Amazon Go

These are purely assumptions based on what I saw on the video alone:

>If my friend and I walk out side by side, how does it know (if it does) which items are mine and which are hers?

Looks like the entry/exit is the same type of set-up you find at most large office buildings with the tap in/out gates (see screenshot: http://i.imgur.com/e7fDglY.jpg). I assume it only lets one person out at a time which also suggests your friend could only enter the store if they had an amazon go account and tapped in themselves.

>What happens when someone picks up an item and leaves without first doing whatever check-in/registration/setup is necessary?

Again, based on those gates, I'd assume you can't actually access the store without going through the necessary set-up first.

Personally, I'd like to know what happens if/when my phone battery dies in-store.

Or if you lose your internet connection. or if your app crashes.

It looks like you don't need to scan anything when exiting. So your phone could die once you're in the store and you'd still be able to check out.

I'm guessing the Amazon Go app generates a one-time-use QR code that gets scanned on entry and then security/tracking cameras follow you all over the store. If the cameras see you in front of an item when its RFID sensor detects it was picked up, they make an educated guess that you picked it up. Then they can re-scan all the RFIDs as you exit for extra confirmation.

i believe this is it too, mainly because of their mentions of machine learning and computer vision in the post.

I think you meant http://i.imgur.com/e7fDglY.png

They probably don't have RFID sensors on the shelving, and if they did they would only be able to detect when you walk several feet away from the shelf, and would have huge problems if two people were standing right next to each other at the shelf.

It looks like they have a row of cameras along the top of each shelf that will be used to detect when you pick up and place items back down.

The RFID tags are mostly useful at the gates to confirm the visual data and feed back into the machine learning algorithm.