Amazon Picking Challenge | Hacker News Reader

Amazon Picking Challenge

amazonpickingchallenge.org

74 pointssigna1149 comments

I saw this competition at RoboCup in Leipzig. So many improvised-looking stuff going on, some with home vacuum cleaners to generate vacuum for suction cups.

Apparently, the winner, Team Delft from the Netherlands, had 0 errors on recognizing the objects. They used some sort of deep neural net for that.

http://www.tudelft.nl/nl/actueel/laatste-nieuws/artikel/deta...

First prize is $25,000 USD. Doesn't seem worth it given that it appears one must upload their code, eg: https://github.com/amazon-picking-challenge

In comparison, Netflix gave away $1,000,000 for a better recommendation algorithm.

It seems worth it if you want 25000$ and are willing to do this work for it, no?

I think what was meant was that a winning solution to Amazon's problem is probably worth a lot more to them than $25000 to them. Eg. cents saved picking item would rack up to millions saved annually. Amazon bought Kiva for $775 million.[1] a top rank robotic picking and packing solution would be worth as much too.

[1] http://www.bloomberg.com/news/articles/2012-03-19/amazon-acq...

Amazon won't be using any of this technology. Heck, nobody will be using this technology. This is just an opportunity to encourage people to get together and further research.

If you're capable of creating something that can win first place, and you do the work, then the result is worth far more than $25,000; so if you want $25,000 and are willing to do this work, then appropriate reaction would be to do that outside of the competition and sell it for much, much more than that.

You don't want to sell your IP to one competitor, what you really want to do is freely license it and then go consult for everyone in the market. Michael Hanack got extremely wealthy on that with his phthalocyanine semiconductors.

I think the total potential winnings could be $50,000 because it's $25k top event prize for "each task". The 2 tasks I see are: "stow task" and "pick task".

But to your point, $50k doesn't have quite the same glamour as $1 million.

What people don't seem to understand is that picking is hard even for humans. The technology isn't yet close to replacing human labor imho.

The bins used here are the best case scenario- every item visible, every item is actually in the bin it's supposed to be in, the bin only 20℅ full by volume. A human can pick at ten times the speed of these robots given those conditions. Fill the bin to the point that nothing more will fit in, and now try competing with a human.

Amazon Robotics isn't paying the winners $25,000 and actually using what they built next week. They're just trying to put research on the right path.

There appears to be a stray object replacement character (U+FFFC) in the title (right after the 'n' in Amazon). Perhaps someone can edit it out?

I assumed that was intentional - I read it as "Amazon Object Picking Challenge"

Something that seems really odd is they are working off the false assumption that current picking bins are suitable for the job. It seems every team was effectively competing against the bins, not devising an efficient way to put a subset of objects in storage into a container.

How do you know it's a false assumption and not a tested fact?

How do you know if it's a tested fact? And even if it was, why not experimenting without human constraints?

I don't know, that's why I'm asking!

And even if it was, why not experimenting without human constraints?

Who says they didn't? Really, all I'm asking is: why is everyone assuming that Amazon/Kiva didn't think about these things? Because unless one has a good reason to think they are taking a wrong approach, it just seems like hubris to flat-out claim so.

Because all of the teams are building very contrived contraptions to work around them.

Currently, the human workers work with those bins. Its easier to make the transition from human to robot labor if they work with the same bins, I presume.

It is kind of sad to see the animated gif at the bottom of the page. Robots are coming for his job.

http://amazonpickingchallenge.org/kivapick2.gif

It's sad that any of us has to have jobs. Can't wait for the day all of the humans are served by Robots (Yes.. reading too many Asimov novels of late :) )

The economic and societal forces that require having a job will not, on their own, magically change over upon the arrival of better robots. Without those changes, all you've done is taken away somebody's source of income.

The question is will we iteratively destroy each job until 5% of the current workers still can afford an opulent living while the other 95% are crushed by misery and accusations of inadequacies, or will we, as a society, realize that we can suppress the need for work altogether and try to make the smoothest transition possible?

This is not a technical question but a political one. Seeing how most politicians paint unemployed as subhumans that barely deserve to live, I am not that optimistic :-/

And it's interesting how the social democracies seem to have more of a plan on how to proceed than the purely capitalist nations.

Which purely capitalist nations are you thinking of?

I think it's a question that this election (in the US) is addressing, but only on the periphery. We're more likely to address the symptoms and misattribute them to other forces.

> The question is will we iteratively destroy each job until 5% of the current workers still can afford an opulent living while the other 95% are crushed by misery

If 95% are crushed by misery, where will the 5% get their money from? I'm guessing money will only be exchanged between the 5%, meaning the economy will be a lot different.

Also, what will keep the 95% from revolting? That amount of people will surely form a formidable army. What will save the 5%?

> Seeing how most politicians paint unemployed as subhumans that barely deserve to live, I am not that optimistic :-/

Changing the way we treat unemployed people is surely a necessary step forward, but it's not black and white. There are people who don't want to work because they're lazy or incompetent. There are people who want to work and are very competent, but can't find a suitable job. There are many types of people in between.

> If 95% are crushed by misery, where will the 5% get their money from?

I am not sure I see the problem. The 95% need goods and food. They buy it from the 5%. The 5% decide of the amount of money they accept to redistribute to the 95%. That can go on for a while.

> Also, what will keep the 95% from revolting?

Absolutely. But I am arguing that if your path toward a positive income has revolt as a necessary step, it is improvable. And even so, there is a risk:

> What will save the 5%?

What makes you think that soldiers will be sparred from automation?

> There are people who don't want to work because they're lazy or incompetent. There are people who want to work and are very competent, but can't find a suitable job.

It sounds to me that society would then benefit by having the lazy or incompetent people kept out of the work pool and paid to not work instead of having to make themselves absolutely necessary.

If you have worked in any medium or big organization I am sure that you can think of people who would have actually saved their employer's money by receiving a salary and not come to work.

>Also, what will keep the 95% from revolting?

Security robots!

More seriously, as robots do more and more, people will start to place a premium on having a human do the work where they notice the difference (so, not Amazon warehouses but probably the local Starbucks). With this factored in, 95% unemployment is unlikely until artificial general intelligence.

Would you pay a premium for being served by a human? Do you pay a premium to have a driver, to have a maid, to have a chef, when you can do without?

I know I would not. If a machine gives me accurate medical advice, I wont go to the doctor. If a machine makes good burgers, I won't pay a random student to flip burgers for me.

It would actually feel awkward and quite contemptful to ask a human to do the task of a robot.

I'll answer your specific questions first. Driver, no. Maid, no. (But keep in mind that robot maids/janitors are considerably farther away than self-driving cars. Cleaning environments not specifically designed for it is really complicated and I expect humans to outperform robots at it for longer than at most other kinds of physical labor.) For a fancier meal than burgers, yes, I'd pay a premium to have a human chef cook it. Doctor, definitely not in most cases.

As for it being "awkward and contemptful," I suspect you are imagining humans perform a task exactly as a robot would. That will be the minority of cases, because the economics are strongly against it in most. Instead, you'll pay humans for their ability to customize their work or the products they produce in a way that isn't economical to automate. (Burger flipping is doomed for this reason -- not much to differentiate humans form robots there.) Hence, chefs, plumbers, tailors, auto mechanics, specialist doctors, interior decorators, etc. will be assisted by "weak" AI but won't be replaced by it.

I have seen this kind of denial for more than 20 years now. I have seen people argue that driving could not be automated, that people would never accept an automated cashier, that webshops would never compete against physical shops.

The level of performance and customization that robots can achieve in every single of the task you quote is far superior to humans'. I expect none of your list items to present any technical problem in 10 years. Maybe plumber, give it 20.

I think you do my argument a disservice by equating it to the general trend of people denying X can be automated until X is automated. It is more specific: I say that some jobs (roughly, those requiring an agent to rapidly gain an understanding of existing unstandardized systems) need AGI to automate to human satisfaction in a way that doesn't require constant human supervision. It predicts, for instance, wide replacement of cashiers (which is already happening) and construction workers (which isn't yet).

>Maybe plumber, give it 20.

Perhaps we have different standards for "automated." Do you expect remote operators standing by ready to take over for a plumberbot or giving it hints at least once per service operation on average in 2036? If so, I find that technically plausible. I also expect new buildings to be built for automated service.

If we are talking fully automated human-quality repair and improvement of legacy plumbing, do you to expect a long time to get from there to AGI?

> roughly, those requiring an agent to rapidly gain an understanding of existing unstandardized systems

So, basically, machine learning.

Look, the tech is not there yet for an automated plumber, which is why it does not exist yet. 20 years ago we did not have the tech for automated driving but it finally got there.

I see machine learning improve every year, These last few years have seen a huge rise of popularity of deep learning approaches. And since the DeepDream gizmo, we hear less people talking about how creativity can't be automated.

Yes, in 20 years the tech will be here for a piece of software to figure out where the pipes must be in a house, make plans to test different hypothesis, plan a procedure of intervention, wiggle tubes together to make them fit without any other human instructions than "this tap does not work, go fix it". I have some supicions on what the correct way to go is, and am fairly confident that some other ways exist and that we will find one that will work.

I can't comment on AGI from this point. It really depends on your definitions. Some make AGI an unachievable goal. Under common and reasonnable definitions, yes, we will be close by then but we already are not very far.

And in my earlier comments I talk about 95% of unemployement just to cut some of the philosophical debate, but naturally I expect that we will reach 100% pretty quickly.

Right now western societies could easily go on with 30 hour workweeks. It would reduce unemployment and people who are working would have more time for their families or to do voluntary stuff. Another idea is a basic income which allows a simple and good life. White old men in suits are part of the problem why there's no real progress on these ideas and education is part of the solution.

My belief is that right now we give incentive to people to prevent the automation of their work. Were we to do the opposite (e.g. by promising a guaranteed life-time salary to every worker who got fired because of automation) it would not take 10 years to make the world work out of voluntary work alone.

"The chief aim of their constitution and government is that, whenever public needs permit, all citizens should be free, so far as possible, to withdraw their time and energy from the service of the body, and devote themselves to the freedom and culture of the mind. For that, they think, is the real happiness of life."

hmm... human or context sensitive robot post?

I was about to write the same thing. He's the last guy at the factory. Surrounded by other robots impatiently waiting on this slow human.

I used to think Amazon was great when they sold books. Now I'm not so sure.

but I suppose it's too late. Once humans are removed, they can drop prices lower than anyone with humans could possibly provide. yikes.

So when Amazon replaced all those cashiers, cleaners and rest of retail personnel with a site they were great, but replacing a few warehouse jobs is sad?

you make a good point, but it wasn't so obvious back then what was happening. the world was still dialup.

increase the wages and ROI for robots will look more attractive

On the long term issue of unemployment, instead of talking about how jobs were lost to robots, we should focus on how to make robots work for us, the rest of the population, not just Amazon. People should invest in these technologies so they get to share in the dividends.

Otherwise, there is a second path - vertically integrated communities, with their own farming, schooling, health and manufacturing, so that people have a job and provide for themselves.

the prize 25k is not relevant since most of the equipment can be taken from a sponsor and the participants are mostly universities, which means that every single cost is paid either by a sponsor or by amazon, we entered last year and ended up paying a total of 250€ for 5 members

personally i love the fact that this competition exists even if only to push research in the field

Why are people from Quebec not allowed to enter?

(Challenge is void in Brazil, Quebec, Crimea, Cuba, Iran, Syria, North Korea, Sudan, and where prohibited by law.)

Legal hurdles. Quebec has some interesting laws about conducting business in/with their province, particularly around doing business in both French and English.

I'm not sure about why Brazil and Crimea are not allowed to participate, but I know Cuba, Iran, Syria, North Korea, and Sudan will all be prevented due to US sanctions against those countries.

Google approaching this kind of challenge with neural networks[0]. Which is, perhaps, connected to why they are putting Boston Dynamics up for sale.

[0]https://www.youtube.com/watch?v=fuoX_866UFg

I've thought that maybe a better idea for amazon woud be containerising their products. Is it easier to delevop a picking robot that can deal with odd shaped objects, or a packing robot that can put an individual object into a cardboard box, with a universal way to grip?

They having this contest in germany for a reason - but if they are so cheap with the price money I don't think employees should feel threatened - yet.

I don't understand why this site is hosted on BlueHost...

Reply on news.ycombinator.com