Mark-It: Amazon Go, without the AI (HackGSU F18)

Oct 30, 2018

Amazon Go stores are cool–you walk in, take an item, and walk out, and the cameras in the ceiling figure out how much to charge you.

All this camera and AI stuff is good PR for Amazon, but we decided to try to do better: this system of operation requires a lot of expensive gear to be installed throughout the store, and requires signficant investment in R&D. Its cost also rises linearly with the number of shelves that your store has.

So we designed a three-part system:

A tag inside a donut box
A tag inside a donut box

We were able to put all this together using stuff we already had–we taped our IDs to donut boxes, used my phone as a simulated door sensor, and enabled experimental NFC support in Chrome for the shopping cart.

One issue that we didn’t address in our demo is theft: what if someone walks through the door without adding an item to their cart?

Modern 860-960MHz RFID tags have a range of 2 meters. The door scanners would be ultimately responsible for determining which items the user is purchasing; the user would recieve a friendly notification that they may have not intended to purchase the item!