Image: Amazon / Gizmodo

Amazon is still struggling to get its automated grocery store, Amazon Go, to function right. Who knew it would be so hard to build a brick-and-mortar store with sensors and gadgets instead of cashiers? But in a recent test of the company’s experimental Seattle store, some of Jeff Bezos’ employees got a little weird. They dressed up in Pikachu costumes to go shopping.

This bizarre detail appeared in a new Bloomberg report about the progress of Amazon Go. The stunt not only sounds ridiculous, it also sounds misleading:

Employees have tried to fool the technology. One day, three enterprising Amazonians donned bright yellow Pikachu costumes and cruised around grabbing sandwiches, drinks and snacks. The algorithms nailed it, according to a person familiar with the situation, correctly identifying the employees and charging their Amazon accounts, even though they were obscured behind yellow polyester.

So the ridiculous part is easy to spot. Amazon employees dress up as characters from an old Game Boy game and people who read about the experiment might assume that magical sensors somehow spotted the humans underneath their costumes. It’s a whimsical tale to be sure.

Now for the misleading part. We actually know from previous reports and Amazon’s own promotional video that customers scan their smartphones upon entering the store. We can only assume that Amazon Go uses some sort of beacon technology to track where customers’ through their smartphones as they wander the store and stop to collect items. Amazon hasn’t revealed exactly how the system works, but the Amazon Go promotional video features terms like “COMPUTER VISION” and “DEEP LEARNING ALGORITHMS” and “SENSOR FUSION.”

But despite the success of the Pokémon experiment, the store still has issues with the more existential questions of grocery shopping that are typically resolved by a human cashier. Bloomberg goes on to report:

The system is working well for individual shoppers but still struggles to accurately charge people who are moving around in groups, such as families with grabby kids, the person says. Go engineers have been studying families shopping together and are tweaking their sensors to recognize when a child eats an item while wandering around the store. Engineers are also figuring out which person to charge when a couple goes shopping together. Amazon has encouraged employees to enter the store in pairs and buy lunch.

Does the system get confused when two or more smartphones are close to each other in the store? And without a checkout line, who pays for items picked up by (sometimes non-adult) members of a group? It sounds like Amazon’s DEEP LEARNING ALGORITHMS aren’t sure.

Advertisement

We’re still waiting for Amazon Go to become a reality after The Wall Street Journal reported that its planned public launch in March was being delayed due to bugs, and hey, what a futuristic reality it would be. Some speculate that Amazon wants to install this technology in Whole Foods stores, which would really jazz up the premium grocery shopping experience. Until we see the Amazon Go system deployed in an actual store with actual customers (read: not Amazon employees dressed up as Pokémon characters), the whole thing remains an experiment.

[Bloomberg]