Obstacle Avoidance for Mobile Robots Using End-to-End Learning
Jeong-Jung Kim, Doo-Yeol Koh, J.-G. Park · Journal of Institute of Control Robotics and Systems · 2019
The purpose of autonomous driving is to help a mobile robot move from a starting point to a target point without any collisions. A mobile robot requires global path planning and local path planning to generate a path, where the local path planning guides the robot along the global path while taking action to avoid obstacles. In this paper, end-to-end approach to obstacle avoidance for a mobile robot is proposed. The approach uses a depth data to produce a straight motion, left rotation, or right rotation. The only data that the learning model requires is the depth image and control commands from the human operator. The method does not perform any additional operations like feature extraction on the raw sensor input values, resulting in a relatively low computing burden and intuitive learning without needing to know the robot model. Experimental results show that the proposed method successfully avoids obstacles.