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

NVIDIA Details AI-Assisted Workflow for Robotics Simulation

NVIDIA has detailed a structured, five-step workflow that uses frontier AI models to automate the conversion of complex CAD assets into simulation-ready digital twins for robotics. The process leverages the NVIDIA Omniverse platform and SimReady Foundation specifications to address a critical bottleneck in robotics development: the time-consuming manual preparation of assets for accurate physics-based simulation. By using an AI agent, exemplified by GPT-6 Astra, developers can direct the conversion process using natural language prompts, significantly reducing the manual scripting and configuration required to prepare a robot for virtual testing.

The workflow was demonstrated using an ABB YuMi robot, converting it from industry-standard STEP files into a validated OpenUSD asset within NVIDIA Isaac Sim. The AI agent assists at each stage by generating Python code that calls on specific Omniverse tools and skills. This includes importing the geometry, matching visual appearance from reference images, configuring complex physics properties based on manufacturer datasheets and URDF files, and validating the final asset against SimReady standards. The resulting digital twin was then successfully tested in a simulated pick-and-place task, confirming its readiness for development.

• Platform: NVIDIA Omniverse and Isaac Sim 6.1
• AI Agent Example: GPT-6 Astra
• Asset Standard: OpenUSD with SimReady Foundation specifications
• Key Agentic Tools: NVIDIA Omniverse AI Agent skills (e.g., CAD-to-SimReady)
• Physics Configuration: Joint axes, limits, and mass derived from datasheets and a public URDF; contact friction set to 0.8 static and 0.6 dynamic.

This integration of agentic AI into a specialized engineering pipeline marks a notable step in the practical application of large language models beyond content generation. By providing an AI with a library of specific 'skills'—executable functions within the Omniverse ecosystem—the system can translate high-level human intent into a series of precise technical operations. This approach could lower the barrier to entry for creating sophisticated robotics simulations, enabling faster design iteration and more robust testing of robot behaviors before deploying physical hardware.

The true value proposition here is not the AI model itself, but its integration with a structured, tool-rich environment like Omniverse. By providing AI agents with specific 'skills' to call upon, NVIDIA is creating a blueprint for how generative AI can move beyond text and image generation to perform complex, multi-step engineering tasks, effectively bridging the gap between human intent and machine execution in digital twin creation.