HeteroTask / README.md
LiFanxing's picture
Add a placeholder license
f7eacb6 verified
|
Raw History Blame Contribute Delete
1.82 kB
---
license: other
license_name: placeholder
pretty_name: HeteroTask assets
tags:
- robotics
- isaac-lab
- usd
---
# HeteroTask assets
Assets of the heterogeneous vision-language-action pick-and-place tasks of Isaac Lab
(`isaaclab_tasks.contrib.heterogeneous_tasks`, registered as `IsaacContrib-VLA-PickPlace<NNN>-Franka` and
`IsaacContrib-VLA-Heterogeneous-Franka`), as complete USD files. The robot is in the separate
[franka_robotiq](https://huggingface.co/datasets/LiFanxing/franka_robotiq) dataset. The tasks load these assets from
`https://huggingface.co/datasets/LiFanxing/HeteroTask/resolve/main`, or from the location in the
`VLA_BENCHMARKS_ASSETS_DIR` environment variable.
## Layout
| Path | Content |
|---|---|
| `objects/<name>/<name>.usd` | An object or piece of furniture with its physics: rigid body, mass and inertia, and convex-hull colliders (objects) or box colliders (furniture). |
| `objects/<name>/materials/` | The textures and MDL materials of the object. |
| `scene/franka_table/` | Table that the robot stands on. |
| `scene/home_office.exr` | Dome light texture. |
## Sources
- Objects and furniture: the [Omniverse](https://omniverse-content-production.s3-us-west-2.amazonaws.com/Assets/)
SimReady and ArchVis libraries, and the RoboLab object library (YCB, HOPE, HOT3D, HANDAL, Objaverse and other
objects). The ArchVis models, which have no physics, got rigid bodies, masses and colliders; colliders decomposed
into convex hulls were baked into at most 32 hulls each (16 for ArchVis objects), and furniture collides with boxes
of its boards and panels.
- Robot table and dome light texture: RoboLab.
## License
Placeholder until the licenses are reviewed: each asset keeps the license of its source listed above, and the
dataset may only be used as those licenses allow.