Researcher

Haoyi Zhu

Also known as 朱皓怡

Haoyi Zhu is an AI researcher whose work spans pose estimation, embodied agents and video world models. He earned a B.S. in artificial intelligence at Shanghai Jiao Tong University in 2023. He coauthored MineDojo, a Minecraft framework for open-ended embodied agents, and AlphaPose, a real-time whole-body pose system; MineDojo won the 2022 NeurIPS Outstanding Paper Award. Aether won an Outstanding Paper Award at the ICCV RIWM workshop in 2025, and he coauthored SANA-WM for minute-scale 720p video generation in 2026. Since July 2026, he is a research intern at Alibaba ATH.

Now at
Alibaba
Badges

Timeline

  1. 2019–2023

    AI studies and pose estimation

    In 2020, he coauthors Multi-Person Full Body Pose Estimation, which reports 51.5 mAP on its manually annotated validation set. He earns a B.S. degree in Artificial Intelligence Honor Class at Shanghai Jiao Tong University in 2023, advised by Prof. Cewu Lu.

  2. February 2022–February 2023

    Research at NVIDIA GEAR

    From February 2022 to February 2023, he works as a research intern with Jim Fan's Team (GEAR Lab) at NVIDIA.

  3. 2022

    MineDojo wins at NeurIPS

    He coauthors MineDojo, a Minecraft framework with thousands of open-ended tasks and an internet-scale knowledge base. The paper wins the 2022 NeurIPS Outstanding Paper Award.

  4. 2022

    AlphaPose tracks whole bodies

    In 2022, he coauthors AlphaPose, a real-time system for whole-body pose estimation and multi-person tracking.

  5. September 2023–November 2025

    USTC PhD and Shanghai AI Lab

    In September 2023, he begins a PhD in Computer Science at USTC, advised by Prof. Tong He and Prof. Chunhua Shen, and joins Shanghai AI Lab as a research intern. His Shanghai AI Lab internship runs through November 2025.

  6. 2025

    Aether earns an ICCV award

    He coauthors Aether, a framework for 4D reconstruction, action-conditioned video prediction, and goal-conditioned visual planning. The paper wins an Outstanding Paper Award at the ICCV 2025 workshop on Reliable and Interactive World Models (RIWM).

  7. December 2025–July 2026

    NVIDIA research and SANA-WM

    From December 2025 to July 2026, he works as a research intern with NVIDIA's SANA team in the Efficient AI Lab. In May 2026, he coauthors SANA-WM, an open-source world model for one-minute 720p video generation with precise camera control.

  8. Since July 2026

    Research intern at Alibaba ATH

    Since July 2026, he is a research intern at Alibaba ATH, advised by Dr. Xinlong Wang.

Career

  1. Since 2026AlibabaResearch Intern
  2. 2023–2025Shanghai Artificial Intelligence LaboratoryResearch Intern
  3. 2022–2026NVIDIAResearch Intern

Education

  1. 2023Shanghai Jiao Tong UniversityB.S.
  2. From 2023University of Science and Technology of ChinaPhD

Sources and credits

Photo: We have no published photo of Haoyi Zhu yet. Send us a free-licensed photo.

  1. Haoyi Zhu haoyizhu.site. Checked 21 October 2026.
  2. Multi-Person Full Body Pose Estimation arxiv.org. Checked 21 October 2026.
  3. NeurIPS 2022 Awards neurips.cc. Checked 21 October 2026.
  4. AlphaPose: Whole-Body Regional Multi-Person Pose Estimation and Tracking in Real-Time arxiv.org. Checked 21 October 2026.
  5. Aether: Geometric-Aware Unified World Modeling arxiv.org. Checked 21 October 2026.
  6. SANA-WM: Efficient Minute-Scale World Modeling with Hybrid Linear Diffusion Transformer arxiv.org. Checked 21 October 2026.