Abstract
Schnapp and Warren (VSS 2007 doi:10.1167/7.9.758) showed that observers in a virtual maze could navigate successfully between remembered objects despite the presence of 'wormholes' that took participants from one location to another without the participant noticing and which made it impossible to reconstruct a consistent metric map of the maze. In our experiment, participants in immersive virtual reality had to collect 4 coloured targets in a specified order. Targets were placed in a virtual maze, hidden behind low walls so that the participant had to come close to see and collect them. From the last target, participants had to point in the direction of the other targets. There were 5 repeats with the same target order ('learning' phase) and three repeats with a different target order where the shortest route between pairs of targets was often different from the learned routes. We tested participants in two different maze configurations and for each we tested them with 0, 1 or 3 wormholes. The wormholes increased the length and number of turns between pairs of locations in the maze but did not alter the topological structure. After the learning phase, participants could navigate through the labyrinth efficiently: participants chose the shortest routes in 97% cases for metric scenes, 83% and 75% with 1 and 3 wormholes. However, pointing accuracy was severely affected by the addition of wormholes. The standard deviation of the pointing errors was 13°, 47° and 51° for 0, 1 and 3 wormholes and this spread depended whether the route between a pair of targets was consistent with a metric reconstruction (19°) or not (49°). Participants may not use a single representation for both tasks; at least, the extent of metric information they use depends on the task they are set.
Meeting abstract presented at VSS 2017