German researchers have developed a way of training a flexible robot to learn movements by “goal babbling”.
German automation company Festo’s Bionic Handling Assistant was originally released in 2010, and based on an elephant’s trunk, with a high level of strength yet able to perform delicate tasks. The Assistant is made up of 3D printed segments forming its “trunk”, controlled by pneumatic tube “muscles”.
New Scientist reports that Dr Matthias Rolf and Professor Jochem Steil, from Bielefeld University have found a way to improve the ability of these robots, which did not originally have precision control software.
Using “goal babbling”, similar to the way babies learn movement, the researchers have developed a way to train the robot arms, so that after being manipulated into a series of positions, these are remembered and can be repeated autonomously.
“…the robot remembers what happens to the trunk's position when tiny changes are made to the pressure in the thin pneumatic tubes feeding the artificial muscles,” according to New Scientist.
“This creates a map that relates the trunk's precise position to the pressures in each tube.”
Festo’s arms were originally developed to make robots safer for working alongside people in factories, and were developed in cooperation with the Fraunhofer Institute for Manufacturing Engineering and Automation.