Yuhao Pan

News

• 2026.10Our paper PrismWF was accepted by IEEE Transactions on Dependable and Secure Computing (TDSC).
• 2026.09Our paper SpatialBench was accepted to NeurIPS 2026.
• 2026.09Our paper Cross-Receiver Generalization for RF Fingerprint Identification was accepted by IEEE Transactions on Cognitive Communications and Networking (TCCN).
• 2026.08We released W²-VLA, a world-to-wrist VLA model for fine-grained robot manipulation.
• 2026.05We released SpatialBench, a cross-paradigm benchmark for spatial foundation models.
• 2026.03We released PrismWF, a multi-granularity method for robust website fingerprinting attacks.
• 2026.03Our tutorial on learning-based radio map construction is available on arXiv.

Publications

Selected papers. * denotes equal contribution.

Embodied AI

Preprint · 2026

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

Yuhao Pan*, Haosong Peng*, Zhengshen Zhang, Zhengyang Yan, Yalun Dai, Fushuo Huo, Chujie Wang, Tianyu Qi, Xiucheng Wang, Nan Cheng, Wenchao Xu

World-to-Wrist introduces task-conditioned future wrist modeling for multi-view vision-language-action systems. By anticipating how the robot’s wrist-camera observations will evolve before actions are executed, the model strengthens spatial grounding, long-horizon reasoning, and fine-grained manipulation in both simulation and real-world tasks.

Paper ↗Code ↗Project ↗

NeurIPS · 2026

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

Haosong Peng*, Hao Li*, Jiaqi Chen*, Yuhao Pan*, Runmao Yao, Yalun Dai, Fushuo Huo, Fangzhou Hong, Zhaoxi Chen, Haozhao Wang, Dingwen Zhang, Ziwei Liu, Wenchao Xu

SpatialBench provides a unified evaluation of spatial foundation models across 19 datasets, 546 scenes, six model paradigms, and five application domains. Its deterministic multi-density protocol enables reproducible comparisons and exposes practical gaps in long-sequence memory, out-of-domain generalization, and training-data quality.

Paper ↗Code ↗

Wireless Communication & Network Security

IEEE TDSC · 2026

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

Yuhao Pan, Wenchao Xu, Fushuo Huo, Haozhao Wang, Xiucheng Wang, Nan Cheng

PrismWF addresses multi-tab website fingerprinting by learning traffic patterns at multiple temporal granularities. Convolutional branches extract features with different receptive fields, while hierarchical interactions pass fine-grained details to coarser representations and use router tokens to exchange information within and across granularities. This design combines local traffic details with broader patterns to identify websites in mixed traces, including under website fingerprinting defenses.

Paper ↗Code ↗

IEEE TCCN · 2026

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

Yuhao Pan, Xiucheng Wang, Fushuo Huo, Nan Cheng, Wenchao Xu

This work improves RF fingerprint identification when training and deployment use different receivers. The proposed framework disentangles transmitter and receiver characteristics and applies adversarial training to suppress receiver-specific artifacts while preserving identity cues that generalize to previously unseen hardware.

Paper ↗

Preprint · 2026

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

Xiucheng Wang, Yuhao Pan, Nan Cheng, Çağkan Yapar, Ruijin Sun, Zhisheng Yin, Conghao Zhou, Wenchao Xu, Yuxiang Zhang, Jianhua Zhang, Shuguang Cui, Xuemin Shen

This tutorial organizes learning-based radio map construction into source-aware forward prediction and source-agnostic inverse reconstruction. It reviews representative datasets and learning paradigms, then explains how propagation physics and electromagnetic knowledge can improve accuracy, data efficiency, and generalization.

Paper ↗Code ↗

IEEE Internet of Things Journal · 2024

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

Yuhao Pan, Xiucheng Wang, Zhiyao Xu, Nan Cheng, Wenchao Xu, Jun-jie Zhang

This paper models dynamic relationships among UAVs with graph neural networks and integrates the learned representation into a multi-agent reinforcement learning framework. The approach enables distributed coordination under partial observations and improves age-of-information management as network topology and information freshness change over time.

Paper ↗Code ↗

Journal of Information and Intelligence · 2026

CRAFT: Complexity-Routed Attention with Frequency-Temporal Fusion for Multi-Tab Website Fingerprinting

Yuhao Pan, Fushuo Huo, Haosong Peng, Kang Wei

CRAFT addresses variations in traffic overlap during multi-tab browsing. It combines temporal features with frequency-domain patterns through learned gating, then uses a lightweight router to estimate the mixing complexity of each token and adapt its attention range. The model focuses on nearby context in simpler segments and expands its scope in heavily overlapping regions, capturing both traffic rhythms and changing dependencies.

Paper ↗

Academic Service