NVIDIA releases AI agent tools and skills to accelerate physical AI development for robotics, autonomous vehicles, and industrial systems

NVIDIA revealed open-source physical AI skills and tools at GTC Taipei and Computex, designed to help developers of robotics, autonomous vehicles, visual AI, and industrial digital twins reduce costs, time, and complexity of building physical AI workflows. The company is turning its libraries, models, and frameworks into agent-callable tools, including Cosmos 3 world foundation models, Omniverse libraries, Isaac for robotics, Metropolis for vision AI, Alpamayo for autonomous driving, and Jetson for edge AI.

The skills enable agents to automate repeatable development processes: robotics developers can generate perception and mobility training data and run simulations; autonomous vehicle teams can reconstruct fleet data into simulations and run closed-loop reinforcement learning; industrial software developers can convert CAD data into digital twin simulation environments. Developers can safely deploy agents using the NVIDIA NemoClaw blueprint and OpenShell runtime, which provides policy-based security and privacy governance.

Robotics companies including 1X Technologies, Agile Robots, Agility, FieldAI, Hexagon Robotics, NEURA Robotics, Skild AI, and Universal Robots are already using the stack. Pegatron reduced model training and deployment time by 67% using synthetic data from the Defect Image Generation skill; Delta Electronics improved detection rate by 17%; Inventec reduced defect data collection effort by 30%. Alpamayo 2 Super has been downloaded over 500,000 times.