UCLA-VAIL/UrbanVerse-Training-Scenes

UrbanVerse Training Scenes (Urban Cousins)

A collection of ready-to-simulate urban 3D scenes in OpenUSD for NVIDIA Isaac Sim / Isaac Lab, released by the VAIL-UCLA lab. Each scene is a self-contained USD stage with all of its materials and textures, so it can be opened and simulated directly.

The scenes are generated with UrbanVerse — Scaling Urban Simulation by Watching City-Tour Videos (Liu et al., ICLR 2026, arXiv:2510.15018, project page) — whose UrbanVerse-Gen pipeline reconstructs metric-scale urban simulations from city-tour videos using retrieved assets from the UrbanVerse-100K database. This repository provides the "Urban Cousins" procedural variants of those real-world walking/driving routes.

• ~375 scenes spanning 64 base real-world locations across 6 continents (Africa, Asia, Europe, Middle East, North America, South America, Oceania).
• Each base location (e.g. Europe_France_Paris_drive_01) has multiple procedurally re-composed _Cousin_NN variants.
• ~1.2 TB total, stored via Xet/LFS.

Scene layout

Every scene is one top-level directory. Do not flatten it — the USD stages reference their textures and materials through relative paths.

<Region>_<Country>_<City>_<walk|drive>_<NN>_Cousin_<NN>/
├── World0.usd                 # top-level stage (open this)
├── .collect.mapping.json      # provenance: original asset path -> local path + hashes
└── SubUSDs/
    ├── *.usd                  # component / instanced sub-stages
    ├── materials/*.mdl        # MDL PBR materials
    └── textures/*.png|*.jpg|*.hdr   # textures and HDRI sky maps

Naming: Region_Country_City_{walk|drive}_NN_Cousin_MM, e.g. Asia_Japan_Kyoto_walk_01_Cousin_03.

Usage

Download

With the helper script (resumable):

# one scene
python download.py --scenes Asia_Japan_Kyoto_walk_01_Cousin_01 --dest ./scenes
# first 10 scenes
python download.py --limit 10 --dest ./scenes
# everything (incl. archives)
python download.py --all --include-tars --dest ./UrbanVerse-Training-Scenes

Or directly with huggingface_hub:

from huggingface_hub import snapshot_download

# a single scene
snapshot_download(
    repo_id="UCLA-VAIL/UrbanVerse-Training-Scenes",
    repo_type="dataset",
    allow_patterns=["Asia_Japan_Kyoto_walk_01_Cousin_01/**"],
    local_dir="./scenes",
)

Open in Isaac Sim

from isaacsim import SimulationApp
app = SimulationApp({"headless": False})
import omni.usd
omni.usd.get_context().open_stage("./scenes/Asia_Japan_Kyoto_walk_01_Cousin_01/World0.usd")

Provenance

These scenes originate from the UrbanVerse project (see Citation). Each scene ships a .collect.mapping.json recording, per asset, its original source path plus source/target SHA-1 hashes — produced by the Omniverse Collect workflow when the scene was localised from the isaacurban / cousin_assets_package release.

License

Released under CC BY 4.0. Please attribute VAIL-UCLA when using these scenes.

Citation

These scenes are produced with UrbanVerse; please cite:

@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 (ICLR)},
  year      = {2026},
  url       = {https://arxiv.org/abs/2510.15018}
}