A robotic arm that can dextrously manipulate objects has been developed by a team at the University of Washington in what is being hailed as a major breakthrough.
You may think robots are capable of pretty much any task these days. After all, they can perform surgery, go to Mars, detect bombs and even act as your receptionist. There's one thing, however, that they haven't been able to do – hold a pencil and spin it around.
The University of Washington team's hand can "not only perform dexterous manipulation but also learn from its own experience", the team explains. And this means it doesn't always need human direction. "Hand manipulation is one of the hardest problems that roboticists have to solve," said lead author Vikash Kumar. "A lot of robots today have pretty capable arms but the hand is as simple as a suction cup or maybe a claw or a gripper."
The team used a computer simulation to analyse the movements of a human hand spinning a tube in real time, before applying the algorithm to a real life robotic hand.
As the hand attempted the task it became progressively more capable of spinning the tube, using machine learning algorithms to teach itself how to more adeptly handle the object, which the team compared to someone "going home, doing your homework and coming back to school a little more intelligent".
"There are a lot of chaotic things going on and collisions happening when you touch an object with different fingers, which is difficult for control algorithms to deal with," said co-author Sergey Levine. "The approach we took was quite different from a traditional controls approach."
Other teams have developed similar projects; London based Shadow-Robot developed a "Dexterous Hand", which it said "provides unique capabilities for problems that require the closest approximation of the human hand currently possible".
This article was originally published by WIRED UK
