LIO-GVM: An Accurate, Tightly-Coupled Lidar-Inertial Odometry With Gaussian Voxel Map
Xingyu Ji, Shenghai Yuan, Pengyu Yin, Lihua Xie
Abstract
This letter presents a probabilistic voxel-based LiDAR Inertial Odometry framework for accurate and robust pose estimation. The framework addresses the correspondence mismatching issue by representing the LiDAR points as a set of Gaussian distributions and evaluating the divergence in variance for outlier rejection. Based on the fitted distributions, a new residual metric is proposed for the filter-based Lidar inertial odometry by incorporating both the distance and variance disparities, further enriching the comprehensiveness and accuracy of the residual metric. With the strategic design of the residual, we propose a simple yet effective voxel-solely mapping scheme, which only requires the maintenance of one centroid and one covariance matrix for each voxel. Experiments on different datasets demonstrate the robustness and high accuracy of our framework for various data inputs and environments.