5 Steps to Create SimReady Assets for Robotics with Frontier AI Models

NVIDIA published a five-step workflow for converting CAD assets into simulation-ready (SimReady) OpenUSD assets for robotics, demonstrated on an ABB Robotics YuMi robot. The walkthrough, run in Isaac Sim 6.1 with Codex CLI launched inside the Isaac Sim directory, used GPT-6 Astra as one example of a frontier AI model that writes Python code calling NVIDIA Omniverse libraries at each stage.

The steps cover STEP-file conversion, matching visual appearance against ABB reference images and videos, configuring physics, validating against SimReady requirements, and running a pick-and-place task with both arms and grippers. Physics configuration relied on manufacturer datasheets and a public URDF for joint axes, limits and the zero configuration, with per-link masses estimated from STEP geometry volumes and normalized to specified totals of 38 kg for the robot and 0.28 kg per gripper. Contact properties assumed static friction of 0.8, dynamic friction of 0.6, and zero restitution; NVIDIA states these parameters were not calibrated against physical YuMi measurements.

SimReady defines requirements for preparing OpenUSD assets for specific simulation use cases; developers pick a profile and validate against its requirements, and missing simulation properties can let objects pass through a gripper or prevent coordinated motion. SimReady Foundation checks covered units, asset dependencies, geometry, materials, rigid bodies, joints, drives and articulation, finding all 21 rigid bodies had positive masses and admissible inertia tensors. Runtime testing covered arm and gripper motion, contact, release and collisions along the tested trajectories, but NVIDIA notes these checks did not establish agreement with real-world robot dynamics or collision safety across every pose.

In the final task, both arms and grippers completed four pick-and-place cycles in a 122.2-second physics simulation using 45 mm, 40 g cubes, with each arm grasping, lifting, holding for two seconds, transferring and releasing onto a matching color target. A second demonstration used a Sharpie marker modeled from a reference image, estimated at 140 mm and 10 g. Finger contact and friction carried the cubes without attachment joints, kinematic holds or direct cube-pose updates. NVIDIA cautions results may vary with the frontier AI model, reasoning effort and prompts used.

Why it matters: Developers preparing robot assets for simulation can follow a documented path for conversion, physics setup and validation rather than manual preparation. Public assets for the ABB YuMi walkthrough, the SimReady Foundation repository and the CAD-to-SimReady skill are available to follow along.