A simulation of evacuation dynamics of pedestrians in case of emergency by a two-dimensional cellular automaton method

In this paper, using the multi–agent cellular automaton method, we construct a simulator for evacuation dynamics of pedestrians in case of emergency, and show some natural and acceptable qualitative results. Each pedestrian decides autonomously his next target cell in his Moore neighborhood, considering environmental information in his cognitive region. Such information contains the directions of various flows in the region, the position of obstacles, indicators to the exits, the shortest paths to the exits and diffusing smoke, which are used to evaluate each cell in his Moore neighborhood. Each of these information elements has a direction and is expressed to be a vector, so called an information vector. In our simulation model two types of agents are considered, one is a leader agent who knows the shortest paths to the exits and leads non–leader agents to the exits, and other is a non–leader agent who moves depending on the flow of other agents in the cognitive region and his/her own driving force making him/her to keep the present direction. The driving force is also formulated to be an information vector. When evaluating a possibly move–to cell, we use the differences between the information vectors and the vector from the pedestrian’s present position to the move–to cell, because we think that we do not necessarily move to the object showing the direction but we move along the direction. Taking account of the simple and unified approach using the difference of two vectors, we can emerge typical, easily acceptable and realistic patterns of evacuation flows and our simulator is considered to be a starting point to develop more sophisticated simulator in the future, which may be an important tool for understanding and decision making approach to complex phenomena emerged from many autonomously and interactively moving agents, for which mathematical methods may be hardly applied.