Avoidance behaviors of pedestrians in a virtual-reality-based experiment

Pedestrians have to take actions to avoid collisions when passing by other pedestrians or obstacles. In this work, we focus on the avoidance behavior when a pedestrian passes by a non-perception intruder in a virtual environment. The process is divided into three stages and we focus on the transitions among these stages by analyzing the starting avoidance point, the maximum lateral deviation point, and the preference of avoiding side. It is found that most of pedestrians start avoiding within 3∼5 m away from the potential collision point (CP). Furtherly, the fitting formulas of the reverse cumulative probability for the avoidance are given. The average of the maximum lateral offset distance (MLD) for the detouring decreases with the increasing intrusion angles. Under 30°, 60° and 90° intrusion angles, more pedestrians detour in front of the Non-player character (NPC) since it requires shorter MLD than from the back. Under the 180° intrusion angle, the number of the pedestrians with both kinds of detouring is nearly the same. The findings are compared with that from the experiment in a real environment. The mechanism of pedestrian avoidance behaviors obtained in this work is helpful to adjust pedestrian avoidance prediction models and design the humanoid robots.