PhD Defence by Riccardo Franceschini

Abstract

The need for efficient infrastructure inspection has grown due to significant annual investments required for maintenance of the aiging infrastructure across Europe. Traditional inspection methods are not only time-consuming but also hazardous, underscoring the necessity for faster robotic solutions that can perform these tasks more safely and efficiently. Unmanned Aerial Vehicles (UAVs) have emerged as versatile tools in this domain, but their widespread adoption hinges on the ease and efficiency of their operation where teloperation serves as de-facto the standard. For this reason the thesis introduces AeroAssistant, a framework designed to improve the teleoperation of UAVs, trought a combination of shared autonomy and augmented reality elements to ease the pilot enabling even non-expert pilot to control the UAV with confidence. Thus, the research addresses the problem from different aspects, creating the building blocks that will be used by AeroAssistant to provide a comprehensive enhanced teleoperation experience. First, a novel method for efficiently retrieving a path capable of optimizing a combination of environmental factors, which could range from obstacle distance and terrain structure to environmental conditions, is proposed.

Then, the research shifted to proposing a UAV interaction paradigm for aligning with and following any surfaces using a convenient interface that requires minimal user intervention. Subsequently, the effectiveness of a collaborative navigation experience was studied along with user evaluation, where the operator is only responsible for defining the direction and velocity of the UAV while the latter ensures collision-free navigation. Lastly, the latest advancements in artificial intelligence segmentation models are exploited to create an interaction in which the operator efficiently segments and inspects specific areas of interest, extracting a path capable of covering the entire inspection area with just a few clicks. Finally, AeroAssistant is presented by analyzing the core of the framework, highlighting its architecture and how the different features proposed in the published research are integrated as plugins along with other features. For each plugin, an explanation of the integration is provided with both the user interaction paradigm and the operator interface. The underlying features necessary for developing an effective augmented reality interaction are then proposed along with field experiments and ongoing research.

Supervisors

• Principal supervisor: Associate Professor Matteo Fumagalli, Department of Electrical and Photonics Engineering, DTU
• Co-supervisor: Professor Julian Cayero Becerra
• Co-supervisor: Professor Ole Ravn, Department of Electrical and Photonics Engineering, DTU

Evaluation Board

• Associate Professor Silvia tolu, Department of Electrical and Photonics Engineering, DTU
• Professor Emad Samuel Malki Ebeid, University of Southern Denmark, Denmark
• Professor Vincenzo Lippiello, University of Naples, Italy

Master of the Ceremony

• Associate Professor Dimitrios Papageorgiou, Department of Electrical and Photonics Engineering, DTU