Out-of-Distribution Detection with Vision Foundation Models and Post-hoc Methods
The thesis focuses on improving the reliability of deep learning models, particularly in detecting out-of-distribution (OoD) samples, which are data points that differ from the training data and can lead to incorrect predictions. This is especially important in critical fields like healthcare and autonomous vehicles, where errors can have serious consequences. The research leverages vision foundation models (VFMs) like CLIP and DINO, which have revolutionized computer vision by enabling learning from limited data. The proposed work aims to develop methods that maintain the robustness of these models during fine-tuning, ensuring they can still effectively detect OoD samples. Additionally, the thesis will explore solutions for handling changing data distributions over time, a common challenge in real-world applications. The expected results include new techniques for OoD detection and adaptive methods for dynamic environments, ultimately enhancing the safety and reliability of AI systems in practical scenarios.
Distributed multimodal learning for cooperative acoustic source localization and classification
In many complex environments, such as industrial sites, disaster-stricken buildings, or public spaces, it is necessary to automatically detect and localize sound events (falls, alarms, voices, mechanical failures). Mobile platforms equipped with cameras and microphones represent a promising solution, but a single platform remains limited: its microphone array provides an approximate direction towards the source but not a precise position in space, and its camera may be obstructed. This thesis proposes to study how a network of mobile platform, each carrying a calibrated audio-visual unit, can collaborate to localize and classify such events in 3D. Each platform analyses its own audio-visual observations and shares an estimate of the source direction with its neighbours; the network then combines these estimates to reconstruct the position of the event and identify it. The expected outcomes are a cooperative localization system that is robust to occlusions and partial platform failures.
Control & optimization of fuel cell temperature
Proton exchange membrane fuel cells (PEMFC) represent a key technology for the development of clean and sustainable energy systems, particularly for heavy-duty transport applications where their energy density is very attractive. However, in order to represent a viable industrial alternative, a number of obstacles still need to be overcome, including operating costs and, above all, the durability of the systems under real-world conditions. Among the levers for action, optimizing operating conditions is a promising avenue for limiting the degradation phenomena occurring within the cell. The operating temperature is a particularly key parameter because it affects all aspects of the system, from the kinetics of degradation mechanisms to the thermal capacity that the system can dissipate, including the water balance within the fuel cell. Despite the influence of this parameter on durability, it is generally only optimized at the system level to achieve the best performance, the shortest possible response time and to limit the size of the thermal management system.
The aim of this thesis is to work on optimizing the temperature management of a fuel cell within a system, taking into account not only performance but also sustainability criteria. To do this, the impact of operating temperature on degradation mechanisms will be analyzed using various simulation tools already available at LITEN and the teams' fifteen years of experience in studying PEMFC fuel cell degradation. Various thermal architectures will be proposed and evaluated in conjunction with the work on temperature control optimization. The latter will be implemented on a real fuel cell system in order to demonstrate the relevance of the proposed solution using concrete experimental data.
Multi-criteria Navigation of a Mobile Agent applied to nuclear investigation robotics
Mobile robots are increasingly deployed in hazardous or inaccessible environments to perform inspection, intervention, and data collection tasks. However, navigating such environments is far more complex than simple obstacle avoidance: robots must also deal with communication blackouts, contamination risks, limited onboard energy, and incomplete or evolving maps. A previous PhD project (2023–2026) introduced a multi-criteria navigation framework based on layered environmental mapping and weighted decision aggregation, demonstrating its feasibility in simulated, static scenarios.
The proposed thesis aims to extend this approach to dynamic and partially unknown environments, enabling real-time adaptive decision-making. The work will rely on tools from mobile robotics, data fusion, and autonomous planning, supported by experimental facilities that allow realistic validation. The objective is to bring navigation strategies closer to real operational conditions encountered in nuclear dismantling sites and other industrial environments where human intervention is risky. The doctoral candidate will benefit from an active research environment, multidisciplinary collaborations, and strong career opportunities in autonomous robotics and safety-critical intervention systems.