Characteristics of Voice-Based Interfaces for In-Vehicle Systems and Their Effects on Driving Performance

Abstract

A test track experiment was conducted to measure the disruptive effects on driving of complex secondary tasks and to evaluate voice interface design characteristics that may help reduce distraction. Thirty-six drivers in three age groups (18-25, 30-45, 50-60) drove an instrumented vehicle while performing a combination of car following, peripheral target detection, and secondary tasks on a closed test track. Secondary tasks included a simulated navigation task, in which drivers were given specific questions to answer using information obtained from a traveler information system, and a simulated phone task, in which drivers performed the Baddeley short-term memory task. Four variations of the navigation task were used, including combinations of two factors: (1) mode of information acquisition (auditory vs. auditory + visual map), and (2) system reliability (no recognition errors vs. 20% errors). All secondary tasks were implemented with a hands-free voice interface. The results indicated significant degradation of driving performance associated with all secondary tasks. Specifically, drivers exhibited higher levels of steering entropy, which measures the number and magnitude of steering corrections relative to a baseline drive, and slower car-following responses. They also had higher levels of target detection errors, slower target detection response times, and higher levels of subjective workload while performing secondary tasks relative to a baseline condition with no secondary task. Secondary tasks that required drivers to look at a map while interacting with the traveler information system were consistently more disruptive than similar tasks that did not require the map for all performance measure categories. Decreased system reliability did not disrupt driving performance. The study results provide the following implications for voice interface design: (1) Unnecessary or redundant visual displays should be avoided; (2) Drivers appear to be able to tolerate a voice interface with less than perfect reliability. Additional research is recommended to explore the effects of higher system error rates and task pacing (self-paced vs. externally-paced) on the interference induced by secondary task performance.