Autonomous UAV framework for precise localization and mid-air item exchange

<p dir="ltr"><b>Thesis Statement</b>: This thesis proposes a vision-based framework for <b>autonomous </b><b>mid-air item exchange</b> between two quadcopters, ensuring <b>precise localization</b>, stability, and efficiency. </p><p dir="ltr">Uncrewed Aerial Vehicles (UAVs), particularly multirotors, are increasingly deployed in applications such as aerial photography, parcel delivery, and search-and-rescue due to their ability to hover and operate in confined spaces. As UAV applications expand, the need for controlled physical interaction with the environment grows, enabling novel capabilities that extend flight endurance, increase delivery range, and facilitate operations in complex environments. </p><p dir="ltr">This dissertation presents a novel autonomous framework to enhance the localisation and mid-air docking of multirotor UAVs. It focuses on precise localisation and mid-air item exchange. It introduces an integrated approach that combines modular UAV systems with advanced control, computer vision, and positioning techniques to enable autonomous mid-air item transfers. Initial experiments with a commercial offthe- shelf (COTS) UAV highlight critical concepts in aerial robotics, vertical takeoff and landing (VTOL) systems, flight dynamics, and localisation. By employing vision-based positioning systems and fiducial markers, we enhance horizontal and vertical positioning accuracy, laying the groundwork for developing custom UAVs with more advanced autonomy. </p><p dir="ltr">Subsequent work involves designing and testing a custom UAV with an advanced flight controller and onboard computing, enabling sophisticated computer vision and precise outdoor positioning through GPS sensor fusion and ArUco marker-based landing techniques. The research addresses control challenges and operational design, culminating in a multi-UAV system capable of mid-air docking and dynamic item exchanges. An essential contribution is developing a mechatronic system for mid-air docking, utilising a novel visual cross-marker technique that improves GPS efficiency and enhances docking precision in flight. The system analyses challenges such as downwash effects, drone alignment during proximity flight, and the impact of payload placement on flight dynamics. Computational fluid dynamics (CFD) analysis and propeller sizing studies validate the design. The research also introduces a synchronisation technique for dynamic mid-air docking. It uses a ‘Truncated Marker’ approach to achieve centimetre-level precision even during high-velocity maneuvers. This capability is crucial for multi-drone collaborative operations, such as long-range delivery or aerial monitoring, where precise coordination is required. </p><p dir="ltr">A computer vision-based method for enabling mid-air close proximity flight is proposed. This method leverages onboard sensors to provide relative pose estimates for precise mid-air docking. Experimental validation demonstrates centimetre-level docking precision in dynamic conditions, showing the feasibility of continuous in-flight battery swaps. This work demonstrates the potential for increasing UAV autonomy, operational range, and efficiency through experimental validation and hardware testing. It will pave the way for future developments in mid-air item exchange and collaborative multi-UAV operations.</p>