In an hour and a half, scientists from the University of Tübingen, Germany, thanks to a new learning algorithm, taught a robotic hand to play ping-pong.
The training began with the development of a computer simulation, in which the first stage of training a robotic hand took place. Thanks to the simulation run by the researchers, the hand learning algorithm was able to figure out the speed of the ball, the orientation in the space of the racket, and understand how this affects the flight of the ball. Using two cameras to track actions, the system every 7 milliseconds tracked the set parameters, which were subsequently processed by the algorithm, and decided how and where to move the hand.
The speed of robotic hand learning surprised scientists. But despite the visible successes, the researchers note that the algorithm makes mistakes when the ball flies either too fast, or vice versa – slowly. But at the same time, they are pleased with the result, noting that in the future, an improved algorithm will help develop robotic systems of a new level, which will find their application both in the economy and in industry.