Landmarks reset self-localization but not goal-localization: Evidence for interactive error propagation in navigation

Abstract

This study investigated how self-localization errors, specifically position and heading errors, propagate to goal-localization error during navigation. Goal error can be defined as the difference between position and heading errors. The interactive-propagation hypothesis posits that position and heading errors arising from the same cues (e.g., self-motion) are correlated. Because goal computation incorporates both components, shared error can partially cancel, and the degree of cancellation is determined by their correlation. If landmarks reset position but not heading, this selective intervention may weaken their correlation, thereby reducing shared error and limiting the reduction of goal error even when position error itself is reduced. In contrast, the independent-propagation hypothesis holds that resetting position errors alone is sufficient to reduce goal errors. Two experiments in an immersive virtual environment tested these hypotheses. Participants learned five goal locations and two landmark locations arranged around a designated home location. After walking a two-leg path toward one landmark, they replaced the original goal locations either with or without the landmark visible at the test position. In Experiment 1, landmarks provided positional but not heading information; in Experiment 2, they provided both. The results showed that selectively reducing position error without reducing heading error failed to reduce goal error but decreased the correlation between position and heading errors. In contrast, reducing both errors reduced goal error without altering their correlation. These findings demonstrated that goal-localization errors depend not only on the magnitude of individual self-localization errors but also on their covariance structure, supporting the interactive-propagation hypothesis.