Testing the validity of multi participant distributed simulation for understanding and modeling road user interaction

Abstract

Understanding driver-pedestrian interactions at unsignalized locations has gained additional importance due to the advancements that have happened recently in vehicle automation. To investigate these interactions, we previously developed a novel experimental study paradigm and a set of computational models, including four game-theoretic models (i.e., two conventional and two behavioral game-theoretic models) and a logit model. The aim of this study is to validate both the lab study and the models with naturalistic data to understand to what extent the results that come out of the lab are comparable to real traffic data. In the experimental study, several pairs of one driver and one pedestrian interacted with each other under different kinematic conditions and at zebra crossings in a virtual environment. The naturalistic data collection was conducted at two marked crossings (normal versus zebra staggered crossing) using state-of-the-art sensors capturing road user type, trajectory, and speed over time. Overall, the results indicated there is a fine relative validity of the experimental study where road users showed similar patterns of non-verbal communication in both studies. Similar to the lab data, crossing type played a role in interaction outcomes and metrics such as pedestrian crossing speed and vehicle delay. Pedestrians crossed more often and walked faster at staggered zebra compared to normal zebra in real traffic. In both studies, vehicle delay was affected by kinematics and location. However, vehicle delay was longer in the lab compared to real traffic. Also, unlike the lab study, pedestrian approach speed was measured and found to be the predictor of their crossing speed and the delay of the drivers. Additionally, all the computational models performed close to each other and well as opposed to the lab study. The behavioral game-theoretic models performed slightly better than the others in terms of prediction accuracy; this replicated our previous findings about road user behavior complexity which is a pivotal point for the virtual testing of automated vehicles.