Contribution

LiDAR-Based Ground Segmentation with Structured Point Clouds for Multi-Sensor AMRs

---

Abstract

LiDAR-based perception is a popular component of autonomous mobile robots (AMRs) for obstacle avoidance and traversable area detection. Traditional ground Segmentation approaches, such as ring-based methods, often assume a fixed sensor placement and may struggle in multi-LiDAR or tilted sensor configurations. To overcome these limitations, we propose a novel segmentation approach based on the organized point cloud representation, which preserves the spatial arrangement of LiDAR data in a structured 2D format. Our method first organizes the raw point cloud into a structured array, ensuring direct neighborhood accessibility without additional spatial searches. We then use a rolling window over the Array to estimate surface normal vectors. Ground segmentation is performed iteratively by classifying normal vectors based on orientation and height consistency. A likelihood approach is further utilized to segment points by assigning them to their corresponding normal vectors. Furthermore, we evaluate our method through experimental tests on a real-world multi- LiDAR AMR in five different scenarios within unstructured environments, achieving an average accuracy of 0.939

Keywords: Mobile Robots, Scene Understanding, Off- Road Navigation

How to Cite:

Didari, H. & Steinbauer-Wagner, G., (2025) “LiDAR-Based Ground Segmentation with Structured Point Clouds for Multi-Sensor AMRs”, ARW Proceedings 25(1), 97-102. doi: https://doi.org/10.34749/3061-0710.2025.16