Knowledge-based interactive robot: System architecture and dialogue manager

Abstract

Development of robots that interact with people intelligently in the human-friendly manner is still a challenging research topic. Consider welfare and friend robots that will live with us in the long term, their interactions with human are different from those of traditional dialogue systems. They are usually multi-modal, and multi-topic. Robots should be also able to learn through the conversations in order to be capable of new things. We aim to integrate robotics and knowledge technology to achieve such robot. This paper presents its system architecture and dialogue manager. The architecture is distributed. The robot is decomposed into multiple components called primitive agents. The special agent dialogue manager acts as the brain of the system. It perceives changes in the environment and makes actions by inferencing based on the knowledge base. Frame-based knowledge technique is used to represent the world of interest. It is extended to support time-based layer and frames actions priority. The current system can perform state-based and frame-based types of dialogue, and learn simple facts and rules given explicitly by human. The prototype system is developed on a humanoid robot and an example of multi-modal human-robot interaction is shown. © Springer-Verlag Berlin Heidelberg 2004.