SLIDE 1 Characteristics of Voice-Based Interfaces for In-Vehicle Systems and Their Effects on Driving Performance Thomas A. Ranney, TRC Inc. Elizabeth N. Mazzae, NHTSA G. H. Scott Baldwin, TRC Inc. SLIDE 2: Background Categories of Distraction - Peripheral distraction Visual interference Manual interference - Attentional distraction Cognitive interference SLIDE 3: Background Manufacturers use voice interfaces to reduce distraction. Characteristics include: * Text-to-speech and voice recognition capabilities * Hierarchical menu structures Our previous research indicates that voice interfaces: * Reduce peripheral (visual and manual) distraction * Have little effect on cognitive (attentional) distraction SLIDE 4: Program ObjectiveS - Develop protocol and metrics for assessing distraction potential of tasks performed using voice interfaces - Use protocol to assess selected attributes of secondary tasks and voice interfaces SLIDE 5: Traveler Information Systems - Accessible via telephone by dialing ‘511’ - Implemented in parts of the United States since 2000, 511 service will be: * Available in at least 25 states by 2005 * Operational throughout the United States by 2010 SLIDE 6: Traveler Information Systems - Voice-activated navigation of hierarchical menu structures - May be used by commuters and unfamiliar travelers - Provide current information about: * Traffic conditions - accident and construction delays * Road conditions – slippery, snow-covered etc. * Public transit SLIDE 7: Research Objectives - To determine whether selected secondary tasks degrade driving performance * Simulated phone conversation * 511 information acquisition tasks - To evaluate the effects of specific task interface attributes: * Required map use * Voice interface reliability SLIDE 8: Research Approach We developed: * Simulated 511 system using a “Wizard of Oz” approach Human replaces voice recognition component * Navigation questions related to a hypothetical system of interstate roads SLIDE 9: 511 Task: Hypothetical Roadway System [DIAGRAM] SLIDE 10: Example 511 Tasks * Which route has better road conditions, I-34 or I-51? (No Map) * Which segment has more incident-related time delay, segment 3 or segment 7 (Map required) SLIDE 11: 511 Task: Hypothetical Roadway System [DIAGRAM] SLIDE 12: 511 Task: Hierarchal Menu Structure [DIAGRAM] SLIDE 13: Research Approach cont. - Primary (driving) task * Car following * Peripheral detection task (PDT) - Secondary tasks (hands-free phone and voice interface) * Simulated phone conversation task * Simulated navigation task (4 conditions) - Information acquisition mode (auditory vs. auditory + visual map) - System reliability (no voice recognition errors vs. 20% errors) SLIDE 14: Car-Following Task [SINE WAVE DIAGRAM] SLIDE 15: Peripheral Detection Task [STILL PHOTO - THRU-WINDSHIELD VIEW OF ROADWAY] SLIDE 16: Experimental Design - Independent variables: * Driver age group * Secondary task * Lead vehicle speed signal - Dependent measures: * Car following (coherence and delay) * Vehicle control (steering entropy) * Peripheral target detection (% correct, RT) * Subjective workload rating SLIDE 17: Method - Thirty-six drivers (18-25, 30-45, 50-60) - Drivers drove a 1996 Honda Accord and followed a lead vehicle on TRC’s 7.5-mile test track * Data collected on both 2 mile straight segments * Speeds varied between 45 and 60 mph SLIDE 18: Categories of Driving Performance Measures [TABLE] cATEGORY Vehicle Control INTERFERENCE peripheral PRIMARY MEASURE steering entropy Car Following: INTERFERENCE attentional PRIMARY MEASURE coherence, delay Target Detection: INTERFERENCE peripheral PRIMARY MEASURE % detected, Response Time Subjective Rating: INTERFERENCE overall PRIMARY MEASURE RSME Workload SLIDE 19: Overview of Results - Baseline comparisons - Interface design hypotheses Map vs. No Map Error vs. No Error - Differences between Phone and 511 tasks SLIDE 20: Baseline Comparisons [table[ SLIDE 21: Summary of Results: Baseline Comparisons - All Secondary tasks degraded driving performance. 511 tasks degraded all categories of driving performance. - Phone conversation task was least disruptive Phone conversation did not affect car following performance. SLIDE 22: Interface Design Comparisons [table] SLIDE 23: Summary of Results: Interface Design Issues - Tasks requiring map were more disruptive than similar tasks not requiring map. * Impairment observed on 5 of 6 key performance measures - Increased voice recognition errors did not affect driving performance. * No differences observed SLIDE 24: Phone vs. 511 Task [table] SLIDE 25: Summary of Results: Phone vs. 511 Tasks - All categories of 511 tasks revealed greater degradation than phone conversation task on at least one measure * Map Tasks were most different from phone task - Differences most apparent for car-following measures * Suggests 511 tasks were more cognitively demanding - No differences in workload ratings SLIDE 26: Other Results - Effects of secondary tasks were consistent across all three age groups. SLIDE 27: Conclusions * Test track protocol is sensitive to distraction effects of secondary tasks performed with voice interface and hands-free phone * 511 tasks interfered with all aspects of driving performance * Phone conversation task impaired target detection and vehicle control but not car- following performance SLIDE 28: Safety Implications Complex secondary tasks, including those requiring active navigation of hierarchical menu structures, are likely to disrupt driving more than simple phone conversations. SLIDE 29: Design Implications * Avoid unnecessary or redundant visual displays * Drivers can tolerate systems with some voice recognition errors * Minimize complexity of hierarchical menu structures SLIDE 30: Thanks for your attention!