Vector-based Pedestrian Navigation in Cities

How do pedestrians choose their paths within city street networks? Human path planning has been extensively studied at the aggregate level of mobility flows, and at the individual level with strictly designed behavioural experiments. However, a comprehensive, individual-level model of how humans select pedestrian paths in real urban environments is still lacking. Here, we analyze human path planning behaviour in a large dataset of individual pedestrians, whose GPS traces were continuously recorded as they pursued their daily goals. Through statistical analysis we reveal two robust empirical discoveries, namely that (1) people increasingly deviate from the shortest path as the distance between origin and destination increases, and (2) individual choices exhibit direction-dependent asymmetries when origin and destination are swapped. In order to address the above findings, which cannot be explained by existing models, we develop a vector-based navigation framework motivated by the neural evidence of direction-encoding cells in hippocampal brain networks, and by behavioural evidence of vector navigation in animals. Modelling pedestrian path preferences by vector-based navigation increases the model's predictive power by 35%, compared to a model based on minimizing distance with stochastic effects. We show that these empirical findings and modelling results generalise across two major US cities with drastically different street networks, suggesting that vector-based navigation is a universal property of human path planning, independent of specific city environments. Our results offer a simple, unified explanation of numerous findings about human navigation, and posit a computational mechanism that may underlie the human capacity to efficiently navigate in environments at various scales.