A Framework Combining Cellular Automata and Multi-Agents in a Unified Simulation System for Crowd Control

Controlling crowds in airports, train terminals, sporting events, etc., is a complex problem. This particular problem has a great deal of interaction between the entities themselves (i.e. among the individual members of the crowd) and the crowd (or individuals) with the environment in which the crowd is placed. This complexity of this system can be described in the much researched area of artificial life. By combining Cellular Automata (CA) with agents, we can construct a system to capture and control the ebb and flow of a crowd, including the particular characteristics of the individuals in the crowd. To this end, we have developed a prototype crowd control simulation system as a test case for this kind of problem. The imbedded CA provides a framework for flow of people, much like traffic models (Nagel, K., and Rasmussen, S. 1994) and (Blue, V. J. and Adler, J.L, 2000a), while the agent reproduces the behavior of a crowd, including subgroup behaviors, interactions, stochastic decisions of single units etc. This work is an extension of the web-based model in (Bruzzone, A. and Signorile 1999). CELLULAR AUTOMATA SIMULATIONS AND TRAFFIC SIMULATIONS Cellular automata (CA) are simple spatial processing models with their origins in the early architecture of digital computers designed in the 1940 and 1950s. CA has close associations with complexity theory and has been employed in the exploration of a diverse range of urban phenomena, generally to investigate ideas about how real urban systems operate, but from a controlled experimental environment within computer software. Urban applications of CA range from traffic simulation and regional-scale urbanization to land-use dynamics, historical urbanization, and urban development (Center for Advanced Spatial Analysis website). Therefore, CA is particularly useful in simulating complex adaptive systems such as people movement. There has been a great deal of interest in studying traffic flow with Cellular Automata models. CA models are conceptually simple, thus we can use a set of simple CA rules to produce complex behavior. Using CAs we can capture the complexity of interacting traffic pattern behavior. The basic one-dimensional Cellular Automata model for highway traffic flow is described in (Nagel, K., and Rasmussen, S. 1994). The model describes a one-lane traffic road with sequence of grid points, and each grid point is a square representing one vehicle. There are many variations on the basic model (Blue, V.J. and Adler, J.L. 2000b) that consider the effects of acceleration and delay of vehicles with high speed. The actual speed of the car at each time step depends on the “lambda” value that can be adjusted accordingly. This model captures the realistic traffic situations where the car accelerates and decelerates. The rules in (Blue, V.J. and Adler, J.L. 2000b) model are very simple, but we get complex behavior out of a population of these rules. This complexity is defined by methods in statistical physics. Such models lead us away from the view of multi-agent traffic models as fundamentally linear where units are treated in isolation, thus motivating us to look into combining agents and CA. CELLULAR PEDESTRIAN TRAFFIC SIMULATIONS The one dimensional car traffic models motivates us to develop more complex models of movement. The area of pedestrian movement has been used as possible application field for the use of cellular automata (Blue, V. J. and Adler, J.L, 2000a ) and (Blue, V.J. and Adler, J.L. 2000b) These models contain cellular entities that have a forward direction of movement and the idea is to optimize the speed of the agent in a given direction, under a maximum walk speed constraint. Each agent will account for the position of other agents and their direction of forward movement. In the simplest case we could have an environment where each agent is moving in the same direction as every other agent. The next increase in complexity involves flow where two types of agents in the population move in opposing directions. To further increase, complexity, consider moves that cover all possible local moves (Blue, V.J. and Adler, J.L. 2000b). In (Dijkstra, J., A.J. Jessurun, and H.J.P. Timmermans. 2001), using agents as an extension of CA was initially discussed, However, these agents are just extension of the movement rules in the CA, and do not have personable attributes we consider important in crowd behavior. Most of these models are primarily based on CAs to understand emerging behavior of pedestrian’s movement. We are interested in more than just movement models, but also pedestrian behavior models. For example, in a museum setting, each agent/pedestrian makes decisions on moving both on the CA rules defined in (Bruzzone, A. and Signorile 1999) as well as other internal rules. These internal rules could include the, again for the museum example, the desire to linger at a particular exhibit. This individual pedestrian behavior affects the total crowd behavior in interesting ways. As modeling of spatial systems improve and develop, systems can be modeled at finer and finer granularity, or scale. This means that activity can be represented in the model at various levels, (for example, at the individual entity in the systems). Understanding complex systems naturally mean injecting individual behavior into the gross systems. Therefore, individual mobility, and state are inevitably woven into the fabric of the complex system. Therefore, we look at developing models of complex systems that combine cellular automat at the aggregate spatial level and add entity motion to the cells as agents. In this way, we can model gross movement rules (the CS grid rules) and particular individual/group rules (the agents). THE PROBLEM OF MODELING CROWD CONTROL Crowd control can be applied to many different areas, such as police operations (Varner D., Scott D.R. Micheletti J., Aicella G. (1998)), shopping center design, public park re-organization (Bruzzone A.G., Rivarolo D. 1997)),rail station reengineering (Bouvier E., Cohen E. 1995) and epidemic diffusion. In additions, there has been a significant advancement in studying pedestrian traffic patterns in various environments. In (Bierlaire, M., Antonini, G. and Weber,M. 2003), a system based on multiple agents (MAS) was used. The desire of the author was to create a highly flexible system composed of actors that can be modeled individually. In (Helbing, D., and Molnar, Peter, 1998) pedestrian flow is discussed with some features observered. In (Still, K, 2000), the author presents several phenomena about crowd behavior, the most significant being: