Amazon Go

Here's my speculation on the technology behind it:

Tracking. A lot of tracking.

You don't need deep learning and car sensor technology to do simple rfid tag pick up/drop offs to detect what you have in your cart.

No, this technology probably fully tracking your every movement in the store. It may use rfid tags to detect what you have, but that is not the main tech.

When you walk in and scan your phone, a camera array will scan you to create a footprint on who you are and link it to your ID. Then your every movement will be tracked by various cameras throughout the store.

You walked 3 steps, took a step back and looked at the advertising on the right? We recorded it.

You went to the cereal aisle first? Picked up a box of cereal and then put it back in favor of another one? Yup, we recorded it too.

If this is indeed the case, then the correctness of what is in your shopping cart is going to be very, very high and there will be no need for an honor system, randomized checks, or other mechanisms to prevent inaccuracies.

Shoplifters will probably get away with the first shoplifting but will probably get profiled immediately and unable to do it multiple times.

Here's my rough sketch on how this could work:

a) You track every person from multiple camera angles, it should be "relatively easy" to track the position of every customer at all times. I'm not sure if it can handle people in very close proximity (or e.g. after hugging in store) - maybe Amazon actually uses facial recognition to resolve the identities afterwards

b) Only one person can enter the gate at once, and it's easy to authenticate the person at the starting point

c) Leaving is also a one-way gate, and at that point the system can simply mark the shopping as complete and mail the receipt

d) The shelves are specially designed and can sense (maybe with RFIDs) when an item is removed from the shelf, and match that with the person closest to the item. A possible difficulty is handling people who take an item from the back of the shelf (e.g. to get a fresh item).

for d) it could be a combination of rfid, closeness, and cameras detecting who grabbed it.