Evaluation of 3D LiDAR SLAM algorithms based on the KITTI dataset

Safe autonomous driving is the future trend, and achieving it requires precise and real-time simultaneous localization and mapping (SLAM). Many practitioners are concerned about the performance of LiDAR SLAM algorithms, but there is little research work to evaluate LiDAR SLAM algorithms specifically. This paper evaluates LeGO-LOAM, SC-LeGO-LOAM, LIO-SAM, SC-LIO-SAM, and FAST-LIO2 utilizing the 05-10 sequences from KITTI dataset. The experimental results show that: firstly, there is no significant difference among the absolute trajectory error of the five SLAM algorithms. Secondly, LeGO-LAOM has the smallest relative positional error among the six sequences. Thirdly, FAST-LIO2 has the best real-time performance. Our experiments are intended to provide a reference for practitioners in selecting SLAM algorithms.