Because driving automation is quickly increasing, humans’ relation to automated driving evolves accordingly. Trust in automated driving has been shown to strongly correlate with drivers’ acceptance of such systems. There is still much to be learned about which factors determine trust before any interaction with Highly Automated Driving (HAD) systems. Theoretical trust described in automation models proposes several factors to explain how trust builds before and during interactions with automation. Using 844 answers collected through an online survey, this article aims to propose a new scale specifically designed to evaluate initial trust in automated driving. Moreover, we also measure the other factors linked to drivers’ initial level of trust in HAD by operationalizing the trust components proposed by Hoff and Bashir’s theoretical trust model. To better describe what determines trust in HAD, a linear model based on the collected data is proposed. The model not only weights the factors determining trust but is also able to predict the level of trust considering these factors.

We compared automation monitoring evolution of static or adaptive automation for four different reliability levels over 90 minutes. Previous studies have demonstrated degraded human performance when monitoring automation and that it is possible to mitigate this monitoring performance drop by using adaptive automation. We used the Open Multi-Attribute Task Battery to manipulate two type of automation (static automation without manual take-over sessions and adaptive automation with planned take-over sessions) and four levels of reliability. Participants performed three simultaneous tasks, one of which was automated. Our results suggest that a perfectly reliable or a totally unreliable automation led to different strategies by the participants in terms of visual allocation policy. Under static automation, the time spent looking at the automated task in the 0% reliability level increased over the duration of the experiment; however, the opposite was observed for the 100% reliability level. Although similar, the magnitude of this pattern of results was largely diminished under adaptive automation. For static automation, the reported data also showed a direct link between trust in automation and visual scanning strategies. The more the trust increased, the less the automated task was looked at.

Previous research has suggested that information and action automation stages do not imply the same consequences for human performance in the supervision of automated systems. Still, only a few studies have simultaneously investigated these stages. When information and action automation are reliable, both can support performance. However, with unreliable aids, the literature has suggested that action automation tends to be more detrimental than information automation. This study aimed to assess the contributions of imperfect information and action automation on attentional allocation and to investigate a potential monitoring inefficiency in a multitasking environment. Participants (n = 96) completed three Multi-Attribute Task Battery (MATB) tasks. A monitoring task was automated with two types of automation (action or information) of four reliabilities each (0%; 56.25%; 87.5%; 100%). Ocular behaviors and performance were assessed. Results show that reliability of information automation influenced visual resource allocation. When information automation was the most reliable, participants spent the least amount of time sampling the monitoring task. Finally, the reliability of action automation triggered no effect on performance or cumulative dwell times. Our results suggest that in complex multitasking situations where information and action automation occurred simultaneously, participants allocated fewer visual resources to automated task with increased information automation reliability. Similarly, their performance was better only with increased information automation.

Automation regroups a variety of advanced tools meant to improve performance and decrease human workload. This article was designed to investigate how different automation solutions, engaging different human-machine cooperation modes, interact with expertise. Aircraft pilots (i.e., experts) and nonpilots (i.e., novices) were presented with a set of simplified flight piloting tasks monitored simultaneously using the Open MATB (Open Multiattribute Task Battery) in four different automation conditions (manual, assisted, cooperative, and supervisory control). Experts' performances at the Open MATB were higher than those of novices. Experts also exhibited a lower level of workload. Apart from function delegation, where no human performance is required, automation solutions benefit experts and novices similarly. Participants were then asked to choose repeatedly between the four automation conditions. Experts and novices exhibited different strategies. Novices spread their choices over the different automation conditions, whereas experts clearly favor cooperative control, the automation solution that enhanced their expertise.

Lane Departure Warning Systems (LDWS) are known to be effective at preventing lane departures. The current study aimed to further investigate the influence of LDWS incorrect warnings along with the warning onset on driving performances. Performances were considered during missed warned lane departure episodes, as well as correctly warned lane departure episodes following incorrectly warned (false or/and missed warning) episodes. In the reported experiment, the incorrect warnings order of presentation was manipulated along with the warning onsets (partial and full lane departure). Results pointed out that a missed warning is impairing driving performances during the missed warned situation, whereas a false warning is impairing driving performances for the subsequent lane departure. Moreover, if false warning impairment is magnified by the occurrence of a subsequent missed warning, this multiplicative effect is restricted to the following lane departure only. These findings bring both theoretical and practical insights. From a theoretical perspective, the human–machine cooperation LDWS model was refined by the adjunction of a transitional influence of incorrect warnings on the information processing. From a practical perspective, the recommendation to avoid false warnings as much as missed warnings was made.

The ever-increasing degree of automation in vehicles, with drivers delegating part of the manual driving task to automation, sets new research questions in terms of human–machine cooperation. Manual steering requires drivers to gaze at a distant road sections to ensure that the vehicle follows the road curvature and at a near road sections to maintain the vehicle within the lane limits. In this experiment, gaze behaviors engaged under highly automated driving (HAD) and manual driving conditions were compared. The results show a critical decrease of the number of gazes at the near road sections for the HAD condition compared to the manual driving condition. Two human-based highly automated “driving styles” were also compared, but did not translate into significant gaze behaviors modifications. Finally, no after-effect of the HAD was found on subsequent manual driving.