Elizabeth Aje¹ (), Stephanie Shields, Kathryn Bonnen; ¹Indiana University

Eye tracking has been an important method in vision science since its introduction in the 19th century, but for most of that time, eye trackers had to be attached to laboratory setups that kept the participant’s head in a fixed position. Only relatively recently has the development of head-mounted eye tracking allowed for the recording of eye movements while participants move both their head and body. Unfortunately, the freedom head-mounted eye trackers allow also makes accurate calibration more difficult. One potential strategy for improving calibration is the use of artificial intelligence (AI). Baumann & Dierkes (2023) report that the Pupil Labs Neon eye tracker successfully uses a deep learning approach to accurately track eye movements and estimate gaze direction in a wide variety of use cases, without requiring any calibration. We sought to validate their findings by testing the Neon’s gaze estimation accuracy using two screen-based tasks and two walking tasks. To test for an impact of viewing angle, participants performed screen-based tasks in one of three configurations: (1) while sitting, looking straight ahead; (2) standing, looking slightly downward; and (3) standing, looking at the ground. To test for an impact of slippage, participants performed the screen-based tasks both before and after the walking tasks. Overall, our results suggest that the Neon estimates gaze accurately, even without applying participant-specific corrections. Accuracy sometimes varied across viewing angles and across repeats, but the differences we observed were relatively small and were not consistent across participants. We did find differences in the overall level of accuracy across participants, and our next steps will be to explore the cause of those differences. Additionally, future work will test how much of an improvement the Neon provides over mobile eye trackers that require calibration.