LiDAR-Inertial Odometry for Intelligent Vehicles in Complex Environments via InEKF
Year: 2026
Type: article
Abstract: LiDAR-inertial odometry (LIO) for intelligent vehicles remains challenging in complex environments as LiDAR measurements are affected by moving objects, inaccurate ground extraction, and non-Gaussian residual noise. This paper presents a tightly coupled LIO framework based on an invariant extended Kalman filter (InEKF) with a maximum correntropy criterion (MCC) update. A layered ground segmentation module is used before feature extraction to exploit the different point distributions in near and far regions, and the estimated ground normal is further used to build a ground residual. The MCC-based update then reweights the LiDAR and ground residuals during iterative state correction, which reduces the influence of large residuals caused by outliers or non-Gaussian noise. Experiments on M2DGR, SemanticKITTI, and real indoor/outdoor vehicle tests show errors that are competitive with or lower than those of the compared baselines in most tested cases, and the remaining limitations are discussed.
Authors: Ge Guo, Haodong Lin, Guangheng Liu, Jiageng Liu
Institutions: Northeastern University, Shenyang Ligong University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,645
OpenAlex ID: W7202201604