ImCasting: Nonverbal Behaviour Reinforcement Learning of Virtual Humans through Adaptive Immersive Game

Abstract

This research work puts a focus on the user experience of an alternative method to teach nonverbal behaviour to Embodied Conversational Agents in immersive environments. We overcome the limitations of the existing approaches by proposing an adaptive Virtual Reality game, called ImCasting, in which the player takes an active role in improving the learning models of the agents. Specifically, we based our approach on the Human-in-the-loop framework with human preferences to teach the nonverbal behaviour to the agents through the system. We introduce game mechanisms built around all the tasks of this Machine Learning framework, designing how a human should interact within this framework in real-time. The study explores how a game interaction in an immersive environment can improve the user experience in performing this interactive task, sharing the same space with the learning agents. In particular, we focus on the involvement of the players as well as the usability of the system. We conducted a preliminary evaluation to compare the design of our system with a baseline system which does not use any game mechanisms in teaching the nonverbal behaviour to virtual agents. Results suggest that our design concept and the game story are more engaging, increasing the satisfaction usability factor perceived by the participants.