GLO: General LiDAR-Only Odometry With High Efficiency and Low Drift

Abstract

This study proposes GLO, a general LiDAR-only odometry method with high efficiency and low drift. First, we propose a map data structure using multilevel voxels to improve map update efficiency. Each voxel node actively maintains plane features, minimizing redundant fitting and enhancing matching efficiency. By calculating the occupancy probability of each voxel node, dynamic and unstable points in the map can be efficiently removed. Second, we introduce a weighted elastic matching algorithm that adjusts the weights of each matching constraint across multiple dimensions, such as measurement depth, fitting error, occupancy probability, and matching error, making the matching constraints elastic and enhancing accuracy. This letter also proposes a progressive optimization framework combining prediction, coarse matching, and fine matching. Coarse matching ensures matching convergence and corrects point-cloud motion distortion, while fine matching further refines accuracy. Extensive experiments on public datasets and real LiDAR data demonstrate the efficiency and accuracy of the proposed GLO method compared to state-of-the-art methods.