Abstract
This paper describes a novel approach to modeling behavioral communication for humanoid robots that interact with their partners. Communication is established based on reaction through recognition of motion patterns of partners. Previous work of symbolization of motion primitives using Hidden Markov Models (HMMs) allows robots to recognize the observation and generate their own behaviors separately. In this paper we proposes a hierarchical model for communication, where HMMs in a lower layer abstract motion primitives of a robot and its partner and HMMs in a upper layer abstract interaction patterns. In the upper layer, output is recognition result of current interaction and input is generation of interaction. Shortcut between the output and input maintains the current interaction and realizes behavioral communication between the robot and the partner. Experiments of a humanoid robot interacting with its partner in a virtual world validate our principle of fundamental communication.
