NVIDIA Opens SimReady Central, a Free Library of 3D Simulation Assets

NVIDIA announced SimReady Central Library on October 21, 2026, opening a library of simulation-ready 3D assets on Hugging Face for developers building robotics training environments and AI factory digital twins. In the announcement post, authored by John Kosnik, Frank DeLise, and Pomi Lee, NVIDIA describes the library as a “free, ungated starting point” that brings together more than 2,000 SimReady OpenUSD assets.

The library spans infrastructure and equipment ranging from robot arms and grippers to warehouse equipment, racks, and cooling infrastructure. Developers can check which capabilities each asset supports, then combine assets from different suppliers into a scene. The live SimReady Central interface displays 3,599 results across its Industrial (1,214), OEM (1,063), Residential (1,019), Hospital (256), and Lighting (47) categories. Filters cover profile, rigid bodies, mass, license, and dimensions, and listed assets include Standard Bots robot arms, Robotiq 2F-85 and 2F-140 grippers, a dishwasher, a side-by-side refrigerator, and surgical robots.

What Makes an Asset SimReady

SimReady assets are OpenUSD assets prepared with the properties required for simulation workflows. A detailed 3D model, such as CAD, is a useful starting point, but simulation requires additional information about how an object should be interpreted and used. For a robot manipulation task, that can include the object’s mass, collision geometry, and friction properties; for AI factory workflows, it can include equipment geometry and the electrical or thermal metadata an application requires. OpenUSD itself is an open, extensible framework for describing and composing 3D scenes, and using those scenes to train and evaluate autonomous machines also requires information about physical behavior, collisions, and sensor interactions.

SimReady Foundation provides the specifications, profiles, and validation tools that make these requirements explicit. A profile defines the logical grouping of features and capabilities an asset needs for a particular use case. Depending on the asset and profile, those features can include collision geometry and physical properties for contact and physics simulation in a supported runtime, joint structure and limits for articulated robot and equipment movement, semantic labels for object identification and annotated synthetic data, and materials and sensor attributes for rendering and sensor simulation. NVIDIA advises developers to select assets whose profiles match their workflow needs and compose them into an OpenUSD scene.

OEM Partners and the Submission Pipeline

NVIDIA names Vertiv as an AI factory OEM contributing warehouse equipment racks, server racks, cooling equipment, and supporting infrastructure for digital twins. For robotics OEMs Comau, FANUC, Franka Robotics, Robotiq, Standard Bots, and Universal Robots, the library makes robot arms and grippers discoverable by teams building training and evaluation environments. Simulation asset provider Palatial features the NVIDIA Robotics 1K Dataset, released in collaboration with NVIDIA Robotics, alongside articulated assets generated from text, images, scans, datasheets, and CAD. LG Electronics contributes home appliance assets such as refrigerators and dishwashers, and ForgeCAD contributes articulated robotics props with what NVIDIA describes as high-quality textures.

OEMs contribute by using the SimReady Central submission skill to validate assets against the relevant SimReady profiles, package them, and submit them for inclusion while managing their own Hugging Face datasets. The NVIDIA-Omniverse/simready-central GitHub repository, published under an Apache-2.0 license, serves as NVIDIA’s review entry point for these submissions: OEMs store their packages in publicly accessible Hugging Face datasets and use the repository to submit the exact package versions they want NVIDIA to review.

Before submitting, an OEM completes an onboarding process that includes providing a GitHub username to its NVIDIA contact and accepting a collaborator invitation. NVIDIA recommends an agent-guided flow in which the agent follows the submit-simready-assets skill and runs simready-manage submit to validate, package, and upload each asset, then opens a SimReady Central pull request that NVIDIA reviews and merges. Each dataset entry in the repository is a text file of immutable package-definition URLs, with every line pinning one package to the Hugging Face commit at which it was reviewed; changes take effect only after the pull request is reviewed and merged.

Search and AI-Assisted Scene Authoring

NVIDIA USD Search powers natural-language search in SimReady Central Library, and Hugging Face has integrated USD Search into its platform search with indexed USD properties and simulation metadata to help agents discover SimReady assets suited to their physical AI simulation tasks.

AI-assisted authoring adds another path. Developers can use frontier models such as OpenAI’s GPT-6 Astra, combined with asset search and scene-authoring tools, to build environments from SimReady Central; OEMs can use Astra to help prepare and refine OpenUSD assets against SimReady Foundation requirements before validating and submitting them.

In a robotic pick-and-place example targeting Isaac Sim with PhysX, NVIDIA says it expects Astra to query the SimReady Search API to find relevant assets distributed on Hugging Face, reference or payload them into an OpenUSD stage, generate Python code to assemble the workcell, configure the pick-and-place task, and render a preview for user review. In an AI factory example, Astra would assemble selected SimReady assets into two alternative OpenUSD layouts of racks and cooling equipment, then generate scripts that check equipment overlaps and aisle clearances against supplied requirements before a developer refines the preferred configuration.

NVIDIA lists several ways to extend either environment: keep layout and task-specific changes in separate OpenUSD layers so the same source assets support multiple experiments; introduce additional assets and vary placements, equipment configurations, or surrounding infrastructure; configure robotics training and evaluation tasks in Isaac Lab, or connect an AI factory scene to the equipment data and simulation tools needed for analyses such as power distribution or cooling performance; and carry validated layouts into operations by connecting the scene to live BMS and IT telemetry. These environments can support projects such as Microduck, an open-source robot from Hugging Face and Pollen Robotics.

Roadmap and Developer Resources

NVIDIA says SimReady Central’s roadmap includes more OEM partners and a growing asset library, and it asks the community to comment on which assets they need, which runtimes they use, and where they encounter integration gaps. For developers starting out, NVIDIA points to the Isaac Sim and Isaac Lab quickstarts, a free OpenUSD learning path, free physical AI courses, CAD-to-SimReady skills, and the SimReady Central Library documentation, and it asks for issues in SimReady Foundation for bugs or SimReady requirements feature requests.