NVIDIA expands Cosmos toolkit for building world models

Summary

At GTC Berlin, NVIDIA expanded Cosmos, its platform for world models that let physical AI systems understand and simulate the real world. Training, optimization, and deployment recipes are now packaged as agent skills that work with open models such as GR00T and QWEN, alongside new blueprints for synthetic training data and Cosmos 3 Nano variants for simulation. A company-cited example says Porsche should be able to cut repeated physical test drives by up to 90%.

Why it matters for our work

It accelerates the shift from physical trials to synthetic data and simulation in robotics and autonomous vehicle development. Open tooling for iteration-heavy industrial research can directly reduce time and cost.

Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.