Amazon crowns winner of first warehouse robot challenge | Hacker News Reader

Amazon crowns winner of first warehouse robot challenge

engadget.com

91 pointsoillio51 comments

That's a nice bin-picking system. Bin-picking is a classic robotic manipulation problem [1], and they're doing a good job. They target the object and approach it from an angle where pickup is possible. The unusual thing here is the range of objects picked. Most bin-picking has a far narrower range.

It's a vacuum picker, which is limited; they couldn't pick up a perforated metal pencil holder. But over 90% of the items tested were vacuum-pickable. That's probably a reasonable figure for Amazon's inventory, since books are vacuum-pickable. Amazon can sort their inventory into machine-pickable and non-machine pickable.

The new system is really slow, but that can be fixed.

[1] https://www.youtube.com/watch?v=Wh2GRA9enN4

While the pick up tool did use a vacuum pump are you sure it is using that for suction? It actually looked like the end effector was something along the lines of the 'coffee grounds' pick up tool. See (http://www.news.cornell.edu/stories/2010/10/researchers-deve...)

If it is that sort of gripper then it could pick up the perforated boxes just as easily.

the article explicitly stated that the perforated metal mesh could not be gripped by their vacuum effector

    > since books are vacuum-pickable

Really? I'd have thought they were one of a class that weren't. If you latch on to the outer cover, you'll splay the book, which can damage hard backs.

Perhaps all the books are shrink-wrapped to hold them shut.

How can self-driving cars handle going down the road at 70mph, yet bin-picker robots have to move so slowly to be accurate?

Seems like a contradiction to me.

It comes down to path planning degrees of freedom and dimensionality. Cars have two degrees of freedom and two dimensions, where most of these robots use 6 degrees of freedom in 3 dimensions.

Could it be accelerated with massive amounts of computing power?

Is the limitation here sensor speed or computing speed?

Certainly not motor speed, I see once the robot determines what to do, it immediately moves much more quickly.

Well just off the top of my head bin-pickers probably need to have millimeter accuracy when grabbing something so as to get close enough to suction it and not stab through it. Cars on the other hand are not functioning at that level of precision when traveling 70 mph, they are probably not even using that level of precision when parking, and they do not park at 70 mph.

A bin-picker robot needs to be accurate on the scale of a few millimeters while a car is probably fine with a few hundred centimeters of accuracy.

Think about that again. A few meters of accuracy is not going to drive down a lane safely, let alone park, negotiate an intersection, or pass a cyclist in traffic. A handful of centimeters, not a few hundred.

Perhaps sub-par mechanics is a factor. When building some robot arms I've discovered it's relatively hard to combine speed, precision and power.

Here's a demo of speed and accuracy using standard industrial robots:

https://www.youtube.com/watch?v=cR-YlZ9NdIA

They don't. It is a convergence problem. If you've ever seen a pick-and-place robot in a factor you know they can be quite fast.

Simply put, the convergence problem is getting to the correct orientation and position to positively pick up an item. Factory robots finesse that by putting parts in a narrow area in the work space but modern car assembly robots can pull parts in a wide variety of orientations and positions.

Current research has focused on visual systems as it is pretty cheap these days to put a couple of cameras near the end effector to provide a stereoscopic view to the effector converging on the target.

Here's a pretty sweet pick and place robot for reference: https://www.youtube.com/watch?v=vzHpDDttIBU

Drives me nuts to look at it for too long. And it looks like a spider.

That reminds me of the famous Lucille Ball chocolate scene [0], except in this case the robot is keeping up just fine!

[0] https://www.youtube.com/watch?v=8NPzLBSBzPI

When you're playing golf, why don't you use your 3 wood on the green?

In robotics, a lot of speed problems can be solved by spending more money on a bigger arm with better acceleration and backlash characteristics, reducing vibration by bolting it down to a stronger bit of floor, making your vision system less blurry by adding more and brighter lighting, avoiding people hitting their heads with a proper guarding system, and possibly throwing more CPU at your processing problem.

These issues are all eminently solvable, but the solutions take time and make the system harder to work on and more expensive, without generating a contribution to knowledge or academic papers.

Better to have a slow system you can transport to demonstrations than a fast system bolted to a lab floor that you can't demo anywhere :)

Congrats to Team RBO!

And whatever is the antonym of congrats to Amazon's warehouse workers!

Based on stories about those positions, even though they'll lose their job, hopefully they'll have a better position.

Amazon warehouse positions have high attrition and turnaround

Sure the robots can pick items out of bins, but can they wait in a 30 minute long security screening line at the end of their shift?

It's not just warehouse, it's engineering. Amazon as a company treats employees (outside of the executive suite) like dirt. IT's a retailer, more like Walmart than Microsoft, and so they have no respect for engineering. The only reason AWS happened was Vogels had enough power to isolate his people from the rest of Amazon (though I bet that team is suffering now too.)

My boss was trained to be a prison guard and couldn't operate spreadsheets, after a re-org, his boss was, I kid you not, one of those DMV ladies who has no interest in doing anything but covering her butt. (I mean, literally worked at the DMV and then got a job at Amazon managing dozens of engineers.)

They like to say things like "only hire A players" but their entire management is D players. It's hard to get a job as an engineer there but as an engineering manager all you need is a college degree in basket weaving and a demonstrated ability to kiss ass.

I do agree Amazon is just another retailer in the end. As to being a cutting edge technology company, I don't put Amazon in the same league as any of Musk's companies, but what do I know? I know I'm tired of UPS trucks delivering one package. I don't feel shopping on Amazon is good for the environment. Maybe one day--when the drones deliver all the stuff we really don't need? Sorry, I don't like these huge stores. In Amazons defense, they do allow little guys to sell through their store. Is Amazon as evil as Wallmart--no!

it's "condolences"

Cool, it would probably be less complex if the items were at one place all the time, like with a robotic warehouse: https://youtu.be/6EmR0KHBj_M?t=1m

I have built an automated warehouse very similar to the one in the video; you get different items in the same totes - any other approach is simply not feasible from a space perspective.

say no to spec work

I wonder if it would be better to have a smart bin that would simply push products down a slide to a conveyor belt instead.

Yes, it does seem like they're using a robot to solve a human problem. If the robots were the ones who decided how the shelves were initially stacked then surely there would be more elegant approaches to this? picking and packing is a tiny part of the whole ordering and delivery system which would probably be best re-designed from the top down if all parts are to be conducted automatically.

That said, I'm all for businesses encouraging research to be done in this area. Automation will swallow us all eventually, but it's going to take a while yet...

There's nothing like giving away something for tens of thousands of dollars that they could have sold for tens of billions.

Does amazon now own the IP? If not, then this sounds like a great idea, giving away highly useful ideas for everyone to use.

I don't think parent was referring to Amazon as having given anything away. Still, if you're not in a position to use this sort of tech, "tens of billions" is probably an exaggeration.

If you had a robot that could stow, pick, and pack any item in a warehouse, I have zero doubt you could negotiate $10B, simply for the fact that you could get Amazon, Staples, WalMart, Google, JD.com, Tesco, and AliBaba into a bidding war...and if none of them decided to pay you $10B, you could just offer your own FBA-like service for a fraction of what Amazon charges and make your own $10B.

Amazon is acting like a Pawn Shop owner taking advantage of the guy with the original Picasso by telling him that it was a common forgery. Nobody in their right mind should sign up for these contests, and it pains me to see intelligent people duped into it for "the exposure".

If you had a robot that could stow, pick, and pack any item in a warehouse...

TFA was pretty clear that none of the teams at the competition had anything close to this. The winner, which was far ahead of competitors, was able to move ten of a group of twelve specially-selected objects in twenty minutes. It's quite possible that some of the competitors will eventually create what you're talking about, but it doesn't seem that they've given away their grandchildren's inheritance just yet.

This presumes aquiring large quantities of money is a required motivation for developing new technology for one to be qualified as 'in their right mind'. To me this comes off like saying 'No one in their right mind should be a teacher and give knoweledge to others that they could use to get rich'.

Believe it or not some people enjoy the competition, the challenge of solving an interesting problem you had not previously considered, and over all advancing the technology of the world they live in far more than gaining absurd amounts of monetary wealth.

> Nobody in their right mind should sign up for these contests, and it pains me to see intelligent people duped into it for "the exposure".

This attitude disappoints me greatly.

The team that won was the robotics research unit at the Technical University of Berlin: https://www.robotics.tu-berlin.de, second place was MIT.

I hardly think they were "duped" into competing.

These people do research. Having a specified task that multiple teams "compete" on is there just to give a good comparison of the different approaches, otherwise everyone is picking their own tasks and it's hard to see what benefits / cons each system has.

What next, complaining that researchers publish open access papers for "the exposure" rather than selling the ideas?

Edit -

Here's a list of the teams: http://amazonpickingchallenge.org/teams.shtml

I have no problem with competition nor open research, which is probably how the researchers viewed this competition. My problem was how strikingly small the reward was for custom tailoring months of their efforts and research towards helping a single company eliminate hundreds of thousands of jobs and save billions of dollars. The top prize wasn't even enough budget for a lab assistant for a year! It sounds more like an evil plot from a sinister villain than it does an interesting research opportunity. If you want to do interesting research, don't let multinational companies with obvious agendas tell you what is interesting unless they are also willing to share the reward that they get from your hard work.

Amazon bought Kiva for $775M, so it may be close.

I know, I assumed they meant the researchers. I meant if the competition didn't require that the researchers give the IP to Amazon, then what has happened is just that Amazon have funded some research / a competition which has resulted in some things being released for all to use.

Can you skip the snarky comment and explain why this kind of contest is bad? To me it seems like a great idea for everyone. The contestants:

1. Can win money

2. Gain exposure

3. Network with other contestants

4. Potentially get more funding

The general problem with doing gigs "for exposure" is that it typically only gets you other gigs "for exposure".

These are talented grad students doing a gig "for exposure" on Amazon's dime with Amazon recruiters in the room hungry for talent. Many other recruiters were there (including Uber - first time I've seen them at an academic robot conference).

edit: should add that not only recruiters were there. Amazon engineers including their head of robot R&D were involved and available on the floor. Nice people.

(I'm not affiliated with Amazon and I don't do picking research)

The teams were almost all Universities. This is research, not business.

Pretty much the only one I spotted in the team list that wasn't, was mitsubishi. Unless they're really struggling these days, I don't think that exposure is an issue.

http://amazonpickingchallenge.org/teams.shtml

If you could teleport packages to my front porch, would you be willing to give that to Amazon in exchange for $25k and "exposure"? Amazon becomes the first trillion dollar company in history and you get your name on a press release and some snack money. I would never want any sort of exposure that told people that I'm easily duped about the value of my brain.

That's a bit of a ridiculous analogy. People are making incremental improvements. Perhaps or technology is good enough to get vc funding.

The problem is very very far from solved. Amazon is supporting students to work on problems Amazon cares about, and these events are very important for Amazon to recruit engineers. Some of these young people will go work for Amazon and some of them will start companies that Amazon will buy. I don't think they are being duped.

Amazon's robotics group is the Kiva Systems startup they bought for a huge amount of money a few years ago.

I was at the conference and dropped in to this event. Lots of energy and committed teams. Many/most teams did pretty badly. I heard (but couldn't confirm) that only one item was correctly picked on the first day. It's not easy.

Do they fund the research groups?

Having been a grad student (EDIT: Hey, it turns out my research group has published at ICRA! But not while this has gone on.), I get why these events are interesting/important. However it seems like AMZ could do a little more than <1/8th of an engineer in prize money (not counting the cost of running the conference).

I believe they did not directly fund these group's research. Just the challenge.

Amazon does in-house R&D and is precious about the IP, like most companies. Academic research needs to have flexible IP deals.

Most of the Amazon funding for this event went to supporting student travel and registrations, plus gear shipping. Details are probably confidential but think roughly an order of magnitude more than the cash prize. Shipping is expensive.

Funding challenges is smart. Some groups are doing funded research anyway, on more-or-less related things like grasping, vision, manipulation, planning, whatever. The challenge gets teams to focus on a more complete and realistic task, thus guiding the development of their general tech in Amazon's direction.

Bottom line: Amazon, the historically conservative low-margin retailer, showed up at the premier IEEE robotics research conference, brought some money and hired some grad students. Students worked hard, learned a lot. Amazon and winners got some good press. Awesome.

It's interesting how the best team was way ahead of everyone else. I'd be very interested to see what Amazon can do internally.

"Amazon, the historically conservative low-margin retailer"

They are slowly raising prices. It's off topic, but I have noticed their prices are slowly rising. I saw this happen to Home Depot, and Costco. I'm not knocking Amazon, I just got too comfortable clicking away, and should have been looking around for the best price.

Rarely have the comments on HN made me so sad.

These research groups are building useful tools and giving the results away to the world for free. If you'd invented the teleport system, it'd be so wide-rangingly amazingly useful that I hope it'd be released openly.

I generally agree with this sentiment but it's important to note that they don't know beforehand whether their technology is worth anything or not.

In a way, entering such contests is a way of lower risk in exchange for more modest rewards.

If they wanted to setup a startup instead it would have taken them ages to get the same kind of recognition this contest immediately gives them.

Reply on news.ycombinator.com