Task-related gaze behaviour in face-to-face dyadic collaboration: Toward an interactive theory?

Visual routines theory posits that vision is critical for guiding sequential actions in the world. Most studies on the link between vision and sequential action have considered individual agents, while substantial human behaviour is characterized by multi-party interaction. Here, the actions of each person may affect what the other can subsequently do. We investigated task execution and gaze allocation of 19 dyads completing a Duplo-model copying task together, while wearing the Pupil Invisible eye tracker. We varied whether all blocks were visible to both participants, and whether verbal communication was allowed. For models in which not all blocks were visible, participants seemed to coordinate their gaze: The distance between the participants' gaze positions was smaller and dyads looked longer at the model concurrently than for models in which all blocks were visible. This was most pronounced when verbal communication was allowed. We conclude that the way the collaborative task was executed depended both on whether visual information was available to both persons, and how communication took place. Modelling task structure and gaze allocation for human-human and human-robot collaboration thus requires more than the observable behaviour of either individual. We discuss whether an interactive visual routines theory ought to be pursued.