NVIDIA has detailed performance specifications and real-world deployment cases for its Jetson edge AI and robotics platform, highlighting compact hardware modules capable of running generative AI models directly on physical devices.
Venture capital investor Sarah Guo featured the platform in a video demonstration, showing how developer kits like the Jetson Orin Nano Super fit inside a handbag while delivering edge compute power for robotics and autonomous systems.
Hardware Specifications
The Jetson platform spans multiple performance tiers designed for different edge workloads, according to NVIDIA:
- Jetson Orin Nano Super: Delivers 67 trillion operations per second (TOPS) of AI performance for generative AI and computer vision prototyping.
- Jetson AGX Orin: Provides 275 TOPS of AI compute, targeting advanced robotics, autonomous navigation, and industrial automation.
- Jetson AGX Thor: Offers up to 2,070 FP4 teraflops of AI compute and 128GB of memory for humanoid robots and complex real-time reasoning.
Real-World Deployments
NVIDIA outlined several projects built on the Jetson hardware. A toy electric vehicle model called SidewalkPilot uses the Jetson Orin Nano Super to execute autonomous maneuvers. The Reachy Mini Jetson Assistant runs a low-latency voice and vision assistant locally on the same module without requiring cloud connectivity or internet access at runtime. Developer Lewis constructed a robot using the open-weight Mistral model on Orin Nano Super, while Asier Arnaz built a Yocto-powered video podcast platform featuring two interacting AI models.
For higher-compute workloads, a robotics team at Carnegie Mellon University deployed the Jetson AGX Orin to map 3D environments while searching for survivors in rescue scenarios. At the University of Illinois Urbana-Champaign, the SIGRobotics team used Jetson AGX Thor to run the NVIDIA Isaac GR00T N1.5 model on two robotic arms that autonomously prepared and whisked matcha tea.
To support deployment, NVIDIA introduced Jetson Device Skills and Jetson BSP Skills to help developers use coding AI agents for edge optimization. The company is also launching a three-module livestream series focused on running generative AI, building claw agents, and deploying vision-language-action models on Jetson hardware. NVIDIA did not publish pricing updates for the developer kits in its report.
