Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
Stratégies auto-reconfigurables basées sur la détection pour les systèmes robotiques modulaires autonomes
No full textModular robotic systems (MRSs) have become a highly active research today. It has the ability to change the perspective of robotic systems from machines designed to do certain tasks to multipurpose tools capable of accomplishing almost any task. They are used in a wide range of applications, including reconnaissance, rescue missions, space exploration, military task, etc. Constantly, MRS is built of “modules” from a few to several hundreds or even thousands. Each module involves actuators, sensors, computational, and communicational capabilities. Usually, these systems are homogeneous where all the modules are identical; however, there could be heterogeneous systems that contain different modules to maximize versatility. One of the advantages of these systems is their ability to operate in harsh environments in which contemporary human-in-the-loop working schemes are risky, inefficient and sometimes infeasible. In this thesis, we are interested in self-reconfigurable modular robotics. In such systems, it uses a set of detectors in order to continuously sense its surroundings, locate its own position, and then transform to a specific shape to perform the required tasks. Consequently, MRS faces three major challenges. First, it offers a great amount of collected data that overloads the memory storage of the robot. Second it generates redundant data which complicates the decision making about the next morphology in the controller. Third, the self reconfiguration process necessitates massive communication between the modules to reach the target morphology and takes a significant processing time to self-reconfigure the robotic. Therefore, researchers’ strategies are often targeted to minimize the amount of data collected by the modules without considerable loss in fidelity. The goal of this reduction is first to save the storage space in the MRS, and then to facilitate analyzing data and making decision about what morphology to use next in order to adapt to new circumstances and perform new tasks. In this thesis, we propose an efficient mechanism for data processing and self-reconfigurable decision-making dedicated to modular robotic systems. More specifically, we focus on data storage reduction, self-reconfiguration decision-making, and efficient communication management between modules in MRSs with the main goal of ensuring fast self-reconfiguration process.Les systèmes robotiques modulaires (MRS) font aujourd’hui l’objet de recherches très actives. Ils ont la capacité de changer la perspective des systèmes robotiques, passant de machines conçues pour effectuer certaines tâches à des outils polyvalents capables d'accomplir presque toutes les tâches. Ils sont utilisés dans un large éventail d'applications, notamment la reconnaissance, les missions de sauvetage, l'exploration spatiale, les tâches militaires, etc. Constamment, MRS est constitué de "modules" allant de quelques à plusieurs centaines, voire milliers. Chaque module implique des actionneurs, des capteurs, des capacités de calcul et de communication. Habituellement, ces systèmes sont homogènes où tous les modules sont identiques ; cependant, il pourrait y avoir des systèmes hétérogènes contenant différents modules pour maximiser la polyvalence. L’un des avantages de ces systèmes est leur capacité à fonctionner dans des environnements difficiles dans lesquels les schémas de travail contemporains avec intervention humaine sont risqués, inefficaces et parfois irréalisables. Dans cette thèse, nous nous intéressons à la robotique modulaire auto-reconfigurable. Dans de tels systèmes, il utilise un ensemble de détecteurs afin de détecter en permanence son environnement, de localiser sa propre position, puis de se transformer en une forme spécifique pour effectuer les tâches requises. Par conséquent, MRS est confronté à trois défis majeurs. Premièrement, il offre une grande quantité de données collectées qui surchargent la mémoire de stockage du robot. Deuxièmement, cela génère des données redondantes qui compliquent la prise de décision concernant la prochaine morphologie du contrôleur. Troisièmement, le processus d'auto-reconfiguration nécessite une communication massive entre les modules pour atteindre la morphologie cible et prend un temps de traitement important pour auto-reconfigurer le robot. Par conséquent, les stratégies des chercheurs visent souvent à minimiser la quantité de données collectées par les modules sans perte considérable de fidélité. Le but de cette réduction est d'abord d'économiser de l'espace de stockage dans le MRS, puis de faciliter l'analyse des données et la prise de décision sur la morphologie à utiliser ensuite afin de s'adapter aux nouvelles circonstances et d'effectuer de nouvelles tâches. Dans cette thèse, nous proposons un mécanisme efficace de traitement de données et de prise de décision auto-reconfigurable dédié aux systèmes robotiques modulaires. Plus spécifiquement, nous nous concentrons sur la réduction du stockage de données, la prise de décision d'auto-reconfiguration et la gestion efficace des communications entre les modules des MRS dans le but principal d'assurer un processus d'auto-reconfiguration rapide
Nouvelle méthode objective pour extraire une mesure quantitative de la transparence cornéenne à partir d’images résolues en profondeur
No full textInternational audienc
Flexible Plenoptic X-ray Microscopy
No full textInternational audienceX-ray computed tomography (CT) is an invaluable technique for generating three-dimensional (3D) images of inert or living specimens. X-ray CT is used in many scientific, industrial, and societal fields. Compared to conventional 2D X-ray imaging, CT requires longer acquisition times because up to several thousand projections are required for reconstructing a single high-resolution 3D volume. Plenoptic imaging—an emerging technology in visible light field photography—highlights the potential of capturing quasi-3D information with a single exposure. Here, we show the first demonstration of a flexible plenoptic microscope operating with hard X-rays; it is used to computationally reconstruct images at different depths along the optical axis. The experimental results are consistent with the expected axial refocusing, precision, and spatial resolution. Thus, this proof-of-concept experiment opens the horizons to quasi-3D X-ray imaging, without sample rotation, with spatial resolution of a few hundred nanometres
Third-order nonlinear femtosecond optical gating through highly scattering media
No full textInternational audienceDiscriminating between ballistic and diffuse components of light propagating through highly scattering media is not only important for imaging purposes but also for investigating the fundamental diffusion properties of the medium itself. Massively developed to this end over the past 20 years, nonlinear temporal gating remains limited to ∼10−10 transmission factors. Here, we report nonlinear time-gated measurements of highly scattered femtosecond pulses with transmission factors as low as ≈10−12. Our approach is based on the third-order nonlinear cross-correlation of femtosecond pulses, a standard diagnostic used in high-power laser science, applied to the study of fundamental light scattering properties
Underwater Eversion Robot Growth for Underwater Operations with an Emphasis on Underwater Mine Hunting
No full textInternational audienc
Shallow-water waveguide acoustic analysis in a fluctuating environment
No full textInternational audienceThe Acoustic Laboratory for Marine Applications (ALMA) is a deployable and autonomous acoustic system, designed by DGA Naval Systems, to address problems in underwater acoustics, such as sound propagation in fluctuating environments. In this article, data from the ALMA-2016 at-sea campaign are used to analyze the ocean fluctuation's influence on sound propagation in a shallow-water waveguide. The experiment took place on the continental shelf of the island of Corsica in November 2016. A source and a receiver array were 9.3 km apart in a nearly constant water depth of 100 m. The source emitted a variety of signals from which the chirp (1–13 kHz) is used to extract the waveguide eigenrays. To do so, a time-domain beamforming is performed on the match-filtered received signals with an automatic detection of local maxima in the time of arrival/direction of arrival (TOA/DOA) domain. A 2 min acquisition period of more than 13 h duration shows significant fluctuations in eigenray TOAs/DOAs. Qualitative comparisons with synthetic signals obtained from simulations in two and three dimensions permit reproduction of the observed eigenray fluctuations without including range dependence of the sound-speed profile
Drone Detection and Classification Using Physical-Layer Protocol Statistical Fingerprint
No full textInternational audienceWe propose a novel approach for drone detection and classification based on RF communication link analysis. Our approach analyses large signal record including several packets and can be decomposed of two successive steps: signal detection and drone classification. On one hand, the signal detection step is based on Power Spectral Entropy (PSE), a measure of the energy distribution uniformity in the frequency domain. It consists of detecting a structured signal such as a communication signal with a lower PSE than a noise one. On the other hand, the classification step is based on a so-called physical-layer protocol statistical fingerprint (PLSPF). This method extracts the packets at the physical layer using hysteresis thresholding, then computes statistical features for classification based on extracted packets. It consists of performing traffic analysis of communication link between the drone and its controller. Conversely to classic drone traffic analysis working at data link layer (or at upper layers), it performs traffic analysis directly from the corresponding I/Q signal, i.e., at the physical layer. The approach shows interesting properties such as scale invariance, frequency invariance, and noise robustness. Furthermore, the classification method allows us to distinguish WiFi drones from other WiFi devices due to underlying requirement of drone communications such as good reactivity in control. Finally, we propose different experiments to highlight theses properties and performances. The physical-layer protocol statistical fingerprint exploiting communication specificities could also be used in addition of RF fingerprinting method to perform authentication of devices at the physical-layer
Explaining Aha! moments in artificial agents through IKE-XAI: Implicit Knowledge Extraction for eXplainable AI
No full textInternational audienceDuring the learning process, a child develops a mental representation of the task he or she is learning. A Machine Learning algorithm develops also a latent representation of the task it learns. We investigate the development of the knowledge construction of an artificial agent through the analysis of its behavior, i.e., its sequences of moves while learning to perform the Tower of Hanoï (TOH) task. The TOH is a wellknown task in experimental contexts to study the problem-solving processes and one of the fundamental processes of children's knowledge construction about their world. We position ourselves in the field of explainable reinforcement learning for developmental robotics, at the crossroads of cognitive modeling and explainable AI. Our main contribution proposes a 3-step methodology named Implicit Knowledge Extraction with eXplainable Artificial Intelligence (IKE-XAI) to extract the implicit knowledge, in form of an automaton, encoded by an artificial agent during its learning. We showcase this technique to solve and explain the TOH task when researchers have only access to moves that represent observational behavior as in human-machine interaction. Therefore, to extract the agent acquired knowledge at different stages of its training, our approach combines: first, a Q-learning agent that learns to perform the TOH task; second, a trained recurrent neural network that encodes an implicit representation of the TOH task; and third, an XAI process using a post-hoc implicit rule extraction algorithm to extract finite state automata. We propose using graph representations as visual and explicit explanations of the behavior of the Q-learning agent. Our experiments show that the IKE-XAI approach helps understanding the development of the Q-learning agent behavior by providing a global explanation of its knowledge evolution during learning. IKE-XAI also allows researchers to identify the agent's Aha! moment by determining from what moment the knowledge representation stabilizes and the agent no longer learns
Techniques d'accélération d'une méthode de Branch-and-bound pour l'optimisation parcimonieuse
No full textInternational audienceLes problèmes d'ajustement de modèles de faible cardinalité ont trouvé de nombreuses applications en statistique, en finance et en traitement du signal. Au sein de ces problèmes, nous nous intéressons au problème de l'ajustement par moindres carrés, pénalisé par la cardinalité de la solution. Nous utilisons un algorithme branch-and-bound pour trouver l'optimum global de ce problème NP-complet. Au sein de cet algorithme, les bornes inférieures évaluées à chaque noeud sont calculées par la résolution de problèmes en norme 1, qui disposent d'une large panoplie de méthodes dédiées. Dans cette communication, nous exposons deux techniques exploitant la dualité convexe pour, d'une part, éviter de résoudre certains problèmes de relaxation jusqu'à l'optimalité, permettant d'accélérer le calcul des bornes inférieures, et d'autre part réduire la dimension de ces problèmes par une stratégie de screening. Une étude expérimentale valide la pertinence de ces techniques pour réduire le temps de calcul