A Microscopic Agent-Based Model for Pedestrian Dynamics at Scramble Crossings

Abstract

Simulating pedestrian behavior using microscopic agent-based models helps urban planners optimize traffic flow and improve safety. To reduce pedestrian casualties at traffic intersections with high foot traffic, urban planners may choose to install scramble crossings, which halt all vehicular traffic and facilitate diagonal crossings of streets. Standard models, however, primarily focus on unidirectional or bidirectional pedestrian flow rather than the turbulence of multidirectional pedestrian flow. This NetLogo-based pedestrian model intends to quantify the critical pedestrian density at which orderly free flow of pedestrians collapses into disorderly jamming behavior. In this simulation, the movement of each agent is governed by the rules of the Headed Social Force Model, wherein a pedestrian is treated as an anisotropic point particle driven by a combination of attractive local driving forces and repulsive social forces that guide it towards a desired corner of the intersection. The pedestrian model was validated against CCTV footage of Sukiyabashi Crossing in PeTrack by using the YOLOv8x object detection model to track individual pedestrian trajectories to generate heatmaps of pedestrian motion, which are compared to similar heatmaps generated from the pedestrian model with SSIM. Simulations were then run across a range of pedestrian densities from 0.1 ped/m² to 2.0 ped/m². Between 0.3 ped/m² and 0.6 ped/m², the mean directed mobility and mean nearest-neighbor distance between pedestrians plummets, dropping by 16% and 46%, respectively, indicating a transition from order to disorder. Knowing this threshold will inform more resilient urban designs that aim to lower pedestrian density.