Simulating human interactions in supermarkets to measure the risk of COVID-19 contagion at scale

TL;DR

This paper presents a novel agent-based simulation model for retail environments to assess COVID-19 transmission risk, utilizing space transformation and historical data to generate customer interactions at scale.

Contribution

It introduces a theoretical model for collision probability, a method to generate customer paths from basket data, and a calculation for simulation requirements, enabling large-scale risk assessment.

Findings

Developed a space transformation to the Torus for collision modeling

Created a method to generate customer paths from historical data

Calculated the number of simulations needed for statistical significance

Abstract

Taking the context of simulating a retail environment using agent based modelling, a theoretical model is presented that describes the probability distribution of customer "collisions" using a novel space transformation to the Torus $Tor^2$. A method for generating the distribution of customer paths based on historical basket data is developed. Finally a calculation of the number of simulations required for statistical significance is developed. An implementation of this modelling approach to run simulations on multiple store geometries at industrial scale is being developed with current progress detailed in the technical appendix.