InternVLA-N1-System2

InternVLA-N1 Model Series

---

Model Description

InternVLA-N1 is a state-of-the-art navigation foundation model built on a multi-system design. Within this framework, it introduces a dual-system approach that joint trains the System 2 for high-level reasoning and System 1 for low-level action and control. This asynchronous architecture enables smooth, efficient, and robust instruction-following navigation in both simulated and real-world environments.

---

🔗 Resources

---

Key Features

• 🧩 Modular Multi-System Support
  Combines System 2 (reasoning/planning) with System 1 (action/control) in an asynchronous framework, delivering the first Dual-System Vision-Language Navigation (VLN) Foundation Model.

• 🚀 Zero-Shot Sim2Real Generalization
  Trained exclusively on simulation data (InternData-N1) while generalizing effectively to real-world deployments.

• 🏆 State-of-the-Art Performance
  Achieves leading results on multiple VLN benchmarks, including VLN-CE R2R/RxR and VLN-PE.

• ⚡ Asynchronous Inference
  Enables smooth execution and dynamic obstacle avoidance during navigation.

---

Model Variants

The previously released version is now called InternVLA-N1-wo-dagger. The lastest official release is recommended for best performance.

---

Usage

For inference, evaluation, and the Gradio demo, please refer to the InternNav repository.

---

Citation

If you find our work helpful, please consider starring this repository 🌟 and citing:

@misc{internvla-n1,
    title = {{InternVLA-N1: An} Open Dual-System Navigation Foundation Model with Learned Latent Plans},
    author = {InternVLA-N1 Team},
    year = {2025},
    booktitle={arXiv},
}
@misc{internnav2025,
    title = {{InternNav: InternRobotics'} open platform for building generalized navigation foundation models},
    author = {InternNav Contributors},
    howpublished={\url{https://github.com/InternRobotics/InternNav}},
    year = {2025}
}
@misc{wei2025groundslowfastdualsystem,
      title={Ground Slow, Move Fast: A Dual-System Foundation Model for Generalizable Vision-and-Language Navigation}, 
      author={Meng Wei and Chenyang Wan and Jiaqi Peng and Xiqian Yu and Yuqiang Yang and Delin Feng and Wenzhe Cai and Chenming Zhu and Tai Wang and Jiangmiao Pang and Xihui Liu},
      year={2025},
      eprint={2512.08186},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2512.08186}, 
}