Towards automatic and robust adjustment of human behavioral parameters in a pedestrian stream model to measured data

People die or get injured at mass events when the crowd gets out of control. Urbanization and the increasing popularity of mass events, from soccer games to religious celebrations, enforce this trend. Thus, there is a strong need to better control crowd behavior. Here, simulation of pedestrian streams can be very helpful: Simulations allow a user to run through a number of scenarios in a critical situation and thereby to investigate adequate measures to improve security. In order to make realistic, reliable predictions, a model must be able to reproduce the data known from experiments quantitatively. Therefore, automatic and fast calibration methods are needed that can easily adapt model parameters to different scenarios. Also, the model must be robust. Small changes or measurement errors in the crucial input parameters must not lead to disproportionally large changes in the simulation outcome and thus potentially useless results. In this paper we present two methods to automatically calibrate pedestrian simulations to the socio-cultural parameters captured through measured fundamental diagrams. We then introduce a concept of robustness to compare the two methods. In particular, we propose a quantitative estimation of parameter quality and a method of parameter selection based on a criterion for robustness. We discuss the results of our test scenarios and, based on our experience, propose further steps.