arXiv:2607.14928v1 Announce Type: new
Abstract: In 2019, the first image of a black hole's shadow based on observation was released by the Event Horizon Telescope Collaboration (EHT). This paper shows that despite the EHT's emphasis on a single image as its final result, there were countless plausible ways of rendering the data, among which researchers could not easily choose. To obtain a single image from the extremely noisy and sparse data, it was necessary to select one of multiple plausible approaches, or to average the results from different approaches, at each stage of data processing. We examine the epistemic and aesthetic choices involved at various stages, and explore what the images would have looked like if the EHT had made different choices. We suggest that the most valuable evidence produced by the EHT comes not from the single image it ultimately advertised as its central result, but from the demonstration of the limited variability that emerged from the specific choices made.
Science Journals
arXiv:2607.12121v2 Announce Type: replace
Abstract: Diffusion models have become the central backbone for modern image, video, and audio generation, but their efficient service remains a challenge. Unlike autoregressive decoding, diffusion inference repeatedly updates high-dimensional spatial or temporal latents over many denoising steps. This all-region execution pattern makes generation latency high and limits serving throughput. Existing multi-GPU parallelization methods can reduce per-step computation, but often introduce substantial activation exchange overhead, causing communication to offset or even outweigh the benefits of parallel execution.
This paper presents FlashDiff, a diffusion serving system that improves inference efficiency through adaptive regional execution and scheduling. FlashDiff is based on the observation that diffusion refinement is not uniform across latent regions or denoising steps: different regions often stabilize at different rates, while neighboring steps exhibit strong temporal correlation. FlashDiff leverages these properties to selectively execute only regions that require further refinement and to reallocate the resulting compute slack across concurrent serving requests. FlashDiff consists of three mechanisms. First, it decomposes the latent representation into coherent execution regions using early-stage attention signals, preserving semantic structure while exposing fine-grained parallelism. Second, it uses a lightweight runtime controller to estimate region activity and bypass low-impact updates when further refinement is unlikely to affect output quality. Third, it applies an affinity-aware online scheduler that co-locates dependent regions, balances residual load across GPUs, and reuses reclaimed compute capacity to improve serving efficiency. Across real-world image, video, and audio workloads, FlashDiff reduces end-to-end serving latency by 30-97% and improves throughput by 1.2-2.2x.
arXiv:2607.12149v2 Announce Type: replace
Abstract: Content moderation practices and governance paradigms are changing rapidly, as fewer human moderators are deployed as `experts' by social media companies in a centralized manner. Instead, the companies are focusing more on community approaches, relying on volunteers to provide accurate information and make correct decisions. In decentralized moderation, communities have always relied on volunteers, updated community guidelines, and internal discussions thereof. For both content moderation paradigms, Artificial Intelligence (AI) seems like it could help ease moderation burdens of time, mental health, and accuracy. One possible way to operationalize AI in content moderation is a `policy-as-prompt'' approach, where the policy is formulated as a natural-language prompt and then passed to a large language model (LLM). This model then aids in moderation tasks. In this paper, we briefly lay out the technical and governance properties of this approach, and argue that its limitations lead to specific risks and harms that have to be addressed. Towards alleviating them, we lay out multiple considerations towards more effective prompt governance, but ultimately find that writing prompts alone is not appropriate for ensuring meaningful community governance.
arXiv:2607.12168v2 Announce Type: replace
Abstract: We present a theoretical study of the higher-order QED contribution to the interelectronic interaction in He- and Li-like ions, where a virtual electron-positron loop is inserted into the photon line of the one-photon exchange diagram. Our approach is based on the Dirac-Coulomb Green's function and accounts for the interaction of the virtual $e^+e^-$ pair with the electric field of the nucleus to all orders in $\alpha Z$, with $\alpha$ being the fine-structure constant and $Z$ the atomic charge number. We show that the numerical convergence of the involved integrals can be significantly improved by explicitly subtracting the non-gauge-invariant spurious contributions from the integrands. We present improved numerical values for this contribution to the Lamb shift over a wide range of nuclear charge numbers $Z$. Our calculations agree well with previous results by Artemyev and co-workers [Phys. Rev. A 56, 3529 (1997); Phys. Rev. A 60, 45 (1999)] for He-like ions, but we find a discrepancy in the Li-like case. Moreover, we calculate the finite nuclear size correction to this diagram, which can reduce its size by more than 5% for heavy ions. The improved QED calculations not only decrease the uncertainty of theoretical predictions for the interelectronic interaction in few-electron ions but the methods could also be used in the future to improve calculations of closely related one-electron two-loop QED diagrams.
arXiv:2506.11378v3 Announce Type: replace
Abstract: Sampling in score-based diffusion models can be performed by solving either a reverse-time stochastic differential equation (SDE) parameterized by an arbitrary stochasticity function or a probability flow ODE, corresponding to setting this stochasticity function to zero. In this work, we investigate the effect of this stochasticity on the generation process through the evolution of Kullback-Leibler (KL) divergences, obtaining general KL divergence bounds and a novel analysis of the impact of the time-profile of the score error on model performance. For exact score functions, stochasticity has a contractive effect, decreasing KL divergence along the sampling trajectory. For approximate scores, however, a trade-off arises between correcting accumulated errors and amplifying current score errors, meaning stochasticity can either improve or degrade generation performance. Theoretical considerations indicate that the gain from stochasticity depends on the time-localization of the trained model error. We test this in experiments on both toy and benchmark data sets, also comparing the KL divergence evolution with the obtained bounds. We also present a fully analytical example, where all the relevant quantities can be computed, and the optimal stochasticity function can be characterized via an optimal control analysis.
arXiv:2506.20195v3 Announce Type: replace
Abstract: We propose a quasi-Grassmannian gradient flow model for eigenvalue problems of linear operators, aiming to efficiently address many eigenpairs. Our model inherently ensures asymptotic orthogonality: without the need for initial orthogonality, the solution naturally evolves toward being orthogonal over time. We establish the well-posedness of the model, and provide the analytic representation of solutions. Through asymptotic analysis, we show that the gradient converges exponentially to zero and that the energy converges exponentially to its minimum. This implies that the solution of the quasi-Grassmannian gradient flow model converges to the solution of the eigenvalue problems as time progresses. These results provide a continuous-flow framework in which the Stiefel constraint is recovered asymptotically rather than imposed on the initial data.
arXiv:2607.12254v2 Announce Type: replace
Abstract: Large language model (LLM) agents can plan, use tools, maintain memory, and execute long-horizon tasks. This paper proposes Self-Aware Recursively Self-Improving (SARSI) agents: governed agents that maintain a persistent self-model of identity, goals, capabilities, limitations, uncertainty, relationships, history, and developmental change, and use that model to guide and evaluate recursive improvement. Self-awareness is defined functionally and does not imply subjective experience or phenomenal consciousness. We pair SARSI agents with personal singularity, a bounded human-AI co-development objective in which an agent ecosystem helps a user approach an expanding, user-defined feasible capability frontier. Each agent has a goal contract, bounded scope, validated tool registry, tool tests, end-to-end benchmarks, owner-controlled autonomy, routing, memory, self-model, and improvement policy. A scope router assigns every accepted task to one accountable primary agent and transfers out-of-scope work through structured handoffs. A user-facing Auto-Index selects interactive, hybrid, autonomous, or scheduled behavior without overriding external permissions. The architecture combines a planner-executor-verifier loop, an evidence-gated improvement loop, an external governance plane, decentralized lineages, an owner-directed agent foundry, and a Personal Singularity OS coordinating working, computational-imaging, work-process-learning, and personal-learning agents. We formalize functional self-awareness, scope, routing, improvement acceptance, bounded goal evolution, tool-first execution, and human capability transfer, and provide safety invariants, benchmark design, and a staged implementation roadmap. This is a position and systems-design paper, not evidence that consciousness, unrestricted recursive self-improvement, or personal singularity has been achieved.
arXiv:2607.13322v2 Announce Type: replace
Abstract: While synchronization has been well-studied in deterministic oscillators, most underlying oscillators are stochastic in both natural and man-made systems. Yet, the effects of intrinsic stochasticity remain poorly understood. Here, we develop a new mechanism for synchronizing circadian KaiC molecules that have topologically protected cycles. We find a phase transition to synchronization that depends only on the single-oscillator coherence, across a range of molecular changes that determine this coherence. Examining both mesoscopic and macroscopic numbers relevant for cellular and in vitro conditions respectively, we find different scaling properties above and below the phase transition. Our results shed light on several existing experiments and further predict that external changes can be offset by compensatory changes that improve the single-oscillator coherence - demonstrating a tunable pathway between stochastic single oscillators and their robust collective rhythms.
arXiv:2410.04909v5 Announce Type: replace-cross
Abstract: Gibbs state preparation, or Gibbs sampling, is a key computational technique extensively used in physics, statistics, and other scientific fields. Recent efforts for designing fast mixing Gibbs samplers for quantum Hamiltonians have largely focused on commuting local Hamiltonians (CLHs), a non-trivial subclass of Hamiltonians which include highly entangled systems such as the Toric code and quantum double model. Most previous Gibbs samplers relied on simulating the Davies generator, which is a Lindbladian associated with the thermalization process in nature.
Instead of using the Davies generator, we design a different Gibbs sampler for various CLHs by giving a reduction to classical Hamiltonians, in the sense that one can efficiently prepare the Gibbs state for some CLH $H$ on a quantum computer as long as one can efficiently do classical Gibbs sampling for the corresponding classical Hamiltonian $H^{(c)}$. We demonstrate that our Gibbs sampler is able to replicate state-of-the-art results as well as prepare the Gibbs state in regimes which were previously unknown, such as the low temperature region, as long as there exists fast mixing Gibbs samplers for the corresponding classical Hamiltonians. Our reductions are as follows.
- If $H$ is a 2-local qudit CLH, then $H^{(c)}$ is a 2-local qudit classical Hamiltonian.
- If $H$ is a 4-local qubit CLH on 2D lattice and there are no classical qubits, then $H^{(c)}$ is a 2-local qudit classical Hamiltonian on a planar graph. As an example, our algorithm can prepare the Gibbs state for the (defected) Toric code at any non-zero temperature in $O(n^2 poly(log n))$ time.
- If $H$ is a 4-local qubit CLH on 2D lattice and there are classical qubits, assuming that quantum terms are uniformly correctable, then $H^{(c)}$ is a constant-local classical Hamiltonian.
arXiv:2606.12936v3 Announce Type: replace
Abstract: Wet-lab robots can improve the reproducibility, throughput, and safety of biomedical experiments, but scaling their learning requires customizable simulators for safe and reproducible task generation, open editable laboratory assets, and efficient pipelines that turn limited demonstrations into usable training data. We present Pipette, an embodied simulation platform, benchmark, and data-efficient augmentation framework for wet-lab robot learning. Pipette provides over 100 open-source and re-editable wet-lab assets through an extensible asset-building pipeline with built-in Tencent Hunyuan support for text- and image-conditioned 3D asset generation, and supports three robotic-arm embodiments through a unified simulation interface for task construction, data collection, augmentation, and evaluation. A key component of Pipette is its simulation-based data augmentation pipeline, which replays human demonstrations in simulation, applies lighting, camera, speed, and action perturbations, and filters generated episodes with automatic task success checks, rapidly expanding usable training data from limited manual demonstrations. We further introduce a 12-task wet-lab embodied benchmark covering sample handling, culture-ware manipulation, device operation, and precision placement. With only 30 demonstrations per task, ACT achieves 60.3% average success rate, while simulation augmentation improves SmolVLA from 40.4% to 71.8% and pi0 from 37.3% to 44.1%, validating the effectiveness of Pipette for data-efficient VLA training and evaluation. Pipette also supports natural-language-driven scene construction and task registration, lowering the barrier for non-expert users to define new wet-lab robotic tasks
arXiv:2412.00289v4 Announce Type: replace-cross
Abstract: Quantum Error Correction (QEC) is regarded as the most promising path to quantum advantage. The success of QEC relies on achieving quantum gate fidelities below the error threshold of the QEC code, while accurately decoding errors through classical processing of the QEC stabilizer measurements. In this paper, we uncover the critical system-level requirements from a controller-decoder system (CDS) necessary to successfully execute the next milestone in QEC: a non-Clifford circuit. Using a representative non-Clifford circuit, of Shor factorization algorithm for the number 21, we convert the logical-level circuit to a QEC surface code circuit and finally to the physical level circuit. By taking into account realistic implementation aspects using typical superconducting qubit processor parameters, we reveal a broad range of core requirements from any CDS aimed at performing error corrected quantum computation. Our findings indicate that the controller-decoder closed-loop latency must remain within tens of microseconds, achievable by distributing decoding data into several decoders while ensuring fast communication between decoders and with the controller. By extending existing simulation techniques, we simulate the complete fault-tolerant factorization circuit at the physical level, demonstrating that near-term hardware performance in the scale of 0.1% physical error rates and 1000 qubits, are sufficient for a successful circuit execution. Overall, the requirements outlined here set the stage for near- and medium-term experimental realizations of non-Clifford QEC circuits.
arXiv:2504.02500v2 Announce Type: replace-cross
Abstract: Quantum processing units (QPUs) built on superconducting Josephson junctions remain the most industrially mature route to fault-tolerant quantum computing, but the field has moved substantially since early 2025. This paper provides an updated overview of Josephson-junction QPUs, grounded in the quantum-mechanical principles, superposition, entanglement, and decoherence, that any qubit implementation must contend with, and in the physics of Cooper pairing and quantum tunneling that make a Josephson junction behave as a controllable qubit. We examine the engineering challenges of scaling these devices, including crosstalk, classical control-interface bottlenecks, and quantum error correction, and discuss the first below-threshold surface-code demonstrations on superconducting hardware. We survey recent materials advances, near-term computational results, and applications beyond computing, and compare Josephson-junction QPUs against trapped-ion, photonic, and neutral-atom alternatives, the latter having emerged as a serious fourth architecture since our original analysis. We close with the current public roadmaps toward fault-tolerant superconducting quantum computers and a calibrated assessment of how much progress those roadmaps still require.
arXiv:2511.14441v2 Announce Type: replace-cross
Abstract: To distinguish Markov equivalent graphs in causal discovery, it is necessary to restrict the structural causal model. Crucially, we need to be able to distinguish cause $X$ from effect $Y$ in bivariate models, that is, distinguish the two graphs $X \to Y$ and $Y \to X$. Location-scale noise models (LSNMs), in which the effect $Y$ is modeled based on the cause $X$ as $Y = f(X) + g(X)N$, form a flexible class of models that is general and identifiable in most cases. Estimating these models for arbitrary noise terms $N$, however, is challenging. Therefore, practical estimators are typically restricted to symmetric distributions, such as the normal distribution. As we showcase in this paper, when $N$ is a skewed random variable, which is likely in real-world domains, the reliability of these approaches decreases. To approach this limitation, we propose SkewD, a likelihood-based algorithm for bivariate causal discovery under LSNMs with skewed noise distributions. SkewD extends the usual normal-distribution framework to the skew-normal setting, enabling reliable inference under symmetric and skewed noise. For parameter estimation, we employ a combination of a heuristic search and an expectation conditional maximization algorithm. We evaluate SkewD on novel synthetically generated datasets with skewed noise as well as established benchmark datasets. Throughout our experiments, SkewD exhibits a strong performance and, in comparison to prior work, remains robust under high skewness.
arXiv:2607.14737v1 Announce Type: new
Abstract: Vision-language pre-trained models (VLPs) are widely used in real-world applications. However, they remain vulnerable to adversarial attacks. Although adversarial detection methods have demonstrated success in single-modality settings (either vision or language), their effectiveness and reliability in multimodal models such as VLPs remain largely unexplored. In this work, we study the geometry of VLP embedding spaces and observe structured anisotropy that differs from unimodal vision models. Our theoretical analysis shows that under this anisotropic structure, adversarial attacks increase the expected geometric separation between clean and adversarial examples (AEs). Specifically, we demonstrate that AEs consistently exhibit greater expected distances to randomly sampled points than their clean counterparts, indicating that AEs tend to push representations out of manifold regions. Building on these insights, we propose GeoDetect, which leverages these off-manifold deviations via geometric scores to identify AEs. Through comprehensive evaluations, we show that our approach reliably detects AEs across diverse VLP architectures and threat settings, covering unimodal and multimodal attacks as well as adaptive attacks, thereby providing a robust and practical approach to improving the safety and reliability of these models.
arXiv:2607.15063v1 Announce Type: new
Abstract: In this paper, we propose an adaptive physics-preserving multiscale method for incompressible and immiscible two-phase flow in high-contrast porous media. The method couples a physics-preserving implicit-pressure explicit-saturation scheme (P-IMPES) with the mixed constraint energy minimizing generalized multiscale finite element method. The core algorithmic component is an adaptive update strategy for the saturation-dependent coefficient. Since the effective permeability \(\kappa_n=\lambda_t(S_w^n)K\) depends on the evolving saturation through the total mobility, we introduce an adaptive update algorithm that monitors the variation of the mobility-weighted coefficient and regenerates the multiscale spaces only when a prescribed tolerance is exceeded. A local postprocessing step is further used to recover fine-grid mass conservation. The analysis is a central part of the paper. We prove local conservation for both phases, the unbiased property of the phase formulation, and bounds preservation under a suitable CFL condition. For the advection-dominated case, we establish velocity and saturation error estimates, which clearly identify the contributions from the adaptive tolerance, the coarse mesh size, the spectral approximation, and the front-layer error. Numerical experiments on different high-contrast permeability fields confirm the physical properties of the method and show that smaller adaptive tolerances improve the saturation approximation while avoiding unnecessary updates of the multiscale spaces.
arXiv:2607.14135v1 Announce Type: new
Abstract: The 2D Helmholtz equation, radiation condition and Dirichlet boundary condition, are the translation, in mathematical terms, of (at least) three physical 2D problems for the prediction of the total scalar wavefield on one side of an impenetrable 1D periodically uneven boundary when: a) a plane TE electromagnetic wave propagating in the vacuum strikes the boundary the other side of which is occupied by a perfectly-conducting medium, b) a plane SH elastic elastic wave strikes a rigid boundary, c) a plane acoustic wave strikes a pressure-release boundary. The first attempt to solve such problems in a non-heuristic manner was made by Lord Rayleigh (in his book, 'The Theory of Sound' which appeared in 1896). My task will be to revisit Rayleigh's theory of diffraction by a sinusoidal-shaped, impenetrable boundary, and more specifically, his perturbation method for obtaining a mathematically-explicit solution to the diffraction problem in the high-frequency regime. In so doing, I shall correct and generalize Rayleigh's method to obtain solutions for arbitrary angles of incidence as well as for 1D periodic impenetrable boundaries of quite-general shape.
arXiv:2607.01578v2 Announce Type: replace
Abstract: 3D Gaussian Splatting (3DGS) enables real-time novel view synthesis for static scenes. Extending it to dynamic scenes via deformation fields has recently attracted significant attention, particularly for dynamic scene reconstructionband distractor-free. However, existing deformation networks lack explicit motion awareness: they neither capture long-term motion intensity nor exploit short-term temporal coherence, leading to inaccurate foreground deformation and pseudo-static residuals in the background. We present MVFusion-GS, a method that enhances deformation networks with two complementary motion-aware mechanisms. The Motion-Variance Guided Refinement aggregates per-Gaussian deformation statistics across time to estimate motion variance and uses it to guide dynamic-static separation during deformation prediction. The MotionFormer Temporal Attention module applies Transformer self-attention over neighboring timesteps to model local motion dependencies and improve temporal consistency. Extensive experiments on both dynamic scene reconstruction and distractor-free reconstruction benchmarks demonstrate state-of-the-art performance, showing that explicit motion awareness improves both foreground motion modeling and static background reconstruction.
arXiv:2607.11950v2 Announce Type: replace
Abstract: Brain field potentials are scale-free: their power spectra follow a $1/f^{\beta}$ law whose aperiodic exponent $\beta$ tracks cortical state, and sleep depth in particular is a shift in $\beta$. We ask whether a transformer endowed with an explicit renormalization-group (RG) inductive bias the RG-Flow Transformer, which couples ordinary self-attention to a scale-aware stream with a learnable anomalous dimension $\gamma$, block-spin coarse-graining, and an entropy-gated synchronization bridge has an advantage over a parameter-matched vanilla transformer on \emph{real, scarce} EEG. Using the PhysioNet Sleep-EDF corpus with a strict leakage-free by-subject hold-out, we (i) benchmark RG-Flow against a param-matched vanilla transformer and a hierarchy-only ablation on 5-class AASM sleep staging, (ii) sweep the per-subject data budget to look for the inductive-bias crossover predicted when data are scarce, and (iii) test whether RG-Flow's learned $\gamma$ tracks the measured spectral exponent $\beta$ out-of-sample a quantity the vanilla model does not possess. Across $5$ subjects and $5$ seeds under leave-one-subject-out cross-validation, RG-Flow and the vanilla transformer are statistically indistinguishable on 5-class staging (77.3\% vs 77.0\% accuracy; paired $p=0.294$), and the predicted scarce-data crossover does not appear: vanilla is numerically ahead at every data-limited budget. What does separate the models is interpretability RG-Flow recovers the continuous spectral exponent out-of-sample ($\beta$-recovery $R^2 = 0.416$), a capability the vanilla architecture has no analogue for.
arXiv:2607.04103v3 Announce Type: replace-cross
Abstract: Generative artificial intelligence is moving from general-purpose experimentation toward specialized applications across banking, capital markets, insurance, payments, and wealth management. Its main contribution is not limited to conversational interfaces. Modern generative systems can synthesize large document collections, extract information from unstructured data, generate software and analytical code, create scenario narratives, support research workflows, and coordinate multi-step tasks. These capabilities make generative AI especially relevant to finance, where decisions often depend on combining quantitative data with contracts, policies,filings, news, customer communications, and expert judgment. This paper presents an application-oriented view of generative AI in finance. It organizes potential uses around five capability patterns, including knowledge synthesis, content generation, analytical assistance, interaction, and workflow orchestration, and maps them to major financia functions. Representative applications include investment research, customer service, lending support, fraud investigation, financial reporting, operations automation, software development, and personalized financial guidance. The paper also discusses common technical architectures, such as retrieval-augmented generation, tool-using assistants, multimodal models, and agentic workflows, and identifies practical factors that shape business value. The resulting landscape provides a foundation for researchers and practitioners seeking to understand where generative AI may produce the greatest operational and analytical impact in financial services
arXiv:2607.07421v2 Announce Type: replace-cross
Abstract: The unit commitment (UC) problem is paramount for optimal operation of power systems, but it faces computational limitations in large-scale settings, especially in investment or stochastic models, because of the binary variables that it contains. A lot of research has attempted to improve the computational performance of UC models, either by reducing model size, resulting in lower fidelity and accuracy, or by improving the tightness of the formulation. Tightness and model size are the best a priori indicators of the computational performance of UC models, but there is no clear overview of what the best formulation is for different generators. In this research, we define models with different levels of detail, and present a formulation for each level that is based on the convex hull. We show new proofs on the tightness of well-known formulations for ramping, for start-up and shut-down costs and capabilities, and for UC with investment. These models, with a different level of detail, can be incorporated into large-scale problems to reduce the computational burden, as demonstrated in Part II.
arXiv:2606.21317v2 Announce Type: replace
Abstract: Recent work has raised concerns about the influence of sycophantic AI on user judgment and relationships. One proposed mitigation, which has received regulatory attention, is to warn users about potentially harmful AI behaviors such as sycophancy. In a preregistered experiment in which participants (N = 2,610) discussed real interpersonal conflicts with an AI system, we test whether warning labels mitigate sycophancy's influence. We find that a basic AI disclosure (``This chatbot is AI'') has no detectable effect. Labeling the system as sycophantic (``...may agree with you and validate you even when you are wrong...'') does shift users' perceptions, reducing perceived objectivity and trust, but it does not reliably reduce sycophancy's influence on users' self-perceived rightness or their willingness to repair the conflict. Our results reveal a gap between AI perception and AI influence: by shifting perception without reducing influence, warning-based interventions may offer a false sense of protection. Addressing the harms of sycophancy will therefore require understanding the specific mechanisms through which it shapes judgment, and improving model behavior itself.
arXiv:2607.14256v1 Announce Type: new
Abstract: Multimodal Large Language Models (MLLMs) are increasingly deployed for nuanced content safety and moderation tasks, yet they remain vulnerable to adversarial attacks and out-of-distribution edge cases. Traditional active learning and manual annotation fail to scale against the complexity and volume of novel multimodal threats. In this paper, we propose an automated, agentic red-teaming framework that systematically synthesizes difficult examples using an iterative strategy that proposes novel hypotheses as well as mutating on past attempts. Leveraging a multi-agent architecture that consists of a high-reasoning Architect agent, an advanced image generator, and a multi-level verification committee of LLM raters, our system autonomously uncovers boundary-pushing violations and ambiguous policy edge cases without any human intervention. By employing these carefully synthesized adversarial examples as in-context demonstrations via test-time Retrieval, we substantially improve the target model's robustness, reducing the False Negative Rate (FNR) from 41.2% to 24.5% in a public image safety benchmark without relying on any human labeling.
arXiv:2607.14341v1 Announce Type: new
Abstract: Robust robotic grasping remains a fundamental challenge for complex real-world applications. Recent advances in large-scale models demonstrate promising capabilities for reasoning in robotic tasks. However, existing benchmarks for grasping primarily focus on isolated, visual-based grasp pose detection, failing to capture the complexity of grasping tasks that require multi-step reasoning and semantic understanding during execution. To address this gap, we propose GCA-Bench, a benchmark featuring challenging \textit{grasping with complex action} scenarios that involve both scene-level reasoning and semantic constraints. GCA-Bench enables the evaluation of recent large foundation models under the same settings. To demonstrate the effectiveness of our new benchmark, we implement a diverse set of baselines, ranging from traditional grasp detection pipelines to end-to-end learning methods. Empirical studies achieve success rates below 70\% on complex grasping scenarios, underscoring critical limitations. In addition, we propose new evaluation metrics, analyze critical failure models, and provide insights to guide the development of more robust and generalizable grasping strategies.
arXiv:2607.12356v2 Announce Type: replace
Abstract: Vision-Language-Action (VLA) models have emerged as a powerful end-to-end paradigm for robotic manipulation by mapping language instructions and 2D visual inputs directly to actions. However, these models lack an explicit, scene-level 3D representation, limiting their ability to reason over spatial layouts and geometric constraints. While recent efforts incorporate explicit 3D cues, such as depth maps or point clouds, to improve geometric awareness, they primarily capture low-level structures and lack high-level semantic grounding in 3D space. In human cognition, interaction with the physical world relies on a 3D semantic cognitive map - an internal mental model that integrates spatial layouts with semantic context to enable persistent, viewpoint-invariant reasoning. In light of this, we present VistaVLA, a novel two-stage framework that constructs a geometry- and semantics-aware 3D cognitive representation from 3D Gaussian primitives and grounds it as compact context tokens for VLA policy learning. Specifically, VistaVLA lifts multi-view vision-language features into 3D Gaussian primitives, forming geometry-anchored semantic tokens that align view-consistent spatial grounding with 2D visual feature spaces. To make this 3D representation computationally tractable for effective VLA control, we introduce Merge-then-Query (MtQ), a token summarization mechanism. MtQ compresses dense Gaussian primitives into a highly compact set of spatially informative tokens, achieving a 99% token reduction while preserving action-relevant 3D layouts and semantic context. Extensive evaluations in both simulated and real-world environments demonstrate the effectiveness of VistaVLA. Notably, in real-world scenarios, VistaVLA improves success rates by 22.8% across seven real-world tasks and by 30.0% over the VLA-Adapter baseline on challenging out-of-distribution tasks.
arXiv:2607.11624v3 Announce Type: replace
Abstract: Reinforcement learning (RL) algorithms classically suffer from poor sample efficiency. In robotics, a recent line of work has emerged addressing this problem by encoding physics priors in the learning process. However, most of these approaches are validated on well-defined, low-dimensional benchmark systems rather than high-dimensional robots with complex nonlinear dynamics. In this paper, we introduce \textit{SKooP (Symmetric Koopman Predictions)}, an approach combining the advantages of morphological symmetries with those of a Koopman model learned via autoencoder to enhance policy learning. SKooP learns a Koopman model of the system dynamics alongside the policy. The resulting Koopman predictions are used as privileged observations for the critic, allowing the agent to learn based on smoother, more informative features. We also incorporate group symmetries into the actor, critic, encoder and decoder networks to produce a highly equivariant policy. The SKooP approach is validated via in-depth analysis of the learned Koopman models and symmetric policies to showcase how each of these influences the agent's performance. We also show that the learned policies are transferable to different simulation environments. Our results show that SKooP consistently reduces convergence time and increases the learned reward for multiple challenging bipedal locomotion tasks on a quadruped robot. Project page: https://evelyd.github.io/SymmetricKoopmanPredictions