Automotive UX · Human–AI Interaction · Voice User Interfaces
Designing Transparent In-Vehicle Intelligent Agents
2021–2022 | Virginia Tech | Published Research
Overview
How should an automated vehicle explain its decisions to the driver?
We studied whether drivers respond differently when an in-vehicle agent explains its actions upfront or waits for a request. We also tested how the accuracy of those explanations shapes trust, workload, situation awareness, and takeover performance.
Background & Research Questions
When automation is in control, drivers may have little insight into why a vehicle slows down or changes lanes. Voice explanations can make those actions easier to understand. Their timing and accuracy may shape the driver’s experience.
- How does proactive versus on-demand information delivery affect driver experience and performance?
- How does information accuracy influence driver trust, workload, and situation awareness?
- Does the effect of an information delivery strategy depend on reliability?
Research Approach
In a driving simulator, 25 participants experienced both voice strategies. One group received entirely accurate information, while the other received accurate information 60% of the time.
We collected driving performance measures, situation awareness assessments, standardized questionnaires, and user preferences.
Proactive · Push
AgentI’m slowing down. There’s a stopped car ahead.
On-Demand · Pull
AgentI’m slowing down.
DriverWhy are we slowing down?
AgentThere’s a stopped car ahead.
Key Insights
21 of 25 participants chose Push, and 4 chose Pull.
- Participants rated Push as more intelligent and competent. They rated Pull as warmer.
- Participants reported higher physical demand with Pull. The extra step of requesting information may contribute to this difference.
- Information delivery and reliability jointly affected peak lateral acceleration during takeover.
| Mean score | High reliability | Low reliability |
|---|---|---|
| Awareness of surroundings (SAGAT Level 1) | 81% | 64% |
| Reported effort (NASA-TLX) | 26.4 | 41.0 |
Both differences were statistically significant. Trust ratings showed no statistically significant difference by reliability.
UX Design Implications
- Reduce interaction effort. Offer timely context when drivers have limited attention to ask for it.
- Design for reliability. Account for the quality of available information when deciding what the agent should explain.
- Preserve user control. Make additional detail easy to request when drivers have time and need more context.
My Contributions
UX & Human Factors Research
- Experimental Design
- Data Collection
- Analyzed driving performance, workload, situation awareness, and subjective ratings.
- Translated the results into design implications and authored the publication.
Publication
Zang, J., & Jeon, M. (2022). The Effects of Transparency and Reliability of In-Vehicle Intelligent Agents on Driver Perception, Takeover Performance, Workload and Situation Awareness in Conditionally Automated Vehicles. Multimodal Technologies and Interaction, 6(9), 82.