Answers are grounded in this paper and cite their sources.
Poster presented at the 9th Iberian Robotics Conference (ROBOT2026, Barcelona, Spain). Autonomous livestock herding in extensive farming environments requires robots to operate reliably across large outdoor areas subject to changing environmental conditions. In particular, flooding caused by heavy rainfall or irrigation can alter terrain traversability, making existing roads and pasture areas unsuitable for livestock movement. This work in progress explores the use of UAV imagery to provide updated terrain information for flood-aware livestock route planning. We use the FloodNet dataset, originally developed for post-disaster scene understanding, to train and evaluate two semantic segmentation architectures: DeepLabV3+ and YOLO11n-seg. Despite its original purpose, FloodNet provides relevant aerial imagery and flood-related semantic classes that can support the identification of potentially unsafe terrain. The experimental comparison considers both segmentation quality and inference speed, addressing the different requirements of accurate environmental perception and computational efficiency in UAV-based systems. The resulting segmentation maps distinguish flooded and non-flooded areas, as well as relevant terrain categories, providing spatial information that can be incorporated into geographic planning algorithms. This approach aims to improve the ability of autonomous herding systems to respond to temporary environmental changes without relying exclusively on previously generated
No takes yet. Share an insight, caveat, or question.
Miñambres-Calvo et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: