A Vision-Guided Hovering System for Autonomous UAV Package Pickup

Abstract

Reliable object picking is a key requirement for delivery drones, but precise hovering above a pickup point remains difficult in outdoor environments because of GPS inaccuracy, lighting variation, occlusions, and temporary target loss. This highlights the need for an integrated hovering system that can provide accurate and stable alignment for real-world object-picking tasks. To address this need, this study presents a vision-guided hovering system for autonomous package pickup using an unmanned aerial vehicle. The proposed approach uses a planar ArUco marker for target detection and relative pose estimation, and integrates a mission-level state machine with body-frame PID control to achieve closed-loop lateral, vertical, and yaw alignment during hover. The system further incorporates tilt-corrected altitude estimation and a hover-validation logic to improve alignment reliability before pickup. The method is validated through real-world flight experiments on a UAV platform. Experimental results show that, compared with the baseline P controller, the PID-based system improves hover quality, achieving a mean yaw error of 1.7 degrees and an altitude RMSE of 0.0572 m.