The advent of autonomous driving is set to revolutionize the future of mobility, enhancing safety and efficiency across densely populated areas. Within this vision, the AIDA project focuses on developing autonomous driving technologies tailored for urban environments, where the ability to navigate complex, dynamic scenarios is of paramount importance. The Longitudinal Planner represents a crucial element of the software architecture, responsible for generating the vehicle's reference speed profile over a predefined route. The optimal velocity and acceleration values are determined at each instant, taking into account the limitations imposed by traffic regulations, road characteristics and other contextual factors. Beyond merely generating a viable speed profile, the Planner also addresses passenger comfort and minimizes the strain on vehicle actuators. To this end, a Quadratic Programming (QP) optimization problem is formulated, incorporating a cost function and a set of constraints designed to ensure smooth speed transitions. The complexity of urban environments is further enhanced by the presence of dynamic elements such as traffic signals, intersections and pedestrian crossings, which may necessitate stopping based on their real-time status. An algorithmic tactical evaluation integrated with the onboard supervisor inputs enables the Planner to address stop-and-go situations in a seamless manner, integrating them into the vehicle's speed profile. Given the need for real-time responsiveness, computational efficiency has been a key focus of this Thesis to ensure the Planner meets the architecture's performance demands. The system was validated through experimental testing on both closed tracks and in high-profile automotive events, including the 2024 1000 Miglia and Vallecamonica Hill Climb. These trials validated the Planner’s robustness and reliability in complex scenarios, highlighting its capability for effective navigation in real-world urban environments and establishing a foundation for future advancements in autonomous mobility.
L'avvento della guida autonoma promette di rivoluzionare il futuro della mobilità in aree densamente popolate, migliorando sicurezza ed efficienza. In quest'ottica, il progetto AIDA si concentra sullo sviluppo di tecnologie specifiche per la guida autonoma in contesti urbani, dove la capacità di affrontare scenari complessi e dinamici è fondamentale. Il Planner Longitudinale è un elemento cruciale dell'architettura software, con il compito di generare il profilo di velocità di riferimento del veicolo lungo un percorso prestabilito. In ogni istante vengono determinati i valori ottimali di velocità e accelerazione, tenendo conto delle limitazioni imposte da normative sul traffico, dalle caratteristiche della strada e da altri fattori contestuali. Oltre a generare un profilo di velocità valido, il Planner considera anche il comfort dei passeggeri e minimizza lo stress sugli attuatori del veicolo. A tal fine viene formulato un problema di ottimizzazione mediante Programmazione Quadratica che include una funzione di costo e un insieme di vincoli destinati a garantire transizioni di velocità fluide. La complessità degli scenari urbani è ulteriormente amplificata dalla presenza di elementi dinamici come semafori, incroci e attraversamenti pedonali, che possono richiedere uno stop in base al loro stato in tempo reale. Un algoritmo di valutazione tattica, integrato con gli input del supervisore di bordo, consente di gestire in modo efficace le situazioni di stop-and-go, incorporandole nel profilo di velocità. Considerata la necessità di eseguire l'algoritmo in tempo reale, l’efficienza computazionale è stata un aspetto chiave affrontato in questa Tesi per garantire che il Planner Longitudinale soddisfi i requisiti prestazionali dell’architettura software. Il sistema è stato validato tramite test sperimentali su piste chiuse e in eventi automobilistici di rilievo, come la Mille Miglia 2024 e la Cronoscalata della Vallecamonica. Questi test hanno confermato la robustezza e l’affidabilità del Planner in scenari complessi, evidenziandone la capacità di navigare efficacemente in ambienti di realtà urbana e gettando le basi per futuri sviluppi nel campo della mobilità autonoma.
Real-time longitudinal planning for autonomous driving through QP optimization and tactical management in urban scenarios
GALIMBERTI, ANDREA
2023/2024
Abstract
The advent of autonomous driving is set to revolutionize the future of mobility, enhancing safety and efficiency across densely populated areas. Within this vision, the AIDA project focuses on developing autonomous driving technologies tailored for urban environments, where the ability to navigate complex, dynamic scenarios is of paramount importance. The Longitudinal Planner represents a crucial element of the software architecture, responsible for generating the vehicle's reference speed profile over a predefined route. The optimal velocity and acceleration values are determined at each instant, taking into account the limitations imposed by traffic regulations, road characteristics and other contextual factors. Beyond merely generating a viable speed profile, the Planner also addresses passenger comfort and minimizes the strain on vehicle actuators. To this end, a Quadratic Programming (QP) optimization problem is formulated, incorporating a cost function and a set of constraints designed to ensure smooth speed transitions. The complexity of urban environments is further enhanced by the presence of dynamic elements such as traffic signals, intersections and pedestrian crossings, which may necessitate stopping based on their real-time status. An algorithmic tactical evaluation integrated with the onboard supervisor inputs enables the Planner to address stop-and-go situations in a seamless manner, integrating them into the vehicle's speed profile. Given the need for real-time responsiveness, computational efficiency has been a key focus of this Thesis to ensure the Planner meets the architecture's performance demands. The system was validated through experimental testing on both closed tracks and in high-profile automotive events, including the 2024 1000 Miglia and Vallecamonica Hill Climb. These trials validated the Planner’s robustness and reliability in complex scenarios, highlighting its capability for effective navigation in real-world urban environments and establishing a foundation for future advancements in autonomous mobility.File | Dimensione | Formato | |
---|---|---|---|
2024_12_Galimberti_Tesi.pdf
non accessibile
Descrizione: Tesi
Dimensione
71.86 MB
Formato
Adobe PDF
|
71.86 MB | Adobe PDF | Visualizza/Apri |
2024_12_Galimberti_Executive_Summary.pdf
non accessibile
Descrizione: Executive Summary
Dimensione
5.86 MB
Formato
Adobe PDF
|
5.86 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/10589/230546