AMR-LOAM: Adaptive Multi-Resolution LiDAR Odometry and Mapping in Complex Urban Environments

Abstract

In complex urban environments, achieving globally consistent LiDAR-based Simultaneous Localization and Mapping (SLAM) is crucial for reliable autonomous navigation. Conventional methods typically employ fixed-resolution maps for computational efficiency, which fail to accommodate the inherent density disparity of point clouds at varying distances and consequently degrade the distinctiveness of place recognition descriptors and loop closure reliability. To address these issues, an adaptive LiDAR odometry and mapping method, AMR-LOAM, is proposed. AMR-LOAM employs an adaptive multi-resolution voxel map strategy to balance point cloud density across near-range regions and far-range regions. The resulting density-balanced point clouds are then utilized to generate scan context descriptors with enhanced geometric distinctiveness, thereby enabling reliable loop closure detection and effectively eliminating cumulative drift. Experimental results on the KITTI dataset demonstrate that the proposed method reduces RMSE by up to 71.96%, 74.00%, and 49.39% compared with A-LOAM, LeGO-LOAM, and SC-A-LOAM, respectively. Evaluations on the MulRan dataset further confirm the superior performance of the proposed method in constructing globally consistent maps, achieving the lowest RMSE of 1.88 m among all evaluated methods.