Abstract
Can human crowd behavior be explained as an emergent property of local rules, as in flocking (Reynolds 1987) and fish schooling (Huth & Wissel 1992)? Here we derive one such possible ‘rule:’ a dynamical model of following another pedestrian. We collected position data from pairs of pedestrians walking in a 12m x 12m room, using an inertial/ultrasonic tracking system (IS-900, 60 Hz). The ‘leader’ (a confederate) walked in a straight path. After 3 steps at a constant speed, the leader would (a) speed up, (b) slow down, or (c) remain at the same speed for a variable number of steps (3, 4, or 5), and then return to his original speed. The ‘follower’ (a subject) was instructed to maintain a constant distance from the leader (1m or 3m). We evaluate several candidate following models, in which the follower's acceleration is controlled by (a) nulling change in distance, (b) nulling change in relative speed, or (c) more complex functions of these variables, drawing inspiration from studies of vehicle-following in traffic (Brackstone 1999). For each model, we cross-correlate the predicted acceleration of the follower with the observed acceleration in the tracking data. Future work will investigate the visual information that serves as input to the model. Once a control law for following is characterized, it can be integrated with other components for steering toward goals, avoiding obstacles, and intercepting moving targets (Fajen & Warren 2003, 2007). This will allow us to empirically determine whether human crowd behavior does indeed emerge from such local interactions.