If they continue to raise and add Amazon competitors as customers, they'll be in a similar position as Instacart. Selling to anyone whose in competition w/ Amazon
I see no reason Amazon would want to acquire them: Amazon is a few years ahead in their technology and already has the logistics back-end and branding.
Hi, I'm the lead investor in this round for Standard Cognition. From what we can tell, Amazon has actually over-invested in sensor technology that results in expensive reconfiguration of shelves and stores. It works for Amazon since they are building stores from scratch, but existing retail partners don't want to put that kind of capital into reconfiguration.
This is where a camera-only approach really shines, which is why the team has gotten so much adoption from retailers. This tech is literally the way incumbents can fight back against Amazon's encroachment.
We're delivering autonomous checkout to Amazon's competitors across the world. Amazon can only crush us if they crush the entire retail sector, which will be a tall order, especially with the slew of startups emerging to help defend retailers from Amazon's entry into the brick and mortar world.
I see Instacart people in the store all the time. However, I wouldn't use them. I like to pick my own vegetables, and meat. It's definitely not all the same (meat, vegetables, fruits), and having someone just throw whatever they can grab into a cart in the fastest way possible isn't something I'm interested in.
Is that what it looks like in the store? That's not how it feels at home. Shoppers frequently label items as 'of low quality' to alert me that the bell pepper selection or whatever isn't great.
Also, despite my 30+ years on this Earth, I vastly more trust a 'professional' shopper to pick out good veggies than I do myself. I respect that you're different, but I can say with 100% certainty that the produce picked out by my instacart shoppers is excellent and better than I could do. Never a single complaint there, at least from me.
Do they really rush? From my side, they seem to take their sweet time. They text frequently, asking questions, suggesting replacements not shown in the app, etc. The whole process feels slow and relaxed.
I don't eat meat so I don't know about that. But I would assume the same as veggies. Instacart does a top-top job picking quality ingredients for me. Definitely better than I would know how to do myself.
Who is more likely to know the freshest spinach? Not me, that's for sure. An instacart shopper who is there all day? Much more likely to get good results.
I do have issues with instacart, but none of the concerns you've mentioned have in any way harmed me. My main issue is that the delivery times seem to be not communicated to the driver. So it will tell you 10am - 11am, but the driver will show up at 9:50 and not know that they are early. Sometimes also, the selection available on the website will be mislabelled in category (tomatoes in bath care, or something).
Otherwise it has been really incredible for me, overall.
Instacart has always been excellent in my experience, they have alerted me of better deals, recommendations for different equivalent products. I've managed to get power drill delivered to my doorstep in about an hour. And for meat/produce, the shoppers seem to have a good eye for quality. Amazon go falls flat in comparison.
Hi, lead investor in this round here. Instacart is actually also an Initialized portfolio company and I funded Apoorva while I was a partner at Y Combinator. We think these technologies are both very complementary — either way, retailers need options to fight against Amazon. They can't hire world-class engineers but Instacart and Standard Cognition sure can.
I did something slightly similar to this for class project years ago. I think my project was called "Object Removal Detection in a Retail Environment".
I wasn't trying to detect which item was removed, just that an item was removed, so it was a good bit simpler. I was using Matlab to analyze with old school computer vision algorithms to analyze the video (this was before deep learning was common).
It mostly worked--it could highlight when an item was removed from a shelf, but it wasn't useful enough for much other than maybe helping security to manually review tapes for evidence of shoplifting.
This is very surprising given the limitations of the technology. How they can handle occlusions from the ceiling cameras? How the can know for sure if a person is grabbing a product from the back of the shelf rather than a product that's in front? Unless they're using some high-frequency cameras to "X-Ray" the whole store, I'd say this approach is doomed to failure. Even Amazon Go, after many years of R&D, uses a combination of sensors to ensure that they charge correctly and that they have an accurate inventory of the whole store.
Why is it so surprising? Presumably they can setup as many cameras as needed. I don’t see it being so much of an improvement over the current self checkout systems to warrant the billions of investment being thrown at it currently though.
These are actually very valid points and why most solutions tend to go for a sensor fusion approach that leverages the best of computer vision along with other sensor modalities. Using only ceiling cameras will not get you to the kind of accuracy that most retailers require to confidently rely on the technology.
Even if you placed more cameras (not only ceiling cameras, but also in the shelves and other areas of the store pointing at different angles), leaving aside the cost (which adds up), people will always be able to occlude the items from the cameras (even unintentionally). To put it in numbers (as an example): computer vision can get you 80% of the way in terms of accuracy / detecting items grabbed, for the rest you need other sensors.
The main benefit over other self checkout systems is the time customers wait in line. If you have been to Amazon Go, the experience of walking out without waiting is quite magical.
Occlusion is a big vision problem here. The customer facing problem is being stuck at checkout with a "Wait for Attendant" message on screen.
There's "LaneHawk", which has a camera mounted near the floor looking at the bottom level of shopping carts.[1] This is to catch bags of dog food, cases of beer, and similar big items the cashier might miss. Came out around 2010. Saves about $10 per lane per day. That's probably the most successful system in this area right now.
We were initially unsure that we could achieve high accuracy with overhead cameras only. It's definitely not immediately obvious that it should be feasible, and we weren't convinced until we arrived at a working solution (and a working store!). Getting our action recognition models to high enough accuracy was a big part of the crux, which, not coincidentally, is an area where the literature isn't super mature yet.
Our system still makes mistakes, of course, but we're in the high 90's of accuracy, which is good enough economically to start deploying and getting real world users. And there's plenty of signal to squeeze out still.
Jordan from Standard here. The overhead camera approach is definitely more challenging. We've been able to achieve high accuracy nonetheless. The proofs in the pudding though, of course, which is why we opened a store in SF last month! On the flip side, by using overhead cameras only and avoiding lining the shelves with expensive sensors, we're able to deliver a cheaper and more flexible system. Ultimately that means rolling out to more stores, faster.
The “re-humanizing retail” line left a bad taste in my mouth. When I think of lots of cameras in a store designed to get you in and out as quickly as possible without having to talk to a cashier or a stocker, it feels de-humanizing and sterile. It is also one of the ways I interact with people at different income levels. That’s why people go to farmers markets and do CSA boxes - they want to feel connection to their food providers and community.
I don’t doubt this technology is the future or will be a big market, but let’s not kid ourselves that this will generate more human connection.
This is a fair initial take, and we should clarify why we think we're improving the social experience. A big part of why Lyft or Uber feels different than taking a cab is that there's no transactional portion of your interaction with the driver. Get into the car, chat a bit, say goodbye, and be on your way. By removing transactional mechanics you can focus on the human element. If you look at the Amazon Go stores, or the Standard Store, you see a similar effect. You walk in, and rather than immediately seeing bulky machines manned by people with the sole intent of transacting, instead you see people walking around chatting and helping you find what you need. They're not there to take your money, they're there to help you, exclusively. That changes the nature of the interaction, and is the experience we're trying to deliver across retail.
Let's not forget just how terrible the taxi industry was before Uber came along. The bar was really low. A large part of their success is attributed to their initial business strategy and subsequent legal and regulatory achievements.
I mean, that is a good pitch in theory, that just doesn't resonate with my experience as a consumer in retail.
Go to any store that has self-checkout currently. Usually there is one attendee servicing 4 machines. No one comes up to you to chat, they only come over when theres a problem. When you go to a manned-checkout, they usually ask how you are doing, if you found everything ok, if you want to join a loyalty program...etc. Some of the checkout people I know very well over the years in my community. One even gave my infant daughter free socks because she has grandchildren.
You wont get that in this future. Those checkers will be gone. You'll just have employees minding their own business tending to their work, like you will with yours.
I think the idea is that instead of being tied to a cash register, former checkout people would be able to get to know their community better by being free to go anywhere in the store. The Apple Store experience comes to mind for me.
If the software screws up and the item count is incorrect in the favor of the customer (ie, charged for 3 rather than 4+ widgets), who eats the cost?
Have you hired sleight of hand artists to field test the machines? I’d be very impressed if a determined actor could not fool a set of overhead cameras.
To the few comments above: stores with automatic self-checkouts were a thing in Japan for a decade, but the trend has reversed. Properly trained cashiers were faster than machines, nor had issues with "please wait for attendant every second checkout."
The biggest problem for them were that they after all, all required some amount of human intervention for work. Because of that, the store had to buy both a self-checkout machines, then hire a guy to click buttons on the machine, then a security guy who looks if nobody tries to do any tricks and rough up ones who do.
I think the better way is how it is done in China: you have gigantic (shipping container sized) wending machines placed all around residential blocks and such. Works well for packaged products, but not for fresh produce. When it needs service, a truck comes in, picks it up whole, and a new, fully restocked one is placed in its place in a few hours by the same truck with a crane.
31 comments
[ 2.8 ms ] story [ 86.4 ms ] threadThis is where a camera-only approach really shines, which is why the team has gotten so much adoption from retailers. This tech is literally the way incumbents can fight back against Amazon's encroachment.
We're delivering autonomous checkout to Amazon's competitors across the world. Amazon can only crush us if they crush the entire retail sector, which will be a tall order, especially with the slew of startups emerging to help defend retailers from Amazon's entry into the brick and mortar world.
Also, despite my 30+ years on this Earth, I vastly more trust a 'professional' shopper to pick out good veggies than I do myself. I respect that you're different, but I can say with 100% certainty that the produce picked out by my instacart shoppers is excellent and better than I could do. Never a single complaint there, at least from me.
Do they really rush? From my side, they seem to take their sweet time. They text frequently, asking questions, suggesting replacements not shown in the app, etc. The whole process feels slow and relaxed.
I don't eat meat so I don't know about that. But I would assume the same as veggies. Instacart does a top-top job picking quality ingredients for me. Definitely better than I would know how to do myself.
Who is more likely to know the freshest spinach? Not me, that's for sure. An instacart shopper who is there all day? Much more likely to get good results.
I do have issues with instacart, but none of the concerns you've mentioned have in any way harmed me. My main issue is that the delivery times seem to be not communicated to the driver. So it will tell you 10am - 11am, but the driver will show up at 9:50 and not know that they are early. Sometimes also, the selection available on the website will be mislabelled in category (tomatoes in bath care, or something).
Otherwise it has been really incredible for me, overall.
Seems those things would be based on the same skill.
I wasn't trying to detect which item was removed, just that an item was removed, so it was a good bit simpler. I was using Matlab to analyze with old school computer vision algorithms to analyze the video (this was before deep learning was common).
It mostly worked--it could highlight when an item was removed from a shelf, but it wasn't useful enough for much other than maybe helping security to manually review tapes for evidence of shoplifting.
These are actually very valid points and why most solutions tend to go for a sensor fusion approach that leverages the best of computer vision along with other sensor modalities. Using only ceiling cameras will not get you to the kind of accuracy that most retailers require to confidently rely on the technology.
Even if you placed more cameras (not only ceiling cameras, but also in the shelves and other areas of the store pointing at different angles), leaving aside the cost (which adds up), people will always be able to occlude the items from the cameras (even unintentionally). To put it in numbers (as an example): computer vision can get you 80% of the way in terms of accuracy / detecting items grabbed, for the rest you need other sensors.
The main benefit over other self checkout systems is the time customers wait in line. If you have been to Amazon Go, the experience of walking out without waiting is quite magical.
There's "LaneHawk", which has a camera mounted near the floor looking at the bottom level of shopping carts.[1] This is to catch bags of dog food, cases of beer, and similar big items the cashier might miss. Came out around 2010. Saves about $10 per lane per day. That's probably the most successful system in this area right now.
[1] https://www.youtube.com/watch?v=rpHqWSYTF2s
Our system still makes mistakes, of course, but we're in the high 90's of accuracy, which is good enough economically to start deploying and getting real world users. And there's plenty of signal to squeeze out still.
I don’t doubt this technology is the future or will be a big market, but let’s not kid ourselves that this will generate more human connection.
This is a fair initial take, and we should clarify why we think we're improving the social experience. A big part of why Lyft or Uber feels different than taking a cab is that there's no transactional portion of your interaction with the driver. Get into the car, chat a bit, say goodbye, and be on your way. By removing transactional mechanics you can focus on the human element. If you look at the Amazon Go stores, or the Standard Store, you see a similar effect. You walk in, and rather than immediately seeing bulky machines manned by people with the sole intent of transacting, instead you see people walking around chatting and helping you find what you need. They're not there to take your money, they're there to help you, exclusively. That changes the nature of the interaction, and is the experience we're trying to deliver across retail.
Go to any store that has self-checkout currently. Usually there is one attendee servicing 4 machines. No one comes up to you to chat, they only come over when theres a problem. When you go to a manned-checkout, they usually ask how you are doing, if you found everything ok, if you want to join a loyalty program...etc. Some of the checkout people I know very well over the years in my community. One even gave my infant daughter free socks because she has grandchildren.
You wont get that in this future. Those checkers will be gone. You'll just have employees minding their own business tending to their work, like you will with yours.
Have you hired sleight of hand artists to field test the machines? I’d be very impressed if a determined actor could not fool a set of overhead cameras.
The biggest problem for them were that they after all, all required some amount of human intervention for work. Because of that, the store had to buy both a self-checkout machines, then hire a guy to click buttons on the machine, then a security guy who looks if nobody tries to do any tricks and rough up ones who do.
I think the better way is how it is done in China: you have gigantic (shipping container sized) wending machines placed all around residential blocks and such. Works well for packaged products, but not for fresh produce. When it needs service, a truck comes in, picks it up whole, and a new, fully restocked one is placed in its place in a few hours by the same truck with a crane.