An Ideal Navigator Model of Human Wayfinding: Learning One’s Way Around a New Town

Searching for a target in an unfamiliar environment requires acquiring, storing, processing, and recalling spatial information. We use datasets from two virtual reality navigation experiments to design and validate a simple biologically-inspired computational model of these processes. Subjects in both datasets play taxicab drivers, picking up and delivering passengers in a series of small virtual towns. After just a few deliveries, subjects learn to generate minimumdistance, novel paths between passengers and their destinations. Subjects’ rapid learning and near-ceiling performance inspires the creation of an ideal navigator model, which makes optimal use of sensory information to learn navigationally relevant spatial information about the environment as quickly as possible. We use MAGELLAN, a simple, two-parameter, ideal navigator model, as a benchmark against which to assess possible sources of subject error. By systematically degrading the ideal navigator’s vision and memory, MAGELLAN accounts for human subjects’ mean performance in both datasets, and correctly predicts the difficulty that subjects encounter in navigating different environments.