Humanoid behaviour learning through visuomotor association by self-imitation

Abstract

Learning by imitation renders a means for more natural human-robot interaction and is potentially the primary form of teaching. Imitation provides a potential means of automatically programming complex systems without extensive trials. This paper presents a method of imitation learning based on mapping the demonstrator's motion patterns to the observer self-posture. The self-posture is acquired through observing and recognizing the mirror image of its own body posture while performing the action in front of a mirror. First, different kinds of behavioural actions are learned through motor babbling by randomly performing different actions. The actions are learned through a novel probabilistic model called Topological Gaussian Adaptive Resonance Hidden Markov Model. During learning, the observer also extracts visual features through optical flow from self-observation in a mirror image. Second, after learning, a visuo-motor association is developed through novel Topological Gaussian Adaptive Resonance Associative Memory. Finally, after learning the demonstrator performs a similar action in front of the robot and the he robot recalls the corresponding motor command from the memory.