Abstract
It seems that actively walking around a new city leads to better spatial knowledge than passively riding in a taxi, yet the literature is decidedly mixed. “Active” exploration has several components that are often confounded. We test four contributions to spatial learning: (1) visual information, (2) vestibular information, (3) motor/proprioceptive information, and (4) cognitive decision-making. Participants learn the locations of 8 objects in an ambulatory virtual maze environment, and are tested on their survey or graph knowledge. Six learning conditions are crossed with two test conditions, yielding 12 groups: (a) Free Walking: participants freely explore the environment for 10 minutes, providing all active components. (b) Guided Walking: participants are guided along the same paths, removing decision-making. (c) Free Wheelchair: participants in a wheelchair press buttons to steer through the maze, minimizing motor/proprioceptive information. (d) Guided Wheelchair: participants are wheeled through the maze along paths that match the Free Walking condition, providing visual and vestibular information. (e) Free Video: participants press buttons to steer through a desktop VR maze, providing visual information and decision-making. (f) Guided Video: participants watch a participant's-eye video of the Free Walking condition, providing only visual information. In the test phase, participants are wheeled to object A and instructed to walk to the remembered location of object B: (i) Survey task: the maze disappears and participants take a direct shortcut from A to B. (ii) Graph task: participants walk from A to B within the maze corridors, with detours. In the survey task, shortcuts tend to be more precise when participants walk during exploration, indicating that motor/proprioception is critical for active learning. In the graph task, decision-making also aids learning, but only in conjunction with motor-proprioceptive information. These results suggest that desktop VR may be insufficient for full spatial learning.
Supported by NSF BCS-0843940.