Amazon Go doesn’t need any checkout staff. This is the technology making it possible
Despite its online dominance, Amazon isn’t a stranger to bricks-and-mortar shops. Back in 2015 the firm opened a physical bookshop in Seattle and it’s promised to open a further 400 more in the next few years.
Seattle, the home of Amazon’s current headquarters, is also the testing ground for its first physical grocery store (if you discount those Whole Foods stores it purchased last year). On January 22, Amazon Go will open for business.
In many ways, it’s similar to any other convenience store. There are some packaged sandwiches, ready-meals and booze spread across the shop’s 1,800 square feet of floorspace. But unlike other stores it has no checkouts or human cashiers.
Instead of having to queue at tills, people’s Amazon accounts are automatically billed based on the items they’re carrying when they leave the store. All you have to do is download the Amazon Go app before you enter.
Amazon boasts it has “created the world’s most advanced shopping technology” to allow the system to work. It’s dubbed the tech ‘Just Walk Out’ and it says the systems work similarly to those in self-driving cars.
To remove human cashiers Amazon has created a store that’s able to run a high-level of surveillance on the people inside it. The New York Times reports hundreds of cameras are placed around the store. They’re on shelves and above aisles, but according to TechCrunch, they don’t use facial recognition technology.
Instead, the cameras use computer vision – the process of allowing machines to “see” what is in front of them and determine what an object is – to detect when an item has been taken from a shelf by a customer and who has taken it. The system is also able to remove an item from a customer’s virtual basket if it is put back on the shelves. By using a network of cameras, Amazon is able to track people in the store at all times, ensuring it bills the right items to the right shopper when they walk out, without having to use facial recognition.
Underpinning the computer vision is deep learning. At their simplest, the systems are basically advanced pattern recognition and allow for machines to draw conclusions from vast datasets.
As Recode has previously reported, there are some human employees working behind the screens at the Go store to help train the algorithms and confirm when they have correctly identified a product. Humans also restock shelves, help with product locations and are employed as fresh food chefs. However, the majority of the data collected by cameras is analysed in the same way the Amazon Echo recognises voices, in the firm’s large data centres.
The firm also says it uses “sensor fusion” during the shopping process. It’s likely this systems involves combining data from many sensors – these include weight sensors in the shelves to track individual products.
Amazon hasn’t announced what its plans for the Go store are in the longer-term. While it says more physical stores won’t be introduced in the immediate future, it’s unlikely the firm will have spent five years creating a complex system to just be used in one store. And global expansion may be on the cards: in 2017 Amazon registered Go trademarks in the UK.