This repo serves as the replication package for the paper "Robustness of LiDAR-Based Pose Estimation: Evaluating and Improving Odometry and Localization Under Common Point Cloud Corruptions". We generated 18 types of corruptions for KITTI Odometry LiDAR scans and evaluated 4 SOTA LiDAR-only SLAM systems and one SOTA LiDAR-based global localization system.
You can find the arXiv preprint of the paper at https://arxiv.org/abs/2409.10824.
Since the generated LiDAR corruption data is too large, it is impractical to share them directly. Please refer to our code for generating corrupted LiDAR data.
In this work, we studied the robustness of four LiDAR-only SLAM systems, MULLS, kiss-icp, Delora and NeRF-LOAM, and one LiDAR-based map-based global localization, LocNDF.
The instructions for replicating the evaluation in our paper, as well as the necessary code are provided under the corresponding folder.
We uploaded the raw results, i.e., the poses produced by subject systems, and the evaluation results, i.e., relative pose errors (RPE), under the corresponding folder.
The figures used in the paper are generated at https://colab.research.google.com/drive/1KWtEsUrdcGOSuh7NxSEXOpzioMnWu1JQ?usp=sharing.
The evaluation code and scripts are under eval/.