NVIDIA has extended its Jetson edge platform into agentic AI with the release of JetPack 7.2 and support for the NemoClaw agentic framework, bringing production-grade agent capabilities to robotics, inspection and industrial automation. The announcement, made at COMPUTEX and timed with GTC Taipei’s Build-a-Claw event, packages operating system improvements, developer-focused agent skills and NemoClaw integration on Jetson hardware including Orin and Thor.

Platform updates and performance gains
JetPack 7.2 delivers several foundational upgrades: Yocto-based OS support for lean, customizable Linux builds; NVIDIA CUDA 13 on Jetson Orin; a performance uplift for Jetson AGX Orin 32GB to 241 TOPS (about 20% above the original specification); and Multi-Instance GPU (MIG) support alongside a real-time kernel on Jetson Thor to reserve deterministic GPU resources for mission-critical perception workloads.

Fonte: Generated by ABN – Copyright free

Agent skills layer for faster deployment
Between the OS and NemoClaw, JetPack 7.2 introduces a new layer of agent skills that automates common developer tasks such as Linux customization, memory optimization and model benchmarking. These skills are derived from NVIDIA’s documentation and design guides and are deployable as agents, shortening integration timelines from weeks to days for many Jetson-based projects.

NemoClaw on Jetson: agentic AI at the edge
NemoClaw, NVIDIA’s agentic AI framework, now deploys to Jetson with a single command, enabling agent-driven automation on a production-ready robotics and vision stack. The pairing allows developers to combine onboard reasoning, perception and action, and to augment systems with Metropolis VSS blueprint skills for visual reasoning agents that interpret and act on camera feeds.

Industry deployments and partner use cases
Several companies are already adopting the stack. Solomon leverages NemoClaw for active perception on humanoid robots, integrating reasoning, sensor fusion and manipulation. Advantech is building an agentic factory brain with Nemotron 3 and Jetson Thor to automate fleet management and defect detection. SandStar reports nearly 40% memory optimization using Jetson Orin NX and NemoClaw, enabling migrations from 16GB to 8GB devices. NoTraffic, GROOVE X, Zipline and other robotics and automation firms are also deploying Yocto-based JetPack 7.2 in production scenarios.

Yocto ecosystem and vendor support
A range of Yocto ecosystem partners — including Balena, Konsulko Group, Peridio, RidgeRun and Wind River — offer distro products, engineering services and long-term support to help customers ship Yocto-based Jetson deployments. Hardware partners such as AAEON, ASUS, Avermedia, Connect Tech and YUAN have validated Yocto OS on their production edge systems.

Suggested Links
NVIDIA JetPack software page → https://developer.nvidia.com/embedded/develop/software
NVIDIA NemoClaw page → https://www.nvidia.com/en-us/ai/nemoclaw/
NVIDIA Jetson product page → https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/
Build-a-Claw event page → https://www.nvidia.com/en-us/ai/build-a-claw/
Jetson Orin product page → https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-orin/
Jetson Thor product page → https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-thor/