Demonstration of virtual reality simulation as a tool for understanding and evaluating pedestrian safety and perception at midblock crossings

Recent advancements of virtual reality (VR) technology (such as the HTC Vive or Oculus Quest) have opened the door for lower cost and lower risk methods to study pedestrians' behavior and perception of safety, offering a higher degree of data resolution and level of realism compared to previous pedestrian simulators. This research aims to address the lack of research conducted to compare pedestrian midblock crossing behavior within VR simulators against naturalistic, real-world behavior, as well as provide a low-risk framework for conducting these analyses. In addition to comparing pedestrian crossing speeds and gap acceptance in the replica virtual environment, post-experiment questionnaires also asked participants about their perception of realism in the virtual environment. Chi-squared analysis showed no statistically significant differences in the gap acceptance distribution nor median accepted gap size (real-world = 8.05sec, VR = 8.20sec, p = 0.358), indicating similar gap size acceptance between environments. Logistic regression analysis showed similarity with respect to the rate at which participants' decision to cross or not changes, indicating consistent and naturalistic decision-making processes between the environments. Furthermore, experiment results showed no significant difference in the means (real-world = 3.32mph, VR = 3.52mph, p = 0.138) and medians of crossing speeds (real-world = 3.28mph, VR = 3.43mph, p = 0.358) in either environment and no difference in crossing speed variance. Additionally, survey results showed that subjects felt their sense of movement and risk inside the simulator were realistic and reported high levels of immersion. Through the use of a low-risk and time-efficient study framework, this paper demonstrates the efficacy of VR simulation for use in replicating naturalistic pedestrian crossing behavior in immersive virtual environments modeled after real-world locations.