NVIDIA opens SimReady Central: 2,000+ free simulation-ready OpenUSD assets on Hugging Face, from FANUC arms to Vertiv racks
The library bundles assets with the mass, collision and joint data physics engines need, searchable in natural language. The post, written by NVIDIA, gives no quality or speed numbers yet.
SimReady Central Library brings together more than 2,000 simulation-ready (SimReady) OpenUSD assets on Hugging Face, giving developers a free, ungated starting point for physical AI training environments and AI factory digital twins.
Comprehensive take
Useful plumbing, not a model release. If you build robot training or evaluation scenes in Isaac Sim or Isaac Lab, SimReady Central saves asset prep and tells you up front what each asset supports. Treat the claims as NVIDIA's own until independent users report coverage, fidelity and licensing.
Use cases
Robot training and evaluation scenes
Populate Isaac Sim or Isaac Lab environments with arms, grippers and warehouse equipment already carrying collision, mass and friction data, then vary placements to test behaviours.
AI factory digital twins
Assemble racks, cooling and supporting infrastructure from OEM assets and check overlaps and aisle clearances, as in the post's layout-comparison example.
Equipment manufacturers
Validate CAD-derived assets against SimReady profiles and publish them so simulation teams can find and test your product.
Specs
| Library size | More than 2,000 SimReady OpenUSD assets source |
|---|---|
| Access | Free and ungated on Hugging Face source |
| Named contributors | Vertiv, Comau, FANUC, Franka Robotics, Robotiq, Standard Bots, Universal Robots, Palatial, LG Electronics, ForgeCAD source |
| Search | Natural-language search via NVIDIA USD Search, also indexed in Hugging Face platform search source |
| Runtimes shown | Isaac Sim with PhysX; Isaac Lab for training and evaluation tasks source |
Insights
- Axion reading: the contribution that matters is the profile system, which makes asset capabilities explicit; the 2,000-asset count is a starting size.
- Axion caution: the post is vendor-authored and reports no fidelity, coverage or timing measurements; the GPT-6 Astra scene-assembly examples are illustrations, not results.
- Axion note: the post does not state per-asset licence terms; check each dataset card before commercial use.
What launched
NVIDIA published SimReady Central on the Hugging Face blog: a free, ungated library of more than 2,000 simulation-ready OpenUSD assets, hosted as datasets on Hugging Face, aimed at robotics training environments and AI factory digital twins. The post is vendor-authored (NVIDIA staff are the bylined authors), so the claims below are NVIDIA's own and not independently verified.
The pitch is about the gap between a 3D model and something a physics engine can use. A CAD model gives geometry; simulation also needs mass, collision geometry, friction and, for articulated equipment, joint structure and limits. NVIDIA's SimReady Foundation defines specifications and profiles that spell out what an asset must carry for a given use case, plus validation tools. The library collects assets that have been prepared against those profiles so developers can check what each asset supports and combine parts from different suppliers into one scene.
Who is in it
The post names equipment makers as contributors: Vertiv for AI-factory racks and cooling; Comau, FANUC, Franka Robotics, Robotiq, Standard Bots and Universal Robots for arms and grippers; Palatial, with the NVIDIA Robotics 1K Dataset; LG Electronics for home appliances such as refrigerators and dishwashers; and ForgeCAD for articulated props. Manufacturers can validate assets against the profiles using a SimReady Central submission skill and submit them for inclusion.
How developers and agents are meant to use it
Search is built on NVIDIA USD Search, which supports natural-language queries across participating datasets; the post says Hugging Face has also integrated USD Search into its platform search with indexed USD properties and simulation metadata. The worked examples are a pick-and-place scenario in Isaac Sim using PhysX with FANUC hardware, and a comparison of AI factory layouts with equipment-overlap and aisle-clearance checks. In both, the post has OpenAI's GPT-6 Astra find assets, write the code that assembles the scene and generate variations. Those are illustrations of a workflow described by NVIDIA, not benchmarked results; the post gives no success rates, timings or asset-quality measurements.
What it means for practitioners
The value is in not rebuilding the same warehouse shelf and robot gripper for the hundredth time, and in profiles that tell you in advance whether an asset will behave in your runtime. The open questions the post leaves for readers to test are coverage and fidelity. Two thousand assets is a start for factory and warehouse scenes, but the roadmap only promises more OEM partners, and the post does not state licence terms per asset or how profile validation is audited. The workflow is also tied closely to NVIDIA's own stack (Isaac Sim, Isaac Lab, USD Search), although OpenUSD itself is an open framework.
Who should try it, who should wait
Robotics teams already on Isaac Sim or Isaac Lab can try it today: it is free and ungated. Teams on other simulators should check which profiles their runtime supports before investing. Hardware makers considering contributing should read the submission guide and decide whether discoverability on Hugging Face is worth the validation effort.