GroundingPI is a 4B visual grounding foundation model developed by XPeng and academic researchers. It represents locations as quantized points and boxes, supports detection, referring and OCR, and releases code and weights with simulation and open-loop transfer evaluations.

Overview

Model type
Vision-language model
Company
Xpeng Robotics · Tsinghua University · The Hong Kong University of Science and Technology (Guangzhou) · National University of Singapore · The University of Hong Kong · UC Berkeley
Recorded versions
GroundingPI
Listed versions
GroundingPI

Papers & reports (1)

GroundingPI: A Grounding Foundation Model towards Physical Intelligence with Visual Primitives

Paper available

arXiv: 2609.39601DOI: 10.48550/arxiv.2609.39601
Authors & affiliations (26)

Author order: Original order

  1. Qize Yu

    XPeng Inc. · Peking University

  2. Lianrui Fan

    XPeng Inc.

  3. Boyu Chen

    XPeng Inc. · The University of Hong Kong · Tsinghua University

  4. Jiaqi Liang

    Peking University

  5. Xini Ding

    XPeng Inc.

  6. Yue Chen

    Peking University

  7. Zetian Song

    XPeng Inc. · Peking University

  8. Yuran Wang

    Peking University · National University of Singapore

  9. Yi Zou

    XPeng Inc.

  10. Kaixuan Wang

    The University of Hong Kong

  11. Tianxing Chen

    The University of Hong Kong

  12. Wenxuan Song

    HKUST (GZ)

  13. Bohan Zhou

    Peking University

  14. Mingleyang Li

    Peking University

  15. Siqiao Huang

    Tsinghua University

  16. Yuqi Ye

    Peking University

  17. Caigao Jiang

    XPeng Inc.

  18. Wei Wei

    XPeng Inc.

  19. Ruihai Wu

    University of California, Berkeley

  20. Hang Zhang

    XPeng Inc.

  21. Yixiao Ge

    XPeng Inc.

  22. Shuchang Zhou

    XPeng Inc.

  23. Shilong Liu

    Princeton University

  24. Xianming Liu

    XPeng Inc.

  25. Ping Luo

    The University of Hong Kong

  26. Shiyu Huang

    XPeng Inc.

First published
2026-09-30
Latest paper revision
v2 · 2026-10-16
Venue
Not provided

Covered model versions

Paper ↗PDF ↗
Submit a source or correction