SatNav: A Scalable Benchmark for Long-Horizon UAV Vision-Language Navigation from Satellite Imagery
Paper • 2609.31507 • Published
SatNav-Scenes-v0.1
Satellite scenes accompanying SatNav: A Scalable Benchmark for Long-Horizon UAV Vision-Language Navigation from Satellite Imagery.
Code · Paper · Project page · Documentation · Episodes
This dataset provides 59 satellite GeoTIFF scenes, totaling 64.79 GB (60.34 GiB). Use these scenes with the SatNav codebase and SatNav-Episodes-v0.1 to run navigation episodes, generate training trajectories, and evaluate policies. SatNav's simulator, SatSim, renders RGB observations from the GeoTIFF scenes as the agent moves through geographic space.
| Resource | Contents |
|---|---|
| SatNav | Simulator, navigation environment, data tools, training, and evaluation code |
| SatNav documentation | Installation, examples, configuration, and model integration guides |
| SatNav-Episodes-v0.1 | Navigation instructions, routes, starting poses, goals, and dataset splits |
| SatNav Model Zoo | Baseline model checkpoints |
| SwiftVLN | SwiftVLN implementation and setup for SatNav |
Use with SatNav
Install SatNav following the documentation,
then download the episodes and scene files. Set SATNAV_SCENES_DIR to the local
scenes/ directory as shown below. The scene filenames match the scene identifiers
used by the episodes.
Request access
- Sign in to your Hugging Face account.
- Enter your name, institution, institutional email, and research purpose.
- Read the access terms and select the agreement checkboxes.
- Submit the form. Access is granted after your request passes the system checks.
Download
Use the same Hugging Face account to authenticate and download the dataset:
pip install -U huggingface_hub
hf auth login
hf download Eku127/SatNav-Scenes-v0.1 --repo-type dataset --local-dir data/satnav_datasets/SatNav-Scenes-v0.1
The repository contains one GeoTIFF per scene:
SatNav-Scenes-v0.1/
|-- README.md
|-- scenes_list.yaml
|-- SHA256SUMS
`-- scenes/
|-- Amsterdam-1.tif
`-- ...
scenes_list.yaml lists the scene identifiers and geographic bounds.
SHA256SUMS contains the SHA256 checksum for each GeoTIFF.
On Linux, verify the downloaded files from the dataset directory:
cd data/satnav_datasets/SatNav-Scenes-v0.1
sha256sum -c SHA256SUMS
When running SatNav, set SATNAV_SCENES_DIR to the absolute path of the
downloaded scenes/ directory:
export SATNAV_SCENES_DIR="/absolute/path/to/SatNav-Scenes-v0.1/scenes"
Access Terms v1.0
By requesting access, you agree to these terms:
- Purpose. Use the scenes for non-commercial academic research.
- Access and sharing. Access is granted to individual Hugging Face accounts. Each collaborator requests access through this page. Redistribution, resale, and public mirrors of the scene files are prohibited.
- Attribution. Preserve the Google Maps and imagery provider attribution and copyright notices. Include the applicable attribution when displaying imagery in publications or presentations.
- Citation. Cite the SatNav paper when reporting research that uses this benchmark, using the reference below.
Request information
The maintainers receive your form responses and the username and email associated with your Hugging Face account. These details are used to manage access and communicate dataset updates or changes to the access terms.
Citation
Jiajun Jiang, Chunliang Hua, Zichun Chen, Yanxing Wu, Zeyuan Yang, Jie Song, and Xiao Hu. SatNav: A Scalable Benchmark for Long-Horizon UAV Vision-Language Navigation from Satellite Imagery. NeurIPS 2026, Track on Evaluations and Datasets (accepted, to appear). Paper on arXiv.
@inproceedings{jiang2026satnav,
title = {{SatNav}: A Scalable Benchmark for Long-Horizon {UAV} Vision-Language Navigation from Satellite Imagery},
author = {Jiang, Jiajun and Hua, Chunliang and Chen, Zichun and Wu, Yanxing and Yang, Zeyuan and Song, Jie and Hu, Xiao},
booktitle = {Advances in Neural Information Processing Systems},
year = {2026},
note = {Track on Evaluations and Datasets; accepted, to appear},
url = {https://arxiv.org/abs/2609.31507}
}
- Downloads last month
- 25