(OS invited talk) Live Crowd Simulation

Abstract

Agent-based crowd simulation is a powerful tool for investigating crowd dynamics. It has been applied to study various problems in which experiments are infeasible. However, crowd simulations have been conventionally limited to offline what-if analyses and have rarely been used for real-time applications. This is because real crowd movements involve enormous uncertainty and are thus difficult to predict. Recently, a concept of agent-based simulations incorporating real-time observation data, i.e. live simulations, was proposed. Live simulations can accurately forecast an ongoing event to facilitate timely interventions. The available real-time crowd observation data are continuously enhanced with the increasing number of observation channels. Live crowd simulations incorporating such observation data can be a transformative tool to facilitate better crowd management. This paper first discusses a limitation of conventional offline crowd simulations followed by a review of recent progress in live crowd simulations.