In recent years, autonomous driving has emerged as one of the most transformative and multidisciplinary challenges in robotics and control engineering. Enabling a vehicle to navigate complex, dynamic, and uncertain environments without human intervention demands seamless integration of perception, planning, decision making, and control. Among these components, the design of a robust and safe control layer is fundamental: it must ensure collision avoidance, adherence to vehicle constraints, and responsiveness to unpredictable agents such as pedestrians or other vehicles. In urban settings especially, the density of obstacles, frequent intersections, occlusions, and highly variable traffic dynamics raise significant difficulties for classical control architectures. While many approaches focus on perception, mapping, or high-level planning, this work concentrates on the control problem. To achieve this goal, different types of Control Barrier Functions (CBFs) have been adopted to encode safety constraints and represent the interaction between the vehicle and surrounding obstacles. The use of pairwise CBFs allows one to handle multiple dynamic agents and obstacles, effectively mapping their relative behavior and enforcing collision avoidance in simulation. However, when transitioning from a kinematic to a dynamic vehicle model, necessary to represent higher speed and more realistic scenarios, the standard CBF formulation becomes insufficient. In such cases, Higher Order Control Barrier Functions (HOCBFs) are employed to incorporate higher order dynamics. Nonetheless, HOCBF based formulations often face limitations in feasibility when control inputs are bounded like in the real world. To overcome these issues, this research explores the use of viability theory, which inherently guarantees the existence of safe control actions that keep the system trajectories within a viable set. Although conceptually powerful, direct implementation of viability based approaches is computationally demanding and unsuitable for real time control. Therefore two works have been developed to approximate the viability kernel and exploit its safety guarantees in a more efficient manner. The first decoupled the dynamic bicycle model to make it affine and the second leverages the system’s differential flatness to significantly reduce the complexity of the control problem. Since many robotic systems are differentially flat, this solution can be a new approach to transform the original nonlinear system into a lower-dimensional representation, where safety and viability constraints can be enforced more efficiently. The proposed framework demonstrates that it is possible to combine the formal guarantees of control barrier functions with the generality of viability theory while maintaining computational tractability. This hybrid approach ensures both safety and feasibility in autonomous driving tasks, even under dynamic and uncertain urban conditions. The developed control algorithms have been extensively validated both in a simulation environment and through real-world experimental tests, confirming the effectiveness and robustness of the proposed methods. The resulting control architecture provides a viable foundation for future extensions toward cooperative multi vehicle scenarios and real world experimental validation.

Negli ultimi anni, la guida autonoma si è affermata come una delle sfide più complesse e multidisciplinari nell’ambito della robotica e dell’ingegneria del controllo. Consentire a un veicolo di muoversi in modo autonomo in ambienti complessi, dinamici e incerti richiede un’integrazione armoniosa tra percezione, pianificazione, decisione e controllo. Tra questi elementi, la progettazione di un livello di controllo sicuro e robusto è cruciale: esso deve garantire l’evitamento delle collisioni, il rispetto dei vincoli cinematici e dinamici del veicolo e una risposta efficace alla presenza di agenti imprevedibili, come pedoni o altri veicoli. In particolare, negli scenari urbani, l’elevata densità di ostacoli, la presenza di incroci frequenti, le occlusioni e la variabilità del traffico rappresentano una sfida significativa per le architetture di controllo tradizionali. Mentre gran parte della ricerca si concentra su percezione, mappatura o pianificazione ad alto livello, questo lavoro affronta direttamente il problema del controllo. A tale scopo, sono state adottate diverse Control Barrier Functions (CBF) per modellare i vincoli di sicurezza e descrivere l’interazione tra il veicolo e l’ambiente circostante. L’uso di pairwise CBF consente di gestire più agenti dinamici contemporaneamente, descrivendone il comportamento relativo e garantendo l’evitamento delle collisioni in simulazione. Tuttavia, passando da un modello cinematico a uno dinamico, necessario per rappresentare scenari più realistici e a velocità elevate, la formulazione standard delle CBF risulta spesso inadeguata. In questi casi, si ricorre alle Higher Order Control Barrier Functions (HOCBF), che tengono conto delle dinamiche di ordine superiore, ma che a loro volta presentano limiti di fattibilità quando gli ingressi di controllo sono vincolati, come avviene nel mondo reale. Per superare tali limiti, questa ricerca esplora l’impiego della viability theory, che fornisce garanzie formali sull’esistenza di azioni di controllo in grado di mantenere le traiettorie del sistema all’interno di un insieme sicuro. Sebbene molto potente dal punto di vista teorico, l’applicazione diretta di questo approccio risulta tuttavia troppo onerosa dal punto di vista computazionale per l’uso in tempo reale. Per rendere il metodo più efficiente, sono stati sviluppati due approcci che approssimano il viability kernel mantenendone le proprietà di sicurezza. Il primo propone un decoupling del modello dinamico a bicicletta per renderlo affine, mentre il secondo sfrutta la differential flatness del sistema per ridurre drasticamente la complessità del problema di controllo. Poiché molti sistemi robotici sono flat, questa soluzione apre nuove prospettive per la riduzione della dimensionalità e l’applicazione efficiente dei vincoli di sicurezza e viabilità. Il framework proposto dimostra come sia possibile coniugare le garanzie formali delle Control Barrier Functions con la generalità della teoria della viabilità, mantenendo una complessità computazionale gestibile. Questo approccio ibrido assicura sicurezza e fattibilità nelle manovre di guida autonoma, anche in contesti urbani dinamici e incerti. Gli algoritmi di controllo sviluppati sono stati ampiamente validati sia in simulazione sia attraverso test sperimentali reali, confermando l’efficacia e la robustezza dei metodi proposti. L’architettura risultante rappresenta una base solida per futuri sviluppi verso scenari cooperativi multi-veicolo e ulteriori validazioni in contesti reali.

Progettazione di un controllo sicuro e fattibile per veicoli autonomi mediante Control Barrier Function e Viability Theory / Filippo Bernabei , 2026 May 22. 38. ciclo, Anno Accademico 2024/2025.

Progettazione di un controllo sicuro e fattibile per veicoli autonomi mediante Control Barrier Function e Viability Theory

BERNABEI, FILIPPO
2026

Abstract

In recent years, autonomous driving has emerged as one of the most transformative and multidisciplinary challenges in robotics and control engineering. Enabling a vehicle to navigate complex, dynamic, and uncertain environments without human intervention demands seamless integration of perception, planning, decision making, and control. Among these components, the design of a robust and safe control layer is fundamental: it must ensure collision avoidance, adherence to vehicle constraints, and responsiveness to unpredictable agents such as pedestrians or other vehicles. In urban settings especially, the density of obstacles, frequent intersections, occlusions, and highly variable traffic dynamics raise significant difficulties for classical control architectures. While many approaches focus on perception, mapping, or high-level planning, this work concentrates on the control problem. To achieve this goal, different types of Control Barrier Functions (CBFs) have been adopted to encode safety constraints and represent the interaction between the vehicle and surrounding obstacles. The use of pairwise CBFs allows one to handle multiple dynamic agents and obstacles, effectively mapping their relative behavior and enforcing collision avoidance in simulation. However, when transitioning from a kinematic to a dynamic vehicle model, necessary to represent higher speed and more realistic scenarios, the standard CBF formulation becomes insufficient. In such cases, Higher Order Control Barrier Functions (HOCBFs) are employed to incorporate higher order dynamics. Nonetheless, HOCBF based formulations often face limitations in feasibility when control inputs are bounded like in the real world. To overcome these issues, this research explores the use of viability theory, which inherently guarantees the existence of safe control actions that keep the system trajectories within a viable set. Although conceptually powerful, direct implementation of viability based approaches is computationally demanding and unsuitable for real time control. Therefore two works have been developed to approximate the viability kernel and exploit its safety guarantees in a more efficient manner. The first decoupled the dynamic bicycle model to make it affine and the second leverages the system’s differential flatness to significantly reduce the complexity of the control problem. Since many robotic systems are differentially flat, this solution can be a new approach to transform the original nonlinear system into a lower-dimensional representation, where safety and viability constraints can be enforced more efficiently. The proposed framework demonstrates that it is possible to combine the formal guarantees of control barrier functions with the generality of viability theory while maintaining computational tractability. This hybrid approach ensures both safety and feasibility in autonomous driving tasks, even under dynamic and uncertain urban conditions. The developed control algorithms have been extensively validated both in a simulation environment and through real-world experimental tests, confirming the effectiveness and robustness of the proposed methods. The resulting control architecture provides a viable foundation for future extensions toward cooperative multi vehicle scenarios and real world experimental validation.
Safe and Feasible Control of Autonomous Vehicles: A Control Barrier Function and Viability Approach
22-mag-2026
SECCHI, Cristian
File in questo prodotto:
File Dimensione Formato  
Bernabei.pdf

Open access

Descrizione: Bernabei.Filippo
Tipologia: Tesi di dottorato
Dimensione 2.16 MB
Formato Adobe PDF
2.16 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

Licenza Creative Commons
I metadati presenti in IRIS UNIMORE sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono rilasciati con licenza Attribuzione 4.0 Internazionale (CC BY 4.0), salvo diversa indicazione.
In caso di violazione di copyright, contattare Supporto Iris

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11380/1409088
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact