Investigating Causality in Parent-Toddler-Robot Interaction

Omer Gvirsman, Goren Gordon

Abstract: Human robot interaction is often a very dynamic scenario, where both robots and humans perform verbal and non-verbal behaviors in response to each other. However, most user studies do not analyze the intricate dynamics of the interaction. Here, we present a novel methodology of a multi-participants, multi-dimensional, time-series Granger-causality analysis of a parent-toddler-robot triadic interaction. Our results show a specific causal flow of behaviors, namely, that the parent mediates the toddler-robot interaction by both verbal and non-verbal behaviors. Our proposed novel analysis can highly enrich understanding of the complex dynamics of human robot interaction.