Gangwei Xu

Gangwei Xu | 许刚伟

I am a final-year Ph.D. student at Huazhong University of Science and Technology, advised by Prof. Xin Yang (expected to graduate in June 2027). As first author, I have published 11 papers in top-tier journals and conferences such as T-PAMI, NeurIPS, CVPR, and ICCV. My first-author papers have received over 1600 citations on Google Scholar. I was awarded funding from the National Natural Science Foundation of China PhD Student Research Program and was honored with the title of “Academic Star” at HUST.

Currently, I am a research intern at Robbyant, Ant Group, mentored by Yinghao Xu and Qihang Zhang, focusing on Physical AI and world models.

I am actively looking for full-time positions. Feel free to reach out!

News
[2026-07]We release LingBot-VA 2.0, a native video-action pretraining model for generalizable robot control.
[2026-06]We release Next Forcing, a world action model with faster convergence and more consistent physical dynamics.
[2025-09]Pixel-Perfect Depth accepted by NeurIPS 2025.
[2025-05]IGEV++ accepted by TPAMI 2025.
[2024-10]Honored as “Academic Star” (学术新星) at HUST.
[2024-05]Granted the National Natural Science Foundation of China (NSFC) PhD Student Research Program.
Selected Publications
LingBot-VA 2.0: Native Video-Action Pretraining for Generalizable Robot Control
Qihang Zhang, Lin Li, …, Gangwei Xu, …, Yujun Shen, Yinghao Xu
Tech Report, 2026
Project Paper

Native video-action pretraining that jointly learns future-frame prediction and action generation, enabling generalizable robot control.

Next Forcing: World Action Modeling with Multi-Chunk Prediction
Gangwei Xu, Qihang Zhang, Jiaming Zhou, Xing Zhu, Yujun Shen, Xin Yang, Yinghao Xu
arXiv Preprint, 2026
Project Paper

A multi-chunk prediction (MCP) framework for causal WAMs that improves training convergence speed, final accuracy, and inference efficiency via MCP-accelerated parallel chunk generation.

Pixel-Perfect Depth with Semantics-Prompted Diffusion Transformers
Gangwei Xu, Haotong Lin, Hongcheng Luo, Xianqi Wang, Jingfeng Yao, Lianghui Zhu, Yuechuan Pu, Cheng Chi, Haiyang Sun, Bing Wang, Guang Chen, Hangjun Ye, Sida Peng, Xin Yang
NeurIPS 2025
Project Code

A monocular depth foundation model with pixel-space diffusion transformers; estimated depth maps recover high-quality, flying-pixel-free point clouds.

IGEV
Iterative Geometry Encoding Volume for Stereo Matching
Gangwei Xu, Xianqi Wang, Xiaohuan Ding, Xin Yang
CVPR 2023
Paper Code

A new architecture combining the complementary advantages of filtering-based and optimization-based methods.

IGEV++
IGEV++: Iterative Multi-range Geometry Encoding Volumes for Stereo Matching
Gangwei Xu, Xianqi Wang, Zhaoxing Zhang, Junda Cheng, Chunyuan Liao, Xin Yang
TPAMI 2025
Paper Code

An effective and efficient method for handling large disparities and extensive ill-posed regions.

Fast-ACVNet
Accurate and Efficient Stereo Matching via Attention Concatenation Volume
Gangwei Xu, Yun Wang, Junda Cheng, Jinhui Tang, Xin Yang
TPAMI 2024
Paper Code

A novel cost volume representation (Fast-ACV) for real-time stereo matching.

All Publications
Experience
[01/2026 - now]Research Intern, Robbyant, Ant Group — video world models / embodied AI
[02/2025 - 12/2025]Research Intern, Xiaomi EV — depth foundation models
[05/2023 - 12/2023]Research Intern, Shanghai AI Laboratory — computational photography
Honors & Awards
Academic Services

Conference Reviewer: CVPR 2023, ICCV 2023, SIGGRAPH Asia 2023, CVPR 2024, ECCV 2024, ICRA 2024, ICLR 2025, AAAI 2025, CVPR 2025, ICCV 2025, NeurIPS 2025, AAAI 2026, ICLR 2026, CVPR 2026.

Journal Reviewer: TPAMI, IJCV, TIP, RA-L.