UD-LIOM: A Universal Tightly Coupled Direct LiDAR-Inertial Odometry and Mapping Framework

LiDAR-based SLAM, as a key high-precision pose measurement technology for autonomous driving and embodied intelligence, has attracted considerable attention. In addition, to accommodate diverse platforms, various types of LiDARs have emerged in recent years. Thus, it is imperative to design SLAM systems capable of adapting to different types of LiDARs. However, variations in scanning patterns among different LiDAR types impose significant challenges to the cross-LiDAR compatibility of existing SLAM methods. Moreover, to enhance the robustness and accuracy of these LiDAR SLAM systems, many studies have integrated inertial data into the framework. However, existing LiDAR-inertial fusion methods are mostly tailored for specific LiDAR types, and the residual-level deep fusion of IMU constraints with universal LiDAR processing pipelines remains to be further explored. While state-of-the-art (SOTA) methods have realized tight coupling of IMU and LiDAR in back-end optimization, their poor adaptability to diverse LiDAR scanning patterns and shallow fusion at the factor/filter level restrict practical deployment in multi-sensor scenarios. To address these challenges of cross-LiDAR type universality and multi-sensor data synergy in LiDAR SLAM, we propose UD-LIOM, a universal tightly coupled LiDAR–inertial odometry and mapping framework that makes two major contributions. First, unlike existing direct LiDAR SLAM methods that process each LiDAR point uniformly, UD-LIOM leverages the common scanning “ring” structure across different LiDAR types to develop a universal LiDAR measurement data classification strategy. By dividing each LiDAR scan into edge and planar points and performing dedicated registration with edge/planar submaps, UD-LIOM builds more precise geometric constraints, thus enabling cross-type compatibility with diverse LiDAR scanning patterns and enhancing pose estimation accuracy. Second, in terms of IMU fusion, UD-LIOM integrates IMU pre-integration for front-end motion distortion correction and fuses IMU pre-integration residuals with LiDAR measurement residuals into a unified multi-constraint optimization model, realizing residual-level deep fusion to boost system robustness across complex scenarios. Experimental results on public and self-collected datasets demonstrate that UD-LIOM achieves superior measurement accuracy and system robustness compared with other SOTA methods. For full reproducibility, all code and data are available at https://cslinzhang.github.io/UD-LIOM.