Teaming with Your Car: Redefining the Driver–Automation Relationship in Highly Automated Vehicles

Coordination

Coordination is arranging the timings of tasks and resources to enhance the performance of team members (Table 2). The arrangement involves ordering activities and allocating resources so their executions are aligned to seamlessly support common goals. Organization management research suggests teammates need to serve specific roles at a particular location, time, and order during coordination (O’Brien, 1968). This asserts that members of an organization develop a formal relationship where they practice previously agreed-upon and clearly defined roles and responsibilities (McNamara, 2012). On the other hand, the definitions from human teams and human–automation research focus more on the arrangement of resources, tasks, and activities rather than the formality of the relationship. Human teams research defines coordination as the management of interdependencies between goals, tasks, and resources of various human agents (Malone & Crowston, 1990). Previous studies on coordination in human–automation interaction stressed the alignment of activities between agents. That is, coordination is performed by meta-operations to organize subordinate operations (Hoc, 2001). Also, coordination refers to a process of synchronizing the interdependent activities of humans and machines (Sarter, 2002).

OPEN IN VIEWER

Table 2:

Definitions of Coordination (Emphasis Added to the Original)

Authors (Year)	Characteristic	Definition
Malone and Crowston (1990)	Managing interdependency	“The act of managing interdependencies between activities performed to achieve a goal” (p. 361).
Hoc (2001)	Organizing operations	“Coordination is performed by meta-operations that apply to operations to organize them so that they fulfill certain criteria” (p. 513).
Sarter (2002)	Synchronizing interdependent activities	“Coordination is the process of synchronizing the interdependent activities of various agents. Three important components of coordination are: (a) the management of simultaneity constraints, (b) effective resource allocation, and (c) the communication of intermediate results and difficulties with task performance (Malone & Crowston, 1990)” (p. 20).
Wilson et al. (2007)	Performance orchestration
	Correct and timely actions and contributions	“… coordination is contingent on correct and timely actions and contributions by all team members. Through appropriate coordinating mechanisms, team members are able to sequence, synchronize, integrate, and complete tasks without wasting valuable resources. Therefore, coordination is the behavioral mechanism team members use to orchestrate their performance requirements” (p. 249).
McNamara (2012)	Assistance to achieve organizational goals	“[Coordination] is defined as an interaction between participants in which formal linkages are mobilized because some assistance from others is needed to achieve organizational goals” (p. 391).
Chiou and Lee (2021)	Timing or arrangement of joint decisions	“Therefore, coordination is about the timing or arrangement of joint decisions…” (p. 10).

In driver–ADS teaming, coordination can be considered an arrangement of resources for operational driving tasks and task interdependency between the driver and ADS. Transfer of controls (TOC) in SAE Level 2–3 vehicles can be an example of coordination. For a successful TOC, both the driver and ADS need to align their actions (e.g., hand over by ADS and take over by the driver in a time budget—the time between a takeover request and system limit) without conflicts.

Cooperation

Cooperation uses negotiation in addition to coordination to resolve conflicts that arise between teammates’ individual goals and the joint goal. Conflicts require teammates to adjust their individual goals, thereby incurring costs to the goals. Accordingly, cooperation involves attitudinal and social processes such as positive attitudes toward collaboration (e.g., disposition to collaborate) as well as attitudes toward others (e.g., trust) that motivate teammates to consider the team goals over individual goals, despite the cost.

We found three key characteristics of cooperation from existing definitions (Table 3). First, cooperation involves the pursuit of common goals (Flemisch et al., 2014; Skjerve & Skraaning, 2004). Success in achieving the common goal depends on individual teammates’ goals, intentions, and behaviors (Biester, 2009; Chiou & Lee, 2021; Pellicone et al., 2019). Second, a negotiation process is required to resolve conflicts between teammates’ individual goals. The teammates’ individual goals do not always align completely with the common goals (Pellicone et al., 2019) and may even conflict (Chiou & Lee, 2021). The conflict or misalignment of individual goals needs to be resolved through negotiation (Chiou & Lee, 2021) by adjusting teammates’ activities according to the common goals (Skjerve & Skraaning, 2004; Vanderhaegen, 2012). This negotiation process is called various names, such as the “co-ordinated pursuit” (Schulte et al., 2009), “exchange” (Biester, 2009), and “negotiation” (Chiou & Lee, 2021; Biester, 2009). Third, as teammates cooperate, they develop predispositions to work with others in the future. For example, Chiou and Lee (2021) mentioned that social processes like trust increase in importance as teams cooperate. Similarly, Bedwell et al. (2012) stated that cooperation shapes teammates’ inclination to collaborate.

OPEN IN VIEWER

Table 3:

Definitions of Cooperation (Emphasis Added to the Original)

Authors (Year)	Characteristic	Definition
Skjerve and Skraaning (2004)	Aspiration	“To engage in cooperative activity thus essentially implicates that each of the parties involved has decided to join with the others based on the aspiration to achieve something in common” (p. 651).
	Achievement of something in common
Hoc (2001), Millot (2009), Millot and Debernard (2007)	Interference management	“Two agents are in a cooperative situation if they meet two minimal conditions. (1) Each one strives towards goals and can interfere with the other on goals, resources, procedures, etc. (2) Each one tries to manage the interference to facilitate the individual activities and/or the common task when it exists” (Millot, 2009, p. 7).
	Common task
Biester (2009)	Exchange process	“Cooperation is defined as a continuous exchange process between driver and automobile, with the exchange process creating a common knowledge base and goals, which in turn ensure a transferability of experiences by means of educability” (p. 11).“Keeping in mind the goal of an optimal realization of driving tasks, both participants are responsible for certain parts of the commonly defined whole task, which is the result of a continuous coordination process” (p. 11).
	Commonly defined whole task
	Coordination
Schulte et al. (2009)	Coordination	“Human-machine and machine–machine co-operation, i.e., to achieve the common top-level goals of teaming by a co-ordinated pursuit of the common mission objective” (p. 72).
	Common mission objective
Vanderhaegen (2012)	Mutual demand	“However, cooperation between agents implies possible mutual cooperation demands when the agents want to cooperate to facilitate their own activities. As a matter of fact, the agents of a given human-machine system cooperate together in order to facilitate the activities of other human-machine systems and/or the activity of their own group” (p. 62).
	Desire (“want”) to cooperate
Bedwell et al. (2012)	Attitudinal construct	“… we define cooperation as an attitudinal construct that helps to facilitate the process of collaboration. Salas, Sims, and Burke (2005) describe a construct known as team orientation, which, in essence, is an individual’s propensity to consider other’s behavior when interacting in a group and the belief in the importance of the team’s goals over individual goals. In other words, team orientation describes the extent to which individuals are predisposed to work well with others; the extent to which they are cooperative” (p. 136).
	Predisposition to work with others
McNamara (2012)	Work together within existing structures and policies	“… cooperation is defined as an interaction between participants with capabilities to accomplish organizational goals but chose to work together, within existing structures and policies, to serve individual interests” (p. 391).
	Individual interests
Flemisch et al. (2014)	Common goal	“Cooperative (human-computer) guidance and control of vehicles (in short: cooperative control) is understood in this article as the action or process of working together of at least one human and at least one computer on the guidance and control of at least one vehicle. Cooperative (guidance and) control can also make use of cooperative technical systems, such as vehicles and infrastructure that communicate and work together towards a common goal (e.g., Car2Car or Car2Infrastructure)” (p. 345).
Chiou and Lee (2021)	Negotiation and alignment of individual goals	“… cooperation is about negotiating and aligning individual goals when they differ from the joint goal” (p. 10).
	Joint goal

The main difference between coordination and cooperation is the degree of goal conflict between agents. While cooperation accounts for a conflict between agents and negotiations to resolve it, coordination assumes situations where there is no conflict between individual agents’ goals and stresses the arrangement of joint decisions (Chiou & Lee, 2021). Also, compared to coordination, cooperation is more suitable to frame driver–ADS teaming with a longer time constant (e.g., maneuver-level driving tasks in Michon, 1985).

Cooperation can be applied to driver–ADS teaming where team goals and individual goals are specified, and potential conflicts exist. Like the driver may have “explicit goals” and “implicit goals,” the automation contains “pre-programmed goals” as well (Linegang et al., 2006). An example of cooperation in driver–ADS teaming is speed negotiation between two agents. A driver’s goal of arriving at the destination as early as possible may contradict the ADS’s goal of safe driving (i.e., maintaining speed below the posted speed limit). To settle the conflict, the driver can either approve or reject the ADS’s request to lower the set speed for automated driving. Other examples of cooperation between the driver and ADS in the literature include negotiating conflicts between ADS’ ideal route and the driver’s preferred route (Payyanadan & Lee, 2019; Weßel et al., 2017) and driver and ADS communicating their judgment for the next maneuver and negotiating a maneuver when the judgments conflicts (Guo et al., 2019).

Collaboration

Collaboration is often used interchangeably with cooperation, but collaboration is an ongoing participatory process in which all teammates act with comparable authority and make many decisions jointly over an extended time (Table 4). As a result of collaboration, teammates develop shared rules, norms, and agreements, especially for undefined situations.

OPEN IN VIEWER

Table 4:

Definitions of Collaboration (Emphasis Added to the Original)

Authors (Year)	Characteristic	Definition
Wood and Gray (1991)	Autonomous stakeholder	“Collaboration occurs when a group of autonomous stakeholders of a problem domain engage in an interactive process, using shared rules, norms, and structures to act or decide on issues related to that domain” (p. 146).
	Interactive process
	Shared rules, norms, and structures to act or decide
Fong et al. (1999, 2001)	Peers who work together	“In this new collaborative control model, a human operator and robot are peers who work together, collaborating to perform tasks and to achieve common goals. Instead of a supervisor dictating to a subordinate, the human and the robot engage in constructive dialogue to exchange their ideas and resolve their differences” (Fong et al., 1999, p. 34).
	Common goal
	Dialogue exchange
Rule and Forlizzi (2012)	Dialogue exchange	“Collaborative control was defined as humans and robots engaging in dialog to exchange ideas, resolve differences, perform tasks, and achieve goals. Here, human and robot are seen as equals in collaboration; the human is an imprecise, limited source of information for the robot” (p. 98).
	Equals in collaboration
McNamara (2012)	Shared interests and a collective responsibility	“… collaboration is defined as an interaction between participants who work together to pursue complex goals based on shared interests and a collective responsibility for interconnected tasks which cannot be accomplished individually” (p. 391).
Bedwell et al. (2012)	Evolving process	“We define collaboration as an evolving process whereby two or more social entities actively and reciprocally engage in joint activities aimed at achieving at least one shared goal” (p. 130).
	Social entities
	Shared goal
Mainprice et al. (2014), Phillips-Grafflin et al. (2016)	Peer of the operator	“[Collaborative control] corresponds to high-level supervision where the robot is considered to be a peer of the operator. The role of the operator shifts from an operator who dictates every movement, to a supervisor who guides at a high-level. This approach is often used for unmanned systems controlled from a central command post” (Phillips-Grafflin et al., 2016, p. 344).
Kridalukmana et al. (2020)	Independent SA	“Not only does the ADAS (advanced driver assistance systems) provide cognitive support for drivers, but it also has been granted more access to perform driving tasks, including taking over manual control as a backup in case of emergency. Thus, such an intelligent system in ADAS is transforming into an autonomous agent having a high level of autonomy. Using this perspective, driving becomes a multiple-agent-based activity involving two SA (situation awareness) processes (from the driver and ADAS), which can be described as collaborative driving” (p. 2).

In collaboration, automation is considered as a peer (Fong et al., 1999, 2001; Mainprice et al., 2014; Phillips-Grafflin et al., 2016) or equal (Rule & Forlizzi, 2012) that requires only minimum supervision from its human partners (Mainprice et al., 2014; Phillips-Grafflin et al., 2016). Moreover, automation engages in active communication and coordination with human partners to exchange ideas, resolve differences, perform tasks, and achieve goals (Fong et al., 1999; 2001; Rule & Forlizzi, 2012). This communication supports adaptive behaviors; accommodates human operators of various skills, knowledge, and experience (Fong et al., 2001); and coordinates team members (Hoffman & Breazeal, 2007). As a result, collaborating teams form a participatory structure in which teammates make decisions jointly (McNamara, 2012). In collaboration, teammates share rules, norms, and agreements (shared values) for current and future situations. Then, the shared values reorient teammates’ actions, processes, and decisions toward the task that brought them together (McNamara, 2012; Wood & Gray, 1991). For example, in driver–ADS teaming, functional purposes of driving, such as efficient transport, safe transport, or present driving experience, can be the values shared by the driver and ADS. Each agent may have conflicting values (e.g., speeding for efficiency vs. following the posted speed limit for safety) but through collaboration they can share and agree on the values.

Both cooperation and collaboration address individual goals, team goals, and conflict; however, cooperation is more focused on parallel and deterministic goals, whereas collaboration is more about shared and dynamic goals, which require active communication and negotiations for goal adjustment. In addition, compared to cooperation, collaboration can be considered a long-term relationship between agents.

In driver–ADS teaming, collaboration requires a lateral control structure. The ADS will serve increasing roles (e.g., a co-driver) with involvement in not only vehicle control tasks but also in driver management (e.g., monitoring driver state and intervening if needed) and decision-making throughout a trip. Repeated interactions between the driver and ADS over time will create a continuing relationship and lead to horizontal collaboration. Collaboration in driver–ADS teaming suggests proactive and responsive driving automation in SAE Levels 2 and 3, where ADS acts as a co-driver. Proactive and responsive automation require advanced means of communication to allow drivers to take advantage of the intelligent teammate during decision-making and goal-setting processes. Through communication, the driver–ADS team can set driving goals jointly, restructure the ways teammates act, and make decisions based on the new goals to respond to undefined and untrained road events. Accordingly, communication between drivers and ADS will become essential for the driver–ADS collaboration. Currently, the human–machine interface (HMI) is considered as the primary medium for driver–vehicle interaction and some interactions rely on implicit communication channels (e.g., inferring based on a teammate’s behavior); however, driver–ADS collaboration will require active and bi-directional communication channels.

The panarchy of coordination, cooperation, and collaboration

Coordination, cooperation, and collaboration (3Cs) are highly interrelated and distinct components of driver–ADS teaming in SAE Levels 2 and 3. Based on the findings, we propose a panarchy framework of coordination, cooperation, and collaboration, in which the 3Cs are connected by adaptive cycles of different time scales (Figure 2). Panarchy is a conceptual model of a complex system in which its components are dynamically organized and structured across different time and spatial scales (Holling, 2001). Imagine a gear train with different sizes of gears rotating at different speeds (but they work together as a single entity). Similarly, the panarchy framework was initially proposed to explain how adaptive cycles in ecosystems or socio-ecological systems interact across different time scales to constitute a resilient system (Holling, 2001). Resilience refers to a complex system’s vulnerability to unexpected threats and its capability to adapt continuously to the dynamically changing environment (Chiou & Lee, 2021; Holling, 2001). In driver–ADS teaming, resilience implies the adaptability of the driver–ADS team to accommodate unexpected road or in-vehicle situations. Similar to our understanding of the relationship between 3Cs, a panarchy is comprised of faster and smaller-scale cycles that provide thrusts and innovations to slower and larger-scale cycles. In this structure, influence can travel both from bottom to top and top to bottom, representing how a complex system reacts to disruptions.

OPEN IN VIEWER

Figure 2.

The panarchy of coordination, cooperation, and collaboration. This framework posits that the 3Cs are highly interrelated but distinct concepts that operate at different granular levels to support driver–ADS teamwork.

In our framework, the 3C panarchy contains two adaptive cycles moderated by the 3Cs of teamwork: the smaller and faster coordination-cooperation cycle (lower cycle) and the larger and slower cooperation-collaboration cycle (upper cycle). In this framework, the accumulated interactions in a shorter time scale (lower cycles) shape longer-term relationships (upper cycle), while longer-term relationships guide short-term interactions. The lower cycle starts with coordinating the interdependent driving resources, tasks, and control authorities between the driver and ADS. Coordination activities comprise basic action units for the higher-level construct of cooperation. Moreover, by leveraging implicit and explicit communication, coordination enables the prediction of teammates’ needs that support cooperation (Salas et al., 2005). In turn, cooperation serves to align individual team members’ goals with the team’s driving goal. At the cooperation level, the activities become more social in nature; thus, cooperation shapes the members’ propensity to collaborate in the longer-term (Bedwell et al., 2012). Specifically, team members develop trust in each other that is updated as a function of iterations of cooperation and coordination (Chiou & Lee, 2021). From here, cooperation can influence coordination via the back loop by updating the criteria that govern coordination (passed down from collaboration via the back loop). In addition, cooperation affects how team members communicate and enables or interferes with deliberate coordination (Salas et al., 2008). Cooperation can also influence collaboration via the front loop and the slower upper cycle by affecting the driver’s propensity to collaborate with teammates (Bedwell et al., 2012). At the collaboration level, the driver–ADS relationship becomes horizontal, and teammates are assigned less clearly defined roles (Salas et al., 2005). Teams transition toward the participatory structure, where teammates engage in iterations of interactions (coordination and cooperation). The team’s social processes (e.g., trust, disposition to collaborate) evolve over iterations and guide team members’ behaviors (Chiou & Lee, 2021; McNamara, 2012) and teammates share values (Wood & Gray, 1991), such as driving priorities and goals. If the teams confront surprises, teams refer to the shared values to select new goals among multiple alternative goals to adapt to the new environment. Lastly, collaboration disseminates criteria or rules, updated according to the new goals, via the back loop to reorganize coordination and cooperation.

The 3C panarchy framework shows how to achieve resilient driver–ADS collaboration, where the driver–ADS team can have joint adaptation to the road environment more generally (e.g., shaping behavior to be safer and more efficient). Driving a vehicle with SAE Levels 2 and 3 functionalities will face edge cases in traffic situations, which drivers and ADS are not trained to handle. However, increasing ADS’s automation level can only address a part of the uncertainties in the driving environment. Therefore, it will become crucial to design the driver–ADS team to be capable of adjusting its goals and tuning the collaboration process in response to the road environment. The 3C panarchy framework provides insights into how to achieve resilient driver–ADS teaming. The 3C panarchy suggests designers consider interactions across time and the relationship between the driver and the ADS that is more than a series of independent events. Based on this framework, we conducted the second literature review with a focus on driver–ADS teaming design.

Unilateral communication to mutual communication

Problem summary: Mutual communication is essential for coordination, cooperation, and collaboration to share goals and activities of each teammate; however, in current driving automation applications, communication channels are often unilateral or fragmented.

Rationale: Teammates must infer and predict their partners’ behaviors and coordinate efficiently. In driving automation, the driver and ADS rely on different sensory mechanisms to collect information. Therefore, without mutual communication, they are likely to arrive at different assessments of the same situation (Endsley, 2017; Walch et al., 2017). Moreover, unilateral communication does not support conflict negotiations and goal adjustment, which are requirements in cooperation and collaboration.

Solution: Implement mutual communication channels between the driver and ADS. The mutual communication channel can allow drivers’ proactive information requests (e.g., a driver requests information about the automation status when needed) instead of passively receiving information shared by the system. Similarly, the automation can ask driver’s intention or status update instead of only relying on sensor-based estimation. The mutual communication channel should support continuous bilateral interaction with the automation over time and this communication channel will shape long-term relationship of the driver–ADS team and lead to driver–ADS collaboration.

Usage: Through the mutual communication system, the driver should be able to notify potential failure of the assigned role (e.g., degraded monitoring performance due to fatigue); similarly, the automation system can notify the driver regarding any impending failure of the system. Mutual understanding through mutual communication supports understanding and predicting the intention and behavior of human drivers and ADS. Furthermore, this leads ADS to adapt its behavior based on mutual understanding and increases acceptance by the human driver (Xing et al., 2020; 2021).

Source: Team communication (Cooke et al., 2020; Demir et al., 2017; Mosier et al., 2017; Xing et al., 2020); mutual understanding (Chen et al., 2018; Xing et al., 2020; 2021); maintaining common ground (Klein et al., 2004); shared mental state (Cohen et al., 1997; Mosier et al., 2017); shared situation awareness (Cohen et al., 1997; Endsley, 2017; Xing et al., 2020); and shared situation representation (Walch et al., 2017).

Opaque and brittle automation to observable and directable automation

Problem summary: For the driver–ADS team, the driver should be able to see what the ADS is doing, what it will do, and the reasons for its intention and action (e.g., Christoffersen & Woods, 2002). The driver also needs to be able to redirect the ADS behaviors as needed.

Rationale: Teammates’ actions are intertwined such that one’s actions elicit the other’s response (Bratman, 1992). Therefore, observability and directability are deemed as two fundamental characteristics for supporting cooperative and collaborative driver–ADS teams. Observability can refer to the ability to observe the automation’s behaviors, intentions, and expectations over time. Observability is essential for coordination and more broadly for teaming because coordination facilitated by observability supports team members to achieve a shared goal, especially for cooperation and collaboration (Hoffman & Breazeal, 2004; Skjerve & Skraaning, 2004). Directability can refer to the ability to influence automation’s activities (Christoffersen & Woods, 2002), and it is highly associated with observability.

Solution: Enhance observability and directability of driving automation by delivering automation’s situation awareness to the driver and allowing options to influence automation’s behavior. This helps drivers assess and predict automation’s actions in order to plan their own (Christoffersen & Woods, 2002; Klein et al., 2004; Walch et al., 2017). Although the word observability implies a specific modality, visual, it does not have to be implemented using visual modality. For example, drivers can “feel” ADS’s intention and action using haptic shared control (e.g., haptic gas pedal or haptic steering wheel; Abbink et al., 2012). As the driver–ADS team evolves to collaboration after repeated interactions over time, observability and directability can be applied to not only the automation but also the driver (i.e., making the driver observable and directable for the automation), and this will support a responsive automation.

Usage: For example, Tesla vehicles display what they see and sense through displays (e.g., surrounded cars and traffic lights). This visualizes automation’s perception and helps drivers to understand the vehicle’s behavior and coordinate accordingly. Similarly, GM’s Robotaxi displays what it sees and senses along with its intention (e.g., projected path). Regarding directability, GM’s Super Cruise allows continuous and partial takeover instead of discrete and full takeover. During automated driving, drivers can adjust lateral control, longitudinal control, or both, instead of fully disengaging from the automated driving. This feature enhances directability of the system and promotes instant shared control.

Source: Shared goal (Cohen et al., 1997; Hoffman & Breazeal, 2004; Skjerve & Skraaning, 2004; Xing et al., 2020); team adaptability (Mosier et al., 2017); mutual support (Bratman, 1992); observability and directability (Christoffersen & Woods, 2002; Johnson et al., 2014; Klein et al., 2004).

Subordinate automation to responsive automation

Problem summary: Currently deployed driving automations serve as a tool for humans or a subordinate teammate, thus limiting collaborative aspects of the driver–ADS team. This will prevent teams from transitioning to a participatory structure in which drivers and ADS share values and jointly take actions to respond to unexpected threats on the road.

Rationale: When the driver–ADS relationship becomes horizontal and collaborative, its observant and directive aspects should be considered. Observant and directive aspects lead to responsive automation that complements observable and directable automation.

Solution: Expand the scope of driver–ADS team to collaboration and apply the responsive automation concept to design ADS. In driving automation, responsive automation will require advanced capabilities (e.g., intelligence), but it does not have to be associated with higher SAE automation levels (i.e., Levels 4 and 5); it is rather more suitable for SAE Levels 2 and 3, at which teaming is essential. Responsive automation is collegial and a more proactive system (Chiou & Lee, 2021). Applying the responsive automation enables ADS to behave as an equal teammate and be engaged in active communications with drivers. This will lead a driver–ADS team that is more resilient and adaptable to unexpected traffic situations by allowing collaborative driving between the driver and ADS. Consequently, this will allow drivers to use partially automated vehicles (SAE Levels 2 and 3) even beyond the current operational design domain (e.g., complex urban driving area or unmapped area).

Usage: The observant and directive ADS can play a more proactive role by monitoring the driver’s state, reading the driver’s intention, and even changing the driver’s intention to achieve a shared goal (e.g., promoting use of automated driving when the ADS is more capable of controlling the vehicle) for collaboration. Due to the characteristics of collaboration, within this scope, potential conflicts or even disagreements between the driver and the ADS may occur, such as speed choice (driving below the posted speed limit as programmed for the automation vs. driving 9–10 mile/h above the posted speed limit preferred by the driver). Goal alignment and negotiation to support efficient and safe driving will become a key challenge.

Source: Mutual responsiveness (Bratman, 1992); mutual performance monitoring (Mosier et al., 2017); mutual directable (Klein et al., 2004); mutual predictable in the action (Klein et al., 2004), mutual predictability (Walch et al., 2017); and adjustable autonomy (Calhoun et al., 2018; Myers & Morley, 2001, 2003; Walch et al., 2017).

Binary function allocation to dynamic interdependency

Problem summary: Binary function allocation between ADS and the human driver may support simple coordination between the driver and ADS but does not properly support a collaborative driver–ADS team that requires adaptation to a changing environment over time.

Rationale: The design of driver–ADS teams should consider not only their tasks allocation but also their interdependent relationship. Interdependence refers to how two individuals influence each other’s outcomes through their interactions, and it depends on interpersonal properties of situations and individual responses to the properties (Kelley et al., 2003a). Each person in an interdependent relationship is required to coordinate their actions and communication, which will facilitate their actions in the course of their interaction. However, identifying abstract patterns of interdependence can be less obvious when we consider a HAT instead of a human–human team. For example, changing lanes in a SAE Level 2 automated vehicle can be considered as solely ADS’s task and responsibility (i.e., substitution for the human driver) with minimum engagement of the human driver (i.e., initiating lane change request); however, the driver in that situation is more responsible than just initiating the lane change request and still needs to work together with the ADS by complementing dependency of the automation (e.g., monitor objects or situation that the ADS may fail to detect and supervise lane change behavior).

Solution: The relationship between the human driver and the ADS has to be specified beyond identifying who is driving (e.g., you are driving vs. you are not driving in SAE J3016). The dynamic interdependency between the driver and ADS across different time scales needs to be identified and supported for designing collaborative driver–ADS teams.

Usage: Coactive design defines the interdependency in a joint system based on dependencies of each agent (Johnson et al., 2014, 2018). The coactive design stresses the reciprocal and mutually constraining nature of actions and guides interdependence analysis that focuses on core interdependent relationships—observability, predictability, and directability— and provides information about what needs to be observable, predictable, and directable to designers, including when situation awareness is adequately supported and when it is not. Applying coactive design to the driver–ADS team can define driver–ADS relationships, which vary by time and situation, and identify each agent’s roles across various tasks and situations. Analysis of interdependence in the driver–ADS team can provide the basis for understanding complex coordination in the team and support designing collaborative driver–ADS teaming.

Source: Interpersonal situations and interdependence (Kelley et al., 2003b, 2003c); coactive design (Johnson et al., 2014; 2018).

When and where to apply driver–ADS teaming concept?

Although driver–ADS teaming is a promising concept, it may not be the best for every driving task and level of automation. In terms of levels of automation, conditional automation stages (e.g., SAE Levels 2 and 3) will be where the driver–ADS teaming is needed the most. Within the same level of automation, driving tasks can be further specified using time scales (note that while the 3C panarchy addresses a broader spectrum of time scale such as a long-term relationship between the driver and ADS for even over years, this section addresses time scales during vehicle operation from milliseconds to hours). Michon (1985) introduced three levels of driving tasks that have different time scales: strategic-level tasks, maneuver-level tasks, and control-level tasks. Table 6 listed forms of ADS supports and relevant teaming concepts across Michon’s three levels of driving tasks and time spans based on two research (Flemisch et al., 2019; Weyer et al., 2015). Weyer et al. (2015) introduced a typology of driving assistance system with Michon’s three levels of driving tasks and implied that maneuver-level tasks, in which more frequent interventions by the human driver are expected in partially automated vehicles, are more suitable for driver–ADS cooperation. Flemisch et al. (2019) provided a different view compared to Weyer et al. (2015) in terms of the scope of driver–ADS teaming. They suggested that the scope of the human–machine teaming can be expanded from high-level cooperational metacommunication to low-level operational tasks (Flemisch et al., 2019). They further specified the teaming concept and compared relationships between “shared control,” “shared and cooperative guidance and control,” and “human–machine cooperation.” From their view, shared control can be applied across strategic-level, maneuver-level, and control-level tasks; shared and cooperative guidance and control can be applied between maneuver-level and control-level tasks; human–machine cooperation can be applied across all three levels but is more applicable for higher-level tasks (i.e., strategic-level tasks). Based on our understanding of coordination, cooperation, and collaboration in driver–ADS teams, we concur with this view that the teaming concept can be applied beneficially from shorter time spans (milliseconds to few seconds) to longer time spans (minutes to hours). In driving, strategic-level teaming has not yet been fully explored and may bring benefits of applying the teaming concept. For example, in using haptic shared control (e.g., haptic gas pedal or haptic steering wheel; Abbink et al., 2012), the driver and the ADS can control the vehicle together for a longer time span (e.g., through an entire trip) while feeling and knowing other teammate’s behavior and intention (as a way for communication and negotiation). Similarly, in the control-level driving task, ADS’s role was mainly considered as an intervention (e.g., evasive maneuver initiated by the ADS) in (Weyer et al., 2015). However, driver–ADS teaming concept, specifically coordination, can be applicable at control-level driving tasks using shared control. For example, in vehicles with SAE Level 2 and 3 functionalities, the driver and ADS can transfer the control of the vehicle. Most previous research considered the transfer of control to be a discrete process, but shared control can allow continuous transitions (i.e., arrangement of joint decision—control transfer) and even provide feedback of transition with haptic cue at a shorter time constant (e.g., feeling of transition when the ADS received the authority for control from the driver as a sign of successful transition).

OPEN IN VIEWER

Table 6:

Typology of Driving Assistance Systems and Relevant Teaming Concepts Across Time Spans

Driving Task	Time Span	Support by ADAS (Weyer et al., 2015)	Relevant Teaming Concept (Flemisch et al., 2019)
Strategic-level	Long-term	Recommendation (e.g., route guidance)	Human–machine cooperation (high); shared control (moderate)
Maneuver-level	Medium-term	Cooperation (e.g., adaptive cruise control)	Human–machine cooperation (medium) shared and cooperative guidance and control (moderate); shared control (moderate)
Control-level	Very short-term	Intervention (e.g., automatic emergency braking)	Human–machine cooperation (low); shared and cooperative guidance and control (moderate); shared control (moderate)

Observability, communication, and directability

Observability consists of shared representations of the problem and representations of the activities of other teammates (e.g., what automation is working on, intention about what to do next and reason for a specific decision). Thus, observability supports awareness of the situation and other team members, and it prevents misunderstanding, misbehavior, mode confusion, or an inappropriate level of trust (Pokam et al., 2019). For example, Skjerve and Skraaning (2004) empirically tested a set of HMIs to increase the observability of the autonomous system in nuclear power plants. Observability was implemented using visual and verbal feedback on the autonomous system’s activity. The new HMI with enhanced observability was compared to a conventional HMI, and the result showed that the HMI with observability improved human–automation cooperation quality, as measured by subjective judgments. In driving automation, multiple methods of detecting and inferring drivers’ states have been proposed (Hecht et al., 2018). Providing drivers the information about ADS’s capability in handling a traffic situation has been studied (e.g., Beller et al., 2013; Feldhütter et al., 2018; Holländer & Pfleging, 2018). These efforts mainly focused on achieving observability by providing the ADS’s capability information (e.g., Holländer & Pfleging, 2018) or calibrating trust in ADS (e.g., Helldin et al., 2013). Components such as shared goals, team agreement, and team leadership also contribute to the observability. However, the interaction methods to satisfy these requirements have gained relatively less attention in ADS research.

Supporting mutual understanding and communication will be a key challenge for designing driver–ADS teaming. Multiple research projects support the needs for a medium such as “common ground” or “common workspace” in teaming. Based on the previous research, the common ground can be shared by all members in the joint team activity; includes information of task objectives, task state, and state of other members (Hoffman & Breazeal, 2004); provides information about the environment and other members’ current and future activity (Flemisch et al., 2019); and needs to support team situation awareness (Flemisch et al., 2019). Eventually, common ground will allow each member to understand messages and signals that help cooperate joint actions (Klein et al., 2004) and collaboration. Team situation awareness is considered a significant part of the common ground. Flemisch et al. (2019) mentioned that an “agent that is able to update the common situation awareness may influence other agents by its/his/her own understanding of the situation (validation by others) or may trigger other agents’ reaction (disagreement, explanation asked by others)” (p. 559). Similarly, developing team situation awareness can be considered the most important task for human–automation teaming. Johnson et al. (2014) viewed the common ground from the interdependence between agents and stated that the interdependent relationships will determine what common ground should entail. Thus, building a common ground for efficient coordination and communication will be key for designing driver–ADS teaming.

Communication involves building mutual understanding, a mental model, common ground, or a shared mental state (information sharing) that facilitates human–automation teaming. Moreover, the information sharing mechanism needs to be updated using explicit, implicit, and bi-directional communication (information updating) in accordance with the task environment and goal. In ADS research, few studies tested the bi-directional communication to support driver–ADS teaming. Castellano et al. (2020) implemented the “AutoMate” HMI on a vehicle dashboard, and the system negotiated with drivers on whether to enter a roundabout given a driver’s judgment about the available space. Similarly, Wang et al. (2020) proposed a framework in which drivers negotiated with ADS for the next maneuver using gaze and speech. Lastly, Kuribayashi et al. (2021) tested an augmented reality HMI that communicated the locations of the pedestrians, whose movement the ADS failed to predict, thereby asking for driver input. In this way, drivers were able to leverage their object recognition capability to prevent the ADS from executing an abrupt maneuver (e.g., braking) to avoid the road user if the user’s projected path did not conflict with the vehicle. However, these approaches only cover the communication support for tactical tasks (e.g., avoidance) and leverage visual display (e.g., augmented reality head-up display, dashboard) as main modes of communication.

Directability can be achieved by enabling human teammates to assess and change the partners’ behaviors (Christoffersen & Woods, 2002; Klein et al., 2004; Walch et al., 2017). Such assessment and prediction help teammates understand the consequences of the directions to their partners (Klein et al., 2004) and determine the appropriate direction for a given situation. Therefore, mutual monitoring is an important factor for enhancing directability. Moreover, the development of the adjustable automation teammate should be considered to lower the barrier to directability in driver–ADS teams (Myers & Morley, 2001). Adjustable autonomy enables human operators to adapt their role to the varying cognitive state and lower the barrier to directablity (Calhoun et al., 2018; Myers & Morley, 2001; 2003; Walch et al., 2017). One can imagine a situation in which operators’ workload level is low, as well as their situation awareness. This would delay operators’ reactions to the sudden development of unexpected situations. In such cases, the HAT performance can be enhanced by motivating human operators to engage in more active roles like giving manual directions (e.g., steering). In contrast, there could be a situation in which operators’ workload is high. In this case, giving a higher-level order of predefined actions or delegating most of the roles to automation can compensate for operators’ performance decrease due to high workload. One example of this approach is the flexible automation by the U.S. Air Force, which has shown the potential to improve human–automation teaming by reducing the barrier to directability (Calhoun et al., 2018). This approach provided operators with a wide spectrum of methods for controlling unmanned vehicles, from joystick control to the Playbook, an approach of choosing between sets of predefined tasks (Miller et al., 2005). Therefore, the input method could be selected based on the cognitive state of the human operator, which in turn enhanced the directability of the automation.

Needs for alternative communication channel for the driver–ADS teaming

Carsten and Martens (2019) mentioned HMI is the main communication channel for driver–ADS teaming, meaning HMI is crucial for safe collaboration between the human driver and vehicle-based automated system. However, the issue is that the traditional vehicle HMI is no longer sufficient to support the cooperation or collaboration of the driver and ADS because it requires more cohesive communication, active information sharing (Christoffersen & Woods, 2002), negotiations, and goal adjustments. The lack of information about the ADS due to inadequate HMI also makes it challenging to understand its activities (Skjerve & Skraaning, 2004). Moreover, recent studies suggest interaction methods that depend less on HMIs (e.g., haptic shared control; Abbink et al., 2012). To avoid putting new wine in old bottles (Hollnagel & Woods, 1999), advanced strategies need to be developed to support efficient driver–ADS teaming. This will require expanding the research scope from one-way communication using the traditional vehicle HMI to two-way cooperation design with enhanced observability, key for designing driver–ADS teams.

Regarding the implementation of cooperative control, multiple studies examined shared control, especially haptic shared control. Abbink et al. (2018) described how “in shared control, human(s) and robot(s) are interacting congruently in a perception-action cycle to perform a dynamic task that either the human or the robot could execute individually under ideal circumstance” (p. 511). Flemisch et al. (2003) introduced H(Horse)-Metaphor to describe shared control and transition of authority, along with H-Mode, a haptic-multimodal shared control mode inspired by the H-Metaphor. Characteristics of the H-Metaphor and H-Mode have the potential to solve some of the teaming requirements described in the Driver–ADS Teaming Design Patterns section. The authors wrote that “in H-Mode, two points on the assistance and automation continuum are addressed. In assisted mode (Tight Rein), the automation decreases its influence, i.e., it only acts with low forces at the haptic interface. In the highly automated mode (Loose Rein) automation is the main actor in the vehicle guidance and control, while the driver is still connected with the vehicle and the automation (e.g., by a haptic coupling with an active interface)” (Flemisch et al., 2014, p. 15–16). This will allow continuous and dynamic distribution of control, and both the driver and the automation can initiate transition of control and closely cooperate to achieve the shared goal (Flemisch et al., 2008; 2014). Moreover, haptic shared control allows smooth shifts in control authority through physical interaction. For example, by adjusting the resistance of the steering wheel, any level of automation between full manual control (i.e., no guiding force) and full automation (i.e., very large guiding force) can be manipulated. This automation continuum provides continuous shifts instead of on/off binary conditions. With the haptic steering wheel, the driver can vary their responses to guiding forces by ignoring them, resisting them, or amplifying them (Abbink et al., 2012). More importantly, in terms of designing driver–ADS teaming, this feature enables blending communication into controlling—that is, haptic shared control allows communication with the automation by controlling at the same time. Also, Flemisch et al. (2008) noted three advantages of haptic shared feedback compared to other feedback channels: haptic feedback can be directly linked to the actuator, where the driver reaction is required (e.g., lane departure warning vibration on the steering wheel); haptic feedback can be used to indicate required reaction type (e.g., “steering wheel can be turned into the correct direction by the automation to trigger a steering reaction of the driver”); and haptic interaction is bi-directional (i.e., continuous communication between the driver and the vehicle). The effectiveness of the haptic shared control was supported by empirical studies. Flemisch et al. (2008) showed that haptic shared feedback enhanced driving performance, reduced driver workload, and increased acceptance of the automation system. Likewise, Petermeijer et al. (2015) showed that haptic shared control mitigates the loss of situation awareness and skills as the driver is continuously involved in the vehicle control loop.

However, there are limitations of haptic shared control. Johns et al. (2016) experimented with haptic shared control with a pair of participants (one served as a driver and the other served as a driving agent). The results showed that drivers were able to interpret the intention of the driving agent through steering wheel torque for simple trajectory intentions but not for higher-level ideas such as availability. The authors suggested that another communication channel (visual, auditory, or other forms of haptic feedback) may support high-level communication. Additionally, Abbink et al. (2018) mentioned that shared control is not the best choice for every human–robot (or human–automation) interaction. For some situations or tasks, traded control can be a better option. In the driving domain, based on automation levels, driving situations, and other factors, traded control can be preferred. However, for a situation where the expected time budget—the time between a takeover request and ADS limit—is short or the automation requires frequent intervention by the human driver, shared control would serve better.