Large-scale 3D outdoor mapping and on-line localization using 3D-2D matching

Abstract

Map-based outdoor navigation is an active research area in mobile robots and autonomous driving. By preparing a precise map of an environment or roadside, a robot or a vehicle can localize itself based on a matching between the map and a sequence of sensor inputs. This paper describes a campus-wide mapping and localization of a mobile robot with 2D and 3D LIDARs (Laser Imaging Detection and Ranging). For mapping, we use a 3D data acquisition system with a 2D LIDAR and a rotation mechanism and takes a sequence of point clouds. We adopt an NDT (Normal Distribution Transform)-based ego-motion estimation method for pose graph generation and optimization for loop closing. For localization, we propose to use a 2D LIDAR on a robot for being matched with a 3D map for a fast and low-cost localization. The mapping and the localization method are validated through the experiments in our campus.