Nvidia Releases 472GB Collection of 2,575 Simulation Ready 3D Assets on Hugging Face for Robotics and Digital Twins
NVIDIA just put 2,575 simulation-ready 3D assets on Hugging Face 🤯 A 472GB collection covering everyday objects, tools and industrial equipment. It’s called NVIDIA SimReady Assets. The models include physics settings and semantic labels for use in simulated environments. - Train robots to pick up and move objects - Build virtual factories and warehouses - Generate labelled training images - Populate 3D scenes with ready-made objects Useful for robotics research, digital twins and anyone building detailed virtual environments. Download the collection below 👇
Context
Openly releasing thousands of physically calibrated 3D objects with preconfigured collision meshes, mass properties, and semantic labels removes a major data bottleneck for training manipulation robots and building industrial digital twins.
Training embodied AI and robotic manipulation policies in simulators like Nvidia Isaac Sim requires 3D meshes that behave accurately under gravity, friction, and contact forces rather than purely visual game assets that lack physical metadata.
Deep dive
Nvidia has published a 472 gigabyte dataset called Nvidia SimReady Assets on Hugging Face containing 2,575 simulation ready 3D models, as highlighted on X by Oliver Prompts on Sunday night. The collection spans everyday household objects, hand tools, warehouse shelving, pallets, and industrial machinery built in Universal Scene Description format.
Unlike standard 3D meshes used in video games or visual effects, each SimReady asset comes preconfigured with physical properties such as mass, center of mass, friction coefficients, and accurate collision geometry, alongside semantic class labels for computer vision training. That allows robotics researchers to drop the objects directly into physics simulators to train robotic arms on grasping and rearrangement tasks without spending weeks manually rigging collision boxes and material properties.
Developers can also use the labeled 3D catalog to generate synthetic camera, depth, and segmentation datasets for computer vision models, or to populate virtual factories and logistics warehouses for industrial digital twin simulations.
What to expect next
Robotics labs using Nvidia Isaac Lab, MuJoCo, and Blender based synthetic data pipelines will incorporate the 2,575 SimReady objects to diversify domain randomization for sim to real grasping policies and warehouse navigation models. Standardizing physical mass, friction, and semantic annotations across thousands of free USD objects should also lower the barrier for academic labs and smaller robotics startups that cannot afford to build proprietary 3D asset libraries from scratch.