Architecture and representation for handling dialogues in human-robot interactions

Human-robot interaction is an important component for robots operating in human environments and verbal interaction is in many cases the most intuitive and effective solution for humans. Managing dialogues between physical agents interacting in a physical environment brings additional challenges to virtual dialogue systems (eg. Siri or Google Now). More channels of information are available, as gaze or hands movements, which can modify or support verbal information. Also exophoric references to different parts of the environment can occur along the conversation. In this article we focus on the problem of extending the representation of the dialogue context to a physical environment and using this representation for resolving exophoric references. We also describe an architecture for integrating an open source dialogue manager in a service robot. In this architecture, the aforementioned representation is jointly built by different modules and it's used by the dialogue manager to ground utterances happening in the conversation. Finally we describe several experiments performed for assessing the utility of this architecture with actual robots in physical scenarios.