An Empirically-Grounded Emergent Approach to Modeling Pedestrian Behavior

Realistic models of locomotion, accounting for both individual pedestrian behavior and crowd dynamics, are crucial for crowd simulation. Most existing pedestrian models have been based on ad-hoc rules of interaction and parameters, or on theoretical frameworks like physics-inspired approaches that are not cognitively grounded. Based on the cognitively-plausible behavioral dynamics approach, we argue here for a bottom-up approach, in which the local control laws for locomotor behavior are derived experimentally and the global crowd behavior is emergent. The behavioral dynamics approach describes human behavior in terms of stable, yet flexible behavioral patterns. It enabled us to build an empirically-grounded model of human locomotion that accounts for elementary locomotor behaviors. Based on our existing components, we then elaborate the model with two new components for wall avoidance and speed control for collision avoidance. We show how the model behaves with many stationary obstacles and interacting agents, and how it can be used in agent-based simulations. Five scenarios show how complex individual behavioral patterns and crowd dynamics patterns can emerge from the combination of our simple behavioral strategies. We argue that our model is parsimonious and simple, yet accounts realistically for individual locomotor behaviors while yielding plausible crowd dynamics, like lane formation. Our model and the behavioral dynamics approach thus provide a relevant framework for crowd simulation.