Predictive Models of Driver Deceleration and Acceleration Responses to Lead Vehicle Cutting In and Out

Jason Hu, Brian Lin, Jim Vega, Nathan Tsiang

Abstract:

A common maneuver drivers perform and experience on the road is changing lanes. Autonomous vehicles are required to
engage a lane change safely and to react to the other road users’ lane changes. To develop autonomous vehicles that change
lanes or respond to the lead vehicle’s lane changes in a safe and human-like way, one should investigate the factors that affect
human driver responses. By reviewing the literature to identify potential factors, this study extracted these factors from a
naturalistic driving data set and associated them with driver deceleration and acceleration responses to the lead vehicle’s cutin and cut-out to develop predictive models for the impact of the events on traffic flow. After the events were verified as
accurate, the variables associated with the events, including range, range rate, speed, lateral position in the lane, and average
acceleration were analyzed using logistic regression, support vector machines (SVM), and two forms of decision trees. In
total, 799 cut-in events and 684 cut-out events with the necessary variables were applied for analysis. The significant variables
influencing driver behavior were found, and using these variables, the predictive models achieved around 80% accuracy for
cut-ins, and 73% accuracy for cut-outs on test data. These results will assist in the future design of autonomous vehicle control to minimize detrimental effects on traffic when changing lanes and safe longitudinal control when responding to a lead
vehicle’s lane changes, allowing for safe integration with human drivers, and better design of driver assistance systems.