Abstract
The present experiments tested three hypotheses about the structure of spatial knowledge used for navigation: the Euclidean, Topological, and Stability Hypotheses. We selectively destabilized three geometric properties of the environment during learning: metric, neighborhood, or graph structure. If spatial knowledge is primarily Euclidean, performance on all tasks should deteriorate when metric structure is destabilized; if spatial knowledge is primarily topological, performance should reflect the graph when metric and neighborhood structure are destabilized; if navigators acquire whatever geometric properties are stable during learning, performance should reflect the stable structure in each environment. Method. During learning, 10 groups of participants (N=120) walked in one of four virtual hedge mazes containing 10 target objects: (1) The Control Maze preserved all three properties. (2) Elastic Maze I preserved the place graph but destabilized Euclidean and neighborhood structure by stretching some hallways across neighborhood boundaries 50% of the time. Metric information (from path integration) but not visual information for these boundaries was available. (3) Elastic Maze II added visual information for these neighborhood boundaries. (4) The Swap Maze preserved neighborhoods but destabilized Euclidean and graph structure by swapping certain object locations 50% of the time. In the test phase, spatial knowledge was probed in one of three navigation tasks: (a) metric shortcut task, to assess Euclidean knowledge, (b) neighborhood shortcut task, to assess knowledge of neighborhoods, (c) route task, to assess graph knowledge. Results are inconsistent with Euclidean and Stability hypotheses, but support the Topological hypothesis. Metric shortcuts are highly unreliable (Angular Deviations ~24° in Control maze) and neighborhood shortcuts suffer even when neighborhoods are stable (21.4% incorrect in Swap maze). But the place graph is learned in all environments (96% accuracy on route task). Spatial knowledge appears to be primarily topological, consistent with a labeled graph that incorporates approximate local distances and angles.
Meeting abstract presented at VSS 2014