Development of a Statistical Method for Predicting Human Driver Decisions

As autonomous vehicles enter the fleet, there will be a long period when these vehicles will have to interact with human drivers. One of the challenges for autonomous vehicles is that human drivers do not communicate their decisions well. However, the kinematic behavior of a human-driven vehicle may be a good predictor of driver intent within a short time frame. The authors analyzed the kinematic time-series data (e.g., speed) for a set of drivers making left turns at intersections to predict whether the driver would stop before executing the turn or not. The authors used principal components analysis (PCA) to generate independent dimensions that explain the variation in vehicle speed before a turn. These dimensions remained relatively consistent throughout the maneuver, allowing the authors to compute independent scores on these dimensions for different time windows throughout the approach to the intersection. The authors then linked these PCA scores to whether a driver would stop before executing a left turn using the Bayesian additive regression trees (BART). The model achieved an area under the receiver operating characteristic curve (AUC) of more than 0.90 by -25m away from the center of an intersection.