This paper presents an autonomous unmanned aerial vehicle (UAV)-based system for cleaning glass building facades. The proposed solution combines a real-time computer vision subsystem, built upon the YOLO (You Only Look Once) object detection architecture for accurate window localization, with an embedded control module that actuates high-pressure water nozzles to perform the cleaning task. A custom dataset composed of both real-world and synthetic images of glass facades was built and used to train and evaluate several YOLO models. Among them, YOLOv12 achieved the highest performance, attaining a mean Average Precision (mAP50) of 80% for the “glass window” class. The integrated system was thoroughly validated through simulations in Gazebo/ROS2 and real-world field experiments using a dedicated UAV platform (Skyclean drone), demonstrating reliable detection performance and safe operation. Results confirm that the approach effectively automates facade maintenance, significantly improving cleaning operations, reducing costs and human risk exposure, and providing a robust, extensible framework for future fully autonomous facade cleaning systems.
Monteiro et al. (Wed,) studied this question.