Visual coupling to multiple neighbors in a crowd influences walking speed and direction

Abstract

Pedestrians in a crowd are visually coupled to nearby neighbors, yielding a common speed and direction of travel (heading). Previously, we derived a visual control law for walking together with one neighbor (Rio & Warren, VSS 2011; Page & Warren, VSS 2013). Here, we extend this framework to investigate how multiple neighbors combine to influence a pedestrian. We collected data from a participant (N=10) walking with 3 confederates in a diamond configuration across a 12x14m room, while recording head position with an ultrasonic tracking system. On some trials, confederates received covert instructions to change their direction (turn left, turn right) or speed (speed up, slow down), and the effect on participant's final lateral position or speed, respectively, was measured. The number of manipulated confederates (1, 2, or 3), their positions (left, center, right), and their initial distance from the participant (1 or 2 m) were varied. For both direction and speed manipulations, there was a main effect of the number of confederates manipulated (direction: F(8,64)=20.17, p<0.001; speed: F(8,64)=2.29, p<0.05), with more confederates eliciting larger participant responses. For the speed manipulation, there was also a main effect of neighbor distance (F(8,64)=2.29, p<0.05), although this was not the case for direction (F(1,8)=0.033, p>0.05). Multiple linear regression on participants' final position reveals that the confederate directly ahead [center] is weighted more heavily than confederates to the left or right (Beta=0.32, 0.17, 0.26, respectively; R2=0.47); the same holds for multiple regression on participant speed (Beta=0.40, 0.26, 0.20, respectively; R2 = 0.46). The results indicate that the influences of multiple neighbors are linearly combined, weighted by their position and, in the case of speed, their distance. We extend our previous control law to model this multiple visual coupling, which may be applied in simulations of human crowd behavior. Meeting abstract presented at VSS 2014