Coral is a range of products from Google Research that allows you to create your own AI projects for prototype or production products. They share occasional projects they’ve cooked up in the lab, publishing most recently a teachable sorter project that uses machine learning to sort objects, specifically marshmellow cereal.

Gautam Bose and Lucas Ochoa created the contraption to sort their favorite cereal, separating the marshmellows bits from the icky non-marshmellow bits. Overall, it’s a pretty clever (and simple) bit of AI mechanism making.

There are four components to the device: the Singulator (where all the bits to sort are separated into each item), the Decider (that views each item and determines what it is), the Tippything (that sends the item one way or the other), and Start! (that starts the training or sorting).

After the build is complete, they trained the sorter using the Teachable Machine website, a web-based tool to create and export machine learning models. A collection of images of each item to sort were taken then collated into a model that is uploaded to the Coral sensing module used as the inference engine for the sorter. The module is completely off-line that helps the Decider run very fast, tell the Tippything what to do, and sort until your sorting desires are met.

Before you start your own build, you’ll need a Coral USB Accelorator ($59.99), a Raspberry Pi ($60) and power supply ($10), and a 1/2″ rod (they used acrylic). Lucas and Gautum provide all the system hardware details (with links), the STL files for modding/3D printing, and step-by-step instructions to complete the build yourself.

Have a project you’ve built or find interesting? Share it with us here!

This post features affiliate links which helps support SolidSmack through a small commission earned from the sale at no extra cost to you!

Author