> You place all your items on the white shelf with some space between them. Although they were clearly designed to be a self-checkout experience, the stadium had a staff member rearrange your items, then for about 30 seconds the kiosk would be thinking. After, it would pop up all items on the menu, and the staff member would have to tap to confirm what each item was.
Maybe we're just calling all forms of automation and computer vision "AI" these days because it's sexy. Anyway: any automation that requires a human staff member to intervene to complete every run is not automation: It's just adding unnecessary technology and making the process worse. Imagine if each grocery store self-checkout required a human staff member to scan items, re-arrange things, and confirm checkout.
> Maybe we're just calling all forms of automation and computer vision "AI" these days because it's sexy.
Funny thing is, at first it was the other way around! 'Computer Vision' has always been a sub-field of AI, but the term was more widely used by academics during a previous AI winter as a way to avoid the tainted 'AI' label.
They do that at Circle K[0] today using the same tech from Mashgin. It's meant to be a self-checkout, but you literally have one employee standing and watching this one checkout (sometimes 2-3, but usually 1-2). It's not always accurate, requires some hand-holding at times, and slows down the already slow lines at Circle K. It's a bit faster than the article implies and does not require a staff member, but still slower than a human would be.
Meanwhile over at QuikTrip, there's one person checking out two people at a time. Suffice to say, if both stores are available, I will always choose the QT.
I use circle-K because it's like 3 blocks from me. The self-checkout seems to work fine. I buy alcohol on the rare occasion I drink, and even then the cashier just steps to the side for one moment to check my ID, presses a button, and then goes back to what he's doing with someone/something else. It never overcharges me, at least, and it always seems to pick things up in just a couple moments. I suppose I take the time to space them out a bit, but I always seem to have a perfectly reasonable experience.
That being said, I see the same one-person-two-registers thing at 7-eleven, and it's very, very fast.
> any automation that requires a human staff member to intervene to complete every run is not automation
Not strictly true. Barcode readers are used by humans and are definitely automation. The ironic part though is that the automation going on here is literally object classification, which humans are good at.
The play may be to collect data and make their system better.
I came here to paste that quote. It sounds like a disaster of poor HCI around not-ready-for-prime-time tech implementation. That's not even a great PoC demo to get funding, much less deploy into production.
Maybe they were emboldened because many companies still can't even do decades-old UPC barcode scanner self-checkouts well?
The closest self-checkout to working reasonably well I see regularly would be at Whole Foods Market, at least just using the low-tech UPC and scale. I only have a few nits about it.
(Though, within the last week, the usual duct-taped-on off-the-shelf hand scanner apparently saw the wrong barcode on the front of the product label (yes, some brand did that), which wedged the station, and the employee who came over seemed like they might think I was trying to defraud the store. I've coded for a few of those scanners before, and they provide a mix of automagical easy high-value behavior and major pitfalls. There are a few kinds of interfaces, and a large fleet of settings, and you really have to wrestle the device to the ground, to make every scenario bulletproof. If the integrator wasn't careful, for some of them, you can even reprogram or brick them with an in-band barcode, and disabling this feature is buried among the numerous settings.)
The worst self-checkout I'm currently exposed to is the dumpster-fire of a major chain, which goes out of its way to fill the UI with garbage, and then doesn't do even basic sensing and state flow right. They really need to look at WFM design, and then go even further in that direction, and get the state model right. While making sure that no one's bonus is tied to garbage and dark patterns on the screen.
(Also, for return customers who nope right out of the self-checkout headache, and go to the human checkout, or get directed to it by the attendant who's glaring at all the self-checkouts, they need clean their CC terminal keypad that's visibly caked with crud, like maybe it hasn't been disinfected in a year. Especially since they mandate repeated use of it when it should default to working with just a card tap/swipe, for a high-traffic location for many sick people.)
> Imagine if each grocery store self-checkout required a human staff member to scan items, re-arrange things, and confirm checkout.
They always have at least one person going between each self-checkout kiosk helping confused and upset customers. Meanwhile, 1 traditional checkout lane is open with a long line. Self checkout is great to use if you know how and have a handful of items, but it sucks with a full cart due to space constraints and the bag scales being finicky.
I wish. The Wal-Mart near me no longer has staffed checkouts between 6am (opening time) and 8am. That's two hours in the morning of robots-only. I don't know about in the evenings.
traditional checkout lane is open with a long line.
I use the traditional check-out line whenever I can because where I live, the self-check line is almost always longer. It's not hard to keep an eye on the last person in the self-check line when you go to a real register to see which is faster.
I'm not a fireman on call. I'm OK spending an extra 45 seconds in a traditional line to keep a low-skilled human being employed.
I don't even think it's about going faster. In my experience, a traditional checkout staff is much faster than the self-check out. First of all, it's fantastically batched and pipelined. While the person in front of me is being checked out, I load all of my groceries onto the conveyer. Then, when it's my turn, the clerk does one motion over and over very quickly, and puts it on another conveyer that whisks it out of the way. When there's dedicated bagging staff, it's even more parallel.
Contrast that with self-checkout: There is no conveyer. You have to reach into your cart, grab one item, run it across the scanner, and place that item in a bag, then reach back over into the cart, grab one item, and so on. No pipelining at all.
I go the traditional staffed checkout route for the speed.
This probably doesn't work for groceries, but Uniqlo (a clothing retailer) tags all their items with RFID. You put all your items in the "checkout box" at the same time and just pay, no scanning required.
Australia has had self-checkout in supermarkets and larger retailers for nearly 2 decades now.
Usually you will have a single staff member responsible for 4-10 checkouts to override the machine when your product doesn't register a weight, or you move items off the scale too quickly, and it wants them to check it.
It generally just works and is a lot quicker if you're just scanning a few items; surprised it hasn't really taken off in the US.
Most of the issues these days are when they introduce new features like less tolerance on the weight (sometimes adjusting already scanned items trips an error) or auto-scanning fruit and vegetables.
I think the major difference is whether or not the machine weighs the bagging area. When it's not weighing it, you can scan fast and any weirdness is ignored. If it does check the weight, you have to not only deal with wrong weights in the system but also moving anything around or very light items will trip the alert. Generally the nicer the neighborhood, the better chances you'll have finding a non-weighing self-checkout.
I'm not stupid, I know why these measures exist but there's likely a smarter way to let the small things go. Find a way to add percentages to the alerts so that it won't trigger if I rearrange the bags. Factor in the price of an item as well so that it's triggering on meat that doesn't weigh right versus a can of beans.
Maybe we're just calling all forms of automation and computer vision "AI" these days because it's sexy. Anyway: any automation that requires a human staff member to intervene to complete every run is not automation: It's just adding unnecessary technology and making the process worse. Imagine if each grocery store self-checkout required a human staff member to scan items, re-arrange things, and confirm checkout.