Announcing Cloud-Native Support for Jetson Platform

Key Takeaways

you're looking at a demo of new cloud native support for the Nvidia Jetson platform the demo is built around a service robot use case and this overview will highlight how cloud native technologies can enable rapid and ongoing feature improvements cloud native support allows you to use the same technologies and workflows that revolutionized cloud platforms to build deploy and manage AI applications on jets and edge devices when we launched the Jetson platform targeting AI at the edge a few years ago we focused on accelerating convolutional neural networks but Jetsons GPU architecture has the flexibility to accelerate both today's and tomorrow's algorithms we've continued each year to add support for the latest and greatest networks and we've continued to introduce new jets and modules all supported by the same software stack these are compact power efficient computers with the performance to run real-time AI applications Jetson Xavier NX is the latest member of the family with more than 10 times the performance and a smaller form factor than its widely adopted predecessor jetsam tx2 at the same price we're using the Jets at Xavier and X Developer Kit for this demo which is also available for the Jetson AG X Xavier everything you'll see runs locally on Jetson the demo includes four containerized applications the benefit of containers is that each can be independently developed deployed and updated for seamless improvements these four containers accelerates seven different neural networks at the same time as well as pre and post processing each covers a basic service robot need you can see for 1080p streams being processed and visualized by the first containerized app a service robot will use cameras to avoid colliding with people as it moves or in a retail setting for crowd counting or other purposes a great way to interact with a service robot is through speech and this container has an entire conversational AI stack running locally including a Bert model for natural language processing Bert enables very close to human accuracy and running it previously required powerful Cloud Servers now we can do it on device with Jetson Xavier pose recognition can help the robot better understand the meaning of a scene for example is a person pointing at something while asking a question again it's not just the AI inference that's accelerated video decoding and display are good examples of accelerated pre and post-processing lastly we have gaze recognition which requires three different AI models to detect a person's face locate their eyes and see what they're looking as with pose estimation this is useful for interaction if a person is looking at the robot what they are saying is likely intended for the robot technologies and workflows originally created for the cloud now let you develop deploy and manage AI applications at scale with the jetson platform check out the Nvidia jets and developer site to purchase a dev kit download the demo and join the developer forums