NVIDIA has released NemoClaw, a blueprint for building safer AI agents that can automate physical AI workflows on local machines or in the cloud. The toolkit, unveiled at GTC Taipei and Computex in early June, includes new agent-callable skills designed to speed up robotics development, autonomous vehicle testing, visual inspection, and industrial automation. Companies already using these tools report significant efficiency gains, from faster model training to reduced defect detection times.

What Is NemoClaw and Why Does It Matter for Robotics?

NemoClaw is part of NVIDIA's broader Agent Toolkit, a collection of open-source tools that turn complex physical AI development tasks into repeatable instructions that coding agents can follow. Rather than requiring engineers to manually set up simulations, generate training data, and deploy models, agents can now handle these steps autonomously. The blueprint works alongside NVIDIA's OpenShell runtime, which provides policy-based security and privacy governance, allowing agents to run safely on edge devices like NVIDIA Jetson platforms or cloud infrastructure.

The significance lies in speed and scale. Physical AI systems, which power robots and autonomous vehicles, traditionally require massive amounts of training data collected in diverse environments. By automating data generation, simulation, and model training through agent-callable skills, developers can compress workflows that once took weeks into processes that now take hours.

"With stack updates in NVIDIA Isaac GR00T, new end-to-end workflows can be set up in hours versus weeks," explained Rev Lebaredian, vice president for physical AI simulation at NVIDIA.

Rev Lebaredian, Vice President for Physical AI Simulation at NVIDIA

How Are Companies Using NemoClaw and Agent Skills Today?

• Manufacturing Defect Detection: Pegatron reduced model training and deployment time by 67% using synthetic data generated from the Defect Image Generation skill. Delta Electronics improved defect detection rates by 17% on metal busbars, while Inventec cut defect data collection effort by 30% for laptop chassis manufacturing.
• Autonomous Vehicle Development: Self-driving companies Li Auto, Afari, and DeepRoute.ai are using NVIDIA Omniverse neural reconstruction models to generate photorealistic driving scenarios. These teams have produced more than 1,000 reconstructions and over 300,000 renders and simulations per day to train safer autonomous systems.
• Hospital Robotics and Healthcare: Foxconn and Compal are deploying NVIDIA Isaac for Healthcare to accelerate robot development in clinical settings. Foxconn is scaling its Nurabot across multiple hospitals and long-term care facilities, while Compal is advancing its PolyMedX robot toward hospital-wide orchestration platforms that integrate simulation, AI, and real-world operations.
• Industrial Digital Twins: SK hynix is implementing semiconductor fab digital twins using NVIDIA Omniverse as part of its "Autonomous Fab 2030" roadmap. The chipmaker is collaborating with NVIDIA and SK Telecom to validate the NVIDIA Agent Toolkit specifically for manufacturing-scale physical AI applications.

What Skills and Tools Power NemoClaw?

NVIDIA optimized its entire physical AI stack to work with agents by converting libraries, models, and frameworks into agent-callable tools. These include Cosmos 3, a world foundation model that understands videos and text to predict physical outcomes and generate actions; Omniverse libraries for simulation and digital twins; Isaac for robotics simulation and robot learning; Metropolis for vision AI; and the Jetson platform for edge AI development.

The agent skills themselves are designed to automate specific development tasks. For robotics, skills can accelerate perception and mobility training data generation, simulation, navigation training, and Jetson-based edge system tuning. For autonomous vehicles, skills can reconstruct fleet-captured data into simulation environments and run closed-loop reinforcement learning to expand training coverage. For vision AI agents used in automated inspection, skills can generate synthetic training data, fine-tune models, and automate labeling.

"AI agents are revolutionizing software development, and that shift is now coming to physical AI, extending into the systems that will transform transportation, manufacturing, healthcare, and robotics," said Jensen Huang, founder and CEO of NVIDIA.

Jensen Huang, Founder and CEO of NVIDIA

Which Companies Are Already Adopting NemoClaw?

Early adoption spans robotics, autonomous vehicles, and industrial software. Robotics companies including 1X Technologies, Agile Robots, Agility, FieldAI, Hexagon Robotics, NEURA Robotics, Skild AI, and Universal Robots are using NVIDIA's agent-ready physical AI stack. In electronics manufacturing, TSMC and Pegatron are fine-tuning visual inspection models with agent-generated synthetic data. Industrial software leaders Cadence, Dassault Systèmes, Siemens, and Synopsys are using NVIDIA Omniverse libraries and skills for engineering data inspection and interactive digital twins.

The real-world impact is measurable. Foxconn, working with DeepHow, improved manufacturing efficiency by catching errors early, boosting first pass yield by approximately 3%. These gains suggest that agent-driven automation is moving beyond proof-of-concept into production environments where efficiency directly affects profitability.

How to Get Started With NVIDIA Agent Skills for Physical AI

• Assess Your Workflow: Identify which parts of your physical AI development pipeline are most time-consuming, such as data generation, simulation setup, model training, or deployment tuning. Agent skills are designed to automate these repetitive tasks.
• Explore the Agent Toolkit: Access NVIDIA's open-source Agent Toolkit and review available skills for your use case, whether robotics, autonomous vehicles, vision AI, industrial digital twins, or healthcare applications.
• Deploy With NemoClaw and OpenShell: Use the NemoClaw blueprint to safely build and deploy autonomous agents on local Jetson hardware or cloud infrastructure, leveraging OpenShell's policy-based security and privacy governance to ensure safe execution.
• Combine Skills for Complex Workflows: Stack multiple skills together to orchestrate end-to-end workflows such as data generation, simulation, optimization, inference tuning, and continuous evaluation across your entire development pipeline.

NVIDIA's release of NemoClaw and agent-callable skills represents a shift in how physical AI systems are built. By automating the time-consuming, repetitive parts of robotics and autonomous vehicle development, agents can free engineers to focus on higher-level design and validation. Early results from manufacturers, autonomous vehicle teams, and healthcare providers suggest that this approach can deliver substantial gains in speed, efficiency, and quality.