Skip to content

Dataset · nvidia

NVIDIA OpenUSD simulation 3D assets

nvidia/USDCraft-10K

1,000 to 10,000 movable and rigid 3D objects with physics and joints for robot training.

What's inside

The NVIDIA dataset contains 1,000 to 10,000 simulation-ready OpenUSD 3D objects, as well as images and texts. The assets are generated using a language model via the USDCraft system as programs and compiled with physics in mind. Each object includes details, constrained movable joints, collisions, and mass.

What you can do with it

  • Training robotics and object manipulation algorithms in simulators
  • Training 3D asset generation models from text descriptions or code
  • Testing physically correct scenes in the OpenUSD environment
Tasks
Robotics
Size
10K–100K rows
Downloads this month
76.2K
Likes
149

How to load it

from datasets import load_dataset

ds = load_dataset("nvidia/USDCraft-10K")