Localization Accuracy of Autonomous Mobile Robots: A Comprehensive Evaluation of KISS-ICP Odometry

Abstract

This paper emphasizes the importance of accurate localization for the appropriate behavior of autonomous mobile robots. In particular, it provides a rigorous evaluation of the accuracy of the KISS-ICP algorithm, a lightweight lidar-based pose estimation algorithm known for its simplicity with minimal setup parameters. The algorithm works only with lidar data; unlike other more sophisticated SLAM algorithms, it does not use IMUs. The algorithm is shown to work under a variety of conditions and sensors. To comprehensively evaluate how the algorithm performs under varying conditions, several sensors were used in the experiments. The evaluation included a series of 3D lidars, namely Ouster OS0, OS1, and OS2. These lidars are characterized by different field-of-view settings and operating modes with different point per row. In order to facilitate this evaluation, a data set with extensive robot traversals over a distance of more than 9 km in two different environments, each equipped with different sensor types, was prepared.