Abstract:
Crowd simulation is listed under many practical applications in computer industry; such as safety modelling, pre-planning building architectures, urban modeling and entertainment software. Most of these current simulations are created by extending deterministic models such as particle systems, cellular automata and fluid motion. However, extending a crowd simulation model to support an emergency situation still remains as a key challenge. The reason lies behind the difficulty of simulating the unpredictable nature of crowd behaviour during panic; since a computer algorithm approaches a solution by parameterizing predictability within a problem.
It is evident from literature that multi-agent technology has proven success in modeling complex systems interacting among many entities that are distributed and operated under lot of uncertainty. Therefore it is postulated that multi-agent systems technology can model the uncertainty of a scenario such as crowd behaviour raised during a panic situation.
The proposed solution provides an agent-based framework to simulate crowd behaviour during an emergency situation. By considering evacuation of a crowd from a building during fire as a sample input scenario; each individual is modeled as an agent associated with a local ontology. The local ontology of an agent is a collection of simple rules, representing the knowledge known to each individual; prior to occuring the emergency. The knowledge embedded within these rules is exchanged (i.e. shared) among individuals as they communicate with each other during the emergency. As a result unpredictable global behaviour patterns emerge within the crowd; which is similar to observations of a real crowd facing a real emergency situation. Output of the system is a visualization of crowd behaviour during the emergency along with statistics recorded per each simulation session, indicating evacuation related information for each individual. The solution is evaluated by implementing a prototype and comparing the statistics recorded from the prototype with statistics recorded from real world crowd behaviour during panic. Hence it is concluded that a multi-agent based knowledge sharing approach is well suited for modeling a crowd in panic.
