High Level Planning in Crowd Simulation

Abstract

One of the main challenges in Crowd Simulation is that a problem does not only need to be solved, but it often needs to be solved for the large number of agents that are simulated. In this project we want to let all agents determine what steps they need to perform to reach their personal goal. While Artificial Intelligence has many methods to determine such steps, most of them are not suited to do this for the large number of agents in Crowd Simulation. In this thesis we show that it is possible to use an Hierarchical Task Network to solve high level planning problems in Crowd Simulation. Compared to the lower levels of planning, the relative time required to plan this for a Train Station scenario is very small (<0.1%). In addition, we show that by using the proposed Agent Profiles approach, we can solve extremely complex planning problems for large numbers of agents (up to one million) in a time that is short enough for a real-time simulation.