UC Berkeley’s Robot Breaks Speed Record ... for Folding Laundry

November 11, 2022 by Seth Price

UC Berkeley sets a world record for robotic laundry folding with SpeedFolding, a methodology for teaching a robot to take on the boring and repetitive task of folding clothes.

Robots have long been accused of stealing jobs from humans. Some jobs, however, may be worth stealing due to their undesirable nature. Any job that is repetitive, boring, dangerous, or requires high precision is ripe for automation. Along these lines, consider how many people have relegated the tasks of washing and drying laundry to machines, only to do what they can to avoid folding laundry by hand. SpeedFolding hopes to change all of that with their latest robot.


Artist conception of ABB bimanual robot using SpeedFolding

Artist’s conception of SpeedFolding. Image used courtesy of SpeedFolding


The Challenge of Automating Laundry Folding

Folding laundry, for its boring and repetitive nature, is quite challenging to automate. Different types of clothing require different folds. Some items are thin and difficult to grasp with robotic grippers, and often items will static cling together. Their arrangement in the laundry basket is random and a long sleeve or pant leg removed from the stack can dump other items on the floor. While the human brain takes all of the necessary steps to prevent these problems, training a robot to do so is much more difficult.

Identification of laundry items is the most difficult part. Once the item has been properly identified, it can be moved out of the laundry basket and into a pre-programmed location. The act of folding the laundry can be performed uniformly by a well-trained robot. In essence, the actual motions required to fold the laundry are not much different than other machine-tending tasks found in an industrial setting.



SpeedFolding is not the robot itself, but rather a methodology for teaching a robot to fold laundry. University of California, Berkeley developed a system for analyzing the way humans fold laundry—4300 actions, to be exact. From there, they were able to teach robots to anticipate shapes of laundry from a crumpled mass, smooth out the item, and then fold them along user-defined fold lines. All of this is performed at the less-than-human speed of 30-40 garments an hour. While this may not sound impressive, it is lightning fast compared to the previous record of 3-6 folds per hour. 


folding template example used by SpeedFolding

Example of several folding templates used by SpeedFolding. Image used courtesy of SpeedFolding


UC Berkeley's AUTOLAB developed, tested, and wrote a paper on this system. The published results show that their test robot was able to fold laundry at 30-40 garments an hour, most pieces taking under 2 minutes, at a success rate of 93%. It should be noted that this system was not compared to the rate of the average teenager, but it would likely be faster and more successful. 

In their methodology, they discussed the use of a neural network system called BiManual Manipulation Network (or BiMaMa). This system communicates the necessary action to two separate manipulator arms in much the same way as the human brain communicates with both arms. Two arms allow the robot to smooth out wrinkles and make them into interpretable shapes, recognizable in its library. Humans do this automatically.


ABB bimanual robot using SpeedFolding to fold a t-shirt that was extracted from a laundry basket in a random orientation. Video used courtey of Griffig


The Future

Hopefully, this innovative code and integration will lead to future developments in laundry-folding machines. While the ABB robot’s $58,000 price tag still keeps it out of the author’s (and many others’) homes, it still shows promise. Perhaps this form of household drudgery will be a thing of the past as new laundry automation unfolds.