335598 research outputs found
Sort by
DIME: Diffusion-Based Maximum Entropy Reinforcement Learning
No full textMaximum entropy reinforcement learning (MaxEnt-RL) has become the standard approach to RL due to its beneficial exploration properties. Traditionally, policies are parameterized using Gaussian distributions, which significantly limits their representational capacity. Diffusion-based policies offer a more expressive alternative, yet integrating them into MaxEnt-RL poses challenges—primarily due to the intractability of computing their marginal entropy. To overcome this, we propose Diffusion-Based Maximum Entropy RL (DIME). DIME leverages recent advances in approximate inference with diffusion models to derive a lower bound on the maximum entropy objective. Additionally, we propose a policy iteration scheme that provably converges to the optimal diffusion policy. Our method enables the use of expressive diffusion-based policies while retaining the principled exploration benefits of MaxEnt-RL, significantly outperforming other diffusion-based methods on challenging high-dimensional control benchmarks. It is also competitive with state-of-the-art non-diffusion based RL methods while requiring fewer algorithmic design choices and smaller update-to-data ratios, reducing computational complexity
HOGraspFlow: Exploring Vision-based Generative Grasp Synthesis with Hand-Object Priors and Taxonomy Awareness
No full textWe propose Hand-Object (HO)GraspFlow, an affordance-centric approach that retargets a single RGB with hand-object interaction (HOI) into multi-modal executable parallel jaw grasps without explicit geometric priors on target objects. Building on foundation models for hand reconstruction and vision, we synthesize SE(3) grasp poses with denoising flow matching (FM), conditioned on the following three complementary cues: RGB foundation features as visual semantics, HOI contact reconstruction, and taxonomy-aware prior on grasp types. Our approach demonstrates high fidelity in grasp synthesis without explicit HOI contact input or object geometry, while maintaining strong contact and taxonomy recognition. Another controlled comparison shows that HOGraspFlow consistently outperforms diffusion-based variants (HOGraspDiff), achieving high distributional fidelity and more stable optimization in SE(3). We demonstrate a reliable, object-agnostic grasp synthesis from human demonstrations in real-world experiments, where an average success rate of over 83% is achieved
Constraining four-heavy-quark operators with top-quark, Higgs, and electroweak precision data
No full textWe establish constraints on the dimension-six four-heavy-quark operators in the Standard Model Effective Field Theory (SMEFT) by synthesising LHC measurements of top-quark and single-Higgs production with electroweak precision observables. We scrutinise the choice of the scheme in single-Higgs calculations, demonstrating its non-negligible impact on SMEFT fits
Explaining Themselves and Making Friends: Towards Formalising the Sociability of Autonomous Agents
No full textWhile autonomous systems are integrated into more and more close-to-human application domains, we investigate sociability as a necessary extra-functional system property to ensure their integrability into diverse societies. To enable formalisation and formal validation of sociability of autonomous systems, we derive requirements for social rules from interdisciplinary sources. We further discuss explainability as a tool
for understanding social actions of autonomous systems
The aftermath of the pandemic: how the COVID-19 pandemic affected physical activity, fitness, health, and body fat in first-year students in Norway
No full textPublic health measures to limit the spread of COVID-19 included restricting physical activity (PA). Here we described the impact of pandemic restrictions and reduction in PA on physical fitness and health and body composition amongst first-year students, and the associations to body fat and total PA at the end of their first year. “On your own feet” is a longitudinal study exploring changes in lifestyle habits amongst first-year students. Questionnaires for assessment of perceived restriction, PA behaviour and fitness and health were administered at the start and end of the first year at university. Body composition (bioelectrical impedance analysis) and total PA (Actigraph®) were recorded at both time-points. In multivariable models we identified factors associated to body fat and total PA. We included 150 students aged 18-22 years, 53% of whom reported restrictions and 34% a reduction in PA due to the COVID-19 pandemic. Students reporting restrictions had comparable fitness, health, body composition and PA level at baseline and follow-up, compared to those without restrictions. Students with reduced PA less often reported “good” fitness (30% vs. 56%, p < 0.001) and health (54% vs. 70%, p = 0.046) and had higher mean body fat percentage (27% vs. 23%, p = 0.009) and lower total PA (314 vs. 420 cpm, p < 0.001) at baseline, compared to those without reduction in PA. At follow-up, they less often reported “good” physical fitness (26% vs. 54%, p = 0.005), while body composition and total PA were comparable. We concluded that students who report pandemic reduction in PA may need targeted interventions to improve fitness
Observation of a low energy nuclear recoil peak in the neutron calibration data of an AlO crystal in CRESST-III
No full textThe current generation of cryogenic solid state detectors used in direct dark matter and CENS searches typically reach energy thresholds of (10) eV for nuclear recoils. For a reliable calibration in this energy regime a method has been proposed, providing monoenergetic nuclear recoils at low energies ∼100 eV–1 keV. In this work we report on the observation of a peak at (1113.6) eV in the data of an AlO crystal in CRESST-III, which was irradiated with neutrons from an AmBe calibration source. We attribute this monoenergetic peak to the radiative capture of thermal neutrons on Al and the subsequent deexcitation via single emission. We compare the measured results with the outcome of Geant4 simulations and investigate the possibility to make use of this effect for the energy calibration of AlO detectors at low energies. We further investigate the possibility of a shift in the expected energy scale of this effect caused by the creation of defects in the target crystal
Evaluation of stratospheric transport in three generations of Chemistry-Climate Models
No full textThe representation of stratospheric transport in Chemistry-Climate Models (CCMs) is key for accurately reproducing and projecting the evolution of the ozone layer and other radiatively relevant trace gases. We evaluate stratospheric transport in CCMs that have participated in three model intercomparison initiatives (CCMVal-2, CCMI-1, and CCMI-2022) over the last ~15 years using modern satellite datasets and reanalyses. Key long-standing model biases persist across generations, with some worsening in recent simulations. Transport remains overly fast in the models, with a global mean age of air young bias of ~1 year for the CCMI-2022 median. It is argued that this bias could be associated with too fast tropical upwelling in the lower stratosphere, insufficient horizontal mixing and/or excessive vertical diffusion. In the springtime southern polar stratosphere, the final warming is delayed (~3 weeks), downwelling is underestimated (~25 %), and the depth of the ozone minimum is overestimated (~10 DU) on average in the most recent models. The tropopause is too high in all generations, and the tropical cold point tropopause is too warm in the latest generation (~1–2 K). Long-term trends in transport and over 1980–1999 are consistent across model generations and highlight the crucial role of ozone depletion in contributing to accelerate the Brewer-Dobson circulation and delaying the southern polar vortex breakdown
