Mobilit.AI returns for its 6th edition on October 16–17, 2025, in Toulouse!
A unique international event
Co-organized by IVADO, IRT Saint Exupéry, IID, CRIAQ, and ANITI, Mobilit.AI is the leading event dedicated to Artificial Intelligence applied to critical transport and mobility systems. Each year, the event brings together an international community of experts : researchers, industry leaders, engineers, and specialists in critical embedded systems, to address the major AI challenges in sectors such as aerospace, automotive, rail, space, and drones.
2025 Program
Welcome
Opening Remarks
Speaker – Katarzyna Kapusta – Cybersecurity expert working on AI & Cyber at Thales.
Katarzyna is a cybersecurity researcher specializing in AI at Thales’ cortAIx Labs. She leads the cross-functional team “AI Friendly Hackers,” which aims to develop countermeasures against AI-specific attacks, particularly those attempting to misuse AI or steal training data. She acts as a technical expert on AI cybersecurity in collaborative research projects and European standardization groups, such as ETSI TC Securing Artificial Intelligence. She earned a PhD in computer science from Télécom Paris in 2018, where she continues to teach. She is the author of several publications in journals and conferences, as well as patents.
Poster Session
Roundtable on Cyber & AI: From Research to Industry
12h30-14h Lunch break et Session Posters
During this session, we will address the topic of explainable artificial intelligence (AI) through three presentations from academia and industry, followed by a roundtable discussion. In high-stakes domains such as mobility, energy production, and distribution, it is essential to audit AI systems to ensure their reliability and certify their use. The presentations will provide a broad overview, from interpretation methods to practical applications tackling major industrial challenges. The roundtable will then offer a privileged space for discussion on the perspectives and challenges of this rapidly evolving field.
Speaker – Thomas Fel, Researcher at the Kempner Institute, Harvard University
“From Methods to Phenomenology: Rethinking Interpretability in Vision Models”
Thomas Fel is a researcher at Harvard University’s Kempner Institute, specializing in explainable AI. He takes an interdisciplinary approach at the intersection of computer science, mathematics, and neuroscience. He completed a CIFRE PhD between SNCF, Brown University, and the DEEL project (ANITI) in Toulouse. He has also completed several research internships at Google and GoPro. His research focuses on the explainability of deep neural networks in computer vision. By combining computational methods with neuroscience knowledge, he seeks to reveal and understand the internal mechanisms of these models. This interdisciplinary approach aims not only to deepen our understanding of AI but also to use it as a tool to explore human intelligence
Speaker – Clément Bénard, AI Researcher at Thales cortAIx-Labs
“Explainable AI for Performance Monitoring in Air Traffic Management”
Clément has ten years of experience in machine learning research across aerospace, aviation, and technology sectors, with a strong interest in healthcare applications. He holds a PhD in machine learning and mathematical statistics from the University of Sorbonne, focusing on explainable AI (XAI). His research areas include explainable AI, causal inference, random forests, deep learning, and uncertainty quantification. His work covers algorithm design, mathematical analysis, and industrial application
Speaker – Nicolas Bousquet, Senior Researcher at EDF R&D and Director of the EDF–TotalEnergies–Thales Industrial AI Lab
“Overview of AI Explainability Work for Critical Systems at EDF”
Nicolas Bousquet is an applied mathematics researcher specializing in statistical modeling, machine learning, and decision support in uncertain environments, leveraging the fusion of heterogeneous information sources. He is currently a senior researcher at EDF R&D and leads the joint industrial AI laboratory EDF–TotalEnergies–Thales, focusing on foundational AI for critical industrial systems. He is also an associate professor at the University of Sorbonne.
16:30–17:15 – Roundtable on Explainability
17:15–21:00 – Networking Evening
October 17
Welcome
This panel will explore the role of generative AI in mobility and critical systems. While these models open the door to new forms of interaction, design, and automation, they also raise unprecedented questions about reliability, robustness, and safety. Our speakers will discuss the opportunities offered by these technologies, as well as the precautions necessary to ensure responsible use in sensitive contexts.
Speaker – Julie Hunter, Researcher in Linguistics and NLP at Linagora
“Building Language Models from Scratch: Challenges of Interpretability”
Julie holds a PhD in philosophy and linguistics from the University of Texas at Austin, specializing in the semantic and pragmatic properties of natural language. She conducted research on conversation and multimodality before joining Linagora’s R&D team in 2019, contributing her linguistic expertise to corpus-based projects. She is currently research lead for NLP R&D projects and technical coordinator of the OpenLLM France research project funded by BPI, aimed at developing fully transparent, open-source language models with a focus on French. Her current research focuses on multilingualism in language models.
Poster Session
Speaker – Céline Hudelot, Professor of Computer Science at CentraleSupélec
“Generative AI for Critical Societal Domains”
Céline Hudelot is a professor of computer science at CentraleSupélec and director of the MICS laboratory since October 2022. She earned her PhD in computer science from INRIA and the University of Nice Sophia Antipolis in 2005 and her Habilitation to Supervise Research from Paris-Sud University in 2014. She has supervised over 20 AI theses (completed or ongoing). She leads the AI research axis of the lab and co-directs the AI track of the CentraleSupélec engineering program. Her research focuses on the intersection of machine learning, knowledge representation, and reasoning, with current work on explainability, learning from unstructured data (images, text, documents) under various learning paradigms (few-shot, semi-supervised, active, incremental), hybrid intelligence, and foundation models.
Speaker – Luce Lefeuvre, Head of AI Innovation Projects at SNCF
12:30–14:00 – Lunch break and Best Poster Presentation
Agentic AI Panel
Presentation – Swarnadeep Bhar, PhD Student at ANITI
“The Audio Mobility 2030 Conversational System: A Blueprint for More Capable Cobots”
This presentation explains how research at ANITI contributes to creating systems capable of interacting with users to perform a wide range of tasks. The system demonstrates how agentic approaches with advanced conversational capabilities allow cobots to accomplish tasks more efficiently and reliably. We will present work from both the AM2030 project and the associated CoCoreli project, which focuses on assembly tasks in low-data environments.
Swarnadeep Bhar is a final-year PhD student at the Artificial and Natural Intelligence Toulouse Institute (ANITI), under the supervision of Nicholas Asher. He develops conversational and agentic AI systems enabling collaborative robots (cobots) to interact more reliably with humans in complex, low-data environments. He is behind platforms such as CoCoBots and CORELI, designed to enhance trust in human–AI interactions, particularly in assembly and mobility domains.
Robotics and Simulations Panel
Speaker – Nicolas Mansard, Researcher at LAAS-CNRS
Nicolas is a permanent researcher at LAAS-CNRS (to lead the Gepetto team in 2026) and a member of Toulouse’s AI excellence cluster (ANITI), where he holds a chair and sits on the scientific committee. A recognized robotics expert, he has made major contributions in optimal control and reinforcement learning.
He collaborates regularly with researchers at the Czech Technical University (CTU), New York University (NYU), and INRIA Paris. He coordinates the European Memmo project (10 partners, €4M, recognized as an ERF 2023 success story and awarded a French Ministry of Research “Étoile de l’Europe”), and the AGIMUS project (8 partners, €5M, ongoing until 2026). His work has been awarded the CNRS Bronze Medal in 2013. The ENTRACT project he coordinated was recognized by the ANR in 2015 as the best French computer science project. He also maintains close ties with industry: coordinator of the joint DYNAMOGRADE lab with PAL Robotics, principal investigator of the ROB4FAM lab with Airbus, and collaborator with companies such as Wandercraft. In 2025, he was awarded the Nokia-France Chair of Excellence.
More information about previous editions: MOBILIT AI – MobiliT.AI