Two-Dimensional LiDAR SLAM with Multi-Resolution Distance-Field Matching, Adaptive Keyframes, and Loop Closure

In two-dimensional laser SLAM exploration and mapping in complex indoor environments, scan matching is sensitive to the initial pose, error accumulation, and redundant map updates. To address these issues, an MDF-ALC-SLAM method is proposed. Firstly, a multi-resolution distance field is constructed at the front end, and a coarse-to-fine scan matching strategy is adopted to optimize the pose estimation method. Second, it selects adaptive keyframes based on translation displacement, rotation angle, and matching error. Third, the back end uses polar-coordinate-space geometric descriptors to search for candidate closed loops and performs pose graph optimization after geometric verification, achieving loop detection and map correction. Finally, offline replay simulation experiments and SIASUN smart park physical experiments were conducted. The offline experiment used manually annotated reference positions and headings, with trajectory deviation ATPE being 0.042 m and ATHE being 1.54°, and the average processing time per frame was 68 ms. The on-site experiment conducted 30 autonomous return tests and used a laser tracker to measure the return pose. ATPE was 0.053 m, and ATHE was 1.94°; on the RK3399 platform, the algorithm’s average processing time per frame was 73 ms. The on-site return error reflects system performance under the joint action of positioning and motion control, which is not equivalent to SLAM’s overall trajectory positioning error. It demonstrates the correctness and superiority of the proposed method.