Event-based Evacuation in Outdoor Environment

Evacuation in outdoor environment is of great practical interest due to its significant impact on saving human lives under emergency conditions. Comparing with the large population to evacuate, the number of guides is much smaller. Thus the evacuation policy optimization usually suffers from partial information, partial control, and the pervasive uncertainty in the evacuation process, and is nontrivial. We consider this important problem in this paper and make the following major contributions. First, we model the evacuation policy optimization problem as an event-based optimization, in which actions are taken only when people pass by the guides. Second, a simulation-based policy improvement method is developed to improve from given event-based evacuation policies. Third, the performance of this policy improvement method is demonstrated through numerical results. We hope this work brings more insight to large-scale outdoor evacuation.