Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
Fall detection and prevention systems of homecare for the elderly: myth or reality?
No full textInternational audienceThere is an exponential increase in the range of digital products and devices promoting aging in place in particular devices aiming at preventing or detecting falls. However, their deployment is still limited and few studies have been carried out in population-based settings. Such a matter of fact is due to the technological challenges that remain to overcome but also to the barriers that are specific to the users themselves such as the generational digital divide and acceptability factors specific to the elderly population. To date, scarce studies take into account these factors. In order to capitalize technological progress, the further step should be to better take into account these factors and to deploy, in a broader and more ecological way, these technologies designed for home care seniors, in order to assess their effectiveness in real life
Physically motivated structuring and optimization of neural networks for multi-physics modelling of solid oxide fuel cells
No full textInternational audienceNeural network models for complex dynamical systems typically do not explicitly account for structural engineering insight and mutual interrelations of various subprocesses that are related to the multiphysics nature of such systems. For that reason, they are commonly interpreted as a kind of data-driven, black box modelling option that is in opposition to a physically inspired equation-based system representation for which suitable parameters are subsequently identified in a grey box sense. To bridge the gap between datadriven and equation-based modelling paradigms, this paper proposes a novel approach for a physics-inspired structuring of neural networks. The derivation of this kind of structuring, an optimal choice of network inputs and numbers of neurons in a hidden layer as well as the achievable modelling accuracy are demonstrated for the thermal and electrochemical behaviour of hightemperature fuel cells. Finally, different network structures are compared against experimental data
Analyse de la motivation intrinsèque au cours d’une activité de résolution de problèmes.
No full textOur aim is to help researchers in digital sciences (computer science and applied mathematics),cognitive neurosciences and educational sciences to join their forces to better understand, in a specific framework, how learners learn, in the case of a problem-solving task related to computational thinking learning.More precisely, we explore the research carried out on the intrinsic motivation of the learner, during a learning task, in order to understand to what extent it is involved in such process, mainly reporting the flowers research team work, summarized here, in an accessible way for colleagues from these different disciplines, including going to the technical details of modeling.This research raises an issue, requiring a more precise definition of the concepts essential to understanding the mechanisms involved in learning, especially because of their multidisciplinary, and often polysemous, use.The aim of a review of the literature is thus to explore the research carried out on the intrinsic motivation of the learner during learning in order to understand to what extent it is involved in it. This will help to improve the learner model and getting closer to understanding the mechanisms of human learning.Carried out as part of a research internship in bioinformatics, it is neither exhaustive nor generalist, but specific to the fields covered. However, it represents a unique production in terms of reviewing the French-speaking literature on these subjects.On cherche ici à aider des chercheuses et chercheurs en sciences du numérique (informatique et mathématiques appliquées), desneurosciences cognitives et des sciences de l’éducation à s’allier pour tenter de mieux comprendre, dans un cadre précis, comment les personnes apprenantes apprennent, dans le cas d’une tâche de résolution de problème en lien avec l’apprentissage de la pensée informatique.Plus précisément, on explorer ici les recherches menées sur la motivation intrinsèque de l’apprenant au cours de l’apprentissage afin de comprendre dans quelle mesure elle y est impliquée, au sein de l'équipe flowers, que nous résumons ici de manière accessible pour les collègues des différentes disciplines, en allant jusqu'au détails techniques de la modélisation. Ces recherches ont soulevé une problématique nécessitant de définir plus précisément les concepts indispensables à la compréhension des mécanismes impliqués dans l’apprentissage, tout particulièrement du fait de leur usage pluridisciplinaire, et souvent polysémique.Une revue de la littérature a pour alors but d’explorer les recherches menées sur la motivation intrinsèque de l’apprenant au cours de l’apprentissage afin de comprendre dans quelle mesure elle y est impliquée. Cela permettra de contribuer à l’amélioration du modèle de l’apprenant et ainsi de s’approcher davantage de la compréhension des mécanismes de l’apprentissage humain.Réalisée dans le cadre d’un stage de recherche en bioinformatique, elle n'est ni exhaustive, ni généraliste, mais spécifique aux domaines traités. Elle représente cependant une production unique en matière de revue de la littérature francophone sur ces sujets
Études numériques et expérimentales des impacts hydrodynamiques primaires et secondaires lors du tossage de sections de carènes
No full textThis work involves modeling and studying the phenomena of flow separation and secondary hydrodynamic impact that can occur during ship slamming. A hydrodynamic impact is a shock between a liquid and a solid. This is a very common phenomenon in nature and is a problem in many industrial fields, including marine and offshore engineering. The main objective of this work is to study secondary impacts, that is to say impacts on the upper parts of a ship after flow separation, following the primary impact. 2D and 3D numerical models using the VOF method and an Euler-Lagrange coupling technique were developed. Two configurations were studied: a ship section with a roll angle and wedge with a bulb on their lower part. In parallel, experimental tests were carried out using a hydraulic impact machine. Comparisons between tests and simulations have generally supported the numerical approach. An interesting phenomenon of secondary force peak related to the rapid contraction of the air cavities generated by the impact was demonstrated, both numerically and experimentally.Ces travaux portent sur la modélisation et l'étude des phénomènes de séparation d'écoulement et d'impact hydrodynamique secondaire qui peuvent se produire lors du tossage des navires. Un impact hydrodynamique correspond à un choc entre un liquide et un solide. Il s'agit d'un phénomène très courant dans la nature et qui constitue une problématique dans de nombreux domaines industriels, notamment l'ingénierie navale et offshore. L'objectif principal de ces travaux est l'étude des impacts secondaires, c'est à dire des impacts ayant lieu sur les parties hautes d'une carène après séparation de l'écoulement, consécutif à l'impact primaire. Des modèles numériques 2D et 3D, utilisant la méthode VOF et une technique de couplage Euler-Lagrange ont été mis au point. Deux configurations ont été étudiées : une section de carène avec un angle de roulis et des dièdres présentant un bulbe sur leur partie inférieure. En parallèle, des essais expérimentaux ont été réalisés à l'aide d’une machine de choc hydraulique. Les comparaisons entre essais et simulations ont globalement conforté l'approche numérique. Un phénomène intéressant de pic secondaire d'effort lié à la contraction rapide des cavités d'air générées par l'impact a été mis en évidence, à la fois numériquement et expérimentalement
Robust Semantic Segmentation with Superpixel-Mix
No full textInternational audienceAlong with predictive performance and runtime speed, robustness is a key requirement for real-world semantic segmentation.Robustness encompasses accuracy, predictive uncertainty, stability under data perturbation and distribution shift, and reduced bias. To improve robustness, we introduce Superpixel-mix, a new superpixel-based data augmentation method with teacher-student consistency training. Unlike other mixing-based augmentation techniques, mixing superpixels between images is aware of object boundaries, while yielding consistent gains in segmentation accuracy. Our proposed technique achieves state-of-the-art results in semi-supervised semantic segmentation on the Cityscapes dataset. Moreover, Superpixel-mix improves the robustness of semantic segmentation by reducing network uncertainty and bias, as confirmed by competitive results under strong distributions shift (adverse weather, image corruptions) and when facing out-of-distribution data
The Tornado Project: An Automated Driving Demonstration in Peri-Urban and Rural Areas
No full textInternational audienceThis paper presents the results of a two-week robottaxi service demonstration in peri-urban and rural areas. A fully robotized Renault ZOE was available for general public use in the Rambouillet Territory in France. The driving zone included several complex scenarios as two-way narrow road driving, a tunnel crossing with lane reduction from two-way road up to a single narrow lane or roundabouts, allowing to evaluate the maturity of the vehicle for such application. This paper describes the scientific and technical development especially from perception, navigation and control point of view to carry out such demonstration. Results indicate that even if the vehicle was able to autonomously navigate through peri-urban and rural areas, there are still some technical challenges that limit its integration with the transport system
Physics-based auralization of wind turbine noise
No full textInternational audienceAmplitude modulation of wind turbine noise is known to be a potential source of annoyance for people living in the vicinity of wind farms. To better understand this auditory annoyance, we propose to auralize the sound that is generated by the wind turbines, rather than to observe a visual representation of the sound levels. It is desirable for the developed auralization tool to be physically-based rather than sample-based. This allows control over the prevailing physical parameters. In our work, the auralization tool is based on Amiet's theory in the frequency domain, and considers the main broadband aerodynamic noise sources, namely trailing edge noise and turbulent inflow noise. For the auralization of the full wind turbine noise, the power spectral density for each blade segment and each position is considered along with the appropriate time shift due to the propagation between the moving blades and the fixed observer. In this study, an efficient method is discussed for the conversion of the frequency-domain power spectral density into a time domain signal. The appropriate time delay due to propagation is accounted for. Finally, a proper implementation of energy conserving cross-fading between consecutive signal grains is proposed. The complete auralized signal for the wind turbine noise in free field is then computed with different receiver orientations and meteorological conditions and compared with the original results in the frequency domain. This auralization tool combined with Virtual Reality/ Augmented Reality can help in building the wind farms while also accounting for auditory annoyance factor in the design phase
An Electrochemical Study of Bis(cyclopentadienyl)titanium(IV) Dichloride in the Presence of Magnesium Ions, Amides or Alkynes
No full textInternational audienceIn tetrahydrofuran, the electrochemical reduction of Cp2TiIVCl2 (2 mM) generated three titanium(III) complexes which were in equilibrium: [Cp2TiCl2]•−, [Cp2TiCl]• and (Cp2TiCl)2. Although the anion radical [Cp2TiCl2]•− was the main species produced under these conditions, cyclic voltammetry investigations clearly showed that the proportion of the three electrogenerated TiIII complexes can be modified as a function of the amounts of chloride ion present in the solution. Accordingly, the presence of Mg2+ ions, which led to the consumption of chloride ions through the formation of MgCl2, favoured the formation of [Cp2TiCl]• and, consequently, of the corresponding dimer (Cp2TiCl)2. The electrochemical behaviours of Cp2TiIVCl2 and of the electrogenerated low-valent Ti complexes were also investigated in the presence of amide and alkyne derivatives. Under these conditions, titanium complexes could not only interact with the amide carbonyl group, but also with the alkyne triple bond, provided the latter was not sterically hindered. Interestingly, the carbonyl group and the triple bond had antagonist effects on redox properties of titanium(III) complexes
Agents Autotéliques Vygostkiens : Buts, Langage et Apprentissage Intrinsèquement Motivé
No full textBuilding autonomous machines that can explore large environments, discover interesting interactions and learn open-ended repertoires of skills is a long-standing goal in artificial intelligence. Inspired by the remarkable lifelong learning of humans, the field of developmental machine learning aims at studying the mechanisms enabling autonomous machines to self-organize their own developmental trajectories and grow their own repertoires of skills. This research makes steps towards that goal.Reinforcement learning methods (RL) train learning agents to control their environment by maximizing future rewards and, thus, seem adapted to our purpose. Although it achieved impressive results in the last decade---beating humans at video games, chess, go or controlling robotic agents---it falls short of solving our goal. Indeed, RL agents demonstrate low autonomy and open-endedness because they usually target a (small) set of pre-defined tasks characterized by hand-defined reward functions. In this research, we transfer, adapt and extend ideas from a developmental framework called intrinsically motivated goal exploration process (IMGEP) to the RL setting. The resulting framework builds on goal-conditioned RL techniques to design autotelic RL agents: agents that are intrinsically motivated to represent, generate, pursue and master their own goals as a way to grow repertoires of skills.The efficient acquisition of open-ended repertoires of skills further requires agents to creatively generate novel goals out of the domain of known effects (creative exploration), to readily generalize their understanding of known skills to similar ones (systematic generalization), and to compose known skills to form new ones (composition). Inspired by developmental psychology, we propose to use language as a cognitive tool to support such properties.We organize the manuscript around these two notions: goals and language. The first part focuses on goals. It covers foundational concepts and related work on intrinsic motivations, reinforcement learning and developmental robotics before introducing our framework, goal-conditioned intrinsically motivated goal exploration process (GC-IMGEP), the intersection of RL and IMGEPs. Building on this framework, we present three computational studies of the properties of autotelic agents. We first show that we can use autotelic exploration to solve external hard-exploration tasks (study 1: GEP-PG and 2: ME-ES). We then move on to reward-free environments and propose CURIOUS, an autotelic agent that targets a diversity of goals, transfers knowledge across skills and organizes its own learning trajectory by pursuing goals associated with high learning progress (study 3).The second part focuses on language. Inspired by the pioneering work of Vygotsky and others, we first discuss existing communicative and cognitive uses of language for goal-directed artificial agents. Language facilitates human-agent communications, abstraction, systematic generalization, long-horizon control, but also creativity and mental simulations. In two subsequent computational studies, we propose to implement these two last cognitive uses of language. IMAGINE uses language both to learn goal representations from social interactions (communicative use) and to imagine out-of-distribution goals used to drive its creative exploration and enhance systematic generalization (cognitive use). In our last study, LGB trains a language-conditioned world model to generate a diversity of possible futures conditioned on linguistic descriptions. This leads to behavioral diversity and strategy-switching behaviors.Concevoir des machines autonomes qui explorent des environnements larges, découvrent des interactions pertinentes et développent des répertoires de comportements non-bornés est un des défis majeurs en intelligence artificielle. Inspiré par le remarquable apprentissage de l'humain, l'apprentissage machine développemental étudie les mécanismes permettant aux machines d'auto-organiser leurs trajectoires développementales et de développer des répertoires de comportements. Notre recherche progresse vers ce but.L’apprentissage par renforcement (RL) entraîne des agents à contrôler leur environnement de sorte à maximiser des récompenses et apparaît donc adapté à notre objectif. Malgré ses récent succès---battre l’humains à certains jeux vidéos, aux échecs, au go ou contrôler des robots---le RL ne saurait être suffisant : les agents RL sont peu autonomes et montrent des comportements bornés car ils s'attaquent à de (petits) sets de tâches pré-définies, caractérisées par des fonctions de récompenses pré-codées. Dans cette recherche, nous proposons de transférer, d'adapter et d'étendre des idées issues d'une approche de robotique développementale appelée processus d'exploration de buts intrinsèquement motivés (IMGEP) aux méthodes de RL. Notre nouveau cadre algorithmique étend les techniques de RL conditionné par des buts pour développer des agents RL autotéliques: des agents intrinsèquement motivés à représenter, générer, poursuivre et maîtriser leurs propres buts en vue de développer des répertoires de comportements.L'acquisition efficace de répertoires de comportements non-bornés nécessite une génération créative de buts en dehors de la distribution des effets connus (exploration créative), la généralisation de comportements connus à des comportements nouveaux (généralisation systématique) et la capacité à composer des comportements connus pour en former de nouveaux (composition). Inspiré par la psychologie développementale, nous proposons d'utiliser le langage comme un outil cognitif de sorte à soutenir ces propriétés.Ce manuscrit est construit autour de deux notions: les buts et le langage. La première partie se concentre sur les buts. Elle couvre les concepts fondamentaux et la littérature associée traitant des motivations intrinsèques, de l'apprentissage par renforcement et de la robotique développementale avant d'introduire notre framework: les processus d'exploration de buts intrinsèquement motivés avec des politiques conditionnées par des buts (GC-IMGEP). À partir de ce cadre, nous présentons trois études computationnelles des propriétés des agents autotéliques. Nous montrons d'abord que l'exploration autotélique peut être utilisée pour résoudre des tâches nécessitant une importante exploration (étude 1: GEP-PG et 2: ME-ES). Nous proposons ensuite CURIOUS dans un environnement sans récompense: un agent autotélique qui vise une diversité de buts, transfère de l'information entre compétences et organise sa trajectoire d'apprentissage en poursuivant les buts liés à de forts progrès (étude 3).La seconde partie se concentre sur le langage. Inspirés par les travaux de Vygostky et d'autres, nous discutons des utilisations des capacités communicatives et cognitives du langage dans le cadre d'agents dirigés par des buts. Le langage facilite les interactions humain-agent, l'abstraction, la généralisation systématique, le contrôle à long horizon temporel, mais aussi la créativité et la simulation mentale. Dans les deux études computationnelles qui suivent, nous implémentons ces deux dernières capacités. IMAGINE utilise le langage pour apprendre des représentations de buts (usage communicatif) et pour imaginer de nouveaux buts de sorte à diriger une exploration créative (usage cognitif). Dans notre dernière étude, LGB entraîne un modèle du monde à générer une diversité de futurs possibles à partir de descriptions linguistiques. Cela mène à une plus grande diversité comportementale et à des comportements de changement de stratégie
