Learning the Inverse Dynamics of a Robot Arm by Auto-Imitation

Auto-imitation is a type of unsupervised learning which enables a controller to quickly acquire the inverse dynamics of plants. The basics of the method are demonstrated using a one-segment robot arm. If embedded into a learner-operator schema, the inverse model can be updated while the robot performs purposive or even arbitrary movements. So, the proposed algorithm would be an alternative to supervised learning, for instance via feedback-error. It is presumed that autoimitation like learning governs especially social learning in humans.