Generate Synthetic Data for Physical AI with NVIDIA Cosmos World Foundation Models

This hands-on lab introduces roboticists, AI developers, and pipeline engineers to NVIDIA Cosmos world foundation models (WFMs) for synthetic data generation in robot learning pipelines. Participants will learn how to leverage simulation scenes from Isaac Sim and real-life images and videos to provide initial datasets and augment them with Cosmos Transfer, then evaluate data quality with Cosmos Reason—all within a production-ready robotics pipeline. Through this session, you'll experience an end-to-end data pipeline workflow, leveraging Cosmos WFMs to generate synthetic data from real-world and physically-accurate simulated data, and using Cosmos Reason as an intelligent critic to validate data quality. By completing this lab, you'll unlock the ability to generate high-quality training datasets, accelerating your action model development.