README.md

---
language:
- en
license: other
license_name: uniwam-and-upstream-model-licenses
license_link: https://github.com/Fysics-AI/UniWAM/blob/main/THIRD_PARTY_NOTICES.md
pipeline_tag: robotics
inference: false
base_model: Wan-AI/Wan2.2-TI2V-5B
tags:
- robotics
- navigation
- world-action-model
- uniwam
- map-bench
---

<p align="center">
  <img src="assets/uniwam.png" width="280" alt="UniWAM" />
</p>
<p align="center">
  <a href="https://arxiv.org/abs/2609.39388">Paper</a> ·
  <a href="https://fysics-ai.github.io/uniwam.github.io/">Project</a> ·
  <a href="https://github.com/Fysics-AI/UniWAM">GitHub</a> ·
  <a href="https://huggingface.co/datasets/Fysics-AI/MAP-Bench">MAP-Bench</a>
</p>

**UniWAM** unifies navigation and manipulation through a shared world-action backbone and Manipulation Anchor Pose (MAP) supervision. This repository provides the checkpoints and shared components for navigation on **MAP-Bench**:

- **UniWAM-Base and UniWAM-Scaled**, with inference configurations and normalization files.
- **π₀.₅** navigation baseline checkpoint.
- Shared **Wan2.2 VAE, UMT5-XXL text encoder, and tokenizer**.

The released UniWAM checkpoints use **a single navigation expert and one forward RGB view**, since MAP-Bench evaluates navigation only. The full mobile manipulation architecture and training/inference code are available on [GitHub](https://github.com/Fysics-AI/UniWAM/blob/main/training/ROBOT.md).

## Download

Run from your workspace directory to download the complete model package (approximately **50 GB**):

```bash
python -m pip install -U huggingface_hub
hf download Fysics-AI/UniWAM --local-dir ./UniWAM-weights
```

This includes both UniWAM checkpoints, the π₀.₅ comparison checkpoint, and the shared components. For selective downloads, use the [GitHub download helper](https://github.com/Fysics-AI/UniWAM#checkpoints-and-data).

## Usage

See GitHub for [environment setup](https://github.com/Fysics-AI/UniWAM#installation-and-downloads), [inference and MAP-Bench evaluation](https://github.com/Fysics-AI/UniWAM#map-bench-evaluation), and [training](https://github.com/Fysics-AI/UniWAM/blob/main/training/README.md).

UniWAM weights use [Apache-2.0](LICENSE). Bundled third-party models retain their [upstream licenses](THIRD_PARTY_NOTICES.md).

## Citation

```bibtex
@article{xue2026uniwam,
  title   = {UniWAM Technical Report: Unified Mobile Manipulation via Mixed-Stream World-Action Modeling and Manipulation Anchor Pose Supervision},
  author  = {Wei Xue and Keliang Liu and Mingzhang Cui and Jinhua Xie and Jinjie Wei and Jianan Hou and Jingcheng Lu and Lintao Wang and Kaixiang Qiu and Yizhou Liu and Xinghai Ye and Jinghang Han and Mingcheng Li and Jie Gu and Shunli Wang and Lihua Zhang and Dingkang Yang},
  journal = {arXiv preprint arXiv:2609.39388},
  year    = {2026},
  url     = {https://arxiv.org/abs/2609.39388}
}
```