Cultivating Precision: Unmanned Aerial Vehicle Retrieval Enhancement through Computer Vision in Ground Robot Recovery Missions

Robot exploration is a rapidly advancing domain within robotics, with applications ranging from disaster relief to space exploration. Unmanned aerial vehicles (UAVs), commonly called drones, have become essential tools in these scenarios due to their agility and extensive workspace coverage. However, their limited battery life remains a challenge. To address this, collaborations between UAVs and ground robots have gained traction. This article focuses on the retrieval operation pro-cess in such collaborative systems. Specifically, it addresses the localization of the retrieval component attached to the drone. Prior study used ArUco markers and computer vision to detect the drone's position. However, due to the environmental factors, such as sunlight, makes it hard to operate outside and rely on this method. Thus, this study aims to address this with three key objectives: designing and fabricating a prototype retrieval component, training object detection model using YOLOv4-tiny, and evaluating the performance of the retrieval mechanism process. The results showcase the system's adeptness in detecting the retrieval component using the trained model with an accuracy of 89%, achieving a mere 1.41% average error in area estimation and a commendable 1.28% average error in estimating the retrieval component's distance to the ground robot's center. These findings underscore the system's precision, ith potential implications for diverse real-world scenarios.