Slashdot Mirror
Building Amazon a Better Warehouse Robot
Nerval's Lobster writes: Amazon relies quite a bit on human labor, most notably in its warehouses. The company wants to change that via machine learning and robotics, which is why earlier this year it invited 30 teams to a "Picking Contest." In order to win the contest, a team needed to build a robot that can outpace other robots in detecting and identifying an object on a shelf, gripping said object without breaking it, and delivering it into a waiting receptacle. Team RBO, composed of researchers from the Technical University of Berlin, won last month's competition by a healthy margin. Their winning design combined a WAM arm (complete with a suction cup for lifting objects) and an XR4000 mobile base into a single unit capable of picking up 12 objects in 20 minutes—not exactly blinding speed, but enough to demonstrate significant promise. If Amazon's contest demonstrated anything, it's that it could be quite a long time before robots are capable of identifying and sorting through objects at speeds even remotely approaching human (and thus taking over those jobs). Chances seem good that Amazon will ask future teams to build machines that are even smarter and faster.
Amazon's contest demonstrated anything, it's that it could be quite a long time before robots are capable of identifying and sorting through objects at speeds even remotely approaching human (and thus taking over those jobs).
This is from the summary, and it is wrong.
It implies that it will be awhile before the robots are as fast as humans and thus replacing their jobs.
This is not it at all. The robot doesn't have to be faster than humans, just cheaper. If the robot can only handle 12 items in 20 min, but it costs 50 cents an hour to run, then you can simply order up more robots.
---
I was watching a show on The Discovery Channel recently that was profiling new robots, and one of them is a new learning robot that you teach to do things by just showing it. It isn't fast, it takes maybe 5 minutes to fold a shirt for example, but you can teach it to fold a shirt by just showing it, the same way you'd show another human. Make some adjustments and corrections as it tries to do it and then it has "learned" how to do it. That robot costs just $30,000 to purchase and is expected to last for years. It can also be taught how to do other things.
It doesn't have to be fast if the job is one that you can scale up and just buy more robots, that robot will work 24/7/365, it never calls in sick, it never asks for a raise, and it doesn't complain about working conditions.
---
Amazon ships stuff from warehouses, it can simply order up 500,000 such robots if needed.
In some circles this is also being referred to as the "Put mommy out of a job contest"
ôó
The effect of 'contests' and 'rewards' is often a bunch of people coming up with an expensive one-off stunt that does exactly what is required for the prize money and nothing more, and does not really advance the state of the art. The various turing test contests are an example, as well as the Ansari X prize.
I agree with you, but not completely. For contrast, the Darpa grand challenge led to Google's self-driving car, which is poised to put 3 million truck drivers out of work.
The original grand challenge might actually be the problem - people looked at the success and tried to emulate it.
The differences might stem from problem specifications, or proper choice of problem. I remember the Darpa prize for building a machine to ascend the space elevator powered by a big searchlight at the bottom. The contest rules specifically required solar cells and electric motors, completely cutting out thermodynamic engines of various type (steam engines, stirling-cycle, other mechanical types). With so little room for innovation, it became a simple cutting-edge engineering chore.
Another prize involved a machine that can ride (and pilot) a tractor, dismount and walk into a building, find and turn a valve, and return. That doesn't quite fire the imagination as much as building a self-driving car, and the requirements are quite specific.
The Turing Test has no fundamental basis in theory, but it's led to some interesting algorithms like ELIZA, insights into human interaction (ie - that you don't actually have to be intelligent to keep up a conversation), and clarified the definition of AI a little.
So there's definitely value in having prizes, but I agree with you that it's not a 1-to-1 ratio of prize money to return.
(complete with a suction cup for lifting objects)
What I'm getting from this is, these robots suck. But they might replace some jobs due to a vacuum in the market for free workers.
Don't waste your vote! Vote for whoever you want, unless you live in a swing state it won't matter anyways
This thing about having a robot detect stuff and select it sounds error prone. If I was the one creating this gigantic vending machine, I would think it would be easier to put things in predictable places. (like they do in a vending machine) This way, you know that the thing in address x is a y. A step better would be to the have the robots put things away in their correct places. Some sort of easy to grab container could be used for each object to reduce the complexity of the hardware picking up the merch. Add a few routines for validating the weight of each object and packing them into boxes and you're all set.
I think the future of massive warehouses will involve RFIDs and standardised packaging. Products are packed far too randomly ... if standard packages were deployed, robots dexterity could perhaps be reduced?
Chances seem good that Amazon will ask future teams to build machines that are even smarter and faster.
The chances that Amazon will want future warehouse robots to be "even smarter and faster" are "good"?? Okay, I suppose the probability that they will want future robots to be dumber and slower is technically non-zero, but I'd at least revise "good" to "almost certain".
HAL still won't open that damned pod bay door.
Table-ized A.I.
I, for one, welcome our new Amazonian robot overlords.
It won't happen next year, or even 5 years from now... but at some point... all those drivers, from taxis to trucks, will become unemployable through no fault of their own. They simply will not be able to compete with the cost of a robot.
Oh, I agree with you and that sentiment completely.
To be specific, take a look at Manna, by Marshall Brain. It's an easy read, and it shows in frighteningly clear steps the two different ways the economy can go.
I'm all for automation, and I've worked on automation projects before. By and large, automation takes away those jobs that humans don't really want to do. Boring, repetitive, dehumanizing things like crop harvesting or long-haul driving.
While I recognize that automated production is the way of the future, I'm not quite sure how to get there. If there were some clear path, I'd be advocating it.
The best I can come up with at the moment is to point out how we're going to be in tough straits when 3 million people find themselves without a job in the next 5 years (and 2 1/2 million after that when short-haul driving is mostly automated, automated drone delivery of packages and mail and such.)
Pointing out the problems might nudge us into rethinking how economics works. That's all I can think of at the moment.
Do you have any ideas on how we can be part of that transition?
Exactly. A warehouse is nothing but a physical database.
Goods are placed and retrieved at specific locations. If something's wrong a manual check and correction can be done, or maybe have one or two robots with good visual recognition randomly check locations and the contents for correctness.
home
Completely missing the point. Robots don't need to be faster than humans at the job. they just need to be more cost effective. If a robot design was 1/2 the speed of a human picker but a quarter the cost (operating cost, not capital cost) then those jobs are gone.
Why invest time and money into research and development when you can just organize a competition and get everyone to do the work for you
There you go - YouTube video of the iHerb "vending machine"... pretty cool to it work - skip to around 1:18 to see the "vending machine" ... watch the whole thing to see how optimising the packing, before it gets stocked helps speed things up.
I saw an automated warehouse at Kodak in Rochester, New York, in 1975. What am I missing? Why is it so difficult now when it was done in 1975? The computers remembered where stuff was stored and the pickers just went to the spot and got the item. Some details omitted here, of course, but that was a long time ago when the relevant technology was relatively primitive.
... like most of Climatedot's article...
If I catch a robot handling my book like shown in the video I'm gonna wreck it and sell it for scrap.
Or mod it into my personal coffee machine.
We suffer more in our imagination than in reality. - Seneca
Their winning design combined a WAM arm (complete with a suction cup for lifting objects) and an XR4000 mobile base into a single unit
The chief designer, a Dr Davros of Skaro, CA, welcomed his 'supreme victory' in the competition, but questioned Amazon's decision not to proceed with the immediate replacement of their entire human workforce with his creation: 'Do you believe that I would let a lifetime's work be ended by the will of spineless fools like you? You have won nothing. I allowed this charade to be played out for one reason only. To find those men who were truly loyal to me and to discover those who would betray me! WE... I WILL GO ON!' Amazon officials, earlier invited to a demonstration of the improved 'Mark III Travel Machine', could not be reached for comment.
I guess CMU won't be fielding a team this year. /joke
As corporate-profits motivate more and more replacement of humans with automation, societal attitudes about employment are still moored in the Industrial Revolution age.
If you don't have a job, you're a bum - doubly so if you are a man, just to pile on the Puritanistic guilt.
I can't recall where I saw it - perhaps the Daily Dot or the Daily Beast, but it was a map of the most common job in every state of 2015 America. The most common in total for the country? Truck Driver.
In about ten years, the only jobs left will be in the private riot police squads, trying to hold back the tsunami of the unemployed assaulting the One-Percenter citadels to extract their revenge.
Quite a few in that mob will also have CS degrees.
Modularity.
This is their current situation. Stuff comes in different sized packages, and placement is not perfect.
Of course they could get improvements, even for human workers, if stuff came pre-checked, correctly classified and stuff. The thing is that's not their current status. The idea is to get rid of the picking human, without changing anything other than the human.
Self driving cars would be easy with the strategy you propose, just build intelligent roads, wired roads with wireless navigation, no people. Close to what a train is. It makes it a lot easier, but it just can't replace all driving, unless you change the whole infrastructure at once.
The purposes of mass surveillance is to make successful revolution impossible. The purpose of a middle class was to distract the masses until the security infrastructure was functioning at a revolt-proof level. Guess what? It has reached that point. What poses as the middle class is the "security thug" class. People in jobs that require a handgun carry permit, security clearance and/or a personal pedigree that disallows dual citizenship are the occupations that make up this new class.
The problem is that with empiricism (reality = space + time + matter + energy + chance and/or only that which senses and instruments detect) governing modern worldviews, the only two possible positions are Rand (ambition) and Marx (laziness). If you do not believe in "Fuck you, I got mine", then you are assigned the guilt of the deaths of tens of millions of people during the twentieth century by default.