Preparing CAD assets for robotics simulation goes beyond converting geometry to OpenUSD: developers must also configure and validate materials, collision geometry, joints and other physics properties before testing robot behavior. NVIDIA Omniverse libraries, the SimReady Foundation specifications, and Omniverse AI Agent skills provide a structured workflow for converting and validating SimReady (simulation-ready) assets. Frontier AI models can assist at each stage by interpreting references and invoking Omniverse tools, reducing manual preparation.
What is SimReady and why it matters
SimReady defines requirements for preparing OpenUSD assets for particular simulation use cases. Developers pick a SimReady profile for their intended task and validate the asset against those requirements. Profile-specific checks examine an asset’s structure, materials and physics properties and report issues for remediation.
A model can look correct visually but still fail in simulation if required simulation properties are missing: missing collision geometry can let objects pass through a gripper, or incorrect joints can prevent coordinated motion. Parameters such as mass and friction directly affect whether a grasp remains stable.
Technologies and tools used
- OpenUSD: common representation for geometry, assembly structure, materials and simulation properties.
- SimReady Foundation: specifications and validation guidance for simulation-ready assets.
- NVIDIA Omniverse libraries: tools agents and developers use to convert, inspect and prepare OpenUSD content.
- NVIDIA Isaac Sim: environment for configuring physics and testing robot behavior.
- NVIDIA Omniverse AI Agent skills (for example CAD-to-SimReady): guide agents through conversion, material and physics assignment and validation.
SimReady robotics workflow: ABB YuMi example (five steps)
This walkthrough applies the SimReady workflow to an ABB Robotics YuMi robot and demonstrates five steps: STEP-file conversion, appearance validation, physics configuration, SimReady validation, and a final pick-and-place validation task in Isaac Sim. GPT-6 Astra is shown as an example frontier AI model that can generate Python code to call Omniverse libraries, but the workflow is applicable to other models and robots.
Preparation
- Install Isaac Sim (the example used Isaac Sim 6.1).
- Create a reference folder inside the Isaac Sim directory and place robot specifications, STEP files, reference images/videos, the SimReady Foundation spec repository and the CAD-to-SimReady skill there.
- Install and configure your chosen frontier AI model (the example used GPT-6 Astra) and launch Codex CLI inside the Isaac Sim directory so the agent can find the necessary skills.
- Start Isaac Sim in windowed mode with the remote Python server enabled so the agent can communicate via Python with the running Isaac Sim instance.
At each step, adapt prompts for the frontier AI agent, then review simulation outputs and validation feedback to refine the asset.
1) Import robot from STEP files
Place the ABB Robotics YuMi STEP files and reference media in the reference folder. Ask the agent (using CAD-to-SimReady skill and related Omniverse skills) to convert the STEP geometry to OpenUSD and import the robot into Isaac Sim, ensuring both grippers are imported and mounted at the correct arm points.
Check: the asset is imported and available for inspection; required inputs are understood and missing information is identified.
2) Match visual appearance
Inspect the imported asset to confirm materials and textures loaded correctly. Compare the model with the supplied ABB images and video for scale, orientation and part placement. Ask the agent to add or refine materials and textures based on those references and report any visual discrepancies.
Assumption: the agent adjusted color, metallic response and roughness visually while preserving CAD geometry and mount transforms. Material settings were visual estimates rather than measured reconstructions.
Check: scale, orientation, parts and appearance match expectations; any missing materials or textures are documented.
3) Configure physics properties
At this stage the OpenUSD asset has visual geometry but may lack joints, rigid bodies and collision geometry. Ask the agent to configure and validate physics properties for both arms and grippers. Use the included datasheet to set joint ranges and speed limits.
If manufacturer data is incomplete, the agent can identify missing parameters and consult technical manuals or public URDF descriptions. Where mass properties are unavailable, estimate them from STEP geometry volumes using documented density or mass-distribution assumptions following NVIDIA’s guidance.
Assumptions and specifics used in the example workflow:
- The agent retrieved a public URDF to identify joint axes and zero configuration.
- Per-link masses were estimated from STEP volumes and normalized to manufacturer-specified totals: robot total 38 kg, each gripper 0.28 kg.
- Centers of mass and inertia were calculated assuming solid parts of uniform density; internal motors and gearboxes were not modeled separately.
- Contact properties assumed static friction 0.8, dynamic friction 0.6, and zero restitution.
- Collision geometry approximated with convex shapes; surface roughness and material response were approximated.
- Parameters were not calibrated against physical YuMi measurements.
NVIDIA Skills used: usd-articulation, physics-simulation, isaac-sim-validator, isaac-sim-remote.
Check: joints move about expected axes and within limits; base mounting and gripper motion behave as configured; estimated properties and validation failures are documented.
4) Validate SimReady properties
Run the selected SimReady Foundation checks on the OpenUSD asset. Ask the agent to apply the SimReady schema and any related skills to the USD file, review findings and revise the asset as needed.
Validation outcomes in this workflow:
- SimReady Foundation checks covered units, asset dependencies, geometry, materials, rigid bodies, joints, drives and articulation against SimReady requirements.
- Isaac Sim asset checks verified scale, orientation, part placement, gripper mounting, material coverage and resolved texture dependencies using the Isaac Sim validation engine.
- Mass and inertia: all 21 rigid bodies had positive masses and physically admissible inertia tensors; runtime values matched authored values.
- Joints: axes, frames, limits, drives, finger-mimic behavior and fixed-base mounting were checked.
- Runtime behavior: arm and gripper motion, contact, release and collisions were tested along examined trajectories.
Note: these checks validate the simulation implementation but do not establish agreement with real-world robot dynamics or guarantee collision safety across every pose.
Check: the selected SimReady checks pass; tuning changes, estimated properties and unresolved issues are recorded.
5) Final robotic simulation task (pick-and-place)
Combine the validated visual, physics and SimReady properties and run a pick-and-place scenario in Isaac Sim.
Example 1 — cube pick-and-place
- The agent added colored targets and a box to the scene, defined target locations and color metadata, and used both arms and grippers to execute pick, lift, hold, transfer and release sequences, selecting matching color targets via scene metadata.
- Simulation assumptions included friction coefficients (static 0.8, dynamic 0.6) and test cubes sized 45 mm with 40 g mass.
Demonstrated results:
- Completed cycles: both arms and grippers completed four pick-and-place cycles total during a 122.2 second physics simulation.
- Task sequence: each arm grasped its cube, lifted it, held for two seconds, transferred it and released it onto its matching color target, then picked it up again and placed it in the box.
- Contact behavior: finger contact and friction carried the cubes without attachment joints, kinematic holds or direct cube pose overrides during simulation.
- Acceptance: all cycles passed the defined grasp, hold, transfer, release and placement criteria.
- Recording: MP4 and GIF were generated from the simulation.
Example 2 — pick-and-place with an image-derived object (Sharpie marker)
- The agent created a 3D Sharpie marker model from a reference image. Assumed dimensions were 140 mm and 10 g, with the center of mass at the geometric center and collision geometry approximated with cylinders for cap, barrel and rear plug.
- Result: YuMi’s gripper lifted the marker and released it into a blue tray in Isaac Sim.
Note: successful tests demonstrate behavior under the tested conditions; results may vary by frontier AI model and prompt.
Skills used in the final steps: manipulation-ik, physics-simulation, isaac-sim-validator, isaac-sim-remote, isaac-sim-rendering.
Conclusion and how to try the workflow
SimReady Foundation plus NVIDIA Omniverse tools help turn CAD files and reference materials into assets that can be configured, validated and tested in simulation. The YuMi walkthrough is one concrete example of the five-step SimReady robotics workflow using GPT-6 Astra as an assisting frontier AI model.
To adapt the approach to other robots or AI models, ensure access to the required tools and skills, supply the appropriate reference materials, and repeat asset and task validation. Outcomes will vary depending on the model used, prompt engineering and reasoning effort.
Additional resources and event
Start with the SimReady Foundation repository and the CAD-to-SimReady skill to guide asset preparation. The ABB Robotics YuMi STEP file, reference video and image are available to follow the walkthrough. Explore Isaac Sim, Isaac Lab, the free OpenUSD learning path and physical AI courses for deeper study.
Join NVIDIA GTC Berlin on October 20–22, 2026 to learn more about developments in physical AI and robotics and to connect with developers, researchers and industry practitioners.
