On-the-fly behavior coordination for interactive virtual agents: a model for learning, recognizing and reproducing hand-arm gestures online

In human conversation, verbal and nonverbal behaviors are coordinated by the interlocutors on the fly. To participate in this, artificial conversational agents must be able to create, adopt, and adjust behaviors flexibly and autonomously. We present a novel approach to learning behavioral patterns online, Ordered Means Models (OMMs), that meets the demands of dynamic behavior coordination in interaction. We describe how OMMs enable the virtual agent VINCE to engage in playing Rock-Papers-Scissors games, in which he learns, adapts to, and recognizes every human opponent's gestures on-the-fly such that he becomes unbeatable after only a few rounds. An evaluation study demonstrating this is presented.