Abstract
Studying peripheral vision requires sufficiently complex stimuli and well-defined tasks. Here for this purpose, we explore the use of mental maze solving -- tracing a path through a maze without a pen or a finger. Mental maze solving involves both cognitive and perceptual processes and can provide a controlled environment to study visual crowding, a well-known limiting factor in peripheral vision. The task requires observers to make a series of eye-movements and combine multiple views of the maze into a stable representation, which could then enable observers to decide where to look next. In the current study, we investigated visual crowding by measuring maze solving performance as a function of maze appearance, recording eye-movements, and modeling the results. Human observers solved a series of 2D mazes while we recorded the time to solve each maze and eye-movements. The perceptual features of mazes were manipulated to alter the level of crowding in each maze (i.e., path thickness, visual complexity). Experiments 1&2 showed that observers were faster at solving mazes with thinner and less-complex paths, suggesting visual crowding is a significant factor in determining maze solving performance. In Experiment 3, we tested whether a crowding model (The Texture Tiling Model; TTM) can predict fixation allocation during maze solving. Observers were slower and made a larger number of eye-movements while solving crowded mazes, as predicted by TTM, although TTM underpredicts the number of fixations. To examine this underestimation tendency, in Experiment 4, we tested whether observers could detect targets placed on/off the maze paths near the model fixation locations. Preliminary data suggest that observers were over 70% accurate in detecting targets, even though targets were located on average 2.7 times farther than their fixation locations. These findings suggest visual crowding as a significant factor in mental maze solving.