UrbanVerse-CraftBench Scenes
The dataset contains 12 hand-crafted realistic urban scenes designed for benchmarking embodied AI in diverse urban environments. Each scene is a fully constructed, standalone IsaacSim-ready USD stage.
π¬ See preview videos here π¬
Current release version: 0.1.0
Total size: 16.8 GB
License: CC BY-NC 4.0
Folder structure:
Scene folders follow the naming convention: scene_<id>_<place_type>_<topology>_<primary_attribute>_<secondary_attribute>.
UrbanVerse-CraftBench/
βββ scene_01_residential_street_dual_sidewalks_dense_bollards_trash_bags/
βββ scene_02_residential_street_single_sidewalk_door_to_street_buildings/
βββ scene_03_residential_street_dual_sidewalks_clean/
βββ scene_04_commercial_street_asymmetric_sidewalks_benches_bicycles/
βββ scene_05_japanese_tourist_street_dual_sidewalks_blocked_by_vehicles/
βββ scene_06_bar_street_dual_sidewalks_obstructed_fallen_scooters/
βββ scene_07_narrow_alley_outdoor_dining/
βββ scene_08_commercial_street_bike_lane_dual_sidewalks/
βββ scene_09_cbd_t_intersection_construction_sites/
βββ scene_10_cbd_cross_intersection_diverse_obstacles/
βββ scene_11_urban_park_dogs_sports_area/
βββ scene_12_courtyard_plaza_central_fountain/
Per-scene contents:
Each scene folder contains:
| File | Description |
|---|---|
Collected_export_version.tar |
Archived USD scene. Extract this to get the full scene directory. Open Collected_export_version/export_version.usd in IsaacSim to load the scene. |
preview_video.mp4 |
Flythrough video previewing the scene. |
preview_front.png |
Front-view screenshot of the scene. |
preview_topdown.png |
Top-down overview of the scene. |
preview_closeup.png |
Close-up screenshot highlighting scene details. |
cam0_to_world.txt |
Camera-to-world transformation matrices used for filming the preview video. |
Get Started
UrbanVerse-CraftBench Dataset Usage Toolkit: We provide a Python toolkit urbanverse-scene for easy, robust downloading and exploring the full dataset.
Request Access: First, you need to get access to this repository approved. Visit the repository page, fill out the form and click "Submit". Once your request is approved, you will be able to use our toolkit or any other method you prefer to download and access the files.
Install & Authentication
Hardware Requirements: Windows and Ubuntu are all supported for using most APIs, as it requires IsaacSim 4.5.0 for openning the scenes. See details of IsaacSim requirements here.
Install the toolkit package:
pip install urbanverse-scene --extra-index-url https://pypi.nvidia.com
Note: The --extra-index-url flag is required because isaacsim is hosted on NVIDIA's PyPI index.
Authenticate your machine with your Hugging Face account that have access to this HF repository approved:
hf auth login
You might be prompted to paste your Hugging Face access token, if you are not logged-in before. You can create a token here: https://huggingface.co/settings/tokens
After login, verify the current account:
hf auth whoami
Quick start
We recommend using the provided urbanverse-scene toolkit to quickly and efficiently get started with downloading and using UrbanVerse-CraftBench scenes.
import urbanverse_scene as uvs
(Optional) Set and override cache directory (default: ~/.cache/urbanverse_scenes/)
uvs.set("~/my_data/urbanverse_scenes")
List available CraftBench scenes and download specific scenes:
descs = uvs.craftbench.get_descriptions()
# Download & extract specific scenes
result = uvs.craftbench.load_scenes(descs[8:10])
usd_path = result[descs[8]] # Path to .usd scene
Download all 12 CratBench scenes:
# Download
uvs.craftbench.download_all()
# Check download integrity & repair missing files
uvs.craftbench.check_integrity()
uvs.craftbench.repair()
Preview scenes in the browser or open them in IsaacSim simulator:
# Preview scenes in the browser (video + images)
uvs.craftbench.preview(descs[8:10])
# Open scenes in IsaacSim simulator
uvs.open(descs[8:9])
uvs.open(descs[8:9], headless=True, rtx=True)
Toolkit API overview
| Function | Description |
|---|---|
uvs.set(path) |
Set the local cache directory |
uvs.open(descriptions, headless, rtx) |
Open scenes in IsaacSim |
uvs.craftbench.get_descriptions() |
List all 12 CraftBench scene names |
uvs.craftbench.load_scenes(descriptions) |
Download & extract specific scenes |
uvs.craftbench.download_all() |
Download the full CraftBench dataset |
uvs.craftbench.check_integrity() |
Verify download completeness |
uvs.craftbench.repair() |
Re-download missing files |
uvs.craftbench.preview(descriptions) |
Browser-based scene preview (video + images) |
Citation
If you use this dataset or find our work helpful, please cite our paper:
@inproceedings{
liu2026urbanverse,
title={UrbanVerse: Scaling Urban Simulation by Watching City-Tour Videos},
author={Mingxuan Liu and Honglin He and Elisa Ricci and Wayne Wu and Bolei Zhou},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
}
- Downloads last month
- 112