Conference on Cognitive Computational Neuroscience

Goal-directed navigation requires continuously integrating noisy cues about self-motion and position relative to landmarks, representing them internally into a sense of location and heading direction, path planning, and executing motor actions sequentially. Previous studies have uncovered numerous idiosyncratic and seemingly inconsistent patterns of errors in navigation. Here, we show that both the broad successes and specific errors people make when navigating to a target dynamically arise from the continuous interaction of sensory uncertainty, uncertain internal spatial beliefs, and movement variability. Specifically, we find that a computational model of navigation based on optimal feedback control under uncertainty simultaneously predicts variability in trajectory endpoints, explains how sequential egocentric landmark observations form an uncertain allocentric cognitive map, and how this internal map is used in route planning toward navigational goals. Importantly, this model provides a unifying account for numerous phenomena, explains how and why different navigational strategies are adopted, and reconciles previous reports of seemingly sub-optimal cue integration behavior.