Generating pedestrian walking behavior considering detour and pause in the path under space-time constraints

Abstract

Recently, the demand for public open spaces in cities including pedestrian-oriented streets, squares, and plazas has been growing, for better social and psychological health of modern communities. To create lively public spaces, understanding various pedestrian behaviors has been increasing in importance. For example, we need to consider pedestrians’ behavior taking a detour to see beautiful scenery or stopping a walk to watch a street performance, in addition to usual goal-oriented behavior. Recent massive trajectory data from location-aware technologies, such as GPS, Wi-Fi, and Bluetooth technology, makes data-driven approaches promising. Inverse reinforcement learning, one of the data-driven approaches, is a powerful tool to infer the utility function with respect to points of interest (POI) from observed pedestrian trajectories. This approach can generate trajectories based on attractiveness of the open space with any POI distributions because the attractiveness is represented by the utility functions of the POIs. In this paper, we propose a two-step approach to generate the various trajectories. The first step infers the utility functions of the POIs and stochastic policy by inverse reinforcement learning, and the second step generates trajectories under the constraints on given destinations and arrival time. We numerically demonstrated that our approach generated not only goal-oriented but also detour-and-pause behaviors under given destination and time constraints in two types of environments representing an open public space.