Structuring Human-Robot Interactions via Interaction Conventions

Abstract

Interaction conventions (e.g., using pinch gestures to zoom in and out) are designed to structure how users effectively work with an interactive technology. We contend in this paper that successful human-robot interactions may be achieved through an appropriate use of interaction conventions. We present a simple, natural interaction convention-"Put That Here"-for instructing a robot partner to perform pick-and-place tasks. This convention allows people to use common gestures and verbal commands to select objects of interest and to specify their intended location of placement. We implement an autonomous robot system capable of parsing and operating through this convention. Through a user study, we show that participants were easily able to adopt and use the convention to provide task specifications. Our results show that participants using this convention were able to complete tasks faster and experienced significantly lower cognitive load than when using only verbal commands to give instructions. Furthermore, when asked to give natural pick-and-place instructions to a human collaborator, the participants intuitively used task specification methods that paralleled our convention, incorporating both gestures and verbal commands to provide precise task-relevant information. We discuss the potential of interaction conventions in enabling productive human-robot interactions.