Location

Manchester Village, Vermont

Date

29-6-2017

Session

Session 6 — Hybrid Presentations

Abstract

Drivers’ expectations influence their responses to events in complex ways. In particular, covert and sustained hazards, like crosswinds, might require complex vehicle control adaptations. We investigated differences between drivers’ lateral responses in unexpected and expected (repeated) crosswind events using probabilistic topic modeling. First, each driver’s event-based steering wheel movements (angle and rate, 5 Hz) were transformed into symbolic words. Then, probabilistic topic modeling was used to discover patterns in the steering wheel movement data across the event conditions. Results indicate that drivers may make fewer abrupt steering wheel movements when they encounter unexpected crosswinds. On the contrary, drivers are more likely to make continuous faster steering corrections to compensate crosswinds when they are expected. The topic models also classify unexpected and expected crosswind events better than traditional models that use single aggregated values across events (maximum steering wheel angle and rate). These preliminary insights show an advantage for granular, time-series based analysis of driving data, and suggest a viable machinelearning based technique to conduct such investigations.

Rights

Copyright © 2017 the author(s)

DC Citation

Proceedings of the Ninth International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design, June 26-29, 2017, Manchester Village, Vermont. Iowa City, IA: Public Policy Center, University of Iowa, 2017: 375-381.

Share

COinS
 
Jun 29th, 12:00 AM

Exploring Driver Responses to Unexpected and Expected Events Using Probabilistic Topic Models

Manchester Village, Vermont

Drivers’ expectations influence their responses to events in complex ways. In particular, covert and sustained hazards, like crosswinds, might require complex vehicle control adaptations. We investigated differences between drivers’ lateral responses in unexpected and expected (repeated) crosswind events using probabilistic topic modeling. First, each driver’s event-based steering wheel movements (angle and rate, 5 Hz) were transformed into symbolic words. Then, probabilistic topic modeling was used to discover patterns in the steering wheel movement data across the event conditions. Results indicate that drivers may make fewer abrupt steering wheel movements when they encounter unexpected crosswinds. On the contrary, drivers are more likely to make continuous faster steering corrections to compensate crosswinds when they are expected. The topic models also classify unexpected and expected crosswind events better than traditional models that use single aggregated values across events (maximum steering wheel angle and rate). These preliminary insights show an advantage for granular, time-series based analysis of driving data, and suggest a viable machinelearning based technique to conduct such investigations.