Recommendations for Designing Driver- Autonomous Vehicle Interaction

The following recommendations are made to improve driver experiences with SAE Level 2 AVs.

Firstly, mode awareness (Revell et al. 2020; Stanton, Dunoyer and Leatherland 2011) was addressed, including engagement and disengagement, which might relate to issues of situation awareness. In order for drivers to use the system appropriately, there were various things to remember, including conditions for activation and deactivation of ACC and steering assist, and limitations of the automation. Thus, they had to constantly pay attention to the road and the system to intervene when necessary (SAE International 2018). Therefore, it seems that clear presentation about the automation status, including a significant information for longitudinal and lateral control, is needed, such as the lead vehicle, lane markings, and the ACC speed (Revell et al. 2018). This could also help reduce mode confusion which was reported as one of the issues of partially automated driving (Banks et al. 2018). Provision of multimodal information for changes in automation status (such as a mixture of visual and auditory) seems more effective especially for takeover requests (Yoon, Kim and Ji 2019). Furthermore, the addition of vibrotactile stimulus could be considered because it was confirmed to shorten steer-touch times (Petermeijer et al. 2017). For example, it would be helpful to convey a takeover request while drivers are fixing their gaze at the road ahead and are not able to check the cluster information easily.

Secondly, an issue relevant to manoeuvrability was observed that may be linked to usability. The incorrect usage of the ACC and indicating levers was addressed in the cases of the Mercedes and Tesla with participants 1 and 4. The positions of the levers for two different features mattered: one for longitudinal and the other for lateral control. This seemed to require a careful selection of the lever according to their intentions (Revell et al. 2020). This arrangement was not effective especially when the drivers’ cognitive resources were allocated to monitoring tasks, e.g. looking at the road ahead, and the system concurrently. Therefore, it seems reasonable to put the ACC control, e.g., in a different place, or in a different shape. This could in turn enable intuitive usage as buttons with different functions located close to each other could hinder ease of use, as observed in the cases of the Mercedes and Tesla (Clarkson 2008).

Thirdly, issues related to drivers’ insufficient knowledge were identified, which may be associated with driver-AV communication (Kazi et al. 2007). The cases of participants 4 and 6 showed that a certain level of knowledge of an autonomous feature (ACC) was essential for comfortable and safe autonomous driving. It needs to be remembered that AVs can accelerate and decelerate quickly to match the ACC set speed. Thus, the drivers need to be ready either to check the speed, knowing where to look, or to adjust the speed accordingly (Stanton and Young 2005). Inappropriate usage for activation led to confusion of participant 8. Flowever, this could have been avoided if the driver had been able to build knowledge on appropriate usage considering the vehicles’ automation status which showed its availability. Driver training may be effective for drivers to make the most of autonomous features (Stanton et al. 2007), and it was suggested as a useful tool to develop drivers’ mental model (Krampell, Soh's-Marcos and Hjalmdahl 2020). Even though essential information was provided on the interface displays, it may not have been acquired as effectively as it was supposed to be from a driver’s point of view. Drivers’ information behaviour needs to be considered, including information seeking, retrieval, and usage (Wilson 2000) for the training.

In conclusion, these problems seem to be mainly related to maintaining the appropriate level of situation awareness, and understanding of the vehicle’s mental model, and the system that could assist the processes, apart from the issues caused by technical limitations. Considering the challenges of collaboration between the human and automation in relation to the handover of authority between the human and the system in the Level 2 autonomy (SAE International 2018), the driver’s role seems critical based on the observations made in the naturalistic setting. In the current system, it seems that the driver needs to identify when to give authority to the vehicle, and when to take it back, and even when the vehicle has the authority, the driver still had to monitor the system as well as the environment. If the vehicle requires the human intervention (such as the Level 2 and 3 systems), the interface should be able to help reduce the workload by providing relevant information to keep them in the loop. Further, this may lead to enhancement of perceived usefulness and satisfaction about the use of the vehicle.

Overall Summary

This study was conducted to assess drivers’ perceptual responses to naturalistic driving of SAE Level 2 AVs (SAE International 2018), and to identify events that may have affected the reactions. Drivers conducted automated and manual driving tasks in highway and urban environments, and in three different SAE Level 2 AVs: I-PACE, Mercedes, and Tesla. Interpretations were synthesised based on comparisons of subjective workload and acceptance among the driving modes, and environments. The main findings were the drivers’ workload was higher and acceptance was lower in automated than in manual driving both in highway and in urban environments, apart from a few exceptions.

Higher workload and lower acceptance in automated mode may be construed as follows. Firstly, the drivers were required to pay attention to the traffic conditions and be ready to intervene at all times in automated mode. Secondly, they needed to observe the vehicle’s behaviour to see whether it reacted appropriately in relation to the surroundings, if not to intervene. This process led to immediate actions to ensure safety. Once automation was deactivated, drivers needed to think about conditions required to be met to put the vehicle back into autonomous mode successfully. The processes for monitoring and supervising may have led to an increase in workload, especially frustration and mental demand compared to manual driving. Further, they were not perceived as useful and satisfying as much as manual driving.

Some instances that could have influenced the perceptions were described. They seemed to be related to the issues about awareness of the mode (manual or automated), automation status (activation and the status of ACC and steering assist: targeted vehicles and lane markings), and mental model (vehicles’ behaviour and the intention behind it). The more the systems relied on the driver’s memory, the more the system could be perceived to be demanding. Moreover, if the drivers were not aware of necessary information, the events could have been seen as automation surprise (Stanton, Dunoyer and Leatherland 2011; Sarter, Woods and Billings 1997). It is suggested to enable a clear communication between AV and the driver in relation to mode limitation. The interface needs to be designed to better support to shape the driver’s mental model to narrow the gap between the driver’s expectation and the vehicle’s behaviour and the information conveyed through the interfaces.