Robots Are Really Bad At Folding Towels : Planet Money Machines are surprisingly bad at doing things an average 8-year-old can do with ease.

Robots Are Really Bad At Folding Towels

  • Download
  • <iframe src="https://www.npr.org/player/embed/407736307/408010770" width="100%" height="290" frameborder="0" scrolling="no" title="NPR embedded audio player">
  • Transcript

MELISSA BLOCK, HOST:

We hear it all the time - robots are going to take our jobs. But Steve Henn from our Planet Money podcast reports there are some tasks that still stump the cleverest computers. One word - laundry.

STEVE HENN, BYLINE: We've been racing humans against robots over at Planet Money, and in our first race, we wanted to start with something simple - really simple. Folding a towel, right, just a towel. Pieter Abbeel is a computer scientist at Berkeley. Seven years ago, he set out on a quest - to teach a robot how to fold laundry. Abbeel's robot looks like a little person. He's about 5 feet 2 inches tall. His head has kind of an E.T. shape. He has big eyes, a prominent chin, a long neck, and like E.T., he can stretch to get taller. And he has these two arms with grippers on the end for picking up laundry. This robot even has a name.

PIETER ABBEEL: BRETT - B-R-E-T-T - and it stands for the Berkeley Robot for the Elimination of Tedious Tasks.

HENN: To Abbeel, Brett is perfect. His team just needed to teach BRETT how to decipher the chaos contained in every American's laundry basket. And when they started, BRETT just kept poking at his pile of towels, looking for a corner to pick up. Pieter showed me this picture of the robot standing there looking confused.

ABBEEL: It's not - not figured it out yet at this point here (laughter).

HENN: He looks so sad.

ABBEEL: I think it's very easy to start feeling for him.

HENN: To a computer, it's hard to see where underwear stops and where a towel begins. Every pile of dirty laundry is different. So Abbeel's team spent months staring at laundry baskets, holding towels up in the air, taking pictures of laundry. The solution was super complicated.

ABBEEL: Can you use multiple images to build a 3-D model of the current shape? 'Cause once you can do that then you can analyze that 3-D shape, find where the corners are, and so that's what, effectively, we spent a lot of our time on is figuring out that process.

HENN: But once that 3-D corner recognition software was working, the towel problem was basically solved - well, kind of. BRETT could fold the towel.

ABBEEL: So our robot initially would take about 20 minutes per towel.

HENN: What? Twenty minutes per towel.

ABBEEL: So yes, 20 minutes.

HENN: Before I showed up in Pieter Abbeel's lab, I searched YouTube for videos of people folding towels.

I found one of a toddler.

ABBEEL: (Laughter).

HENN: OK, so she looks - how old do you think she is - maybe 18 months old?

ABBEEL: I would say so, yeah, about 18 months old. She's...

(SOUNDBITE OF YOUTUBE VIDEO)

UNIDENTIFIED WOMAN: You can do it - again.

UNIDENTIFIED CHILD: No.

ABBEEL: She's trying to learn to fold this towel.

(SOUNDBITE OF YOUTUBE VIDEO)

UNIDENTIFIED WOMAN: She's folding a towel.

HENN: Look, she grabbed the corner.

ABBEEL: She did find the corner pretty quickly there.

HENN: Oh, look at that. Look how fast that was.

(SOUNDBITE OF YOUTUBE VIDEO)

UNIDENTIFIED CHILD: Grandma, I did it (unintelligible).

UNIDENTIFIED WOMAN: All right, again.

HENN: That took seven seconds.

ABBEEL: (Laughter) Impressive.

HENN: And she's 18 months old. How long did you work on teaching your robot?

ABBEEL: Two years (laughter) but two years of several students spending a lot of time on it.

HENN: And even after they perfected laundry in the lab, if you brought BRETT home and threw something in he had never seen, like a bundled up sock or an inside-out onesie, BRETT could end up stumped.

ABBEEL: I think that's one of the most surprising things. Once you start working in robotics, you realize that the things that kids learn to do - maybe up to age 10 - what happened - whatever we learned before age 10 are actually the hardest things to get a robot to do.

HENN: Machines need rules and one of the ways to figure out if a robot is going to take your job is to ask yourself what are the rules in what I do every day? Is my job a series of decisions based on a pattern, based on rules? Or is my job really more like a giant pile of messy laundry? Steve Henn, NPR News.

Copyright © 2015 NPR. All rights reserved. Visit our website terms of use and permissions pages at www.npr.org for further information.

NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.