The traditional manufacturing workflow—design, build, test—rested on a core belief: Only real-world testing could guarantee reliability. That assumption is rapidly giving way. High-fidelity simulation now produces synthetic training data accurate enough for production-grade AI, enabling perception systems, reasoning models, and agentic workflows to thrive in live factory environments. The catalyst? OpenUSD, which has emerged as the connective standard that makes this practical, and manufacturers adopting it are already seeing measurable results.

SimReady: The Foundation for Physical AI

As Physical AI becomes integral to industrial operations, a foundational challenge persists: 3D assets do not travel reliably between pipelines. Every time an asset moves from a computer-aided design (CAD) tool to a simulation platform, critical physics properties, geometry, and metadata are lost—forcing teams to rebuild from scratch. SimReady, the content standard built on OpenUSD, defines what physically accurate 3D assets must contain to work consistently across rendering, simulation, and AI training pipelines. Combined with NVIDIA Omniverse libraries—which provide a physics-accurate, photorealistic simulation layer—AI models can be trained and validated before deployment, bridging the gap between digital and physical worlds.

Real-World Applications of the NVIDIA Physical AI Stack

Manufacturers are already applying this simulation-first approach with impressive results. Two leading examples illustrate how the NVIDIA Physical AI stack is driving efficiency, cost savings, and accuracy.

ABB Robotics Closes the Sim-to-Real Gap at 99% Accuracy

ABB Robotics has integrated NVIDIA Omniverse libraries directly into RobotStudio HyperReality, its simulation platform used by over 60,000 engineers worldwide. The platform represents robot stations as USD files running the same firmware as their physical counterparts, enabling training robots, testing part tolerances, and validating AI models before a production line exists. Synthetic training variations—such as lighting conditions and geometry differences—can be generated at scale, covering scenarios that would be impractical to replicate manually. As Craig McDonnell, Managing Director of Business Line Industries at ABB Robotics, explains: We’ve managed to vertically integrate the complete technology stack and optimize it to a point where we’re now achieving 99% accuracy on the simulated version. The downstream outcomes are significant: up to 50% reduction in product introduction cycles, up to 80% reduction in commissioning time, and a 30–40% reduction in total equipment lifecycle cost.

JLR Compresses Aerodynamic Simulation from Four Hours to One Minute

JLR applied the same simulation-first principle to vehicle aerodynamics. Engineers trained neural surrogate models on more than 20,000 wind-tunnel-correlated computational fluid dynamics (CFD) simulations across the vehicle portfolio. As a result, 95% of aero-thermal workloads now run on NVIDIA GPUs, slashing simulation time from four hours to under one minute. This dramatic speedup allows designers to test far more iterations, optimize fuel efficiency, and accelerate development cycles without sacrificing accuracy.

These examples underscore a broader trend: Manufacturing’s simulation-first era has arrived. By adopting OpenUSD and the NVIDIA Physical AI stack, companies can reduce costs, speed time to market, and build more reliable production systems—all while maintaining the fidelity needed for real-world deployment.