---
license: apache-2.0
library_name: transformers
pipeline_tag: image-text-to-text
tags:
  - navigation
  - embodied-ai
  - vln
  - llava
  - qwen2
  - snav
  - navspace
base_model:
  - lmms-lab/LLaVA-Video-7B-Qwen2
---

# SNav-7B

**SNav-7B** is the navigation baseline from
[NavSpace: How Navigation Agents Follow Spatial Intelligence Instructions](https://arxiv.org/abs/2510.08173)
(ICRA 2026).

- **Architecture:** LLaVA-Video / Qwen2-7B + SigLIP (`LlavaQwenForCausalLM`)
- **Vision tower:** [`google/siglip-so400m-patch14-384`](https://huggingface.co/google/siglip-so400m-patch14-384)
- **Code & benchmark:** [TidalHarley/NavSpace](https://github.com/TidalHarley/NavSpace)
- **Project page:** [https://navspace.github.io/](https://navspace.github.io/)

## Quick start

```bash
# download
huggingface-cli download TidalYang/SNav-7B --local-dir ./SNav-7B

# evaluate on NavSpace (see NavSpace repo docs)
python evaluation/eval_snav.py \
  --model-path ./SNav-7B \
  --vision-tower-path /path/to/siglip-so400m-patch14-384 \
  --hm3d-base-path /path/to/hm3d_v0.2 \
  --task environment_state
```

Loading uses the LLaVA / StreamVLN-style stack from the NavSpace repository
(not plain `transformers.AutoModel` alone).

## Citation

```bibtex
@misc{yang2026navspacenavigationagentsfollow,
  title={NavSpace: How Navigation Agents Follow Spatial Intelligence Instructions},
  author={Haolin Yang and Yuxing Long and Zhuoyuan Yu and Zihan Yang and Minghan Wang and Jiapeng Xu and Yihan Wang and Ziyan Yu and Wenzhe Cai and Lei Kang and Hao Dong},
  year={2026},
  eprint={2510.08173},
  archivePrefix={arXiv},
  primaryClass={cs.RO},
  url={https://arxiv.org/abs/2510.08173}
}
```