According to UC Berkeley designers and a subsequent writeup earlier this week via ArsTechnica, SpeedFolding utilizes a neural network called BiManual Manipulation Network, a pair of industrial robot arms, and an overhead camera system to analyze each wrinkly, unfolded garment. From there, it can subsequently arrange the fabric into shape on average in under two minutes, with a 93 percent success rate. SpeedFolding learned how to properly and quickly fold through studying 4,300 human and machine-assisted examples. Coupled with the extra arm (previous, similar robots usually only employed one limb or none at all), SpeedFolding essentially blows all past iterations out of the water, given that the previous record only clocked in at between 4-6 folds per hour.
