Archive ouverte de l'ENAC
Not a member yet
3458 research outputs found
Sort by
Covering a rectangle with 6 circles: a reliable mathematical programming approach
No full textInternational audienceWe present a rigorous global optimization-based approach to a problem arising in discrete geometry: covering a rectangle with six identical circles while minimizing their radius. Our main contribution lies in formulating and solving this problem using mathematical programming combined with exact global optimization relying on interval-based computation. This approach not only enables the numerical proof of a theoretical result, but also certifies the accuracy of computed values by enclosing them within verified bounds, thus certifying decimal digits. This brings a strong evidence of the role of mathematical programming and rigorous exact global optimization in addressing geometric covering problems with provable precision
Bias Influence on AI Accuracy: The Case of Air Traffic Controllers’ Experience
No full textInternational audienceThis study examines the effect of a biased dataset in the Artificial Intelligence (AI) learning process. Biases are almost inevitable and history has shown that it may significantly influence AI models, leading potentially to erroneous results and even dangerous consequences. We propose, through the case of air traffic control, an evaluation of the influence of bias in AI model training and performance. Air traffic control requires years of operational experience to master the environment and develop effective conflict resolution strategies. The level of experience is known to influence significantly the aircraft collision avoidance strategy. Our study is based upon a dataset collected with controllers of various levels of experience, leading to the creation of a bias in their avoidance strategy, the validation of this bias through statistics and the AI model development and simulation. Our approach compares two iterations of the model; one without any action to handle the bias and one by integrating it as a feature. The main findings show that declaring the bias as a feature does not necessarily impact AI model learning and overall accuracy, but can clearly influence the results over specific classes. The results of the second iteration contributing to higher alignment with specific preferences of the experienced controllers, versus the novice ones
Star-Burst paradigm: implementation of an “invisible” dry-EEG reactive BCI
No full textInternational audienceCode Visually Evoked Potentials (c-VEP) have become increasingly popular in the rBCI community, leveraging pseudo-random visual flickers that offer shorter calibration times compared to Steady State VEP [1]. However, the application of c-VEP-based reactive BCIs has largely remained confined to laboratory settings due to the reliance on wet EEG systems andsynchronous paradigms with fixed decoding times. To address these challenges, our team used innovative repetitive visual stimuli called StAR (Stimuli for Augmented Response). These stimuli are engineered with specific, mostly invisible textures that elicit neural responses ranging from retinal ganglion cells (contrast detection) to visual cortex cells (orientation selectivity) [2]. Our StAR stimuli are activated using a burst-code VEP paradigm, featuring brief, aperiodic visual flashes presented at aslower rate of three flashes per second. This approach elicits stronger visual evoked responses compared to traditional maximum length sequences [3]. Each stimulus (e.g., a letter or digit) is presented with its own unique pseudo-random code comprising an alternating sequence of '1' (on) and '0' (off). This innovative approach dramatically reduces calibration time to under one minute, as the algorithms only need to differentiate brain responses to the presence (visual ERP) or absence of aflash (no visual ERPs).Material, Methods and Results: The online StAR-Burst rBCI was developed using Timeflux framework [4]. This system was specifically designed for an 11-class classification task to predict participants' attention in real time based on visual stimuli. It utilizes a combination of XDawn spatial filtering and Riemannian-based tangent space classifiers for optimal performance. The 11 commands corresponding to the T9 keypad were encoded using 11 unique burst codes [3], carefully designed to maximize discrimination between commands. The classification pipeline was followed by a correlation-based accumulation method, allowing flexible, self-paced decoding time. Eighteen participants were equipped with an 8-channel dry EEG system (Enobio), with electrodes placed over the occipital and parieto-occipital cortex areas (PO7, O1, Oz, O2, PO8, PO3, POz, PO4) to capture visually evoked potentials (VEPs). They underwent an 40-second calibration procedure beforeperforming an online T9 pinpad self-paced task consisting of 10 sequences, each containing four targets, resulting in a total of 40 targets per participant. The StAR-Burst rBCI demonstrated high performance, achieving a mean accuracy of 96.3% (SD = 4.79) and a mean decoding time of 4.2 seconds (SD = 5.5). A video showcasing the BCI can be seen here https://nextcloud.isae.fr/index.php/s/dxLqYXRAMEep98C.Conclusion: Collectively, these findings highlight the transformative potential of StAR-Burst paradigm driving the evolution to make BCIs more user friendly and efficient. Our implementation achieved high accuracy levels with a dry EEG system, requiring only minimal calibration data (40s). This paradigm, characterized by comfort and subtle perceptibility in peripheral vision, show potential for applications in various reactive BCI paradigms such as P300 speller, SSVEP, and oddball-basedBCI. The application of the proposed StAR approach may be extended beyond technological innovation to fundamental cognitive neuroscience research, providing a valuable avenue for exploring cognition.Acknowledgments and Disclosures: This work is funded by Agence Innovation Defense (VIPER)References:[1] Martínez-Cagigal, V., Thielen, J., Santamaria-Vazquez, E., Pérez-Velasco, S., Desain, P., & Hornero, R. (2021). Brain–computer interfaces based on code-modulated visual evoked potentials (c-VEP): a literature review. Journal of Neural Engineering, 18(6), 061002.[2] Dehais, F., Cabrera Castillos, K., Ladouce, S., & Clisson, P. (2024). Leveraging textured flickers: a leap toward practical, visually comfortable, and high-performance dry EEG code-VEP BCI. Journal of neural engineering, 21(6), 10.1088/1741-2552/ad8ef7[3] Castillos, K. C., Ladouce, S., Darmet, L., & Dehais, F. (2023). Burst c-VEP based BCI: optimizing stimulus design for enhanced classification with minimal calibration data and improved user experience. NeuroImage, 284, 120446.[4] P Clisson, R Bertrand-Lalo, Marco Congedo, G Victor-Thomas, J Chatel-Goldman. Timeflux: an open-source framework for the acquisition and near real-time processing of signal streams. BCI 2019 - 8th International Brain-Computer Interface Conferenc
Assessing and Visualizing Pilot Performance in Traffic Patterns: A Composite Score Approach
No full textInternational audienceObjective measurement of pilot performance has long been a research challenge.This study introduces a new composite score that combines various flight metrics, alongwith its visual representation through an online application. Thirty general aviation pilotscompleted flight simulator scenarios under different Flight Rules (VFR: Visual Flight Rulesvs. IFR: Instrument Flight Rules) and difficulty levels (Low vs. High). Workload wasassessed using subjective and objective indicators. The composite score was developedusing flight parameter compliance, approach stability, and landing quality. Workloadindicators confirmed the scenario difficulties, showing significant increases under IFRcompared to VFR and in High vs. Low difficulty conditions. As predicted by multipleresources theory, the composite score correlated negatively with workload, particularly inIFR conditions, demonstrating its effectiveness in assessing pilot performance. In a followupquestionnaire, pilots rated the online application positively, highlighting its usefulnessin understanding their performance and recognizing its potential for pilot training
Design of low-noise arrival and departure procedures in Paris TMA using Simulated Annealing
No full textAir traffic continues to grow and there is a risk of congestion in the near future. The areas most affected by this growth are the terminal manoeuvring areas (TMAs), which are the bottlenecks in the airspace. In order to mitigate this phenomenon, it is necessary to design optimised departure/arrival routes to smooth the air traffic flow and reduce controller workload. This study addresses the problem of designing Standard Instrument Departure and Standard Terminal Arrival Routes (SID and STAR) taking into account the noise impact. The proposed algorithm is based on the Simulated Annealing method and Dubins curves. The method is tested at Paris Charles-De-Gaulle (CDG) airport and compared with previous work. Then, a noise impact study is carried out to propose short and not very noisy routes. Finally, Orly airport is added to the CDG scenario and the method is tested again. The results obtained seem satisfactory in view of the current state of air traffic management and previous research
Partitionnement de Graphe pour l'Identification de Goulots d'Étranglement Partagés
No full textInternational audienceUn réseau IP peut être représenté sous la forme d'un graphe où les noeuds sont des routeurs et les arêtes des liens de communication IP. Ce graphe aide à analyser les interactions et les flux d'information au sein du réseau. Chaque routeur, agissant comme une file d'attente, gère le trafic avec une capacité de mémoire tampon et un débit de sortie. Lorsque le trafic entrant dépasse cette capacité, une congestion se produit, dégradant le service. Identifier ces goulots d'étranglement est crucial pour évaluer la performance du réseau. Cet article explore une méthode de partitionnement de graphe permettant de regrouper les flux partageant un goulot commun à l'aide d'un modèle probabiliste plus général que ceux de la littérature
Exploring Forest Vertical Structure With TomoSense: GEDI and SAR Tomography Insights
No full textInternational audienceExploring vertical forest structures worldwide via remote sensing faces challenges. Recent technologies like waveform light detection and ranging (LiDAR) from NASA’s global ecosystem dynamics investigation (GEDI) and SAR tomography (TomoSAR) from future European Space Agency (ESA) BIOMASS offer promising solutions. This article assesses the performance of spaceborne GEDI and TomoSAR airborne data from an ESA’s TomoSense campaign to highlight the important role of GEDI measurements in BIOMASS algorithm training and establishing precise site-specific processing parameters. Our study in Germany’s Eifel National Park delves into the precision of GEDI and P-band TomoSAR in measuring surface [digital terrain model (DTM)] and vegetation [canopy height model (CHM)] heights. Results demonstrate that GEDI and P-band TomoSAR offer high-resolution and precise surface and vegetation heights and vertical profile measurements. While GEDI relative height (RH) at 98% (RH98) was previously recommended for tropical forests, our findings advocate for RH85 as the optimal metric for temperate forests. The research supports improving the accuracy of both DTM and CHM utilizing GEDI beams with full-power lasers coupled with high sensitivity and signal-to-noise ratio (SNR). Ground elevation measurements are more accurate than canopy height estimates for temperate forests, with DTM RMSE about 2 m and CHM RMSE about 3 m for GEDI and TomoSAR measurements. By analyzing the vertical structure of monthly GEDI data, we note a 1-m shift in the volume peak between GEDI’s leaf-on and leaf-off periods. At the same time, TomoSAR consistently exhibits a lower volume peak by about 2 m compared to GEDI during leaf-on seasons. In conclusion, our research underscores the complementary roles of TomoSAR and GEDI in accurately mapping diverse forest types, thereby bolstering the effectiveness of the BIOMASS mission
Approche robuste pour la modélisation dans le domaine temporel de la propagation acoustique en milieu poreux
No full textMatériaux poreux et métamatériaux acoustiques; GAPSUS - Acoustique Physique, Sous-Marine et Ultra-Sonore: GVB - Vibro acoustique et Contrôle du Bruit: GABE - Acoustique du Bâtiment et de l'EnvironnementNational audienceLes matériaux poreux sont utilisés en acoustique pour réduire le bruit en dissipant l'énergie des ondes sonores. En aéronautique, ils sont soumis à des signaux acoustiques de forte intensité, transitoires et de large bande fréquentielle, ce qui impose leur modélisation dans le domaine temporel. Ces milieux peuvent être décrits comme des fluides équivalents où les phénomènes de dissipation visqueuse et thermique sont respectivement représentés par les tortuosité et compressibilité dynamiques. Ces propriétés sont généralement modélisées dans le domaine fréquentiel, mais leur transposition en formulations temporelles adaptées aux simulations numériques acoustiques et aéroacoustiques reste un défi. Une démarche classique, l'approximation multipôle, consiste à représenter ces comportements par une somme de fractions rationnelles, analogues à des filtres passe-bas dont les fréquences de coupures sont directement reliées aux pôles. Ces fractions peuvent ensuite être directement converties dans le domaine temporel. Cependant, les méthodes actuelles de détermination des coefficients de ces sommes, comme le vector-fitting, présentent des limitations. Elles posent notamment des problèmes liés à la passivité du modèle et à l'apparition de pôles hors de la bande de fréquences d'intérêt. Nous proposons une méthode alternative qui garantit la passivité du modèle approché. Cette méthode optimise les pôles dans une bande de fréquences cible, grâce à une approche d'optimisation en deux étapes. Une première étape de moindres carrés linéaires sous contrainte de réalité sur les coefficients initialise la seconde étape d'optimisation non-linéaire avec contrainte de positivité. En exploitant les asymptotes basses et hautes fréquences de la fonction à approcher, connues analytiquement, cette méthode assure également l'extension stable de l'approximation au-delà de la bande fréquentielle d'intérêt. La validité de la modélisation obtenue avec cette méthode est démontrée par des simulations temporelles et des mesures expérimentales en tube à impédance sur divers échantillons acoustiques de complexités variées, incluant des matériaux à deux échelles de porosité
Resilient conflict detection and resolution for high-uncertainty constrained urban airspace operations
No full textInternational audienceThe concept of urban air mobility is rapidly advancing, with much research being dedicated towards the development of the air traffic management services required for such operations. An important component of unmanned air traffic management (U-space/UTM) is conflict detection and resolution (CD&R), tasked with ensuring the operational safety of such systems. Strategic flight plan optimisation and tactical CD&R methods have generally been studied independently, leading to suboptimal performance when deployed simultaneously in simulated high-density very-low-level constrained urban airspace environments. Furthermore, the limited flexibility of pre-departure 4D trajectory planning methods towards dynamic and uncertain environmental and operational conditions (i.e., wind and delay) produces a degradation in safety that is difficult to mitigate using tactical manoeuvring. In this work, we design a traffic-flow capacity strategic optimisation method that aims to achieve robustness against flight plan deviations and to better complement tactical CD&R manoeuvring. The performance of the proposed strategic and tactical deconfliction module is tested within constrained urban airspace traffic scenarios simulated using the BlueSky Open Air Traffic Simulator. The results are compared with other methods, such as 4D trajectory planning and state-based CD&R.</div
Navigating Partial Automation in Firefighting with Drones: Trust, Take-Over, and Human-Drone Teaming
No full textInternational audienceIn the past years, the use of drones has been increasingly introduced to firefighting operations. Drawing on concrete examples from interviews with Thai firefighting professionals and recent field trials, as well as prior research, this paper examines the challenges of integrating (partially) autonomous, AI-enhanced drones into firefighting operations. Our findings reveal that, despite the promise of automation, on-field operators still prefer communication via a dedicated drone pilot-a preference driven by unresolved trust issues and concerns over information overload. We discuss challenges such as trust in automation and adaptive take-over. These inform our proposals for design recommendations on adaptive communication, transparent take-over mechanisms, trust calibration, physical handover and mapping of multiple data sources in human-drone teaming for firefighting