Realism Matters. Eye movements in VR reveal prior tool knowledge, intended interaction, and end-state comfort planning

Seminal studies on human cognitive behavior have been conducted in controlled laboratory settings, demonstrating that visual attention is mainly goal-directed and allocated based on the action performed. However, it is unclear how far these results generalize to cognition in more naturalistic settings. The present study investigates active inference processes revealed by eye movements during interaction with familiar and novel tools with two levels of realism of the performed action. We presented participants with 3D tool models that were either familiar or unfamiliar, oriented congruent or incongruent to their handedness, and asked participants to interact with them by either lifting or using. Importantly, we used the same experimental design in two setups. In the first experiment, participants interacted with a VR controller; in the second, they performed the task with a more realistic interaction setup that allowed differentiated hand and finger movements. We used linear mixed models to determine the odds of fixations on the tool effector vs. handle before action initiation. The results show that participants fixate more on the tool9s effector part before action initiation during the use task for unfamiliar tools. Furthermore, under more natural interaction conditions, subjects fixate more on the tool9s handle as a function of the tool9s orientation, well before the action was executed. Secondly, the spatial viewing bias on the tool reveals early fixations are associated with the task and familiarity of the tools. In contrast, later fixations are associated with the manual planning of the interaction. In sum, the findings from the experiments suggest that fixations are made in a task-oriented way to extract the mechanical properties of the tools to plan the intended action well before action initiation. Further, in more natural conditions, fixations are made towards proximal goals of optimally planning the grasp even when the perceived action on the tools is identical. Thus, the realism of the action in a virtual environment seems to be as important as the sensory stimulation.