Autonomous Underwater Vehicles (AUVs) face critical operational limitations in localization and communication, challenges that can be overcome through a cooperative paradigm with an Unmanned Surface Vehicle (USV). The viability of this approach, however, is entirely dependent on the USV’s capacity for robust, autonomous navigation. This thesis addresses this requirement by presenting the end-to-end design and implementation of a navigation and control stack for the BlueBoat USV, tailored for the mission of closely following an AUV while circumnavigating obstacles in unknown and cluttered environments. To rigorously develop and validate this system, a foundational contribution of this work is the development from scratch of a high-fidelity simulation environment in Gazebo, which integrates realistic hydrodynamics, buoyancy, and environmental disturbances including waves, wind, and current. This platform served as the essential testbed for designing and rigorously validating the system’s perception and planning architecture. The system’s perceptual layer is founded on a rigorous comparative analysis of three distinct 3D LiDAR processing paradigms: a voxel-based pipeline employing RANSAC and DBSCAN for high-fidelity object segmentation, a computationally efficient pillar-based architecture, and a probabilistic occupancy mapping approach for representing complex non-convex structures. For long-range guidance, the framework is centered on a state-of-the-art Real-Time Rapidly-exploring Random Tree* (RT-RRT*) planner. This high-level layer guides reactive local planners, including implementations of the Vector Field Histogram (VFH) and an optimization-based method inspired by FASTER. To establish a comprehensive benchmark, these architectures are systematically compared against other configurations using planners like Improved Particle Swarm Optimization (IPSO) and the Dynamic Window Approach (DWA), with performance evaluated under both PID and Sliding Mode Control (SMC) low-level controllers. Extensive validation within the developed simulation environment confirms the distinct operational advantages of each perception strategy and demonstrates the superior, well-balanced performance of the hybrid architectures built upon RT-RRT*.
I Veicoli Autonomi Sottomarini (AUV) affrontano critiche limitazioni operative in termini di localizzazione e comunicazione, sfide che possono essere superate attraverso un paradigma cooperativo con un Veicolo di Superficie Autonomo (USV). La fattibilità di tale approccio, tuttavia, dipende interamente dalla capacità dell'USV di navigare in modo robusto e autonomo. Questa tesi affronta tale requisito presentando la progettazione e l'implementazione end-to-end di uno stack di navigazione e controllo per il BlueBoat USV, concepito per la missione di seguire da vicino un AUV, aggirando ostacoli in ambienti ricchi di ostacoli e sconosciuti. Per sviluppare e validare rigorosamente questo sistema, un contributo fondamentale di questo lavoro è la creazione da zero di un ambiente di simulazione ad alta fedeltà in Gazebo, che integra idrodinamica, galleggiamento e disturbi ambientali realistici, tra cui onde, vento e corrente. Tale piattaforma è servita come banco di prova essenziale per la progettazione e la validazione rigorosa dell'architettura di percezione e pianificazione del sistema. Il livello percettivo del sistema si basa su un'analisi comparativa rigorosa di tre distinti paradigmi di elaborazione LiDAR 3D: una pipeline basata su voxel che utilizza RANSAC e DBSCAN per la segmentazione di oggetti ad alta fedeltà, un'architettura basata su pillar computazionalmente efficiente e un approccio di mappatura probabilistica di occupazione per rappresentare strutture complesse non convesse. Per la guida a lungo raggio, il framework è incentrato su un pianificatore allo stato dell'arte Real-Time Rapidly-exploring Random Tree* (RT-RRT*). Questo livello di alto livello guida pianificatori locali reattivi, includendo implementazioni del Vector Field Histogram (VFH) e un metodo basato sull'ottimizzazione ispirato a FASTER. Per stabilire un benchmark completo, queste architetture sono confrontate sistematicamente con altre configurazioni che utilizzano pianificatori come l'Improved Particle Swarm Optimization (IPSO) e il Dynamic Window Approach (DWA), con prestazioni valutate sotto il controllo di basso livello sia di tipo PID che Sliding Mode Control (SMC). Una validazione estensiva all'interno dell'ambiente di simulazione sviluppato conferma i distinti vantaggi operativi di ciascuna strategia di percezione e dimostra le prestazioni superiori e ben bilanciate delle architetture ibride basate su RT-RRT*.
Real-Time Motion Planning for a USV Tracking an AUV through Cluttered Unknown Environments
ALBIERO, VITTORIO
2024/2025
Abstract
Autonomous Underwater Vehicles (AUVs) face critical operational limitations in localization and communication, challenges that can be overcome through a cooperative paradigm with an Unmanned Surface Vehicle (USV). The viability of this approach, however, is entirely dependent on the USV’s capacity for robust, autonomous navigation. This thesis addresses this requirement by presenting the end-to-end design and implementation of a navigation and control stack for the BlueBoat USV, tailored for the mission of closely following an AUV while circumnavigating obstacles in unknown and cluttered environments. To rigorously develop and validate this system, a foundational contribution of this work is the development from scratch of a high-fidelity simulation environment in Gazebo, which integrates realistic hydrodynamics, buoyancy, and environmental disturbances including waves, wind, and current. This platform served as the essential testbed for designing and rigorously validating the system’s perception and planning architecture. The system’s perceptual layer is founded on a rigorous comparative analysis of three distinct 3D LiDAR processing paradigms: a voxel-based pipeline employing RANSAC and DBSCAN for high-fidelity object segmentation, a computationally efficient pillar-based architecture, and a probabilistic occupancy mapping approach for representing complex non-convex structures. For long-range guidance, the framework is centered on a state-of-the-art Real-Time Rapidly-exploring Random Tree* (RT-RRT*) planner. This high-level layer guides reactive local planners, including implementations of the Vector Field Histogram (VFH) and an optimization-based method inspired by FASTER. To establish a comprehensive benchmark, these architectures are systematically compared against other configurations using planners like Improved Particle Swarm Optimization (IPSO) and the Dynamic Window Approach (DWA), with performance evaluated under both PID and Sliding Mode Control (SMC) low-level controllers. Extensive validation within the developed simulation environment confirms the distinct operational advantages of each perception strategy and demonstrates the superior, well-balanced performance of the hybrid architectures built upon RT-RRT*.| File | Dimensione | Formato | |
|---|---|---|---|
|
Albiero_Vittorio.pdf
accesso aperto
Dimensione
20.91 MB
Formato
Adobe PDF
|
20.91 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/93331