Kewei Zhang

Me at Kaifeng City

Hi there! My name is Kewei Zhang (张可为). I am currently a graduate student in Computer Science at Peking University, supervised by Prof. Daquan Zhou, and a member of DAGroup. Before joining PKU, I studied Computer Science at Wuhan University under the supervision of Prof. Zheng Wang.

My research interests lie in AIGC, Embodied AI, and Efficient AI. Additionally, I am passionate about AI-driven computer applications and have accumulated technical expertise in AI-related software development.

Visit my google scholar profile for most recent publications.

For any suggestions or collaborations, you are always welcome to reach out via email or .

news

Jan 26, 2026	MHLA is accepted by ICLR 2026.
Oct 01, 2025	I am honored to have received the DiDi Scholarship (Top 2%).
Jul 01, 2025	FAB-Attack is accepted by ACM MM 2025.
Jun 01, 2025	One paper is accepted by ICCV 2025.
Oct 01, 2024	One paper is accepted by NeurIPS 2024.
Sep 01, 2024	I am honored to have received the Lei Jun Computer Scholarship (Top 1%).
Oct 01, 2023	One paper is accepted by ACM MM 2023.
Sep 01, 2023	I am honored to have received the National Scholarship (Top 1%).

selected publications

arXiv

Open-HumanWAM: An Open Recipe for Scaling Human and Robot Experience for Embodied Foundation Models

Juncheng Ma, Jiaxin Li, Jiankai Tu, Bo Liang, Guodong Lv, Yufan Deng, Junzhe Tang, Liangzhu Yang, and 26 more authors

arXiv preprint arXiv:2611.06817, Nov 2026

arXiv Bib Code Website

@article{openhumanwam,
  title = {Open-HumanWAM: An Open Recipe for Scaling Human and Robot Experience for Embodied Foundation Models},
  author = {Ma, Juncheng and Li, Jiaxin and Tu, Jiankai and Liang, Bo and Lv, Guodong and Deng, Yufan and Tang, Junzhe and Yang, Liangzhu and Bian, Chen and Gong, Shukai and Zhang, Yintianrun and Cao, Zicong and Zhu, Baohao and Bi, Jianxin and Yan, Tengfei and Zhang, Yinyuan and Zhang, Kewei and Chen, Xinglong and Chen, Kaiqi and Wang, Duomin and Tang, Xiaotian and Tan, Liang and Li, Li and Xu, Yuan and Yang, Haosen and Wang, Yuqi and Xia, Jingfei and Huang, Songyan and Dou, Zhiyang and Dong, Zhen and Xie, Enze and Matusik, Wojciech and Chua, Tat-Seng and Zhou, Daquan},
  journal = {arXiv preprint arXiv:2611.06817},
  month = nov,
  year = {2026},
  primaryclass = {cs.RO},
}

arXiv

PyRUA-DataFarm: LLM Agent as a Cost-Efficient Teleoperator for Real-Ready Robot Data

Kewei Zhang, Juncheng Ma, Guodong Lv, Duomin Wang, Feixiang Ruan, Ruiyang Si, Chuanrui Zhang, Yuxuan Chen, and 6 more authors

arXiv preprint arXiv:2611.05320, Nov 2026

arXiv Bib Code

@article{pyruadatafarm,
  title = {{PyRUA-DataFarm}: {LLM} Agent as a Cost-Efficient Teleoperator for Real-Ready Robot Data},
  author = {Zhang, Kewei and Ma, Juncheng and Lv, Guodong and Wang, Duomin and Ruan, Feixiang and Si, Ruiyang and Zhang, Chuanrui and Chen, Yuxuan and Chen, Hao and Liu, Yichen and Zhang, Yintianrun and Deng, Yufan and Xie, Enze and Zhou, Daquan},
  journal = {arXiv preprint arXiv:2611.05320},
  month = nov,
  year = {2026},
  primaryclass = {cs.RO},
}

arXiv

Fewer Tokens, Better Action: GPT-6 Astra Robot Agents with 14% Higher Success Rate but 65% Fewer Tokens

Ruiyang Si, Jianxin Bi, Shunyu Yang, Rui Ni, Wenbo Huang, Qiang Wang, Shulong Jiang, Yintianrun Zhang, and 7 more authors

arXiv preprint arXiv:2610.01939, Oct 2026

arXiv Bib Code

@article{pyrua,
  title = {Fewer Tokens, Better Action: {GPT-6 Astra} Robot Agents with 14\% Higher Success Rate but 65\% Fewer Tokens},
  author = {Si, Ruiyang and Bi, Jianxin and Yang, Shunyu and Ni, Rui and Huang, Wenbo and Wang, Qiang and Jiang, Shulong and Zhang, Yintianrun and Ruan, Feixiang and Zhang, Kewei and Wang, Duomin and Li, Xiuyu and Feng, Haiwen and Dong, Zhen and Zhou, Daquan},
  journal = {arXiv preprint arXiv:2610.01939},
  month = oct,
  year = {2026},
  primaryclass = {cs.RO},
}

arXiv

Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving

Kewei Zhang^*, Jin Wang^*, Sensen Gao, Chengyue Wu, Yulong Cao, Songyang Han, Boris Ivanovic, Langechuan Liu, and 4 more authors

arXiv preprint arXiv:2605.23163, 2026

*Co-first author

arXiv Bib

@article{fastddrive,
  title = {Fast-dDrive: Efficient Block-Diffusion VLM for Autonomous Driving},
  author = {Zhang, Kewei and Wang, Jin and Gao, Sensen and Wu, Chengyue and Cao, Yulong and Han, Songyang and Ivanovic, Boris and Liu, Langechuan and Pavone, Marco and Han, Song and Zhou, Daquan and Xie, Enze},
  journal = {arXiv preprint arXiv:2605.23163},
  year = {2026},
  primaryclass = {cs.CL},
  note = {*Co-first author},
}

arXiv

HumanScale: Egocentric Human Video Can Outperform Real-Robot Data for Embodied Pretraining

Juncheng Ma, Jianxin Bi, Yufan Deng, Xuanran Zhai, Kewei Zhang, Ye Huang, Bo Liang, Shukai Gong, and 14 more authors

arXiv preprint arXiv:2606.20521, 2026

arXiv Bib

@article{humanscale,
  title = {HumanScale: Egocentric Human Video Can Outperform Real-Robot Data for Embodied Pretraining},
  author = {Ma, Juncheng and Bi, Jianxin and Deng, Yufan and Zhai, Xuanran and Zhang, Kewei and Huang, Ye and Liang, Bo and Gong, Shukai and Tu, Jiankai and Tang, Xiaotian and Li, Jiaxin and Chen, Kaiqi and Wang, Duomin and Wang, Yuqi and Kang, Bingyi and Huang, Eric and Dou, Zhiyang and Dong, Zhen and Xie, Enze and Matusik, Wojciech and Chua, Tat-Seng and Zhou, Daquan},
  journal = {arXiv preprint arXiv:2606.20521},
  year = {2026},
  primaryclass = {cs.CV},
}

ICLR

MHLA: Restoring Expressivity of Linear Attention via Token-Level Multi-Head

Kewei Zhang^*, Ye Huang^*, Yufan Deng, Jincheng Yu, Junsong Chen, Huan Ling, Enze Xie, and Daquan Zhou

In The Fourteenth International Conference on Learning Representations, 2026

*Co-first author

arXiv Bib Code Website

@inproceedings{mhla,
  title = {MHLA: Restoring Expressivity of Linear Attention via Token-Level Multi-Head},
  author = {Zhang, Kewei and Huang, Ye and Deng, Yufan and Yu, Jincheng and Chen, Junsong and Ling, Huan and Xie, Enze and Zhou, Daquan},
  booktitle = {The Fourteenth International Conference on Learning Representations},
  year = {2026},
  primaryclass = {cs.CV},
  note = {*Co-first author},
}

ACM MM

FAB-Attack: Fabric-Aware Adversarial Attacks on Person Detectors under Motion Blur

Jiaqi Hou^*, Kewei Zhang^*, Tianyu Yang, Chengyu Jia, Qiqi Lin, Hui Wei, and Zheng Wang

In Proceedings of the 33st ACM International Conference on Multimedia, 2025

Oral Presentation, *Co-first author

HTML
ICCV

Subjective Camera 1.0: Bridging Human Cognition and Visual Reconstruction through Sequence-Aware Sketch-Guided Diffusion

Haoyang Chen, Dongfang Sun, Caoyuan Ma, Shiqin Wang, Kewei Zhang, Zheng Wang, and Zhixiang Wang

In 2025 IEEE/CVF International Conference on Computer Vision, 2025

arXiv
NeurIPS

Revisiting adversarial patches for designing camera-agnostic attacks against person detection

Hui Wei^*, Zhixiang Wang^*, Kewei Zhang^*, Jiaqi Hou, Yuanwei Liu, Hao Tang, and Zheng Wang

In The Annual Conference on Neural Information Processing Systems, 2024

*Co-first author

Blog
ACM MM

Moiré backdoor attack (MBA): A novel trigger for pedestrian detectors in the physical world

Hui Wei^*, Hanxun Yu^*, Kewei Zhang, Zhixiang Wang, Jianke Zhu, and Zheng Wang

In Proceedings of the 31st ACM International Conference on Multimedia, 2023

*Co-first author

HTML