A Cognitive Architecture for Flexible Imitative Interaction Using Tools and Objects

Imitative responses constitute an important basis of rich interaction between an infant and a care-giver. This paper proposes a cognitive robotic architecture for generating a variety of flexible imitative responses in situations where a robot interacts with humans using various toy objects. The architecture is based on a cognitive scientific hypothesis that "true imitation" capability consists of a set of sub-functions called "pseudo-imitation". Examples of pseudo-imitation are: (1) assimilation of bodily movements, (2) evoking stereotyped actions associated with a viewed object (tool). We propose that the synergy of these pseudo-imitations leads to higher level imitation. The above hypothesis is realized by implementing the individual pseudo-imitation components and the synergy function as parallel real-time modules. The robot flexibly exhibits various pseudo-imitation as well as their synergy, called "action unit imitation", by the dynamic collective workings of components and the function