Scientific Publications of the University of Toulouse II Le Mirail
Not a member yet
92205 research outputs found
Sort by
Comparative analysis of dynamic balance descriptors in humanoids and humans during perturbed bipedal locomotion and fall
No full textInternational audienceThis study identifies a robust parameter for quantifying instability in general biped systems by comparing three mechanical stability descriptors: the distance between the center of mass to the minimal moment axis (d CoM-MMA ), the margin of stability (MoS), and whole-body angular momentum (WBAM) in both humans and humanoid biped robots. We analyzed these metrics during normal and perturbed walking, including robot falls, a dynamic whose observation is limited in human trials due to safety concerns. Our comparative analyses demonstrate that d CoM-MMA is more predictive of different levels of instability and shows a clearer distinction between fall and non-fall states, compared to MoS and WBAM. These findings were consistent for both humans and biped robots, regardless of gait variability or the type and intensity of the perturbation methods. These qualities highlight its potential use in unified stability analysis in both fields, offering insights that can inform the design of exoskeletons, fall monitoring systems, and other gait-assistive devices for aging populations
Analyse multimodale de scène : vers une intégration des données contextuelles pour une compréhension plus approfondie
No full textThis thesis addresses the integration of contextual data in scene analysis. While traditional scene analysis methods generally rely on visual (image sequences), textual (subtitles), or audio (speech) data, they seldom consider the richness of information conveyed by contextual cues-such as the relationships between characters-which are nonetheless crucial to a deep understanding of the dynamics between entities present in a scene.As part of our work, we first propose a generic data model designed to structure and represent contextual information in a coherent, modular, and extensible manner, regardless of the application domain (e.g., video surveillance, audiovisual content, etc.). This model is based on principles of semantic and conceptual modeling, making the contextual data easier to interpret and reuse.We then introduce an automatic annotation framework that leverages large language models (LLMs), combined with computer vision and speech recognition techniques, to extract a broad range of contextual data from scenes. This annotation process aims not only to enrich training datasets but also to improve the overall quality of scene analysis, while also accounting for biases introduced by automated tools. Particular attention is paid to detecting and mitigating the biases inherent in the systems we employ.Finally, we propose an approach based on graph neural networks to effectively integrate the various modalities and contextual data into multiple scene analysis tasks, and to assess the relative importance of each type of information. Experimental results, conducted on tasks such as social interaction classification and interpersonal relationship recognition, demonstrate that the inclusion of contextual data significantly improves performance and enhances the interpretability of results, thereby validating our initial hypothesis.This research opens several promising perspectives, including advanced retrieval of audiovisual content based on complex semantic queries, automated audio description, and the development of new evaluation metrics for scene annotation. As such, this thesis contributes to the advancement of automated systems that better grasp the complexity of human interactions, that are more equitable, and that rely on more adaptive data representation frameworks.Cette thèse explore la problématique de l'intégration des données contextuelles dans l'analyse de scène. Bien que les méthodes d'analyse de scènes s'appuient généralement sur des données visuelles (séquences d'images), textuelles (sous-titres) ou audio (parole), elles considèrent rarement la richesse d'information que peuvent apporter certaines informations contextuelles telles que les relations entre les personnages, pourtant essentielles à une compréhension fine des dynamiques entre les entités présentes dans une scène.Dans le cadre de nos travaux, nous proposons dans un premier temps un modèle de données générique permettant de structurer et de représenter ces informations contextuelles de manière cohérente, modulaire et extensible, quel que soit le domaine d'application (vidéosurveillance, audiovisuel, etc.). Ce modèle repose sur des principes de modélisation sémantique et conceptuelle, facilitant l'interprétation et la réutilisation des données.Ensuite, nous introduisons un cadre d'annotation automatique, tirant parti des larges modèles de langage (LLMs), combinés à des techniques de vision par ordinateur et de reconnaissance vocale, pour extraire un large éventail de données contextuelles à partir de scènes. Cette annotation vise non seulement à enrichir les bases de données d'apprentissage, mais aussi à améliorer la qualité de l'analyse, tout en tenant compte des biais introduits par les outils automatiques. À ce titre, une attention particulière est portée à la détection et à la réduction des biais véhiculés par les systèmes que nous utilisons.Enfin, nous proposons une approche reposant sur des réseaux de neurones en graphes pour intégrer de manière efficace les différentes modalités et données contextuelles dans divers problèmes d'analyse de scène et jauger l'importance de chaque information. Les résultats expérimentaux, menés sur des tâches telles que la classification d'interactions sociales ou la reconnaissance de relations interpersonnelles, montrent que l'ajout du contexte améliore significativement les performances et permet une meilleure interprétabilité des résultats, validant ainsi notre hypothèse de départ.Cette recherche ouvre plusieurs perspectives prometteuses, telles que la recherche avancée de contenu audiovisuel à partir de requêtes sémantiques complexes, l'audiodescription automatisée, ou encore le développement de nouvelles métriques d'évaluation pour l'annotation de scènes. Cette thèse constitue ainsi une contribution vers des systèmes automatisés capables de mieux saisir la complexité des interactions humaines, plus équitables, mais aussi des systèmes de représentation des données plus adaptatifs
: Social status of children in the Neolithic period: the case of the Beaufort child
No full textInternational audienc
Soft, transparent and bioresorbable microelectrode array for transient electrophysiological recordings
No full textTransparent microelectrode arrays that enable multimodal investigation of spatiotemporal electrophysiological activity are critical tools for advancing the understanding of excitable tissues such as the brain, heart, and peripheral nerves. Traditional implantable devices are engineered for chronic use but require surgical removal when they fail or are no longer needed.In contrast, bioresorbable systems that naturally dissolve after serving temporary functions offer a compelling alternative, eliminating the risks and costs of extraction procedures. Here, we present the design, fabrication, and validation of a soft, fully bioresorbable, and optically transparent MEA platform for transient, bidirectional interfacing with living tissues. The device provides high-resolution electrical mapping of dynamic activity. We report precise characterization of electrochemical performance, mechanical properties, bioresorption kinetics, and biocompatibility. While validated in models of cardiac function, this platform establishes a versatile foundation for bioresorbable electrophysiological technologies with applications ranging from postsurgical monitoring of transient conditions to the study and treatment of neurological and neurodegenerative disorders.</p
Cyclists route choice modeling from trip duration data in urban areas
No full textThe lack of GPS data limits the ability to reconstruct the actual routes taken by cyclists in urban areas. This article introduces an inference method based solely on trip durations and origin-destination pairs from bike-sharing system (BSS) users. Travel time distributions are modeled using log-normal mixture models, allowing us to identify the presence of distinct behaviors. The approach is applied to 3.8 million trips recorded in 2022 in the Toulouse metropolitan area, with observed durations compared against travel times estimated by OpenStreetMap (OSM). Results show that, for many station pairs, trip durations align closely with the fastest route suggested by OSM, reflecting a dominant and routine practice. In other cases, mixture models reveal more heterogeneous behaviors, including longer trips, detours, or intermediate stops. This approach highlights both the stability and diversity of cycling practices, providing a robust tool for usage analysis in data-limited contexts, and offering new insights into urban mobility dynamics without relying on spatially explicit data
Study of the stationary Stokes system with mixed boundary conditions in non-convex curvilinear polygonal domains
No full textInternational audienceIn this paper, we consider the stationary Stokes system with mixed boundary conditions, of Dirichlet and Neumann types, in a bounded non-convex curvilinear polygonal domain of . We prove, in particular, a precise regularity result in heterogeneous Sobolev spaces taking into account the fact that the expected regularity is of different nature near the corners of the domain and near the Dirichlet-Neumann transition points. Then, we prove the analyticity of the semigroup generated by the Stokes operator in an appropriate functional setting. We also give a characterization of the stationary Stokes system as an operator equation.Those results, that can be useful in various situations, are in particular motivated by the analysis of a fluid-structure interaction system investigated by the authors in a forthcoming paper, and for which they represent an essential step
Alimentation autonome sans fil et sans batterie par électrothermogénération
No full textL'électrothermogénération est bien adaptée pour récupérer de l'énergie électrique en présence d'un gradient thermique permanent mais il y a toujours quelques problèmes techniques comme présenté dans cet article basé sur la réalisation d'une alimentation électrique destiné à un capteur d'O2 en milieu industriel.International audienc
"De la passion inactive. Dominique de Fromentin et L'Education sentimentale de Flaubert dans le sillage d'Oberman (Senancour)"
No full textInternational audienceRedécouvert dans les années 1830, Oberman de Senancour a légué à Fromentin dans Dominique et à Flaubert dans L’Éducation sentimentale le modèle, présenté comme un défi littéraire, d’un épuisement du romanesque qui passe par la neutralisation de la passion commandée par un présent médiocre. Mais cette atonie doit être nuancée par l’intensité au contraire décuplée de la sensation, de la mémoire et de l’imagination, dans des romans où s’essaient des formes de bonheur toujours ambivalentes
Interacting Kalman Filters for Linear Systems with Coupling Based on Empirical Covariances
No full textInternational audienceThe problem of designing estimators for stochastic linear systems with distributed observations is considered. Each observation process is associated to a node in an undirected graph, which is used to compute a local estimate at the node. The injection gain used in the filter at each node is obtained from the empirical covariance of all the estimates available at that node. This way the underlying idea comes from the theory of ensemble filtering and we analyze the evolution of the coupled covariances over the entire graph. After providing a detailed derivation of the evolution of covariance matrix, we observe that the drift term in the differential equation for the coupled covariances has some inherent stability structure in case of regular graphs, which leads to the fluctuations around the steady state. We provide an illustration of our algorithm on an academic example while comparing it with centralized and ensemble Kalman filters
Mapping the Landscape: A Scoping Review of Research on Neurodevelopmental Disorders in Higher Education
No full textThis project aims to conduct a scoping review of existing research on neurodevelopmental disorders (NDDs) in higher education, including conditions such as autism, ADHD, and learning disorders. The primary objectives are to analyze research trends over time, identify and synthesize both transversal and condition-specific thematic areas within the literature, and highlight existing gaps. This work will inform future research directions