HAL Portal UTC Université de Technologie de Compiègne
Not a member yet
11652 research outputs found
Sort by
Navigation autonome multi-risques tenant compte de l’incertitude induite par les véhicules électriques légers personnels
No full textModern transportation is being transformed by autonomous vehicles (AVs), which promise enhanced safety, efficiency, and convenience. These vehicles rely on complex algorithms to perceive their environment, make decisions, and control motion in real time. A critical component is the risk assessment and management unit, which analyzes traffic scenarios for potential hazards and determines the safest control outputs. This process requires predicting and simulating multiple road scenarios, leading to various hypotheses and algorithms for estimating the future states of surrounding road users.The emergence of Personal Light Electric Vehicles (PLEVs), such as electric scooters and bikes, has increased uncertainty on the road. These vehicles exhibit multi-modal behavior—moving like pedestrians at low speeds and accelerating rapidly—and their small size complicates detection. Two main challenges arise: uncertainty in predicting the states (position, velocity, size) of surrounding agents, and managing multi-modal motion predictions without overly conservative control actions. This PhD thesis, conducted under the ANNAPOLIS (AutoNomous Navigation Among Personal mObiLity devIceS) project, addresses these challenges to ensure safe and comfortable AV navigation in dynamic and constrained urban environments. The research aims to design and validate a Multi-Risk Assessment and Multi-level Motion Optimization \& Control (MiRA-MOC) architecture as a comprehensive multi-risk assessment and management system.Specifically, the thesis addresses autonomous driving challenges posed by uncertainties in the dynamic, possibly erratic motion of surrounding agents, including PLEVs. It introduces a decision-making and control strategy based on a multi-level motion optimization framework that handles motion uncertainties using a Fusion of stochastic Predictive Inter-Distance Profiles (F-sPIDP). This continuous multi-risk assessment metric projects uncertainty in agents’ motion onto the predicted inter-distance between the AV and its surroundings.To address the first challenge, F-sPIDP extends the Predictive Inter-Distance Profile (PIDP) concept to a stochastic version (sPIDP) that captures uncertainty in predicted states. The second challenge in multi-modal prediction is handled by fusing multiple sPIDPs. An optimal trajectory is then selected from possible maneuvers using a combination of safe global trajectory sampling and F-sPIDP. Finally, local control optimization computes control actions that minimize collision risk while respecting vehicle dynamics. The proposed strategy has been validated through extensive simulations and real-vehicle experiments.Le transport moderne est transformé par les véhicules autonomes (VA), qui promettent une meilleure sécurité, une efficacité accrue et une plus grande commodité. Ces véhicules utilisent des algorithmes complexes pour observer leur environnement, prendre des décisions et contrôler leur mobilité en temps réel. Un élément clé de ces algorithmes est constitué des unités d’évaluation et de gestion du risque, qui analysent les scénarios de circulation afin de détecter les dangers potentiels, notamment vis-à-vis des autres usagers, et de calculer les meilleures commandes pour une navigation sûre. Cela nécessite de prévoir et de simuler de multiples scénarios routiers, ce qui a conduit au développement de nombreuses hypothèses et algorithmes de prédiction pour estimer les états futurs des usagers.Cette incertitude s’est accentuée avec l’introduction des Véhicules Personnels Légers Électriques (PLEVs), comme les trottinettes et vélos électriques. Ces dispositifs peuvent présenter des comportements multimodaux, se déplacer à basse vitesse comme des piétons puis accélérer soudainement, et leurs petites tailles compliquent leur détection. Deux défis majeurs se présentent : l’incertitude liée à l’état prédit (position, vitesse, taille) des agents environnants, et la difficulté de gérer les prédictions multi-modales sans adopter une approche trop conservatrice. Cette thèse de doctorat, menée dans le cadre du projet ANNAPOLIS (AutoNomous Navigation Among Personal mObiLity devIceS), s’attaque à ces scénarios complexes pour garantir une navigation sûre, confortable et pas excessivement conservatrice dans des environnements urbains dynamiques et contraints. L’objectif principal est de proposer et de valider une architecture globale de contrôle/commande (intitulée MiRA-MOC), permettant d'évaluer simultanément plusieurs risques possibles et agir en conséquent via des processus d'optimisation et de contrôle multi-niveaux, constituant la partie management du risque de MiRA-MOC.Plus précisément, cette thèse aborde les défis de la conduite autonome liés aux incertitudes du mouvement dynamique et parfois erratique des agents environnants, y compris les PLEVs. Elle introduit une stratégie de décision et de contrôle fondée sur une méthode d’optimisation du mouvement multi-niveaux, capturant ces incertitudes via une Fusion of stochastic Predictive Inter-distance Profile (F-sPIDP). En utilisant le F-sPIDP comme métrique continue d’évaluation multi-risques, les incertitudes du mouvement sont projetées sur la distance inter-prévisionnelle entre le VA et son environnement. Pour répondre au premier défi, le F-sPIDP étend le Predictive Inter-Distance Profile (PIDP) à une version stochastique (sPIDP) intégrant l’incertitude dans l’état prédit des agents. Le second défi, la prédiction multi-modale, est traitée par la fusion de plusieurs sPIDPs. Une trajectoire optimale est ensuite sélectionnée parmi les manœuvres possibles grâce à la combinaison de l’échantillonnage de trajectoires sûres et du F-sPIDP. Enfin, les actions de contrôle minimisant le risque de collision et respectant la dynamique du VA sont calculées par une optimisation du contrôle local. La stratégie a été validée par des simulations intensives et des essais sur un véhicule autonome réel
Soft learning probabilistic circuits
No full textInternational audienceProbabilistic Circuits (PCs) are prominent tractable probabilistic models, allowing for a wide range of exact inferences. This paper focuses on a main algorithm for training PCs, LearnSPN, arguably a gold standard due to its efficiency, performance, and ease of use, in particular for tabular data. We show that LearnSPN is a greedy likelihood maximizer under mild assumptions. While inferences in PCs may use the entire circuit structure for processing queries, LearnSPN applies a hard method for learning PCs, propagating at each sum node a data point through one and only one of the children/edges as in a hard clustering process. We propose a new learning procedure named SoftLearn, that induces a PC using a soft clustering process. We investigate the effect of this learning-inference compatibility in PCs. Our experiments show that SoftLearn outperforms LearnSPN in many situations, yielding better likelihoods and arguably better samples. We also analyze comparable tractable models to highlight the differences between soft/hard learning and model querying
L’agilité : une religion managériale sans croyants ?: Rituels, dogmes et sécularisation dans les pratiques agiles.
No full textInternational audienceModernity, often perceived as a secularization of Christianity, has seen the emergence of social and managerial structures that incorporate religious elements in secular forms. Scientific management, a product of the Industrial Revolution, is one example. However, little attention has been paid to the religious or theological dimension of management. In this context, agile methods, particularly Scrum, offer fertile ground for exploring this transposition of the sacred into the professional world. In this presentation, we will examine the parallels between agile and religious practices and observe how only practitioners still believe in their evangelization.Analyzing agile methods through the lens of secularization highlights the transposition of religious structures into the field of management. Rituals, authority figures, foundational texts, and doctrinal debates recall the dynamics of established religions. However, the limited adherence of practitioners suggests a form of meaningless ritualization, where practices are maintained without the faith that initially justified them. Thus, agility could be perceived as a contemporary managerial religion, marked by a secularization of forms without the substance of belief.La modernité, souvent perçue comme une sécularisation de la religion chrétienne, a vu émerger des structures sociales et managériales qui reprennent des éléments religieux sous des formes profanes. Le management scientifique, produit de la révolution industrielle, en est un exemple. Cependant, peu d'attention a été portée à la dimension religieuse ou théologique du management. Dans ce contexte, les méthodes agiles, notamment Scrum, offrent un terrain fertile pour explorer cette transposition du sacré dans le monde professionnel. Dans cette communication, nous étudierons les parallèles entre pratiques agiles et religieuses, et observerons comment seul les praticiens croient encore en leur évangélisation.L'analyse des méthodes agiles à travers le prisme de la sécularisation met en lumière la transposition de structures religieuses dans le domaine du management. Les rituels, les figures d'autorité, les textes fondateurs et les débats doctrinaux rappellent les dynamiques des religions instituées. Cependant, l'adhésion limitée des praticiens suggère une forme de ritualisation vide de sens, où les pratiques sont maintenues sans la foi qui les justifiait initialement. Ainsi, l'agilité pourrait être perçue comme une religion managériale contemporaine, marquée par une sécularisation des formes sans la substance de la croyance
Full Whittle inference for weak FARIMA models: Full Whittle inference for weak FARIMA models
No full textThis paper investigates statistical inference for weak FARIMA models in the frequency domain. We estimate the asymptotic covariance matrix of the classical Whittle estimator to achieve full inference, thereby addressing an open question posed by Shao, X. (2010). Additionally, we introduce a fast alternative to the Whittle estimator based on a one-step procedure. This method refines an initial Whittle estimator computed on a subsample using a single Fisher scoring step. The resulting estimator retains the same asymptotic properties as the Whittle estimator computed on the full sample while significantly reducing computational time
Optimal Electrohysterography Signal Preprocessing for Delivery Term Prediction Using Hypergraph Neural Networks
No full textInternational audienceThe robust prediction of the infant delivery term through the cooperation of artificial intelligence (AI) and electrohysterogram (EHG) would enable the appropriate early medication for possible premature delivery, thus avoiding death risk or sequels. This paper focuses on unraveling the best preprocessing scheme to be used when dealing with the classification of uterine muscular contraction signals. In addition, the study discusses the impact of several EHG denoising techniques on the prediction outcome. Hypergraph neural network (HGNN) is employed to evaluate the different preprocessing steps. The results show that it is better to start by segmenting the contractions and concatenating them, then standardizing the resulting signal or normalizing it between -1 and 1, and finally segmenting the contractions again in order to characterize them by a set of features that are used to finally train the classifier. The accuracy achieved by the HGNN given this methodology was 89.2%. Moreover, the use of the conventional EHG denoising methods such as canonical correlation analysis (CCA), empirical mode decomposition (EMD), and their combination (EMD-CCA) did not improve the prediction results. As a conclusion, proper preprocessing of the EHG signals would lead to a great improvement in the prediction process.Clinical Relevance— This study presents a new effective approach for preprocessing EHG signals to improve the prediction of delivery term
Regularized variational formulation and 3D discrete approximation with enhanced performance in thermo-elasticity and thermo-damage
No full textInternational audienceAbstract In this paper, we present four-field variational formulation of thermo-elasticity and thermo-damage with non-symmetric stress tensor that is particularly suitable for long-term simulations in thermodynamics needed for fatigue studies. We propose the regularization of this variational formulation in order to enforce the stability of mixed formulation with respect to the Babushka-Brezzi condition. This further allows to eliminate the rotation field, and obtain hybrid-stress format for dynamics and the hybrid heat-flux for transient heat conduction. The regularized variational formulation is combined with a 3D discrete approximation based on the Raviart-Thomas vector space for both mechanic and thermal fields to ensure continuity across tetrahedron finite element faces for the stress vector and normal component of heat-flux. The proposed formulation and the chosen discrete approximations in this study provide superior accuracy for both thermal and mechanical fields compared to the classical FE methods. Both energy-conserving and energy-decaying schemes are utilized for time discretization to ensure robustness in long-term computations. The proposed method is validated through several numerical simulations in both statics and dynamics, including stationary and non-stationary heat conduction, for isotropic thermo-elastic and thermo-damage materials. The potential use in thermal fatigue studies is also demonstrated
Quadrotor Fleet Autonomous Navigation; Fusing Virtual Points Control and Nonlinear Potential Fields
No full textInternational audienceThis paper introduces a multi-layer navigation algorithm for a fleet of unmanned aerial vehicles. The proposed architecture consists on a fusion of virtual point controllers and potential field techniques. On one hand, a potential function is constructed for every agent such that its position smoothly and robustly converges to a virtual guidance point while avoiding collisions with other agents. The virtual points, on the other hand, are controlled to fulfill a swarm control goal such as target tracking, station keeping or search and rescue missions. Therefore, the suggested system has two levels of hierarchy, but the algorithm can be generalized for multiple levels. The vehicle translational and rotational dynamics are controlled using an internal loop based on gradient-tracking, and sliding-mode controllers. The architecture is validated in simulations and real-time experiments, showing good performance for the closed-loop system
Wearable Solutions for Mental Stress Monitoring
No full textInternational audienceStress significantly impacts both mental and physical health, and wearable devices provide anopportunity for continuous monitoring of physiological variables like heart rate and skin conductance.These devices, coupled with advanced noise and artifact-filtering algorithms, can inform users ofhealth and behavior indicators. The goal of this study was to assess wearable devices for theireffectiveness in mental stress monitoring. To achieve this, we conducted a market study, reviewingrelevant articles and examining the accessibility of data from available devices. We specificallyfocused on the Fitbit Sense 2 and Empatica Embrace Plus, evaluating their capabilities in measuringphysiological parameters like heart rate and skin conductance. This evaluation also considered thequality of their data processing algorithms and how the raw data can be used for accurate stressmeasurement
Fine-scale model of concrete composite for long-term cycling transient loads incorporating nonlinear hardening and softening effects
No full textInternational audienceConcrete modeling under long-term cyclic transient loading has always been a challenging task due to the highly heterogeneous nature of the material. In this context, the present study proposes a fine-scale model of concrete by representing it as a two-phase material composed of aggregates and mortar. For this, the material domain is discretized into Voronoi cells connected by cohesive links using Delaunay triangulation, and an aggregate assignment algorithm is proposed to associate these links to either aggregates or mortar. Following this, the cohesive links representing mortar are modeled as 2D Timoshenko beam elements incorporating a distinct nonlinear kinematic hardening model along with isotropic softening model, separately in tension, compression, and shear. Softening in the material is introduced using the Embedded Discontinuity Finite Element Method (EDFEM) to model localized discontinuities.In contrast, the cohesive links associated with aggregates are modeled as elastic 2D Timoshenko beam elements.Lastly, compressive cyclic and three-point bending tests are performed on the concrete specimen, and the results are compared with experimental data reported in the literature, showing very good agreement with the experiments.Additionally, an active Bayesian Optimization (BO) procedure is performed to determine the optimal set of parameters for the three-point bending test by minimizing the Mean Squared Error (MSE) of energy between the present model and the experimental results.</p
Stimulation mécanique en 2D : Un puissant accélérateur de la minéralisation de la matrice chez les cellules chondrogéniques ATDC5
No full textInternational audienceBackgroundMatrix mineralization is a key process in endochondral ossification and cartilage maturation. While optimized biochemical protocols using ATDC5 chondrogenic cells have shortened mineralization timelines, the role of mechanical stimulation in enhancing this process remains underexplored. This study investigates whether dynamic mechanical stimulation in 2D monolayer cultures can accelerate mineralization and influence extracellular matrix (ECM) composition.MethodsATDC5 cells were cultured on PDMS membranes coated with collagen and subjected to cyclic tensile strain (12 % at 0.05 Hz for 1 h/day) using a uniaxial bioreactor over 5 days. The culture protocol included supplementation with β-glycerophosphate and ascorbic acid. Mineralization and ECM production were assessed at days 7, 17, and 23 using Alizarin Red and Alcian Blue staining. SEM/EDX confirmed calcium-phosphate deposition. A phenomenological finite element model was developed to correlate mechanical stimuli with hypertrophy using the osteogenic index.ResultsMechanical stimulation led to a 32 % increase in Alizarin Red staining compared to controls (P < 0.001), indicating faster mineralization. GAG production was reduced under mechanical loading (P < 0.05), consistent with early mineralization. SEM/EDX confirmed more uniform mineral deposition in stimulated samples. Morphologically, stimulated cells aligned along the loading axis and displayed a nearly 10-fold increase in hypertrophic cell area. The numerical model showed elevated osteogenic index values and stress peaks in the chondrocyte membrane under mechanical loading.ConclusionsDynamic tensile stimulation in 2D culture significantly accelerates ECM mineralization in ATDC5 cells. This effect appears to be mediated through enhanced chondrocyte hypertrophy and alignment of ECM components. The combined use of mechanical and biochemical cues provides a promising strategy for optimizing in vitro models of endochondral ossification and developing tissue engineering therapies