Next Leap for Robots: Picking Out and Boxing Your Online Order

From The Wall Street Journal:

Robot developers say they are close to a breakthrough—getting a machine to pick up a toy and put it in a box.

It is a simple task for a child, but for retailers it has been a big hurdle to automating one of the most labor-intensive aspects of e-commerce: grabbing items off shelves and packing them for shipping.

Several companies, including Saks Fifth Avenue owner Hudson’s BayCo. and Chinese online-retail giant Inc. have recently begun testing robotic “pickers” in their distribution centers. Some robotics companies say their machines can move gadgets, toys and consumer products 50% faster than human workers.

Retailers and logistics companies are counting on the new advances to help them keep pace with explosive growth in online sales and pressure to ship faster. U.S. e-commerce revenues hit $390 billion last year, nearly twice as much as in 2011, according to the U.S. Census Bureau. Sales are rising even faster in China, India and other developing countries.

That is propelling a global hiring spree to find people to process those orders. U.S. warehouses added 262,000 jobs over the past five years, with nearly 950,000 people working in the sector, according to the Labor Department. Labor shortages are becoming more common, particularly during the holiday rush, and wages are climbing.

. . . .

Picking is the biggest labor cost in most e-commerce distribution centers, and among the least automated. Swapping in robots could cut the labor cost of fulfilling online orders by a fifth, said Marc Wulfraat, president of consulting firm MWPVL International Inc.

“When you’re talking about hundreds of millions of units, those numbers can be very significant,” he said. “It’s going to be a significant edge for whoever gets there first.”

. . . .

In RightHand Robotics’ Somerville, Mass., test facility, mechanical arms hunt around the clock through bins containing packages of baby wipes, jars of peanut butter and other products. Each attempt—successful or not—feeds into a database. The bigger that data set, the faster and more reliably the machines can pick, said Yaro Tenzer, the startup’s co-founder.

Hudson’s Bay is testing RightHand’s robots in a distribution center in Scarborough, Ontario.

“This thing could run 24 hours a day,” said Erik Caldwell, the retailer’s senior vice president of supply chain and digital operations, at a conference in May. “They don’t get sick; they don’t smoke.”

Link to the rest at The Wall Street Journal (Link may expire)

8 thoughts on “Next Leap for Robots: Picking Out and Boxing Your Online Order”

  1. Any bet Amazon is already miles ahead of them? And isn’t it funny that the only place they mention Amazon is at the very end?

    “Some companies hope to speed development by making some research public. Inc. will hold its third annual automated picking competition at a robotics conference in Japan later this month. For the first time, entrants won’t know in advance all the items the robots will need to pick.”

    “At the University of California, Berkeley, a team is simulating millions of attempts to pick 10,000 objects. Funded by Amazon, Siemens AG and others, the project is meant to build an open-source database for use in any automation system, said Ken Goldberg, the professor leading the project.”

    Something that strikes me as ‘off’ is the way they’re going about it. While for ‘testing’ throwing random stuff at it can show you where you may need to work on something, in the warehouse you would ‘train’ the system as to what to expect and how to grab it (and how to tell that it did a successful ‘pick’ and how to call for ‘help’ if things go wrong.) A little training and the system should be able to handle 99% of it’s job, only needing ‘help’ when something major goes wrong.

    “Robot developers say they are close to a breakthrough—getting a machine to pick up a toy and put it in a box.

    It is a simple task for a child …”

    Someone either never had kids or has forgotten how many times they had to ‘help’ those kids put their things in the box (and has forgotten all those things that ended up in that box that shouldn’t have! 😉 )

    • I suspect that this problem is like Optical Character Recognition, 99% accuracy is unusable.

      • Like Al said below, if they can get 99% they let the bots do all but the 1% they might have trouble with (and they could let the bots try playing in the 1% – with a human double-checking their work.)

        Already boxed things should be easy enough – so long as they’re in the bin they’re suppose it be in (but even humans can grab what looked like the right item that was in the right place.)

        If ‘I’ was testing one of those bots for the worse case scenario, I’d get one of those round bins Walmart sells the $5 DVDs in full of those mixed DVDs in at all sorts of angles, throw in a dozen (still sealed) candy bars and a couple of toys and tell the bot to go get me the DVD that I know there’s only a couple of in the bottom and see how it solves that problem! 😉

    • Some items are easier for robots than others. Robots can be limited to orders that contain items they can handle. As they get better, the scope of their operations can expand to meet their abilities.

      Manufacturers will also be making their goods robot-friendly. Put a hard point on packages, and box the items to make it easier for the robot.

  2. I suspect if they could get 99% consistently, they could rearrange the warehouse to pick 99% by machine and hand-pick the 1% until machine hands got better…

    I hope they adopt machine picking to mattresses… I temped at a mattress company warehouse and that job was a killer.

    Al who had an odd variety of temp jobs

  3. The “99% is not good enough” point is important. If you know where the 1% errors will occur, they are easy enough to cope with, but if a human checker has to look over the robot’s shoulder to spot random-appearing errors, you might as well dump the robot let the human do it.
    If a human proof-reader still has to perform a full proof-read to correct the random errors in a computer proof-read, the computer proof-read did not gain you much.
    Review the literature on control charts and systematic errors to get an idea of the challenge.

Comments are closed.