Datalake embarqué fiable et fusion multi-échelle pour la prise de décision collaborative humain-robot en environnements critiques // Reliable Embedded Data Lake and Multi-Scale Fusion for Collaborative Human–Robot Decision-Making in Critical Environments

Cette thèse vise à concevoir une architecture edge-native distribuée et fiable capable d'intégrer, structurer et exploiter en temps réel des données issues d'opérateurs humains et de flottes de robots hétérogènes évoluant dans des environnements critiques et dynamiques. Les travaux porteront sur le traitement adaptatif et la synthèse de flux multi-capteurs (LiDAR, caméras, IMU, données sémantiques), la fusion multi-échelle de données incomplètes et incertaines, ainsi que le développement d'un datalake embarqué résilient fonctionnant sous fortes contraintes de connectivité, d'énergie et de calcul. La thèse explorera également l'intégration d'IA embarquée, notamment les Small Language Models (SLMs) et les modèles Perception-Language-Action (PLA), afin d'améliorer l'interaction humain–robot, le raisonnement autonome et la prise de décision collaborative en conditions dégradées.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------

This PhD aims to design a reliable distributed edge-native architecture capable of integrating, structuring, and exploiting in real time data generated by human operators and heterogeneous robotic fleets operating in dynamic and critical environments. The research will focus on adaptive processing and summarization of multi-sensor streams (LiDAR, cameras, IMU, semantic data), uncertainty-aware multi-scale data fusion, and the development of a resilient embedded datalake operating under strong bandwidth, energy, and computational constraints. The thesis will also investigate embedded AI paradigms, including Small Language Models (SLMs) and Perception-Language-Action (PLA) models, to enhance human-robot interaction, autonomous reasoning, and collaborative decision-making under degraded conditions.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Début de la thèse : 01/10/2026

Contrat doctoral

Concours pour un contrat doctoral

Université de Limoges

653 Sciences et Ingénierie

- Master's degree or Engineering degree in Robotics, Computer Science, Artificial Intelligence, Data Science, Embedded Systems, or a closely related field - Strong background in at least several of the following domains is expected: - multi-sensor data fusion, robotic perception and autonomous systems, - data management and streaming architectures, - machine learning or online learning, - distributed systems and edge computing, - Good experience with C/C++, Python, ROS, and data streaming tools (Apache Kafka, Apache Spark, Flink, or River/scikit-multiflow for online learning) and familiarity with deep learning tools such as Cuda, Keras, TensorFlow or PyTorch - Good software engineering practices - Experience with data preprocessing, augmentation, and utilizing various AI frameworks - Experience with robotic middleware and software such as ROS/ROS2, real-time systems, or distributed robotic architectures would be considered a strong asset - Knowledge of modern AI approaches for robotics, including foundation models, vision-language-action models, Small Language Models will be highly appreciated - Excellent written English and communication skills - Scientific curiosity, autonomy, and the ability to work independently and collaboratively within a multidisciplinary research environment The project requires a strong interest in interdisciplinary research at the intersection of robotics, AI, distributed information systems, and field experimentation. The candidate should be able to conduct both theoretical and applied research, including algorithm development, system integration, experimentation, and scientific evaluation.
- Master's degree or Engineering degree in Robotics, Computer Science, Artificial Intelligence, Data Science, Embedded Systems, or a closely related field - Strong background in at least several of the following domains is expected: - multi-sensor data fusion, robotic perception and autonomous systems, - data management and streaming architectures, - machine learning or online learning, - distributed systems and edge computing, - Good experience with C/C++, Python, ROS, and data streaming tools (Apache Kafka, Apache Spark, Flink, or River/scikit-multiflow for online learning) and familiarity with deep learning tools such as Cuda, Keras, TensorFlow or PyTorch - Good software engineering practices - Experience with data preprocessing, augmentation, and utilizing various AI frameworks - Experience with robotic middleware and software such as ROS/ROS2, real-time systems, or distributed robotic architectures would be considered a strong asset - Knowledge of modern AI approaches for robotics, including foundation models, vision-language-action models, Small Language Models will be highly appreciated - Excellent written English and communication skills - Scientific curiosity, autonomy, and the ability to work independently and collaboratively within a multidisciplinary research environment The project requires a strong interest in interdisciplinary research at the intersection of robotics, AI, distributed information systems, and field experimentation. The candidate should be able to conduct both theoretical and applied research, including algorithm development, system integration, experimentation, and scientific evaluation.