A Pedestrian Agent Model Using Relative Coordinate Systems

Abstract

Pedestrian agent simulation is useful for not only the analysis of an accident but also designing pedestrians’ spaces. Agent Simulation of Pedestrian Flow (ASPF) has been evolved from a CA model; ASPF uses a pedestrian agent simulator with a cell space model. However, ASPF expresses an agent’s position by a cell unit, which restricts the movement of a pedestrian to only four directions; this causes unnatural behaviors of pedestrians. In addition, it does not allow individual differences in the walking speed of pedestrians. This paper is to address a pedestrian agent model — ASPF ver.3 -- using a relative coordinate system by revising ASPF ver.2, and also demonstrates the special characteristics of this model by the simulation of the intersection with straight and diagonal crosswalks in front of Shibuya Station.

The pedestrian agent model proposed in this paper allows an agent to walk in any direction and also accurately simulates individual differences of walking speed. From the simulation results, it can be shown: (1) in a pedestrian agent model where multidirectional flows exist, simulated behavior closely matches actual pedestrian behavior, as demonstrated by the successful modeling of a stratification phenomenon; and (2) due to the introduction of individual differences of walking speed, a trial calculation of the percentage of remaining pedestrians has become possible.