Predicting Crowd Dynamics Using Local Structure

Unstable and active disordered materials exhibit interesting collective properties and nontrivial dynamics. While the behavior of amorphous solids under shear is relatively well-understood, the instabilities in active systems remain difficult to characterize and predict. In the context of dense crowd dynamics, existing work has analyzed position fluctuations in a self-propelled particle (SPP) model to identify Goldstone modes and soft spots in models for human crowds. This analysis requires time-resolved trajectory information in order to form predictions for collective behavior, which can be cumbersome. To address this issue, we have developed a novel method to generate static packings in an artificial potential that reproduce the packing structures in a class of point-of-interest active SPP crowd simulations. These static packings then allow us to precisely identify local structural defects that govern dynamical group behavior, so that we can predict the locations of material-like failures in dense, active SPP models. Unlike previous methods, these predictions can be derived from a single snapshot and could be relevant to preventing dangerous emergent phenomena in real crowd systems.