Avis de Soutenance
Monsieur Riccardo BONALLI
Soutiendra publiquement ses travaux d’habilitation à diriger les recherches
« Safe–against–uncertainty Control of Autonomous Systems »
Soutenance prévue le mardi 09 juin 2026 à 14h00
Lieu : CentraleSupélec Bâtiment Eiffel 3 rue Joliot-Curie F-91192 Gif-sur-Yvette Cedex
Salle : Amphi I
Lien Teams : https://teams.microsoft.com/meet/377174908117630?p=c01glYCVGBUFfD5fIk
Jury:
Jérôme MALICK, Directeur de Recherches CNRS, Laboratoire Jean Kunztmann (rapporteur).
Christophe PRIEUR, Directeur de Recherches CNRS, Gipsa–Lab (rapporteur).
Hasnaa ZIDANI, Professeur, INSA Rouen Normandie (rapporteure).
Francis BACH, Directeur de Recherches INRIA, Centre de Recherche INRIA de Paris (examinateur).
Daniel KUHN, Professeur, EPFL (examinateur).
Abdel LISSER, Professeur, Université Paris–Saclay, Laboratoire des Signaux et Systèmes (examinateur)
Abstract:
Autonomous Systems (AS) operating in safety–critical circumstances, such as driving and space robots, step up to tackle ever–more complex tasks in unpredictably dynamic situations, e.g., safely driving in trafficked roads. They must ensure reliability throughout a large range of uncertainties, including hazardous disturbances due to uncertain dynamic environments. This makes provably safe–against–uncertainty control of AS a relevant key challenge. Specifically, AS need more efficient and reliable algorithms to compute controls that not only enhance performance but also uphold safety standards across the aforementioned large range of complex, often dangerously undermodelled uncertainties. This manuscript shows how risk–averse stochastic optimal control, geometric and statistical analysis, and machine learning can be brought together to conceive such control algorithms, offering theoretical guarantees of performance. Specifically, this manuscript’s main contributions are on 1. efficiently computing risk–averse controls, 2. efficiently approximating safety constraints, and 3 safely controlling stochastic dynamics from data. Building upon these results, the manuscript ends with a chapter on relevant future research directions to unlock real–world safe–against–uncertainty control of AS.
