arXiv subjects

Yuhao Pan

Publications and source records attributed to Yuhao Pan.

13 recordsLinked to original sources

arXiv2026

World-to-Wrist: Task-Conditioned Future Wrist Modeling for Fine-Grained Robot Manipulation

Vision-language-action (VLA) models often treat main-view and wrist-view observations as parallel visual inputs, overlooking their distinct roles in robot manipulation. Fine-grained manipulation, however, benefits from anticipating how wrist-local interactions may evolve under the global task context. To address this limitation, we present World-to-Wrist VLA (W2-VLA), a VLA model for fine-grained robot manipulation with task-conditioned future wrist modeling. Given current multi-view observations and a task instruction, W2-VLA contextualizes a set of latent modeling tokens as a compact interface between the vision-language model and the wrist predictor. Conditioned on this interface and the observed wrist history, the predictor forecasts future wrist latents, which are transformed into future-aware context for action prediction. In addition, we introduce W2-CoT, a synthesis pipeline that produces structured annotations describing manipulation progress, physical transition cues, and wrist-local evidence. These annotations provide auxiliary supervision that shapes the task-conditioned latent interface. Experiments on LIBERO, RoboTwin 2.0, and real-world manipulation tasks demonstrate improved fine-grained and contact-sensitive manipulation across both single-arm and bimanual settings, while maintaining action-generation rates above 80 Hz.

cs.RO↗

arXiv2026

SpatialBench: Is Your Spatial Foundation Model an All-Round Player?

While spatial foundation models have demonstrated impressive performance on standard datasets, a critical question remains: are they truly all-round players capable of generalizing robustly across diverse downstream tasks, arbitrary viewpoints, shifting scene domains, varying input densities, and specific hardware constraints? Answering this overarching question requires a holistic assessment, yet current models are mainly evaluated on specific domains for which they were specifically designed or trained. Such evaluations are intrinsically limited by narrow paradigm coverage, limited scene domains, and arbitrary frame sampling, making it fundamentally difficult to assess their true generalization capabilities. To address this gap, we present SpatialBench, a cross-paradigm, domain-diverse benchmark for spatial foundation models with deterministic sampling. SpatialBench features unprecedented scale and rigorous deterministic design, comprising 19 datasets and 546 scenes across 5 diverse spatial domains. It comprehensively evaluates 41 models across 6 paradigms on 5 task suites under 4 different input density settings. Our extensive evaluation reveals that current models are not yet all-round players, and uncovers crucial insights for future advancement. Specifically, we demonstrate that full-context attention maximizes accuracy while bounded-memory strategies unlock long-sequence scalability. Moreover, our empirical evaluations in challenging embodied and egocentric tasks demonstrate that strict domain alignment and high data quality are far more critical to performance than simple dataset scaling. Furthermore, to address the largest data gap identified in our analysis, we go beyond evaluation by introducing a large-scale dataset, DA-Next-5M, and a strong baseline model, DA-Next, pushing the boundaries of spatial representation learning.

cs.CV↗

arXiv2026

PrismWF: A Multi-Granularity Patch-Based Transformer for Robust Website Fingerprinting Attack

Tor is a low-latency anonymous communication network that protects user privacy by encrypting website traffic. However, recent website fingerprinting (WF) attacks have shown that encrypted traffic can still leak users' visited websites by exploiting statistical features such as packet size, direction, and inter-arrival time. Most existing WF attacks formulate the problem as a single-tab classification task, which significantly limits their effectiveness in realistic browsing scenarios where users access multiple websites concurrently, resulting in mixed traffic traces. To this end, we propose PrismWF, a multi-granularity patch-based Transformer for multi-tab WF attack. Specifically, we design a robust traffic feature representation for raw web traffic traces and extract multi-granularity features using convolutional kernels with different receptive fields. To effectively integrate information across temporal scales, the proposed model refines features through three hierarchical interaction mechanisms: inter-granularity detail supplementation from fine to coarse granularities, intra-granularity patch interaction with dedicated router tokens, and router-guided dual-level intra- and cross-granularity fusion. This design aligns with the cognitive logic of global coarse-grained reconnaissance and local fine-grained querying, enabling effective modeling of mixed traffic patterns in WF attack scenarios. Extensive experiments on various datasets and WF defenses demonstrate that our method achieves state-of-the-art performance compared to existing baselines.

cs.CR↗

arXiv2026

A Tutorial on Learning-Based Radio Map Construction: Data, Paradigms, and Physics-Awareness

Radio maps (RMs) provide the digital representation of the wireless propagation environment, mapping complex geographical and topological boundary conditions to critical spatial-spectral metrics that range from received signal strength to full channel state information matrices. The integration of artificial intelligence into next generation wireless networks further necessitates the accurate construction of RMs as a foundational prerequisite for electromagnetic digital twins. This paper presents a comprehensive survey of learning-based RM construction, systematically addressing three intertwined dimensions: data, paradigms, and physics-awareness. From the data perspective, we review physical measurement campaigns, ray tracing simulation engines, and publicly available benchmark datasets, identifying their respective strengths and fundamental limitations. From the paradigm perspective, we establish a core taxonomy that categorizes RM construction into source-aware forward prediction and source agnostic inverse reconstruction, and examine five principal neural architecture families spanning convolutional neural networks, vision transformers, graph neural networks, generative adversarial networks, and diffusion models. We further survey optics-inspired methods adapted from neural radiance fields and 3D Gaussian splatting for continuous wireless radiation field modeling. From the physics-awareness perspective, we introduce a three-level integration framework encompassing data-level feature engineering, loss-level partial differential equation regularization, and architecture level structural isomorphism. Open challenges including foundation model development, physical hallucination detection, and mortized inference for real-time deployment are discussed to outline future research directions. The project page is at https://github.com/UNIC-Lab/Awesome-Radio-Map-Categorized.

eess.SY↗

arXiv2025

Cross-Receiver Generalization for RF Fingerprint Identification via Feature Disentanglement and Adversarial Training

Radio frequency fingerprint identification (RFFI) is a key technique for wireless network security, leveraging intrinsic hardware imperfections to enable transmitter identification. Although deep neural networks are effective at extracting discriminative RF features, their performance is significantly affected by receiver-induced variability in practical deployments. In real-world scenarios, RF signals inherently entangle transmitter-specific characteristics with receiver-dependent distortions, leading models to capture receiver-related patterns when training and evaluation are conducted on the same device. Consequently, replacing the receiver during deployment often results in notable performance degradation. To address this issue, we propose a cross-receiver robust RFFI framework that explicitly disentangles transmitter-specific and receiver-specific representations. The proposed method integrates adversarial domain alignment with receiver-aware regularization to suppress residual receiver information in transmitter features while enforcing intra-receiver consistency in receiver-specific representations. A feature separation constraint is further introduced to decouple the two components in the latent space. Extensive experiments on multi-receiver WiFi datasets demonstrate that the proposed method consistently outperforms state-of-the-art baselines under cross-receiver evaluation and significantly improves robustness to receiver replacement.

cs.LG↗

arXiv2024

Interlayer coupling rotatable magnetic easy-axis in MnSe2 mono- and bi-layers

Interlayer coupling plays a critical role in tuning the electronic structures and magnetic ground states of two-dimensional materials, influenced by the number of layers, interlayer distances, and stacking order. However, its effect on the orientation of the magnetic easy axis remains underexplored. In this study, we demonstrate that interlayer coupling can significantly alter the magnetic easy-axis orientation, as shown by the magnetic easy-axis of monolayer 1T-MnSe2 tilting 67{\deg} from the z-axis, while aligning with the z-axis in the bilayer. This change results from variations in orbital occupations near the Fermi level, particularly involving nonmetallic Se atoms. Contrary to the traditional focus on magnetic metal atoms, our findings reveal that Se orbitals play a key role in influencing the easy-axis orientation and topological Chern numbers. Furthermore, we validated our conclusions by changing stacking orders, introducing charge doping, applying in-plane biaxial strains, and substituting non-metallic atoms. Our results highlight the pivotal role of interlayer coupling in tuning the magnetic properties of layered materials, with important implications for spintronic applications.

cond-mat.mtrl-sci↗

arXiv2024

GNN-Empowered Effective Partial Observation MARL Method for AoI Management in Multi-UAV Network

Unmanned Aerial Vehicles (UAVs), due to their low cost and high flexibility, have been widely used in various scenarios to enhance network performance. However, the optimization of UAV trajectories in unknown areas or areas without sufficient prior information, still faces challenges related to poor planning performance and low distributed execution. These challenges arise when UAVs rely solely on their own observation information and the information from other UAVs within their communicable range, without access to global information. To address these challenges, this paper proposes the Qedgix framework, which combines graph neural networks (GNNs) and the QMIX algorithm to achieve distributed optimization of the Age of Information (AoI) for users in unknown scenarios. The framework utilizes GNNs to extract information from UAVs, users within the observable range, and other UAVs within the communicable range, thereby enabling effective UAV trajectory planning. Due to the discretization and temporal features of AoI indicators, the Qedgix framework employs QMIX to optimize distributed partially observable Markov decision processes (Dec-POMDP) based on centralized training and distributed execution (CTDE) with respect to mean AoI values of users. By modeling the UAV network optimization problem in terms of AoI and applying the Kolmogorov-Arnold representation theorem, the Qedgix framework achieves efficient neural network training through parameter sharing based on permutation invariance. Simulation results demonstrate that the proposed algorithm significantly improves convergence speed while reducing the mean AoI values of users. The code is available at https://github.com/UNIC-Lab/Qedgix.

cs.IT↗

arXiv2024

Trapezoidal Gradient Descent for Effective Reinforcement Learning in Spiking Networks

With the rapid development of artificial intelligence technology, the field of reinforcement learning has continuously achieved breakthroughs in both theory and practice. However, traditional reinforcement learning algorithms often entail high energy consumption during interactions with the environment. Spiking Neural Network (SNN), with their low energy consumption characteristics and performance comparable to deep neural networks, have garnered widespread attention. To reduce the energy consumption of practical applications of reinforcement learning, researchers have successively proposed the Pop-SAN and MDC-SAN algorithms. Nonetheless, these algorithms use rectangular functions to approximate the spike network during the training process, resulting in low sensitivity, thus indicating room for improvement in the training effectiveness of SNN. Based on this, we propose a trapezoidal approximation gradient method to replace the spike network, which not only preserves the original stable learning state but also enhances the model's adaptability and response sensitivity under various signal dynamics. Simulation results show that the improved algorithm, using the trapezoidal approximation gradient to replace the spike network, achieves better convergence speed and performance compared to the original algorithm and demonstrates good training stability.

cs.AI↗

arXiv2020

Local probe of the interlayer coupling strength of few-layers SnSe by contact-resonance atomic force microscopy

The interlayer bonding in two dimensional materials is particularly important because it is not only related to their physical and chemical stability but also affects their mechanical, thermal, electronic, optical, and other properties. To address this issue, we report the direct characterization of the interlayer bonding in 2D SnSe using contact-resonance atomic force microscopy in this study. Site specific CR spectroscopy and CR force spectroscopy measurements are performed on both SnSe and its supporting SiO2 substrate comparatively. Based on the cantilever and contact mechanic models, the contact stiffness and vertical Young's modulus are evaluated in comparison with SiO2 as a reference material. The interlayer bonding of SnSe is further analyzed in combination with the semi-analytical model and density functional theory calculations. The direct characterization of interlayer interactions using this nondestructive methodology of CR AFM would facilitate a better understanding of the physical and chemical properties of 2D layered materials, specifically for interlayer intercalation and vertical heterostructures.

cond-mat.mtrl-sci↗

arXiv2019

Bethe-Slater-curve-like behavior and interlayer spin-exchange coupling mechanisms in two-dimensional magnetic bilayers

Layered magnets have recently received tremendous attention, however, spin-exchange coupling mechanism across their interlayer regions is yet to be revealed. Here, we report a Bethe-Slater-curve (BSC) like behavior in nine transition metal dichalcogenide bilayers (MX2, M=V, Cr, Mn; X=S, Se, Te) and established interlayer spin-exchange coupling mechanisms at their van der Waals gaps using first-principle calculations. The BSC-like behavior offers a distance-dependent interlayer anti-ferromagnetic (AFM) to ferromagnetic (FM) transition. This phenomenon is explained with the spin-exchange coupling mechanisms established using bilayer CrSe2 as a prototype in this work. The Se pz wavefunctions from two adjacent interfacial Se sublayers overlap at the interlayer region. The spin alignment of the region determines interlayer magnetic coupling. At a shorter interlayer distance, Pauli repulsion at the overlapped region dominates and thus favors anti-parallel oriented spins leading to interlayer AFM. For a longer distance, kinetic energy gain of polarized electrons across the bilayer balances the Pauli repulsion and the bilayer thus prefers an interlayer FM state. In light of this, the AFM-FM transition is a result of competition between Pauli and Coulomb repulsion and kinetic energy gain. All these results open a new route to tune interlayer magnetism and the revealed spin-exchange coupling mechanisms are paramount additions to those previously established ones.

cond-mat.mtrl-sci↗

arXiv2018

Charge tunable structural phase transitions in few-layer tellurium: a step toward building mono-elemental devices

Recently, a covalent-like quasi-bonding was unveiled for the inter-chain interaction in a promising semiconductor, few-layer Tellurium. Such quasi-bond offers comparable bond lengths and strengths with those of a typical Te-Te covalent bond, which may lead to much easier transformations between the quasi-bonds and the covalent bonds. Here, we show a few structural phase-transitions among four Te allotropes in few-layer Te under charge doping. In particular, the semiconducting {\alpha}-phase can transform into a smaller-bandgap {\beta}-phase, even-smaller-bandgap {\gamma}-phase and then metallic {\delta}-phase in a Te bilayer. In a tri-layer, a metallic chiral {\alpha}+{\delta} phase is more stable under initial electron doping, leading to the appearance of chirality. Variations of electronic structures aside, these transitions are accompanied by the emergence or elimination of inversion centers ({\alpha}-{\beta}, {\alpha}-{\gamma}, {\alpha}-{\alpha}+{\delta}), structural anisotropy ({\alpha}-{\gamma}, {\gamma}-{\delta}) and chirality ({\alpha}-{\alpha}+{\delta}), which could result in substantial changes in optical and other properties. In light of this, this work opens the possibility toward building mono-elemental electronic and optoelectronic heterostructures or devices. It also offers a platform for studying relations between charge doping and electric/optical properties.

cond-mat.mtrl-sci↗

arXiv2017

Few-layer Tellurium: one-dimensional-like layered elementary semiconductor with striking physical properties

Few-layer Tellurium, an elementary semiconductor, succeeds most of striking physical properties that black phosphorus (BP) offers and could be feasibly synthesized by simple solution-based methods. It is comprised of non-covalently bound parallel Te chains, among which covalent-like feature appears. This feature is, we believe, another demonstration of the previously found covalent-like quasi-bonding (CLQB) where wavefunction hybridization does occur. The strength of this inter-chain CLQB is comparable with that of intra-chain covalent bonding, leading to closed stability of several Te allotropes. It also introduces a tunable bandgap varying from nearly direct 0.31 eV (bulk) to indirect 1.17 eV (2L) and four (two) complex, highly anisotropic and layer-dependent hole (electron) pockets in the first Brillouin zone. It also exhibits an extraordinarily high hole mobility (~10$^5$ cm$^2$/Vs) and strong optical absorption along the non-covalently bound direction, nearly isotropic and layer-dependent optical properties, large ideal strength over 20%, better environmental stability than BP and unusual crossover of force constants for interlayer shear and breathing modes. All these results manifest that the few-layer Te is an extraordinary-high-mobility, high optical absorption, intrinsic-anisotropy, low-cost-fabrication, tunable bandgap, better environmental stability and nearly direct bandgap semiconductor. This "one-dimension-like" few-layer Te, together with other geometrically similar layered materials, may promote the emergence of a new family of layered materials.

cond-mat.mtrl-sci↗

arXiv2017

Layer- and doping-tunable long range ferromagnetic order-ing in two-dimensional CrS2

Interlayer coupling is of vital importance for manipulating physical properties, e.g. electronic bandgap, in two-dimensional materials. However, tuning magnetic proper-ties in these materials is yet to be addressed. Here, we found a striped antiferromag-netic (sAFM) to ferromagnetic (FM) transition undergoing from monolayer to bilayer and thicker CrS2. This transition is attributed to charge sharing of interlayer S atoms and its resulting charge transfer from Cr d eg to t2g orbitals. The transferred charge reduces a portion of Cr4+ to Cr3+, which enhances the double-exchange mechanism favoring FM rather than sAFM ordering. Doping of electrons shares the same effect with stacking induced charge sharing. Therefore, charge doping could either manip-ulate the magnetic ordering between sAFM and FM or tune CrS2 between p- and n-doped magnetic semiconductors. We also proposed several prototype devices ena-bling to use external electric field for manipulating magnetic orderings of CrS2 layers. These results manifest the role of interlayer coupling in modifying magnetic proper-ties of layered materials

cond-mat.mtrl-sci↗