4D Radar-Based Pose Graph SLAM With Ego-Velocity Pre-Integration Factor

4D imaging radars (4D radars) provide point clouds with range, azimuth, elevation as well as Doppler velocity. They are much cheaper sensors than LiDARs and can operate under extreme weather conditions. However, its drawbacks of high noise and sparsity would pose great challenges for SLAM. In this paper, we present a 4D radar-based SLAM framework based on pose graph optimization. In order to get a cleaner radar point cloud for registration, the raw 4D radar data is first filtered to reduce ghost and random noise. Next, we estimate the linear and angular ego-velocity using the Doppler velocity. Based on this, we design a new ego-velocity pre-integration factor for pose graph optimization to achieve more accurate and robust pose estimation. Finally, a real-world dataset is collected in different challenging environments. The experimental results demonstrate the precision and robustness of our proposed framework.

Automated summary

This paper presents a 4D radar-based SLAM framework that uses pose graph optimization to achieve accurate and robust pose estimation. The framework filters the raw radar data to reduce noise and estimates ego-velocity to improve registration accuracy. Experimental results demonstrate the precision and robustness of the proposed framework.