In context
In 2024, indoor mobile robots and AGVs increasingly relied on LiDAR SLAM for positioning, as GPS is unreliable indoors and traditional guide tapes or QR codes require fixed environments and extra maintenance. Accurate point cloud registration remained a bottleneck for real-time localization, with existing ICP variants suffering from correspondence model flaws and sensitivity to initial poses.
What was reported
Researchers from Nanjing University of Aeronautics and Astronautics proposed a graph search-based point-to-tangent ICP method for high-precision 2D LiDAR point cloud registration. The method first builds a graph structure using depth features of scan points to efficiently search for nearest neighbors, avoiding errors common in jump-table-based searches. It then introduces a new correspondence model: instead of using existing points or projections, it simulates laser rays and uses their intersections with tangents of environmental sampling points as correspondences, better reflecting actual observation geometry.
To address initial value sensitivity, a coarse registration method based on pose particle probability is implemented. Possible pose transformations are generated as particles, scored, and the highest-scoring pose is used as the initial input for fine registration. The error function is solved with a closed-form solution using Lagrange multipliers.
Experiments with an R2000 LiDAR (distance accuracy 0.001 m, angular resolution 0.0043°) showed the method completes matching of two frames in about 100 ms, achieving high-precision results. The graph-based search reduced nearest-neighbor search time to 0.00168 s per frame pair, outperforming other methods in the study.
Why it mattered
This work offered a practical improvement for real-time LiDAR SLAM in industrial settings, potentially enhancing AGV and mobile robot localization accuracy without extra hardware. The point-to-tangent model and graph search could influence future registration algorithms, especially in environments with repetitive structures or sparse features.
“The results show that, compared to other methods in the experiment, the proposed method can complete the matching of two frames of 2D LiDAR data in a short time (100 ms) and obtain high-precision estimation results.”
Source: 《机器人》期刊 (robot.sia.cn) · Published 2024-09-15 · “基于图搜索点对切线ICP的高精度激光点云配准方法”