As autonomous vehicles (AV) have begun to populate roads, ensuring passenger experience and user safety has become essential. Just as the roads with the highest accident rates in traditional cars are unsignalized intersections without traffic lights and signs, the situation is the same for AVs. Therefore, it has become necessary to control the crossing of connected autonomous vehicles (CAV) at unsignalized crossroad intersections through communication technologies, such as vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I). In this context, a distributed model predictive control (DMPC) is proposed to enable multiple vehicles to cross intersections simultaneously, safely, and efficiently. Each agent locally and simultaneously solves a nonconvex optimal control problem while V2V communication allows for trajectory adjustments. A prioritization mechanism based on each agent’s time to react at potential collision points is implemented in order to facilitate parallel computation of this solution. To ensure feasible solutions for the non-convex problem, semidefinite programming relaxation with randomization is employed. Various two-vehicle and four-vehicle scenarios were created and tested in a simulation environment. In collaboration with the University of Sussex in the UK, this distributed approach was also compared with a centralized offline lane-free method addressing the same problem. Both methods were analyzed using different metrics, showing similar results in terms of safety, providing collision avoidance. However, DMPC demonstrated superior results in terms of passenger comfort and fuel consumption, while the centralized method enabled a faster intersection crossing. The DMPC’s adherence to the reference speed to prioritize passenger experience accounts for the slightly slower crossing time. Additionally, DMPC stands out for its adaptability to sudden changes through MPC and its independence from an intersection management system. These findings indicate that DMPC is a viable solution for the control of CAVs at unsignalized intersections, providing a safer, more comfortable, more environmentally friendly and less costly driving experience.
Man mano che i veicoli autonomi (AV) iniziano a circolare sulle strade, garantire una buona esperienza dei passeggeri e la sicurezza degli utenti è diventato essenziale. Secondo le statistiche le strade con i tassi di incidenti più elevati per i veicoli tradizionali sono le intersezioni non segnalate, e lo stesso vale per i AV. Di conseguenza, è necessario controllare accuratamente l'attraversamento di incroci non segnalati da parte di veicoli autonomi connessi (CAV), attraverso tecnologie di comunicazione veicolo-a-veicolo (V2V) e veicolo-a-infrastruttura (V2I). In questo contesto, si propone un controllo predittivo distribuito (DMPC) per consentire a più veicoli di attraversare le intersezioni in modo sicuro ed efficiente. Ogni agente risolve localmente e in parallelo un problema di controllo ottimale non convesso, mentre la comunicazione V2V consente aggiustamenti di traiettoria. Per facilitare il calcolo parallelo, è implementato un meccanismo di priorità basato sul tempo di reazione di ciascun agente nei punti di potenziale collisione. Per garantire soluzioni fattibili, si utilizza una tecnica di semidefinite programming relaxation con randomizzazione. Diversi scenari con due e quattro veicoli sono stati testati in una simulazione per valutare l’efficacia dell’algoritmo sviluppato. In collaborazione con l’Università del Sussex (UK), questo approccio distribuito è stato anche confrontato con un metodo centralizzato offline per lo stesso problema. Entrambi i metodi hanno dato risultati simili in termini di sicurezza, garantendo la prevenzione delle collisioni. Tuttavia, il DMPC ha mostrato migliori risultati in termini di comfort dei passeggeri e consumo di carburante, mentre il metodo centralizzato ha permesso un attraversamento più rapido. La rigorosa aderenza del DMPC alla velocità di riferimento spiega il suo tempo di attraversamento lievemente più lungo. Inoltre, il DMPC si distingue per la sua adattabilità a cambiamenti improvvisi tramite controllo predittivo e per la sua indipendenza da un sistema di gestione dell’incrocio. Questi risultati indicano che il DMPC è una soluzione valida per il controllo dei CAV agli incroci non segnalati, offrendo un’esperienza di guida più sicura, confortevole, ecologica e meno costosa.
Model predictive control design for connected autonomous vehicles under intersection traffic optimization
Arca, Mehmet Sinan
2023/2024
Abstract
As autonomous vehicles (AV) have begun to populate roads, ensuring passenger experience and user safety has become essential. Just as the roads with the highest accident rates in traditional cars are unsignalized intersections without traffic lights and signs, the situation is the same for AVs. Therefore, it has become necessary to control the crossing of connected autonomous vehicles (CAV) at unsignalized crossroad intersections through communication technologies, such as vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I). In this context, a distributed model predictive control (DMPC) is proposed to enable multiple vehicles to cross intersections simultaneously, safely, and efficiently. Each agent locally and simultaneously solves a nonconvex optimal control problem while V2V communication allows for trajectory adjustments. A prioritization mechanism based on each agent’s time to react at potential collision points is implemented in order to facilitate parallel computation of this solution. To ensure feasible solutions for the non-convex problem, semidefinite programming relaxation with randomization is employed. Various two-vehicle and four-vehicle scenarios were created and tested in a simulation environment. In collaboration with the University of Sussex in the UK, this distributed approach was also compared with a centralized offline lane-free method addressing the same problem. Both methods were analyzed using different metrics, showing similar results in terms of safety, providing collision avoidance. However, DMPC demonstrated superior results in terms of passenger comfort and fuel consumption, while the centralized method enabled a faster intersection crossing. The DMPC’s adherence to the reference speed to prioritize passenger experience accounts for the slightly slower crossing time. Additionally, DMPC stands out for its adaptability to sudden changes through MPC and its independence from an intersection management system. These findings indicate that DMPC is a viable solution for the control of CAVs at unsignalized intersections, providing a safer, more comfortable, more environmentally friendly and less costly driving experience.File | Dimensione | Formato | |
---|---|---|---|
2024_12_Arca_Thesis_01.pdf
solo utenti autorizzati a partire dal 11/11/2027
Descrizione: Thesis
Dimensione
6.64 MB
Formato
Adobe PDF
|
6.64 MB | Adobe PDF | Visualizza/Apri |
2024_12_Arca_ExecutiveSummary_02.pdf
solo utenti autorizzati a partire dal 11/11/2027
Descrizione: Executive Summary
Dimensione
3.39 MB
Formato
Adobe PDF
|
3.39 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/10589/229924