Init-LOAM: LiDAR-based Localization and Mapping with a Static Self-Generated Initial Map

Abstract

In GPS-denied environments, Simultaneous Localization and Mapping (SLAM) is a key technology for the navigation of autonomous robots. 3D LiDAR sensors are particularly suitable in this context as they enable accurate localization and high-quality mapping of a previously unknown environment. In this work, we propose an approach to improve 3D LiDAR-based SLAM performance by creating an initial map of the environment prior to exploration by rotating a 3D LiDAR sensor while the robot, on which the sensor is mounted, is still static. Our approach assumes a static environment and is implemented as a modification of the well-known LOAM framework. We provide a detailed algorithm description of the initial map creation using LOAM. The approach is validated on three simulated datasets with a 3D LiDAR sensor mounted on a UAV via a 1-DoF gimbal. The datasets feature indoor and outdoor visual inspection scenarios. We compare the case where the entries of the initial map remain unchanged during exploration with the case where the initial map is updated during the movement of the mobile platform. Our results show a reduction in trajectory error when creating an initial map before exploration compared to state-of-the-art LOAM and superior results when using an unchangeable initial map.