Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
Evaluation of two boundary integral formulations for the Eddy current nondestructive testing of metal structures
No full textInternational audienceWe investigate two bounary integral formulations for the resolution of the Maxwell equations in the Eddy Current (EC) regime in a context of nondestructive testing (NdT). The first one, based on an approximation of the Maxwell equations, requires a loop-star decomposition of the surface currents and the global loops are constructed manually for non-simply connected domains. The second formulation is stabilized by using quasi-Helmholtz projectors, thus avoiding the definition of global loops
Contributions à l’étude de la cohérence spatiale bi-dimensionnelle dans le cadre du développement d’un sonar à ouverture synthétique basse fréquence
No full textFirst applied to light, the notion of spatial coherence is related to the degree of correlation between two signals acquired at two points in space. In the field of sonar, spatial coherence is a central concept that affects the spatial resolution of systems as well as the performance of processes such as interferometry and aperture synthesis. The configuration of the sonar system, the signal processing parameters, and external disturbances such as changing environmental conditions all affect spatial coherence. In the field of mine countermeasures, low-frequency synthetic aperture sonars are currently being developed, in particular to detect buried objects. The aim of this thesis is to develop an analytical model of the spatial coherence of such systems. To this end, existing models developed for vertical acquisition geometry have been extended to be compatible with lateral acquisition. However, in lateral geometry, geometric decorrelations not taken into account in the model reduce the level of coherence. For this reason, numerical modelling of the signals is proposed in order to study the influence of these decorrelations and to estimate the coherence levels after compensation for the geometric decorrelation phenomena.Appliquée pour la première fois à la lumière, la notion de cohérence spatiale exprime le degré de corrélation entre deux signaux acquis en deux points de l’espace. Dans le domaine sonar, la cohérence spatiale est un concept central qui affecte la résolution spatiale des systèmes ainsi que les performances de procédés tels que l’interférométrie et la synthèse d’ouverture. La configuration du système sonar, les paramètres de traitement du signal, et les perturbations externes telles que des conditions environnementales changeantes affectent la cohérence spatiale. Dans le domaine de la lutte contre les mines, des sonars à ouverture synthétique basse fréquence sont en cours de développement en particulier afin de permettre la détection d’objets enfouis. L'objectif de cette thèse est de développer une modélisation analytique de la cohérence spatiale de tels systèmes. Pour cela, des modèles existants développés pour la géométrie d’acquisition verticale ont été étendus pour être compatibles d’une acquisition latérale. Cependant, en géométrie latérale, des décorrélations géométriques non prises en compte dans le modèle réduisent le niveau de cohérence. Pour cela, une modélisation numérique des signaux est proposée afin d’étudier l’influence de ces dernières et d’estimer les niveaux de cohérence après compensation des phénomènes de décorrélation géométriques
Online Learning a Symbolic Abstraction of Actions in Hierarchical RL with Formal Methods
No full textInternational audienceReasoning about actions, planning in open-ended learning requires a hierarchical abstractions of action models and benefits from the use of symbolic methods for goal representation as they structure knowledge for efficient and transferable learning. However, the existing Hierarchical Reinforcement Learning (HRL) approaches relying on symbolic reasoning are often limited as they require a manual goal representation. A symbolic goal representation must preserve information about the environment dynamics. We propose an automatic subgoal discovery via an emergent representation that abstracts (i.e., groups together) sets of environment states that have similar roles in the task, using formal verification methods. We create a HRL algorithm that learns online this representation along with the policies. We show on navigation tasks that the learned representation is interpretable and results in data efficiency
Generative Pretrained Embedding and Hierarchical Irregular Time Series Representation for Daily Living Activity Recognition
No full textInternational audienceWithin the evolving landscape of smart homes, the precise recognition of daily living activities using ambient sensor data stands paramount. This paper not only aims to bolster existing algorithms by evaluating two distinct pretrained embeddings suited for ambient sensor activations but also introduces a novel hierarchical architecture. We delve into an architecture anchored on Transformer Decoder-based pre-trained embeddings, reminiscent of the GPT design, and contrast it with the previously established state-of-the-art (SOTA) ELMo embeddings for ambient sensors. Our proposed hierarchical structure leverages the strengths of each pre-trained embedding, enabling the discernment of activity dependencies and sequence order, thereby enhancing classification precision. To further refine recognition, we incorporate into our proposed architecture an hour-of-the-day embedding. Empirical evaluations underscore the preeminence of the Transformer Decoder embedding in classification endeavors. Additionally, our innovative hierarchical design significantly bolsters the efficacy of both pre-trained embeddings, notably in capturing inter-activity nuances. The integration of temporal aspects subtly but distinctively augments classification, especially for time-sensitive activities. In conclusion, our GPT-inspired hierarchical approach, infused with temporal insights, outshines the SOTA ELMo benchmark
La motivation intrinsèque d'apprendre par renforcement et imitation des tâches séquentielles: Analyse de mouvement, analyse d'activité de la vie quotidienne et motivation intrinsèque d'un robot pour apprentissage par renforcement hiérarchique
No full textMy work in the field of developmental cognitive robotics aims to devise a new domain bridging between reinforcement learning and imitation learning, with a model of the intrinsic motivation for learning agents to learn with guidance from tutors multiple tasks, including sequential tasks. My main contribution has been to propose a common formulation of intrinsic motivation based on empirical progress for a learning agent to choose automatically its learning curriculum by actively choosing its learning strategy for simple or sequential tasks: which task to learn, between autonomous exploration or imitation learning, between low-level actions or task decomposition, between several tutors. The originality is to design a learner that benefits not only passively from data provided by tutors, but to actively choose when to request tutoring and what and whom to ask. The learner is thus more robust to the quality of the tutoring and learns faster with fewer demonstrations.With my previous advisors, colleagues and my collaborators, we developed the framework of socially guided intrinsic motivation with machine learning algorithms to learn multiple tasks by taking advantage of the generalisability properties of human demonstrations in a passive manner or in an active manner through requests of demonstrations from the best tutor for simple and composing subtasks. The latter relies on a representation of subtask composition proposed for a construction process, which should be refined by representations used for observational processes of analysing human movements and activities of daily living. With the outlook of a language-like communication with the tutor, we investigated the emergence of a symbolic representation of the continuous sensorimotor space and of tasks using intrinsic motivation. These works proposed within the reinforcement learning framework, a reward function for interacting with tutors for multi-task learning. The formulation of this reward unifies interactive learning, multi-task learning and hierarchical learning through the same empirical competence progress measure as the intrinsic motivation to choose the different aspects its learning strategy.The socially guided intrinsic motivation framework could be used for intelligent tutoring systems to be applied in socially assistive robotics for a robot coach for physical rehabilitation of low-back pain patients analysing whole body movements, or for a robot coach for ASD children in an imitation game. The project I wish to develop in cognitive developmental learning pertains both to the reinforcement learning framework of machine learning, to control learning in the perception-action loop of cognitive robotics and the theory of intrinsic motivation from developmental psychology in cognitive science. The objective is to propose new reinforcement learning algorithms for robots to learn by actively interacting with their environment, enabling robots to master control tasks of growing complexity, while extending the current theory of intrinsic motivation towards a unified model of motivations for interacting with tutors. The project relies on two axes : a model of interaction with tutors relying on an emerging representation of tasks by the robot, and a model of motivation of human learners through the use case of a robot coach for physical rehabilitation. This work will enrich the formulation of the motivation for social guidance with various motivational drives of learners to interact with tutors, taking the perspective of communication and emotions. This will also highlight the importance of emerging internal representations for understanding social guidance, thus emphasising the interdependence between autonomous learning and social learning
Automatic Image Annotation for Mapped Features Detection
No full textInternational audienceDetecting road features is a key enabler for autonomous driving and localization. For instance, a reliable detection of poles which are widespread in road environments can improve localization. Modern deep learning-based perception systems need a significant amount of annotated data. Automatic annotation avoids time-consuming and costly manual annotation. Because automatic methods are prone to errors, managing annotation uncertainty is crucial to ensure a proper learning process. Fusing multiple annotation sources on the same dataset can be an efficient way to reduce the errors. This not only improves the quality of annotations, but also improves the learning of perception models. In this paper, we consider the fusion of three automatic annotation methods in images: feature projection from a high accuracy vector map combined with a lidar, image segmentation and lidar segmentation. Our experimental results demonstrate the significant benefits of multi-modal automatic annotation for pole detection through a comparative evaluation on manually annotated images. Finally, the resulting multi-modal fusion is used to fine-tune an object detection model for pole base detection using unlabeled data, showing overall improvements achieved by enhancing network specialization. The dataset is publicly available
Comparative performances of CNN models for SAR Targets classification
No full textInternational audienceThis article focuses on the development of classification architectures for Synthetic Aperture Radar (SAR) image analysis, particularly for target recognition. Several methods based on deep artificial neural networks are explored and compared. Specifically, this work investigates the study of pretrained architectures such as Xception and DenseNet, which were originally developed and trained on the ImageNet database containing optical images. However, adaptation of these architectures is necessary within the context of SAR radar images. The goal was to take advantage of the powerful feature extraction capabilities of these models to effectively classify objects in radar images. We evaluated and compared the classification performances of each technique using the publicly MSTAR dataset. This led to the proposal and development of new architectures based on the Xception and DenseNet models. Using these models, it is possible to achieve impressive recognition rates close to 99.5% on the test dataset, surpassing several benchmarks reported in the scientific literature on the same dataset
Range Cell Migration Processing Loss in Automotive Radar
No full textInternational audienceFrequency Modulated Continuous Wave (FMCW) automotive radar using stretch-processing usually relies on a fast-chirp signal model which assumes range and Doppler decoupling between fast and slow-time dimensions. To achieve high range and speed resolutions, a large bandwidth and a long coherent processing interval are used. However, if multiple range resolution cells are crossed during the integration duration, fast-chirp signal model assumption becomes inaccurate, resulting in misfocusing in both fast and slow-time dimensions. This misfocusing leads to the misestimation of target parameters and a loss of coherent integration. In this paper, Range Cell Migration (RCM) processing loss is studied and illustrated using a FMCW signal model. Furthermore, a windowing in both dimensions is proposed to mitigate this loss
Towards Bio-inspired Heuristically Accelerated Reinforcement Learning for Adaptive Underwater Multi-Agents Behaviour
No full textInternational audienceThis paper describes the problem of coordination and collaboration of an autonomous Multi-Agent System (MAS) which aims to solve the coverage planning problem in a complex environment. The considered applications are the detection and identification of objects of interest while covering an area. These tasks, which are highly relevant for space applications, are also of interest among various domains including the underwater context, which is the focus of this study. In this context, coverage planning is traditionally modelled as a Markov Decision Process (MDP) where a coordinated MAS, i.e. a swarm of heterogeneous autonomous underwater vehicles, is required to survey an area and search for objects. This MDP is associated with several challenges: environment uncertainties, communication constraints, and an ensemble of hazards, including time-varying and unpredictable changes in the underwater environment. Multi-Agent Reinforcement Learning (MARL) algorithms can solve highly nonlinear problems using deep neural networks and display great scalability against an increased number of agents. Nevertheless, most of the current results in the underwater domain are limited to simulation due to the high learning time of MARL algorithms. For this reason, a novel strategy is introduced to accelerate this convergence rate by incorporating biologically inspired heuristics to guide the policy during training. The Particle Swarm Optimization (PSO) method, which is inspired by the behaviour of a group of animals, is selected as a heuristic. It allows the policy to explore the highest quality regions of the action and state spaces, from the beginning of the training, optimizing the exploration/exploitation trade-off. The resulting agent requires fewer interactions to reach optimal performance. The method is applied to the Multi-Agent Soft Actor-Critic (MASAC) algorithm and evaluated for a 2D covering area mission in a continuous control environment. The results demonstrate that the proposed method reduces the training time
EELS hyperspectral images unmixing using autoencoders
No full textInternational audienceSpatially resolved Electron Energy-Loss Spectroscopy conducted in a Scanning Transmission Electron Microscope enables the acquisition of hyperspectral images. Spectral unmixing is the process of decomposing each spectrum of a hyperspectral image into a combination of representative spectra (endmembers) corresponding to compounds present in the sample along with their local proportions (abundances). Spectral unmixing is a complex task, and various methods have been developed in different communities using hyperspectral images. However, none of these methods fully satisfy the spatially resolved Electron Energy-Loss Spectroscopy requirements. Recent advancements in remote sensing, which focus on Deep Learning techniques, have the potential to meet these requirements, particularly Autoencoders. As the Neural Networks used are usually shallow it would be more appropriate to use the term “representation learning”. In this study, the performance of these methods using autoencoders for spectral unmixing is evaluated, and their results are compared with traditional methods. Synthetic hyperspectral images have been created to quantitatively assess the outcomes of the unmixing process using specific metrics. The methods are subsequently applied to a series of experimental data. The findings demonstrate the promising potential of autoencoders as a tool for Electron Energy-Loss Spectroscopy hyperspectral images unmixing, marking a starting point for exploring more sophisticated Neural Networks