arXiv subjects

Yue Hu

Publications and source records attributed to Yue Hu.

At least 91 records · Page 5Linked to original sources

arXiv2026

CityCube: Benchmarking Cross-view Spatial Reasoning on Vision-Language Models in Urban Environments

Cross-view spatial reasoning is essential for embodied AI, underpinning spatial understanding, mental simulation and planning in complex environments. Existing benchmarks primarily emphasize indoor or street settings, overlooking the unique challenges of open-ended urban spaces characterized by rich semantics, complex geometries, and view variations. To address this, we introduce CityCube, a systematic benchmark designed to probe cross-view reasoning capabilities of current VLMs in urban settings. CityCube integrates four viewpoint dynamics to mimic camera movements and spans a wide spectrum of perspectives from multiple platforms, e.g., vehicles, drones and satellites. For a comprehensive assessment, it features 5,022 meticulously annotated multi-view QA pairs categorized into five cognitive dimensions and three spatial relation expressions. A comprehensive evaluation of 33 VLMs reveals a significant performance disparity with humans: even large-scale models struggle to exceed 54.1% accuracy, remaining 34.2% below human performance. By contrast, small-scale fine-tuned VLMs achieve over 60.0% accuracy, highlighting the necessity of our benchmark. Further analyses indicate the task correlations and fundamental cognitive disparity between VLMs and human-like reasoning.

cs.CV↗

arXiv2026

MemGovern: Enhancing Code Agents through Learning from Governed Human Experiences

While autonomous software engineering (SWE) agents are reshaping programming paradigms, they currently suffer from a "closed-world" limitation: they attempt to fix bugs from scratch or solely using local context, ignoring the immense historical human experience available on platforms like GitHub. Accessing this open-world experience is hindered by the unstructured and fragmented nature of real-world issue-tracking data. In this paper, we introduce MemGovern, a framework designed to govern and transform raw GitHub data into actionable experiential memory for agents. MemGovern employs experience governance to convert human experience into agent-friendly experience cards and introduces an agentic experience search strategy that enables logic-driven retrieval of human expertise. By producing 135K governed experience cards, MemGovern achieves a significant performance boost, improving resolution rates on the SWE-bench Verified by 4.65%. As a plug-in approach, MemGovern provides a solution for agent-friendly memory infrastructure.

cs.SE↗

arXiv2026

Machine-Learning Estimation of Energy Fractions in MHD Turbulence Modes

Magnetohydrodynamic (MHD) turbulence plays a central role in many astrophysical processes in the interstellar medium (ISM), including star formation and cosmic-ray transport and acceleration. MHD turbulence can be decomposed into three fundamental modes-fast, slow, and Alfvén-each contributing differently to the dynamics of the medium. However, characterizing and separating the energy fractions of these modes was challenging due to the limited 2D information available from observations. To address this difficulty, we use 3D isothermal and multiphase MHD turbulence simulations to examine how mode energy fractions vary under different physical conditions. Overall, we find that the Alfvén and slow modes carry comparable kinetic-energy fractions and together dominate the turbulent energy budget in multiphase media, while the fast mode contributes the smallest fraction. Relative to isothermal conditions, multiphase simulations exhibit an enhanced fast-mode energy fraction. We further introduce a machine-learning-based approach that employs a conditional Residual Neural Network to infer these fractions directly from spectroscopic data. The method leverages the fact that the three MHD modes imprint distinct morphological signatures in spectroscopic maps owing to their differing contributions to density and velocity fluctuations. Our model is trained on a suite of isothermal and multiphase simulations covering typical ISM conditions. We demonstrate that our machine learning model can recover the mode fractions from spectroscopic observables, achieving mean relative normalized errors of approximately 0 and standard deviation of 0.01 - 0.02 for seen data and 0.1 - 1.8 for unseen data.

astro-ph.GA↗

arXiv2026

Unveiling the 3D structure of the central molecular zone from stellar kinematics and photometry: The 50 and 20 km/s clouds

The central molecular zone (CMZ), surrounding the Galactic centre, is the largest reservoir of dense molecular gas in the Galaxy. Despite its relative proximity, the 3D structure of the CMZ remains poorly constrained, primarily due to projection effects. We aim to constrain the line-of-sight location of two molecular clouds in the CMZ -- the 50 and 20 km/s clouds -- and to investigate their possible physical connection using stellar kinematics and photometry. This study serves as a pilot for future applications across the full CMZ. We estimated the line-of-sight position of the clouds by analysing stellar kinematics, stellar densities, and stellar populations towards the cloud regions and a control field. We find an absence of westward moving stars in the cloud regions, which indicates that they lie on the near side of the CMZ. This interpretation is supported by the stellar density distributions. The similar behaviour observed in the two clouds, as well as in the region between them (the ridge), suggests that they are located at comparable distances and are physically linked. We also identified an intermediate-age stellar population (2-7 Gyr) in both regions, consistent with that observed on the near side of the CMZ. We estimated the line-of-sight distances at which the clouds and the ridge become kinematically detectable (i.e. where the proper motion component parallel to the Galactic plane differs from that of the control field at the 3 sigma level) by converting their measured proper motions parallel to the Galactic plane using a theoretical model of the stellar distribution. We find that the 50 and 20 km/s clouds are located at $43\pm8$ pc and $56\pm11$ pc from Sgr A*, respectively, and that the ridge lies at $56\pm11$ pc; this supports the idea that the clouds are physically connected through the ridge.

astro-ph.GA↗

arXiv2025

Obtaining Magnetization of Super-Alfvénic Turbulence with the Structure Functions of Gradient Directions

Super-Alfvénic turbulence is widespread in astrophysical environments, including molecular clouds and the diffuse plasma of galaxy clusters. At large scales, magnetic fields play only a minor dynamical role; however, for sufficiently extended turbulent cascades, the motions transition into the MHD regime at a characteristic scale $l_A$. We introduce a new diagnostic based on the structure functions of the gradient directions, which can be obtained directly from spectroscopic and synchrotron intensity observations. We demonstrate that the new measure robustly recovers the transition scale $l_A$. Building on this result, we propose a generalized expression that replaces the traditional Davis-Chandrasekhar-Fermi (DCF) method for estimating magnetic field strength in the super-Alfvénic regime, where the DCF approach fails. We further show how the magnetization and magnetic field strength of diffuse media, such as the intracluster medium, can be inferred using synchrotron intensity maps. Our theoretical predictions are validated through numerical simulations.

astro-ph.GA↗

arXiv2025

Sparse-Tuning: Adapting Vision Transformers with Efficient Fine-tuning and Inference

Parameter-efficient fine-tuning (PEFT) has emerged as a popular solution for adapting pre-trained Vision Transformer (ViT) models to downstream applications by updating only a small subset of parameters. While current PEFT methods have achieved fine-tuning efficiency, they overlook the efficiency of computation and GPU memory during inference, falling short of practical requirements. To address this limitation, we propose Sparse-Tuning, an efficient and effective framework that leverages popular token sparsification (TS) techniques to reduce information redundancy in images and videos, thereby significantly improving computational and memory efficiency. However, TS often compromises performance due to inevitable information loss. To address this limitation, we further introduce Dense Adapters (DA) to compensate for the information losses incurred by token sparsification. DA integrates comprehensive token information from shallow layers into the retained tokens of deeper layers, ensuring minimal performance degradation. Through the integration of TS techniques and DA, Sparse-Tuning achieves a significant reduction in computation and memory overhead while maintaining performance. Empirical results on VTAB-1K, three image datasets, and two video datasets show that Sparse-Tuning reduces GFLOPs to 66\% of the original ViT-B while achieving state-of-the-art performance compared to full fine-tuning and other PEFT baselines.

cs.CV↗

arXiv2025

ACES: The Magnetic Field in Large Filaments in the Galactic Center

The Galactic Center (GC) is an extreme region of the Milky Way that is host to a complex set of thermal and non-thermal structures. In particular, the GC contains high-density gas and dust that is collectively referred to as the Central Molecular Zone (CMZ). In this work, we study a subset of HNCO filaments identified in band 3 ALMA observations of the GC obtained by the ALMA CMZ Exploration Survey (ACES) that are comparable to high density filaments identified in the Galactic Disk. We compare the orientation of the magnetic field derived from 214 um SOFIA and 850 um JCMT observations with the filament orientation to determine which mechanisms dominate the formation of these filaments. We observe a large range of magnetic orientations in our observed filaments indicating the complex environments the filaments are located in. We also compare the observational results to synthetic data sets created using an MHD model of the GC. Our analysis reveals that the dominant mechanisms local to the HNCO filaments vary throughout the GC with some filaments being dominated by supersonic turbulence and others by subsonic turbulence. The comparison to synthetic observations indicates that the observed filaments are in magnetically dominated environments that could be supporting these filaments against collapse. Our results on the CMZ filaments are also compared to results obtained on similar filaments located in the Galactic Disk, and we find that the filaments studied here are possible CMZ analogs to the dense filamentary "bones" observed previously in the Galactic Disk.

astro-ph.GA↗

arXiv2025

Descrip3D: Enhancing Large Language Model-based 3D Scene Understanding with Object-Level Text Descriptions

Understanding 3D scenes goes beyond simply recognizing objects; it requires reasoning about the spatial and semantic relationships between them. Current 3D scene-language models often struggle with this relational understanding, particularly when visual embeddings alone do not adequately convey the roles and interactions of objects. In this paper, we introduce Descrip3D, a novel and powerful framework that explicitly encodes the relationships between objects using natural language. Unlike previous methods that rely only on 2D and 3D embeddings, Descrip3D enhances each object with a textual description that captures both its intrinsic attributes and contextual relationships. These relational cues are incorporated into the model through a dual-level integration: embedding fusion and prompt-level injection. This allows for unified reasoning across various tasks such as grounding, captioning, and question answering, all without the need for task-specific heads or additional supervision. When evaluated on five benchmark datasets, including ScanRefer, Multi3DRefer, ScanQA, SQA3D, and Scan2Cap, Descrip3D consistently outperforms strong baseline models, demonstrating the effectiveness of language-guided relational representation for understanding complex indoor scenes. Our code and data are publicly available at https://github.com/jintangxue/Descrip3D.

cs.CV↗

arXiv2025

Unveiling the Impact of Data and Model Scaling on High-Level Control for Humanoid Robots

Data scaling has long remained a critical bottleneck in robot learning. For humanoid robots, human videos and motion data are abundant and widely available, offering a free and large-scale data source. Besides, the semantics related to the motions enable modality alignment and high-level robot control learning. However, how to effectively mine raw video, extract robot-learnable representations, and leverage them for scalable learning remains an open problem. To address this, we introduce Humanoid-Union, a large-scale dataset generated through an autonomous pipeline, comprising over 260 hours of diverse, high-quality humanoid robot motion data with semantic annotations derived from human motion videos. The dataset can be further expanded via the same pipeline. Building on this data resource, we propose SCHUR, a scalable learning framework designed to explore the impact of large-scale data on high-level control in humanoid robots. Experimental results demonstrate that SCHUR achieves high robot motion generation quality and strong text-motion alignment under data and model scaling, with 37\% reconstruction improvement under MPJPE and 25\% alignment improvement under FID comparing with previous methods. Its effectiveness is further validated through deployment in real-world humanoid robot.

cs.RO↗

arXiv2025

Shuffle algebras and their integral forms: specialization map approach in types $C_n$ and $D_n$

We construct a family of PBWD bases for the positive subalgebras of quantum loop algebras of type $C_n$ and $D_n$, as well as their Lusztig and RTT integral forms, in the new Drinfeld realization. We also establish a shuffle algebra realization of these $\mathbb{Q}(v)$-algebras (proved earlier in arXiv:2102.11269 by completely different tools) and generalize the latter to the above $\mathbb{Z}[v,v^{-1}]$-forms. The rational counterparts provide shuffle algebra realizations of positive subalgebras of type $C_n$ and $D_n$ Yangians and their Drinfeld-Gavarini duals. While this naturally generalizes our earlier treatment of the classical type $B_n$ in arXiv:2305.00810 and $A_n$ in arXiv:1808.09536, the specialization maps in the present setup are more compelling.

math.QA↗

arXiv2025

EnerVerse: Envisioning Embodied Future Space for Robotics Manipulation

We introduce EnerVerse, a generative robotics foundation model that constructs and interprets embodied spaces. EnerVerse employs a chunk-wise autoregressive video diffusion framework to predict future embodied spaces from instructions, enhanced by a sparse context memory for long-term reasoning. To model the 3D robotics world, we adopt a multi-view video representation, providing rich perspectives to address challenges like motion ambiguity and 3D grounding. Additionally, EnerVerse-D, a data engine pipeline combining generative modeling with 4D Gaussian Splatting, forms a self-reinforcing data loop to reduce the sim-to-real gap. Leveraging these innovations, EnerVerse translates 4D world representations into physical actions via a policy head (EnerVerse-A), achieving state-of-the-art performance in both simulation and real-world tasks. For efficiency, EnerVerse-A reuses features from the first denoising step and predicts action chunks, achieving about 280 ms per 8-step action chunk on a single RTX 4090. Further video demos, dataset samples could be found in our project page.

cs.RO↗

arXiv2025

PhaseWin Search Framework Enable Efficient Object-Level Interpretation

Attribution is essential for interpreting object-level foundation models. Recent methods based on submodular subset selection have achieved high faithfulness, but their efficiency limitations hinder practical deployment in real-world scenarios. To address this, we propose PhaseWin, a novel phase-window search algorithm that enables faithful region attribution with near-linear complexity. PhaseWin replaces traditional quadratic-cost greedy selection with a phased coarse-to-fine search, combining adaptive pruning, windowed fine-grained selection, and dynamic supervision mechanisms to closely approximate greedy behavior while dramatically reducing model evaluations. Theoretically, PhaseWin retains near-greedy approximation guarantees under mild monotone submodular assumptions. Empirically, PhaseWin achieves over 95% of greedy attribution faithfulness using only 20% of the computational budget, and consistently outperforms other attribution baselines across object detection and visual grounding tasks with Grounding DINO and Florence-2. PhaseWin establishes a new state of the art in scalable, high-faithfulness attribution for object-level multimodal models.

cs.CV↗

arXiv2025

Genie Envisioner: A Unified World Foundation Platform for Robotic Manipulation

We introduce Genie Envisioner (GE), a unified world foundation platform for robotic manipulation that integrates policy learning, evaluation, and simulation within a single video-generative framework. At its core, GE-Base is a large-scale, instruction-conditioned video diffusion model that captures the spatial, temporal, and semantic dynamics of real-world robotic interactions in a structured latent space. Built upon this foundation, GE-Act maps latent representations to executable action trajectories through a lightweight, flow-matching decoder, enabling precise and generalizable policy inference across diverse embodiments with minimal supervision. To support scalable evaluation and training, GE-Sim serves as an action-conditioned neural simulator, producing high-fidelity rollouts for closed-loop policy development. The platform is further equipped with EWMBench, a standardized benchmark suite measuring visual fidelity, physical consistency, and instruction-action alignment. Together, these components establish Genie Envisioner as a scalable and practical foundation for instruction-driven, general-purpose embodied intelligence. All code, models, and benchmarks will be released publicly.

cs.RO↗

arXiv2025

Nonthermal Pressures: Key to Energy Balance and Structure Formation Near Sgr A* in the Milky Way

The circumnuclear region of the Galactic Center offers a unique laboratory to study energy balance and structure formation around Sgr A$\star$. This work investigates thermal and nonthermal processes within 7 pc distance from Sgr A$\star$. Using MeerKAT 1.3 GHz radio continuum data and ALMA H40 radio recombination line emission from the ACES survey, we separate free-free and synchrotron components at $\sim$0.2 pc resolution. With a thermal fraction of $\simeq$13%, the 1.3 GHz emission shows tight correlations with the Herschel PACS infrared data. The correlation between the equipartition magnetic field and molecular gas traced by JCMT $^{12}$CO (J=3$\rightarrow$2) observations reveals a balance between the magnetic field, cosmic rays, and molecular gas pressures south of the circumnuclear disk on $\sim$0.7 pc scales. Unlike the magnetic field and ionized gas, the molecular gas density declines in the cavity (R$\leq$2 pc) toward the center, likely due to feedback from Sgr A$\star$. We find that nonthermal pressure from turbulent gas nearly balances magnetic and cosmic ray pressures and exceeds thermal pressure by two orders of magnitude. The medium surrounding Sgr A$\star$ is filled by a low-$β$ (thermal-to-magnetic energy), supersonic plasma, with an Alfvén Mach number $\simeq$ 4 (assuming equipartition). Analysis of the mass-to-magnetic flux ratio suggests that the circumnuclear region is mostly subcritical and, therefore, the magnetic field can help stabilize gas clouds against gravitational collapse.

astro-ph.GA↗

arXiv2025

From Canada to Japan: How 10,000 km Affect User Perception in Robot Teleoperation

Robot teleoperation (RTo) has emerged as a viable alternative to local control, particularly when human intervention is still necessary. This research aims to study the distance effect on user perception in RTo, exploring the potential of teleoperated robots for older adult care. We propose an evaluation of non-expert users' perception of long-distance RTo, examining how their perception changes before and after interaction, as well as comparing it to that of locally operated robots. We have designed a specific protocol consisting of multiple questionnaires, along with a dedicated software architecture using the Robotics Operating System (ROS) and Unity. The results revealed no statistically significant differences between the local and remote robot conditions, suggesting that robots may be a viable alternative to traditional local control.

cs.RO↗

arXiv2025

Manual2Skill: Learning to Read Manuals and Acquire Robotic Skills for Furniture Assembly Using Vision-Language Models

Humans possess an extraordinary ability to understand and execute complex manipulation tasks by interpreting abstract instruction manuals. For robots, however, this capability remains a substantial challenge, as they cannot interpret abstract instructions and translate them into executable actions. In this paper, we present Manual2Skill, a novel framework that enables robots to perform complex assembly tasks guided by high-level manual instructions. Our approach leverages a Vision-Language Model (VLM) to extract structured information from instructional images and then uses this information to construct hierarchical assembly graphs. These graphs represent parts, subassemblies, and the relationships between them. To facilitate task execution, a pose estimation model predicts the relative 6D poses of components at each assembly step. At the same time, a motion planning module generates actionable sequences for real-world robotic implementation. We demonstrate the effectiveness of Manual2Skill by successfully assembling several real-world IKEA furniture items. This application highlights its ability to manage long-horizon manipulation tasks with both efficiency and precision, significantly enhancing the practicality of robot learning from instruction manuals. This work marks a step forward in advancing robotic systems capable of understanding and executing complex manipulation tasks in a manner akin to human capabilities.Project Page: https://owensun2004.github.io/Furniture-Assembly-Web/

cs.RO↗

arXiv2025

The JWST Galactic Center Survey -- A White Paper

The inner hundred parsecs of the Milky Way hosts the nearest supermassive black hole, largest reservoir of dense gas, greatest stellar density, hundreds of massive main and post main sequence stars, and the highest volume density of supernovae in the Galaxy. As the nearest environment in which it is possible to simultaneously observe many of the extreme processes shaping the Universe, it is one of the most well-studied regions in astrophysics. Due to its proximity, we can study the center of our Galaxy on scales down to a few hundred AU, a hundred times better than in similar Local Group galaxies and thousands of times better than in the nearest active galaxies. The Galactic Center (GC) is therefore of outstanding astrophysical interest. However, in spite of intense observational work over the past decades, there are still fundamental things unknown about the GC. JWST has the unique capability to provide us with the necessary, game-changing data. In this White Paper, we advocate for a JWST NIRCam survey that aims at solving central questions, that we have identified as a community: i) the 3D structure and kinematics of gas and stars; ii) ancient star formation and its relation with the overall history of the Milky Way, as well as recent star formation and its implications for the overall energetics of our galaxy's nucleus; and iii) the (non-)universality of star formation and the stellar initial mass function. We advocate for a large-area, multi-epoch, multi-wavelength NIRCam survey of the inner 100\,pc of the Galaxy in the form of a Treasury GO JWST Large Program that is open to the community. We describe how this survey will derive the physical and kinematic properties of ~10,000,000 stars, how this will solve the key unknowns and provide a valuable resource for the community with long-lasting legacy value.

astro-ph.GA↗

arXiv2025

MultiFoodhat: A potential new paradigm for intelligent food quality inspection

Food image classification plays a vital role in intelligent food quality inspection, dietary assessment, and automated monitoring. However, most existing supervised models rely heavily on large labeled datasets and exhibit limited generalization to unseen food categories. To overcome these challenges, this study introduces MultiFoodChat, a dialogue-driven multi-agent reasoning framework for zero-shot food recognition. The framework integrates vision-language models (VLMs) and large language models (LLMs) to enable collaborative reasoning through multi-round visual-textual dialogues. An Object Perception Token (OPT) captures fine-grained visual attributes, while an Interactive Reasoning Agent (IRA) dynamically interprets contextual cues to refine predictions. This multi-agent design allows flexible and human-like understanding of complex food scenes without additional training or manual annotations. Experiments on multiple public food datasets demonstrate that MultiFoodChat achieves superior recognition accuracy and interpretability compared with existing unsupervised and few-shot methods, highlighting its potential as a new paradigm for intelligent food quality inspection and analysis.

cs.CV↗