Markus Solbach¹, Mohit Goyal², Sakar Khattar², Jayant Varma¹, Bjorn Vlaskamp³, Konstantine Tsotsos⁴, John Tsotsos¹; ¹York University, Toronto, Ontario, Canada, ²Google, Mountain View, California, USA, ³Google, Seattle, Washington, USA, ⁴Google, Toronto, Ontario, Canada

Functional vision relies on active gaze to acquire task-relevant information, yet how the visual system adapts to environmental instability remains poorly understood. This study investigates adaptive strategies during a complex 3D assembly task under dynamic constraints. We recruited 90 participants to replicate 3D magnetic block models on a central turntable rotating at 0, 5, 10, and 15 RPM. Participants were monitored under "Free" or "Restricted" mobility conditions using synchronized binocular eye-tracking and high-fidelity motion capture to record objective performance, oculomotor metrics, and 6D head-hand kinematics. Results reveal a critical interaction between mobility and workload. At lower speeds (5 RPM), the "Free" condition provided a clear performance advantage with faster response times. However, at the highest workload (15 RPM), this trend reversed: The quality of performance observed solution methods degrades as speeds increases. Conversely, the "Restricted" condition demonstrated greater stability, maintaining consistent efficiency even as difficulty increased. Longitudinal analysis indicates that mobility significantly influences adaptation (p < .001). Performance was best at 0 RPM, and participants were allowed to move "Free." Generally, the "Free" group showed better learning compared to the "Restricted" group. However, head translation revealed that while movement remained consistent at lower speeds, for the 15 RPM condition, a drastic increase appeared in cumulative head translation (> 35m) for "Free" moving participants. Additionally, oculomotor analysis revealed a processing shift: fixation counts increased linearly with speed while saccade amplitude contracted (p < .001), implying a transition from global scanning to localized tracking. We demonstrate that the benefit of physical mobility is not universal. While allowing participants to move generally aids adaptation, it becomes counterproductive during extreme dynamic tasks. By quantifying visual behaviour in a complex 3D environment, we move beyond simply observing that performance declines to documenting exactly what fails.

Acknowledgements: This research was supported by grants to the senior author (John K. Tsotsos) from the Air Force Office of Scientific Research USA and the Natural Sciences and Engineering Research Council of Canada and Google Canada.