par Balachander, Mrudula 
Président du jury Heim, Rachelle
Promoteur Raskin, Jean-François
Co-Promoteur Filiot, Emmanuel
Publication Non publié, 2026-02-09
Président du jury Heim, Rachelle
Promoteur Raskin, Jean-François
Co-Promoteur Filiot, Emmanuel
Publication Non publié, 2026-02-09
Thèse de doctorat
| Résumé : | Reactive systems continuously interact with their environment in real-time, requiring high standards of correctness, reliability, and efficiency. Designing such systems is challenging due to their inherent complexity and the high risk of unexpected behaviour or severe failures arising from unforeseen interactions and timing constraints. Therefore, rigorous formal methods and verification are needed to improve the reliability of reactive systems and ensure their correctness. The synthesis problem goes a step further and aims to automate the construction of reactive systems that are guaranteed to meet a given specification, typically expressed as logical formulas. This dissertation advances the field of reactive system synthesis along two main themes. First, we address the limitations of automatic LTL synthesis by introducing “LTL synthesis with hints” — a novel variant of the LTL synthesis problem that incorporates user guidance in the form of traces to steer synthesis algorithms toward more practical and efficient solutions. By generalising the user-provided example traces, our approach combines passive learning techniques with existing LTL synthesis algorithms to bridge the gap between abstract specifications and real-world implementations, as demonstrated through an implementation and a series of case studies.Second, we investigate efficient and effective representations for controllers. Traditional LTL synthesis algorithms output large finite-state Mealy machines. To address the need for concise yet efficient representation, we explore the class of register automata, which are finite state machines equipped with registers, recognising the class of regular languages over infinite alphabets. Our work introduces a novel variant of deterministic register automata, namely Register Automata with Permutations (pDRA), which aims to strike a balance between succinctness and computational efficiency. Additionally, we provide a Myhill– Nerode-like characterisation that enables the construction of canonical minimal representations. Furthermore, we present a polynomial-time passive learning algorithm for a subclass of these automata, demonstrating the efficient identifi-cation of this class in the limit.Together, these contributions enhance the practicality and scalability of synthesising reactive systems through LTL specifications, paving the way for more robust and manageable implementations in real-world applications. |
