In a massive breakthrough, a UC Berkeley scientist has built a robot that can … fold a dish towel.

Since Dr. Pieter Abbeel first posted a video of his team’s PR2 robot folding towels on YouTube in March, over a half million viewers have watched the industrious robot pick up, smooth out, fold and stack a messy pile of towels.

Although fascinating to watch, it’s hard not to be underwhelmed. Folding? A towel? Human origami experts can make a snake with 1,000 scales folded out of a single piece of paper! They can fold 3-D tessellations that would make M.C. Escher cross-eyed!

Hasn’t robotics advanced beyond the laundry room yet?

But as it turns out, in order to make the PR2 fold towels, Abbeel and his team had to grapple with a big challenge in robotics: Robots suck at randomness. Your family’s towels strewn here and there – that chaotic mess of wet cotton is a roboticist’s nighmare.

As Prof. Abbeel told a BBC interviewer recently, “It’s surprisingly difficult for a robot to do any laundry task.”

I was skeptical, so I rang up a scientist who has dedicated much of his career to folding – physicist and origami master Robert J. Lang. Bay Area resident Lang, who unites mathematics with origami and has diagrammed over 500 origami models, including some of the most complex designs ever created, was already familiar with Abbeel’s robot. “I was impressed,” he said.

It all comes down to how hard it is to manipulate what Abbeel calls “non rigid objects.” Or in Lang’s parlance, “disordered structures.” Piles of towels are hard for robots to deal with because no two look alike.

In origami, whether folding a simple crane or one of Lang’s advanced creations, such as Rattlesnake, Opus 55, or his mind-bending Molecular Tessellation, the starting orientation of the paper is uniform. But not so in the world of laundry, where towels, sweat pants and duvet covers lay about in random, stinky heaps. The challenge for the robot lies in image recognition, in interpreting what the device “sees” with its cameras so the device can find and identify the corners and other features of the towel.

“The impressive part is the thinking part of it, not the folding part,” Lang said.

Copyright by Robert J. Lang

Rattlesnake, opus 539 by Robert J. Lang, 2008, folded out of one uncut rectangle of Thai unryu paper.

The thinking part is, of course, an algorithm, and Abbeel’s algorithm has proven to be robust. The PR2 can fold 50 towels in a row without a mishap. While accurate, it’s not exactly a speed demon: The team has increased the robot’s performance to four minutes per towel, up from the 25 minutes it took it back in March.

Besides having proved a champ with non-rigid objects, the towel-folding robot is part of a larger movement in open source robotics.

The PR2, built by Willow Garage of Menlo Park, runs on the open source ROS robotics framework, which Abbeel and other research teams across the country are using to develop new robotics capabilities. While towel-folding is perfected at UC Berkeley, other scientists are programming the PR2 – Willow Garage has distributed 11 in a new pilot program -- to lift heavy objects, open doors and do other discreet operations. The research results are available to the ROS community at large.

Next up for Abbeel’s team is expanding the PR2’s suite of capabilities to the entire laundry room so it can not only fold towels, but tackle the whole dirty clothes basket -- sorting, washing, drying and folding and stacking all your irregularly shaped clothes. Algorithms similar to the ones behind folding laundry may also have other applications, such as tying knots in a surgical setting, Abbeel explained.

But what about robots and origami? Robots already exist that can fold basic origami shapes, and computer programs such as Lang’s own TreeMaker software are used widely in advanced origami design. Lang and others also use computer-controlled lasers to score paper when extremely precise folds are required.

But there’s one element that’s especially hard for the PR2, or any robot or computer: creativity. While specialized software can map out a very complex origami form, there’s no guarantee of its artistry. “Aesthetic judgments are much harder to automate and turn into an algorithm,” Lang said.

I asked Prof. Abbeel if he had considered aesthetics when he created the towel-folding PR2. Surprisingly, there’s a sort of delicate beauty to watching the clumsy looking robot crisply shake, smooth out and neatly stack each towel.

“What I’ve noticed is that when people see the robot fold the towel, they perceive it as the robot really taking care of the towel,” Abbeel said.

What does Abbeel himself see? The robot doesn’t care of course. It’s just doing what it’s programmed to do. But Abbeel sees something most of us wouldn’t even consider: the personality of the student who contributed this or that bit of code.

“When I see it do something," Abbeel said, "I tend to recognize the character of the student who’s been working with the robot, more than seeing it as the robot character.”

Much mainstream media coverage of Abbeel’s robot has used the towel-folding PR2 to poke a little fun at robotics – is this really all you brainy scientists can do?

But watching it fold and fold, I now notice something else: Prof. Abbeel’s determination. And creativity.