README.md

---
license: cc-by-nc-4.0
language:
- en
base_model: Wan-AI/Wan2.2-TI2V-5B
base_model_relation: finetune
tags:
- arxiv:2610.25690
- robotics
- navigation
- video-diffusion
- image-text-to-action
- safetensors
library_name: scalenav
pipeline_tag: robotics
widget:
- example_title: 'ScaleNav: full navigation demo'
  output:
    url: replay.mp4
datasets:
- Yangyihui/ScaleNav-Data
---

# ScaleNav

**Scalable Internet Video Pretraining for Language-Conditioned Local Navigation**

[Paper](https://arxiv.org/abs/2610.25690) · [Code](https://github.com/MachEmbodied/ScaleNav) · [Data](https://huggingface.co/datasets/Yangyihui/ScaleNav-Data) · [Project page](https://zju3dv.github.io/ScaleNav/) · [Collection](https://huggingface.co/collections/Yangyihui/scalenav-6ad591e8ff62bb75ec75f645)

ScaleNav learns navigation priors through action-free pretraining on Internet
videos, then turns an RGB observation and a language instruction into robot actions.
It navigates to a target visible in the initial view and stops at the instructed location.

**5B parameters · 10.02 GB checkpoint · [CC BY-NC 4.0](LICENSE)**

Watch the full demo in the video preview, or visit the
[project page](https://zju3dv.github.io/ScaleNav/#demos).

## Download

Complete the [installation](https://github.com/MachEmbodied/ScaleNav/blob/main/docs/INSTALL.md#environment),
then run from the ScaleNav code repository root:

```bash
python scripts/download_checkpoint.py --output checkpoints/scalenav
```

The downloader resumes interrupted transfers and verifies the pinned
[checkpoint checksums](https://github.com/MachEmbodied/ScaleNav/blob/main/docs/checkpoint-download-manifest.json).
No HF login is required for this checkpoint.

Also download the **Wan VAE, text encoder and tokenizer** (14.20 GB) following
the [checkpoint setup](https://github.com/MachEmbodied/ScaleNav/blob/main/docs/INSTALL.md#checkpoints).
For inference, use a BF16-capable NVIDIA GPU; **24 GB GPU memory and 64 GB system RAM**
are recommended.

<details>
<summary>Download with the Hugging Face CLI</summary>

```bash
hf download Yangyihui/ScaleNav --revision 8808b3589142b52e3bf0719aa0ecff8f86715a4a --local-dir checkpoints/scalenav \
  --include 'transformer/*' ar_model.pt train_meta.pt SHA256SUMS.json
```

</details>

## Inference

Use ScaleNav's `awm` CLI with your own image and an instruction describing a target
visible in it:

```bash
awm check --checkpoint checkpoints/scalenav --sha256
awm infer --checkpoint checkpoints/scalenav \
  --base-model checkpoints/Wan2.2-TI2V-5B \
  --image /path/to/observation.png \
  --prompt "Walk to the chair visible in the initial view." \
  --output outputs/example
```

Replace the image path and instruction with a matching pair. This saves nine
per-frame action deltas in `observation_trajectory.npy` and a visualization in
`observation_trajectory.png`. Actions are forward/right/clockwise-yaw displacements
in metres/metres/radians. Add `--decode-video` to save the nine predicted RGB frames.

For continuous navigation, use [`awm serve`](https://github.com/MachEmbodied/ScaleNav/blob/main/docs/INFERENCE.md#http-api).
See the [inference guide](https://github.com/MachEmbodied/ScaleNav/blob/main/docs/INFERENCE.md)
for paired examples, outputs and the robot adapter.

## Model settings

| Setting | Value |
| --- | --- |
| Observation size | 288 × 512 |
| Observation/action rate | 4 Hz |
| Chunk length | 9 frames |
| Sampling steps | 4 |
| Video / action guidance | 5 / 1 |
| Video / action scheduler shifts | 5 / 1 |
| History | Initial-frame anchor and two recent chunks |
| Action normalization | Symlog, c = 0.01 |

## Checkpoint contents

| File | Contents |
| --- | --- |
| `transformer/diffusion_pytorch_model.safetensors` | ScaleNav model weights |
| `transformer/config.json` | Model architecture |
| `train_meta.pt`, `ar_model.pt` | Deployment metadata |
| `SHA256SUMS.json` | File checksums |

Total checkpoint size: **10.02 GB**. The separate Wan components occupy about
14.2 GB. The code repository includes [training](https://github.com/MachEmbodied/ScaleNav/blob/main/docs/TRAINING.md),
[evaluation](https://github.com/MachEmbodied/ScaleNav/blob/main/docs/EVALUATION.md)
and [trajectory visualization](https://github.com/MachEmbodied/ScaleNav/blob/main/docs/INFERENCE.md#outputs).

The released checkpoint is the final joint-fine-tuning checkpoint at step 6500.
It contains model weights and deployment metadata. Optimizer, scheduler and RNG
state for exact training resumption are not included. The Wan VAE, text encoder
and tokenizer are downloaded separately.

## License

The ScaleNav fine-tuned checkpoint is released under [CC BY-NC 4.0](LICENSE)
for noncommercial use. When sharing it or adaptations, provide attribution,
link to the license, and indicate any changes.

ScaleNav modifies [Wan2.2-TI2V-5B](https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B)
through navigation pretraining, action alignment and joint fine-tuning. The
CC BY-NC 4.0 terms apply to the ScaleNav contributions; upstream Wan materials
retain their [Apache-2.0 license](LICENSE-WAN). The
[code](https://github.com/MachEmbodied/ScaleNav/blob/main/LICENSE) is separately
licensed under Apache-2.0, and [dataset components](https://huggingface.co/datasets/Yangyihui/ScaleNav-Data#sources-and-license)
retain their respective terms.