Teaching robots like babies

Muscle memory isn’t just for beings that have muscles anymore. German engineers Jochen Steil and colleague Matthias Rolf have worked with a robotic segmented tentacle-like arm to develop a way to teach the robot motions by example.

In the past, programming the robotic movements yielded less than accurate movements, limiting use of these robotics to experts. Inaccuracies of a few centimeters made the robots unusable in the real world. But with the advances that Steil and Rolf have made, the arm can now be “taught” its movements, precisely and easily.

Their new process, called “goal babbling”, requires the user to manually teach the movements to the arm. By physically moving the arm to create the movement, the robot learns the pressure and positions that are required for each specific move. This mimics the way that a baby learns movements, gradually learning what muscles need to move.

Using this method, the arm can be taught various movements, execute them accurately, and repeat these movements to achieve a variety of tasks.

Source
Source