Ground-Optimized 4D Radar-Inertial Odometry via Continuous Velocity Integration using Gaussian Process

TL;DR

This work introduces a ground-optimized 4D radar–inertial odometry framework that jointly addresses radar noise and sensor asynchrony. It combines uncertainty-aware ground filtering with zone-based modeling and a Gaussian-Process–based continuous velocity preintegration to tightly fuse radar velocity and IMU data, alongside cluster-weighted ICP and pose-graph optimization. The approach achieves superior elevation accuracy and robust performance across diverse datasets, outperforming state-of-the-art 4D radar–inertial methods and enabling near real-time operation. The open-source release facilitates adoption and further research in radar-inertial odometry under challenging conditions.

Abstract

Radar ensures robust sensing capabilities in adverse weather conditions, yet challenges remain due to its high inherent noise level. Existing radar odometry has overcome these challenges with strategies such as filtering spurious points, exploiting Doppler velocity, or integrating with inertial measurements. This paper presents two novel improvements beyond the existing radar-inertial odometry: ground-optimized noise filtering and continuous velocity preintegration. Despite the widespread use of ground planes in LiDAR odometry, imprecise ground point distributions of radar measurements cause naive plane fitting to fail. Unlike plane fitting in LiDAR, we introduce a zone-based uncertainty-aware ground modeling specifically designed for radar. Secondly, we note that radar velocity measurements can be better combined with IMU for a more accurate preintegration in radar-inertial odometry. Existing methods often ignore temporal discrepancies between radar and IMU by simplifying the complexities of asynchronous data streams with discretized propagation models. Tackling this issue, we leverage GP and formulate a continuous preintegration method for tightly integrating 3-DOF linear velocity with IMU, facilitating full 6-DOF motion directly from the raw measurements. Our approach demonstrates remarkable performance (less than 1% vertical drift) in public datasets with meticulous conditions, illustrating substantial improvement in elevation accuracy. The code will be released as open source for the community: https://github.com/wooseongY/Go-RIO.

Figures (7)

• Figure 1: Top: Our uncertainty-aware ground filtering efficiently eliminates the noise (red) from radar. Middle: Mapping result of RURAL_A2. Our continuous velocity integration with the Gaussian Process (GP) proficiently handles the sharp turns and roundabouts. Bottom: The proposed method shows the lowest elevation error (18m over 2.5km of path length, only 0.72%).
• Figure 2: The overall pipeline of our algorithm.
• Figure 3: Comparison between naive plane fitting and our zone-based approach in the sloped environment (yellow box). (c) Many true negatives are found when using naive plane fitting (TN, red). (d) The proposed model effectively filtered ground points even at slope as true positives (TP, green).
• Figure 4: As the vehicle moves $t$ to $t+1$, prominent features maintain the structures in radar (yellow, cyan). The higher weight is allocated to correspondences within the consistent cluster (green line).
• Figure 6: (a) Qualitative analysis of the proposed odometry in LOOP_A0. Point cloud map based on our odometry shows well-alignment with the satellite image. (b) Sudden ego-velocity drift can occur due to large dynamic objects in LOOP_A0. While other radar-inertial baselines fail to generate accurate trajectories, our method effectively handles the challenging scenarios. (c) Detailed analysis in elevation on URBAN_A0, LOOP_A0, RURAL_A2, respectively. In RURAL_A2, EKF-RIO is omitted for clarity.