RB-LIO: A SLAM Solution Applied to Large-Scale Dynamic Scenes with Multiple Moving Objects

Simultaneous Localization And Mapping (SLAM) is a critical technology for autonomous driving in urban environments. However, in environments with many moving objects, currently available LiDAR-based SLAM methods cannot effectively detect loops as they assume a static environment, resulting in unreliable trajectories. Therefore, in this paper, we propose RB-LIO, a LiDAR SLAM system based on the LIO-SAM framework. The proposed system utilizes a dynamic object segmentation module to mitigate the influence of moving objects with tight-coupled LiDAR inertial odometry. It also corrects the complete 6-DoF loop closure with BoW3D and performs pose graph optimization using GTSAM-ISAM2. We tested RB-LIO on public datasets (KITTI and MulRan) and self-collected datasets and compared it with state-of-the-art SLAM systems such as A-LOAM, LeGO-LOAM, LINS, LIO-SAM, and Fast-LIO2. The experimental results indicate that RB-LIO achieves more than 40% improvement in accuracy and a significant improvement of map quality.