How do drivers negotiate intersections with pedestrians? Fractional factorial design in an open-source driving simulator
Paper in proceedings, 2017
Forward collision warning (FCW) and autonomous emergency braking (AEB) systems are increasingly available and promise to prevent or mitigate collisions by alerting the driver or autonomously braking the vehicle. Threat-assessment and decision-making algorithms for FCW and AEB aim to find the best compromise for safety by intervening at the “right” time: neither too early, potentially upsetting the driver, nor too late, possibly missing opportunities to avoid the collision.
Today, the extent to which intervention times for FCW and AEB should depend on factors such as pedestrian speed and lane width is unknown. To guide the design of FCW and AEB intervention time, we employed a fractional factorial design, and determined how seven factors (crossing side, car speed, pedestrian speed, crossing angle, pedestrian size, zebra presence, and lane width) affect the driver’s response process and comfort zone when negotiating an intersection with a pedestrian. Ninety-four volunteers drove through an intersection in a fixed-base driving simulator, which was based on open-source software (OpenDS). Several parameters, including pedestrian time-to-arrival and driver response time, were calculated to describe the driver response process and define driver comfort boundaries.
Linear mixed-effect models showed that driver responses depended mainly on pedestrian time-to-arrival and visibility, whereas factors such as pedestrian size, zebra presence, and lane width did not significantly influence the driver response process. Some drivers changed their negotiation strategy to minimize driving effort over the course of the experiment. Experienced drivers changed more than less experienced drivers; nevertheless, all drivers behaved similarly, independent of driving experience. The flexible and customizable driving environment provided by OpenDS proved to be a viable solution for behavioural experiments in driving simulators.
Results from this study suggest that visibility and pedestrian time-to-arrival are the most important factors for defining the earliest acceptable FCW and AEB activations. Fractional factorial design effectively compared the influence of several factors on driver behaviour within a single experiment; however, this design did not allow in-depth data analysis. In the future, OpenDS may became a standard platform, enabling crowdsourcing and favouring repeatability across studies in traffic safety. Finally, this study may guide future design and evaluation of FCW and AEB by highlighting which factors deserve further investigation and which ones do not.