Thesis of Nourdine Ait Tmazirte

Localization for Safety Critical Land Transportation Systems: Contributions to the Acceptability of Satellite-Based Solutionsfor Safety Critical Land Transportation Systems: Contributions to the Acceptability of Satellite-Based Solutions

Since the dawn of humanity, the ability to position and navigate has constituted a decisive evolutionary advantage. From Polynesian navigators using constellations to cross the Pacific to European mariners orienting themselves with sextants, navigation has always relied on observing celestial signals. GNSS systems represent the culmination of this millennial quest: generalizing the principle of star-based localization using artificial satellites, providing for the first time a global, precise and accessible spatio-temporal reference. Yet this revolution remains unfinished. Despite decades of improvements, GNSS struggle to cross the acceptability threshold of critical terrestrial applications for safety. This limitation reveals conceptual misunderstandings, gradually erected into incompatibilities, between development paradigms of localization functions based among others on satellite technology and safety requirements of critical terrestrial applications. Our work focuses on the railway and road/automotive sectors, where safety stakes are critical. The limitations reveal three structural inadequacies. The first is a geographical paradox: environments where GNSS are most disturbed correspond to locations where requirements are maximal. The second concerns behavioral unpredictability: GNSS constitute variable-performance sensors whose reliability is a function of uncontrolled external factors. The user faces a different sensor at each instant, with error variance that can vary by a factor sometimes exceeding 10. This instability complicates establishing deterministic specifications, the foundation of safety assessment methods. The third inadequacy concerns performance discontinuities: environmental transitions generate abrupt switches contradicting required graceful degradation principles. We introduce acceptability as a structuring framework: the capacity of a localization function to satisfy certification processes while remaining economically viable. This acceptability is evaluated across four dimensions: technical, normative, operational and economic. It depends on the Operational Design Domain (ODD): requirements vary by several orders of magnitude according to contexts, excluding any universal solution. Facing these barriers, we develop a tripartite systemic approach unified by information theory. Our assertiveness does not bear on the specific superiority of this or that filter or architecture, but on the systemic approach simultaneously addressing three complementary levers: parallel multi-sensor fusion architectures (2 out of 3 with voting) avoiding common failure modes; a multi-parametric weighting model (C/N0, elevation, DOP, etc.) constrained by design through sigmoid functions guaranteeing regularity and predictability; integration of a diagnostic layer based on informational divergences (mutual information, Kullback-Leibler, α-Rényi) for adaptive diagnosis. Our philosophy privileges predictability and explainability over algorithmic sophistication, recognizing that acceptability depends on integration into certification processes. The challenge is to design optimally dimensioned fault-tolerant architectures, validatable and viable, not to build through massive redundancy. Complete acceptability exceeds the scope of this thesis. Our work claims neither to have certified a system, nor to have demonstrated the complete acceptability of an architecture. We present methodological contributions aimed at lifting certain identified scientific barriers, providing reusable and validatable building blocks in a progressive approach. Acceptability remains the horizon guiding our methodological choices, not a claimed achieved result.

Jury

- M. Maan EL BADAOUI EL NAJJAR Professeur des universités Université de Lille Directeur de thèse, - M. Denis POMORSKI Professeur des universités Université de Lille Co-encadrant de thèse, - M. Ludovic MACAIRE Professeur des universités Université de Lille Examinateur, - M. Jean-Christophe POPIEUL Professeur des universités INSA Hauts-de-France Examinateur, - Mme Nadine LE FORT-PIAT Professeure émérite SUPMICROTECH-ENSMM Examinatrice, - Mme Julie BEUGIN Chargée de recherche Université Gustave Eiffel Examinatrice, - M. Philippe BONNIFAIT Professeur des universités Université de Technologie de Compiègne Alliance Sorbonne Université Rapporteur, - M. Roland CHAPUIS Professeur des universités Université Clermont Auvergne Rapporteur.

Thesis of the team ToSyMA defended on 23/11/2026