NVIDIA Enables the Next Era Of Physical AI Research With New Agent Skills Powered by Cosmos 3

NVIDIA is unveiling new physical AI agent skills at CVPR. The skills help researchers and developers speed the development of autonomous vehicles, robots and vision AI systems. They address the challenge of fragmented workflows in physical AI research. NVIDIA announced NVIDIA Cosmos 3 earlier this week. It is an open frontier model for physical AI. The model is the world's first full omnimodel unifying vision reasoning, world and action generation. For autonomous vehicle research, the skills let AI agents automate scene reconstruction from fleet data and generate synthetic scenarios. Neural Reconstruction skills turn fleet-captured data into editable 3D scenes for simulation. NVIDIA Alpamayo 2 Super is an open 32-billion-parameter reasoning vision language action model. New NVIDIA Metropolis skills help generate synthetic visual scenarios including anomalies. The Defect Image Generation skill creates examples of defects across different surfaces. Skills for video search and summarization extract insights from massive volumes of video data. NVIDIA Isaac Sim 6.0 includes agent-friendly skills and connectors. Isaac mobility skills automate navigation workflows. Cosmos-H-Surgical-Simulator generates realistic surgical robotics data for policy training. NVIDIA technologies were referenced in the majority of accepted CVPR 2026 papers. NVIDIA researchers are presenting work at the event running June 3 to 7 in Denver. The company is running open research challenges including the AI City Challenge and new benchmarks for autonomous driving.