A Reliable Robot Localization Method Using LiDAR and GNSS Fusion Based on a Two-Step Particle Adjustment Strategy

Abstract

Accurate localization is essential for robot autonomous navigation. The localization methods that rely overly on the global navigation satellite system (GNSS) are not reliable in urban environments where GNSS signals are vulnerable to occlusion. In this work, we fuse data from IMU, LiDAR, and GNSS with a particle filter, presenting a novel method based on a two-step particle adjustment strategy. Our algorithm first uses GNSS data to evaluate the current particles and adjust their distribution if necessary. Subsequently, we use laser measurements to evaluate old particles and the reliability of the GNSS data, adjusting the particle distribution for correction. In addition, we use statistical features of point clouds for laser measurements, which transform the global map into a series of normal distribution models, and use these models to match with 3-D laser scans for particle state evaluation. Our method improves the processing efficiency of 3-D point cloud data and fully utilizes its 3-D features during localization. Experimental results demonstrate that our algorithm achieves higher localization accuracy on the publicly available KITTI dataset and in real campus environments. In addition, our algorithm consistently delivers precise localization in both open areas and GNSS-unavailable scenarios, showcasing superior reliability.