GaRLIO: Gravity enhanced Radar-LiDAR-Inertial Odometry

TL;DR

This paper addresses vertical drift in LiDAR-Inertial Odometry by introducing gravity-aware estimation that leverages radar Doppler velocity. The GaRLIO framework fuses radar velocity with LiDAR and IMU data, applying a velocity-aware gravity residual in a two-stage update and using radar to filter dynamic LiDAR points. Key contributions include a novel velocity-based gravity estimation mechanism, radar-guided dynamic removal, and comprehensive experiments across diverse datasets showing improved elevation accuracy and robustness in dynamic and downhill scenarios. The work demonstrates practical impact for robust SLAM in challenging environments and provides open-source code to foster further development.

Abstract

Recently, gravity has been highlighted as a crucial constraint for state estimation to alleviate potential vertical drift. Existing online gravity estimation methods rely on pose estimation combined with IMU measurements, which is considered best practice when direct velocity measurements are unavailable. However, with radar sensors providing direct velocity data-a measurement not yet utilized for gravity estimation-we found a significant opportunity to improve gravity estimation accuracy substantially. GaRLIO, the proposed gravity-enhanced Radar-LiDAR-Inertial Odometry, can robustly predict gravity to reduce vertical drift while simultaneously enhancing state estimation performance using pointwise velocity measurements. Furthermore, GaRLIO ensures robustness in dynamic environments by utilizing radar to remove dynamic objects from LiDAR point clouds. Our method is validated through experiments in various environments prone to vertical drift, demonstrating superior performance compared to traditional LiDAR-Inertial Odometry methods. We make our source code publicly available to encourage further research and development. https://github.com/ChiyunNoh/GaRLIO

Figures (8)

• Figure 1: Top: Trajectories of GaRLIO and other methods with ground truth (white) overlaid on the TLS map. Bottom: Elevation plot along path length. Our method (blue) reported only $\unit{1.21}{m}$ vertical mean error over $\unit{2.045}{km}$ path length.
• Figure 2: GaRLIO is divided into four primary modules. Each module contributes to achieving the optimal state by removing LiDAR dynamic points and calculating both pointwise velocity and velocity-aware gravity residuals.
• Figure 3: Gravity estimation using the velocity-ignorant (yellow) and velocity-aware method (ours, blue) on different platforms.
• Figure 4: Estimated $z$ (top) and $xy$ (bottom) trajectory for the $\textit{loop3}$ sequence. $\textbf{Top}$ : From 120 to 350 seconds, GaRLIO accurately predicts elevation during the downhill descent, outperforming other methods. $\textbf{Bottom}$ : GaRLIO results closely match the ground truth. The black dot represents the start point.
• Figure 5: Estimated trajectory for the $\textit{iaef}$ sequence and GaRLIO demonstrated superior performance over such a long sequence.