Crowd evacuation using multi-agent system with leader-following behaviour

Abstract

Crowd evacution is a major safety issue in many public places. Efforts have been made to design the best evacuation plans using simulations and modeling of certain evacuation scenarios, in order to help prepare for the actual event. In some scenarios where the majority of evacuees are not familiar with the layout of the building, evacuees most familiar with the layout of the building should be appointed as group leaders. These leaders guide their evacuee group to safety, usually outside the building. This research employs leader-following behaviour, and multi-agent system with Reciprocal Velocity Obstacles (RVO) for navigation, thus creating leader-follower relationship between evacuees. Each evacuees, i.e. follower, tries to navigate to follow their respective group leader to reach their escape point together. Experiment results demonstrate the navigation of hundreds of agents in an evacuation scenario, which shows that the evacuees safely reached the evacuation point between 92.597 to 129.71 seconds.