Abstract
This thesis presents an in-depth study of predictive modeling of tire grip during high-speed cornering, with particular reference to autonomous Formula 1 vehicles. Developed in collaboration with the Technology Innovation Institute in Abu Dhabi, the research is part of a pioneering project aimed at surpassing human capabilities in motorsport through intelligent systems capable of making real-time decisions based on environmental data captured via sensors. The main problem addressed, known as the Grip Problem, is to enable an artificial intelligence model to determine, in real time, whether a vehicle can accelerate or should brake during a turn without exceeding grip limits. To this end, the work integrates advanced telemetry data, vehicle dynamics models, and state-of-the-art machine learning techniques to estimate grip availability and optimize cornering performance. The experimental methodology consisted of two main phases. The first involved a simulated environment, using Assetto Corsa, in which data were collected, datasets constructed, physical variables analyzed, and numerous machine learning models developed. These models were evaluated both in terms of grip condition classification and the prediction of optimal driving strategies and dynamic variables. The results highlighted the crucial importance of temporal modeling. The second phase focused on the analysis of real datasets from autonomous driving sessions. Through signal fusion, the processing of dynamic indices such as drift angles and friction coefficients, and the application of regression models, the transferability of simulation-developed models to real-world contexts was demonstrated. A real-time graphical user interface for visualizing grip status was also developed, evaluated through usability testing, and designed for practical use in both motorsport and automotive environments. The work demonstrates the feasibility of integrating predictive grip estimation into autonomous driving systems.