Stefano Bucaria
Physics-Informed Neural Network approaches for control-oriented vehicle dynamics modelling.
Rel. Aldo Sorniotti, Stefano De Pinto, Cecilia Formento, Matteo Depalo. Politecnico di Torino, Corso di laurea magistrale in Automotive Engineering (Ingegneria Dell'Autoveicolo), 2026
Abstract
Accurate estimation of vehicle lateral dynamics is a key requirement for control-oriented applications involving stability control, motion prediction, and vehicle adaptation under varying operating conditions. This thesis investigates the use of a Physics-Informed Neural Network framework that combines fundamental vehicle dynamics, based on a single-track model and the Pacejka Magic Formula, with a data-driven neural network to estimate Pacejka coefficients under real-time conditions. The objective is to obtain an interpretable and adaptive predictive model capable of supporting more reliable control-oriented vehicle dynamics applications. The proposed methodology is based on an initial offline training phase using a sufficiently large and representative dataset, designed to promote generalization across a wide range of driving scenarios, followed by an online retraining procedure applied to improve prediction accuracy and robustness..
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Informazioni aggiuntive
Corso di laurea
Classe di laurea
Aziende collaboratrici
URI
![]() |
Modifica (riservato agli operatori) |
