en.Wedoany.com Reported - NVIDIA has released physical AI skills and tools that break down development tasks in areas such as robotics, autonomous vehicles, and industrial digital twins into fine-grained tasks that AI agents can autonomously execute. These skills, launched as part of the NVIDIA Agent Toolkit, allow agents to directly call NVIDIA's libraries, models, and frameworks to accelerate data generation, simulation runs, model training, result evaluation, and final deployment.
NVIDIA founder and CEO Jensen Huang stated that AI agents are transforming software development, and this shift is extending into the physical AI domain, deeply impacting transportation, manufacturing, healthcare, and robotics systems. When agents can directly use NVIDIA's libraries, models, and frameworks, physical AI development will accelerate.
NVIDIA is restructuring its physical AI technology stack around agents, transforming libraries, models, and frameworks into tools that agents can directly invoke. This includes the Cosmos world foundation model, Omniverse libraries, Isaac robot simulation, Metropolis visual AI, Alpamayo autonomous driving, and the Jetson edge AI platform. The new skills introduced in the Agent Toolkit convert physical AI development into repeatable, step-by-step instructions for coding agents to follow, specifying the tools to call, the outputs to generate, and the methods to verify result reliability.
In terms of security, developers can build and deploy autonomous agents using the NVIDIA NemoClaw blueprint and the NVIDIA OpenShell runtime, adding security and privacy governance based on policies.
These skills accelerate agent development in the following areas: robotics and edge AI, autonomous vehicles, real-time visual AI agents, industrial AI, and healthcare. Robotics developers can accelerate processes from perception data generation to simulation training, navigation automation, and edge system deployment; autonomous driving developers can reconstruct fleet data into simulation environments, generate realistic driving scenarios, and run closed-loop reinforcement learning; visual AI teams can generate synthetic training data, fine-tune models, automate annotation, and build video AI agents; industrial software developers can convert engineering data into CAD assets required for digital twin simulations and optimize large-scale OpenUSD scenes; healthcare teams can complete digital twin creation of hospital environments, simulation-to-reality data generation, and software-in-the-loop strategy testing before clinical deployment. These skills can be combined and integrated into larger agent systems, supporting developers in orchestrating and automating complex workflows, including data generation, simulation, optimization, inference tuning, and continuous evaluation.
NVIDIA's physical AI agent tools and skills are now open-sourced via GitHub and skills.sh, available for use by any coding agent.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
