Automated Vehicle Behavior Design for Pedestrian Interactions at Unsignalized Crosswalks

Abstract

A significant challenge for real-world deployment of automated vehicles (AVs) is their safe interaction with other road users, particularly pedestrians, whose behavior can change almost instantaneously. To ensure safety, AVs are expected to drive cautiously around pedestrians. However, conservative AV behavior can encourage pedestrians to be careless. Additionally, riders inside the AV expect a comfortable ride and to reach their destinations without excessive delay (AV performance). These objectives of safety, performance, and comfort are opposing in nature and thus have to be balanced in a way that is accepted by both the riders and the pedestrians. For example, in the case of pedestrian crossing at an unsignalized crosswalk, the AV might want to prioritize comfort normally and gradually slow down to stop, but at a school zone might want to prioritize safety and have a stronger deceleration. Existing approach of fixing the weights of the objectives can be constrictive as in reality, the objectives can vary depending on the situation. To overcome this limitation, we incorporated a hybrid systems pedestrian model that is interactive and depends on the vehicle behavior (distance and speed). Our findings suggest the possibility of characterizing driving behaviors by varying the weights of the driving objectives. We believe that by understanding the objective performance of the different driving behaviors with an interactive pedestrian model, the AV-decision making can be expressed by varying the weights of the objective function. The primary contribution of this research is the objective evaluation of a spectrum of driving behaviors represented by the different weight combinations of the three AV driving objectives – safety, performance, and comfort.