Abstract
Human binocular vision is adapted to statistical regularities in the natural environment such that depth perception and binocular eye movements are precise, fast, and comfortable. We measured for the first time the statistics of fixation, disparity, and vergence-accommodation conflict in virtual reality (VR) and compared them to statistics when viewing the natural world. Specifically, we collected eye-position and 3d-scene-geometry data during tasks performed in VR (first-person shooter games, environmental simulation, and beat/rhythm games) and the natural environment (making a sandwich, ordering coffee, taking indoor and outdoor walks, and editing text). We found that gaze is biased towards straight ahead and farther distances in VR relative to the natural environment (VR: 1.25m; natural environment: 0.67m). We also measured the statistics of the vergence-accommodation conflict (VAC) and found that ~80% of fixations in VR produce significant conflict. From this, we determined the optimal screen distance to minimize discomfort due to VAC. In the natural environment, the vertical horopter and natural-disparity statistics exhibit a top-back pitch, a pattern not present in the VR environment. Specifically, the VR produces significantly smaller near (crossed) disparities in the lower visual field than the natural environment, a difference as large as 900arcsec. Finally, we tested whether observers prefer VR content that is consistent as opposed to inconsistent with the statistics of the natural world. We found that content that violates the top-back pitch of the natural world generates more discomfort and reduced performance. We conclude that the mismatch between the statistics of the VR and natural environments leads to discomfort and reduced performance in VR headsets. Our findings inform improvements to VR headset design and content in order to be more consistent with the statistics of the natural environment.