Conversational Voice Agents are Preferred and Lead to Better Driving Performance in Conditionally Automated Vehicles

Abstract

In-vehicle intelligent agents (IVIAs) can provide versatile information on vehicle status and road events and further promote user perceptions such as trust. However, IVIAs need to be constructed carefully to reduce distraction and prevent unintended consequences like overreliance, especially when driver intervention is still required in conditional automation. To investigate the effects of speech style (informative vs. conversational) and embodiment (voice-only vs. robot) of IVIAs on driver perception and performance in conditionally automated vehicles, we recruited 24 young drivers to experience four driving scenarios in a simulator. Results indicated that although robot agents received higher system response accuracy and trust scores, they were not preferred due to great visual distraction. Conversational agents were generally favored and led to better takeover quality in terms of lower speed and smaller standard deviation of lane position. Our findings provide a valuable perspective on balancing user preference and subsequent user performance when designing IVIAs.