance motions while considering the moving direction of an obstacle, in this paper, we propose a motion planner that utilizes optical flow images as inputs. Through autonomous navigation experiments, we demonstrate that the proposed motion planner enables the robot to successfully avoid both dynamic and static obstacles.