Autonomous vehicles are expected to revolutionize transportation by enhancing road safety, sustainability, and accessibility. This thesis presents innovative solutions to two pivotal challenges in autonomous driving: large-scale mapping and vehicle localization. The aim of this work is the development of a mapping procedure capable of delineating a viable trajectory for an autonomous vehicle to follow over expansive distances, specifically the MilleMiglia route. Unlike conventional methods that rely on high-definition, proprietary maps, this study exclusively utilizes publicly accessible data from OpenStreetMap. This approach ensures scalability and adaptability across various geographic locations without incurring significant costs. Complementing the mapping process, the thesis introduces a novel particle filter architecture aimed at aligning the generated trajectory with the vehicle’s real-time pose. The proposed architecture matches the Localization Map, a representation of broad road geometry data, with the Explored Area, which details the vehicle’s immediate surroundings in a Bird’s Eye View. This alignment is critical for seamless integration within the autonomous vehicle's software stack and ensures that the vehicle can effectively interpret and navigate its environment. This research delivers a comprehensive pipeline for addressing localization and mapping in large-scale scenarios by capitalizing on the vast repository of open-source geographical data. The resulting framework demonstrates the potential of publicly available maps in supporting the operational needs of autonomous vehicles, paving the way for their widespread adoption in complex urban and suburban environments.
I veicoli autonomi stanno diventando sempre più centrali nel panorama attuale grazie alla loro capacità di rispondere a sfide importanti dei nostri sistemi di trasporto, come la sicurezza stradale, la sostenibilità ambientale e la mobilità individuale. Questa tesi affronta il problema della mappatura su larga scala con l'obiettivo di costruire un percorso navigabile da un veicolo autonomo. Il contributo principale consiste nello sviluppo di una procedura che funzioni su un scenario molto ampio, specificamente su tutto il percorso della MilleMiglia, utilizzando solo mappe open-source come OpenStreetMap. Il secondo obiettivo del lavoro è l'allineamento della traiettoria generata alla posizione del veicolo, in modo che possa essere utilizzata all'interno del software di guida unitamente al modulo di Pianificazione. È stata sviluppata una nuova architettura basata su metodi di localizzazione Monte Carlo, che cerca di far corrispondere la Mappa di Localizzazione, ovvero una mappa con informazioni generiche sulla struttura delle strade, con l'Area Esplorata, una griglia che rappresenta l'ambiente circostante il veicolo in vista dall'alto. In conclusione, questa tesi presenta una procedura completa per affrontare sia il problema della localizzazione che della mappatura in scenari su larga scala, sfruttando unicamente informazioni generiche della rete stradale provenienti da OpenStreetMap.
Large-scale localization and mapping for autonomous driving in urban scenarios using publicly available maps
Giacalone, Alberto
2023/2024
Abstract
Autonomous vehicles are expected to revolutionize transportation by enhancing road safety, sustainability, and accessibility. This thesis presents innovative solutions to two pivotal challenges in autonomous driving: large-scale mapping and vehicle localization. The aim of this work is the development of a mapping procedure capable of delineating a viable trajectory for an autonomous vehicle to follow over expansive distances, specifically the MilleMiglia route. Unlike conventional methods that rely on high-definition, proprietary maps, this study exclusively utilizes publicly accessible data from OpenStreetMap. This approach ensures scalability and adaptability across various geographic locations without incurring significant costs. Complementing the mapping process, the thesis introduces a novel particle filter architecture aimed at aligning the generated trajectory with the vehicle’s real-time pose. The proposed architecture matches the Localization Map, a representation of broad road geometry data, with the Explored Area, which details the vehicle’s immediate surroundings in a Bird’s Eye View. This alignment is critical for seamless integration within the autonomous vehicle's software stack and ensures that the vehicle can effectively interpret and navigate its environment. This research delivers a comprehensive pipeline for addressing localization and mapping in large-scale scenarios by capitalizing on the vast repository of open-source geographical data. The resulting framework demonstrates the potential of publicly available maps in supporting the operational needs of autonomous vehicles, paving the way for their widespread adoption in complex urban and suburban environments.File | Dimensione | Formato | |
---|---|---|---|
2024_04_Giacalone_Executive_Summary.pdf
non accessibile
Descrizione: Executive Summary
Dimensione
1.59 MB
Formato
Adobe PDF
|
1.59 MB | Adobe PDF | Visualizza/Apri |
2024_04_Giacalone.pdf
non accessibile
Descrizione: Thesis
Dimensione
20.18 MB
Formato
Adobe PDF
|
20.18 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/218205