Estimating Visual Focus of Attention in Dyadic Human-Robot Interaction for Planar Tasks

Visual focus of attention (VFOA) has been shown to be a strong indicator of human intent and engagement, and thus, an important feature to human-robot interaction (HRI). As a mature research problem in computer vision, VFOA recognition has been implemented in environments as complex as group meetings. In this paper, we define and leverage the relatively simple and highly-constrained environments of recent dyadic HRI for planar tasks, and validate the application of simple statistical methods for VFOA classification. We discuss the potential of VFOA in emerging HRI research and platforms.