UniMapGen: A Generative Framework
for Large-Scale Map Construction
from Multi-modal Data

🎉🎉AAAI 2026 Oral🎉🎉

Yujian Yuan^1,2,*, Changjie Wu^1,*, Xinyuan Chang^1,*, Sijin Wang^1,*, Hang Zhang¹, Shiyi Liang^1,3, Shuang Zeng^1,3, Mu Xu^1,†,

¹Amap, Alibaba Group, ²The Hong Kong University of Science and Technology,
³Xi’an Jiaotong University

^{*Equal Contribution †Corresponding author}

UniMapGen is a generative unified framework that autoregressively generates smooth and topologically consistent vectorized maps from multi-modal inputs, enabling scalable, occlusion-robust city-scale mapping without costly on-site data collection.

Release Plan

• Inference code
• Visualization code
• Dataprocess codes
• UniMapGen checkpoints
• UniMapGen code

📜 Citation

If you find UniMapGen is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry:

@article{yuan2025unimapgen,
  title={UniMapGen: A Generative Framework for Large-Scale Map Construction from Multi-modal Data},
  author={Yuan, Yujian and Wu, Changjie and Chang, Xinyuan and Wang, Sijin and Zhang, Hang and Liang, Shiyi and Zeng, Shuang and Xu, Mu},
  journal={arXiv preprint arXiv:2509.22262},
  year={2025}
}

🙏 Acknowledgement

Our work is primarily based on LLaMA-Factory and our dataset comes from OpenSatMap. We are sincerely grateful for their work.