A comprehensive comparison study of different 4D radar SLAM systems

Abstract

With the development of 4D Radar, it can replace LiDAR for localization and navigation in bad weather. In this dissertation, 4D radar SLAM technology is explored, and a novel ground extraction-based 4D radar point cloud filtering method is introduced to improve the quality of the input point cloud for the SLAM process. The study reviews and compares the major radar SLAM systems including reve-IMU, EKF-RIO, 4DRadarSLAM and Fast-LIO. For the 4DRadarSLAM system, this dissertation makes some improvements by adding ground constraints in the back-end optimization part. Comparative experiments with indoor and outdoor datasets demonstrate the superiority of the improved 4DRadarSLAM, especially in terms of accuracy, highlighting its potential in various environments.