New Robust LiDAR-Inertial Odometry Method

rsasaki0109 · x · 2026-07-17

BIEVR-LIO introduces a robust LiDAR-Inertial Odometry method designed for complex environments.

The core idea represents surfaces as "voxel-level directional height maps" that can be directly used for registration and efficiently updated.
It also proposes a map information-based point sampling strategy, focusing computation on geometrically rich areas to improve stability in feature-scarce environments while reducing computational overhead.
Experiments across various sensors, platforms, and environments show it achieves SOTA in well-constrained scenarios, significantly outperforms baselines in difficult ones, and even prevents divergence where some methods fail.
The paper also demonstrates that the fine-grained geometric details captured by this method can be utilized for downstream tasks, such as generating elevation maps for robot navigation.

More from Embodied