Hacker News Reader

Comment by ChuckMcM | Hacker News Reader

I love this story, especially about ordering dough pizzas.

It reminded me of this twitter thread: https://twitter.com/meslin/status/1225834920611848192?lang=e...

In which the author tries to order the Uline "box of boxes", a box of twenty-five (25) 6" x 9" x 6" boxes, only to have Amazon deliver a 6" x 9" x 6" box containing some random product. The collection product from Uline has the same bar code as the box itself, so the pick up robot would scan the shelf for the box, find something that SOME OTHER VENDOR had put into a 6x9x6 Uline box, and pick that to satisfy the query.

Adding automation to a process that any human with visibility to the whole process would say, "Wait, that can't be right." ends up in misbehavior.

View on HN

The automated system should say "weight, that can't be right". Works fairly reliably in supermarkets, especially if you allow a bigger margin of uncertainty on sku weight. Or verify the volume/shape - or both, like volumetric weight that shipping companies use. I'm guessing amazon has some system to automatically assign boxes based on the dimensions of stuff that's being picked.

Not all distribution centers validate weight, don't recall if Amazon does but I would be surprised if they don't. High speed scales are expensive.

A few people got a box of rocks instead of a $4k camera awhile back, one guy got two boxes of rocks after returning the first.

Thinking about this more, I'm surprised this couldn't be caught when the items got delivered to Amazon. I find it extremely unlikely that a picking robot accidentally found an item with the same barcode if they weren't already in the wrong place.

Seems like when the items arrived at the fulfilment centre, they got mis-scanned and ended up comingled. Presumably that's the stage where you can check weight and volume - eg does this item fit with the known dimensions.

This check must be made somewhere otherwise people wouldn't bother returning high value electronics with rocks inside (presumably someone does a cursory check of weight before the inventory gets comingled again).

It's not unusual for warehouses to deal with items with multiple barcodes.

For a technology example, you might get a network card with a UPC barcode, but also a MAC address barcode, and a manufacturer's part number barcode. A wholesaler/ manufacturer sells boxes of 20 items to resellers? The outer box will have a barcode. That box got sent by courier? Three barcodes on some mailing labels.

So at a goods-in station, the usual response to "multiple barcodes, some don't make sense" is "Keep trying until you find one that does make sense"

Those multiple barcodes also confuse the crap out of self-check-out stations. Once it picks up on the wrong one, it throws a fault that the attendant has to come back out of, so you can't just back out scan the right one.

What baffles me is that the store/software can't build a dictionary of "known misdirection" barcodes, like "this is the shipping carton not the product itself, fault and tell the user to rescan, but don't just lock up" when they're seen.

Scanners in distribution centers contain a barcode hierarchy that filter out unused barcode types. For example, after something has been picked, a UPS label might be applied to it. Subsequent scanners will be programmed to only read either the internal routing label or the UPS label (or both and let the software handle it).

The scanners in supermarkets are made by the same companies and I would assume run on the same software. However, it might be more difficult to ensure a specific type of barcode is used on 100% of products in the store. So if you see the scanner picking up the wrong barcode at the check-out aisle, it is most likely either the scanner is not programmed correctly, or the scanned barcode is the same type as a valid barcode used somewhere else in the building.

As far as handling faults, that would probably be done on the POS system, not the scanner itself. The scanner software is perfectly capable of handling errors in different ways (for example, sending a specific code to indicate two different non-matching barcodes were read in the same pass), but from a functional standpoint, the scanner is 'dumb'.

Weight can easily be gamed by adding a bag of water.

Verifying shape would be difficult without knowing what to expect beforehand, but can also be gamed.

You won’t get the precision of supermarket weighting if you are dealing with diverse restaurants with a dynamic menu.

Unless I'm missing something, this thread is talking about the automated system making a mistake on its own. No one is trying to game the system.

> Works fairly reliably in supermarkets

No it doesn’t.

Not just because they can be badly calibrated, but also because the range of weight they have to deal with must make it hard to manage any sane range.

The only SCO I’ve seen doing a decent job at dealing with weight use a binary check (“was there any product at all added to the to total weight of the basket ?”) and they still miss products like lollipops or anything too light to pass the range.

Sure much more reliable system could be built, especially at Amazon’s engineering scale. But so far supermarkets are mediocre at best at this game.

Works great in any supermarket I've been to with a self-scanning systems and that's been the standard in my country for many years now.

“Unexpected item in bagging a area” gets me every time.

Same. Have to have a clerk unlock the machine 3 times on bad days. It feels like the shop doesn't want me to shop there.

It used to be like that the first couple of years but the newer systems never give me that and feel very accurate.

I used to pickup lunch from Tesco most days. Occasionally you'd find one item that must have been weighed wrongly at the input stage. It would cause an issue every time you checked out. Usually got fixed within a couple of days.

I wonder if they built in correlation of items which on a mis-weighed basket.

I haven't heard that phrase in years. There were some growing pains when self-checkout systems were first introduced, but every one I've interacted with in recent memory has been re-calibrated to be much more permissive.

Speaking for France, a lot of supermarket who introduced self-checkouts have further invested in developing scanning apps operated by the client on their phone as a parallel solution. Basically a low-tech amazon-go solution, where the client is its own casher.

Those bring their own set of issues, can be difficult to understand and use from the customer, yet they still felt way easier to deal with than the SCO experience.

We've had this for years with Waitrose, who just give you a scanner that you can take round the store. I think more recently they let you do it on the phone. Then you just pay at a special checkout and leave. They also have no-scale checkouts where you scan and leave (these are still attended). Usually if you buy something very expensive it'll flag up for a "random" check which always happens to be the expensive item.

The handheld scanners are easy to use. All it has is a trigger for scanning (which everyone knows how to do), scroll buttons, and a delete button if you make a mistake.

I hated the weight check at Tesco when I lived in the UK. It unnecessarily locked up my self service checkout a few times per week and in general made things slower.

Compared the AH in the Netherlands where you just scan the barcode on all your things and they really don't care about the weight or where you put them after scanning.

I'm sure there will be a tiny percentage extra fraud that Tesco may catch with this, but given the choice I don't shop there due to the shitty user experience. That's got to cost them more in lost revenue than the fraud they stop.

Yes, it's really annoying. It's that bad that I normally just queue for some personal touch with the cashier :)

For vast majority of things amazon sells, a system checking if the weight is within a pound or 25%, whichever is greater, would almost never have a false positive.

So a pound is 453g.

I checked my order history and my last orders are around:

- 350g

- 600g

- 200g

- 500g

- 5kg

- 120g

a box with empty packing would fit within a pound for most of these. I don't intend to nitpick your back of the enveloppe calculation, just that it's not as simple as it seems.

I don't know if I am the typical amazon shopper, but on my 35 orders in the last 6 months the above pattern is repeating with mainly very small items (like cables, dongles etc.) and one big heavy package from time to time.

I think an important part of the equation would be knowing the weight of the boxes you're using as well.

At that level of detail, it also requires the seller to give an exact weight for the goods they are selling, and update it each time it differs by even dozens of grams.

It can be done, just not sure the parties involved are willing to commit to that level of accuracy.

This is called a "tare weight", FYI.

I just had a package from Amazon that literally weighed less than the product that was supposed to be in it. I peeled off the shipping label and found another one underneath that was labeled "weight error". Someone had apparently intercepted it, overridden whatever check was going on, and sent it out anyway.

Weirdly, Amazon still made me mail back the underweight one (I would happily have paid the correct amount for the smaller quantity to save the trouble of remailing) and a few days later sent out the correct one. You'd think in this case they have a record of the actual weight of the package and could sort things out instantly, but apparently not.

A couple of months ago Amazon shipped me a completely empty, sealed envelope labeled as 1 pound. At least Amazon support immediately shipped me the correct item.

Hah that box story is hilarious.

The actual automated fix should be during stock intake I assume. We ship international parcels with DHL etc and it's a clear stipulation there are NO BARCODES at all on the box apart from the mailing barcode that we generate. Being Amazon, nothing probably happened because they have more money than they need, but I assume they have similar rules and would be justified to ding the suppliers for this cost.

But Amazon isn't intentionally trying to lose money here.

I hadn't realized that cardboard boxes were that expensive!

Reply on news.ycombinator.com