Eye Paths and Choices: A Simulated Approach to Maze Navigation Analysis

Abstract

Decision-making during maze navigation requires integrating visual sampling, memory, and planning. We present a reproducible eye-tracking pipeline for grid-maze problem-solving, applied here to simulated data as a validation step. Twenty-four participants were simulated across eight trials each in a 9 × 9 maze with six decision points; 292 fixations were detected via the I-DT algorithm.
Fixations at decision-point AOIs were numerically longer than those elsewhere (114 ms vs. 106 ms), with a medium effect size (*d* = 0.66, 95% CI [0.38, 1.11]), but did not reach significance (*t* = 2.88, *p* = .087). Gaze efficiency showed zero variance and could not be tested. A three-state Hidden Markov Model produced a degenerate solution (98% of fixations in one state). K-means clustering identified three strategy profiles (silhouette = 0.32), with PC1 explaining 47.2% of variance.
The pipeline is released as open, reproducible code. Because the present results are derived from simulated gaze, they establish methodological benchmarks rather than substantive empirical claims. Application to real eye-tracking hardware is the natural next step.