Day 2 – April 23, 2026
Assembling Digital Twins for Physical AI: Accelerate Sim-to-Real using NVIDIA Isaac Sim with Cosmos on Google Cloud
In this hands-on lab, you will explore how to dramatically enhance synthetic training data for Physical AI using NVIDIA’s Cosmos and Isaac Sim, all powered by Google Cloud G4 (RTX Pro 6000) instances. You will step through a complete, simulation-focused data-generation pipeline, starting by exploring a baseline environment in Isaac Sim to capture an initial dataset of robotic demonstrations. Next, you will enrich this baseline data using Cosmos, applying generative AI to create photorealistic, physically consistent variations with diverse lighting, textures, and object placements based on simple text prompts. Finally, you will conduct a side-by-side comparison of the datasets and policy rollouts, evaluating the visual diversity and robustness of models trained purely on standard simulation data versus those trained on the Cosmos-generated dataset to see firsthand how generative modeling on Google Cloud unlocks significant performance gains.
Read more