Image: University of Washington

If you’ve ever tried to learn how to spin a pencil in your hand, you’ll know it takes some concerted effort—but it’s even harder for a robot. Now, though, researchers have finally built a ‘bot that can learn to do it.


The reason that tasks like spinning a stick are hard is that a lot happens in a very short time. As the stick moves, the forces exerted by the hand can easily send it flying out of control if they’re not perfectly co-ordinated. Sensing where the stick is and varying the hand’s motion is an awful lot for even the smartest algorithms to handle based on a list of rules.

But a team from the University of Washington has built an incredibly dextrous five-fingered robotic hand and developed software that allows it to learn from its own experience. Over time, it can sense forces and friction, learning how that feedback can be used to move objects within its grip. So while it starts off slowly, over time it can improve dramatically. Its ability to spin a stick is made possible by the fact that the hand is fast—very fast, as you can see in the video—and the fact that the robot teaches itself rather than having a scientist impose rules upon it.

“Usually people look at a motion and try to determine what exactly needs to happen - the pinky needs to move that way, so we’ll put some rules in and try it and if something doesn’t work, oh the middle finger moved too much and the pen tilted, so we’ll try another rule,” explained Professor Emo Todorov to E&T. “What we are using is a universal approach that enables the robot to learn from its own movements and requires no tweaking from us.”


[University of Washington via E&T]