Abstract:
In underground coal mines, Global Navigation Satellite System (GNSS) signals are completely unavailable and roadway features are sparse, posing challenges to existing Simultaneous Localization and Mapping (SLAM) algorithms, including high computational requirements and cumulative drift over long distances. To address these problems, a lightweight LiDAR-inertial SLAM method for underground mines with Ultra-Wideband (UWB) constraints was proposed. During mapping, segmented dynamic distortion compensation guided by an Inertial Measurement Unit (IMU) was used to address point cloud distortion in solid-state LiDAR. Tightly coupled LiDAR-IMU odometry was developed, and roadway geometric constraints and ground constraints were introduced to improve registration stability in feature-sparse scenes. Global UWB position observations were used to correct pose drift at a low frequency, and incremental local map management was adopted for lightweight mapping. During relocalization, real-time localization was performed by matching point clouds to a prior cloud map, with UWB position observations providing a basis for drift detection and relocalization. When the discrepancy between the position estimated by point cloud matching and the UWB position exceeded a threshold, a warning was triggered and the matching range was adjusted. When the relocalization threshold was exceeded, the UWB position was used to assist localization recovery. Tests under typical underground operating conditions showed that the method produced complete, accurate maps and consistent trajectories and achieved stable decimeter-level localization. The planar position Root Mean Square Error (RMSE) was 0.22 m. Localization accuracy and real-time performance were significantly better than those of mainstream LiDAR-inertial SLAM algorithms, and localization was more stable than with UWB alone. The relocalization mechanism reliably corrected position drift of 2–4 m. When the initial deviation increased to 6 m, the relocalization success rate and correction accuracy decreased, but effective position recovery was still achieved overall.