NVIDIA (NASDAQ: NVDA) announced the release of open source physical AI agent skills and tools designed to streamline development workflows for robotics, autonomous vehicles, vision AI and industrial digital twins, according to a company statement.

The collection includes tools spanning NVIDIA Omniverse, Cosmos, Alpamayo and Metropolis platforms. The physical AI skills convert complex training, evaluation and deployment workflows into agent-executable instructions.

The announcement was made at NVIDIA GTC Taipei. Jensen Huang, NVIDIA's founder and CEO, stated that AI agents are extending into physical AI systems that will affect transportation, manufacturing, healthcare and robotics.

The tools are part of NVIDIA Agent Toolkit and include Cosmos world foundation models, Omniverse libraries for simulation and digital twins, Isaac for robotics simulation, Metropolis for vision AI, Alpamayo for autonomous driving, and the Jetson platform for edge AI development.

Several companies are implementing these tools. TSMC and Pegatron are using them for visual inspection models, with Pegatron reporting a 67% reduction in model training and deployment time using synthetic data. Delta Electronics improved detection rates by 17% for excess soldering detection. Foxconn reported approximately 3% improvement in first pass yield.

In autonomous vehicles, Li Auto, Afari and DeepRoute.ai are using NVIDIA Omniverse NuRec models, generating over 1,000 reconstructions and more than 300,000 renders and simulations daily.

Industrial partners include Cadence, Dassault Systèmes, Siemens, Synopsys and PTC, which are using NVIDIA Omniverse libraries for engineering data inspection and digital twins.

The tools are available through GitHub and skills.sh. Microsoft, CoreWeave and Nebius are integrating these agent skills with their cloud services.