Thèse IA Incarnée dans un Micro Drone d'Assistance H/F Doctorat.Gouv.Fr

Établissement : Université de Toulouse École doctorale : EDMITT - Ecole Doctorale Mathématiques, Informatique et Télécommunications de Toulouse Laboratoire de recherche : IRIT : Institut de Recherche en Informatique de Toulouse Direction de la thèse : Christophe JOUFFRAIS ORCID 0000000207681019 Début de la thèse : 2027-10-01 Date limite de candidature : 2026-11-16T23:59:59 Les mini-robots (par exemple, les drones) sont aujourd'hui considérés comme des dispositifs d'assistance potentiels pour les personnes avec déficience visuelle (PDV). Des travaux récents se concentrent sur l'utilisation de mini ou micro drones pour aider les PDV dans la navigation piétonne, en guidant sur le chemin et en signalant les obstacles. Des techniques d'interaction non visuelles ont été spécifiquement conçues pour ces tâches. Il y a, cependant, un énorme potentiel inexploité pour ce qu'un mini drone peut faire pour aider une PDV. Par exemple, une PDV pourrait déployer le drone pour chercher un objet personnel dans la maison ou vérifier s'il y a un siège vide dans un parc. Mais il pourrait aussi aider à initier des relations sociales avec les tierces personnes (commerçants ou passants par ex) ou même fournir une aide pour revenir aux souvenirs visuels qui ont été enregistrés dans le passé.

Nous envisageons un mini ou micro drone léger et personnel, équipé de compétences en vision par ordinateur, mémoire et techniques d'interaction basées sur l'IA, qui servirait de dispositif d'assistance hautement personnalisable pour une PDV. Un tel drone d'assistance nécessite des techniques d'interaction humain-drone et des interfaces de programmation nettement plus avancées - accessibles et multimodales - que ce que propose l'état de l'art.Cela soulève de nouvelles questions fondamentales sur la façon dont une PDV personnaliserait et utiliserait un tel drone. Vision impairment affects at least 2.2 billion people worldwide1. In recent years, the HCI community has seen a wave of assistive robots designed and developed to support the Blind and Low Vision (BLV) community in their daily lives [3, 11 , 17 ]. For example, robotic canes and robotic guide dogs have been explored to replace their non- technical counterparts [7, 9, 13]. These systems demonstrate that assistive robots have great potential to improve the quality of life for BLV people and, importantly, the form factor of assistive robots significantly affects both the range of tasks they can perform and their overall usability. Drones have emerged as a versatile form of assistive robots, allowing them to support BLV people in a variety of tasks, such as navigation [1, 10 ] and locating objects [ 6 ]. Furthermore, a recent survey found that BLV people would like drones to provide assistance that expands beyond mundane day-to-day activities to leisure activities [4 ]. Unlike prior works where the BLV people must have a physical hold of the robot to use it, there is greater space leniency between users and the drone - for instance, a drone could be flying meters ahead of the user, which opens up new use cases and interactions. While common methods (e.g., joystick) may work for simple tasks (e.g., take-off, landing), they often fall short in supporting complex real-world scenarios. Viewed this way, it is important that BLV people can communicate naturally and flexibly with their drones to fully harness the potential of such novel assistive technology. Past research has highlighted that BLV people have greatly benefited from voice-based interactions [ 2, 16 ]. Furthermore, recent research has successfully shown that large language models (LLMs), such as ChatGPT, can enable users to communicate with robots more intuitively via voice [ 8 ]. For example, Padmanabha et al. have built an LLM-powered feeding robot for the elderly, where the elderly only need to provide natural language commands, and the LLM would parse human intent and generate code for the corresponding robot applications [12 ]. Nonetheless, limited work has explored how to leverage AI and LLMs to build assistive robots for BLV individuals - an area with unique design considerations due to the absence of visual feedback channels to interact with the robot.
Importantly, rather than prioritizing purely technological advancements, the focus is on tailoring assistive technologies to individual needs, based on user-centered design and participatory methods that will involve different stakeholders including BLV people as well as low vision professional from the Institut des jeunes Aveugles de Toulouse.
User studies will validate the usability, accessibility, and reliability of the system, with a focus on BLV users satisfaction and experience. A comprehensive evaluation will assess the impact of the system on users' autonomy, confidence, and quality of life.