Datasets:
droid2sim
The Alchera PanClean kitchen scene, ported to Isaac Sim for DROID manipulation.
What's here
One scene under scenes/panclean/, in three stages from original to runnable:
panclean/
├── source/ # 1. ORIGINAL — as delivered by Alchera (untouched)
│ ├── asset.glb # the kitchen (glTF, 465 MB)
│ └── render01.png # reference render — the look to match
│
├── asset.usd # 2. PORTED — asset.glb converted to USD
├── textures/ # PBR textures for asset.usd
├── assets/ # same scene split into per-object meshes (13 objects)
│
├── scene.usda # 3. ASSEMBLED — asset.usd placed in the DROID robot frame
└── initial_conditions.json # robot/policy start state (from PolaRiS pan_clean)
- source/ — the original Alchera deliverable, nothing changed.
- asset.usd / textures / assets — the port: only the format is converted
(glb → USD, via NVIDIA
omni.kit.asset_converter); meshes are unchanged. - scene.usda — only references
asset.usdand sets position/orientation. It never edits the meshes.
Coordinate note
The source is Y-up, real scale (countertop ~1.1 m). The DROID policy expects Z-up
with the surface ~0.05 m in front of the base, so scene.usda rotates +90° about X
and lowers it into reach.
Reproduce
Docker setup + render/GUI scripts are in the code repo SNU-PI/droid2sim —
clone it, download this scene into data/panclean/, and run scripts/docker/launch.sh.
See that repo's README.
License
Held jointly by Seoul National University PI Lab, LG, and Alchera.
- Downloads last month
- 2