Jeffrey KATAN, Université de Montréal, Canada
Liliana PEREZ, Université de Montréal, Canada
Raja SENGUPTA, McGill University, Canada
The evacuation of individuals during a building emergency remains an undertaking which can prove to be very difficult and particularly complex. Complex phenomena of crowd movement can produce critically dangerous situations. Foremost among them is the bottleneck, where too many individuals inside the building with the same destination crowd together at a narrow passage such as a doorway or staircase and reduce the overall flow rate through it. Such situations have long been observed in reality, simulated and studied extensively in several fields including computer graphics, robotics, traffic engineering, and social sciences. This study proposes an agent-based modelling approach to simulate crowd movement. Agent-based models of crowd simulation require the implementation of algorithms to mimic individuals’ movement relative to immediate concerns such as obstacles, as well as medium-term concerns like pathfinding. Navigation fields provide an efficient pathfinding solution for hundreds or even thousands of agents, where goal-directed agents with an individual knowledge of the environment follow common pathfinding information to reach their goal. In an effort to improve egress time and to reduce the impact of bottlenecks, we have developed a semi-adaptive navigation field algorithm that uses movement history to produce dynamic exit strategies of evacuation in case of an emergency.
Mots clés : Navigation field|building evacuation|agent-based modelling|crowd movement|complex systems