(Reuters) РResearchers develop a robotic hand with dexterity that mimics a human hand and can learn to handle objects better and better without human intervention.  Sharon Reich reports.

This is the closest engineers have ever come to mimicking the dexterity of the human hand.

A team at the University of Washington are developing a five-fingered robotic hand with unprecedented capabilities. It can handle objects and learn from its own mistakes.

Graduate student Vikash Kumar says that there are several variables needed to develop a successful robotic hand.


“One of them is ability to act really fast because in case you are losing control of an object you want to quickly go and re-position yourself … That hand should be capable of doing a lot of different things and second, which is the most important piece of the puzzle, we need to have some kind of an intelligent unit that knows how to exploit that dexterity in order to solve those problems.”

Driving the hand is a series of actuators programmed to control cables that manipulate the hands’ fingers and thumb with human-like reflexes.

What really sets this robotic hand apart from earlier models is its ability to learn as it attempts to carry out a task, in much the same way humans learn ‘on the job’. So with each attempt at a task the hand actually improves its performance.

Assistant Professor Sergey Levine wrote the algorithm that the hand uses to learn.


“If we can get robots that can learn autonomously, they can learn on their own, then we don’t need to figure out ourselves how the robot should perform the skill. We just need to define the skill and have the robot figure it out on its own. I think that’s tremendously powerful because then we can have robots that can figure out on their own how to do a wide variety of behaviors.”

While there’s still a ways to go, the commercial possibilities of a robotic hand that doesn’t need programming and can learn tasks in an unanticipated environment are limitless.


“I think in order to succeed in those unstructured environments learning will be a crucial component because we need robots to deal with uncertainty, to deal with variable environments, with a lot of complexity they simply don’t see in a factory setting.”

For now the hand continues to improve at it’s current task.

Next up, the researchers plan to share their design with engineers from around the world next week at the IEEE robotics conference in Sweden.