Affiliation:
1. Purdue University, West Lafayette, IN, USA
Abstract
Automated vehicles (AVs) are becoming increasingly intelligent. At the same time, researchers are exploring ways to enable transparent communication between drivers and AVs. One line of work has focused on understanding how displaying an AV’s confidence in detecting roadway obstacles influences drivers’ behavior. In this study, we investigate what actions drivers make when presented with AV’s confidence information regarding its obstacle avoidance ability that does not always match its reliability. Twenty participants drove a semi-autonomous vehicle, while being presented with a confidence information, and needed to decide whether to take over or not. Findings suggest that alignment between the vehicle’s reliability and confidence increased the number of correct decisions. Also, drivers self-calibrated their decision strategy such that more correct decisions were made when provided with accurate information. Insights from this work can be used to inform the design of AV driver models and human-machine interfaces to promote safety within AVs.
Funder
The US National Science Foundation