DR-LIOM: Direct Registration LiDAR-Inertial Odometry and Mapping for Uncrewed Surface Vehicles in a Harbor

LiDAR-dominated simultaneous localization and mapping (SLAM) for uncrewed surface vehicles in a harbor is challenged by sparse water-surface returns and motion-induced distortions. This paper presents a direct registration LiDAR-inertial odometry and mapping (DR-LIOM) framework that incorporates inertial measurement unit (IMU) pre-integration for point cloud de-skewing, keyframe-based local map construction, and full-resolution point-to-plane iterative closest point for robust odometry estimation. Loop closure mechanism and multi-source factor graph, integrating LiDAR, IMU, RTK/GPS, and loop constraints, are devised to ensure globally consistent mapping. Field experiments in Linghai Harbor demonstrate that the DR-LIOM achieves absolute trajectory errors around 0.5 m in 2D and 3D domains, significantly outperforming typical SLAM baselines that suffer from localization errors in tens of meters. Despite using dense registration, the DR-LIOM not only maintains real-time feasibility at 12.3 Hz, but also supports integration of surface and underwater data, enabling high-fidelity harbor charting.