Interresting that energy wise it is feasible to transport a whole lot of items that are not being used to the picker and back. But I am pretty sure that this has been taken into account when designing the size of the storage pods.
Pretty impressive!
Yeah I thought this too. I guess treating the warehouse as a 2D grid vastly simplifies everything from algorithms to robotics so much that the reliability out ways any inefficiencies.
Presumably they could make the stacks shorter and have more "floors" to increase efficiency of each trip.
I'm interested in what the wear is on the robots. How often do the parts need to be replaced or repaired? How long can you run without any replacements or other maintenance? What does it cost, in money and resources, to replace the fleet?
It's more obvious when you consider that the alternative is people walking around the FC filling carts with items.
I imagine that the cost of energy is minimal compared to the cost of labour in this case. But even if it wasn't, you still have 60-100kg people walking all over the FC so the difference in energy usage is minimal... but Amazon doesn't (directly) pay for the calories that their employees burn walking around the FC.
However one could imagine a Kiva v2 that figured out some way to not move whole racks at a time. The only reason they move the whole rack is to avoid having to build a robot that can pick individual items off of the shelf individually.
People walking around might be happier than people who have to stand in one place all day, though. It would be interesting to invert the system and build a navigation system for the workers to guide them down the aisles in the most optimal way to pick up the items - it should take the same amount of time, but leave you with workers in better physical condition.
That's what Amazon had before Kiva robots. Here's the job description:
"We go around with a little cart to a conveyor belt and grab 2 plastic tubs. With a scanner in one hand and pushing the cart in the other your job is to go around to wherever the scanner tells you and "pick the items" This involves scanning the tub, then the location bar code, searching for the right item (which most of the time is easy to find but sometimes hard) and scanning it. Then doing the same thing for the next 10hours of your day minus the 2 15minute breaks and the 30min lunch break."
The people are just peripheral devices of the computer systems that run the distribution center.
Ok, so I'm thinking the need for this could become huge. How does one get into this industry as a programmer? What would I need to learn? What kind of diplomas are useful? What should I start practising on?
As someone who's been trying to hire for a robotics programmer for a long time:
Need to have:
Systems/RTOS experience. ie. not spend a week chasing your tail when you stupidly put a printf in an ISR which changes timing so the code works, but then when you pull out the printf it stops working.
Be able to read schematics and datasheets. I'm not expecting you to design a board (we have EEs for that), but you should absolutely be able to be given a schematic and a datasheet and then write a driver for a simple peripheral.
Basic understanding of control theory. Being able to articulate what a PID loop is and why open control loops are a bad idea is probably good enough.
Being a self directed learner. Robotics tends to be fairly low volumes all things considered (or atleast my branch of it). We have issues with vendor support not really caring about what happens to us since our volumes are so low. We therefore have to own significantly more of our tooling and have to be able to peer in to what our vendors would prefer to be treated as black boxes quite a bit. Any RE experience would go a really long way here in my book.
Nice to have:
Good understanding of mechanical engineering/physics terms like stiction. Ostensibly if you have the skillsets written above this you can be taught this, but this lowers the bar for hiring you quite a bit.
Would you mind shooting me an email? I know some folks that are looking for work, and at the very least I'd be interested to hear more about what your business looks like.
It's still a small industry. Kiva is a few hundred people. iRobot is 528 people. Boston Dynamics was about a hundred before the Google acquisition. Some of the industrial automation employees are bigger, but many of their employees are at field sites, installing and fixing systems.
This one shows the picking station, where the humans take things out of bins and put them in other bins. A computer-controlled laser pointer points to the item to be taken out, and a light shows where the item goes, and a bar code scanner checks on the human. The job takes about two minutes to learn, and full productivity for new humans is achieved in about half an hour. There is no possibility of promotion. Machines should think. People should work.
Kiva is a huge success. Before Amazon bought them, they had about 20% of online order fulfillment in the US. It's higher now, with Amazon using them. Kiva is so successful because it's so simple to install. All it needs is a big flat floor with some bar code stickers, a supply of cheap shelving units, and the robots. All the robots are small and interchangeable, so they don't have to be repaired on-site and there's no need for expensive on-site technicians and repair shops. So converting to Kiva robots is fast, cheap, and easy.
Automated warehousing isn't new, but it used to be a lot more complicated and far more custom. The older systems involved conveyors, machines that moved on tracks, extensive site-specific and product-specific engineering, and good onsite maintenance. Here's a state of the art version of a classic automated storage and retrieval system in a frozen foods warehouse. (The frozen foods industry has been heavily into warehouse automation for decades, because they work in a sub-freezing environment.) This is impressive, but look at the sheer complexity and number of moving parts involved. All those belts, motors, lifts, and sensors, and all dedicated - if any of that stuff breaks, it has to be fixed, not just bypassed. With Kiva, any dead robots can be pushed out of the way and dealt with later, off-site.
Compare the mechanical simplicity of the Kiva system.
Kiva was started by one of the executives of Webvan. (Remember Webvan - first dot-com boom?) Webvan offered same-day delivery a decade ago. It was popular. It just cost too much to provide that service. If they could only get rid of all those warehouse employees and complex warehouse machinery... Well, they did. Most of them.
But humans are still needed to take things out of one bin and put them into another. For that, there's the Amazon Bin-Picking Challenge:
why is it that Amazon cant just put a much higher bounty on that task?
of lets say 1m$? it seems that this would increase the competition by a huge amount and if that can get amazon 1 year closer to fully automated picker it should be wort it.
The payed around 700m$ for the kiva bots, so why one 30k$ for that competition?
Before Amazon bought them, they had about 20% of online order fulfillment in the US. It's higher now, with Amazon using them.
I don't think that's right. They only had about $100m in revenue when Amazon acquired them.
Also, it's not a given that their market share has increased since Amazon acquired them. Amazon stopped Kiva's marketing and dedicated all of Kiva's manufacturing to themselves for years. Even today, Kiva produces robots for Amazon.
All non-Amazon customers are simply being maintained. Kiva hasn't signed up a new customer since Amazon acquired them.
Here's a direct competitor to Kiva. This approach, with tiny robot cranes, is less successful. ... Another competitor. This one uses an overhead robotic crane.
That's a single company called Swisslog and they have quite a few products in this space:
http://www.swisslog.com/
They also have a product that's nearly identical to Kiva:
I don't think that's right. They only had about $100m in revenue when Amazon acquired them.
That's what's so striking about Kiva. They basically make one model of robot with minor variations. Kiva was about 500 employees when Amazon bought them. Yet that was enough to handle 20% of online fulfillment. Pets.com, Diapers.com, Abercrombie and Fitch, and many other major brands used Kiva robots.
Most of the other automation companies do vast amounts of custom engineering work. Every customer is different. For Kiva, every customer is the same, and the product line is simple. Kiva is the first company to achieve this in robotics. (Well, there's the Roomba, if you consider that a robot.)
21 comments
[ 2.8 ms ] story [ 54.5 ms ] threadPresumably they could make the stacks shorter and have more "floors" to increase efficiency of each trip.
I imagine that the cost of energy is minimal compared to the cost of labour in this case. But even if it wasn't, you still have 60-100kg people walking all over the FC so the difference in energy usage is minimal... but Amazon doesn't (directly) pay for the calories that their employees burn walking around the FC.
However one could imagine a Kiva v2 that figured out some way to not move whole racks at a time. The only reason they move the whole rack is to avoid having to build a robot that can pick individual items off of the shelf individually.
"We go around with a little cart to a conveyor belt and grab 2 plastic tubs. With a scanner in one hand and pushing the cart in the other your job is to go around to wherever the scanner tells you and "pick the items" This involves scanning the tub, then the location bar code, searching for the right item (which most of the time is easy to find but sometimes hard) and scanning it. Then doing the same thing for the next 10hours of your day minus the 2 15minute breaks and the 30min lunch break."
The people are just peripheral devices of the computer systems that run the distribution center.
Need to have:
Systems/RTOS experience. ie. not spend a week chasing your tail when you stupidly put a printf in an ISR which changes timing so the code works, but then when you pull out the printf it stops working.
Be able to read schematics and datasheets. I'm not expecting you to design a board (we have EEs for that), but you should absolutely be able to be given a schematic and a datasheet and then write a driver for a simple peripheral.
Basic understanding of control theory. Being able to articulate what a PID loop is and why open control loops are a bad idea is probably good enough.
Being a self directed learner. Robotics tends to be fairly low volumes all things considered (or atleast my branch of it). We have issues with vendor support not really caring about what happens to us since our volumes are so low. We therefore have to own significantly more of our tooling and have to be able to peer in to what our vendors would prefer to be treated as black boxes quite a bit. Any RE experience would go a really long way here in my book.
Nice to have:
Good understanding of mechanical engineering/physics terms like stiction. Ostensibly if you have the skillsets written above this you can be taught this, but this lowers the bar for hiring you quite a bit.
you've just eliminated perhaps 90% of the folks who've recently decided to switch to "hacker" as their career! ;-)
This one shows the picking station, where the humans take things out of bins and put them in other bins. A computer-controlled laser pointer points to the item to be taken out, and a light shows where the item goes, and a bar code scanner checks on the human. The job takes about two minutes to learn, and full productivity for new humans is achieved in about half an hour. There is no possibility of promotion. Machines should think. People should work.
Kiva is a huge success. Before Amazon bought them, they had about 20% of online order fulfillment in the US. It's higher now, with Amazon using them. Kiva is so successful because it's so simple to install. All it needs is a big flat floor with some bar code stickers, a supply of cheap shelving units, and the robots. All the robots are small and interchangeable, so they don't have to be repaired on-site and there's no need for expensive on-site technicians and repair shops. So converting to Kiva robots is fast, cheap, and easy.
Automated warehousing isn't new, but it used to be a lot more complicated and far more custom. The older systems involved conveyors, machines that moved on tracks, extensive site-specific and product-specific engineering, and good onsite maintenance. Here's a state of the art version of a classic automated storage and retrieval system in a frozen foods warehouse. (The frozen foods industry has been heavily into warehouse automation for decades, because they work in a sub-freezing environment.) This is impressive, but look at the sheer complexity and number of moving parts involved. All those belts, motors, lifts, and sensors, and all dedicated - if any of that stuff breaks, it has to be fixed, not just bypassed. With Kiva, any dead robots can be pushed out of the way and dealt with later, off-site.
https://www.youtube.com/watch?v=LuvrP1QKsdk
Here's a direct competitor to Kiva. This approach, with tiny robot cranes, is less successful. https://www.youtube.com/watch?v=0-G70CivfLM
Another competitor. This one uses an overhead robotic crane. https://www.youtube.com/watch?v=Peef_5W9nOQ
Compare the mechanical simplicity of the Kiva system.
Kiva was started by one of the executives of Webvan. (Remember Webvan - first dot-com boom?) Webvan offered same-day delivery a decade ago. It was popular. It just cost too much to provide that service. If they could only get rid of all those warehouse employees and complex warehouse machinery... Well, they did. Most of them.
But humans are still needed to take things out of one bin and put them into another. For that, there's the Amazon Bin-Picking Challenge:
http://amazonpickingchallenge.org/
Win up to $26,000 and eliminate 30,000 jobs at Amazon. Entries for the first round closed in October.
The payed around 700m$ for the kiva bots, so why one 30k$ for that competition?
Or dont they beliefe in markets?
At the 700m$ range, high end research now goes towards the driver-less delivery truck.
The remaining work to be done regarding the picker task is low-end software work: how to cut costs, reduce errors, increase robustness, scalability.
http://www.youtube.com/watch?v=Plo7SH9aBgg [IPI Robot Throwing Boxes-IEEE Spectrum]
https://www.youtube.com/watch?v=tlC2O9T9jks [Object Detection and Recogition WillowGaragevideo's channel]
https://www.youtube.com/watch?v=IdQ8lTOSS0U [Industrial Perception Inc 3D Pick and Place - Gary Bradski]
https://www.youtube.com/watch?v=-g2e2WFG7C8 [iHY Robot Hand Can Grip Anything - IEEE Spectrum]
I don't think that's right. They only had about $100m in revenue when Amazon acquired them.
Also, it's not a given that their market share has increased since Amazon acquired them. Amazon stopped Kiva's marketing and dedicated all of Kiva's manufacturing to themselves for years. Even today, Kiva produces robots for Amazon.
All non-Amazon customers are simply being maintained. Kiva hasn't signed up a new customer since Amazon acquired them.
Here's a direct competitor to Kiva. This approach, with tiny robot cranes, is less successful. ... Another competitor. This one uses an overhead robotic crane.
That's a single company called Swisslog and they have quite a few products in this space: http://www.swisslog.com/
They also have a product that's nearly identical to Kiva:
http://www.swisslog.com/Products/WDS/Storage-Systems/CarryPi...
Although Amazon is only using Kiva in 10 of its warehouses. So we're not really talking about all of amazon.
of course, I don't think that any Kiva's former customers were running all their warehouses using Kiva either...
IIRC the reason Kiva stopped selling outside of Amazon was because Amazon literally bought all of the units they were capable of making..
That's what's so striking about Kiva. They basically make one model of robot with minor variations. Kiva was about 500 employees when Amazon bought them. Yet that was enough to handle 20% of online fulfillment. Pets.com, Diapers.com, Abercrombie and Fitch, and many other major brands used Kiva robots.
Most of the other automation companies do vast amounts of custom engineering work. Every customer is different. For Kiva, every customer is the same, and the product line is simple. Kiva is the first company to achieve this in robotics. (Well, there's the Roomba, if you consider that a robot.)