It’s a good month to be a maker in Artificial Intelligence. March sees the launch of Nvidia’s Jetson Nano, the fourth board in their Jetson line for running robots and other autonomous machines. Earlier boards were more and more powerful, leading up to last year’s Jetson Xavier. Now, for the first time, Jetson is affordable to a hobby market — it costs just $99. If you’ve been interested in trying out Deep Learning to make a self-driven robot or other trained-rather-than-programmed creation, Jetson Nano occupies a sweet spot of power and price.
Nano is built for flexibility. It runs all the popular machine learning frameworks. If drawing 10W of power, comparable to a Raspberry Pi, is too much for your creation, it can down-power half its cores to run at 5W. And if you discover that you just can’t do without faster performance after making version 1 of your creation on Nano, the same deep learning models you built on Nano will run on the higher-end Jetsons, no changes needed.
Nvidia has launched Jetson with a few first projects to try out: HelloAI introduces Machine Learning concepts. Makers may want to skip straight to JetBot, a robot that can be put together for $250 in parts, Jetson included. Many hobby robots are limited to simple sensors like ultrasonic and laser rangefinders that tell them only how close the nearest thing in front of them is, so the robot knows when to dodge obstacles without knowing what it’s dodging. JetBot accomplishes the same feat with a camera, using its AI smarts to sort through actual pictures. Not only does it know when to dodge, run a second neural net in parallel and the Nano can know what it dodged, too.