A framework for achieving realism in agent-based pedestrian crowd simulations

Pedestrian crowd simulations are used to predict the behavior of human crowds. Decision makers, however, often feel crowd simulations look mechanized and do not accurately reflect the motion of real crowds. Thus, current research focuses less on computational efficiency and more on improving simulation realism. In this conceptual work, we analyze recent, major contributions in the computer science field to identify current endeavors in crowd simulation research that lead to increased realism. We provide a framework that can be used to identify components in agent-based crowd simulations that contribute towards realism. External and internal factors influence the realism of any crowd simulation. We show that crowd simulations typically address environmental, situational and physiological factors. Agents however are rarely implemented to also consider psychological and cultural factors. As a result, the realism and therefore model accuracy and trustworthiness of crowd simulations is undermined.