A LiDAR–IMU tightly coupled and adaptive loop closure optimization method
Sen Li, Jihao Xu, He Guan, Jialuo Yang, Xiaofei Ma, Hezhao Liu, Shiwen LiPurpose
Indoor environments often contain repetitive structures, glass surfaces and feature-sparse spaces that degrade LiDAR measurements and impair SLAM performance. To address motion distortion, inertial measurement unit (IMU) drift and loop-closure failures under such conditions, this study aims to develop a robust LiDAR–IMU SLAM framework capable of maintaining stable localization accuracy and constructing globally consistent three-dimensional maps in degraded environments.
Design/methodology/approach
A tightly coupled LiDAR–IMU SLAM system is established by integrating high-precision spatiotemporal calibration, IMU-driven motion distortion compensation and a multi-dimensional point-cloud descriptor. An adaptive loop-closure detection strategy is introduced using a hybrid histogram of invariant structural characteristics (HISC) descriptor and a dynamic similarity threshold derived from real-time pose analysis. Global pose refinement is achieved through a factor-graph-based optimization model incorporating covariance-weighted constraints and a sliding-window keyframe strategy. The proposed system is evaluated on the KITTI data set and a self-collected indoor data set.
Findings
Experimental analyses demonstrate that the proposed method improves absolute trajectory accuracy compared with scan context (SC), ISC, A-LiDAR odometry and mapping (LOAM), LeGO-LOAM, LIO-SAM and SC-F-LOAM. The HISC descriptor reduces root mean square absolute trajectory error by up to 15.93% relative to conventional scan-context methods. The adaptive loop-closure module decreases computational time by 60%–78% compared with LIO-SAM while maintaining detection reliability. Mapping experiments show that the system generates geometrically consistent global maps with suppressed drift and improved representation of road structures and indoor spatial features.
Originality/value
This work introduces a unified LiDAR–IMU fusion framework that tightly couples motion compensation, multimodal global descriptors, adaptive loop-closure validation and factor-graph optimization. The proposed HISC descriptor and dynamic thresholding mechanism enhance loop-closure robustness in repetitive or degenerate environments. The system demonstrates strong adaptability across diverse indoor and urban scenarios, offering a high-precision SLAM solution suitable for long-duration operations and robotics applications requiring reliable localization and mapping.