Abstract
The agent-based model is a well-established approach to carry out the simulation of crowd behavior under varying conditions and environments. However, this approach still lacks the ability to be fully automated and to take into account the agents' unfamiliarity and minimum awareness in an unknown environment. In this paper, we propose a multi-level multi-stage agent-based modeling framework that automates the decision-making processes during crowd evacuation in large-scale facilities. We demonstrate that crowd behaviors can be modeled at different stages and at several layers. We do this by integrating a probabilistic model with a dynamic generating process of intelligent guide agents to automatically generate decisions that are nearly optimal for neighboring agents. Specifically, we apply the framework to different evacuation scenarios and analyze its effectiveness in case of a high-density crowd. Our experimental results provide evidence that the hybrid multi-layered approach can be successfully applied to efficiently simulate agent behaviors in intensive crowd environments. The results show that employing our method, which is scalable, outperforms the standard conservative agent-based approaches.