NVIDIA publishes SimReady robot assets on Hugging Face for Isaac Sim pick-and-place workflows

NVIDIA has published SimReady robot and object assets on Hugging Face's SimReady Central, an open asset library for robotics simulation. The release includes validated models for the UR17G robotic arm, Robotiq 2F-140 gripper, and a beverage can, each prepared with physics properties including collisions, mass, joints, and joint drives needed for simulation in NVIDIA Isaac Sim 6.2.

The assets follow the SimReady Foundation specification, which defines profiles, features, and validation requirements to ensure reusable simulation components. Developers can download these assets via Hugging Face's web interface or CLI, assemble them in Isaac Sim, and use AI agents such as GPT-6 Sol to generate Python control code for pick-and-place tasks. The validation reports confirm passing PhysX driven-joint checks, providing a baseline for asset reliability across different simulation scenarios.

SimReady asset structure and Isaac Sim integration

• OpenUSD asset specification: SimReady Foundation defines three specification levels—Profile (use-case feature groups like robot body), Features (capabilities such as controllable joints), and Requirements (validation checks on settings like joint drive parameters)—to standardize simulation-ready assets.
• Validated physics properties: The UR17G arm and Robotiq 2F-140 gripper include passing PhysX driven-joint validation checks, with the arm conforming to Robot-Body 4.0.0 profile and the gripper to Robot-Gripper 3.0.0 profile.
• Asset packaging: Each asset ships as a complete folder with USD models and supporting files; the example gripper path shows a .usda file structure that Isaac Sim loads with dependencies intact.
• Isaac Sim 6.2 integration: The workflow requires launching Isaac Sim with --enable isaacsim.code_editor.python_server for remote Python agent interaction, and assets are assembled programmatically or manually in the simulator.
• AI agent code generation: The example uses GPT-6 Sol with Isaac Sim skills to generate Python task code for arm-gripper assembly, joint inspection, and pick-and-place controls, though generated code requires human inspection before deployment.

Why it matters

This release lowers barriers for robotics researchers and developers by providing validated, reusable simulation assets through an open platform rather than proprietary silos. The Hugging Face distribution model and OpenUSD standardization support interoperability across tools and reduce duplicate asset preparation work in the robotics and AI training pipeline.