Vehicle Automation Field Test: Impact on Driver Behavior and Trust

IEEE Access

Today, and possibly for a long time to come, the full driving task is too complex an activity to be fully formalized as a sensing-acting robotics system that can be explicitly solved through model-based and learning-based approaches in order to achieve full unconstrained vehicle autonomy. Localization, mapping, scene perception, vehicle control, trajectory optimization, and higher-level planning decisions associated with autonomous vehicle development remain full of open challenges. This is especially true for unconstrained, real-world operation where the margin of allowable error is extremely small and the number of edge-cases is extremely large. Until these problems are solved, human beings will remain an integral part of the driving task, monitoring the AI system as it performs anywhere from just over 0% to just under 100% of the driving. The governing objectives of the MIT Advanced Vehicle Technology (MIT-AVT) study are to (1) undertake large-scale real-world driving data collection that includes high-definition video to fuel the development of deep learning based internal and external perception systems, (2) gain a holistic understanding of how human beings interact with vehicle automation technology by integrating video data with vehicle state data, driver characteristics, mental models, and self-reported experiences with technology, and (3) identify how technology and other factors related to automation adoption and use can be improved in ways that save lives. In pursuing these objectives, we have instrumented 23 Tesla Model S and Model X vehicles, 2 Volvo S90 vehicles, 2 Range Rover Evoque, and 2 Cadillac CT6 vehicles for both long-term (over a year per driver) and medium term (one month per driver) naturalistic driving data collection. Furthermore, we are continually developing new methods for analysis of the massive-scale dataset collected from the instrumented vehicle fleet. The recorded data streams include IMU, GPS, CAN messages, and high-definition video streams of the driver face, the driver cabin, the forward roadway, and the instrument cluster (on select vehicles). The study is ongoing and growing. To date, we have 122 participants, 15,610 days of participation, 511,638 miles, and 7.1 billion video frames. This paper presents the design of the study, the data collection hardware, the processing of the data, and the computer vision algorithms currently being used to extract actionable knowledge from the data.

Proceedings of the 8th International Conference on Automotive User Interfaces and Interactive Vehicular Applications - Automotive'UI 16, 2016

This study explores the effects of minor changes in automation level on drivers' engagement in secondary activities. Three levels of automation were tested: manual, semi-autonomous, and fully-autonomous. Potential distractor items were present and participants were instructed they could use them if they felt it was safe. Hand positions and engagement in secondary activities were manually coded. Participants were significantly less likely to engage in a secondary activity in semi-autonomous than fully-autonomous mode. Likewise, they were significantly less likely to use two hands to interact with a secondary activity in semi-autonomous mode than fully-autonomous mode. Gaze classification for each of the driver roles revealed that increasing levels of automation resulted in an increasing percentage of off-road glance durations. These observations suggest that in the event of automation failures, a driver in semi-autonomous driving may be in a somewhat better position to retake control and avoid collisions than during fully autonomous driving.

2015

Automated driving changes the role of the driver from an active operator towards a supervisor during partially automated driving and passenger in the highly automated driving mode. To foster successful interaction between humans and automated systems, feedback on automation stages and behaviors is considered a key factor. The present study used a two-step procedure to investigate drivers' information needs during partially and highly automated driving in comparison to manual driving for highway scenarios. The first step consisted in an expert focus group on expected information needs. Results showed that independent from specific scenarios, information should provide transparency, comprehensibility, and predictability of system actions. This includes the current system status, the remaining time to a change in the level of automation, the fallback level as well as reasons and a preview for ongoing and subsequent maneuvers. In the second step, results from the expert focus group were used to set up a driving simulator study. A sample of 20 participants performed three highway trips on the same route either in the manual, partially automated (hands-on, permanent monitoring, no secondary task) as well as highly automated condition (cloze test on a laptop as secondary task). Questionnaires and interviews about information needs were applied after each trip and glance behavior was analyzed. Information needs showed great variance between the drivers, which can mainly be explained by trust in automation. Partially automated driving was considered more exhausting than the other conditions due to the continuous supervision task. Information needs for the automated conditions were primarily related to the supervision of the system, whereas requested information during manual driving was centered on performing the current driving task. Glance data supported these patterns: during partially automated driving, drivers showed most and longer control glances at the mirrors and instrument cluster. Secondary task engagement during highly automated driving varied in dependence of trust in automation and the perceived complexity of the situation. However, less salient objects in a situation, such as traffic signs, were not perceived and no control glances were performed. It can be concluded that information needs change for partially and highly automated driving. Requested information is primarily focused on the status, transparency and comprehensibility of system action in contrast to driving-task related information during manual driving. These changes need to be considered in the human-machine-interface (HMI) design for automated driving.

Social Science Research Network, 2022

This study reports usage of supervised automation and driver attention from longitudinal naturalistic driving observations. Automation inexperienced drivers were provided with instrumented vehicles with adaptive cruise control (ACC) and lane keeping (LK) features (SAE level 2). Data was collected comparing one month of driving without support to two months where drivers were instructed to use automation as desired. On highways, level 2 automation was used respectively 63% and 57% of the time by Tesla and BMW users, with peak usage during slow stop-and-go traffic (0-30 km/h) and higher speeds (>80 km/h). On roads with speed limits below 70 km/h, automation was used less than 8%, and use on urban roads was incidental rather than habitual. Automation usage increased with time in trip, but no clear time of day effects were found. Head pose data could not classify driver attention, and we recommend gaze tracking in future studies. Head pose deviation was selected as alternative indicator for monitoring activity. Comparing among forms of automation usage on the highway, head heading deviation was smallest during ACC use, but did not differ between automation and baseline manual driving. Head heading deviation during manual driving was smaller in the baseline than the experimental phase, which suggests that motives for manual highway driving may be attention related. Automation usage did not change much over the first 12 weeks of the experimental condition, and there were no longitudinal changes in head pose deviation. 1. Introduction Supervised, or SAE Level 2 partial automation (SAE, 2021) is rapidly deployed in commercial cars. Current systems automate longitudinal control with adaptive cruise control (ACC) and support lateral control with lane keeping (LK). While Level 2 automation is active, the driver has to supervise the automation, and intervene when needed to ensure safety. However recent studies and accidents indicate that drivers occasionally use automation in unsuitable conditions, and are not always monitoring the environment sufficiently (Dutch Safety Board, 2019). Harms, Bingen, and Steffens (2020) found that drivers are not always aware of the abilities and limitations of current systems. Farah et al. (2021) also found that drivers overestimated the operational design domain as defined by the vehicle manufacturer in an on-road study with a Tesla. Banks, Eriksson, O'Donoghue, and Stanton (2018) observed behaviours in a Tesla and noted that drivers occasionally missed notifications from the HMI, leading to mode confusion.

Advances in intelligent systems and computing, 2017

In recent years, advanced driver assistant systems (ADAS) and solutions for automated driving have been introduced by several automotive original equipment manufacturers (OEMs) and suppliers. Currently, these types of automation are designed for partially automated driving, but the step towards higher levels of automation can be expected to be made soon. One of the most commonly addressed use cases is driving on a highway such as the German Autobahn. In this paper, we propose an approach for adapting the automation's behavior to the human's driving preferences, providing a cognitive automation system with a machine-learning algorithm. This system has been implemented in a simulator for automated driving and has been used in a study addressing conditional automation. Within the presented experiment, typical situations for automated driving under varying conditions have been tested in the driving simulator. During cooperative human-machine driving, the automation can learn the human's preferences regarding relevant driving states.

Journal of the Australasian College of Road Safety, 2019

This study evaluated the behavioural validity of the Monash University Accident Research Centre automation driving simulator for research into the human factors issues associated with automated driving. The study involved both on-road and simulated driving. Twenty participants gave ratings of their willingness to resume control of an automated vehicle and perception of safety for a variety of situations along the drives. Each situation was individually categorised and ratings were processed. Statistical analysis of the ratings confirmed the behavioural validity of the simulator, in terms of the similarity of the on-road and simulator data.

IEEE Transactions on Intelligent Transportation Systems

Highly Automated Driving technology will be facing major challenges before being pervasively integrated across production vehicles. One of them will be monitoring drivers' state and determining whether they are ready to take over control under certain circumstances. Thus, we have explored their physiological responses and the effects on trust of different scenarios with varying traffic complexity in a driving simulator. Using a mixed repeated measures design, twenty-seven participants were divided in two reliability groups with opposite induced automation reliability expectations-low and high-. We hypothesized that expectations would modulate participants' trust in automation, and consequently, their physiological responses across different scenarios. That is, increasing traffic complexity would also increase participants' arousal, and this would be accentuated or mitigated by automation reliability expectations. Although reliability group differences could not be observed, our results show an increase of physiological activation within high complexity driving conditions (i.e., a mentally demanding non-driving related task and urban scenarios). In addition, we observed a modulation of trust in automation according to the group expectations delivered. These findings provide a background methodology from which further research in driver monitoring systems can benefit and be used to train machine learning methods to classify drivers' state in changing scenarios. This would potentially help mitigate inappropriate takeovers , calibrate trust and increase users' comfort and safety in future Highly Automated Vehicles.

Theoretical Issues in Ergonomics Science, 2017

Automated driving can fundamentally change road transportation and improve quality of life. However, at present, the role of humans in automated vehicles (AVs) is not clearly established. Interviews were conducted in April and May 2015 with twelve expert researchers in the field of Human Factors (HF) of automated driving to identify commonalities and distinctive perspectives regarding HF challenges in the development of AVs. The experts indicated that an AV up to SAE Level 4 should inform its driver about the AV's capabilities and operational status, and ensure safety while changing between automated and manual modes. HF research should particularly address interactions between AVs, human drivers, and vulnerable road users. Additionally, driver training programs may have to be modified to ensure that humans are capable of using AVs. Finally, a reflection on the interviews is provided, showing discordance between the interviewees' statements-which appear to be in line with a long history of human factors research, and the rapid development of automation technology. We expect our perspective to be instrumental for stakeholders involved in AV development and instructive to other parties.

13th International Conference on Automotive User Interfaces and Interactive Vehicular Applications, 2021

While using highly automated systems, various non-critical automated driving scenarios can be identified in which trust plays a role. In this study, we investigated the change of trust in these scenarios with a digital "Feeling of Trust" indicator, through video-based online experiments simulating automated driving. Initial results show that trust even changes in these scenarios and revealed multiple influential factors. While trust seems to drop consistently in certain cases, we found individual differences in other events. With our experimental setup and findings, we provide a tool to examine trust aspects in an online study. This contributes to the understanding of how to design human-vehicle interactions in highly automated cars with the goal to calibrate trust under ordinary non-critical events. CCS CONCEPTS • Human-centered computing → Human computer interaction (HCI); HCI design and evaluation methods.

Information, 2019

This article reports on a study to investigate how the driving behaviour of autonomous vehicles influences trust and acceptance. Two different designs were presented to two groups of participants (n = 22/21), using actual autonomously driving vehicles. The first was a vehicle programmed to drive similarly to a human, "peeking" when approaching road junctions as if it was looking before proceeding. The second design had a vehicle programmed to convey the impression that it was communicating with other vehicles and infrastructure and "knew" if the junction was clear so could proceed without ever stopping or slowing down. Results showed non-significant differences in trust between the two vehicle behaviours. However, there were significant increases in trust scores overall for both designs as the trials progressed. Post-interaction interviews indicated that there were pros and cons for both driving styles, and participants suggested which aspects of the driving styles could be improved. This paper presents user information recommendations for the design and programming of driving systems for autonomous vehicles, with the aim of improving their users' trust and acceptance.

Energies

As vehicle driving evolves from human-controlled to autonomous, human–machine interaction ensures intuitive usage as well as the feedback from vehicle occupants to the machine for optimising controls. The feedback also improves understanding of the user satisfaction with the system behaviour, which is crucial for determining user trust and, hence, the acceptance of the new functionalities that aim to improve mobility solutions and increase road safety. Trust and acceptance are potentially the crucial parameters for determining the success of autonomous driving deployment in wider society. Hence, there is a need to define appropriate and measurable parameters to be able to quantify trust and acceptance in a physically safe environment using dependable methods. This study seeks to support technical developments and data gathering with psychology to determine the degree to which humans trust automated driving functionalities. The primary aim is to define if the usage of an advanced dri...

Proceedings of the 8th ACM International Symposium on Pervasive Displays

Cooperative, intelligent transportation systems (C-ITSs) have capabilities far beyond what human drivers can achieve. For instance, intelligent systems that plan the exact trajectories of vehicles could increase throughput at intersections, allowing vehicles to pass with high speed and without any need for traffic lights. It is imaginable that C-ITS will manage traffic in fully automated driving (FAD). However, the unpredictable behavior patterns of the automated vehicle (AV) are also likely to make the occupants feel uncomfortable. FAD systems operate based on a vast amount of information. As this information is potentially invisible to individuals, the question is if users trust FAD systems. We hypothesize that users experiencing such scenarios would demand system feedback to anticipate upcoming system decisions and maneuvers. Therefore, we evaluated five augmented reality (AR) user interface (UI) concepts aiming to increase system transparency in a user study (N = 30) in a driving simulation in virtual reality (VR). Our results support the assumption that feedback about the system state (in the form of route information) significantly increases trust in AVs. As trust is highly subjective, we propose to provide experience-based route visualization systems in FAD to meet individuals' needs. CCS CONCEPTS • Human-centered computing → User studies; Mixed / augmented reality.

Technical Report IMS / Department of Industrial and Material Science ;, 2020

The automotive industry is rapidly developing driving automation systems (DAS) with the aim of supporting drivers through automation of longitudinal and lateral vehicle control. As vehicle complexity increases, drivers' understanding of their responsibility and their vehicles' capabilities and limitations becomes significantly more important. In order to motivate manufacturers to adopt a human-centric perspective for the development of driving automation systems, the factors influencing the driver's perception during usage of such systems have to be understood. Therefore, the aim of this thesis is to contribute to the understanding of factors influencing user perception and understanding of driving automation systems in order to guide future design decisions from a human-centric perspective. The research for this thesis is organised into three empirical studies, embedding a mixedmethods research design. Study 1 aimed at investigating usage of DAS during different driving situations by facilitating an online survey. Studies 2 and 3 aimed to explore how drivers motivate their usage of driving automation systems, and which factors affect their understanding. Study 2 adopted an Explanatory Sequential Mixed Methods approach, consisting of a Naturalistic Driving Study and in-depth interviews to elicit knowledge about how users understand the DAS, and which factors influence usage. In Study 3 observations and interviews during an on-road driving session with a Wizard-of-Oz vehicle were conducted to gain insights into how users build an understanding of a vehicle with multiple levels of automation. The results show that the users of such systems, independent of the level of automation, talked about the systems by referring to different elements: the Context, the Vehicle, and the Driver. In addition, eleven recurring aspects describing the drivers' understanding of an automated system were discerned. Furthermore, six factors were identified that influence how drivers perceive driving automation during usage. The six factors are Preconceptions, Perceived Usefulness, Previous Experiences, Trust, System Performance, and Driving Behaviour of the Vehicle. Collectively, the identified aspects and factors constitute the building blocks of a process describing how drivers perceive driving automation systems and how this shapes their consequent understanding. The process is presented as a descriptive unified model. The main contribution of this thesis is twofold: unification of aspects found to shape a driver's understanding of a driving automation system, and the presentation of a unified descriptive model of the process showing how this understanding is shaped through what the driver perceives at the moment of use.

2019

Appropriate user trust is critical in ensuring the acceptance and safe use of Advanced Driver Assistance Systems (ADAS). Despite the prevalence of ADAS on-road today, there is a limited understanding of how trust is affected by a user's first contact with the system on-road. Ten participants without prior experience were introduced to a level 2 system and completed an on-road test drive session. Utilizing a mixedmethods approach including the Trust in Automation (TiA) questionnaire, verbal trust scores, and Facial Emotion Recognition (FER), trust in the system was measured at key milestones. TiA scores increased in a majority of participants, and a significant shift in the factor Reliability/Competence (p<0.05) was observed post-drive. According to FER scores, participants with a gain in TiA post-drive and those with a loss in TiA post-drive, more frequently displayed the emotions happy and angry, respectively. Results indicate that trust increases after a user's first experience with ADAS and further that FER may be predictive of user trust in automation.

Transportation Research Part F: Traffic Psychology and Behaviour, 2020

Driving automation systems are being introduced into mass-market vehicles, but little is known about whether drivers will trust driving automation systems and use the technology. In this study, volunteer drivers operated five vehicles equipped with automated longitudinal and lateral control and completed surveys about their experience. A subset of drivers also documented uncomfortable experiences as they used the automation while driving. Driver agreement that the automation improved the overall driving experience was significantly higher for Vehicle A than the systems implemented in the other four vehicles. Drivers reported significantly higher trust in adaptive cruise control than in lane centering in every vehicle but Vehicle B. Increased agreement that the automation consistently detected lane lines; detected moving vehicles ahead; and made smooth, gentle steering inputs was associated with significant increases in agreement that the automation improved the overall driving experience. Situations where drivers reported feeling uncomfortable with the automation during their drive were dominated by instances where lane centering struggled with common roadway features such as hills and intersections.