Proceedings of the Annual Meeting of the Cognitive Science Society

Human navigation in maze-like environments can be approximated by distinct heuristic strategies, yet which strategy best reflects human behavior appears to depend on environmental structure rather than global optimality alone. We compare three families of navigation models—exit-oriented (EXO), target-angle (Angular), and target-distance (Distance)—across 25 grid mazes and multiple start locations. Model behavior is evaluated using success rate, path efficiency, and human-likeness defined as the correlation between model and human state-visit profiles. Across mazes, Angular heuristics are more human-like than EXO in the majority of environments, while hybrid combinations offer limited additional benefit. As an exploratory diagnostic test of whether maze structure contains information about relative strategy alignment, we label each maze by the EXO–Angular human-likeness gap (excluding ambiguous ties) and evaluate a simple leave-one-maze-out classifier using maze-level features. Results show weak but non-trivial signal under balanced accuracy, highlighting both the possible relevance of maze features and the current limits imposed by the small, strongly imbalanced dataset. These findings support a feature-based view of navigation strategy selection, in which environmental structure shapes the apparent human-likeness of different heuristics.