This thesis presents the design and implementation of a ROS2-based visual data pipeline to support autonomous docking functionalities in small-scale robotic boats. The proposed system focuses on real-time image acquisition using the Raspberry Pi Camera Module 3, integrated via a custom Python ROS2 node running on a Raspberry Pi 5. The captured images are transmitted to the “Open Water” segmentation algorithm, developed with PyTorch by a member of the Autodocking team, which classifies image regions into water, sky and urban elements. Due to the computational limitations of the Raspberry Pi 5, the architecture was progressively evolved to delegate processing tasks to an NVIDIA Jetson Nano, introducing a modular and distributed design. To address system compatibility, Docker containers were used to harmonize dependencies across heterogeneous hardware and operating systems. Field testing in Padua and Chioggia validated system performance in real-world aquatic environments, highlighting key challenges such as thermal management and segmentation inaccuracies caused by environmental reflections. Although full GPU acceleration on the Jetson Nano was not achieved, the groundwork has been laid for future improvements. The system’s modularity and documentation aim to facilitate further development toward autonomous navigation and docking.
Questa tesi presenta la progettazione e l’implementazione di una pipeline di elaborazione visiva basata su ROS2, finalizzata a supportare funzionalità di ormeggio autonomo in imbarcazioni robotiche di piccola scala. Il sistema proposto si concentra sull’acquisizione in tempo reale di immagini tramite il Raspberry Pi Camera Module 3, integrato mediante un nodo ROS2 personalizzato scritto in Python ed eseguito su una Raspberry Pi 5. Le immagini acquisite vengono trasmesse all’algoritmo di segmentazione “Open Water”, sviluppato in PyTorch da un membro del team di Autodocking, il quale classifica le regioni dell’immagine in acqua, cielo ed elementi urbani. A causa delle limitazioni computazionali della Raspberry Pi 5, l’architettura è stata progressivamente evoluta per delegare le operazioni di elaborazione a una NVIDIA Jetson Nano, introducendo un design modulare e distribuito. Per garantire la compatibilità del sistema, sono stati utilizzati container Docker al fine di armonizzare le dipendenze tra hardware ed ambienti operativi eterogenei. I test condotti a Padova e Chioggia hanno validato le prestazioni del sistema in ambienti acquatici reali, evidenziando sfide chiave come la gestione termica e le imprecisioni nella segmentazione causate dai riflessi ambientali. Sebbene non sia stata raggiunta una piena accelerazione GPU sulla Jetson Nano, sono state poste le basi per futuri miglioramenti. La modularità e la documentazione del sistema mirano a facilitare lo sviluppo verso una navigazione e un ormeggio completamente autonomi.
Progettazione e Implementazione di una Pipeline Visiva basata su ROS2 per l'Ormeggio Autonomo: Integrazione e Valutazione delle Prestazioni su Raspberry Pi 5 e Jetson Nano
PAVANETTO, MARCO
2024/2025
Abstract
This thesis presents the design and implementation of a ROS2-based visual data pipeline to support autonomous docking functionalities in small-scale robotic boats. The proposed system focuses on real-time image acquisition using the Raspberry Pi Camera Module 3, integrated via a custom Python ROS2 node running on a Raspberry Pi 5. The captured images are transmitted to the “Open Water” segmentation algorithm, developed with PyTorch by a member of the Autodocking team, which classifies image regions into water, sky and urban elements. Due to the computational limitations of the Raspberry Pi 5, the architecture was progressively evolved to delegate processing tasks to an NVIDIA Jetson Nano, introducing a modular and distributed design. To address system compatibility, Docker containers were used to harmonize dependencies across heterogeneous hardware and operating systems. Field testing in Padua and Chioggia validated system performance in real-world aquatic environments, highlighting key challenges such as thermal management and segmentation inaccuracies caused by environmental reflections. Although full GPU acceleration on the Jetson Nano was not achieved, the groundwork has been laid for future improvements. The system’s modularity and documentation aim to facilitate further development toward autonomous navigation and docking.| File | Dimensione | Formato | |
|---|---|---|---|
|
Pavanetto_Marco.pdf
accesso aperto
Dimensione
1.06 MB
Formato
Adobe PDF
|
1.06 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/20.500.12608/89672