NVIDIA Outlines 5 Steps for Preparing SimReady Assets for Robots Using AI Agents

Preparing CAD assets for robotics simulation requires more than simply converting geometry into OpenUSD. Developers need to configure materials, collision geometry, joints, and other physical properties before testing a robot's behavior. NVIDIA Omniverse libraries, governed by SimReady Foundation specifications and NVIDIA's agent skills, provide a structured workflow for converting and validating SimReady assets.

Frontier AI models, such as GPT-6 Astra, can assist developers at every stage by interpreting reference materials and calling Omniverse tools on their behalf, reducing the amount of manual preparation needed. NVIDIA published a full example of preparing the industrial robot ABB YuMi for simulation using SimReady Foundation and NVIDIA Omniverse.

What SimReady Is and Why It Matters

SimReady defines requirements for preparing OpenUSD assets for specific simulation use cases. Developers choose a SimReady profile based on their intended use and validate the asset against its requirements. A robot might look correct but still fail in simulation if it lacks collision geometry or if its joints are configured incorrectly. Simulation properties such as mass and friction coefficient directly affect grip stability.

The workflow relies on the following technologies: OpenUSD for representing geometry, assembly structure, materials, and simulation properties; SimReady Foundation with its specifications and validation guidance; NVIDIA Omniverse libraries (omniverse-cad-to-simready, omniverse-cad-to-usd, isaac-sim-remote, usd-pipeline, usd-articulation, physics-simulation, isaac-sim-validator) for content conversion and preparation; NVIDIA Isaac Sim 6.1 for physics simulation and testing; Codex CLI for interacting with frontier models.

Five Steps for Preparing a Robot

Step 1: Import from STEP files. The ABB YuMi robot is loaded from its original CAD files along with both manipulator arms and grippers. The CAD-to-SimReady skill helps convert the geometry to OpenUSD and position the grippers at the correct mounting points on the robot's arms.

Step 2: Aligning the visual representation. The asset is checked and compared against reference images and video from ABB Robotics. An AI agent adjusts materials, textures, colors, metallic sheen, and roughness to match the physical prototype.

Step 3: Configuring physical properties. For both arms and grippers, the workflow sets up joints (with axes and motion limits taken from the manufacturer's datasheet or a public URDF), collision geometry, mass properties (robot at 38 kg, each gripper at 0.28 kg), and friction (static 0.8, dynamic 0.6).

Step 4: SimReady validation. The asset is checked against a requirements checklist: units of measurement, geometry, materials, rigid bodies, joints, drives, articulation, and runtime behavior.

Step 5: Task execution. The robot performs four cycles of a pick-and-place task with colored cubes and a Sharpie marker in Isaac Sim 6.1, demonstrating grasping, holding, transferring, and releasing without additional articulated constraints.

Practical Outcome

Frontier models, such as GPT-6 Astra, write Python code that calls Omniverse tools at every stage. This automates the preparation process and reduces the need for manual configuration. NVIDIA provides access to the SimReady Foundation repository, the CAD-to-SimReady skill on GitHub, and the Isaac Sim 6.1 environment (which includes simready-foundation-tier-core) for a full implementation of the workflow.