Modelling distracted agents in crowd simulations

Abstract

Multi-agent simulations can provide useful insights into the movement of pedestrians in arbitrary environments for predictive planning and analysis. The fidelity of such agents is important for the validity of the associated analyses. Current methods tend to employ agent models that are largely homogeneous in both physical abilities and behaviours. However, actual pedestrians exhibit a wide range of locomotion abilities and behaviours. In this work, we take a first step towards identifying and modelling distracted behaviours, such as walking and texting on a cell phone. Our models relate reported changes to the locomotion patterns and sensory abilities of distracted pedestrians to the corresponding parameters of a commonly used crowd simulation steering approach. We demonstrate experimentally that accounting for even a few of these behaviours significantly alters the flow patterns of the simulated agents. This impact affects overall crowd behaviour and is reflected in several crowd statistics including flow rate, effort, and kinetic energy.