Announcing SimReady Central: An Open Library of Simulation-Ready 3D Assets for Robotics and AI Factories

What it claims

SimReady Central Library is a set of more than 2,000 OpenUSD assets on Hugging Face. Each asset carries the simulation properties a raw CAD model lacks, such as mass, collision geometry and friction, or electrical and thermal metadata for AI factory work. Each asset is tagged with the SimReady Foundation profiles it meets. A profile is a named group of features required for one use case, so a developer picks assets whose profiles match the workflow and then composes them into one OpenUSD scene, even when they come from different suppliers. OEMs keep their content in their own Hugging Face datasets.

Named contributors and what they supply:

• Robot arms and grippers: Comau, FANUC, Franka Robotics, Robotiq, Standard Bots and Universal Robots.
• AI factory equipment: Vertiv supplies racks, server racks and cooling equipment.
• Robot training assets: Palatial supplies the NVIDIA Robotics 1K Dataset and articulated assets generated from text, images, scans, datasheets and CAD.
• Household appliances: LG Electronics supplies refrigerators and dishwashers for household robot environments.
• Textured props: ForgeCAD supplies articulated props for perception and manipulation training.

How it works

flowchart LR
  A[OEM CAD model] --> B[CAD-to-SimReady skills]
  B --> C{Validate vs profile}
  C -->|pass| D[Submission skill]
  D --> E[OEM HF datasets]
  E --> F[USD Search index]
  F --> G[Isaac workflows]
  F --> H[Library search]
  F --> I[AI agents]

NVIDIA USD Search indexes the participating datasets and serves natural-language queries. Hugging Face has also added USD Search to its own platform search, with USD properties and simulation metadata indexed. The SimReady Central GitHub repository has "index-to-library" skills for building a custom asset browser.

How to use it

The search endpoint named on the page is nvidia-simready-central.hf.space. The post gives two example prompts for OpenAI's GPT-6 Astra.

The first prompt builds a pick-and-place scene in Isaac Sim with PhysX, using a FANUC arm and a Robotiq gripper. The expected steps are:

1. Query the SimReady Search API.
2. Reference or payload the assets into an OpenUSD stage.
3. Generate Python code that assembles the workcell.
4. Configure the pick-and-place task.
5. Render a .mov preview.

The second prompt builds two AI factory layouts. It checks equipment overlaps and aisle clearances against supplied dimensions and saves each layout as an OpenUSD scene. Both prompts ask for scripts, asset versions and runtime versions so the result can be reproduced.

The post recommends some practices for extending these scenes:

• Keep layout and task changes in separate OpenUSD layers so the source assets serve several experiments.
• Configure robotics training and evaluation in Isaac Lab.
• For AI factories, connect the scene to live building management system (BMS) and IT telemetry.

OEMs contribute through the onboarding guide, the CAD-to-SimReady skills and the submission skill.

Limits

The page says "free, ungated" but names no licence for the assets. The Astra workflows are described as what the model is "expected" to do, and the post reports no success rates, timings or failure cases. The post does not break down the asset count by category or supplier. The list of profile features did not come through in the fetched text. NVIDIA asks for feedback on missing assets, runtimes and integration gaps, which suggests runtime coverage beyond Isaac Sim and PhysX is not established.