NVIDIA Releases 2,000 Free 3D Assets for Robot Training

On October 21, NVIDIA released SimReady Central on Hugging Face, a collection of 3D assets for training robots in simulation. More than 2,000 assets are free to use, including robot arms, grippers, warehouse equipment, server racks, and cooling systems. The target audience is developers who want to train robots in virtual spaces or build digital replicas of AI facilities.

What sets these apart from ordinary 3D models

A visually detailed 3D model alone makes it hard to test how an object actually behaves. Recreating a scene in which a robot picks something up requires information such as weight, friction, and the shape used to judge collisions. SimReady assets come with that information built in, sparing developers the work of adding it piece by piece. Assets for AI facilities can include electrical or thermal information depending on their intended use.

The assets are based on OpenUSD, an open technology for composing 3D scenes. Developers can check which capabilities each asset supports and then mix equipment from multiple manufacturers into a single scene. The requirements for each use case, along with validation tools, are defined by the SimReady Foundation.

From robot arms to LG Electronics appliances

On the robotics side, NVIDIA says developers will find robot arms and grippers from Comau, FANUC, Franka Robotics, Robotiq, Standard Bots, and Universal Robots. Vertiv was presented as an example on the AI facility side, with assets such as server racks and cooling equipment. Manufacturers can prepare and validate models of their own equipment against the requirements, then apply to have them listed in the library. They can also manage their own asset collections directly on Hugging Face.

Image: huggingface.co

South Korea’s LG Electronics is also among the companies featured. NVIDIA pointed to the library as a channel through which LG Electronics, which develops home appliances and household robots side by side, can share assets such as refrigerators and dishwashers. Furnishing a home environment with assets like these makes it usable for training and evaluating robots. According to NVIDIA, such environments can also be used to generate training data in simulation.

Search in plain language, lay out scenes with AI

Developers can search for the equipment they need using everyday sentences. The search is powered by NVIDIA’s USD Search, which Hugging Face has also wired into its platform search. It draws not only on equipment names but also on the properties of the 3D assets and their simulation-related information. The idea is to help developers and AI find assets suited to a given task more quickly.

Image: huggingface.co

NVIDIA also laid out a usage pattern that connects asset search and scene-building tools to OpenAI’s GPT-6 Astra. For a robotics task, that means asking the AI to find an arm and a gripper and set up an environment for picking up and moving objects. For an AI facility, the example given was having the AI produce two layout options for racks and cooling equipment, then run checks for places where equipment overlaps or aisles are too narrow. The developer’s role is to review the results and make corrections.

SimReady Central is focused on reusing existing equipment models to reduce the setup burden of simulation experiments. Developers can vary object positions or facility configurations to test a range of conditions. NVIDIA said it plans to expand both the roster of participating manufacturers and the asset count going forward. Its usefulness will depend on how much of the equipment developers need gets added, and on whether each asset meets the requirements of the simulation tools they use.