Abstract Manager

The fast development of Autonomous Vehicle (AV) technologies has raised numerous questions regarding how safely these vehicles will interact with pedestrians. In most cases, AVs can be easily recognized as they are equipped with several sensors (e.g., multiple cameras, Lidar, radar) to collect data about the surrounding environment, identify dangerous situations, and make decisions accordingly. This variety of sensors makes pedestrians aware that they are interacting with a different type of vehicle, which influences their decisions when in close contact. For example, several previous studies have shown that pedestrians have different perceptions of safety while interacting with AVs when compared to human-driven vehicles (HDVs). These perceptions might be highly associated with how comfortable pedestrians feel to initiate crossings in an urban environment.

Furthermore, AVs have been deployed in different locations to enable training in a wide variety of traffic conditions. This has been done to ensure that AVs know how to behave under distinct interaction scenarios (e.g., jaywalking, signalized intersection, lane changing behavior) and in multiple locations worldwide (e.g., North America, Asia). Testing AVs in different cities makes it possible to evaluate how safely road users from certain locations interact with vehicles as site-specific behavioral conditions have a great impact on traffic safety. For example, what is considered to be a near miss in one location might not be perceived as such in another environment. This is especially important as some environments might be more prepared to incorporate AVs because of the population’s acceptance of these technologies.

This work proposes a random intercept parametric survival model to investigate pedestrian waiting times when interacting with AVs and HDVs. The model is able to incorporate the effects of different covariates that represent road user behavior. Moreover, random intercepts are used to capture the behavior differences from different environments. The study uses data from a large-scale AV dataset collected in North America and Asia. The vehicles contained sensors that enabled collecting data from the vehicle itself and from the surrounding environment. Results indicate the variables that influence pedestrians’ endurance probabilities while waiting to cross, which is useful for designing infrastructure that better accommodates pedestrian traffic and their interactions with AVs.