Gaze in Maze Solving

From seamlessly following moving objects to quickly switching gazes during visual searches, individuals use various eye movement techniques depending on the situation. While eye movements offer valuable insights into cognitive processes, creating generative models of these movements has proven to be incredibly challenging. To date, the computational goals driving eye movements remain largely elusive.In this study, we addressed these challenges in the context of a well-known spatial planning task, maze-solving. We gathered eye movement data from five participants and developed advanced generative models of eye movements using an innovative differentiable framework for gaze fixations and gaze shifts.Our findings reveal that eye movements are most accurately predicted by a model designed not for optimal task performance, but rather for simulating an object's path through the maze.Specifically, it suggests that individuals employ mental simulation as a strategy.Additionally, our study highlights the intricate relationship between maze-solving and gaze behavior. We observed that participants frequently adjusted their gazes to anticipate maze pathways, demonstrating a dynamic interplay between visual search and spatial planning.