Learning and communication via imitation: An autonomous robot perspective

This paper proposes a neural network architecture designed to exhibit learn ing and communication capabilities via imitation. Our architecture allows a "protoimitation" behavior using the "perception ambiguity" inherent in rea l environments. In the perspective of turntaking and gestural communication between two agents, new experiments on movement synchronization in an inte raction game are presented. Synchronization is obtained as a global attract or depending on the coupling between agents' dynamic. We also discuss the n onsupervised context of the imitation process and we present new experiment s in which the same architecture is able to learn perception-action associa tions without any explicit reinforcement. The learning is based on the abil ity to detect novelty or irregularities in the communication rhythm.