Natural head and body orientation for humanoid robots during conversations with moving human partners through motion capture analysis

Abstract

In conversations between humans, a natural body and head orientation towards the interlocutor is important for their social interaction. Humanoids communicating with humans have to learn how to orient themselves properly which becomes a challenging task in the case of moving conversation partners. Studies of conversational behaviour often involve only stationary partners. In this research, we perform a motion capture study to address the scenario of moving subjects. Specifically, study trials were recorded during conversation between a human participant and interlocutor, with a focus on the behaviour of the head, shoulders, and feet. The results help better understand how humans behave while conversing with non-stationary interlocutors. The data from the trials was used to generate a mathematical model describing the relationship of the angle at which the interlocutor is located to the orientations of the head, shoulders and feet while tracking is performed. A new model setup to couple the motion of the interlocutor, the head and the shoulders is introduced, as well as a model to represent stepping in order to better replicate participant behaviour. The models are evaluated and then deployed to the REEM-C Humanoid Robot, for the purposes of generating a natural behavior of the robot and improving human-robot interaction.