Modeling Moderation for Multi-Party Socially Assistive Robotics

Abstract

Socially Assistive Robotics (SAR) is the study and development of algorithms that enable robots to help people achieve their goals through social interaction, such as coaching, therapy, or companionship. One extension of this field to multi-party interactions is the problem of moderation, in which an agent supports a group in achieving its goals. We validate the use of robot moderator in an open-ended interaction. The implemented algorithm does not use speech recognition; the robot moderates the interaction based on who is speaking, not what is said. We show that such a SAR moderator is accepted into human-human interactions, and can improve group cohesion and increase participant speech in those interactions. A number of studies have directly examined the role of a moderator in online or teleconference contexts (e.g., [1]). In the virtual agents community, Bohus and Horvitz developed an approach to enable a virtual agent to take part in group conversations in open environments [2]. There has been substantial work on robot-centered interactions, where the robot provides a service to participants, such as a bartender [3] or game master [4]. Additionally, Mutlu et al. used a social robot to manipulate the roles of the participants [5] and Matsuyama et al. developed a framework for enabling an agent to integrate a participant into a conversation [6]. This work contributes an autonomous controller for a robot moderator, evaluated with participants in an open-ended discussion.