AIAny
Icon for item

USDCraft-10K: Simulation-Ready Articulated 3D Assets in OpenUSD

Provides about 10,000 OpenUSD objects for robotics simulation, including articulated and static assets with physics properties and robot-interaction annotations.

Introduction

Robot manipulation simulations need more than attractive meshes: objects must have plausible parts, joints, collision geometry, and physical properties. This dataset makes those properties explicit across a broad catalog, so it can support interaction tasks where an agent must grasp, open, slide, or otherwise operate everyday objects.

What Sets It Apart
  • Its 10,000 assets span 1,000 categories, from household appliances to industrial equipment, giving simulation tasks a much broader object pool than a narrow benchmark set.
  • 8,222 assets are articulated and 1,778 are static rigid objects; joint limits, collision geometry, masses, and physics materials are authored for simulation use.
  • Object-level annotations describe graspable features, operation methods, and symmetry, providing useful context for robot interaction beyond geometry alone.
  • The assets are generated from text prompts and packaged as flattened PhysX USD stages for NVIDIA Isaac Sim, making them directly relevant to simulation workflows.
Who It’s For—and the Trade-Off

Great fit for robotics researchers who need varied simulation assets for manipulation, scene construction, or sim-to-real research. Look elsewhere if you need scanned replicas of specific real products or validated deformable-body dynamics: these are generated assets, and fidelity for fine structures and flexible objects remains limited.

Information

  • Websitehuggingface.co
  • OrganizationsNVIDIA, Nanyang Technological University, Peking University
  • AuthorsChuanrui Zhang, Zaijia Yang, Duomin Wang, Lu Shi, Daquan Zhou, Ruihua Zhang, Ziwei Wang
  • Published date2026/10/28

More Items

Hugging Face

Pairs metric-scale 3D scans of real environments with egocentric task recordings, estimated body and hand motion, and hierarchical annotations for physical AI research. The first release contains 126 scenes; the dataset is licensed CC BY-NC 4.0.

Turns LLMs’ physical knowledge into video training data using simulator constraints, physical verification, and iterative correction. It also repairs physical errors in generated videos and releases 100K videos spanning 2,500 physical events.

Hugging Face

Tests whether AI agents can interpret complex visual files, synthesize information across documents, and produce finished professional deliverables. Its 10 public tasks span distinct domains and require multimodal perception and multi-step tool use.