Archive ouverte HAL-LAAS
Not a member yet
12189 research outputs found
Sort by
Contact-Robust Trajectory Planning via Parametric Sensitivity Analysis for Hybrid Robotic Systems
No full textInternational audienceIn this paper, we combine first-order approximations of hybrid systems (i.e., the so-called saltation matrix) with previous works on parametric sensitivity for continuous systems to propose a general framework for robust trajectory optimization of hybrid systems subject to parametric uncertainties. A method for computing parametric sensitivities of both continuous dynamics and hybrid events is presented. The obtained "hybrid parametric sensitivity" is then combined with sensitivity-based tubes that encapsulate all possible perturbed states and control trajectories given a known bounded range for the uncertain parameters. The proposed method is then applied to the problem of planning robust trajectories for legged robot systems, which allows obtaining trajectories that remain feasible w.r.t. the contact constraints even in presence of uncertainties in the dynamics, guard conditions, and reset maps. We also illustrate one of the fundamental limitations of first-order approximations, that is, the fact that the sensitivity reset time is fixed, and propose an extension to the sensitivity analysis that can form the basis for future developments
Smart sampling for optimized materials: A Gaussian process framework for region-of-interest active learning
No full textInternational audienceDiscovering materials with properties that meet user-defined performances remains a critical yet resource-intensive challenge in materials engineering, often requiring costly experiments or simulations. This paper introduces a data-efficient Gaussian Process (GP) active learning framework designed to map multidimensional design spaces and efficiently identify subsets of the output space that represent the optimal or acceptable range for a specific application, defined as Regions of Interest (RoIs). Three novel acquisition function families were proposed: (i) Probability-Entropy acquisitions, balancing exploration and exploitation without additional tuning; (ii) Dual Level-Set Gaussian Bound, extending classical straddle heuristics to multiple thresholds; and (iii) Expected Mahalanobis (ExM), generalizing boundary-focused sampling to arbitrary multivariate target distributions. Benchmark evaluations on synthetic functions and a Al/CuO thermite powder mixture design case demonstrate that the ExM acquisition significantly outperforms existing methods, achieving faster RoI discovery, lower uncertainty, and more complete mapping with minimal iterations. We found that ExM removes the need to tune an explicit exploration parameter. It only requires choosing a target distribution within the RoI, for which we provide simple default options (Uniform or truncated Normal) along with practical guidelines. Applied to Al/CuO thermite formulation for welding applications, ExM efficiently identifies diverse optimized compositions within the target RoI, minimizing redundancy and experimental cost. Overall, ExM provides a scalable and flexible RoI-targeted active learning approach that removes exploration-weight tuning for materials design and beyond
Low-latency online estimation of human upper-limb pose and kinematics from a single 360 camera
No full textInternational audienceWe present a fully online framework for streaminghuman upper-limb kinematics estimation from a single 360camera. Incoming frames are processed sequentially throughvertical-boundary-aware tracking, pseudo-perspective rendering,and Neural Localizer Fields to estimate a sparse set of 3Danatomical landmarks in real time. These landmarks are mappedto an OpenSim-compatible biomechanical model, with jointangles computed on the fly via an online inverse kinematicssolver. The system achieves end-to-end latencies as low as22.9 ms on a high-performance setup. Evaluated in a single-participant scenario involving an initial T-pose calibration andrepeated object displacement toward the camera, it demonstratesrobust performance under moderate self-occlusion and sphericaldistortion. While tested in a constrained setting, its modular, real-time design makes it a promising candidate for human–robotinteraction and other motion analysis applications, enablingminimal, markerless, and anatomically interpretable upper-limbtracking from omnidirectional vision
On the Random Schrödinger Equation and Geometric Quantum Control
No full textWe introduce the random Schrödinger equation, with a noise term given by a random Hermitian matrix as a means to model noisy quantum systems. We derive bounds on the error of the synthesised unitary in terms of bounds on the norm of the noise, and show that for certain noise processes these bounds are tight. We then show that in certain situations, minimising the error is equivalent to finding a geodesic on SU (n) with respect to a Riemannian metric encoding the coupling between the control pulse and the noise process. Our work thus extends the series of seminal papers by Nielsen et al. on the geometry of quantum gate complexity
Mie resonant silicon particles via bottom-up synthetic routes and assembled into 2d metasurfaces
No full textInternational audienceMie resonant silicon particles have strong scattering and thus are highly desirable building-blocks in metamaterials and metasurfaces. Most silicon resonators are produced via top-down fabrication methods.We presenttwo bottom-up syntheses followed by particle self-assembly, producing a silicon-based metasurface having a degree of disorder, and yet a high quality factor
Approches dynamiques de gestion du cache mémoire partagé pour un système multi-critique sur processeur multi-coeur
No full textNational audienceCache memory management in multi-core processors is a crucial element for the determinism and performance of a real-time critical system. In multicritical systems, non-critical tasks can affect critical tasks by increasing execution times, particularly through shared cache memory. Indeed, if their data is evicted from the cache, critical tasks must reload it from main memory, which significantly increases data access time. As a result, ensuring execution determinism, i.e., enforcing upper bounds on the execution times of critical tasks, leads to hardware overprovisioning. To limit evictions, several strategies exist. These strategies aim to prevent cache interactions by avoiding concurrent access or by partitioning the cache into dedicated spaces for different tasks. However, these strategies reduce the overall performance of the system. Current approaches systematically or even permanently lock cache partitions, leading to cache over-reservation. Our approach explores new last-level cache management strategies to guarantee real-time deadlines for critical tasks while optimizing the performance of non-critical tasks. The idea is to reserve only the necessary amount of cache for the proper execution of critical tasks and release it as soon as possible. These new approaches rely on dynamic cache allocation mechanisms. Our first contribution is to demonstrate the relevance and feasibility of dynamic cache management mechanisms to improve the performance of non-critical tasks while keeping worst-case execution times of critical tasks under control. Specifically, we conducted experiments on processors equipped with Intel CAT technology. Our second contribution is the development of a dynamic approach based on a precise estimation of the cache required for each occurrence of critical tasks, allowing us to minimize unnecessary reservations. The main challenge is thus to determine the appropriate cache allocation to ensure real-time constraints while maximizing the available space for non-critical tasks. We modeled the scheduling problem of a set of critical tasks while integrating the impact of allocated cache for each occurrence of a task, i.e., each job. The more cache a job has, the shorter its worst-case execution time will be. We propose two dynamic allocation strategies: one per task, where each job receives the same cache allocation, and one per job, where the cache allocation can vary from one job to another. These models are implemented using the extit{Minizinc} modeling language, allowing the use of solvers to optimize the allocation configuration for each job. These models can be used by any real-time engineer seeking to properly size their system and ensure real-time requirements for a task set. To assess the benefits and limitations of our approaches, we conducted a comparative study by generating task sets in a pseudo-random manner. In these case studies, we compare a static allocation strategy (similar to the state of the art) with our dynamic allocation strategies, representing a scientific advancement in this field. Our results show that the dynamic per-job allocation strategy outperforms other strategies at the cost of increased complexity in finding the optimal allocation configuration. From an engineering perspective, the choice of the most suitable strategy strongly depends on the task set to be integrated into the system, particularly on the sensitivity of tasks to cache interference.La gestion du cache mémoire dans les processeurs multi-cœurs est un élément crucial pour le déterminisme et les performances d'un système temps réel critique. Dans le cas des systèmes multicritiques, des tâches non critiques peuvent influer sur les tâches critiques en rallongeant les temps d'exécution, notamment au travers du cache mémoire partagé. En cas d'éviction de leurs données du cache, les tâches critiques doivent les recharger depuis la mémoire principale, ce qui allonge considérablement le temps d'accès aux données. De ce fait, garantir le déterminisme d'exécution, i.e. le respect de bornes maximales des temps d'exécution des tâches critiques, mène au surdimensionnement du matériel. Pour limiter les évictions, des stratégies existent. Ces stratégies cherchent à empêcher les interactions dans le cache en évitant les accès concurrents ou en partitionnant le cache en espaces dédiés aux différentes tâches. Cependant, ces stratégies diminuent les performances globales du système. Les approches actuelles verrouillent le cache de manière systématique, voire permanente, ce qui sur-réserve le cache mémoire. Notre approche consiste à investiguer de nouvelles stratégies de gestion du dernier niveau de cache afin de garantir les échéances temps réel des tâches critiques tout en optimisant les performances des tâches non critiques. L'idée est de ne réserver que la quantité de cache nécessaire au bon fonctionnement des tâches critiques et de la libérer dès que possible. Ces nouvelles approches utilisent des mécanismes d'allocation de cache mémoire dynamique. Notre première contribution est de démontrer l'intérêt et la faisabilité des mécanismes de gestion dynamique de cache pour améliorer la performance des tâches non critiques, tout en contrôlant les pires temps d'exécution des tâches critiques. Nous avons notamment réalisé des expérimentations sur des processeurs équipés de la technologie Intel CAT. Notre seconde contribution est le développement d'une approche dynamique fondée sur un dimensionnement précis de la quantité de cache nécessaire pour chaque occurrence des tâches critiques, ce qui nous permet de limiter au maximum les réservations inutiles. Le principal défi devient donc le dimensionnement des quantités de cache à allouer pour garantir les contraintes temps réel tout en optimisant l'espace disponible pour les tâches non critiques. Nous avons modélisé le problème d'ordonnancement d'un jeu de tâches critiques en intégrant l'impact de la quantité de cache allouée à chaque occurrence des tâches, c'est-à-dire de chaque job. Plus un job dispose de cache, plus son temps d'exécution pire cas sera réduit. Nous proposons deux stratégies d'allocation dynamique : l'une par tâche, où chaque job dispose de la même quantité de cache, et l'autre par job, où la quantité de cache peut varier d'un job à l'autre. Ces modèles sont écrits avec le langage de modélisation extit{Minizinc}, ce qui permet d'utiliser un solveur pour optimiser la configuration de l'allocation pour chaque job. Ces modèles sont utilisables par n'importe quel ingénieur temps réel qui souhaite dimensionner son système et garantir les exigences temps réel d'un jeu de tâches. Pour évaluer les gains et les limites de nos approches, nous avons réalisé une comparaison en générant des jeux de tâches de manière pseudo-aléatoire. Sur ces cas d'école, nous comparons une stratégie d'allocation statique (similaire à l'état de l'art) avec nos stratégies d'allocation dynamique, ce qui constitue une avancée scientifique sur cette problématique. Nos résultats montrent que la stratégie d'allocation dynamique par job domine les autres stratégies, au prix d'une plus grande complexité dans la recherche de la configuration optimale de l'allocation. D'un point de vue ingénierie, le choix de la stratégie la plus adaptée dépend fortement du jeu de tâches à intégrer au système, et en particulier de la sensibilité des tâches aux interférences du cache
Towards a CuSEF (CubeSat Systems Engineering Framework)
No full textCubeSat technology has lowered the barrier to entry into space, enabling small organizations and universities to rapidly design and launch miniature satellites. This trend has also created a market in which companies are jostling to sell these systems and the services they provide.Cubesats projects are very often carried out in very specific contexts with small teams and limited budgets. What's more, in the face of stiff competition, industry players are trying to compress development time to keep costs down and deliver the product as quickly as possible. This leads to a very high mission failure rate.A key factor of this high failure rate is the lack of a rigorous yet lightweight systems engineering process adapted to these small, fast-paced projects. Traditional standards for systems engineering (such as the ISO/IEC/IEEE 15288 lifecycle standard or space agency process standards like ECSS) are often not applied on CubeSat projects, as they can be overly complex, resource-intensive, or ill-suited to the constraints of CubeSat-class missions. NewSpace teams (small start-ups or academic groups with limited budgets and personnel) typically operate in a “garage development” style – ad hoc design-build efforts using off-the-shelf parts – which can result in missed engineering steps and inadequate testing.This paper introduces the CubeSat Systems Engineering Framework (CuSEF), tailored for NewSpace, that bridges the gap between heavyweight standards and the practical realities of CubeSat development. CuSEF combines a simplified process model with modern methods and tools to provide structure without undue burden. By adapting the ISO 29110 standard for very small entities and integrating essential requirements from the ECSS space engineering guidelines, the framework aims to enhance the reliability of CubeSat missions while optimizing development cost and time. The following sections review the state of the art, detail the CuSEF framework (its process, methods, and toolset), discuss expected benefits, and conclude with future perspectives
Progress on IEC 63187: System Safety for Complex Systems in Defence Programmes
No full textInternational audienceDefence programmes deliver capability through systems that are often highly complex, combining system elements with diverse, often cutting-edge technologies, involving multiple stakeholders and complex supply chains, and operating in environments with dynamically changing risk. As well as their inherent hazards, such complex systems have the potential for emergent system interactions to cause unexpected hazards. To address the concern that such complex systems are not sufficiently addressed by current safety assurance standards, the International Electrotechnical Commission is developing a new international standard, IEC 63187. The authors have previously introduced the goals and systems engineering approach of IEC 63187. This position paper gives an update on the progress of the draft standard towards publication and an overview of key areas of development
Moment-SOS hierarchies for arrow-type polynomial matrix inequalities with applications to structural optimization
No full textThe Arrow Decomposition (AD) technique, initially introduced in [Mathematical Programming 190(1-2) (2021), pp 105-134], demonstrated superior scalability over the classical chordal decomposition in the context of Linear Matrix Inequalities (LMIs) if the matrix in question satisfied suitable assumptions. The primary objective of this paper is to extend the AD method to address Polynomial Optimization Problems (POPs) involving large-scale Polynomial Matrix Inequalities (PMIs), with the solution framework relying on moment-sum of square (mSOS) hierarchies. As a first step, we revisit the LMI case and weaken the conditions necessary for the key AD theorem presented in [Mathematical Programming 190(1-2) (2021), pp 105-134]. This modification allows the method to be applied to a broader range of problems. Next, we propose a practical procedure that reduces the number of additional variables, drawing on physical interpretations often found in structural optimization applications. For the PMI case, we explore two distinct approaches to combine the AD technique with mSOS hierarchies. One approach involves applying AD to the original POP before implementing the mSOS relaxation. The other approach applies AD directly to the mSOS relaxations of the POP. We establish convergence guarantees for both approaches and prove that theoretical properties extend to the polynomial case. Finally, we illustrate the significant computational advantages offered by the application of AD, particularly in the context of structural optimization problems
Complexity analysis and scalability of a matrix-free extrapolated geometric multigrid solver for curvilinear coordinates representations from fusion plasma applications
No full textInternational audienceTokamak fusion reactors are promising alternatives for future energy production. Gyrokinetic simulations areimportant tools to understand physical processes inside tokamaks and to improve the design of future plants. Ingyrokinetic codes such as Gysela, these simulations involve at each time step the solution of a gyrokinetic Poissonequation defined on disk-like cross sections. The authors of [14,15] proposed to discretize a simplified differentialequation using symmetric finite differences derived from the resulting energy functional and to use an implicitlyextrapolated geometric multigrid scheme tailored to problems in curvilinear coordinates. In this article, we extendthe discretization to a more realistic partial differential equation and demonstrate the optimal linear complexityof the proposed solver, in terms of computation and memory. We provide a general framework to analyze floatingpoint operations and memory usage of matrix-free approaches for stencil-based operators. Finally, we give anefficient matrix-free implementation for the considered solver exploiting a task-based multithreaded parallelismwhich takes advantage of the disk-shaped geometry of the problem. We demonstrate the parallel efficiency forthe solution of problems of size up to 50 million unknowns