AstraNav-World: World Model for Foresight Control and Consistency

✨ Core Highlights

• 🧠 VLM-Wan Symbiosis: Combines Qwen2.5-VL reasoning with Wan2.2 physical priors to enable robust long-horizon world modeling.
• 🔄 Bidirectional Consistency: Jointly couples vision and action to eliminate cumulative errors and "physical hallucinations."
• 🌍 Zero-Shot Sim-to-Real: Direct deployment to physical robots without any real-world fine-tuning, powered by generalized spatial understanding.

Architecture

Consistency

🔥 Latest News!!

• January 21, 2026: We release the inference checkpoints of action former.
• December 29, 2025: We release the inference code of action former for the R2R, RxR, and OVON benchmarks.

Coming Soon

• training of action former
• inference of diffusion policy
• training of diffusion policy

Quickstart

🧰 Installation

Clone the repo:

git clone https://github.com/amap-cvlab/AstraNav-World.git

Install Training dependencies:

# Ensure torch >= 2.6.0
pip install -r requirements.txt
cd wan 
pip install .

Install habitat-sim and habitat-lab for inference

r2r&rxr and ovon have different version dependencies.
● habitat-sim
r2r&rxr: git clone https://github.com/facebookresearch/habitat-sim.git && cd habitat-sim && git checkout v0.1.7 
Ovon: git clone https://github.com/facebookresearch/habitat-sim.git && cd habitat-sim && git checkout v0.2.3
pip install -r requirements.txt
python setup.py install --headless
● habitat-lab
r2r&rxr: git clone https://github.com/facebookresearch/habitat-lab && cd habitat-lab && git checkout v0.1.7
ovon: git clone https://github.com/chongchong2025/habitat-lab && cd habitat-lab && git checkout v0.2.3_waypoint
python -m pip install -r habitat-baselines/habitat_baselines/rl/requirements.txt
python -m pip install -r habitat-baselines/habitat_baselines/rl/ddppo/requirements.txt
pip install -e .
cd habitat-baselines
pip install -e .

🎁 Model

⚡ Inference

● r2r & rxr

cd infer_r2r_rxr
bash eval_r2r.sh
bash eval_rxr.sh

● ovon

cd infer_ovon
bash eval_ovon.sh

🏛️ Citation

If you find this repository useful, please consider giving a star ⭐ and citation

@article{hu2025astranav,
  title={AstraNav-World: World Model for Foresight Control and Consistency},
  author={Hu, Junjun and Chen, Jintao and Bai, Haochen and Luo, Minghua and Xie, Shichao and Chen, Ziyi and Liu, Fei and Chu, Zedong and Xue, Xinda and Ren, Botao and others},
  journal={arXiv preprint arXiv:2512.21714},
  year={2025}
}
@article{xue2025omninav,
  title={OmniNav: A Unified Framework for Prospective Exploration and Visual-Language Navigation},
  author={Xue, Xinda and Hu, Junjun and Luo, Minghua and Shichao, Xie and Chen, Jintao and Xie, Zixun and Kuichen, Quan and Wei, Guo and Xu, Mu and Chu, Zedong},
  journal={arXiv preprint arXiv:2509.25687},
  year={2025}
}

Acknowledgments

Thanks to OmniNav, Navid, MTU3D, and Ovon for open-sourcing the construction of training data and the closed-loop inference code. Their contributions have significantly enriched the open-source community.