Wait Wait, Nonverbally Tell Me: Legibility for Use in Restaurant Navigation

Abstract

Legible motion is important for robots navigating in social environments, but certain contexts present additional constraints that require us to adapt what actions count as legible. In this work, we explore the legibility of navigation paths used by a robot waiter in a restaurant context. We highlight several areas where the existing formulation needs expansion for this context, with three areas of potential improvement: distribution of highly legible moments to within the field of view of observers, balancing between multiple audience members with different perspectives, and incorporating other social signals relevant to goal inference. We design an experimental scenario to begin examining the effectiveness of these models of improved "audience-aware" legibility.