LiDAR-Inertial SLAM with DEM-driven Global Constraints for Planetary Rover Exploration

Abstract

The positioning capability of a rover is a critical factor that determines the efficiency of a planetary exploration missions. Most rovers primarily rely on visual odometry for trajectory estimation and relative pose determination. However, the distinctive characteristics of planetary environment, particularly limited visibility, can significantly compromise the accuracy of positioning. To address the limitations of visual localization techniques, this study introduces a novel LiDAR-Inertial SLAM framework integrating satellite Digital Elevation Model (DEM) data. A prior map is developed using DEM data through surface fitting after an offline process, followed by three online operational subparts: LiDAR odometry for estimate pose, 3D-Monte Carlo Localization (3D-MCL) adjust the estimate pose using the prior map, and graph optimization for final pose estimation. Experimental results indicate an average absolute trajectory error of approximately 0.6 meters, demonstrating the framework's effectiveness for long-distance navigation.