Whose Drive Is It Anyway? Using Multiple Sequential Drives to Establish Patterns of Learned Trust, Error Cost, and Non-Active Trust Repair While Considering Daytime and Nighttime Differences as a Proxy for Difficulty

Semi-autonomous driving is a complex task domain with a broad range of problems to consider. The human operator’s role in semi-autonomous driving is crucial because safety and performance depends on how the operator interacts with the system. Drive difficulty has not been extensively studied in automated driving systems and thus is not well understood. Additionally, few studies have studied trust development, decline, or repair over multiple drives for automated driving systems. The goal of this study was to test the effect of perceived driving difficulty on human trust in the automation and how trust is dynamically learned, reduced due to automation errors, and repaired over a seven-drive series. The experiment used 2 task difficulty conditions (easy vs. difficult) x 3 error type conditions (no error, takeover request or TOR, failure) x 7 drives mixed design. Lighting condition was used as a proxy for driving difficulty because decreased visibility for potential hazards could make monitoring the road difficult. During the experiment, 122 undergraduate participants drove an automated vehicle seven times in either a daytime (i.e., “easy”) or nighttime (i.e., “difficult”) condition. Participants experienced a critical hazard event in the fourth drive, in which the automation perfectly avoided the hazard (“no error” condition), issued a takeover request (“TOR” condition), or failed to notice and respond to the hazard (“failure” condition). Participants completed trust ratings after each drive to establish trust development. Results showed that trust improved through the first three drives, demonstrating proper trust calibration. The TOR and automation failure conditions saw significant decreases in trust after the critical hazard in drive four, whereas trust was unaffected for the no error condition. Trust naturally repaired in the TOR and failure conditions after the critical event but did not recover to previous levels before the critical event. There was no evidence of perceived difficulty differences between the daytime and nighttime conditions. Thus, a consistent lack of trust differences was found between lighting conditions. This study demonstrated how trust develops and responds to errors in automated driving systems, informing future research for trust repair interventions and design of automated driving systems