HT-LIO: Robust Hybrid Multiscale Planar and Thickness-Constrained LiDAR–Inertial Odometry for Pose Measurement

Abstract

This article presents HT-LiDAR–inertial odometry (LIO), a hybrid multiscale planar LIO framework for robust pose measurement across environments with heterogeneous geometric structures. HT-LIO maintains a large-scale planar voxel map for stable planar structures and a sparse support voxel map that constructs small-scale triangular planar patches on demand when large planes are unavailable or insufficiently observed. This complementary representation combines the reliability of validated large-scale planes with the availability of local geometric constraints in sparse and irregular regions, without requiring explicit mesh connectivity. A thickness-aware planar consistency residual is further proposed to constrain the normal-coordinate dispersion of planar observations, and all residuals are integrated into an error-state iterated Kalman filter (ESIKF). Quantitative experiments on 12 public sequences covering subterranean, cave, corridor, park, and open-field environments show that HT-LIO successfully completes all sequences. It achieves strong tracking robustness and low translational error in tunnel-like subterranean scenes while maintaining competitive estimation accuracy across heterogeneous indoor and outdoor environments.