LLOAM: LiDAR Odometry and Mapping with Loop-closure Detection Based Correction

Abstract

Re-observed places recognition problem has always been a key issue in the areas of mobile robot navigation including Simultaneous Localization and Mapping (SLAM), it is also called loop-closure detection. A global optimization back-end is usually required to correct global poses and map when a loop closure is detected. It is widely acknowledged that a standard SLAM system should be comprised of odometry front-end, loop-closure detection and a back-end responsible to global optimization. The previous SLAM researches are focused on the images based visual SLAM or lidar based odometry systems, there are still few complete 3D LiDAR based SLAM systems. This paper proposes a complete 3D LIDAR based SLAM system, LLOAM, which includes a front-end with a point clouds segmentation matching based loop-closure detection and a back-end based on factor graph optimization. The proposed SLAM system has been evaluated using a KITTI dataset of large-scale outdoor urban street environment. The results indicate that the proposed SLAM system outperforms others odometry system in local accuracy and global consistency.