---
license: apache-2.0
library_name: pytorch
tags:
  - world-model
  - aerial-navigation
  - diffusion
  - computer-vision
---

# ANWM

Aerial Navigation World Model checkpoint released with the paper
[Aerial World Model for Long-horizon Visual Generation and Navigation in 3D Space](https://arxiv.org/abs/2512.21887).

- **Code**: [https://github.com/EmbodiedCity/ANWM.code](https://github.com/EmbodiedCity/ANWM.code)
- **Dataset**: [EmbodiedCity/ANWM-Dataset](https://huggingface.co/datasets/EmbodiedCity/ANWM-Dataset)

## Files

| File | Steps | Notes |
|---|---:|---|
| `0200000.pth.tar` | 200,000 | Paper / released model (`config/anwm.yaml`) |

## Quick start

Clone the code, then download this checkpoint into the path expected by `infer.py`:

```bash
git clone https://github.com/EmbodiedCity/ANWM.code.git
cd ANWM.code

huggingface-cli download EmbodiedCity/ANWM 0200000.pth.tar \
  --local-dir logs/anwm_cdit_airvln/checkpoints

# optional check
python scripts/check_environment.py --require-checkpoint

# inference
torchrun --standalone --nproc_per_node=1 infer.py \
  --exp config/anwm.yaml \
  --ckp 0200000 \
  --datasets airvln_16 \
  --eval_type rollout \
  --rollout_fps_values 1,4
```

This places the weight at:

```text
logs/anwm_cdit_airvln/checkpoints/0200000.pth.tar
```

## Related

- Code: https://github.com/EmbodiedCity/ANWM.code
- Dataset: https://huggingface.co/datasets/EmbodiedCity/ANWM-Dataset
- Model: https://huggingface.co/EmbodiedCity/ANWM
- Paper: https://arxiv.org/abs/2512.21887