Hugging Face adds 2,000-plus simulation-ready assets for robot and data-center builders

Developers building robot training environments or digital models of AI data centers now have a free library of more than 2,000 equipment assets on Hugging Face. Announced October 21, SimReady Central packages 3D models with simulation properties, giving teams a starting point beyond objects that merely look right on a screen.

A robot simulation may need an object’s mass, friction and collision geometry—the shapes used to calculate contact. A data-center simulation may need electrical or thermal information. NVIDIA describes prepared assets as a way to reduce that manual setup.

From a 3D shape to a usable simulation

SimReady Central uses OpenUSD, an open framework for describing and combining 3D scenes. Its contents range from robot arms and grippers to warehouse equipment, server racks and cooling infrastructure. Developers can inspect supported capabilities, then combine equipment from different suppliers in one scene.

The organizing unit is a profile: a defined set of features an asset needs for a particular simulation task. SimReady Foundation supplies those requirements and validation tools. Depending on the profile, assets can include joint structures and movement limits, object labels, or material and sensor attributes.

Finding parts, then assembling a scene

NVIDIA USD Search powers natural-language searches across participating datasets. Hugging Face has also integrated it into platform search, indexing USD properties and simulation metadata to help developers and AI agents find suitable assets.

The announcement illustrates a proposed pick-and-place workflow using a FANUC arm, compatible Robotiq gripper and warehouse assets in Isaac Sim. An agent would search for parts, generate Python code to assemble the scene, configure the task and render a preview.

The data-center example instead proposes alternative rack-and-cooling layouts, with generated checks for overlaps and aisle-clearance violations. Changes can sit in separate OpenUSD layers, preserving source assets while teams vary layouts or task-specific details.

NVIDIA’s accompanying digital-twin walkthrough explains where assembly work stops. Power and cooling studies still require equipment properties, connection data and preparation for the chosen simulation tools. An agent-generated model needs dimensional checks and engineering review; a plausible-looking layout is not enough.

Manufacturers supply the next building blocks

Equipment manufacturers can prepare their CAD models, validate them against relevant profiles, package them and submit them for inclusion. They can manage their own Hugging Face datasets. NVIDIA presents this as a route for customers to test scenarios around manufacturers’ products.

The roadmap calls for more manufacturers and a larger library. NVIDIA is asking developers to identify missing equipment, the simulation runtimes they use and integration gaps, helping guide which assets come next.