HAL Portal UTC Université de Technologie de Compiègne
Not a member yet
11652 research outputs found
Sort by
Uncertainty-Aware Online Extrinsic Calibration: A Conformal Prediction Approach
No full textInternational audienceAccurate sensor calibration is crucial for autonomous systems, yet its uncertainty quantification remains underexplored. We present the first approach to integrate uncertainty awareness into online extrinsic calibration, combining Monte Carlo Dropout with Conformal Prediction to generate prediction intervals with a guaranteed level of coverage. Our method proposes a framework to enhance existing calibration models with uncertainty quantification, compatible with various network architectures. Validated on KITTI (RGB Camera-LiDAR) and DSEC (Event Camera-LiDAR) datasets, we demonstrate effectiveness across different visual sensor types, measuring performance with adapted metrics to evaluate the efficiency and reliability of the intervals. By providing calibration parameters with quantifiable confidence measures, we offer insights into the reliability of calibration estimates, which can greatly improve the robustness of sensor fusion in dynamic environments and usefully serve the Computer Vision community
Contextual-Graph Topology for Efficient and Real-Time Cooperative Intersection Management
No full textInternational audienc
Micro-robot numérique à multiples degrés de liberté avec lévitation diamagnétique et pilotage par moyen optique
No full textThe work presented in this thesis concerns the development of a fully levitated, wireless, and contactless micro-positioning platform based on diamagnetic levitation and optical actuation. The actuator consists of a Pyrolytic Graphite Sheet (PGS) levitating above a Permanent Magnet Array (PMA), with motion induced by localized laser heating achieving actuation without mechanical contact, embedded electronics, or feedback control. Following a bibliographical study on planar and levitating actuators, diamagnetic actuation was selected for its passive stability and energy efficiency. The system’s originality lies in its ability to achieve both discrete and continuous motion across multiple Degrees of Freedom (DOF), reaching up to 6-DOF in its most advanced configuration. A first prototype was modeled and experimentally validated, demonstrating reliable planar actuation with 5 mm stroke (PM pitch), 1.4 µm repeatability, transport of up to 29 mg (5× the PGS mass), 17.48 mm/J energy efficiency, and velocities up to 7 mm/s. A Micro-Electromechanical System (MEMS) mirror was integrated for laser beam steering, enabling simultaneous wireless actuation of multiple PGSs, and achieving better velocities and energy efficiencies. Pathfinding algorithms such as A* and decentralized Multi-Agent Pathfinding (MAPF) were implemented for collision-free motion planning of multiple PGS samples. A multi-resolution platform was then introduced, composed of three PMAs with different pitches (3 mm, 6 mm, and 12 mm), allowing both fast and precise motion within a single workspace. Rotational motion above the smallest PMA enabled 2.5-DOF actuation, and a complete trajectory across the platform was validated experimentally. Finally, advanced actuation was demonstrated using two microstruc-tured Parallel Platforms (PPs) fabricated via two-photon polymerization enabling vertical displacement and fully decoupled 6-DOF motion. Their performances were experimentally characterized and compared. This work establishes a foundation for scalable, reconfigurable, and sensor-free diamagnetic ac-tuation systems with applications in micromanipulation, lab-on-chip devices, and precision robotic.Ce travail de thèse porte sur le développement d’une plateforme de micro-positionnement entièrement lévitée, sans fil et sans contact, basée sur la lévitation diamagnétique et l’actionnement optique. L’actionneur se compose d’une feuille de graphite pyrolytique lévitant au-dessus d’un réseau d’aimants permanents, avec un mouvement induit par un chauffage laser localisé permettant un actionnement sans contact mécanique, sans électronique embarquée ni contrôle en boucle fermée. A la suite d’une étude bibliographique des actionneurs planaires lévitants et non lévitants, l’actionnement diamagnétique a été retenu pour sa stabilité passive et son efficacité énergétique. L’originalité du système réside dans sa capacite à réaliser des mouvements à la fois discrets et continus dans plusieurs degrés de liberté, atteignant jusqu’à six degrés de liberté dans sa configuration la plus avancée. Un premier prototype a été modélisé et validé expérimentalement, démontrant un actionnement plan fiable avec une course de 5 mm (correspondant au pas des aimants), une répétabilité de 1,4 µm, le transport de charges allant jusqu’à 29 mg (cinq fois la masse de la feuille de graphite pyrolytique), une efficacité énergétique de 17,48 mm/J, et des vitesses atteignant 7 mm/s. Un miroir à microsystème électromécanique a été intégré pour diriger le faisceau laser, permettant l’actionnement sans fil simultané de plusieurs feuilles de graphite pyrolytique, avec des performances améliorées en termes de vitesse et d’efficacité énergétique. Des algorithmes de planification de trajectoires, tels que A* et la planification multi-agent décentralisée, ont été mis en œuvre pour assurer un déplacement sans collision de plusieurs échantillons de feuilles de graphite pyrolytique. Une plateforme multi-résolution a ensuite été introduite, composée de trois réseaux d’aimants permanents de pas différents (3 mm, 6 mm et 12 mm), permettant un mouvement à la fois rapide et précis dans un même espace de travail. Un mouvement rotationnel au-dessus du plus petit réseau a permis de réaliser un actionnement à 2,5 degrés de liberté, et une trajectoire complète sur la plateforme a été validée expérimentalement. Enfin, un actionnement avancé a été démontré à l’aide de deux plate-formes parallèles microstructurées fabriquées par polymérisation à deux photons, permettant un déplacement vertical et un mouvement entièrement découplé selon six degrés de liberté. Leurs performances ont été caractérisées et comparées expérimentalement. Ce travail établit les bases de systèmes d’actionnement diamagnétiques évolutifs, reconfigurables et sans capteurs, avec des applications en micromanipulation, dispositifs de type lab-on-chip, et robotique de précision
Leveraging ensemble deep models and llm for visual polysemy and word sense disambiguation
No full textInternational audienceVisual Polysemy Disambiguation (VPD) and Visual Word Sense Disambiguation (VWSD) are challenging tasks for both computer vision and NLP since an image can have diverse contextual interpretations, ranging from visual representations to abstract concepts. In this paper, we propose a novel approach to address the challenges of VPD and VWSD by leveraging ensemble deep models from computer vision to alleviate the problem of VPD and using the strength of LLMs to mitigate the problem of WSD. We first generate visually representative images from textual descriptions through a zero-shot text-to-image generation framework using image scrapping and Google search. We then employ an ensemble of classifiers and a deep network to learn feature representations, classify images into contexts and find the best match. Similarly, we perform the reverse process of generating textual descriptions from images using a Vision Transformer model and calculate the cosine distance with the actual text. Experimental evaluation of benchmark datasets demonstrates the effectiveness of our combined approach in strengthening both text-to-image and image-to-text generation, we improve disambiguation accuracy, providing a robust solution for VWSD with an MRR of 95.77% and a Hit rate of 92.00% surpassing state-of-the-art methods.</div
A survey on TRIX index and machine learning applications in marine pollution context
No full textInternational audienc
A Trust Management Architecture for Energy Efficient Mobile IoT
No full textInternational audienceIn mobile IoT applications, managing connectivity, security, and data exchange presents significant challenges. Mobility introduces critical issues such as seamless handovers between networks, energy efficiency in battery-powered devices, and the maintenance of stable connections despite fluctuating network conditions. Additionally, ensuring the security and privacy of IoT devices becomes even more complex when they are in motion, as they may encounter untrusted networks, unauthorized access attempts, or data interception threats. While IoT offers vast opportunities, mobility introduces an additional layer of complexity. It requires robust network management, strengthened security protocols, and optimized data handling strategies to ensure seamless, reliable, and uninterrupted connectivity A fundamental question arises: when a node moves from one network to another, how can the new network assess its trustworthiness and seamlessly integrate it? In this paper, we propose a novel solution to address this challenge by leveraging clustering, software-defined networks (SDN), and machine learning algorithms. Our approach ensures that mobile IoT nodes are reliably assessed and securely integrated into new networks, all while maintaining seamless connectivity energy efficiency and robust security
Adaptive and Context-Aware Defenses Against Interest Flooding Attacks in CCN-based IoT
No full textInternational audienceThe Internet of Things (IoT), particularly in constrained environments like Wireless Sensor Networks (WSNs), demands scalable and secure networking. Content-Centric Networking (CCN) addresses key IoT challenges through content-based communication, enabling caching, request aggregation, and resilience. However, CCN is vulnerable to Interest Flooding Attacks (IFA), which overload the Pending Interest Table (PIT) with excessive Interest packets. IfNot, a lightweight mitigation approach, adjusts forwarding based on PIT timeouts but suffers from high false positives and limited resilience to adaptive attackers. To improve this, we propose IfNot-R, which penalizes malicious behavior using historical data, and IfNot-FPGuard, which reduces false positives via longterm Interest satisfaction tracking. These enhancements strengthen IFA defense in dynamic IoT contexts
Comparative Study of Conventional and Emerging Extraction Methods for Papaya Seed Oil: Process Optimization and Quality Evaluation
No full textInternational audiencePapaya seed oil ( Carica papaya ) is gaining interest for its potential applications as a functional food ingredient. Papaya seeds, usually discarded as agro-industrial waste, represent a valuable source of oil rich in unsaturated fatty acids. This study evaluated several extraction methods, including Soxhlet extraction, maceration, microwave-assisted extraction (MAE), hydraulic pressing, and ultrasound-assisted extraction (UAE), using pure ethanol and pure isopropanol as green solvents. Soxhlet extraction with hexane for 24 h at 121 °C achieved the highest yield (30.91 ± 0.49%) and was used as the reference. Among the greener techniques, UAE performed at 400 W for 30 min with isopropanol achieved the best compromise between yield (22.50 ± 0.27%) and oil quality, showing the lowest acidity index (9.00 ± 2.16 mg KOH/g). Therefore, UAE was selected for further optimization by varying ultrasound power (0–400 W), extraction time (5–30 min), and solvent composition (isopropanol, 50:50 ethanol–isopropanol mixture, and ethanol). The optimal extraction conditions were 200 W for 30 min using a 1:1 ethanol-isopropanol solvent mixture, which resulted in an extraction yield of 26.34%, corresponding to an oil recovery of approximately 85% relative to the Soxhlet reference. The mixed solvent enhanced extraction efficiency by improving mass transfer through the combined effects of solvent polarity and ultrasonic cavitation, confirming that ultrasound-assisted extraction with green solvent mixtures is an effective method for papaya seed oil recovery