Automated driving is a rapidly evolving technology with the aim to revolutionize transportation by eliminating the need for human drivers. The predicted potentials of automated driving include increased driving comfort and safety, traffic flow optimization, reduced emissions, and enhanced mobility for people with disabilities or limited access to transportation. A common framework to classify different levels of automation is the SAE standard J3016 [1]. The framework categorizes the degree of automation into six levels, starting with Level 0, representing manual driving with no active assistance systems, up to fully automated driving without the need for a driver in Level 5. The introduction of active driver assistance systems up to Level 2 in the past decades has continuously increased the safety of passengers in critical driving situations [2]. Although Level 2 systems are already actively controlling the longitudinal and lateral vehicle guidance simultaneously, the driver is still obligated to monitor the system. The next step towards fully automated driving is, therefore, Level 3 (L3), also described as conditionally automated driving (CAD). L3 is the first level in which the driver is not required to continuously monitor the automated system under specific conditions, enabling users to spend their time with non-driving-related activities.

With the shift of responsibility to the automated system, the role of the driver also changes to that of a passenger [3,4], raising the question of how users want to be driven when they are no longer in control of the driving task. Do users prefer their own or a different driving style in automated driving? Only a few studies have started to investigate this question in recent years with the majority using driving simulators. A common approach for investigating user preferences is to let users experience different driving styles in comparison. The driving styles are either predefined or a prerecording of the user’s own driving style. Refs. [5,6] found that both defensive and aggressive drivers prefer a defensive style. Ref. [7] comes to the same conclusion for defensive drivers, but found no tendency for aggressive drivers. Refs. [8,9] analyzed driving style preferences in relation to participants’ prerecorded own driving style without considering driver types. Whereas participants preferred their own driving style in [8], a more defensive style compared to participants’ personal driving style was preferred in [9] for all driving situations except for lane change maneuvers, for which a more aggressive style was preferred. Refs. [3,10,11] analyzed driving style preferences for lane changes on motorways. Without considering the driver type or recordings of participants’ driving styles, all studies found that participants preferred a defensive driving style.

The presented automated driving styles differ in characteristics of driving dynamics and driving strategies with regard to interaction with other road users and the environment. Under the assumption that drivers will expect their AV to drive like a human, many research papers focus on analyzing [12,13,14] and learning human-like driving behavior for AVs [15,16,17,18,19]. The goal of human-like driving behavior is to include human driving conventions to increase ride comfort and safety as well as predictability for other road users [20]. Human-like features in AVs, including driving behavior, are also studied in many papers under the term anthropomorphism [21,22].

Whether a driving style is preferred for automated driving is often determined by assessments of comfort, safety, and trust. Although definitions for comfort and safety in the field of automated driving can differ depending on the research question, both can be measured by the absence of discomfort and danger or on one dimension from a negative to positive value range representing different degrees of comfort or discomfort. In [23], trust in automated systems is defined as “an attitude that a user is willing to be vulnerable to an action from an automated system”. Trust is seen as a key factor in the acceptance and usage of automated driving technology. Therefore, a considerable amount of literature focuses on understanding trust in the context of automated driving and how it can be increased to meet user expectations [7,24,25,26].

2. Driving Comfort

Driving comfort is an essential component to fulfill customer expectations [27,28]. However, until today, there is no unique definition of comfort [29]. Whereas some studies see comfort and discomfort as two end poles on a single dimension [30,31], others interpret both as independent constructs that can be experienced simultaneously [32,33]. Although no universal definition of comfort exists, Ref. [33] has summarized three aspects of comfort that are commonly shared among a variety of definitions: (1) comfort is subjective and individual; (2) both internal and external factors have an impact on individually experienced comfort; (3) comfort is experienced as a reaction to the environment. In addition to the absence of a common definition, measuring experienced comfort poses another significant challenge in comfort-related research. In the past, different methods have been introduced to measure comfort. In [34,35], comfort is measured via physiological measures such as heart rate, electrodermal activity, galvanic skin response, and others, whereas [36,37] measures the discomfort via a hand controller. The usage of questionnaires is another approach that is often taken in various studies to measure comfort [38,39].

3. Driving Style

In the context of conditional automated driving, the driving style and strategy will become increasingly important for experienced comfort and safety as the driver will have limited predictability of the AV’s trajectory and control over the vehicle guidance [10,36,40]. This has also been described as a “loss of controllability” [4]. So far, no universal definition of driving style exists [41]. However, Ref. [41] has summarized three aspects of a driving style that are commonly shared among various definitions: (1) driving styles are relatively stable and habitual; (2) driving styles differ across groups and individuals; (3) driving styles reflect conscious decisions made by the driver. A common approach to describe a driving style is to use objective parameters such as vehicle kinematics and environment context. In the past, a variety of parameter representations have been introduced, such as time series data [3,42], statistical characteristics [7,43,44], or latent variables, representing a dimensionless combination of different parameter types [9,45,46]. In order to differentiate between various driving styles, classification techniques such as manual or data-driven methods are commonly applied. These techniques typically involve analyzing value ranges or data patterns, which are then followed by subjective categorization using terms such as “defensive”, “moderate”, or “aggressive” [41]. Driving styles can also be differentiated based on their strategies for direct vehicle control, interaction with other traffic participants, or behavior in specific environments, as discussed in various studies [47]. For instance, different deceleration profiles for approaching a slower lead vehicle have been presented as examples of different strategies used for vehicle control [10,36]. In another study, the behavior of an autonomous vehicle affected whether a cut-in vehicle changed lanes in front of or behind the autonomous vehicle [7] as an example of different strategies for cooperation. In the case of specific environments, Ref. [21] presented different approaches for passing through an intersection. One strategy involves cautious, human-like behavior, whereas the other involves machine-like behavior that conveys a sense of prior knowledge about the environment and other oncoming road users. Many studies suggest that human-like driving behavior will increase user trust, acceptance, predictability, and ride comfort and safety [18,22,48]. Human-like driving can be learned from large naturalistic data sets [13,49], expert demonstration [14,16], or random subject collectives [50]. A slightly different approach to learning a driving style was taken by [15], specifically asking participants to demonstrate their desired driving style for an automated vehicle through manual driving instead of recording their everyday driving style. The aim of human-like driving styles is to incorporate common driving conventions of human drivers, including in-lane positioning in curves [51], distance regulation for car-following [17], crossing behavior for intersections [52], and others. The use of virtual agents in AVs with human-like features such as voice and appearance [22,53,54,55] is an additional approach to create an even stronger impression of a human-like automated vehicle, which is also referred to as anthropomorphism in the literature. Preferences for automated driving have been analyzed for different maneuvers such as lane changes [8,10,11,36,43], car-following [56,57,58,59], decelerating to a slower lead vehicle [10,36], or across different maneuvers and scenarios [6,7,25,26]. A common approach by previous studies in this context is to let users experience a predefined driving style in comparison with other driving styles or their own prerecorded driving style, using driving simulators. A Wizard-of-Oz setup was used by [6,36]. The amount of studies investigating driving style preferences for automated driving is considered small. In addition, studies differ in terms of driving environments such as urban, rural, or motorways and whether preferences are evaluated in relation to [5,6,7,8,9] or independently of participants’ personal driving style [10,11,36,43], making it yet difficult to derive a clear picture of how users want to be driven in an automated vehicle. With regard to age-related differences, Refs. [3,60,61] have investigated how age affects the experience of different automated driving styles for various maneuvers on rural and motorway roads. All studies involved participants experiencing their own driving style and up to two predefined automated driving styles. Ref. [3] found that both younger and older participants preferred the driving style of a younger driver, whereas younger drivers in [60] preferred a defensive driving style, showing the greatest rejection of their own driving style in terms of safety and comfort. Results of [61] show that older drivers prefer their own style when compared to a more aggressive automated driving style.

4. Trust

Trust has proven to be a critical factor for the acceptance of automated vehicles [24,62,63]. Without appropriate levels of trust, users may not feel comfortable using and relying on driverless technology, preventing them from fully taking advantage of its potential benefits [64,65,66]. Similar to the terms comfort and driving style, various definitions exist for trust as it is seen as a multidimensional concept. However, Ref. [62] has summarized three aspects that are commonly shared among a variety of definitions: (1) trust is given by a truster to a trustee with something at stake; (2) in order for the trustee to perform the task, an incentive must exist. In the case of technology, the designer’s intended use of the system serves as the primary incentive for the trustee; (3) trust involves the possibility of the trustee failing to perform the task, leading to uncertainty and risk. Trust is commonly assessed using questionnaires [7,10,21]. Ref. [26], on the other hand, measured trust via acceleration and brake pedal inputs by the user, expressing their dissatisfaction with the driving style. Various studies investigated which factors affect users’ trust in an automated vehicle and how it can be increased. References [67,68,69] focused on how information displayed to the driver can help to increase trust, whereas [7,10,21,26,70] analyzed the influence of driving styles on users’ trust, all coming to different conclusions. Whereas the simulator study results of [7,26] suggest that automated driving styles that align with the user’s personal driving style increase trust, no influence of driving style on trust was found in [10,21]. However, it has been observed that trust can increase over time with repeated experience of an AV [21,62,71,72].