Proceedings of the Annual Meeting of the Cognitive Science Society

Computational approaches to human planning propose that people construct simplified mental representations of their environment, known as value-guided construals (VGC), for computational efficiency. The VGC model successfully predicts patterns of external attention when people are required to navigate mazes externally. However, it remains unknown whether internal selection processes untethered to external stimuli (i.e., offline planning and imagination) engage similar VGCs. We developed a novel maze navigation task in which the start and goal locations are revealed after stimulus offset. Despite the physical absence of the maze stimulus, we find that participants remain capable of constructing efficient mental representations aligned with the VGC model. These internal plans demonstrate significant but weaker influences of spatial grouping, similar to the pattern demonstrated for external attention. Together, our work reveals how VGC extend to purely internal, imagined task representations. These findings will help guide future research on how human-like planning may be implemented in artificial systems.