Are Two Heads Better than One? Exploring Two-Party Conversations for Car Navigation Voice Guidance
Late-Breaking Works, ACM CHI Conference on Human Factors in Computing Systems Proceedings (CHI), 2020 Acceptance Rate: 41.8%
In this pilot Wizard-of-Oz study, we explored the use of two-party conversations that is played in the middle of driving a recommended route (leftmost image). The conversations involve two voice agents with two turns each. In this example, the Optimal voice agent continues to suggest a faster route while the Familiar voice agent presents an alternative that uses previously taken routes. The participants in the pilot experiment were recommended three routes: route F (familiar), route O (optimal) and route E (explorer). They drove in simulation environment with the researcher manually playing the voice navigation guidance.
Abstract
Voice guidance for car navigation typically considers drivers as docile actors. Recent works highlight limitations to this assumption which make drivers rely less on given directions. To explore how drivers can make better navigation decisions, we conducted a pilot Wizard-of-Oz study that gives turn suggestions in conversations between two voice agents. We asked 30 participants to drive in a simulation environment using voice guidance that gives three types of suggestions: familiar, optimal, and new routes. We examined their route choices, perceived workload and utterances while driving. We found that while most drivers followed directions appropriate for the given scenarios, they were more likely to make inappropriate choices after hearing alternatives in conversations. On the other hand, two-party conversations allowed drivers to better reflect on their choices after trips. We conclude by discussing preliminary design implications for car navigation voice guidance specifically and recommender systems in general.