Ground Robot Detection for Collaborative UAV-UGV Alignment and Retrieval Operations

Volcanologists face considerable challenges in ex-ploring and monitoring harsh and dangerous environments like volcanoes. To gather essential data, they use robots, such as drones and mobile robots. However, drones have limited capacity and flight time, while mobile robots face mobility constraints. Combining drones and mobile robots has emerged as an inno-vative technology for long-range monitoring of volcanoes. This approach allows for fast deployment of ground robots using transport drones. However, the retrieval operation of the ground robot is challenging due to inaccurate drone positioning. A visual technique for detecting and aligning the ground robot is crucial during this procedure. In this study, the detection of the ground robot utilizing a drone was investigated. The drone-based detection approach involved (i) capturing ground robot images from various conditions, (ii) training a robust detection model, and (iii) assessing its performance. Furthermore, this study employed the You Only Look Once (YOLOv8) architecture for ground robot detection. Three models were investigated: the plate detection model, the ground robot detection model, and the model for both ground robot and plate detection. The trained ground robot detection model achieved a robust detection performance garnering the precision of 98.9% with a respectable recall of 100%, F1 score of 99.94%, and 98.6% mean average precision, compared to the other two trained models. The trained detection model presented a practical approach, which produced good performance and could predict and segment the ground robot during the alignment procedure for the retrieval operation by the drone.