What Makes a Good Team? - Towards the Assessment of Driver-Vehicle Cooperation.

Sebastianus Martinus Petermeijer, NLR, Netherlands, Bastiaan.Petermeijer@nlr.nl
Angelica Tinga, Dutch Institute for Road Safety Research (SWOV), Netherlands, angelique.tinga@swov.nl
Reinier Jansen, SWOV, Netherlands, reinier.jansen@swov.nl
Antoine de Reus, NLR, Netherlands, antoine.de.reus@nlr.nl
Boris van Waterschoot, Rijkswaterstaat, Ministry of Infrastructure and Water Management, Netherlands, boris.van.waterschoot@rws.nl

With the introduction of driving automation, the driving task has become a shared task between driver and vehicle. Today, an increasing amount of driving tasks can be performed by the automation and the view of driver and automation acting as collaborative partners has been well established. Although this notion has been adopted in the research and design domains, means to assess the quality of the driver-automation interaction in a structured way are still lacking. Moreover, most design evaluations are usually addressed either from a technical stance or from a human factors viewpoint, which does not comply with a general acknowledged view of a unified driver-vehicle system. The aim of the current study is therefore to investigate the possibility to quantitatively evaluate the quality of the driver and vehicle cooperation. Seven dimensions indicative for the quality of cooperation are identified, based on a literature survey and expert input during focus groups. This work potentially supports road authorities, legislation, regulation and original equipment manufacturers to monitor, evaluate and design driver-vehicle cooperation.

CCS Concepts:Human-centered computing → HCI theory, concepts and models;

KEYWORDS: Human-Machine Cooperation and Teamwork, Shared Control and Coordination, Joint Cognitive Systems, Automated Driving, Performance Assessment, Shared Driving Task

ACM Reference Format:
Sebastiaan, S.M. Petermeijer, Angelica, A.M. Tinga, Antoine, A. de Reus, Reinier, R.J. Jansen and Boris, B.M. van Waterschoot. 2021. What Makes a Good Team? - Towards the Assessment of Driver-Vehicle Cooperation.. In 13th International Conference on Automotive User Interfaces and Interactive Vehicular Applications (AutomotiveUI '21), September 09-14, 2021, Leeds, United Kingdom. ACM, New York, NY, USA, 14 Pages. https://doi.org/10.1145/3409118.3475153

1 INTRODUCTION

With the wide-spread introduction of advanced driver assistance systems, like adaptive cruise control or lane keeping assistance systems, drivers are routinely sharing the driving task with automated driving systems. The responsibility for the dynamic driving task is shared between these commercially available automated driving systems and the driver. As long as full driving automation is not accomplished, which is argued to take at least 10 years if not more [1,2], drivers will share the driving task with the automation for at least part of the trip. In short, in the foreseeable future, driver and automation will need to cooperate closely to perform the driving task together.

1.1 Practical assessment methods are missing

Even though cooperation plays such an important role in vehicle automation, current evaluation of automated driving systems usually focuses either on the technical aspects of the automated system [3], or on the human factors effects of automation e.g., [4, 5, 6]. In research focussing on the driver-automation interaction it is often argued that vehicle automation has the potential to offer benefits in terms of traffic safety, traffic flow, and comfort [7]. However, the effects of sharing the driving task between human driver and automated driving system on traffic safety and flow, are currently not well-known [8], and insights into the effects of vehicle automation are often limited to anecdotal evidence or theoretical arguments.

Available literature shows that the hybrid situation, where driver and vehicle share autonomy and control, lacks a dedicated assessment tool. For example, Abbink et al. [9] argue that shared control systems need to be evaluated “fairly”, by subjecting them to experimental conditions inside and outside the design boundaries of the automation. Similarly, Carsten and Martens [10] propose design principles, like “stimulate the appropriate level of attention and intervention” for the HMI of automated vehicles of the future. Such frameworks are useful for abstract considerations of driver-automation interaction but offer less insight into practical methods for evaluation of the cooperation between driver and automation. Also, little research is dedicated towards the evaluation of the performance of the human-machine system and thus far a structured approach to assess the quality of cooperation at the driving task level is missing. The main premise for the current research is the view that existing assessment techniques and concepts do not satisfactory address human-machine interaction and the quality of their cooperation as a team effort, which is in agreement with recent findings of the Dutch Safety Board [8].

An assessment tool for driver-vehicle collaboration could be used for monitoring, design, evaluation, regulation and admission purposes. Furthermore, answering the need for an ability to explore and evaluate the hybrid situation complies with the ambition to reveal the effects of increased vehicle automation and potentially contributes to relevant insights, such as the relationship between increased automation, sharing vehicle control and road safety.

1.2 Aim

The aim of this article is to set a first step towards objective evaluation of driver-automation cooperation. To this end, this study identified important dimensions, which characterize the quality of cooperation between driver and automation. The dimensions were identified in two phases. Firstly, a literature survey was performed to identify potentially relevant dimensions. Secondly, three focus group sessions were organized to discuss the view of the experts on assessment of the shared driving task and the results of the literature survey. This resulted in a final set of seven dimensions defining the quality of cooperation between driver and automation. Finally, recommendations for the next steps towards an objective evaluation of the cooperation quality between the driver and automation are provided, with a focus on the identification of quality indicators. These indicators could eventually be used to define minimum requirements to facilitate standardized methods of evaluation regarding the quality of cooperation.

2 BACKGROUND: DRIVER AUTOMATION COOPERATION

2.1 The shared driving task

The current research views the dynamic driving task, with the subtasks “object and event detection and response”, lateral vehicle control, and longitudinal vehicle control, as a shared task between driver and automation. That is, when a subtask is performed by both simultaneously (e.g., a lane keeping assistance system supports the steering task of the driver), or when the allocation of subtasks is shared (e.g., the driver steers, while the automation controls speed), the situation is considered a shared driving task (see fig. 1). Existing literature on the design and evaluation of driver-automation interaction in terms of cooperation and collaborations shows that human computer interaction and driving are relatively well understood. However, the problem of assessing the quality of this collaboration is less well understood.

Figure 1
Figure 1: The dynamic driving task as defined by [3]. When driver and automation perform a single task simultaneously or tasks are allocated to either the driver or automation within the dynamic driving task, it is considered a shared driving task. OEDR: Object and Event Detection and Response.

Driving automation and driver support systems rely on similar processes as their end users (i.e., drivers), which means that they act on perceived information and some sort of decision making. However, while depending on numerous sources of information, their effectiveness highly depends on the compliance of drivers when offered warnings and directions, and the system will only reveal its safety value when drivers act in accordance with the support that is given. This involves matching driving automation with humans to achieve common goals through collaboration, which views driver support as an automated co-driver or team member, rather than substitution [11].

The view of humans and machines cooperating as a joint cognitive system was coined by Rasmussen [12] and Hollnagel & Woods [13] and has been adopted for the context of driving (e.g., [14, 15, 16]). At present, many attempts to describe cooperative driver support have been presented (e.g., [17, 18]) and since one of its first appearances in [19], the concept of automation as a team player has been widely adopted (e.g., [20, 21, 22, 23, 24, 25]).

In the current research, the concept of a shared driving task is highly related to research topics like team performance, joint action, human-machine cooperation and -collaboration and task sharing. In accordance with the suggestion of Flemisch et al. [26] to analyze and develop the concepts of shared control and human-machine cooperation together, the present work views driver and automation as social partners or agents sharing the task of driving. The focus of the present work is aimed at the ability to evaluate the driving task in terms of quality of cooperation or collaboration between these agents.

2.2 Driver and automation as team players

When driver vehicle cooperation is viewed as teamwork and joint action between social partners, research concerning joint action between humans might provide valuable insights to accomplish qualitatively good cooperation. For example, [27] proposed the construct Teamwork Quality (TWQ) which comprises performance-relevant measures of team internal interaction, emphasizing that their study focusses on the quality of teamwork rather than the content of the shared tasks and activities. In order to measure Teamwork Quality the authors identified six concepts - communication, coordination, balance of member contributions, mutual support, effort, and cohesion – addressing questions concerning the proposition that the success of teamwork depends on the quality of interactions within teams and collaboration between team members. Sebanz et al. [28] defined joint action as any form of social interaction whereby two or more individuals coordinate their actions in space and time to bring about a change in the environment. In the authors’ view, successful joint action depends on three abilities. Firstly, the ability to share representations. Through a mechanism of joint attention, a kind of ‘perceptual common ground’ is created which stresses the importance that interacting partners know about each other what they do and do not perceive. Secondly, successful joint action depends on the ability to predict each other's actions. [29] discusses how human agents are able to predict each other's actions and argue that this ability is mediated by a common coding of perceived and performed actions of jointly acting partners. In line with the view that perception‐action links allow us to infer the goals of others’ actions [30], this not only suggests an innate ability to infer and predict the actions of others, it shows how humans are wired to perform tasks together. In line with the findings concerning joint action among humans, [25] explains how difficulties associated with human-machine cooperation are closely related to predictability and the congruence between predicted and actual outcomes of actions. By associating the concept of agency with being in or out of the control loop, it is suggested that ‘priming’ (i.e.,” providing predictive information of what will happen next”) could be helpful in order to reach or restore an appropriate sense of control. By means of two experiments using an aircraft supervision task, Berberian and colleagues explored the question how a sense of control could be preserved when interacting with a highly automated system [31]. Their results suggest that displaying the system's intentions prior to an action maximizes both takeover efficiency and human-machine teaming.

A third important ability for successful joint action that was identified by [28] is the ability to integrate predicted effects of own and others’ actions. Within the context of human and machine cooperation the requirements of successful joint action are strongly related to the requirements that are distinguished by research that aims to make automation and their end users team players. Pioneers in the field stressed that in order for automated agents to become team players, observability and directability [21] together with common ground [22] are fundamental design requirements, which are illustrated by “ten challenges for making automation a team player”[22]. Subsequently, [24] applied the team player approach on car design and concluded that the process of teambuilding might be an important prerequisite or even an alternative view of team play. Recently, [32] proposed four basic requirements for humans and automation to be effective team players: mutual predictability, directability, shared situation representation and calibrated trust in automation.

3 METHOD

This study consisted of two parts, namely a literature survey which identified candidate dimensions which characterize the cooperation between driver and automation; and three focus groups in which the identified dimensions were discussed among experts within the automotive domain.

3.1 Literature survey

Between October and December 2020, relevant literature was searched in the databases of Google Scholar, Web of Science, and Science Direct, by using (a combination of) the following key words:

  • Shared driving / shared control / cooperative driving / human-machine teaming / driver automation teaming
  • Driving assistance system / driving automation / advanced driver assistance system / automated driving system / driver support / adaptive automation
  • Human-machine interaction / human machine cooperation / task sharing / joint action

The results of the literature search were first evaluated based on the manuscript's title, to determine whether it was potentially relevant to include in the literature survey. Next, the abstract was read to determine if the manuscript met any of the inclusion criteria, or should be discarded based on the exclusion criteria. If needed, the complete manuscript was read to decide on the in- or exclusion. The entire process of searching and selecting literature was performed by three of the current study's authors.

The inclusion criteria were as follows:

  • The manuscript is published scientific literature, such as a conference paper, journal article, or book.
  • The cooperation between driver and automation fits the definition of the shared driving task as given in the introduction,
  • The research goal is to improve traffic safety,
  • The research goal is primarily focused on the cooperation between driver and automation,
  • An aspect of the cooperation between driver and automation is quantifiably measured and/or theoretically discussed, and
  • The manuscript is written in English or Dutch.

The exclusion criteria were as follows:

  • The study focuses exclusively on a strategic subtask of the driving task,
  • The automation system studied is an active safety system, like active emergency braking,
  • The cooperation between driver and automation is not the primary focus of the study (e.g., the article describes a system design and performs a relatively small validation study), or
  • The research goal and results match other studies found earlier, and therefore provide no novel insights for the framework.

Based on the title and abstract, 85 studies were selected to be included in the literature study, 43 were excluded based on the exclusion criteria. As a result 42 studies were included in the analysis. The selected manuscripts were summarized, using a word-template, in terms of publication details (i.e., title, authors, full reference), the research goal and domain, and any relevant measurements, results, and conclusions. This information was used to find commonalities between the studies in order to extract generalizable constructs which capture the cooperation between driver and automation (i.e., dimensions), as well as to extract potential performance indicators to quantify these dimensions.

The literature survey yielded fourteen candidate dimensions, which identified by the authors as relevant for the evaluation of cooperation, and served as discussion topics in the focus groups.

3.2 Focus groups

The results from the literature survey (i.e., candidate dimensions with respect to driver-automation cooperation) were discussed and reviewed in three separate focus groups. Participants were recruited by approaching 21 experts from the personal network of the authors, of which fifteen experts agreed to participate in a workshopsession. A total of 15 experts from car manufacturers, academics, and safety institutes (e.g., EURONCAP) participated, consisting of eight full professors, five post-doc research fellows from various European and American universities, an expert on standardization and regulations, and an HMI-expert employed by a major European car manufacturer. Participants with similar expertise were divided as much as possible across the three focus groups to create heterogeneous groups to stimulate discussion. The agenda for each focus group session was identical and as such each session consisted of the following three parts:

  • Replication of the results of the literature survey.

During the first part of the session, the experts were asked to answer the following questions using the interactive presentation software Mentimeter [33].

  • What aspects characterize good coordination between the human and automation in the shared driving task?
  • What aspects characterize good communication between the human and automation in the shared driving task?
  • What other aspects would characterize a shared driving task of good quality?
  • What aspects would characterize a shared driving task of inadequate quality?

Each expert could individually answer the questions in the Mentimeter application. Experts were asked to provide two answers to all of the questions, after which all answers were shown to all participants, and a discussion was led by the moderator of the focus group. When answers were unclear, the moderator would ask the experts to elaborate and explain.

  • Critical discussion of dimensions related to driver-automation cooperation.

The second part of the focus group started with a short introduction and elaboration on the dimensions the authors identified as important with respect to driver-automation cooperation. Experts were then asked to rate, on a scale from 1 to 10, how important they thought each dimension was for the quality of the shared driving task. Finally, a discussion was held about the dimensions in general, and each dimension specifically. In this discussion, the ratings were used to reveal any discrepancies between the opinions of the experts. For example, if a certain dimension was rated of low importance by some experts, while it was rated high by others the discussion leader would instigate a discussion.

  • Identifying quality indicators for the cooperation between driver and automation.

During the final part of the focus group, experts were asked to identify indicators that could be used to assess the quality of cooperation between driver and automation.

During the sessions, one author made notes of the entire session. Moreover, the audio of all sessions was recorded and the answers and ratings which the experts provided were stored for the purpose of posterior inspection of the notes.

4 RESULTS

4.1 Literature survey

The literature survey using the inclusion and exclusion criteria (section 3.1.) yielded 42 studies, consisting of 20 empirical studies, 20 theoretical studies, and 2 standard/norm reports. Included are: [4, 5, 6, 9, 10, 34-71]. The empirical studies mainly focussed on specific effects of the cooperation between driver and automation and measured these effects objectively. For example, [34] evaluated the driver's visual attention in different visual conditions with an eye-tracking device. On the one hand, empirical studies provided valuable insights into the identification of specific quality indicators for cooperation, but their results were often hard to generalize beyond the experiment boundaries. On the other hand, theoretical studies provided an abstract perspective on human-automation cooperation, for example by presenting overviews of the human factors issues, interface guidelines, or interaction frameworks. These studies offered valuable insights into how different aspects of cooperation are interrelated, but contributed less to practical indications for quality of cooperation measurements.

Fourteen candidate dimensions were formulated based on the literature survey, which are listed in Table 1. The concept dimensions were categorized in three categories, namely if the dimensions is related to the driver, automation, or the cooperation between them.

Table 1: List of the concept dimensions which were formulated based on the literature survey and their average importance score across all experts. The importance score was the answer to the question: To what extent do you find each dimension of importance for the quality of the shared driving task (from 1-10)?
Category Concept dimension Average importance score
Driver Situation awareness 8.0
Mental model 8.1
Trust 7.8
Attention 8.5
Cognitive load 8.0
Non driving related activity-characteristics 7.7
Driver state 7.9
Acceptance 7.3
Driver profile 6.0
Automation Automation limits 7.9
Cooperation between driver and automation Conflicts 8.5
Goals 8.1
Communication 8.5
Distribution of responsibility, capability, and authority 9.2

4.2 Focus groups

In general, the experts agreed on what aspects were relevant with respect to the quality of cooperation. The experts highlighted that aspects that are shared between the human and the automation are especially of importance to consider. For example, situational awareness should not only relate to the human driver but also to the automation, both the driver and automation need to share situation awareness of the environment.

The average scores have been included in Table 1. However, these scores should be considered carefully, as they were mainly used to instigate discussion among experts and do not offer much insights beyond that. The scores for most concept dimensions are similar, so experts subjectively rated those as important for the assessment shared driving task quality. The main result from this part of the session was that almost concept dimensions are relevant, but that some are very similar and clustering these dimensions should be considered. The ratings the experts provided in the second part of the focus groups also revealed much coherence among experts on the importance ratings of the dimensions.

Only the driver profile was rated as somewhat less important, as the experts argued that good cooperation between driver and automation should account for many different drivers. Moreover, the experts agree that the framework should focus on dimensions that relate to the driver and automation, and not solely the driver or automation. Some experts suggested that dimensions currently relating to solely the driver or automation could in fact be related to both and should be considered as such.

The third part of the session revealed another common observation between experts, namely that it would be very difficult to determine quality indicators which are objectively quantifiable, as the quality of cooperation is highly situation dependent. Moreover, it was indicated that quality indicators can be related to multiple dimensions of the cooperation. In short, it is challenging to operationalize quality indicators and it will require thorough research and a structured approach to be able to use such indicators for standardized evaluation methods or real-time measurements in commercial vehicles.

4.3 Model for driver automation cooperation

This section describes the dimensions which have been indicated as relevant for the quality of driver-automation cooperation, based on the literature study and focus groups. The resulting model for the cooperation between driver and automation is shown in Figure 2. The large circles illustrate the driver and automation, who both perform Perception (P) – Decision (D) – Action (A) cycles. External events, for example a braking lead vehicle, affect the driver and automation in their shared driving task. Cooperation between the driver and automation is illustrated by the part where the circles overlap.

Figure 2
Figure 2: Illustration of the cooperation between driver and automation in a shared driving context. The model illustrates the quality indicators, which suggest the state of the dimensions and consequently the quality of the cooperation. P = perception, D = Decision, A = Action.

Seven dimensions which capture good cooperation between the driver and automation (illustrated as vertical lines, see section 4.2 for details) were identified. Evaluation indicators are illustrated as small circles on the vertical lines, which are indicative for the quality of the cooperation. Evaluation indicators are not dimension specific, meaning that an evaluation indicator could be indicative for multiple dimensions (illustrated by the circles that connect to multiple vertical lines).

4.4 Dimensions

The following seven dimensions impacting the quality of cooperation between driver and automation, were identified.

4.4.1 Compatible goals. Both the driver and automation have or are programmed with goals related to the driving task. Such goals, like maintaining a constant speed or a safe distance to a vehicle ahead, need to correspond to a certain degree or be complementary to accomplish good cooperation. [48] and [14] refer to compatible goals or interfering goals, and argue that goals of the driver and the automation do not have to match exactly for them to collaborate well. Goals that differ a lot, or opposing goals, are viewed as detrimental to the cooperation.

Goals can be categorized on a strategic, tactical, and operational level. Incompatible goals between the driver and automation on an operational level could, for example, result in discrepancies between the intended steering wheel or gas pedal position [9]. For example, the automation maintains a constant speed, whereas the driver wants to brake. Note that a discrepancy on an operational level does not necessarily originate from a incompatible goal on the operational level, but could be the result of an incompatible goal on a strategic level (e.g. the automation may maintain speed because it wants to overtake a slow vehicle and the driver would rather adjust the speed).

4.4.2 Shared situational awareness. Situational awareness is typically divided into three levels, namely perception, interpretation, and projection [72]. It is crucial that the driver and automation perceive the cues from the driving environment, that are relevant to the driving task each of them is responsible for. When driver and automation share a task, they need to perceive the same cues, and interpret and project those similarly. Alternatively, when a driving task is completely allocated to either driver or automation, their situational awareness does not necessarily have to be similar. Their combined situational awareness, however, does need to be complementary so that it is sufficient for safe performance on the entire dynamic driving task.

4.4.3 Consistent and compatible mental models. For the driver and the automation to collaborate well they should have accurate mutual mental models [73]. In the literature, a mental model of a driver is defined as an internal mental representation of the manner in which the automation operates. People form mental models by interaction and these evolve continuously. Moreover, mental models enable someone to make predictions and explain system behavior, but they can be incorrect [58, 74, 75]. An incorrect mental model could result in over- or distrust, whereas a good mental model facilitates calibrated trust [76]. Hence, a driver with an accurate mental model of the automation is able to anticipate when the system limits will be reached. In other words, a driver requires an accurate mental model to predict when an automation will disengage or when it will fail to operate safely.

For good cooperation, not only the driver needs an accurate model of the automation, the automation also needs an accurate model of the driver's cognitive and physical state, and adapt its behavior accordingly. An automation could adapt the output modalities and the salience of a take-over request to the driver state. For example, by making a warning sound louder when it detects that a driver is drowsy.

4.4.4 Distribution of responsibility, capability, and authority. For the driver and automation to collaborate well, it is crucial that the driving tasks are allocated accordingly [73]. Responsibility is, in the context of human-machine interaction, the degree the (un)successful performance of a task and the consequences thereof can be attributed to the driver or automation. Capability is the capacity of the driver or automation to perform a (part of the) driving task successfully. Authority can be defined as the ability to allow another (for example the automation) to perform a task (or not).

For good cooperation between the driver and the automation the task should be allocated as such that the responsibility, capability, and authority of the driver and automation are complementary. Ideally, they should display adaptive behavior and compensate for each other when one fails to perform its task properly. For example, the automation could (if it is allowed and capable) take over control of the vehicle when the driver is inattentive. However, the driver should always be able to intervene when the automation makes a mistake. [73] state that the responsibility should not exceed the authority, and that the authority and responsibility should not exceed the capability of either the driver or the automation.

4.4.5 Adaptability. Driving happens in a complex and dynamic environment. During nominal driving conditions, adaptability is not crucial for good cooperation between the driver and automation. However, when an unanticipated event occurs, a team that is able to adapt is more likely to better cope than a team with a static approach to cooperation.

Good cooperation between driver and automation requires, therefore, adaptability of both agents. For example, depending on the level of automation [68] the driver needs to fulfill different roles, like supervisor or co-pilot. On the other hand, the automation should be able to adapt its behavior (and perhaps allocated tasks; see section 4.2.4) based on the driver state. Moreover, the automation should be able to handle various types of driver and driving styles.

4.4.6 Conflicts. When the driver and automation share a driving task, like steering, they provide input to the steering wheel simultaneously. A conflict happens, according to Itoh, Flemisch, and Abbink [56], when the input of the driver is not consistent with the input of the automation. As such, conflicts are strongly related to the compatible goals (section 4.2.1). When the automation and the driver collaborate well by complying to the other dimensions, the occurrence of conflicts should be limited to a minimum.

Not only should conflicts be minimized in number and magnitude. When they do occur, there should be a robust approach to resolve them. Small conflicts are unlikely to negatively impact the traffic safety directly, but they can lead to irritation, which in turn can lead to disuse of the automation [73]. Large conflicts can negatively impact the traffic safety, when they are not properly communicated and arbitrated. Nevertheless, the agent (driver or automation) that is responsible for the driving task should also have the authority to resolve the conflict. As long as drivers are legally responsible for traffic safety, this means that overruling the automation should be simple and intuitive, like pressing the brake pedal to disengage an automatic cruise control system.

4.4.7 Communication. Communication between the automation and driver is an overarching concept that happens at all three levels of the cooperation, namely the strategic, tactical, and operational level [9, 48]. Clear and unambiguous communication mitigates or prevents incorrect interpretation of information [74] and consequently facilitates good cooperation. There are two important aspects related to good communication, namely what content is communicated and how does the information impact the automation or driver.

As to what is communicated, it is, for example, important that the driver receives clear information on 1) the (current) automation mode, 2) time budgets to foreseen events, 3) reasons for changes in automation modes, 4) reasons for planned maneuvers, 5) driver responsibilities, and 6) sensor quality of the automation [75].

Regarding the impact of communication, the human-machine-interface should 1) support the driving tasks, 2) facilitate conflict resolution, 3) facilitate transfer of control between driver and automation, 4) update driver and automation on each other's intentions, capabilities, and limits, 6) improve driver readiness through intervention and prevention, 7) facilitate calibrated trust to avoid under- or over-trust, and 8) support adequate mode awareness.

Communication is inherently related to all other dimensions and there is no single correct approach to facilitate communication between the driver and automation. Generally though, it can be argued that information overload and unnecessary distraction should be avoided and that individual driver preferences should be taken into account.

As already mentioned, important prerequisites for coordinating tasks and actions between agents are shared and compatible goals. To achieve this, Hoc et al. [74] stressed the importance of optimizing the communication between human and machine. Similarly, [38] showed how communicating information about the automation, like uncertainty, could contribute to the improvement of driver–automation cooperation. By conducting a driving simulator study, Beller and colleagues [38] found that presenting uncertainty information to participants, time to collision increased in the case of automation failure. Furthermore, participants showed improved situational awareness, better knowledge of fallibility, higher trust ratings and increased acceptance when reliability information about the automation was communicated.

5 DISCUSSION & CONCLUSION

The aim of this paper was to set a first step towards objective evaluation of driver-automation cooperation. Based on a literature survey and focus groups with experts from the automotive domain, seven dimensions which characterize the cooperation between driver and automation were identified. This study builds upon the existing knowledge and frameworks so that ultimately quantifiable metrics to evaluate the cooperation between driver and automation can be determined. However, the framework as presented in this paper has not yet reached that goal.

Currently, the dimensions are abstract constructs, like shared situational awareness, which cannot be measured directly and therefore additional development of the framework is needed. To develop the dimensions further towards a framework that can be used to define evaluation methods, the dimensions need to be operationalized so that they can be measured and quantified. Quantifiable variables, based on behavioral measures, or tests will be able to indicate if cooperation with respect to a certain dimension is going well. For example, eye-tracking can indicate whether drivers are paying attention to cues on the road relevant for the part of the driving task that they are responsible for. Note that eye-tracking is not only an indicator for adequate shared situation awareness, but could also be used to evaluate if mode changes have been perceived. Indicators should be viewed in the context (i.e., relevant traffic cues) to provide sufficient insight into the quality of the cooperation between driver and automation.

In addition, the dimensions are not strictly independent and relate to each other in various degrees. ‘Conflicts’ and ‘compatible goals’ are for example highly related to each other, whereas ‘shared mental models’ and ‘adaptability’ are related to a lesser degree. We argue that the relationship between two dimensions becomes most apparent in the quality indicators that are shared between two dimensions. How much the dimensions are related is currently unclear at present. Future research should investigate which quality indicators are representative for each dimension in order to reveal the relationship between the dimensions. Nonetheless, in its current form they already offer a useful perspective to discuss and set the first steps towards evaluating cooperation between driver and automation on an abstract level. It could, for example, bring a new perspective in evaluating traffic accidents compared to the usual and rather technical approach.

For now, the framework consists of the seven dimensions of cooperation, but lack quantitative evaluation indicators. Hence, further development is needed to transform the dimensions into a framework that can be applied to assess the quality of cooperation between the driver and automation. The first step would be to identify quantifiable quality indicators per dimension, as well as techniques and methods to measure these indicators. Next, evaluation scenarios should be determined that can be used to validate the quality indicators. Evaluation scenarios should entail a range of possible traffic situations, for example nominal highway driving, but also unexpected events. Finally, it should be investigated which (minimum) selection of quality indicators is representative to evaluate the quality of cooperation between driver and automation.

The ultimate goal is to develop evaluation methods to assess the cooperation, which could be used in standardized tests by safety organizations (e.g., EuroNCAP) or perhaps for real-time evaluation to facilitate adaptive automation. The dimensions defined in this paper can be used to serve as a basis for a practical framework to assess cooperation between driver and automation.

ACKNOWLEDGMENTS

This study was funded by Rijkswaterstaat, Ministry of Infrastructure and Water management under contract number 31162195.

REFERENCES

  • Bengler, K., Dietmayer, K., Färber, B., Maurer, M., Stiller, C., & Winner, H. (2014). Three decades of driver assistance systems – Review and future perspectives. IEEE Intelligent Transportation Systems Magazine, 6(4), 6-22.
  • Dia, H. (2021). ‘Self-driving’ cars are still a long way off. Here are three reasons why. The Conversation, April 22, 2021.URL retrieved from: https://theconversation.com/self-driving-cars-are-still-a-long-way-off-here-are-three-reasons-why-159234
  • EuroNCAP, European New Car Assesment Programme(2018), retrieved from https://www.euroncap.com/en/vehicle-safety/safety-campaigns/2018-automated-driving-tests/
  • Breyer, F., Blaschke, C., Farber, B., Freyer, J., & Limbacher, R. (2010). Negative behavioral adaptation to lane-keeping assistance systems. IEEE Intelligent Transportation Systems Magazine, 2(2), 21-32.
  • Endsley, M. R. (2018). Situation awareness in future autonomous vehicles: Beware of the unexpected. In Congress of the International Ergonomics Association (pp. 303-309). Springer, Cham.
  • Jamson, A.H., Merat, N., Carsten, O.M.J., & Lai, F.C.H (2013). Behavioural changes in drivers experiencing highly-automated vehicle control in varying traffic conditions. Transportation Research Part C: Emerging Technologies, Volume 30, Pages 116-125, ISSN 0968-090X, https://doi.org/10.1016/j.trc.2013.02.008.
  • De Winter, J. C. F., Happee, R., Martens, M. H., & Stanton, N. A. (2014). Effects of adaptive cruise control and highly automated driving on workload and situation awareness: A review of the empirical evidence. Transportation research part F: traffic psychology and behaviour, 27, 196-217.
  • Dutch Safety Board, Dutch Safety Board (2019). Who is in Control? Road Safety and Automation in Road Traffic (report). The Hague, The Netherlands: Dutch Safety Board. Retrieved from https://www.onderzoeksraad.nl/en/page/4729/who-is-in-control-road-safety-and-automation- in-road-traffic.
  • Abbink, D. A., Carlson, T., Mulder, M., de Winter, J. C.F., Aminravan, F., Gibo, T. L., & Boer, E. R. (2018). A topology of shared control systems—finding common ground in diversity. IEEE Transactions on Human-Machine Systems, 48(5), 509-525.
  • Carsten, O., & Martens, M.H. (2019). How can humans understand their automated cars? HMI principles, problems and solutions. Cognition, Technology & Work, 21, 3-20.
  • Van Waterschoot, B. M. (2013). The challenge of designing intelligent support behavior: emulation as a tool for developing cognitive systems. Enschede, The Netherlands: University of Twente.
  • Rasmussen, J. (1983). Skills, rules, and knowledge; signals, signs, and symbols, and other distinctions in human performance models. IEEE Transactions on Systems, Man, and Cybernetics, 13(3), 257-266.
  • Hollnagel, E., & Woods, D. D. (1983). Cognitive systems engineering: New wine in new bottles. International journal of man-machine studies, 18(6), 583-600.
  • Hoc, J.M., Mars, F., Milleville-Pennel, I., Jolly, É., Netto, M., Blosseville, J.M. (2006). Human-machine cooperation in car driving for lateral safety: delegation and mutual control. Le travail humain 69(2), 153–182.
  • Biester, L. (2008). Cooperative automation in automobiles. (Doctoral dissertation). Humboldt Universität Berlin, Berlin. Retrieved from http://edoc.hu-berlin.de/dissertationen/biesterlars-2008-12-12/PDF/biester.pdf
  • Flemisch, F., Kelsch, J., Löper, C., Schieben, A. & Schindler, J. (2008). Automation spectrum, inner/outer compatibility and other potentially useful human factors concepts for assistance and automation (1-16). In D. de Waard D, F. O. Flemisch, B. Lorenz, H. Oberheid,& K. A. Brookhuis (Eds.), Human factors for assistance and automation. Maastricht: Shaker Publishing.
  • Flemisch, F. O., Bengler, K., Bubb, H., Winner, H., & Bruder, R. (2014). Towards cooperative guidance and control of highly automated vehicles: H-Mode and Conduct-by-Wire. Ergonomics, 57(3), 343-360.
  • Walch, M., Sieber, T., Hock, P., Baumann, M., & Weber, M. (2016, October). Towards cooperative driving: involving the driver in an autonomous vehicle's decision making. In Proceedings of the 8th international conference on automotive user interfaces and interactive vehicular applications (pp. 261-268).
  • Malin, J. T., Schreckenghost, D. L., Woods, D. D., Potter, S. S., Johannesen, L., Holloway, M., & Forbus, K. D. (1991). Making Intelligent Systems Team Players: Case Studies and Design Issues. (NASA Technical Memorandum 104738). Houston, TX: NASA Johnson Space Center.
  • Sarter, N. B., Woods, D. D., & Billings, C. E. (1997). Automation surprises. Handbook of human factors and ergonomics, 2, 1926-1943.
  • Christoffersen, K., & Woods, D. D. (2002). How to make automated systems team players. Advances in Human Performance and Cognitive Engineering Research, 2, 1-12. Elsevier Science Ltd. doi:10.1016/S1479-3601(02)02003-9.
  • Klein, G., Woods, D. D., Bradshaw, J. M., Hoffman, R. R., & Feltovich, P. J. (2004). Ten challenges for making automation a" team player" in joint human-agent activity. IEEE Intelligent Systems, 19(6), 91-95.
  • Young , M. S. Stanton, N. A. & Harris, D. (2007). Driving automation: Learning from aviation about design philosophies. International Journal of Vehicle Design, 45(3), 323-338.
  • Davidsson, S., & Alm, H. (2009). Applying the" Team Player" approach on car design. In International Conference on Engineering Psychology and Cognitive Ergonomics (pp. 349-357). Springer, Berlin, Heidelberg.
  • Berberian, B. (2019). Man-Machine teaming: a problem of Agency. IFAC-PapersOnLine, 51(34), 118-123.
  • Flemisch, F., Abbink, D., Itoh, M., Pacaux-Lemoine, M. P., & Weßel, G. (2016). Shared control is the sharp end of cooperation: Towards a common framework of joint action, shared control and human machine cooperation. IFAC-PapersOnLine, 49(19), 72-77.
  • Hoegl, M., & Gemuenden, H. G. (2001). Teamwork quality and the success of innovative projects: A theoretical concept and empirical evidence. Organization science, 12(4), 435-449.
  • Sebanz, N., Bekkering, H., & Knoblich, G. (2006). Joint action: bodies and minds moving together. Trends in cognitive sciences, 10(2), 70-76.
  • Sebanz, N., & Knoblich, G. (2009). Prediction in joint action: What, when, and where. Topics in Cognitive Science, 1(2), 353-367.
  • Rizzolatti, G. & Craighero, L. (2004). The mirror-neuron system. Annual review of Neuroscience 27, 169–92.
  • Le Goff, K., Rey, A., Haggard, P., Oullier, O., & Berberian, B. (2018). Agency modulates interactions with automation technologies. Ergonomics, 61(9), 1282-1297.
  • Walch, M., Mühl, K., Kraus, J., Stoll, T., Baumann, M., & Weber, M. (2017). From car-driver-handovers to cooperative interfaces: Visions for driver–vehicle interaction in automated driving. In Automotive user interfaces (pp. 273-294). Springer, Cham.
  • Mentimeter, Menti (2021). Mentimeter features. URL: https://www.mentimeter.com/features
  • Louw, T., & Merat, N. (2017). Are you in the loop? Using gaze dispersion to understand driver visual attention during vehicle automation. Transportation Research Part C: Emerging Technologies, 76, 35-50.
  • Abbink, D. A., & Mulder, M. (2010). Motivation for a neuromuscular basis for haptic shared control. IFAC Proceedings Volumes, 43(13), 299-303.
  • Banks, V.A., & Stanton, N.A. (2015). Discovering Driver-vehicle Coordination Problems in Future Automated Control Systems: Evidence from Verbal Commentaries. Procedia Manufacturing, Volume 3, Pages 2497-2504, ISSN 2351-9789, https://doi.org/10.1016/j.promfg.2015.07.511.
  • Banks, V.A., & Stanton, N.A. (2016). Keep the driver in control: Automating automobiles of the future. Applied Ergonomics, Volume 53, Part B, Pages 389-395, ISSN 0003-6870, https://doi.org/10.1016/j.apergo.2015.06.020.
  • Beller, J., Heesen, M., & Vollrath, M. (2013). Improving the driver–automation interaction: An approach using automation uncertainty. Human factors, 55(6), 1130-1141.
  • Boink, R., Van Paassen, M. M., Mulder, M., & Abbink, D. A. (2014, October). Understanding and reducing conflicts between driver and haptic shared control. In 2014 IEEE International Conference on Systems, Man, and Cybernetics (SMC) (pp. 1510-1515).
  • De Waard, D., & Lewis-Evans, B. (2014). Self-report scales alone cannot capture mental workload. A reply to De Winter, Controversy in human factors constructs and the explosive use of the NASA TLX: a measurement perspective. Cognition, Technology & Work, 16, 303-305.
  • De Winter, J. C. F., & Dodou, D. (2011). Preparing drivers for dangerous situations: A critical reflection on continuous shared control. In 2011 IEEE International Conference on Systems, Man, and Cybernetics (pp. 1050-1056). IEEE.
  • Du, N., Zhou, F., Pulver, E. M., Tilbury, D. M., Robert, L. P., Pradhan, A. K., & Yang, X. J. (2020). Predicting driver takeover performance in conditionally automated driving. Accident Analysis & Prevention, 148, 105748.
  • EASA, European Union Aviation Safety Agency (2007).CS 25.1302 Installed systems and equipment for use by the flight crew. In: Certification Specifications and Acceptable Means of Compliance for Large Aeroplanes CS-25, Amendment 25, 24 June 2020.
  • Endsley, M. R. (2017). Autonomous driving systems: A preliminary naturalistic study of the Tesla Model S. Journal of Cognitive Engineering and Decision Making, 11(3), 225-238.
  • Engström, J., Markkula, G., Victor, T., & Merat, N. (2017). Effects of Cognitive Load on Driving Performance: The Cognitive Control Hypothesis. Human Factors, 59(5), 734–764. https://doi.org/10.1177/0018720817690639
  • Eriksson, A., Petermeijer, S. M., Zimmermann, M., De Winter, J. C. F., Bengler, K. J., & Stanton, N. A. (2018). Rolling out the red (and green) carpet: supporting driver decision making in automation-to-manual transitions. IEEE Transactions on Human-Machine Systems, 49(1), 20-31.
  • Flemisch, F. O., Adams, C. A., Conway, S. R., Goodrich, K. H., Palmer, M. T., & Schutte, P. C. (2003). The H-Metaphor as a guideline for vehicle automation and interaction.
  • Flemisch, F., Abbink, D. A., Itoh, M., Pacaux-Lemoine, M. P., & Weßel, G. (2019). Joining the blunt and the pointy end of the spear: towards a common framework of joint action, human–machine cooperation, cooperative guidance and control, shared, traded and supervisory control. Cognition, Technology & Work, 21(4), 555-568.
  • Franz, B., Haccius, J.P., Stelzig-Krombholz, D., Pfromm, M., Kauer, M., & Abendroth, B. (2015). Evaluation of the SAGAT method for highly automated driving. In: Proceedings 19th Triennial Congress of the IEA, Melbourne, Australia, 09.-14.08.2015.
  • Hancock, P. A. (2017). Imposing limits on autonomous systems. Ergonomics, 60(2), 284-291.
  • Hancock, P. A., Jagacinski, R. J., Parasuraman, R., Wickens, C. D., Wilson, G. F., & Kaber, D. B. (2013). Human-automation interaction research: past, present, and future. ergonomics in design, 21(2), 9-14.
  • Itoh, M., Flemisch, F., & Abbink, D. (2016). A hierarchical framework to analyze shared control conflicts between human and machine. IFAC-PapersOnLine, Volume 49, Issue 19, Pages 96-101, ISSN 2405-8963, https://doi.org/10.1016/j.ifacol.2016.10.468.
  • Kaber, D. B. (2018). Issues in Human–Automation Interaction Modeling: Presumptive Aspects of Frameworks of Types and Levels of Automation. Journal of Cognitive Engineering and Decision Making, 12(1), 7–24. https://doi.org/10.1177/1555343417737203.
  • Kurpiers, C., Lechner, D., & Raisch, F. (2019, June). The influence of a gaze direction based attention request to maintain mode awareness. In Proceedings of the 26th International Technical Conference on the Enhanced Safety of Vehicles, Eindhoven, The Netherlands (pp. 10-13).
  • Larsson, P., Johansson, E., Söderman, M., & Thompson, D. (2015). Interaction Design for Communicating System State and Capabilities during Automated Highway Driving. Procedia Manufacturing, Volume 3, Pages 2784-2791, ISSN 2351-9789, https://doi.org/10.1016/j.promfg.2015.07.735.
  • Lee, J. D. (2018). Perspectives on Automotive Automation and Autonomy. Journal of Cognitive Engineering and Decision Making, 12(1), 53–57. https://doi.org/10.1177/1555343417726476.
  • Marberger, C., Mielenz, H., Naujoks, F., Radlmayr, J., Bengler, K., & Wandtner, B. (2017, July). Understanding and applying the concept of “driver availability” in automated driving. In international conference on applied human factors and ergonomics (pp. 595-605). Springer, Cham.
  • Martens, M. H., & van den Beukel, A. P. (2013, October). The road to automated driving: Dual mode and human factors considerations. In 16th International IEEE Conference on Intelligent Transportation Systems (ITSC 2013) (pp. 2262-2267). IEEE
  • Merat, N., Jamson, A. H., Lai, F. C. H., & Carsten, O. (2012). Highly Automated Driving, Secondary Task Performance, and Driver State. Human Factors, 54(5), 762–771. https://doi.org/10.1177/0018720812442087.
  • Merat, N., Seppelt, B., Louw, T., Engström, J., Lee, J. D., Johansson, E., ... & McGehee, D. (2019). The “out-of-the-loop” concept in automated driving: Proposed definition, measures and implications. Cognition, Technology & Work, 21(1), 87-98.
  • Navarro, J., Mars, F., Hoc, J. M., Boisliveau, R., & Vienne, F. (2006). Evaluation of human-machine cooperation applied to lateral control in car driving. In Proc. 16th World Congress of the International Ergonomics Society (pp. 4957-4962).
  • Östlund, J., Peters, B., Thorslund, B., Engström, J., Markkula, G., Keinath, A., Horst, D., Juch, S., Mattes, S., & Foehl, U. (2005). Driving performance assessment – methods and metrics. Deliverable D2.2.5 of the AIDE project. IST-1-507674-IP.
  • Radlmayr, J., Gold, C., Lorenz, L., Farid, M., & Bengler, K. (2014, September). How traffic situations and non-driving related tasks affect the take-over quality in highly automated driving. In Proceedings of the human factors and ergonomics society annual meeting (Vol. 58, No. 1, pp. 2063-2067). Sage CA: Los Angeles, CA: Sage Publications
  • SAE,Society of Automotive Engineers. (2018). Taxonomy and definitions for terms related to driving automation systems for on-road motor vehicles. J3016 JUN2018. SAE International: Warrendale, PA, USA.
  • Solís-Marcos, I., Ahlström, C., & Kircher, K. (2018). Performance of an additional task during Level 2 automated driving: An on-road study comparing drivers with and without experience with partial automation. Human factors, 60(6), 778-792.
  • Trösterer, S., Meschtscherjakov, A., Mirnig, A. G., Lupp, A., Gärtner, M., McGee, F., ... & Engel, T. (2017, September). What we can learn from pilots for handovers and (de)skilling in semi-autonomous driving: An interview study. In Proceedings of the 9th International Conference on Automotive User Interfaces and Interactive Vehicular Applications (pp. 173-182).
  • Victor, T. W., Tivesten, E., Gustavsson, P., Johansson, J., Sangberg, F., & Ljung Aust, M. (2018). Automation expectation mismatch: Incorrect prediction despite eyes on threat and hands on wheel. Human factors, 60(8), 1095-1116.
  • Endsley, M. R. (2006). Situation awareness. Handbook of human factors and ergonomics, 3, 528-542.
  • Flemisch, F., Heesen, M., Hesse, T., Kelsch, J., Schieben, A., & Beller, J. (2012). Towards a dynamic balance between humans and automation: authority, ability, responsibility and control in shared and cooperative control situations. Cognition, Technology & Work, 14(1), 3-18.
  • Banks, V. A., Plant, K. L., & Stanton, N. A. (2018). Driver error or designer error: Using the Perceptual Cycle Model to explore the circumstances surrounding the fatal Tesla crash on 7th May 2016. Safety science, 108, 278-285.
  • Norman, D. A. (1983). Some observations on mental models. Mental models7(112), 7-14
  • Lee, J. D., & See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human factors, 46(1), 50-80.
  • Parasuraman, A. (1997). Reflections on gaining competitive advantage through customer value. Journal of the Academy of marketing Science, 25(2), 154-161.
  • Wickens, C.D., Lee, J., Liu, Y., & Gordon-Becker, S. (2003) Introduction to Human Factors Engineering. 2nd edition. Pearson, USA.
  • Cleij, D., Bakker, B., Borowsky, A., Christoph, M., Fiorentino, A., van Grondelle, E., & van Nes, N. (2020). Mediator System and Functional Requirements, Deliverable D1.4 of the H2020 project MEDIATOR. In preparation.
  • Hoc, J. M., Young, M. S. & Blosseville, J-M. (2009). Cooperation between drivers and automation: implications for safety. Theoretical Issues in Ergonomics Science, 10(2), 135-160.

 Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from permissions@acm.org. AutomotiveUI '21, September 09-14, 2021, Leeds, United Kingdom

AutomotiveUI '21, September 09–14, 2021, Leeds, United Kingdom

ACM ISBN 978-1-4503-8063-8/21/09...$
DOI: https://doi.org/10.1145/3409118.3475153