Critical Aspects of LiDAR-Based Detection of Floating Obstacles in Waves

Reliable perception of small floating obstacles is essential for autonomous maritime navigation. A RoboSense Ruby Plus LiDAR and ROS2-based perception pipeline are evaluated on a small environmental-monitoring USV. Calm water tests compare detection data of white and black floating balls, at different distances, in terms of target-point count, calibrated reflectivity, centroid range and sphere-fit geometry. Obstacles distance strongly affects LiDAR outputs; in addition, reflectivity and geometry reconstruction depend on colour, sampling density and incidence angle. The comparison between the two clustering methods, DBSCAN and fixed-radius Euclidian, showed similar results in well-sampled cases, but DBSCAN is more robust for sparse, low-reflectivity long-range returns. During wave-disturbed tests carried out at the towing tank of Naples, the same descriptors for target detection are considered and fluctuate with balls motion, depending on wave amplitude and frequency. The collected dataset provides a foundation for future developments involving object classification, sensor fusion and autonomous collision avoidance.