The effect of social groups on the dynamics of bi-directional pedestrian flow: a numerical study

TL;DR

This study uses novel computational models to analyze how social groups influence pedestrian lane formation, velocity, and collision rates in bi-directional flow, highlighting the importance of realistic group behavior in crowd dynamics.

Contribution

It introduces a new model accounting for body shape and rotation during collision avoidance, combined with group behavior modeling, to study their effects on pedestrian flow.

Findings

Groups significantly affect velocity and lane stability.

Presence of groups increases collision rates.

Group behavior modeling is crucial for realistic crowd simulations.

Abstract

We investigate the effect of groups on a bi-directional flow, by using novel computational methods. Our focus is on self-organisation phenomena, and more specifically on the time needed for the occurrence of pedestrian lanes, their stability and their effect on the velocity-density relation. Moreover, we are interested in understanding the amount of physical contact in the crowd. To this end, we use a novel model considering the asymmetrical shape of the human body and describing its rotation during collision avoidance, and combine it to a mathematical model of group behaviour. We configure several scenarios by varying the global density $\rho$ of pedestrians and the ratio $r_g$ describing the percentage of grouped pedestrians in the simulation. Our results show that the presence of groups has a significant effect on velocity and lane organisation, and a dramatic one on collision. We are well aware of the limitations of our approach, in particular concerning (i) the lack of calibration of body rotation in collision avoidance on actual data and (ii) straightforward application of a low density group model to higher density settings. We nevertheless want to stress that it is not our intention to state that our results reproduce the actual effect of groups on bi-directional flow. In particular, it seems highly unrealistic that crowds with groups collide extremely more often. Nevertheless we believe that our results show the great theoretical and practical implication of the consideration of realistic group behaviour in pedestrian models, and suggest that realistic results may hardly be achieved simply by adding together modular models.