At Hughes, we're leading the way in the Managed SD-WAN revolution. But we also like to look at the advancements SD-WAN is supporting. With the increase in bandwidth and improved traffic shaping SD-WAN delivers, front-end technologies like autonomous checkout become more realistic possibilities. The most well-known of these are the Amazon Go stores. How you can bring that technology to your store and improve on Amazon Go's problems?
Question: “What’s wrong with Amazon Go?”
There’s far too much equipment involved
Have you counted the number of cameras? It would make for a great sweepstakes challenge: “Free food for a year to anyone who can correctly guess the number of cameras and sensors involved in an Amazon Go store!”
The more technology involved, the less scalable the solution, the more expensive to operate and maintain.
It’s an artificial retail operational concept.
Outside of vending machines, what other retail concepts can profitably sustain scalable operations based on perfect inventory management? With Amazon Go, every product must be positioned in its proper position AT ALL TIMES!
I did not appreciate the significance of this issue until I mistakenly picked up two chocolate bars and put them back in the wrong place. When I changed my mind and purchased one of bars, I was surprised to find out that I had been charged for the spicy chipotle bar, when I had taken the raspberry dark chocolate bar.
It’s a proprietary solution from Amazon.
Implication: there are 150,000 convenience stores in North America with a deep desire to see Amazon Go "go away."
So what’s the answer?
Simplify the solution with AI-powered computer vision enabled cameras and cheap shelf sensors.
With a single camera per 50 square feet, Zippin has a remarkably lighter technology footprint than Amazon Go. By using standard (i.e. cheap) shelf sensors, Zippin drives down the operational cost of maintaining the technology.
Eliminate the need for shelf sensors with smarter AI!
Standard Cognition’s Autonomous Checkout solution shocked us all with an AI that was smart enough to correctly identify product regardless of where they happened to be sitting on the shelf. It was able to even correctly identify products when your hand was partially covering the label! All of this WITHOUT shelf sensors!
Yes, there is a cost. The Standard Cognition solution requires a significant upfront investment involving actors of all kinds of body types spending days grabbing each product a variety of different ways in order to train the AI to properly identify products. ("pickup with left hand", "pickup with right hand", "pickup with right hand with a few fingers covering the label", etc.) But even this process has been artificially intelligently automated by extrapolating thousands of images into millions of variations of lighting and skin tones.
The elegance of Standard Cognition lies in its ability to deal with the inventory chaos of real-world retail! You don’t have to grab an item that has been artificially positioned on the shelf in an artificial manner. Employees can fill up the shelves with whatever fits where ever. Customer can grab what they need in whatever manner is natural.
And the store “knows” what to do!