Secondary task engagement and vehicle automation – Comparing the effects of different automation levels in an on-road experiment

Secondary task engagement and vehicle automation – Comparing the effects of different automation levels in an on-road experiment

Transportation Research Part F 38 (2016) 67–82 Contents lists available at ScienceDirect Transportation Research Part F journal homepage: www.elsevi...

634KB Sizes 2 Downloads 34 Views

Transportation Research Part F 38 (2016) 67–82

Contents lists available at ScienceDirect

Transportation Research Part F journal homepage: www.elsevier.com/locate/trf

Secondary task engagement and vehicle automation – Comparing the effects of different automation levels in an on-road experiment Frederik Naujoks ⇑, Christian Purucker, Alexandra Neukum Würzburg Institute for Traffic Sciences (WIVW), Robert-Bosch-Straße 4, 97209 Veitshöchheim, Germany

a r t i c l e

i n f o

Article history: Received 28 January 2015 Received in revised form 18 December 2015 Accepted 20 January 2016

Keywords: Automated driving On-road experiment Real traffic Secondary task engagement Individual differences

a b s t r a c t Recent and upcoming advances in vehicle automation are likely to change the role of the driver from one of actively controlling a vehicle to one of monitoring the behaviour of an assistant system and the traffic environment. A growing body of literature suggests that one possible side effect of an increase in the degree of vehicle automation is the tendency of drivers to become more heavily involved in secondary tasks while the vehicle is in motion. However, these studies have mainly been conducted in strictly controlled research environments, such as driving simulators and test tracks, and have mainly involved either low levels of automation (i.e., automation of longitudinal control by Adaptive Cruise Control (ACC)) or Highly automated driving (i.e., automation of both longitudinal and lateral control without the need for continuous monitoring). This study aims to replicate these effects during an on-road experiment in everyday traffic and to extend previous findings to an intermediate level of automation, in which both longitudinal and lateral control are automated but the driver must still monitor the traffic environment continuously (so-called Partial automation). N = 32 participants of different age groups and different levels of familiarity with ACC drove in rush-hour traffic on a highway segment. They were assisted by ACC, ACC with steering assistance (ACC+SA), or not at all. The results show that while subjective and objective driving safety were not influenced by the degree of automation, drivers who were already familiar with ACC increased the frequency of interactions with an in-vehicle secondary task in both assisted drives. However, participants generally rated performing the secondary task as less effortful when being assisted, regardless of the automation level (ACC vs. ACC+SA). The results of this on-road experiment thus validate previous findings from more-controlled research environments and extend them to Partially automated driving. Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction 1.1. Motivation Today’s assistive technologies support the driver at different levels of the driving task, from providing driver information (e.g., route planning and navigation) to active vehicle safety. Commercially available support systems such as Adaptive ⇑ Corresponding author. Tel.: +49 931 780090; fax: +49 931 78009150. E-mail addresses: [email protected] (F. Naujoks), [email protected] (C. Purucker), [email protected] (A. Neukum). http://dx.doi.org/10.1016/j.trf.2016.01.011 1369-8478/Ó 2016 Elsevier Ltd. All rights reserved.

68

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

Cruise Control (ACC) and lane departure assistance help the driver maintain either longitudinal or lateral control (automation level: ‘‘Driver assistance”, Gasser et al., 2012; Gasser & Westhoff, 2012). Only recently have automobile manufacturers introduced systems to the market that are capable of taking over both longitudinal and lateral control simultaneously (automation level: ‘‘Partial automation”, Gasser et al., 2012; NHTSA, 2013). These new systems continue to require that the driver monitors the driving situation continuously to be able to take over vehicle control at any time. Therefore, Partial automation usually does not allow hands-off driving for extended time periods. Higher levels of automation that do not require continuous monitoring by the driver have been demonstrated in test vehicles and in various research projects (e.g., HAVEit, SARTRE, CityMobil, AdaptIVe) and are subject to on-going research (e.g., Gold, Damböck, Lorenz, & Bengler, 2013; Naujoks, Mai, & Neukum, 2014; Wiedemann, Schömig, Mai, Naujoks, & Neukum, 2015). It is commonly agreed that vehicle automation causes a fundamental change in the nature of the driving task, as the role of the driver is shifted from actively controlling the vehicle to monitoring the environment and the behaviour of the automation (cf. Endsley & Kaber, 1999; Flemisch et al., 2012; Parasuraman, Sheridan, & Wickens, 2000). Automation relieves the human operator from previously manually controlled tasks and lowers workload as a consequence, particularly in dualtask situations (Wiczorek & Manzey, 2014). However, numerous studies have demonstrated negative side-effects of automation on human performance in various contexts, such as loss of skill (Stanton & Marsden, 1996), loss of situation awareness (Endsley & Kaber, 1999) or overreliance on the automation (Parasuraman & Riley, 1997). Within the context of vehicle automation, one important aspect is the driver’s attention distribution between monitoring the primary driving task and performing secondary activities while driving. A growing body of literature has established a link between the introduction of assistive technologies and secondary task engagement. Providing assistive technologies to the driver, such as warning systems (Burns, Knabe, & Tevell, 2000; Naujoks & Totzke, 2014) or ACC (de Winter, Happee, Martens, & Stanton, 2014; Dragutinovic, Brookhuis, Hagenzieker, & Marchau, 2005; Rudin-Brown & Parker, 2004), may result in behavioural effects such as an increase in the frequency of performing secondary tasks while driving. While the link between vehicle automation and secondary task engagement can be considered well established, there are still some important knowledge gaps within the research field that will be addressed in the current work:  Most studies have been conducted using driving simulation (cf. de Winter et al., 2014) or test tracks as a research environment. Both research environments provide the possibility of conducting carefully controlled studies and, as a result, can contribute to the internal validity of the results. However, the results may differ to some extent from everyday driving, due to the complexity of realistic driving environments (Carsten, Kircher, & Jamson, 2013) or other reasons of ecological validity, such as the perceived physical danger (de Winter, van Leeuwen, & Happee, 2012).  To date, several studies have addressed the impact of assistance and automation technologies on secondary task engagement while driving. However, prior studies have focused mainly on the automation levels Driver assistance and, to a lesser extent, on Highly automated driving. Partial automation has mainly been a non-issue within this context (Carsten, Lai, Barnard, Jamson, & Merat, 2012; Kircher, Larsson, & Hultgren, 2014).  Moderating factors on the relationship between automation availability and secondary task engagement have rarely been reported. For example, Naujoks and Totzke (2014) have shown that younger rather than older drivers were likely to increase their level of secondary task engagement when being assisted by a congestion tail warning system. Other relevant inter-individual differences, such as familiarity with vehicle automation or personality traits such as sensation seeking or locus of control (cf. Hoedemaeker & Brookhuis, 1998; Rudin-Brown & Parker, 2004), have been addressed occasionally but again not within the context of the Partial automation level. 1.2. Vehicle automation and secondary task engagement During non-assisted driving, drivers continuously observe changes in the traffic situation (Koornstra, 1993; van der Hulst, Meijman, & Rothengatter, 1999) and adapt their driving behaviour and attention allocation accordingly (Brown, Lee, & McGehee, 2000; Cooper, Vladisavljevic, Medeiros-Ward, Martin, & Strayer, 2009; Muhrer & Vollrath, 2010). Consequently, drivers have been shown to increase or decrease their engagement in secondary activities according to the demands of the traffic situation (Schömig, Metz, & Krüger, 2011). Providing assistance to the driver might lead to a perceived increase in driving safety, which may cause drivers to engage in riskier driving styles (e.g., Fuller, 1984; Näätänen & Summala, 1976; Wilde, 1988) or to rely too much on the system’s capabilities (e.g., Rudin-Brown & Noy, 2002; Weller & Schlag, 2004). This, in turn, is likely to result in increased secondary task engagement (e.g., Carsten et al., 2012; de Winter et al., 2014; Dragutinovic et al., 2005; Rudin-Brown & Parker, 2004). Understanding the circumstances that lead to such an increase in secondary task engagement is particularly important with regard to recent and upcoming advances in vehicle automation. During non-assisted driving, an increase in secondary task engagement is likely to go along with compensatory driving behaviour, such as lowering velocity (Horberry, Anderson, Regan, Triggs, & Brown, 2006; Rakauskas, Gugerty, & Ward, 2004) or increasing following distance (Jamson, Westerman, Hockey, & Carsten, 2004; Strayer & Drew, 2004), which might buy the distracted driver more time in case a critical situation emerges. When driving with assistance systems that take over full vehicle control, these types of compensatory actions can no longer be initiated by the driver.

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

69

Various reasons for and consequences of increased secondary task engagement have been discussed thus far. Most prominently, secondary task engagement has been linked to negative effects on primary task performance, such as an increase in crash risk (Bach, Jaeger, Skov, & Thomassen, 2009; Ferdinand & Menachemi, 2014; Green, 1999; Horrey & Wickens, 2007; Klauer, Dingus, Neale, Sudweeks, & Ramsey, 2006), because the visual resource can hardly be divided between spatially separate targets (Wierwille, 1993). In particular, multiple resource theory suggests that visual-manual secondary tasks (such as using a smartphone) are likely to negatively impact a mainly visual-manual primary task (such as driving a car), as both tasks draw from the same cognitive resources (Wickens, 1984). While, according to biased competition theory (Desimone & Duncan, 1995), visual processing can also be impacted by other cognitive operations, it is unlikely that the engagement of purely cognitive secondary tasks – as compared with visual or visual-manual tasks – goes along with similar impairments of the primary task performance, as they do not draw primarily from the same cognitive resource. In accordance with an increased crash risk, secondary task engagement was furthermore associated with a decrease in situation awareness (de Winter et al., 2014; Ma & Kaber, 2005; Rogers, Zhang, Kaber, Liang, & Gangakhedkar, 2011; Young, Salmon, & Cornelissen, 2013). Alternatively, secondary task engagement might also be interpreted as a consequence of a subjective decrease of workload from the driving task (cf. de Winter et al., 2014), which may likely be the result of increased subjective safety or trust in the capabilities of the respective assistance system. Within this context, de Winter et al. (2014) note that secondary task engagement may measure different constructs based on the specific instruction and type of secondary task. According to these authors, performing a self-paced task on an in-vehicle display with the specific instruction to do so may be interpreted as an indicator of workload. In contrast, choosing to deliberately engage in secondary tasks, such as using in-vehicle infotainment systems, while driving might indicate decreased situation awareness (Schömig et al., 2011). In sum, increased secondary task engagement may reflect both positive consequences of assistive technologies, such as decreased workload or increased subjective safety, and negative consequences, such as decreased situation awareness. As diverse as these explanations for an increase in secondary task engagement may be, the consequences are the same: secondary task engagement most likely decreases the capability of monitoring the driving environment and the status of the automation in an anticipatory manner. For example, Schömig et al. (2011) found that drivers decreased their level of engagement in a secondary activity when approaching demanding traffic situations; this can be classified as ‘‘anticipatory control” (Fuller, 2005). However, when drivers become fully immersed in their secondary activities, this anticipatory driving style may be impaired – with obvious consequences. Summing up the current research literature, the following moderator variables can be regarded as contributing to an increased willingness to engage in secondary tasks while driving (cf. Saad, 2006): 1. Level of automation: An increased level of automation is associated with an increase in secondary activities while driving (Carsten et al., 2012). For example, Rudin-Brown and Parker (2004) showed that driving with ACC leads to an increase in the processing of secondary tasks while driving, indicating an attention shift away from the primary driving task. A similar adverse effect on the distribution of attention following the presentation of Forward Collision Warnings (FCWs) has been reported by Wege, Will, and Victor (2013) and Muhrer, Reinprecht, and Vollrath (2012). A recent review by de Winter et al. (2014) has shown that during higher levels of automation (i.e., Highly automated driving), an even stronger tendency to engage in secondary activities during driving is to be expected. 2. Traffic state/velocity: Metz, Landau, and Just (2014) report that the frequency of secondary activities during a recent Field Operational Test (FOT) was negatively correlated with velocity: common secondary activities were mostly performed when the vehicle was idling at standstill. When the vehicle was in motion, drivers tended to perform secondary activities significantly less frequently. In a similar vein, Naujoks and Totzke (2014) show that engagement in a self-paced menu task was higher during congested traffic (lower velocity) than during free-flow traffic (higher velocity). A possible explanation of this effect may be that a higher driving speed is related to greater feelings of risk and higher workload (Lewis-Evans, de Waard, & Brookhuis, 2011; Fuller, 2005), which may inhibit the willingness to perform secondary activities while driving (de Winter et al., 2014; Schömig et al., 2011). 3. Age: Although some authors have proposed that age-related impairments may be compensated for by in-vehicle technologies (Suen, Mitchell, & Henderson, 1998; Vrkljan & Miller-Polgar, 2005), these technologies may also lead to additional distraction, cognitive overload or a change in compensatory driving behaviour of older drivers (e.g., Davidse, 2007; Simões & Pereira, 2009). With this in mind, it might be expected that older drivers tend to engage less in secondary tasks during non-assisted driving than younger drivers (Naujoks & Totzke, 2014). Providing an assistance option may, on the other hand, change the drivers’ tactical decisions not to engage in the secondary tasks, leading to an increase in secondary activity. 4. Prior experience with assistive technology: According to Rudin-Brown and Noy (2002; cf. Rudin-Brown & Parker, 2004), over-trust in the capabilities of an assistive technology may be a necessary condition for the development of a willingness to engage in secondary activities while driving. Trust in technology is a cognitive attitude towards the respective technology that changes over time (Lee & See, 2004; Wickens & Xu, 2002). In line with the foregoing, drivers who frequently use ACC also report more-frequent ACC usage in distracting and risky situations (Wu & Boyle, 2015). With this in mind, it may be possible that prior experience with vehicle automation is an important moderating variable with regard to increased secondary task engagement: if participants are already familiar with vehicle automation, they may be more likely to rely on the assistant system because they know the system’s limitations (Larsson, 2012; Larsson, Kircher, & Hultgren, 2014). This may, in turn, lead to a greater likelihood that they become complacent, directing their attention away from the primary driving task (Bailey & Scerbo, 2007; Parasuraman, 2000).

70

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

1.3. Research questions and hypotheses This on-road experiment aimed to investigate participants’ willingness to engage in a simple and common secondary task while driving with different levels of vehicle automation (Baseline vs. Driver assistance vs. Partial automation). Specifically, the aim of the study was to answer the following research questions:  Does an increase in vehicle automation lead to an increase in secondary task engagement while driving? According to previous research (Carsten et al., 2012; de Winter et al., 2014), an increase in the willingness to engage in secondary tasks based on a higher level of vehicle automation was expected.  Additionally, the impact of different potential moderating factors on the relationship between the level of vehicle automation and secondary task engagement, such as traffic state, age or prior experience with ACC, was taken into account to answer this research question. For example, Kircher et al. (2014) and Jamson, Merat, Carsten, and Lai (2013) have shown that even with higher levels of vehicle automation, drivers still adapt their behaviour to the demands of the traffic situation. We thus expected to find similar effects. Furthermore, we expected that older drivers would engage in secondary activities to a lesser extent than younger drivers (Naujoks & Totzke, 2014), while prior experience with ACC was expected to lead to stronger engagement in secondary activities because of greater familiarity with vehicle automation (Rudin-Brown & Noy, 2002; Rudin-Brown & Parker, 2004).  Does an increase in vehicle automation go along with a change in subjective assessments of the mental effort of performing the secondary task or of the subjective safety of the drives? At this point, we expected that an increase in secondary task engagement would go along with a reduction in reported workload, as it has been argued that an increase in the frequency of secondary activities reflects a lowered workload due to vehicle automation (de Winter et al., 2014). 2. Method The study was conducted with a series-production Mercedes-Benz E-Class in daily traffic (vehicle length: 4.9 m, vehicle width: 1.9 m, sedan car, model year 2013). During the study, driver behaviour was measured, as was engagement in a secondary task while driving. To assess the impact of different automation levels (factor: ‘‘Automation level”) on driver behaviour and secondary task engagement, participants completed drives with a commercially available Adaptive Cruise Control system with steering assistance (ACC+SA), which continuously assists the driver with lateral vehicle control in addition to the longitudinal control by the ACC (automation level: Partial automation). The participants also completed additional drives with ACC alone (without steering assistance functionality, automation level: Driver assistance) and without any assistance (Baseline). In addition to the experimental factor ‘‘Automation level”, the driver’s age as well as prior experience with ACC was taken into account in the selection of the participants. This resulted in a test plan with an experimental factor ‘‘Automation level” as well as two quasi-experimental factors: ‘‘Age” and ‘‘ACC experience” (see Table 1). A within-subject design was chosen, i.e., every participant completed drives in all three automation conditions. The order of the drives was fully balanced. 2.1. Human–machine interface The Human–machine interface (HMI) was located in the central information display in the middle of the tachometer and consisted of a depiction of the lead vehicle (when driving behind another vehicle), line markings and a steering wheel symbol. Via this HMI, the following information about the assistant system was provided to the driver:  Indication of set speed and deviation from set speed on the tachometer.  Indication of set headway.  Indication of whether the lane markings were recognized by showing lane marking symbols in green or white (green: recognized correctly).  Indication of whether the steering support was active or passive by showing a green or white steering wheel (green: active). When driving with activated ACC+SA, steering support was provided if the lane markings were recognized correctly or if a lead vehicle was recognized (indicated by green steering wheel symbol). Otherwise, the steering support was passive

Table 1 Experimental design. Factors

Factor levels

Automation level Age ACC experience

Baseline (BL) vs. ACC (Driver assistance) vs. ACC+SA (Partial automation) > 50 years vs. 6 50 years With vs. without prior experience with ACC

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

71

(indicated by white steering wheel symbol). ACC was continuously active when ACC+SA was activated. The active steering support consisted of a steering torque overlay towards the centre of the lane. While driving on highways, the pre-conditions for active steering support were fulfilled most of the time. During the drives, the HMI screen was recorded on video to evaluate the activity of the steering support. In 8 of the 32 drives with ACC+SA, it was not possible to identify the colour of the steering wheel symbol on the HMI screen from the video recording because of glare due to sunny weather. In the remaining 24 drives, steering support was not available for 18.38% of the driving time (SD = 10.19), on average, when ACC+SA was activated. Availability did not differ between participants with or without prior experience with ACC (t-test for independent samples: t = 1.35, df = 21, p = .191), and there was also no correlation with driver age (r = .290, df = 23, p = .180). 2.2. Vehicle equipment The test vehicle was equipped with the necessary measurement instrumentation. As no direct, unencrypted access to the CAN (controller area network) bus data was available, the relevant data were recorded by external measurement equipment:  Velocity, longitudinal and lateral acceleration were measured via GPS and an external 3-axis acceleration sensor (model: JW24/F14).  Secondary task activity while driving was directly measured via an in-vehicle smartphone-based task.  Furthermore, the vehicle was equipped with several cameras facing the traffic environment (front and back) and the driver. All measures were recorded jointly with a sampling frequency of 100 Hz via SILAB software. A unified time stamp was attached to all measures via SILAB during data recording. To ensure the safety of the participants, the test vehicle was equipped with a dual control pedal system. This allowed the experimenter to take over longitudinal vehicle control directly, if necessary. All experimenters received extensive training in using the dual control pedal system prior to the study. The participants were not informed about the dual control pedal system and, thus, were unaware that the experimenter could intervene in case of a critical driving event. 2.3. Test route The purpose of the study was to investigate driving behaviour while driving in areas of congested traffic. Congested traffic was chosen as the focus area for two reasons. First, driving with ACC requires other road users to be present. Otherwise, the drives would not have been different from driving with ordinary cruise control from a driver’s point of view. Second, a comparison of the secondary task engagement at different velocities but with a comparable level of traffic density was intended. Therefore, the drives were completed in rush-hour traffic (i.e., high traffic density). To control the effects of the number of available lanes and road curvature, the test route had to meet the following requirements: highway with several lanes and mostly straight stretches of road. A suitable road structure was found in the metropolitan area of Stuttgart, which ranks sixth among European cities on the TomTom Congestion Index (TomTom International B.V., 2013). The on-road experiment took place on a fixed section of highways A8 and A81 in the metropolitan area of Stuttgart. The test route consisted of a three-lane highway. Most parts of the test route had no speed limit. All test drives took place during rush hour traffic in the morning or in the evening. 2.4. Procedure and instructions The study consisted of the following parts: 1. After arriving at the test location, participants completed a familiarization drive prior to the main part of the study. This study segment took approximately half an hour and included basic instruction about the system characteristics (switching the assistant on/off), followed by a short familiarization drive in which both ACC and ACC+SA were used by the participants. During the instruction, the participants were merely told that the assistant would keep the set speed and distance (in case of ACC) or that it would additionally assist the driver with lane keeping (in case of ACC+SA). However, specific information about the system (e.g., detailed interpretation of the HMI) was not provided. 2. After the familiarization drive, the secondary task was explained to the participants, and every participant executed 40 tasks while the vehicle was in parking mode. This was done to ensure that no learning effects regarding the secondary task would occur during the test drives. 3. During the subsequent main part of the experiment, the drivers completed three consecutive drives on the chosen highway routes, either with ACC, ACC+SA or without any automation (Baseline drive), each lasting approximately 30 min. Within each of the drives, the secondary task was made available to the participants, which they could engage at their own will. After each drive (ACC, ACC+SA or Baseline), the next available parking place was approached, where the participants had some time to rest and answered standardized questions concerning different aspects of the preceding drive. This study part took approximately two hours.

72

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

Table 2 Distribution of age and years since driver license test; descriptive statistics and t-test for independent samples comparing participants with or without prior experience with ACC. Variable

Prior experience with ACC

N

M

SE

df

p

Age [years]

No Yes

20 12

46.70 48.00

3.78 4.41

t 2.18

30

.829

Years driver license [years]

No Yes

20 12

27.65 28.42

3.64 4.43

1.32

30

.896

Table 3 Distribution of mileage and frequency of car usage. Prior experience with ACC

No Yes

Mileage: ‘‘How many kilometres have you driven in your lifetime?” <100.000 km

100.000–300.000 km

300.000–500.000 km

500.000–1.000.000 km

>1.000.000 km

3 0

4 5

4 0

5 3

4 4

Frequency of car usage: ‘‘How often do you use a car?”

No Yes

<1/week

1–2/week

3–5/week

Daily

1 0

4 1

6 3

9 8

2.5. Sample The participants were recruited from a pool of over 150 potential candidates who were contacted through advertisements in newspapers prior to the study. The participants were chosen (1) to select people with sufficient experience with ACC (i.e., having ACC in their own car or in a company car they use regularly) and (2) to match the ACC groups according to age. The sample consisted of 8 female and 24 male drivers (n = 10 with prior ACC experience, n = 22 without prior ACC experience). None of the drivers had previously participated in experimental studies that addressed similar research questions. The median age of the drivers was 47.19 years (SD = 16.08, MIN = 20, MAX = 70). As evident from Table 2, participants with and without prior experience with ACC did not differ with regard to age and with regard to the years they held their driver license. Participants’ mileage and frequency of car usage is shown in Table 3. Nominal scales were used for these variables to facilitate the participants’ estimations, as it was expected that most participants would not be able to give a precise statement on these variables. Drivers with or without prior experience with ACC did not differ in these two characteristics according to Fischer’s exact test (mileage: p = .232; frequency of car usage: p = .712). 2.6. Secondary task description It must be assumed that the actual effect of the secondary task on driving behaviour as well as the willingness to engage in the secondary activity depends on the type of the secondary task, specifically on how long the task requires the drivers to take their eyes off the road (NHTSA, 2012). The task used in this study was a rather simple visual-manual task that was selfpaced and could be interrupted at any time (cf. Young, Regan, & Lee, 2009 for a discussion on this topic). We expected that the task would be easy to execute while driving and that the drivers would be willing to do so. A recent Field Operational Test (FOT) indicates that drivers actually do perform such everyday tasks while driving (Metz et al., 2014). The task consisted of a smartphone application that was created at the Würzburg Institute for Traffic Sciences for use in experimental studies. The task was specifically designed to be similar to real-world secondary tasks that are actually performed while driving and consisted of retrieving weather-related information within a 1  4  4  4-task menu (see Fig. 1). Interactions with the secondary task consisted of the following steps:  The first screen indicated the type of information that had to be retrieved in order to complete the respective task. Within each task, participants were asked to look up either the minimum temperature, the maximum temperature, the rain probability or the wind velocity at different times of day (in the morning, at noon, in the evening or at night) on different days (today, tomorrow, in two days or in three days). For example, the first screen might indicate the question: ‘‘What is the probability of rain in the evening in two days?”  After touching the screen, the participants were required to select the day (second screen in Fig. 1), the time of day (third screen in Fig. 1) and the type of weather information (fourth screen in Fig. 1).  Subsequently, the next item was displayed on the secondary task screen, e.g., ‘‘what is the minimal temperature in the morning in three days?” The task order was generated randomly. There was no limit to the maximum number of tasks that could be processed.

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

73

Fig. 1. Secondary task (example task: looking up the rain probability on the evening of the same day). First screen: task presentation; second screen: selection of day; third screen: selection of time of day; final screen: selection of specific information.

2.7. Ethical considerations From the perspective of research ethics, investigating interactions with secondary activities during driving requires that the participants are fully aware of the research goals and that their safety is guaranteed during driving. To fulfil these requirements, careful considerations were made regarding (1) the selection and installation of the secondary task, (2) the instructions given prior to the study and (3) the possibility of intervention in case of critical driving situations. First, a secondary activity that would not be expected to exceed the drivers’ capabilities was selected. The smartphone was installed in a fixed position at the upper part of the central infotainment display, which allowed legal usage according to German road traffic regulations (§ 23 section 1a of the German road traffic code (Straßenverkehrsordnung, 2013)). The task itself was originally developed for a study financed by the German Federal Highway Research Institute on the distraction effects of smartphone activities (Schömig, Schoch, Neukum, Schumacher, & Wandtner, 2015). A study on the distraction effects of this task (Schömig et al., 2015) found that single glance durations and total glance time were generally within the acceptable tolerance range according to the visual-manual NHTSA driver distraction guidelines for in-vehicle electronic devices (NHTSA, 2012). Second, the participants were fully informed about the research goals and the procedure of the study and gave informed consent. The participants were told that the goal of the study was to investigate drivers’ willingness to perform secondary activities while driving. They were instructed to engage in the secondary task only when they would normally perform such a task in traffic. They were instructed to work on the task only when they judged the traffic situation to be safe enough to do so and that they could interrupt working on the secondary task at any time. It was emphasized that the decision of when to engage in the secondary task while driving was fully optional, and that choosing not to perform the secondary task would not lead to any disadvantages for them. Specifically, the experimenter told them that choosing not to perform the task at all was ‘‘completely alright”. They were also informed about general driving risks and that their primary goal as a driver in the experiment was to ensure their own driving safety, the safety of the experimenter and that of other road users. In addition, participants were informed that they could abort the experiment at any time without stating a reason and still receive their financial compensation. All participants received a financial compensation of 50 € for taking part in the study, independently of their performance in the secondary task. They were thus not forced to perform the task in any way. Third, it was taken care that the drivers would not endanger themselves or the experimenter during driving. For this reason, one experimenter, whose only task was to monitor the driving situation, could take over the longitudinal vehicle control via a dual control pedal system directly, if necessary. Finally, a participant accident insurance policy was taken out that covered the monetary risks of the experiment. 2.8. Dependent measures and inferential statistics To describe the behaviour of the driver and the resulting reaction of the vehicle, various characteristics of primary task performance were determined (see Table 4).  Lateral and longitudinal control was measured by recording the speed of the vehicle as well as the longitudinal and lateral acceleration. Stronger longitudinal deceleration values indicate that the driver performs hard braking manoeuvres, whereas stronger lateral acceleration values may be indicative of poorer lateral control of the vehicle.  To document the drivers’ attention allocation, inputs into the secondary task device were recorded during the drive. The glance behaviour of the participants was not recorded. This was done for several reasons. First, because the study was conducted in a real vehicle, it was expected that reliable measurement of the eyes-off road time would have required a head-mounted eye-tracking system. The rather long time the drivers would have had to wear such a device might have

74

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

Table 4 Dependent variables. Parameter

Description

Unit

Lateral and longitudinal Velocity Maximum deceleration Lateral acceleration

vehicle control Mean velocity Maximum deceleration values

[km/h] [m/s2]

Maximum acceleration values

[m/s2]

Input in secondary task device per minute; input is defined by touching the secondary task device and moving one step further in the menu task

[inputs/ min]

Rating of perceived effort of performing the secondary task while driving (SEA Scale: 0 = very low effort, 220 = extraordinary high effort; Eilers et al., 1986) Rating of perceived safety of the preceding drive (0 = not safe at all, 15 = very safe)

[0. . .220]

Attention allocation Secondary task engagement Subjective measures Effort Safety

[0. . .15]

resulted in considerable discomfort. Additionally, wearing a head mounted system might also have impaired the drivers’ visibility and moving range, which would have impaired their driving style and their ability to perform the secondary task while driving. Finally, the resulting eyes-off road time associated with the secondary task during non-assisted driving had already been determined in a previous study (Schömig et al., 2015).  In addition to the measures of driving performance, subjective ratings of perceived effort (the so-called SEA scale, Eilers, Nachreiner, & Hänecke, 1986) and perceived safety were collected after each drive. The SEA scale is a uni-dimensional scale that uses verbal labels as anchoring points (e.g., ‘‘extremely strenuous”, ‘‘somewhat strenuous”), ranging from 0 to 220. Driving safety was rated using a 15-point scale that also uses verbal labels as anchoring points. The rating was given in a two-step category scaling procedure: after an initial categorization of driving safety (e.g., ‘‘very safe”), a numerical rating was given (e.g., ‘‘13”; for examples of this two-step approach, see Naujoks & Neukum, 2014; Siebert, Oehl, & Pfister, 2014; Totzke, Naujoks, Mühlbacher, & Krüger, 2012). To assess the influence of the experimental factors on the dependent measures, two statistical procedures were used:  Linear mixed model (LMM): To assess the effects of the experimental factors on the performance measures (vehicle control and attention allocation), an LMM procedure was implemented. Unlike ANOVAs (see below), LMMs can deal appropriately with missing data that are likely to occur in an on-road experiment (caused by random variability in traffic conditions, data recording errors, etc.). The data were analysed by applying an LMM with the fixed effects ‘‘Automation level”, ‘‘Velocity condition”, ‘‘ACC experience” and ‘‘Age”. The participants served as random effects in the model. Driving speed was taken into account as a factor with four levels (‘‘driving at very low driving speed”: <10 km/h; ‘‘jammed traffic”: 10–60 km/h; ‘‘transition from jammed to flowing traffic”: 60–100 km/h; ‘‘flowing traffic”: >100 km/h). In addition to the main effects, two-way interactions between the factors were introduced to the model. Because interaction effects of higher order were not expected, these were omitted in the model.  Analysis of Variance (ANOVA) with repeated measures: To assess the effects of the experimental factors on the subjective (perceived effort and perceived safety) measures, a full factorial ANOVA for repeated measures was applied. 3. Results 3.1. Velocity distribution In sum, the data represent approximately 45 h of valid driving time (i.e., driving on highways with a valid GPS signal; Baseline: 14.93 h; ACC: 15.00 h; ACC+SA: 15.83 h). Approximately 38% of the driving time is spent driving either in stopand-go traffic (10–60 km/h) or in standstill to near standstill (0–10 km/h). The rest of the driving time (62%) is spent in the transition between congested and free-flowing traffic (60–100 km/h) or in free-flowing traffic (>100 km/h). A v2-test on the frequency of observations within these defined velocity ranges indicates that the relative frequencies do not differ among automation levels (v2 = 0.69, df = 6, p = .995; see Table 5). This justifies a comparison of the dependent measures among automation levels in the current study. 3.2. Acceleration/deceleration During assisted and non-assisted drives, no braking intervention by the experimenter was necessary. Fig. 2 depicts the mean maximum longitudinal deceleration (left panel) and the mean maximum lateral acceleration (right panel) per velocity condition. Visual inspection shows that these values depend mainly on the traffic situation.

75

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82 Table 5 Frequency of observations. Automation level

Velocity range [km/h]

Baseline ACC ACC+SA

0–10

10–60

60–100

>100

17 20 19

29 29 29

30 32 31

27 23 27

Maximum longitudinal deceleration

Maximum lateral acceleration

-1.0

2.0 BL ACC

-1.5

1.5

[m/s²]

[m/s²]

ACC+SA

-2.0

-2.5

1.0

0.5

BL ACC ACC + SA

-3.0

0.0 0-10

10-60

60-100

Velocity range [km/h]

>100

0-10

10-60

60-100

> 100

Velocity range [km/h]

Fig. 2. Maximum longitudinal deceleration (left panel) and maximum lateral acceleration (right panel) according to velocity. Mean and standard error are depicted; BL: Baseline; ACC: Adaptive Cruise Control; ACC+SA: Adaptive Cruise Control with steering assistance.

An LMM (see Table 6) confirms that both the maximum longitudinal deceleration and the maximum lateral acceleration strongly depend on the velocity condition (main effect ‘‘Velocity”), whereas the automation level has no influence on these variables. Thus, no negative or positive influence of the ACC or the ACC+SA can be found on the collected measures of lateral and longitudinal vehicle control. Furthermore, no statistically significant influence of age (main effect ‘‘Age”) or prior experience with ACC (main effect ‘‘ACC experience”) is evident. 3.3. Secondary task engagement Table 7 shows different descriptive statistics of the velocity differences between the beginning and end of the secondary tasks. The aim of this analysis is to quantify whether drivers continue processing the menu task if the traffic state changes rapidly. Within this context, it is particularly important to determine whether the drivers would continue processing the secondary tasks despite an increase in driving velocity. Measures of central tendency (mean and median) indicate that the tasks are finished under a comparable velocity condition. More than half of the tasks are processed with zero difference in velocity between their beginning and end, which means they were processed during standstill. The tails of the distribution (5th and 95th percentiles) also indicate that most tasks are finished without a relevant difference in velocity between their beginning and end. The drivers’ engagement with the secondary task when driving with different automation levels is shown in Fig. 3. On a descriptive basis, the following trends are apparent from the figure:  The drivers’ level of engagement in the secondary task depends on the velocity condition. During non-assisted driving, participants show an increase in task processing during low-velocity conditions (0–10 km/h) in comparison to driving under higher-velocity conditions (>10 km/h). Processing of the task is at its lowest while driving under free-flow conditions (>100 km/h).  During assisted driving, the average total number of inputs per minute is comparable to those during non-assisted driving, but the pattern of engagement in the secondary task with regard to driving velocity is shifted. With ACC and ACC+SA, more inputs/min are observed during the velocity conditions 10–60 km/h and 60–100 km/h compared to non-assisted driving. However, during driving in low-speed traffic (0–10 km/h), there are fewer inputs/min during the assisted drives. Independently of the automation level, engagement with the secondary tasks is lowest in free-flow traffic (>100 km/h). To identify statistically significant influences of the independent variables on the secondary task engagement, an LMM was applied (see Table 8). This analysis shows the following:

76

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

Table 6 Inferential statistics of the LMM for the independent variables ‘‘Maximum longitudinal deceleration” and ‘‘Maximum lateral acceleration”. Effect

Maximum longitudinal deceleration

Automation level Velocity Age ACC experience Automation level ⁄ Velocity Automation level ⁄ Age Automation level ⁄ ACC experience Velocity ⁄ Age Velocity ⁄ ACC experience Age ⁄ ACC experience

Maximum lateral acceleration

F

df1

df2

p

F

df1

df2

p

0.01 18.71 0.11 0.62 0.32 1.74 0.98 0.46 0.48 1.15

2 3 1 1 6 2 2 3 3 1

125 119 204 199 87 135 125 146 143 146

.995 <.001 .744 .434 .925 .179 .380 .714 .700 .285

0.10 41.47 0.00 0.24 1.26 0.24 0.37 1.61 0.83 0.98

2 3 1 1 6 2 2 3 3 1

103 123 154 145 86 112 112 126 132 149

.908 <.001 .965 .626 .283 .785 .692 .190 .479 .324

Table 7 Descriptive statistics of differences in velocity at the beginning and end of the secondary tasks; positive values indicate that the tasks were ended in a highervelocity condition. Automation level

Difference in velocity between start and end of secondary tasks [km/h] Ntasks

M

MD

Percentile 5%

Baseline ACC ACC+SA

939 924 991

0.14 0.71 0.57

0.00 0.00 0.00

25%

6.06 8.63 6.01

1.47 1.23 0.81

50%

75%

95%

0.00 0.00 0.00

1.37 1.22 1.40

5.96 7.74 7.87

Secondary task engagement 8

BL ACC ACC + SA

[inputs/min]

6

4

2

0 0-10

10-60

60-100

> 100

Velocity range [km/h] Fig. 3. Secondary task engagement according to velocity condition and automation level; mean and standard error are depicted; BL: Baseline; ACC: Adaptive Cruise Control; ACC+SA: Adaptive Cruise Control with steering assistance.

 First, the analysis confirms the influence of velocity on the secondary task engagement of the participants (main effect ‘‘Velocity”): Drivers adapt their engagement to the traffic state, engaging more in processing the secondary task while driving in low-velocity ranges compared with driving at higher speeds (0–10 km/h: M = 5.19, SD = 5.83; 10–60 km/h: M = 3.30, SD = 4.96; 60–100 km/h: M = 3.07, SD = 3.43; >100 km/h: M = 1.39, SD = 2.46). However, as apparent from Fig. 4 (left panel), this main effect is mainly due to participants having prior ACC experience (interaction effect ‘‘Velocity ⁄ ACC experience”). Participants with prior ACC experience exhibit the above-mentioned trend towards a higher engagement in the secondary task during lower driving speed, whereas participants without prior ACC experience show a low level of task processing independently of the driving speed.  Second, younger participants (<50 years: M = 3.72, SD = 4.85) generally perform more inputs/min than older participants (>50 years: M = 2.41, SD = 3.69), regardless of the assistance condition (main effect ‘‘Age”).  Third, prior experience with ACC has an impact on the level of secondary task engagement: Participants with prior ACC experience perform more inputs in the secondary task device (M = 4.63, SD = 5.65) than those without prior ACC experience (M = 2.15, SD = 3.01, main effect ‘‘Automation level”). Furthermore, only participants with prior ACC experience

77

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82 Table 8 Inferential statistics of the LMM for the independent variable ‘‘Secondary task engagement”. Effect

F

df1

df2

p

Automation level Velocity Age ACC experience Automation level ⁄ Velocity Automation level ⁄ Age Automation level ⁄ ACC experience Velocity ⁄ Age Velocity ⁄ ACC experience Age ⁄ ACC experience

0.25 14.21 12.24 25.10 1.19 0.45 4.89 0.27 5.71 0.00

2 3 1 1 6 2 2 3 3 1

95 122 121 123 93 151 156 130 138 219

.779 <.001 .001 <.001 .321 .641 .009 .847 .001 .974

Secondary task engagement

Secondary task engagement with prior ACC experience

8

8

[inputs/min]

[inputs/min]

6

4

6

4

2

2

0

0 0-10

10-60

60-100

Velocity range [km/h]

> 100

with prior ACC experience without prior ACC experience

without prior ACC experience

Baseline

ACC

ACC+SA

Level of automation

Fig. 4. Secondary task engagement according to velocity condition and ACC experience (left) and according to automation level and prior ACC experience (right); mean and standard error are depicted; BL: Baseline; ACC: Adaptive Cruise Control; ACC+SA: Adaptive Cruise Control with steering assistance.

show an increased level in secondary task processing in assisted drives (Baseline: M = 3.52, SD = 4.52; ACC: M = 4.78, SD = 5.77; ACC+SA: M = 5.48, SD = 6.36; interaction effect ‘‘Automation level ⁄ ACC experience”; see Fig. 4, right panel). However, the presence of steering assistance functionality (ACC+SA) does not lead to a further increase in secondary task engagement compared to driving with ACC alone. 3.4. Subjective measures Fig. 5 (left panel) shows the mean ratings of the perceived effort associated with the processing of the secondary task while driving. Inferential statistics are provided in Table 9. As evident from the figure, the processing of the secondary task is rated as less effortful during assisted drives (main effect ‘‘Automation level”). Planned contrast comparisons between assisted and non-assisted drives reveal that the difference from non-assisted driving is more pronounced in drives with ACC (F = 18.84, df = 1, p < .001) than in those with ACC+SA (F = 3.03, df = 1, p = .093). Despite this difference, the ratings of mental effort are distributed at a rather low range of the rating scale (Baseline: M = 65.47, SD = 42.26, ACC: M = 43.75, SD = 28.06; ACC+SA: M = 52.34, SD = 37.61). The ratings of mental effort are not influenced by age or prior experience with ACC. As apparent from Fig. 5 (right panel), the drivers reported a high level of driving safety (M = 11.69, SD = 2.49). However, the safety ratings differ among the age groups when the automation level is taken into account (interaction ‘‘Automation level ⁄ Age”, see Table 9): Younger drivers report a marginally significantly lower level of driving safety with the ACC+SA system than older drivers (t = 1.71, df = 30, p = .097). During non-assisted driving (t = 1.52, df = 30, p = .139) and driving with ACC (t = 0.35, df = 30, p = .729), there is no such difference in reported safety among the age groups. ACC experience has no influence on the safety ratings. 4. Summary and discussion The main objective of this on-road experiment was to investigate the impact of different levels of vehicle automation on participants’ engagement in a secondary task while driving. The study was conducted using a commercially available vehicle equipped with the assistance systems Adaptive Cruise Control (ACC) and Adaptive Cruise Control with steering assistance

78

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

Rating of perceived safety

Rating of perceived effort 15 200

10

[0-15]

[0-220]

150

100

5 50

< 50 years > 50 years

0

0 BL

ACC

BL

ACC+SA

ACC

ACC+SA

Level of automation

Level of automation

Fig. 5. Rating of mental effort (left panel) and driving safety (right panel); mean and standard error are depicted; BL: Baseline; ACC: Adaptive Cruise Control; ACC+SA: Adaptive Cruise Control with steering assistance.

Table 9 Inferential statistics of the ANOVA for the independent variables ‘‘Rating of mental effort” and ‘‘Rating of driving safety”. Effect

Automation level Automation level ⁄ Age Automation level ⁄ ACC experience Automation level ⁄ Age ⁄ ACC experience Age ACC experience Age ⁄ ACC experience

Mental effort

Safety

F

df1

df2

p

g

9.39 0.88 1.83 0.13 0.40 0.13 0.04

2 2 2 2 1 1 1

27 27 27 27 28 28 28

.001 .425 .180 .877 .531 .717 .844

0.41 0.06 0.12 0.01 0.01 0.01 0.00

2

F

df1

df2

p

g2

0.33 4.34 0.19 0.87 2.20 0.10 0.20

2 2 2 2 1 1 1

27 27 27 27 28 28 28

.720 .023 .830 .430 .150 .753 .655

0.02 0.24 0.01 0.06 0.07 0.00 0.01

(ACC+SA). N = 32 participants of different age groups and with varying degrees of prior experience with ACC drove in daily rush-hour traffic on a highway segment under three automation conditions: non-assisted (Baseline), Driver assistance (ACC) and Partial automation (ACC+SA). A growing body of research suggests that an increase in the level of vehicle automation accompanies an increased level in secondary task activities while driving (e.g., de Winter et al., 2014). Possible reasons for this include an increased level of subjective safety (Fuller, 1984; Näätänen & Summala, 1976; Wilde, 1988), an overreliance on the capabilities of the respective assistance system (Rudin-Brown & Noy, 2002; Weller & Schlag, 2004; Lee & See, 2004; Wickens & Xu, 2002), a decrease in workload due to the assistance system (de Winter et al., 2014) and a decrease in the drivers’ situation awareness due to the increased automation level (de Winter et al., 2014; Schömig et al., 2011). The main research aim of this study was to investigate the circumstances that would cause drivers to change their attention distribution away from the primary driving task and direct it towards the secondary task. The results of this study can be summarized as follows:  An increased level of secondary task engagement was indeed found for both assisted drives, but only among participants with prior experience with ACC. Furthermore, a strong dependency of secondary task engagement on velocity was evident, with tasks mainly being executed during low-speed driving. Additionally, older drivers generally performed fewer secondary tasks than younger drivers.  Independent of age and prior experience with ACC, drivers reported a lower level of mental effort associated with the secondary tasks when driving with ACC than with non-assisted driving. Regarding ACC+SA, a marginally significant reduction in mental effort was reported.  No decrease was found in primary task performance (as measured by the maximum longitudinal deceleration and maximum lateral acceleration) or subjective safety that would have been possible because of the higher level of task engagement during assisted drives. The findings from the on-road experiment thus validate and extend previous research addressing secondary task engagement and vehicle automation. First, findings from more-controlled research environments (cf. de Winter et al., 2014 for an overview) that driving with ACC (automation level: Driver assistance) can lead to an increase in secondary task engagement

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

79

were replicated in an on-road setting. Second, it was evident that the same is true with regard to the ACC+SA (automation level: Partial automation). However, the increase in secondary task engagement was not more pronounced than when driving with ACC only. Furthermore, participants adjusted their level of engagement in the secondary tasks to the traffic state, as secondary task engagement decreased with increasing velocity regardless of the automation level. The study results indicate that the drivers’ tactical decisions regarding when to engage in a secondary task were evident during both non-assisted and assisted drives, validating findings from recent simulator studies (Jamson et al., 2013; Kircher et al., 2014) and FOTs (Metz et al., 2014). This decrease in secondary task engagement may be attributed to a decreased level of subjective safety or subjective workload during higher driving speeds (Lewis-Evans et al., 2011; Fuller, 2005) and can thus be considered situationadaptive behaviour (Schömig et al., 2011). Consequently, the results demonstrate that with Partial automation – as with ACC – an increased level of secondary task engagement is evident, but as speed increases, drivers focus their attention on the primary driving task. This situation-adaptive behaviour may explain why a decrease in primary task performance or subjective safety due to the increased level of secondary task engagement was not found. At this point, some limitations of the study have to be noted:  It must be emphasized that the results of the current study cannot be generalized to higher levels of vehicle automation. For example, Highly automated driving would permit drivers to fully disengage from the driving task, as long as they are not requested to take back manual control of the vehicle in case of system limits. In the current study on Partially automated driving, drivers were still part of the control loop and were not allowed to direct their attention completely away from driving. It can thus be expected that the effects of Highly automated driving on the drivers’ willingness to perform secondary activities during automated driving will differ from our findings.  One shortcoming of the study is that no time-based, objective measures of situation criticality were collected during the on-road trials. In assessing driving safety, only measures of driving performance (acceleration, deceleration) and the participants’ safety ratings were used. A more exhaustive picture would have been available if measures such as time-tocollision or time headway had been recorded. To a certain extent, these measures would also allow one to quantify and evaluate the quality of interaction with other road users.  Another limitation is that local traffic density was not measured directly. All drives were conducted in rush hour traffic in the same geographical area to achieve comparable levels of traffic density (i.e., high traffic density due to rush hour traffic). Nevertheless, variations in traffic density may have influenced the results independently of driving velocity, which may weaken the internal validity of the study. As no objective measure of real-time traffic density was collected (e.g., measurement by detection loops or camera systems), it is not possible to account for this possibility in the current study. With regard to the study design, some methodological issues have to be discussed. Concerning the secondary task used in this study, the following points may weaken the generalizability of the results:  Instruction: It is possible that the increased level of secondary task engagement was partly due to the instructions given to the participants. It is possible that they would not have performed any task at all when driving without any instruction. To control this possible instructional effect, we specifically instructed the participants to only perform the tasks when they judged the driving situation to be safe enough and that not performing the task would not lead to any disadvantages for them. It may also be possible that drivers would engage more heavily in secondary activities when driving on their own. The presence of an experimenter may have altered the participants’ behaviour because they might have felt that it would not be socially acceptable to perform such tasks while driving. Taken together, it must be assumed that the setting of the study may have influenced the results towards lower or higher levels of task engagement.  Task characteristics: Furthermore, the degree to which the study results can be generalized may be limited to the specific secondary task used in this study. The secondary task consisted of a self-paced and rather simple visual-manual task that could be interrupted at any time. It may thus be possible that the results cannot be generalized to more-complex secondary activities that would require the driver to neglect the primary driving task for longer time periods, for example, browsing or text messaging. Additionally, it must be asked whether performing the selected task was sufficiently motivating. It is quite possible that the drivers would have engaged more heavily in processing a secondary task if it would have been more interesting or important for them to complete the tasks. With regard to this, it is an open question of why drivers would choose to perform such a task, even if the task itself is not motivating. It can be speculated that the intrinsic motivation to complete the tasks was to break the boredom and monotony that arise when driving is not performed fully manually, as boredom and inactivity are usually experienced as unpleasant (Wilson et al., 2014).  Usage of driver assistance: Based on the experimental design used in this study, it cannot be ruled out that drivers actually would not use the investigated systems. Because drivers were instructed to activate the respective systems when entering the highway and were asked not to deactivate the system deliberately, it is possible that the results of the study represent an experimental artefact and would thus not translate to everyday driving because of this specific requirement to always use the assistant systems.  Effects of moderator variables: Another methodological issue of this study pertains to our finding that only those drivers with prior experience with ACC increased their level of secondary task engagement during assisted drives. It is important to note that this result cannot be interpreted unambiguously, as the participants may differ in factors other than prior experience with vehicle automation. In this respect, one shortcoming of the study is that the drivers’ experience with

80

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

secondary tasks in general (e.g., smart phone or tablet usage) and while driving (e.g., texting or phoning during driving) were not considered as alternative explanations. On the other hand, all drivers participated an extensive practice period during the study. All of them had to complete 40 tasks prior to the main study. This was done to control practice effects with the secondary task used in the on-road trials. Despite these limitations, the results obtained in the on-road experiment during non-assisted drives (i.e., the frequency of secondary activities decreases with increasing speed) match those reported during an FOT (Metz et al., 2014), which indicates the ecological validity of the results. 5. Conclusion According to previous findings from controlled simulator experiments and test track studies, an increase in the level of vehicle automation may cause an increase in secondary activities while driving. These findings have been validated and extended in the current on-road experiment through the following three findings:  An increase in secondary task engagement was found only among drivers with prior experience with vehicle automation (i.e., using ACC on a regular basis).  Secondary task engagement when driving with Partial automation (ACC + steering assistance) was not more pronounced than when driving with Driver assistance (ACC).  Despite these possibly negative side effects of vehicle automation, participants still exhibited situation-adaptive behaviour, as they decreased the frequency of secondary tasks in both assisted and non-assisted drives with increasing vehicle speed. The third point, in particular, may attenuate concerns that increased vehicle automation will cause a possible dangerous shift of attention away from the primary driving task, especially as we did not find a decrease in subjective safety or primary task performance. References Bach, K. M., Jaeger, M. G., Skov, M. B., & Thomassen, N. G. (2009). Interacting with in-vehicle systems: Understanding, measuring, and evaluating attention. In A. F. Blackwell (Ed.), Proceedings of HCI 2009. People and computers XXIII. Celebrating people and technology (pp. 453–462). Cambridge, United Kingdom: British Computer Society. Bailey, N., & Scerbo, M. (2007). Automation-induced complacency for monitoring highly reliable systems: The role of task complexity, system experience, and operator trust. Theoretical Issues in Ergonomics Science, 8(4), 321–348. Brown, T. L., Lee, J. D., & McGehee, D. V. (2000). Attention based model of driver performance in rear-end-collisions. Transportation Research Record, 1724, 14–21. Burns, P. C., Knabe, E., & Tevell, M. (2000). Driver behavioral adaptation to collision warning and avoidance information. Human Factors and Ergonomics Society Annual Meeting Proceedings, 44(20), 315. Carsten, O. M., Kircher, K., & Jamson, S. (2013). Vehicle-based studies of driving in the real world: The hard truth? Accident Analysis & Prevention, 58, 162–174. Carsten, O. M., Lai, F. C. H., Barnard, Y., Jamson, A. H., & Merat, N. (2012). Control task substitution in semiautomated driving: Does it matter what aspects are automated? Human Factors, 54(5), 747–761. Cooper, J. M., Vladisavljevic, I., Medeiros-Ward, N., Martin, P. T., & Strayer, D. L. (2009). An investigation of driver distraction near the tipping point of traffic flow stability. Human Factors, 51(2), 261–268. Davidse, R. J. (2007). Assisting the older driver: Intersection design and in-car devices to improve the safety of the older driver (Doctoral thesis), Rijksuniversiteit Groningen, Netherlands. de Winter, J., 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, 27B, 196–217. de Winter, J., van Leeuwen, P. M., & Happee, R. (2012). Advantages and disadvantages of driving simulators: A discussion. In A. J. Spink, F. Grieco, O. E. Krips, L. W. S. Loijens, L. P. J. J. Noldus, & P. H. Zimmermann (Eds.), Proceedings of the 8th international conference on methods and techniques in behavioral research (pp. 47–50). Utrecht, The Netherlands: Noldus Information Technology bv. Desimone, R., & Duncan, J. (1995). Neural mechanism of selective visual attention. Annual Review of Neuroscience, 18, 193–222. Dragutinovic, N., Brookhuis, K. A., Hagenzieker, M. P., & Marchau, V. (2005). Behavioural effects of advanced cruise control use. A meta-analytic approach. European Journal of Transport and Infrastructure Research, 5(4), 267–280. Eilers, K., Nachreiner, F., & Hänecke, K. (1986). Entwicklung und Überprüfung einer Skala zur Erfassung subjektiv erlebter Anstrengung. Zeitschrift für Arbeitswissenschaft, 40(4), 215–224. Endsley, M. R., & Kaber, D. B. (1999). Level of automation effects on performance, situation awareness and workload in a dynamic control task. Ergonomics, 42(3), 462–492. Ferdinand, A. O., & Menachemi, N. (2014). Associations between driving performance and engaging in secondary tasks: A systematic review. American Journal of Public Health, 104(3), 39–48. 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. Fuller, R. (1984). A conceptualization of driving behaviour as threat avoidance. Ergonomics, 27(11), 1139–1155. Fuller, R. (2005). Towards a general theory of driver behaviour. Accident Analysis & Prevention, 37(3), 461–472. Gasser, T. & Westhoff, D. (2012). BASt-study: Definitions of automation and legal issues in Germany. In Paper presented at the workshop on the future of road vehicle automation, Irvine, CA. Gasser, T., Arzt, C., Ayoubi, M., Bartels, A., Buerkle, L., Eier, J., et al (2012). Rechtsfolgen zunehmender Fahrzeugautomatisierung (Berichte der Bundesanstalt für Straßenwesen, Reihe F84). Bremerhaven: Wirtschaftsverlag NW. Gold, C., Damböck, D., Lorenz, L., & Bengler, K. (2013). ‘‘Take over!” How long does it take to get the driver back into the loop? Human Factors and Ergonomics Society Annual Meeting Proceedings, 57, 1938–1942.

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

81

Green, P. (1999). Estimating compliance with the 15-second rule for driver-interface usability and safety. Human Factors and Ergonomics Society Annual Meeting Proceedings, 43(18), 987–991. Hoedemaeker, M., & Brookhuis, K. A. (1998). Behavioural adaptation to driving with an Adaptive Cruise Control (ACC). Transportation Research Part F, 1(2), 95–106. Horberry, T., Anderson, J., Regan, M. A., Triggs, T. J., & Brown, J. (2006). Driver distraction: The effects of concurrent in-vehicle tasks, road environment complexity and age on driving performance. Accident Analysis & Prevention, 38(1), 185–191. Horrey, W., & Wickens, C. D. (2007). In-vehicle glance duration: Distributions, tails, and model of crash risk. Transportation Research Record, 2018, 22–28. Jamson, A. H., Merat, N., Carsten, O. M., & Lai, F. C. H. (2013). Behavioural changes in drivers experiencing highly-automated vehicle control in varying traffic conditions. Transportation Research Part C, 30, 116–125. Jamson, A. H., Westerman, S. J., Hockey, G. R. J., & Carsten, O. M. (2004). Speech-based e-mail and driver behavior: Effects of an in-vehicle message system interface. Human Factors, 46(4), 625–639. Kircher, K., Larsson, A. F., & Hultgren, J. (2014). Tactical driving behavior with different levels of automation. IEEE Transactions on Intelligent Transportation Systems, 15(1), 158–167. Klauer, S. G., Dingus, T. A., Neale, V. L., Sudweeks, J. D., & Ramsey, D. J. (2006). The impact of driver inattention on near-crash/crash risk: An analysis using the 100-car naturalistic driving study data. Washington DC: U.S. National Highway Traffic Safety Administration. Koornstra, M. J. (1993). Safety relevance of vision research and theory. In A. G. Gale (Ed.), Vision in Vehicles IV (pp. 3–13). Amsterdam: North-Holland. Larsson, A. F. L. (2012). Driver usage and understanding of adaptive cruise control. Applied Ergonomics, 43(3), 501–506. Larsson, A. F. L., Kircher, K., & Hultgren, A. J. (2014). Learning from experience: Familiarity with ACC and responding to a cut-in situation in automated driving. Transportation Research Part F, 27B, 229–237. Lee, J. D., & See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human Factors, 46(1), 50–80. Lewis-Evans, B., de Waard, D., & Brookhuis, K. A. (2011). Speed maintenance under cognitive load. Implications for theories of driver behaviour. Accident Analysis & Prevention, 43(4), 1497–1507. Ma, R., & Kaber, D. B. (2005). Situation awareness and workload in driving while using adaptive cruise control and a cell phone. International Journal of Industrial Ergonomics, 35(2005), 939–953. Metz, B., Landau, A., & Just, M. (2014). Frequency of secondary tasks in driving. Results from naturalistic driving data. Safety Science, 68, 195–203. Muhrer, E., Reinprecht, K., & Vollrath, M. (2012). Driving with a partially autonomous forward collision warning system how do drivers react? Human Factors, 54(5), 698–708. Muhrer, E., & Vollrath, M. (2010). Expectations while car following. The consequences for driving behaviour in a simulated driving task. Accident Analysis & Prevention, 42(6), 2158–2164. Näätänen, R., & Summala, H. (Eds.). (1976). Road-user behavior and traffic accidents. Amsterdam: North-Holland. Naujoks, F. & Neukum, A. (2014). Timing of in-vehicle advisory warnings based on cooperative perception. In D. D. Waard, K. Brookhuis, R. Wiczorek, F. D. Nocera, R. Brouwer, P. Barham, C. Weikert, A. Kluge, W. Gerbino, & A. Toffetti (Eds.), Proceedings of the human factors and ergonomics society Europe chapter annual meeting (pp. 193–206). Naujoks, F., Mai, C., & Neukum, A. (2014). The effect of urgency of take-over requests during highly automated driving under distraction conditions. In T. Ahram, W. Karowski, & T. Marek (Eds.), Proceedings of the 5th international conference on applied human factors and ergonomics AHFE 2014 (pp. 2099– 2106). Krakow: AHFE Conference. Naujoks, F., & Totzke, I. (2014). Behavioral adaptation caused by predictive warning systems. The case of congestion tail warnings. Transportation Research Part F, 26, 49–61. NHTSA (2012). Visual-manual NHTSA driver distraction guidelines for in-vehicle electronic devices. Notice in the Federal Register, 77. NHTSA (2013). Preliminary statement of policy concerning automated vehicles. Washington, DC: National Highway Traffic Safety Administration. Parasuraman, R. (2000). Designing automation for human use: Empirical studies and quantitative models. Ergonomics, 43(7), 931–951. Parasuraman, R., & Riley, V. (1997). Humans and automation: Use, misuse, disuse, abuse. Human Factors, 39(2), 230–253. Parasuraman, R., Sheridan, T. B., & Wickens, C. D. (2000). A model for types and levels of human interaction with automation. IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans, 30(3), 286–297. Rakauskas, M. E., Gugerty, L. J., & Ward, N. J. (2004). Effects of naturalistic cell phone conversations on driving performance. Journal of Safety Research, 35(4), 453–464. Rogers, M., Zhang, Y., Kaber, D., Liang, Y., & Gangakhedkar, S. (2011). The effects of visual and cognitive distraction on driver situation awareness. In D. Harris (Ed.), Engineering Psychology and Cognitive Ergonomics (pp. 186–195). Heidelberg: Springer. Rudin-Brown, C. M., & Noy, I. (2002). Investigation of behavioral adaptation to lane departure warnings. Transportation Research Record, 1803(1), 30–37. Rudin-Brown, C. M., & Parker, H. A. (2004). Behavioural adaptation to Adaptive Cruise Control (ACC): Implications for preventive strategies. Transportation Research Part F, 7(2), 59–76. Saad, F. (2006). Some critical issues when studying behavioural adaptations to new driver support systems. Cognition, Technology & Work, 8(3), 175–181. Schömig, N., Metz, B., & Krüger, H.-P. (2011). Anticipatory and control processes in the interaction with secondary tasks while driving. Transportation Research Part F, 14(6), 525–538. Schömig, N., Schoch, S., Neukum, N., Schumacher, M., & Wandtner, B. (2015). Simulatorstudien zur Ablenkungswirkung fahrfremder Tätigkeiten. (Berichte der Bundesanstalt für Straßenwesen, Reihe Mensch und Sicherheit, Heft M253). Bremen: Carl Schünemann Verlag. Siebert, F. W., Oehl, M., & Pfister, H.-R. (2014). The influence of time headway on subjective driver states in adaptive cruise control. Transportation Research Part F, 25A, 65–73. Simões, A., & Pereira, M. (2009). Older drivers and new in-vehicle technologies: Adaptation and long-term effects. In M. Kurosu (Ed.), Human centered design (pp. 552–561). Heidelberg: Springer. Stanton, N. A., & Marsden, P. (1996). From fly-by-wire to drive-by-wire: Safety implications of automation in vehicles. Safety Science, 24(1), 35–49. Straßenverkehrsordnung [German road traffic code], §23 (2013). Strayer, D. L., & Drew, F. A. (2004). Profiles in driver distraction: Effects of cell phone conversations on younger and older drivers. Human Factors, 46(4), 640–649. Suen, L., Mitchell, C., & Henderson, S. (1998). Application of intelligent transportation systems to enhance vehicle safety for elderly and less able travellers. In National Highway Safety Advisory Committee (Ed.), Proceedings of the 16th international technical conference on experimental safety vehicles (pp. 386– 394). Washington: NHTSA. TomTom International B.V. (2013). TomTom European Congestion Index. . Retrieved 29.10.13. Totzke, I., Naujoks, F., Mühlbacher, D., & Krüger, H.-P. (2012). Precision of congestion warnings: Do drivers really need warnings with precise information about the congestion tail´s position? In D. Waard, A. H. Jamson, Y. Barnard, & O. M. Carsten (Eds.), Human factors of systems and technology (pp. 235–248). Maastricht: Shaker Publishing. van der Hulst, M., Meijman, T. F., & Rothengatter, J. A. (1999). Anticipation and the adaptive control of safety margins in driving. Ergonomics, 42(2), 336–345. Vrkljan, B. H., & Miller-Polgar, J. (2005). Advancements in vehicular technology: Potential implications for the older driver. International Journal of Vehicle Information and Communication Systems, 1, 88–105. Wege, C., Will, S., & Victor, T. (2013). Eye movement and brake reactions to real world brake-capacity forward collision warnings. A naturalistic driving study. Accident Analysis & Prevention, 58, 259–270. Weller, G., & Schlag, B. (2004). Verhaltensadaptation nach Einführung von Fahrerassistenzsystemen. In B. Schlag (Ed.), Verkehrspsychologie. Mobilität – Verkehrssicherheit – Fahrerassistenz (pp. 351–370). Lengerich: Pabst Science Publishing.

82

F. Naujoks et al. / Transportation Research Part F 38 (2016) 67–82

Wickens, C. D. (1984). Processing resources in attention. In R. Parasuraman & D. R. Davies (Eds.), Varieties of attention (pp. 63–102). New York: Academic Press. Wickens, C. D., & Xu, X. (2002). Automation trust, reliability and attention HMI 02-03. Illinois: Illinois University at Urbana-Champaign. Wiczorek, R., & Manzey, D. (2014). Supporting attention allocation in multitask environments. Effects of likelihood alarm systems on trust, behavior, and performance. Human Factors, 56(7), 1209–1221. Wiedemann, K., Schömig, N., Mai, C., Naujoks, F., & Neukum, A. (2015). Drivers’ monitoring behaviour and interaction with non-driving related tasks during driving with different automation levels. In Paper presented at the 6th international conference on applied human factors and ergonomics (AHFE), Las Vegas, USA. Wierwille, W. W. (1993). Visual and manual demands of in-car controls and displays. In B. Peacock & W. Karwowski (Eds.), Automotive ergonomics (pp. 299–320). London: Taylor and Francis. Wilde, G. (1988). Risk homeostasis theory and traffic accidents: Propositions, deductions and discussion of dissension in recent reactions. Ergonomics, 31(4), 441–468. Wilson, T. D., Reinhard, D. A., Westgate, E. C., Gilbert, D. T., Ellerbeck, N., Hahn, C., et al (2014). Just think: The challenges of the disengaged mind. Science, 345 (6192), 75–77. Wu, Y., & Boyle, L. N. (2015). Drivers’ engagement level in Adaptive Cruise Control while distracted or impaired. Transportation Research Part F, 33, 7–15. Young, K. L., Regan, M. A., & Lee, J. D. (2009). Factors moderating the impact of distraction on driving performance and safety. In M. A. Regan, K. L. Young, & J. D. Lee (Eds.), Driver distraction: Theory, effects and mitigation (pp. 335–354). Boca Raton, FL: CRC Press. Young, K. L., Salmon, P. M., & Cornelissen, M. (2013). Missing links? The effects of distraction on driver situation awareness. Safety Science, 56, 36–43.