---
license: other
pretty_name: AirGoal-10k
task_categories:
- robotics
tags:
- aerial-navigation
- image-goal-navigation
- uav
- world-model
- uncertainty-aware-navigation
- ua-nwm
---

# AirGoal-10k

AirGoal-10k is an aerial image-goal navigation dataset released with
**UA-NWM: Uncertainty-Aware World Model for Aerial Image-Goal Navigation**.

Project page: [https://duryi.github.io/UA-NWM-Project-Page/](https://duryi.github.io/UA-NWM-Project-Page/)  
Code: [https://github.com/DurYi/UA-NWM](https://github.com/DurYi/UA-NWM)  
Paper: [https://arxiv.org/abs/2608.05597](https://arxiv.org/abs/2608.05597)

## Dataset Summary

AirGoal-10k contains 11,000 aerial navigation trajectories for image-goal navigation. Each trajectory
contains 12 RGB observations and trajectory metadata. The test split additionally provides 32
candidate trajectories per case for offline ranking evaluation.

The dataset is stored under the `airgoal10k/` subdirectory in this Hugging Face repository:

```text
airgoal10k/
├── action_stats.json
├── pseudo_goals/
│   └── test/
├── train/
│   └── traj_000000..008999/
├── val/
│   └── traj_000000..000999/
└── test/
    └── traj_000000..000999/
```

## Splits

| Split | Number of trajectories | Notes |
| --- | ---: | --- |
| `train` | 9,000 | Training trajectories |
| `val` | 1,000 | Validation trajectories |
| `test` | 1,000 | Evaluation trajectories with 32 candidates per case |

The repository also includes a small set of pseudo-goal images under `airgoal10k/pseudo_goals/` for
visualization and debugging.

## File Format

Each trajectory folder contains 12 RGB frames and one metadata file:

```text
traj_xxxxxx/
├── 0.jpg
├── 1.jpg
├── ...
├── 11.jpg
└── traj_data.pkl
```

`traj_data.pkl` is a Python pickle dictionary. For train/validation trajectories, it contains:

- `point`: `(12, 3)` array of 3D positions.
- `roll`, `pitch`, `yaw`: `(12,)` arrays of attitude angles.
- `images`: `(12,)` array of image identifiers.

For test trajectories, `traj_data.pkl` additionally contains:

- `candidate_trajectories`: a list of 32 candidate trajectory dictionaries, each with `point`,
  `roll`, `pitch`, and `yaw` fields.

`action_stats.json` stores the dataset-level action normalizer used by the UA-NWM codebase.

## Download

From the root of the UA-NWM code repository, run:

```bash
hf download DurYi/AirGoal-10k \
  --repo-type dataset \
  --local-dir data \
  --include "airgoal10k/**"
```

The dataset is also available from
[Baidu Netdisk](https://pan.baidu.com/s/1xzUz1laMjZY1CyDWNec_FQ?pwd=uawm).

The expected local layout after downloading is:

```text
data/
└── airgoal10k/
    ├── action_stats.json
    ├── pseudo_goals/
    ├── train/
    ├── val/
    └── test/
```

See the [UA-NWM repository](https://github.com/DurYi/UA-NWM) for environment setup, checkpoint
preparation, training, evaluation, and visualization instructions.

## Source and Acknowledgment

AirGoal-10k is built on top of
[AerialVLN](https://github.com/AirVLN/AirVLN) and
[OpenFly](https://huggingface.co/datasets/IPEC-COMMUNITY/OpenFly). Please also respect the licenses and terms of
the original data sources.

## Citation

```bibtex
@misc{zhu2026uanwm,
      title={Uncertainty-Aware World Model for Aerial Image-Goal Navigation},
      author={Deyi Zhu and Haoyu Fan and Yinan Zhu and Weichen Zhang and Shilin Ma and Xinlei Chen and Yansong Tang},
      year={2026},
      eprint={2608.05597},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2608.05597},
}
```