Abstract
The underlying geometry of cognitive maps could take different forms. At one extreme, spatial knowledge might have a Euclidean structure that preserves metric distances and angles (Gallistel, 1980). At the other, it might have a topological graph structure that only preserves the connectivity between known places. We investigate this question by introducing two “wormholes” that link remote places in a virtual hedge maze, making it non-Euclidean. A control maze had the same layout with no wormholes. Participants walked in an immersive virtual environment (10m x10m) with a head-mounted display (80deg FOV) and a sonic/inertial tracking system (50–70ms latency). In the learning phase, they explored the environment and were then trained to a criterion of two trials walking directly from a Home location to each of 10 landmark locations. In the test phase, they walked from Home to location A and thence to location B. On probe trials, the possible routes from A to B included a shortcut through a wormhole that required initially walking away from B. If participants learn both environments with comparable numbers of training trials and errors, and take advantage of wormholes during testing, it suggests that they rely on a graph-like structure for navigation. Conversely, if they have more difficulty learning the wormhole environment, or fail to use wormholes and make errors during testing, it suggests that they acquire more metric spatial knowledge. The results will allow us to better understand the geometrical structure underlying human cognitive maps.
Acknowledgement: Funded by NSF BCS-0214383