Creating Dynamic and Density Dependent Indicative Routes for Crowd Simulation

Abstract

Crowd simulation has emerged as an important research field within computer science. Simulations for new commercial complexes and large events are increasingly becoming the standard. Additionally we are able to show a growing range of realistic specialized crowd behaviors. The latest generation of research has provided us with powerful approaches for exceedingly larger crowds.

Within the field of crowd simulation we see a strong distinction between global and local methods. Unfortunately, combining these micro and macro methods can be difficult and can lead to problems when the results of these methods conflict. In this thesis we will define an approach that attempts to unify two existing global and local approaches by functioning on a meso level between the macro and micro planning. Agents remain individual entities with personal goals and settings.

We will show that currently global and local methods can heavily counteract each other and lead to undesirable behavior. By storing global paths for individual agents that are dynamic, instead of static paths from one specific point location to the next, we achieve increased flexibility. We introduce the concept of paths that are planned from the edge of one cell to the edge of the next cell. To protect the quality and benefits of the original methods, we will show that the resulting paths for individual agents not under the influence of other agents will remain the same. We will implement this approach to obtain an implementation that unifies global and local pathing.

Experiments show that the paths reached with our approach mitigate undesirable and unrealistic results like large clusters within densely agent-filled regions making it a valuable contribution to current research. The concepts can be applied to a range of pathplanning solution that depend on a combination of separate global and local approaches.

This project has been made possible through a collaboration between Utrecht University and INCONTROL Simulation Software. The algorithms have been developed for Pedestrian Dynamics, which is a pedestrian flow simulator by INCONTROL.