Crowd Modeling and Control in a RoboCup Rescue domain

Robocup Rescue is a research initiative on disaster mitigation, search and rescue problems. Rescue Simulation is a sub-domain that domain aims at increasing the understanding of disasters at a macroscopic(i.e.city-wide) level and exploring new ways for disaster management. In this graduation project, we extended the simulation with a model of human crowd behavior. Also explored is, how crowds can be influenced to direct civilians to safe areas in the city. The choice of the modeling approach was determined by the specific characteristics of disaster domains and by the need to model realistically large numbers of civilians, yet keep the computational complexity of the simulation low. The approach chosen, flow modeling, uses the characteristics common to all individuals in a group to describe the group behavior. A probabilistic framework is used to derive group behavior from individual transition probabilities on every location. These individual probabilities correspond to the subjective attractiveness/repulsion of the current location and its neighbor-location(s). They depend on factors, e.g. expected danger and people density, that we have defined based on the literature on disaster behavior. As for the second part of the project, to influence/direct a human crowd a simple strategy is proposed that blocks certain roads in order to guide civilians to a safe place (”refuge”). In sum, the simulation of human crowds shows typical behavior of humans in disaster circumstances. The crowd control study is a main soure for future work where evacuation strategies can be studied in the context of this simulation.