Distributed Simulation of Crowds with Groups in CrowdSim

Simulating large crowds of individuals is socially important, e.g., for developing and studying evacuation or rescuing in dangerous situations. Such simulations remain complex due to the scalability challenge: simulating thousands of virtual characters is computationally expensive, especially when taking into account psychological factors and group-specific behavior that play a crucial role, e.g., in panic situations and highly crowded environments. In this paper, we make two new contributions: 1) we extend the HiDAC agent-based modeling approach with the aspects of group formation and movement, and 2) we implement our approach within the CrowdSim system, including the possibility to distribute the simulation process across several compute servers for better performance. We report experimental results on scaling the distributed simulation of a real-world evacuation scenario in a building using several compute servers.