Scientific Publications of the University of Toulouse II Le Mirail
Not a member yet
92205 research outputs found
Sort by
Coupled Local and Global World Models for Efficient First Order RL
No full textRL has demonstrated strong performance in locomotion through robust sim-toreal transfer from parallelized simulators to hardware. In contrast, extending this simulator-centric approach to real-world robotics manipulation is hindered by high data demands, sim-to-real discrepancies in intricate interactions, and the difficulty of engineering rewards or accurate physics for tasks involving deformable or irregular objects. We introduce a framework that replaces simulators with a diffusion-based world model trained on real robot image data, which captures complex dynamics from data without rule-based modeling, thus addressing the limitations that prevent simulator-style RL from succeeding in manipulation. Our method enables feasible policy training despite the high cost of trajectory unrolls in large-scale image models, through a novel decoupled first-order gradient (FoG) approach: the full world model generates accurate forward trajectories, while a lightweight latent-space surrogate model is learned to approximate local dynamics around the policy's trajectories, providing efficient, low-variance gradients via backpropagation. This surrogate operates on compressed representations, avoiding direct differentiation through pixel-level diffusion processes. Unlike previous model-based RL approaches, this decoupling ensures high-fidelity forward unrolling alongside computationally tractable backward differentiation. Evaluated on a real robotic arm, our method achieves high success rate on the Push-T task, strongly outperforming PPO in sample-and time-efficiency while offering a scalable alternative to simulator-dependent RL for real-world manipulation
Approximating the Uniform Value in Hidden Stochastic Games with Doeblin Conditions
No full textIn \emph{zero-sum two-player hidden stochastic games}, players observe partial information about the state. We address: the existence of the \emph{uniform value}, i.e., a limiting average payoff that both players can guarantee for sufficiently long durations, and the existence of an algorithm to approximate it. Previous work shows that, in the general case, the uniform value may fail to exist, and, even when it does, there need not exist an algorithm to compute or approximate it. Therefore, we consider the \emph{Doeblin condition} in hidden stochastic games, requiring that, after a sufficiently long time, the posterior beliefs have a uniformly positive probability of resetting to one of finitely many neighborhoods in the belief space. We prove the existence of the uniform value and provide an algorithm to approximate it. We identify sufficient conditions, namely \emph{ergodicity} in the blind setting (when the signal is uninformative) and \emph{primitivity} in the hidden setting (when there are multiple signals). Moreover, we show that, in the hidden setting, ergodicity does not guarantee the Doeblin condition. Our results are new even for the one-player setting, i.e., partially observable Markov decision processes
Cost trade-offs in matrix inversion updates for streaming outlier detection
No full textInternational audienceOutlier detection identifies data points that deviate significantly from expected patterns, revealing anomalies that may require special attention. Incorporating online learning further improves accuracy by continuously updating the model to reflect the most recent data. When employing the Christoffel function as an outlier score, online learning requires updating the inverse of a matrix following a rank-k update, given the initial inverse. Surprisingly, there is no consensus on the optimal method for this task. This technical note aims to compare three different updating methods: Direct Inversion (DI), Iterative Sherman–Morrison (ISM), and Woodbury Matrix Identity (WMI), to identify the most suitable approach for different scenarios. We first derive the theoretical computational costs of each method and then validate these findings through comprehensive Python simulations run on a CPU. These results allow us to propose a simple, quantitative, and easy-to-remember rule that can be stated qualitatively as follows: ISM is optimal for rank-1 updates, WMI excels for small updates relative to matrix size, and DI is preferable otherwise. This technical note produces a general result for any problem involving a matrix inversion update. In particular, it contributes to the ongoing development of efficient online outlier detection techniques
Celtic Metrology Reconsidered: Archaeological and Textual Evidence
No full textThis paper explores the question of Celtic metrology through a combined analysis of archaeological and written sources. While the Iron Age has long been considered poorly documented with regard to measurement practices, recent studies, particularly those focusing on weighing equipment, have provided new evidence that allows a more nuanced assessment. Archaeological evidence suggests the existence of structured practices in several domains, particularly concerning weight ; whether they do not demonstrate the existence of a single unified "Celtic" system of measurement, they nevertheless suggest a degree of metrological coherence across large areas of Celtic Europe, allowing practical equivalences between different units.The article also reassesses the long-standing hypothesis that the Celts used a vigesimal counting system. By reviewing and discussing the demonstration of this idea, it shows that the evidence supporting a specifically "Gallic" vigesimal system remains weak, while the archaeological record nor the available historical sources provide convincing support for such a model.</div
Une éducation aux communs, un enjeu démocratique
No full textInternational audienceDemocracy is linked to education through engagement in shared action, requiring participation and deliberation. Organizing politically, particularly on the environment, requires these consensual political objects that are the commons, even if they encounter multiple forms of appropriation. They bring together resources, communities, rules, and cooperation. Their shared management fosters democracy, as E. Oström has shown. Educating about the commons is not just about sharing goods, but also about learning together to manage resources, establish rules, and commit to the community, developing eco-citizenship and new behaviors. Educating about the commons will necessarily involve learning about new democratic concepts.This democratic education involves interdisciplinary approaches and places of experimentation, while promoting cooperation and commitment to the territory and to the ecological transition. The commons are thus a lever for renewing democracy. They link participation, co-obligation, and environmental preservation. Finally, they encourage humble and available citizenship.La démocratie est liée à l’éducation par l’engagement dans un agir commun, nécessitant participation et délibération. S’organiser politiquement, notamment sur l’environnement, suppose ces objets politiques consensuels que sont les communs, même s’ils se heurtent à de multiples formes d’appropriation. Ils rassemblent ressources, communautés, règles et coopération. Leur gestion partagée favorise la démocratie, comme l’a montré E. Oström. Éduquer aux communs, ce n’est pas seulement partager des biens, mais apprendre ensemble à gérer des ressources, établir des règles, s’engager pour la communauté, en développant écocitoyenneté et comportements nouveaux. Éduquer aux communs va supposer d’apprendre nécessairement de nouveaux objets démocratiques. Cette éducation démocratique implique des approches interdisciplinaires et des lieux d’expérimentation, tout en valorisant la coopération et l’engagement sur le territoire et dans la transition écologique. Les communs sont ainsi un levier pour renouveler la démocratie. Ils relient participation, co-obligation et préservation du milieu. Finalement, ils incitent à une citoyenneté humble et disponible
Diderot ou Voltaire : quel modèle pour la Statue équestre de Pierre le Grand par Falconet ?
No full textInternational audienceDans cet article, nous examinons le rôle de Diderot et de Voltaire dans la création de la statue équestre de Pierre le Grand à Saint-Pétersbourg. Si l’amitié entre Diderot et Falconet, le sculpteur, est bien documentée, les écrits de Falconet montrent que Voltaire a profondément marqué son imaginaire. Cette inspiration est souvent négligée, éclipsée par l’idée reçue du rôle central de Diderot dans la création du monument. En même temps, lorsque Catherine II engage Falconet à venir œuvrer en Russie, elle le présente alors comme est « l’ami d’âme de Diderot ». En recrutant un artiste de cette envergure, elle ne fait pas simplement venir un sculpteur, mais également un proche des philosophes qu’elle admire tant. Pourtant, il est possible de remettre en question cette vision qui attribue à Diderot une place centrale
Multi-method constraints on late Pleistocene glacier fluctuations in the Ossau valley (SW France), with wider implications for the Pyrenean icefield during the last glacial maximum
No full textInternational audienceCurrent evidence throughout the Pyrenees indicates that the local last glacial maximum (LLGM) of the Late Pleistocene occurred earlier than the globally recognised LGM period (26-19 ka). Some LGM glaciers subsequently grew to sizes similar to those of their LLGM predecessors, but others fell short. These apparent differences have been attributed either to peculiar palaeoprecipitation patterns across the mountain range or to artifacts among the different dating methods underpinning the glacier fluctuation chronologies. Here we test these two alternatives by presenting 33 new ¹⁰Be and ³⁶Cl surface exposure ages from the terminal moraines of the Ossau valley, by elaborating 3D glacial reconstructions fitted to successive generations of landforms, and by discussing the data in light of published ¹⁴C ages from the nearby Estarrès threshold-lake glaciolacustrine sequence and from several archaeological sites. Results indicate that a first major glacier advance occurred at some time before 35 ka cal BP, after which the Ossau glacier receded by 4 km between 34-30 ka and 31-27 ka cal BP. A subsequent readvance generated LLGM moraines ca. 22.7 ka, followed by post-LLGM recessional moraines (age: 20.2 ka) positioned 2 km to the south. Around 19 ka cal BP, the glacier retreated by a further 30 km. By revealing an interval of maximum glacier growth during the LGM, this new chronology suggests that the distinction between LGM and LLGM in the Pyrenees may increasingly become obsolete as multi-method approaches help to refine Late Pleistocene icefield fluctuation patterns in other valleys
GEOSUR -Lot3 -Test Case Generation for Geolocation Fusion Algorithms
No full textA key component of autonomous agricultural robots is the geolocation system, which relies on multiple sensors (e.g., GNSS -Global Navigation Satellite System -and IMU -Inertial Measurement Unit) combined using fusion algorithms. These algorithms must handle diverse situations caused by environmental conditions, sensor failures, or measurement inaccuracies. One major challenge is to systematically explore these situations with diverse scenarios, in order to generate test cases for validating such algorithms. This paper addresses this challenge by proposing a comprehensive framework that integrates a geolocation fault model with an open-source test case generator (TAF). An experiment conducted on a standard fusion algorithm demonstrates that the proposed framework can generate valid and diverse test cases with very low resource consumption
Antibioprophylaxis in thoracic surgery, thoracic endoscopy, and interventional radiology
No full textInternational audienceThoracic surgery involves, on one hand, pulmonary resection surgeries with opening of the bronchial tree (clean-contaminated surgery, Altemeier class 2), and mediastinal or pleural surgeries without bronchial opening (clean surgery, Altemeier class 1). The indication for antibiotic prophylaxis is clear in both types of surgery (GRADE 1). Its purpose is to reduce surgical site infections (SSI), including wall infections and pyothorax, and for pulmonary resection surgeries, it may also be beneficial in reducing postoperative pneumonias (POP). The incidences of postoperative infections vary: SSI (5%), pyothorax/empyema (1%), POP between 2 and 5%, and up to 13% in certain series. Risk factors for POP include operative technique (open surgery vs. video-assisted), operative duration, and certain patient factors, including having chronic obstructive pulmonary disease (COPD). The most frequently isolated bacteria during POP are staphylococci, S. pneumoniae, H. influenzae, and Gram-negative bacilli (GNB). When the incidence of POP is high, the use of amoxicillin/clavulanate as antibiotic prophylaxis could reduce its incidence, as well as extending prophylactic antibiotics for a maximum of 24 to 48 hours postoperatively, which could also reduce the incidence of POP when it is high. This underpins the proposal made in the table for COPD patients undergoing pulmonary resection
