Walmart drops inventory robots from its stores | Hacker News Reader

Walmart drops inventory robots from its stores

bbc.co.uk

128 pointsvector_spaces99 comments

There's a few comments asking why this can't be done by subtracting sales from ordered inventory. All large retailers do this, I've worked on this problem, but there are lots of things that reduce the accuracy of the data-driven approach:

• Theft

• Spoiled/broken products

• Customers taking a product then browsing for another hour before checking out

• Customers buying multiple flavours of a product with the same price but different UPCs, which the cashier rings up by scanning the first product then hitting "times 10"

• There is stock on the shelf, but it's hidden behind something else or out of reach

• There is stock in the store, but not on the shelf (many retailers don't track backroom → shelf stocking)

• Product is stocked in multiple locations in the store, one of which is empty

• Plus all the usual data issues caused by wrong barcode scanned, data entry error, out-of-date information about units per box, etc

Those issues are all small, but still significant:

• For slow-moving products, being a few units off can mean the difference between out of stock and 1 month stock remaining.

• Even for fast-moving products these errors add up over time, you never really get a ground-truth event to reset the error drift. Stocktake and out-of-stock help but are infrequent and not a perfect reflection of reality.

There's a big unsexy industry working on this problem with data and human checks. The economics of robots, drones, smart shelves, and smart carts don't seem competitive right now. When you have 10k stores in the US, the expense is massive.

> Customers buying multiple flavours of a product with the same price but different UPCs, which the cashier rings up by scanning the first product then hitting "times 10"

I worked at retail (Uniqlo to be specific) and it was definetely a rule to not do this. They wanted us to scan everything one by one even if they were the same color and size. This allowed for a pretty accurate information on stock, e.g. we could see how many of a given sweater in given size and color is still in storage etc.

Probably this is too tedius for some retail (groceries with very small unit price) but very smart rule in general.

The rule is probably on the books at most retail stores, getting everyone to follow it all the time though is another matter. Cashiers and their direct supervisors could easily not understand or care about the rule and ignore it to save a few seconds on big orders.

90% of these problems are fixable with RFID tags, which are relatively cheap to roll out and maintain. Lululemon has been doing it for years[1], any insights about why it's not more popular?

https://www.globenewswire.com/news-release/2017/05/16/125712...

This is a great question...When I was in college in the 2000s RFID was going to be a game changer. The ability to track everything...the ability to just push your cart out and pay without scanning your items. Technology is there, but my only guess is somehow it's cheaper to have stolen items and less accurate counts.

I believe Walmart uses RFID extensively in their logistics/distribution pipeline, and they have a lot of control over their suppliers, so I'd assume they would use it in the store if it was viable.

https://www.atlasrfidstore.com/rfid-insider/walmart-and-rfid...

I'd be interested to know the precise reasons for not using RFID on items though, I suspect it might be something like "tagging $1 items is cost-prohibitive, and the system isn't useful if you don't tag every item".

I'd be more inclined to suspect that the sheer square footage of their stores makes it cost-prohibitive to outfit the space with the sensor infrastructure (comparing to Lululemon, who has much smaller stores). The tags themselves are fractions of a penny (I'm seeing $0.002 on alibaba) and are just applied as stickers, so I don't think that's too much of an issue -- not to mention Walmart probably has the buying power to insist their suppliers pre-tag all items.

> not to mention Walmart probably has the buying power to insist their suppliers pre-tag all items.

That wouldn't stop them from paying for the tagging. If it happens at all, everyone later in the supply chain has to pay for it.

Thieves could just pull off the tag or swap tags with a cheaper item.

If you want it integrated then your store must convince every manufacturerto modify their production line to put a tag inside the product packaging. Reading is still a problem also, RFID works great alone but what about reading a shelf full of products at once?

IMO computer vision (like Amazon Go) is a more likely solution to the problem.

They already can with barcodes. It's harder to forge an RFID then a UPC 128 barcode (print as sticker, apply at store)

Reading a shelf full of products isn't an issue, how do you think they can wave a wand over clothes and know the exact sizes they have and quantity?

They should double down on RFID but many of the manufacturers don't want to incorporate it in their process.

The processing time of applying in the store would be enormous there's a reason very few stores still use the price sticker gun to label items, it's time consuming when there's already a tag on each item.

Also in a number of places walmart and other stores literally just open a box provided by their sources and place it on shelves.

Does anyone have a (well-founded) idea of the amount of extra waste this would create? Or how much this would increase manufacturing costs?

This is a genuine question by the way, I honestly have no idea of the scale of this when applied to a store chain as huge as Walmart.

Bulk raw wafers can be obtained for a couple cents in 120k quantities but the tooling to apply them is where the cost is mostly. Also Walmart is a huge customer but not the sole customer for the brands they stock so they'd have to convince a lot of different companies to include it to make it worth while.

I think the second is a big part of why RFID tags haven't taken hold, stores need enough of their stock to use them to be useful while manufacturers need enough clients to want the tags embedded to go through the trouble of including it in the packaging.

https://www.mouser.com/ProductDetail/NXP-Semiconductors/SL3S...

The tags are still much too expensive for grocery and low-end retail. The ~US$0.10 to buy and apply the tag really eats into profit for those segments, where average price is a few dollars and margins are a few percent.

10 cents is likely closer to being the raw cost of the tag itself; the actual costs including tooling and tracking would likely make this closer to 15-20 cents per unit at scale. It's still too high for a low-margin retail environment, where you're competing with services like Amazon Go's camera systems (including the removal of overhead associated with cashiers) which have a fixed upfront cost regardless of unit volume.

Lululemon sells a small number of luxury items at a margin of 90%. (They don't capture that entire margin for themselves, but the presence of this margin also affects their relationship with their suppliers.)

Walmart sells bulk poverty goods with a margin of 1%. It's often the only store in town it's customers can afford to shop in.

Back in the late 1980s I worked (indirectly) for a clothing rental place and looked at RFID tags to replace the barcodes that were iused to track individual garments.

Barcodes are not ideal on something that is flexible and gets cleaned a lot, but RFID was discarded as a solution then.

My point is that RFID tags have had over 30 years in the market. If they were a better general solution than barcodes, they would be everywhere.

RFID tags are problematic in a lot of environments. Warehousing in particular, especially with containers of fluid or other dense goods that can block radio signals. I can see a lot of problems at a place like Walmart. A shelf full of cleaning products sounds like a nightmare scenario.

Source: I worked at a company selling RFID-based inventory tracking hardware.

RFID is definitely not cheap, especially at scale compared to barcoding.

One unmentioned issue is that RFID is also not universal. In that, each country has it's own part of the EM spectrum regulated for RFID. And most of these bands don't overlap. So you need one tag for the US, one for the EU, one for Japan, etc.

Do you have any firsthand knowledge of how well lululemon's implementation actually works? that seems like the sort of press release where corporate and the implementation partner are very proud of their thing, but the frontline staff who have to actually manage the inventory roll their eyes and fix the computer manually every time it's wrong.

A few other HN comments on what is holding back RFID tags:

https://news.ycombinator.com/item?id=22529512

Yay, more e-waste!

I have some experience of RFID from working for a UK supermarket in their supply strategy team, who did a very limited partial rollout.

The reason is that your original premise is not true - they are not relatively cheap to roll out or maintain (relative to the cost of goods).

They are not cheap to roll out:

- Readers in all stores, including handheld devices, tills, doors e.t.c.

- If you are tagging 100% of products, it affects every single supplier production line, of which there are tens of thousands. A cost to your supplier is a cost to you, because it will inevitably affect the cost of goods sold.

- IT cost implications both internally and at your suppliers.

- If you don't get 100% of suppliers adhering, you don't get as much benefit. Good luck getting Coca-Cola to change their production lines for no benefit to themselves!

They are not cheap to maintain:

- At an estimated 2p per item, if you are shipping 10 billion items that is £200,000,000 a year just in tag costs. Alternatively put, if you sell items for £1 on average and make 4% net profit, that cost of tag cuts deep.

- There is an incremental cost for suppliers to maintain all that equipment to apply tags or build them into the packaging (which comes back in COGS).

- Maintenance of all RF readers

And inventory routines are cheap relative to this cost: - Assuming you have 1000 stores, and spend 5 hours per day on inventory routines @ £12 total cost to employ, you are talking a £20m cost (5 hours assumes gap scanning & code checking). I assume a manual financial account is still required once a year to account for failed tags.

So we have to come up with other benefits, like top-line growth driven by a better customer experience, but customers will have to be willing to pay for this with higher prices.

Self-checkouts I'm sure now screw with the accuracy too..

At least with self-check out, every single items needs to be scanned, as there is no "times 10" button, and they get picky about sensing weight after each item. This can be quibbled of course (skip bagging, theft, etc).

The weight checking drives me up a wall. I use this primarily at hardware stores and I frequently can't fit large/heavy items on the output scale. Or the opposite problem, I'm buying a shim or a screw and it simply doesn't register.

For sure, after having discussed with someone working there I can tell you results are awful.

From someone inside Walmart:

""" The original idea was to have a few extra security cameras and some mirrors. I think it took 2 mirrors per aisle and only a few 4k color security cameras with infrared to cover the fast moving items.

After prototyping we find exactly what you said. Turns out it doesn't matter how well you know you need to stock items, if you don't give enough people-hours to do it, the number of items on the shelf doesn't change.

The robots were probably pitched by the Walmart dot com or Jet dot com guys. Thier projects always were greenlit without any analysis and rarely worked. """

https://old.reddit.com/r/technology/comments/jmx78e/walmart_...

Thanks for the insight, that seems plausible to me! A moving robot doesn't make too much sense in this application since the products and shelving is stationary. It just adds a ton of complexity.

What's interesting is the Amazon store concept which just like this post is heavily based on cameras watching the shelves.

The idea was for the shelf scanning robots to send messages when they found a low stock item or a mess, and then a human could just load up a cart with stock from the back room and head directly to the problem areas.

This would be in contrast to having people regularly walk every aisle and examine everything personally, going back and forth to get stock repeatedly. Much time was wasted.

As a former Walmart employee I was actually excited to see these proposed.

But now with so many people walking the aisles collecting items for pickup, I guess they’re just going to have them send “problem detected” messages instead of the robots. Which is fine I guess. Shopping there has become an even greater hassle with all the employees and their pickup carts, but that’s a whole other issue.

All items are scanned on the way out anyway - you should be able to get an idea when an item is low on stock from that alone, no? Sure there will still be things like people in the store that haven't checked out yet, but...

What am I missing?

Maybe that data is not sufficiently relayed to people on the floor? It also misses theft, although that can eventually become an assumed rate.

That's what I mean, though. They have decades of data at this point, you can likely guess what the theft rate is.

How much do these margins really matter? Probably not enough to have a robot that's very pricy going up and down isles.

When an item comes in from a truck it either goes directly onto the sales floor, into overstock (a high shelf above the other items), or into the backroom. People put things back in random places, stuff get destroyed (think of dropping a can on the ground), stuff get stollen, cashiers make a mistake, people make a mistake in self checkout, probably other stuff I am missing. It is a store, things are chaotic.

Let's take the journey of some cans of Goya black beans. 12 come in on a truck. The shelf capacity for them is 4. The person unloading the truck takes beans to the sales-floor, puts 4 on the shelf, puts 4 into overstock puts 4 into the backroom. Joe walks in takes a can, drops it and it bends. Jane comes along takes a can, continues shopping, the changes her mind and leaves it somewhere in the cereal aisle. Janet comes along and takes a can, takes it to self checkout but forgets to scan it. Jacob comes along takes a can, takes it to the checkout, and 1 can is subtracted from inventory. Inventory shows 11 cans, but there are actually 0 cans out on the sales floor. And I haven't even gotten into stuff being misplaced in the backroom ect.

I work (transitively) for a very large grocery retailer that rhymes with "Broger". Live inventory data is mostly junk -- it's sorta reliable in broad strokes, but you can't rely on it for determining out of stock events. There'd be all sorts of false positives and false negatives.

As someone who has had to clean and analyze inventory data, I can't agree with you enough.

In every company I've worked or consulted for, inventory reconciliation is always an ongoing headache for every business with a retail component. It is unbelievable how much waste and loss comes about from this.

Are their any companies achieving some level of automation? I would be very interested in learning more.

Like 20 years ago I worked at a grocery store when they were introducing "live inventory". Goods got scanned when unloaded from truck, and of course the register.

One day a customer complained we had no more washing powder of the most popular brand. Boss came, said they just got some the previous day, checked the inventory system and saw there should be three "trays" left, each with 12x packs of 1.2kg. He went to check the back, nothing there.

After a bit more searching he had a hunch, and checked the cameras. Sure enough, someone had loaded the three trays into a duffel bag earlier that morning and walked right out.

So yeah, even with all the normal human elements in the logistical chain, kinda hard to keep track of blatant theft.

Theft, misplaced items, etc. Employee grabs it to use for the store, damaged items, etc etc.

With only using in-out the inventory system thinks there are plenty to sell and won't reorder, but if it's wrong the shelf is empty and can't be reordered. There are posts from Walmart people online complaining about this exact problem.

Affects me as a customer, hit their website says they have 10 in stock for pickup today, drive over and the shelf is bare. Ask, they say the inventory is probably wrong, check another store because they can't reorder.

Now they are missing x items + y lost sales. That's why place do inventory.

> That's why place do inventory.

Obviously, we know why places do inventory.

We are talking about paying an expensive robot to go up and down isles to track inventory and if that is worth it.

This doesn't make sense: WalMart doesn't really have a backroom. It comes off the truck and goes onto the shelves. The backroom is too small (compared to the size of the store) to have spares of everything on pallets that are moved to the shelves as needed. This is by design - spares in the backroom are excess inventory which needs to be paid for. I'm sure there exceptions for a few things that sell in large numbers and take up a lot of space (bulk packs of toilet paper for example, keep them on a high shelf and move a few down as needed), but for the most part WalMart is against inventory.

WalMart already knows how much of something they have because they get everything scanned at the register. Sure there is a hour delay between it leaving the shelf and it being subtracted from inventory, but they need more than an hours worth on inventory on the shelf anyway because it takes time to load a truck with whatever they need more of. Shoplifting is a problem that mucks with this (I could actually see this being a useful help to that problem, but I'm not sure if it is useful enough)

Doing inventory isn't just how counting how much of item A exists in a store for accounting purposes. More importantly it is making sure that the there enough items on a shelf for customers to purchase. This is known as "on shelf inventory".

An item in the backroom does not count towards the on shelf count. An item that is on a high shelf to be moved down as needed (overstock) does not count. And there isn't just a bit of overstock, most aisle the top shelf will be overstock. Also Items can get damaged, meat can expire ect. A customer can pick up an item decide he doesn't want to buy it, and throw it back on a random shelf.

The job involves going to an aisle, scanning the items on the shelf, and if the shelf count is not full either moving items down from overstock or from the back room.

But once again you get that data from the register sales. there is a delay, but for the most part there are enough items on the shelf to last through the delay, or you would have sent someone to move items to the shelf anyway. Only when someone buys all of something is there are problem.

I'm extremely curious about the loss rate at the Amazon stores. They're relatively new, but I'm going to be a bit surprised if they don't wind up having half a store walk off on them at some point. I think we've all seen enough examples of what happens when you release algorithmic systems into the wild.

Thank You.

In terms of Warehouse Automation, does anyone knows how well does Walmart compare to Amazon?

Does Walmart need the same automation as Amazon? Does Walmart even need pickers or are they just filling up trucks with pallets and merchandise already packaged?

Brick and mortar warehousing seems like it would be much different than online retail warehousing.

They need something. They get a truck of toilet paper from the paper mill, and a truck of coffee from the roaster to the warehouse. What goes out of the warehouse is a 2 trucks to two different stores, each with half coffee, half toilet paper.

I'm grossly simplifying the above, but the idea should be clear. WalMart doesn't store things in their warehouses so much as take them off of trucks and put a different mix of thing on other trucks. There is lots of room for automation, but it will look very different from Amazon.

There’s still a lot of individual item picking due to (as TFA mentions)

> ...precise inventory is also essential for Walmart’s increasingly popular pickup and delivery services.

...and I can see highly accurate inventory helping that (and as a shopper I’d like it there was warehouse style ‘pick to light’ systems in supermarkets (perhaps using Augmented Reality))

warehouse style ‘pick to light’ systems

Seriously, watching one person pick one order from retail shelves is kind of aggravstingly inefficient, when (and this is probably already starting to happen) they could have everything picked in bulk with an optimized path through the shelves for a batch of orders, then put-to-light each order.

I always get the wrong items from Walmart shipping. I get items I didn’t order. I don’t bother filing a complaint, because I have to wait on their phone line to talk with customer service. Bleh.

Same goes for Target. I sometimes get incorrect products from them.

Amazon has been much better in product accuracy. But I get leery at buying certain items from Amazon, especially if it has a high probability for defects or counterfeits.

the robot as seen in the video makes zero sense, a railing camera on the ceiling or top of the opposite shelf would do the same in a fraction of the time because it doesn't have to be slow to avoid injuring people and will have greater accuracy due the fixed nature of the shelfs spatial relationships, moreover it doesn't have to come with complex navigation software and hardware.

research is fine and all, but this is quite far from something that could have resulted in a production run.

The problem with ceiling camera on rails would be that it won't see stuff in the lower shelves. It would be best to have a camera at human level, or lower.

But going this way, why not put a camera on each shelf, looking at the shelf opposite of it? I.e. if you have shelf A and shelf B opposite each other, the camera on A looks at B, and camera on B looks at A. The goal would be to have each shelf seen by at least one camera.

It now becomes a problem of integrating data, but I can't imagine it being more complex than crowd-navigating robots, or even moving cameras on rails.

> or top of the opposite shelf

angle would be about same from top or ceiling

> on each shelf, looking at the shelf opposite of it

it would cost a lot more than a camera on rail, especially the wiring but also bandwidth

I've run into this issue a number of times, in terms of processes and straight up code.

Management teams often love the idea of things that prompt humans to do a thing, under the guise of saving human work hours due to magical 'efficiencies'. But somehow it is often lost on them that they're actually simply prompting ... more (or even just the same amount) human work hours.

Later the idea is often eventually scrapped when they finally associate it with human work hours.

It's this strange constant cycle.

During my gap year, I ran the backroom at a local Wal-Mart. While I was there, we switched to a new fulfillment system that involved putting all of the stocking responsibilities on one team, instead of spreading it between departments.

The problem isn't the technology, it's the physical amount of product coming in compared to man hours. A typical 8hr shift would look like:

- 3hrs to unload the general merch delivery truck. An hour to unload the two freezer/cooler that would arrive at some point throughout the shift

- 1hr30m of breaks

- 1hr setup for night shift

- 1hr of 'picks', which was stock we had in the backroom that the system said could go out. This doesn't include the actual stocking of said items

That leaves 1hr30m to unload all the stock from the trucks, which doesn't happen.

Even if these robots were in my store and could tell me for certain that products needed to be reshelved, the man power isn't there to do it. It ends up being that all the teams achieve their department goals except for the fulfillment team that gets killed.

edit: formatting

Here is an armchair strategist opinion :).

This seems to be an approach of "let's replicate what humans do with robots", which often fails due to complexity (humans are very versatile). For example, the separate problems of finding empty shelves and restocking those are much more complex if you try to solve them together. Do it with a ground-moving robot, competing for floor space with humans and it may be hopeless.

But if you aim for just finding empty shelves you can probably do so with a few high-res ceiling mounted cameras and algorithms. This also allows for incremental enhancements. If the system initially detects missed inventory at 80% of the shelves it is a good start (there are now fewer shelves for humans to check).

Incremental improvement usually beats trying to solve complex problems in one go. My 2c.

They already have this information from the cash registers though. it isn't up to the second like the system you proposed, but it is "good enough".

Items can run out of stock before hitting the registers- people spent a long time in big department stores like Wal-Mart.

That doesn't matter, it takes more than an hour to get the replacements on the truck and the truck to walmart anyway. WalMart doesn't store much extra inventory in the back room. Everything comes off the truck and goes right onto the shelf.

People steal shit and put shit back in the wrong spot all the time.

I agree, but those are different problems. they are related, but different.

Incorrect reshelving is a different problem. Your internal record of what's in stock -- from the cash registers you mention -- will be correct, you just won't be able to find the items.

Shoplifting is not a different problem; it's the same problem. Shoplifting means the record derived from register transactions is incorrect, because registered sales are not the only type of transaction you participate in.

Actual strategist here (I automate things humans do for a living).

I agree! Building a machine to do what humans are good at is dumb. Building a machine to do what humans are not good at is smart. Humans are great at noticing things as they walk around. Machines are OK, but miss the context way too frequently. If a person can do it alone, it'll be really hard to make a robot better at it, at least in the real world.

We should build more machines that are good at what machines are good at, and let humans fill in the blanks around those machines. For example, a robot that unloads heavy stuff from trucks would save hours of time so that human can move stuff to the shelves (and answer questions from shoppers while they're out there). Lo and behold, the profitable robotics companies here and elsewhere do that: move heavy stuff around (and only that, and only in warehouses where things are controlled).

Now, arguably their case is that they want something "acceptably worse than a human for much less cost", but they erred by missing all the different things a human has to do while on shift, and building up a whole machine for this, the one thing that humans do without thinking.

More "The world's largest supermarket chain said it had ended its partnership with Bossa Nova Robotics, who made the roving robots."

Vs going away from the idea entirely. WalMart labs is heavily interested on automation to reduce human labor needed.

>>WalMart labs is heavily interested on automation to reduce human labor needed.

Well... There really is no question of interest. They'd obviously like to automate more and reduce labour. The interesting questions relate to the trials, tribulations, successes and failures.

It would be a more informative article with some indication of how much they're investing in such initiatives. Why did this initiative fail, etc.

Of course they want to reduce labor. They are the largest (non-govt.) employer in the US and largest employer in most states. They have over 1.5 million employees and over 2 million with contractors.

That's almost 1% of US population employed by a single employer !

> The federal government employs nearly 9.1 million workers, comprising nearly 6 percent of total employment in the United States.

Both are shockingly large.

For a random comparison: The entire German government (federal+state+local) employs 3.7 million people, about 8.8% of total employment. Another 0.77 million (1.8% of total employment) are indirectly employed in public services. The German federal government alone are just 0.47 million, or 1.1% of total employment.

...different kinds of shocking.

The federal government is a sector, comparable to "mining" or whatnot. Big is still big in the one-prick-many-ouchies sense, but sector size is not the same thing.

The "resolution" of private markets is a major characteristic.

That's a reasonable point, but also no other sector has one person (DJT) as CEO.

Eh, the federal government doesn't fall entirely under the President's power. The position doesn't even have authority over the entire executive branch.

Also, as far as the "government sector" of the economy goes -- 87% of government employees work for state or local governments.

I don't see why 1% of the workforce being employed by the same employer is supposed to be shocking?

Suppose you have a village of 400 people. What are the odds that somebody there employs 4 other guys in some capacity? That's more than 1% of the workforce.

...it's not one percent of a village. Many things won't make sense to you if you go around being deliberately obtuse like this.

Why would we expect the employment structure of an empire to be more fragmented than the employment structure of a village? The whole idea of an empire is large-scale cooperation.

Acting as though scale is irrelevant is being deliberately obtuse, and doubling down on that is inane.

I'm not saying scale is irrelevant. I'm saying the influence of scale should push in the opposite direction to what you are implicitly arguing for.

I notice you've never bothered to make any arguments other than "stop being stupid". Could you?

Suppose you have a village of 4 people. What are the odds that one guy employs one other guy in some capacity? That's 50% of the workforce.

Thus 50% is not a shockingly high number. Right?

The odds there are zero; a community of four people is not going to bother with any commercial relationships at all. It is almost certainly a very small family.

Walmart labs is just the group of people running the .com in Sunnyvale. They don't have any actual labs or innovation.

Their ideas of "Robots!" and sending hourly employees to your home with merchandise instead of paying actual delivery drivers were two ideas that came from them recently. Everything they touch is broken.

As someone who worked in a supermarket... this robot is something that definitely can help the staff, especially in large locations. Customers always manage to spill stuff, throw back items from the cooling section randomly in some rack... it's not a bad idea to leave the monotonous patrol duty to a robot and only act when the robot detects something.

The obviously best solution to keep track of inventory would be "smart racks" with RFID tags on each product (side note: that could also be used to deal with tracking stolen products, return policy abusers, people who want to legit return something but lost the receipt, counterfeit issues), but building such a rack is likely enormously expensive...

> The obviously best solution to keep track of inventory would be "smart racks" with RFID tags on each product

As someone who's deployed several industrial RFID systems: I doubt we'll ever see an RFID tag on every item in Walmart.

The first reason is the cost of tags: Even the cheapest EPC Gen2 RFID tags cost more than 1 cent - and nobody selling a 20 cent can of beans is going to pay that. Barcodes, on the other hand, come free with the product's printed packaging.

And the tag's performance varies - the super-low-price tags might work fine for cardboard packages, but not work on metal cans, for example, as sticking a radio antenna to a bit of metal and having it work is a complex business.

The second reason is reliability: Even in a highly controlled environment, where every tag is on an identical item and you're willing to spend 20 cents per tag, RFID is hard to get right.

One thing that causes reliability problems is RFID is tough to test. Reflections and interference patterns mean you'll get areas where reading will fail, even when inches away reading works very well - and if you move metal in the environment around, it all changes and the nulls move.

And that's before you get into having a thousand different products from a hundred different suppliers, all of whom now have to learn this new RFID technology.

Another thing that leads to reliability problems is simple mechanical matters: It's tough for an RFID tag to survive being in between two metal cans that bump together.

And if you're hoping to do something like detecting thefts, a system that is merely 99.99% reliable is going to generate far more false alarms than it does real alarms.

WalMart threw millions into RFID a few years ago. They abandoned it for the problems you stated. What they wanted was the checkout process to be fast enough that your cart doesn't stop as you are leaving the store (unless you stop to bag your items after you paid for them).

> What they wanted was the checkout process to be fast enough that your cart doesn't stop as you are leaving the store (unless you stop to bag your items after you paid for them).

That could be solved by giving the customers handheld barcode scanners.

Which actually is something that exists. Its an option at my local Kroger, although I rarely see anyone using it.

In Sam's club (walmart owned), they have customers scan products via their phone app

As someone else said, this could probably be done more cheaply and reliably with stationary cameras.

No, it cannot - the stationary camera will always be somewhere at the ceiling, which means that it physically can not see the lower levels of a rack.

I heard someone speak of the Bosso Nova at Walmart at a presentation and most of it was off record. They were reluctant to be open about the Walmart robots.

From what I heard/saw I figured there was something more because what it seemed to be achieving was very low hanging fruit.

Walmart is notorious for cutting costs and keeping prices low for shoppers and a razor thin profit margin. Their real strength is their supply chain and I have tremendous respect for what Walmart does...and I don’t think anyone matches Walmart when it comes to supply chain management. I heard even amazon’s best are from Walmart.

Having said that..I could see no way Bossa Nova bots were compatible with Walmart principles. IMO, they should all pivot and work on autonomous Ag bots. Unlike warehouses, there are multiple environments and multiple problems to solve in Ag. And it’s under explored and under developed.

I remember thinking that this(walmart bots) is similar to Ag robotics. Robotics and automation companies coming up with a solution first and then hope that they can define the problem later. They are creating solutions for problems that don’t exist in the first place.

I see an emerging consensus here on HN that automating retail is a hard thing to do. Somehow though, Amazon Go just works. No robots, just cameras and machine learning. It just seems to me Amazon is able to do this and Walmart isn't. For now. My bet is that 15 years from now it's going to be as common as an iPhone now (remember how common was an iPhone in 2005? not very...)

To be fair Amazon Go also doesn't use robots for any sort of inventory management. They literally have humans go around the store as needed, based on information collected by the cameras & tracking systems.

What if they made Walmart be a giant vending machine, automatically fed with conveyor belts from a warehouse in the back?

This might not be true, but I think Walmart makes money by selling shelf space visibility to brands--e.g. General Mills pays Walmart money to have products on an eye-level place on the shelf.

Or, Walmart can pivot to entirely online delivery shopping and save loads of money by not having stores.

Or, Amazon can slide into the groceries space.

I’m surprised that these were able to last as long as they did without human interference and damage. I assumed generally that these robots would not last due to humans messing with them.

How come back doesnt work in browser on BBC? Thats like some scam / ad site stuff.

Works fine for me

It's not exactly that the back button doesn't work. It's because for some reason they seem to be redirecting in some weird non-HTTP way from their `.co.uk` domain to their `.com`, even though you already got a complete `200` response from `.co.uk`.

As a result, clicking the back button only once from the `.com` takes you to the `.co.uk`, which redirects you back to the `.com`. Clicking twice, or if you follow the `.com` link in the first place, does not affect the back button in any way.

I still think we are a decade away from this technology being implemented.

I imagine with OCR advances (and similar) it would make more True Positives, eliminating the need for a worker.

If a worker still needs to check an isle it almost defeats the purpose.

Reply on news.ycombinator.com