Grid-based partitioning for large-scale distributed agent-based crowd simulation

Abstract

Agent-based crowd simulation, which aims to simulate large crowds of autonomous agents with realistic behavior, is a challenging but important problem. One key issue is scalability. Parallelism and distribution is an obvious approach to achieve scalability for agent-based crowd simulation. Parallel and distributed agent-based crowd simulation, however, introduces its own challenges, in particular, effectively distributing workload amongst multiple nodes with minimal overhead. In order to ensure effective distribution with low overhead, a proper partitioning mechanism is required. Generally, human crowds consist of groups or exhibit particular patterns of flow, which are then reflected in simulations. Exploiting this grouping with an appropriate partitioning mechanism should enable efficient distribution of crowd simulation. In this paper we introduce a grid-based clustering algorithm which we compare to previous clustering approaches that used the K-means algorithm.