Autonomous Navigation System with Obstacle Avoidance using 2.5D Map Generated by Point Cloud

Abstract

The development of a robust obstacle avoidance system for the autonomous mobile robot has become important as their applications become widespread. Even though plenty of autonomous systems have been developed, safe driving in complex environments, such as crowded places or a place with many obstacles in the path, is still a challenging task. In this paper, we propose an autonomous navigation system of a mobile robot in a dynamic environment. We build a 2.5D map, that integrates a 2D grid map with dynamic objects 3D geometry information. From the 3D LiDAR point cloud, dynamic points are detected by tracking the occupancy changes over time. Then remained static points are used for generating a 2D grid map by SLAM algorithm. A computation cost is reduced efficiently by reconstructing only the necessary parts for the mobile robot driving into the high resolution of raw point cloud data. In addition, it can be easily utilized in the embedded board since the system does not need any complex calculations. Our experiments demonstrate that the system is possible to simultaneously perform multiple roles of map building, localization, and dynamic obstacle avoidance using real-time incoming 3D point cloud data.