In recent years, self-driving vehicles have become increasingly advanced technologically, reaching unprecedented levels of safety and efficiency. Through the integration of cutting-edge sensors and advanced algorithms, these vehicles can interpret complex environments in real-time. With the potential to reduce human errors that contribute to accidents, autonomous driving vehicles aim to revolutionize the automotive landscape. To achieve autonomous driving, a vehicle must be able to locate itself in its surroundings. The problem of vehicle localization is commonly addressed by relying on GNSS-RTK receivers. Widely adopted for their global reliability and precise accuracy in vehicle localization, these receivers, unfortunately, encounter limitations in tunnels, urban canyons, or areas with poor satellite visibility, where their precision decreases. This thesis addresses the vehicle localization problem in GNSS-limited environments proposing two different approaches. The first approach relies on road infrastructure, utilizing Ultra Wide Band antennas to identify the vehicle's global position. This information, along with data from other vehicle sensors, is fused to provide an accurate and reliable estimate of the vehicle's position. The second approach comprises a road lines detection algorithm based on LiDAR point clouds. A particle filter estimates the local vehicle position relative to the road structure by integrating the detected road lines with the vehicle velocities. The proposed localization methods are extensively tested offline using real data and validated in a real-world scenario with the vehicle autonomously driving on an open-to-traffic highway. The experimental results demonstrated that the precision of the vehicle position estimated with our localization methods is comparable to the one achieved by GNSS-RTK receivers.
Negli ultimi anni, i veicoli a guida autonoma sono diventati sempre più avanzati tecnologicamente, raggiungendo livelli di sicurezza ed efficienza senza precedenti. Grazie all'integrazione di sensori all'avanguardia e algoritmi avanzati, questi veicoli sono in grado di interpretare ambienti complessi in tempo reale. Con il potenziale di ridurre gli errori umani che contribuiscono agli incidenti, i veicoli a guida autonoma mirano a rivoluzionare il panorama automobilistico. Per abilitare la guida autonoma, un veicolo deve essere in grado di localizzarsi nell'ambiente circostante. Il problema della localizzazione del veicolo viene comunemente affrontato affidandosi ai ricevitori GNSS-RTK. Ampiamente adottati per la loro affidabilità globale e la precisione nella localizzazione dei veicoli, questi ricevitori, purtroppo, incontrano limitazioni in gallerie, canyon urbani o aree con scarsa visibilità satellitare, dove la loro precisione diminuisce. Questa tesi affronta il problema della localizzazione dei veicoli in ambienti con limitazioni GNSS proponendo due approcci diversi. Il primo approccio si basa sull'infrastruttura stradale, utilizzando antenne UWB per identificare la posizione globale del veicolo. Queste informazioni, insieme ai dati provenienti da altri sensori del veicolo, vengono fuse per fornire una stima accurata e affidabile della posizione del veicolo. Il secondo approccio comprende un algoritmo di rilevamento delle linee stradali basato sui sensori LiDAR. Un filtro a particelle stima la posizione locale del veicolo rispetto alla struttura stradale integrando le linee stradali rilevate con le velocità del veicolo. I metodi di localizzazione proposti sono stati ampiamente testati offline utilizzando dati reali e convalidati in uno scenario reale con il veicolo che guidava autonomamente su un'autostrada aperta al traffico. I risultati sperimentali hanno dimostrato che la precisione della posizione del veicolo stimata con i nostri metodi di localizzazione è paragonabile a quella ottenuta dai ricevitori GNSS-RTK.
Development of localization algorithms for a self-driving vehicle in GNSS-limited environments
CASTIGLIA, DANILO
2022/2023
Abstract
In recent years, self-driving vehicles have become increasingly advanced technologically, reaching unprecedented levels of safety and efficiency. Through the integration of cutting-edge sensors and advanced algorithms, these vehicles can interpret complex environments in real-time. With the potential to reduce human errors that contribute to accidents, autonomous driving vehicles aim to revolutionize the automotive landscape. To achieve autonomous driving, a vehicle must be able to locate itself in its surroundings. The problem of vehicle localization is commonly addressed by relying on GNSS-RTK receivers. Widely adopted for their global reliability and precise accuracy in vehicle localization, these receivers, unfortunately, encounter limitations in tunnels, urban canyons, or areas with poor satellite visibility, where their precision decreases. This thesis addresses the vehicle localization problem in GNSS-limited environments proposing two different approaches. The first approach relies on road infrastructure, utilizing Ultra Wide Band antennas to identify the vehicle's global position. This information, along with data from other vehicle sensors, is fused to provide an accurate and reliable estimate of the vehicle's position. The second approach comprises a road lines detection algorithm based on LiDAR point clouds. A particle filter estimates the local vehicle position relative to the road structure by integrating the detected road lines with the vehicle velocities. The proposed localization methods are extensively tested offline using real data and validated in a real-world scenario with the vehicle autonomously driving on an open-to-traffic highway. The experimental results demonstrated that the precision of the vehicle position estimated with our localization methods is comparable to the one achieved by GNSS-RTK receivers.File | Dimensione | Formato | |
---|---|---|---|
2023_12_Castiglia_Thesis.pdf
non accessibile
Descrizione: Elaborato Tesi
Dimensione
26.75 MB
Formato
Adobe PDF
|
26.75 MB | Adobe PDF | Visualizza/Apri |
2023_12_Castiglia_Executive_Summary.pdf
non accessibile
Descrizione: Executive Summary
Dimensione
722.25 kB
Formato
Adobe PDF
|
722.25 kB | 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/10589/215521