Autonomous emergency braking systems adapted to snowy road conditions improve drivers' perceived safety and trust

ABSTRACT

Objective: This study investigated drivers' evaluation of a conventional autonomous emergency braking (AEB) system on high and reduced tire–road friction and compared these results to those of an AEB system adaptive to the reduced tire–road friction by earlier braking. Current automated systems such as the AEB do not adapt the vehicle control strategy to the road friction; for example, on snowy roads. Because winter precipitation is associated with a 19% increase in traffic crashes and a 13% increase in injuries compared to dry conditions, the potential of conventional AEB to prevent collisions could be significantly improved by including friction in the control algorithm. Whereas adaption is not legally required for a conventional AEB system, higher automated functions will have to adapt to the current tire–road friction because human drivers will not be required to monitor the driving environment at all times. For automated driving functions to be used, high levels of perceived safety and trust of occupants have to be reached with new systems. The application case of an AEB is used to investigate drivers' evaluation depending on the road condition in order to gain knowledge for the design of future driving functions.

Methods: In a driving simulator, the conventional, nonadaptive AEB was evaluated on dry roads with high friction (μ = 1) and on snowy roads with reduced friction (μ = 0.3). In addition, an AEB system adapted to road friction was designed for this study and compared with the conventional AEB on snowy roads with reduced friction. Ninety-six drivers (48 males, 48 females) assigned to 5 age groups (20–29, 30–39, 40–49, 50–59, and 60–75 years) drove with AEB in the simulator. The drivers observed and evaluated the AEB's braking actions in response to an imminent rear-end collision at an intersection.

Results: The results show that drivers' safety and trust in the conventional AEB were significantly lower on snowy roads, and the nonadaptive autonomous braking strategy was considered less appropriate on snowy roads compared to dry roads. As expected, the adaptive AEB braking strategy was considered more appropriate for snowy roads than the nonadaptive strategy. In conditions of reduced friction, drivers' subjective safety and trust were significantly improved when driving with the adaptive AEB compared to the conventional AEB. Women felt less safe than men when AEB was braking. Differences between age groups were not of statistical significance.

Conclusions: Drivers notice the adaptation of the autonomous braking strategy on snowy roads with reduced friction. On snowy roads, they feel safer and trust the adaptive system more than the nonadaptive automation.

KEYWORDS:

• Autonomous emergency braking
• adaptation to road friction
• safety
• trust
• gender
• age

Introduction

Worldwide, up to 50 million people suffer injuries and more than 1.2 million people die yearly because of road traffic crashes (World Health Organization 2015). In the United States, adverse weather was associated annually with more than 1.5 million vehicle crashes, which result in about 800,000 injuries and 7,000 fatalities (National Research Council 2004). Statistical analyses of real accident data show that snowfall contributes to higher accident rates (Andrey et al. 2003; Eisenberg and Warner 2005; Khattak and Knapp 2001; Shankar et al. 1995). Winter precipitation irrespective of type (e.g., snow, freezing rain, ice pellets, or sleet) was associated with a 19% increase in traffic crashes and a 13% increase in injuries compared to dry conditions (Black and Mote 2015).

Road conditions with reduced friction (e.g., wet, snowy, icy surfaces) contribute to a higher accident rate, although the severity of these accidents varies among gender and age groups (Black and Mote 2015; Morgan and Mannering 2011; Myers et al. 2011; Ulfarsson and Mannering 2004). In adverse weather, older drivers showed an increased likelihood of involvement in property damage accidents only but not in those involving injuries; the reason for this may be due to their greater experience as drivers (Black and Mote 2015). The statistics show that women are less frequently involved in traffic accidents than men (Statistisches Bundesamt (Federal Statistical Office of Germany) 2016; Oltedal and Rundmo 2006; Pulido et al. 2016; Rhodes and Pivik 2011; Scott-Parker and Oviedo-Trespalacios 2017; Tavris et al. 2001). Research also shows age and gender differences in the braking behavior that could explain differences in the accident rates (Kusano et al. 2015; Li et al. 2016; Montgomery et al. 2014). In a naturalistic driving experiment, Montgomery et al. (2014) found that women apply the brakes on average 1.3 s earlier than men and drivers younger than 30 years brake 1.7 s later than drivers older than 30 years. However, there is also research on braking behavior that shows no gender differences (Hancock et al. 2002). Research shows that drivers can adapt to snow by reducing speed and maintaining a longer headway on snowy roads compared to dry roads (Kilpeläinen and Summala 2007). Drivers can use visual cues to anticipate a reduced level of friction and to adapt the driving style before kinesthetic cues signalize the reduced friction (Öberg 1978; Wallman 1997).

The role of automation in preventing the occurrence and reducing the severity of accidents

The deployment of automated driving systems is considered a key measure to reduce the number of accidents and improve road safety (European Commission 2011; European Road Transport Research Advisory Council 2015; U.S. Department of Transportation 2016; World Health Organization 2015). Autonomous emergency braking (AEB) systems were first introduced to the market starting around 2006. The braking intervention of an AEB is coupled with a visual/acoustic/haptic warning when the collision risk is over a certain level determined by the time to collision (Euro NCAP 2015). All AEB systems in production minimize the risk of false positives (e.g., braking in wrong situations). A reliable, robust, and accurate road friction measurement is not currently available (Lex 2015). Engineers thus design systems for high friction. Where collision risk is detected, low times to collision (TTCs) are used. Warnings are typically given between 1.5 and 2.5 s TTC and brake engagement typically between 0.8 and 1.2 s TTC.

Research has shown the potential of AEB to avoid crashes or to reduce the severity of crashes. The AEB is one of the 5 best rated systems with the potentials to prevent 22% of fatal accidents (Eichberger et al. 2010). An analysis of real crash data showed 38% fewer rear-end crashes in vehicles equipped with AEB than in similar vehicles without AEB (Fildes et al. 2015). It was shown that AEB was effective at reducing 40% of fatally injured and 27% of severely injured pedestrians in frontal collisions with cars having a 40° field of view (Rosén et al. 2010). However, research also shows potentially negative effects of automation on the road safety, such as drivers' excessive trust in automation, complacency, and inability to take over the control when required that contribute to accidents (Shen and Neyens 2014). Trust in automation is an attitude that automation will help achieve a goal in conditions of vulnerability and uncertainty (Lee and See 2004). The appropriate level of trust depends on the true capabilities of automation (Kidd et al. 2017). Trust affects the automation use or lack of use (Parasuraman and Riley 1997). If drivers have low trust in automation, as in the case when automation produces many false alarms, they may not use it.

Automated driving functions of SAE level 1 or 2 (SAE International 2014) that are available in series-production vehicles do not require tire–road friction adaption, because the driver is legally responsible to monitor the driving environment at all times and adapt the driving style accordingly. For SAE level 3 and above, the automated function will be required to adapt to the road condition. Within the design of such adaptive functions, both physical and human factors have to be considered, such as imperfect sensor information and occupant's perception of safety and trust in automated functions at different road conditions (Lex et al. 2017). In this study, it was shown that drivers mainly rely on visual cues as well as the vehicle's response when estimating the road conditions, according to their own statements. Being asked for categories they used to differentiate road conditions, 24% of the drivers mentioned dry, 22% wet, 19% icy, and 13% snowy, showing a similar distribution in different age and gender groups (Lex et al. 2017).

To further understand human factors for tire–road friction adaptive driving strategies, a conventional AEB City system is used as an application case. AEB City typically works at lower speeds (10–50 km/h), being designed for urban application (Euro NCAP 2015). The braking interventions of the AEB are activated at a certain TTC, which is calculated by Equation (1)(1) . (1) where Δs is the relative distance and Δv is the relative speed between the ego and the target vehicle. The TTC values currently used by the conventional AEB systems at SAE level 1 (SAE International 2014) can avoid accidents on dry roads with high friction. This is appropriate because the AEB is supposed to react at the latest possible time if the driver fails to react. For autonomous braking interventions on roads with low friction at an SAE level of 3 and higher (SAE International 2014), the TTC values must be higher; for example, TTC/μ in the case of braking.

If the design of automation aims at compensating for human limitations in driving, especially in conditions of reduced friction, then age and gender differences clearly need to be considered. The perceived usefulness, acceptance, and an appropriate level of trust in automated assistance systems by drivers are essential prerequisites for purchasing and using them. Acceptance depends on the fail-safe characteristics of automated systems and on drivers' comfort (Brookhuis et al. 2001). Research shows that advanced driver assistance systems could be improved if the control algorithms included individual preferences and environmental conditions (Koglbauer et al. 2017; Xiong and Boyle 2012). Theoretically, the greatest potential of the driver assistance systems to reduce the number of accidents would be in conditions of limited road friction (Lex et al. 2013; Niederkofler et al. 2011). The increase in similarities between automated and human driving behaviors was found to improve the acceptance of automation (Van Driel et al. 2007). Automated systems thus need a control strategy adapted to the road conditions similar to that of human drivers. In addition, automated systems should be able to assist drivers in perceiving earlier, better understanding, and reacting more appropriately to prevent accidents and reduce the severity of accidents that cannot be avoided. Drivers should be able to trust the automation system and feel safe with it.

The objective of this study is to evaluate adaptive and conventional (nonadaptive) vehicle control strategies of AEB with female and male drivers from different age groups in a driving simulator (simulated high vs. reduced friction and summer vs. winter sceneries). The systems tested are the conventional AEB system that ignores the road friction and a new adaptive AEB with a generic algorithm that includes the road friction and adapts the timing of the braking intervention in a manner to avoid the collision.

Research questions

•	Hypothesis 1: Drivers will notice that the automated braking strategy of the conventional AEB system is not appropriate on snowy roads with reduced friction.
•	Hypothesis 2: Drivers feel less safe and trust the conventional AEB less on snowy roads with reduced friction than on dry roads with high friction. Thus, this hypothesis refers to the conventional AEB in 2 road conditions: Dry and snowy roads.
•	Hypothesis 3: Drivers will notice the difference between adaptive and nonadaptive braking strategies on snowy roads with reduced friction.
•	Hypothesis 4: On snowy roads with reduced friction, drivers will feel safer and have more trust in the adaptive AEB system compared to the conventional AEB system. Thus, this hypothesis refers to the conventional and adaptive AEB in snowy road conditions.

An additional research question refers to age and gender differences in the evaluation of adaptive and conventional AEB in conditions of high and reduced friction.

Method

Participants

The participants were recruited on a volunteer basis, using a newspaper announcement. Ninety-six drivers (48 women) participated in the study. Ten women and 10 men were recruited in each of the age groups 20–29, 30–39, 40–49, and 50–59 years and 8 women and 8 men in the age group 60–75 years. The participants were selected to obtain a similar driving activity between the gender groups within each age group and among the age groups. Males had a mean driving activity of 18,323 km/year and females had a mean driving activity of 18,094 km/year. Each participant was informed about the purpose and procedure of the experiment and signed an informed consent form. This research complied with the tenets of the Declaration of Helsinki.

Equipment

The simulator is a fixed-base full-vehicle mock-up of a Mini Countryman with an autostereoscopic vision system. Sound simulation includes engine, road, and wind noise of the simulated vehicle and other traffic. The pitch at the onset of braking was simulated in the visual system. The drivers were thus able to notice the onset of a braking maneuver. The summer condition used high friction (μ = 1) and showed a completely dry road in a landscape with green grass and trees. The winter scenery used reduced friction (μ = 0.3) and showed a road and landscape partly covered with snow. The coefficient of friction μ = 0.3 is within the typical range for snowy road condition (Gustafsson 1997).

The AEB City function was investigated in the present article in the typical setting of a rear-end collision between the simulated vehicle and the preceding target vehicle in a stopped driving state. The setup was validated with real driving tests. The braking of the conventional AEB was initiated at 1.2 s TTC, together with a visual/acoustic driver warning. The braking of the adaptive AEB was initiated at 1.9 s TTC, together with a visual and acoustic driver warning. Thus, the adaptive AEB brakes earlier on snow than the nonadaptive AEB. The drivers were asked to drive in the city with a speed of 8 m/s. In case of an imminent rear-end collision, the participants were instructed to not react but to observe the collision avoidance actions of the AEB.

Procedure

Each participant received a written briefing about the conventional and adaptive AEB and a familiarization session with the simulator on a dry road with high friction (μ = 1). Subsequently, the participants drove and observed the AEB in 3 rear-end collision scenarios:

•	Conventional AEB, summer scenery, high friction (μ = 1).
•	Conventional AEB, winter scenery, reduced friction (μ = 0.3).
•	Adaptive AEB, winter scenery, reduced friction (μ = 0.3).

Dependent measures

After each experimental block, the drivers evaluated the automated control actions of the AEB. Drivers could choose one of these categories for evaluating the timeliness of the AEB's reaction: Appropriate, too early, or too late. Deceleration was evaluated using the categories appropriate, too strong, and too weak. In addition, the drivers rated their safety and trust in the AEB system on a numerical scale ranging from 1 (very low) to 6 (very high). The usefulness of the AEB was rated after each maneuver block using the categories counterproductive, not useful, neutral, and useful.

Data analysis

Chi-square^{Table 1}tests were calculated for testing differences between frequencies (hypotheses 1 and 3). Analysis of variance was used to evaluate the effect of road conditions on driving with the conventional AEB on drivers' safety and trust in automation (hypothesis 2). For the second hypothesis, the within-subjects factor was the road condition (dry vs. snowy). Analysis of variance was used to evaluate the effect of AEB type on drivers' safety and trust in automation (hypothesis 4). For the fourth hypothesis, the within-subjects factor was the type of AEB (conventional vs. adaptive). There were 2 between-subjects factors: Age and gender. Alpha was set at .05.

Results

Evaluation of the conventional AEB in conditions of high and reduced friction

Significantly more drivers (99%) were of the opinion that the conventional AEB reacted too late on snowy compared to dry roads (54%, χ² = 12.08, df = 1, P < .001). Significantly fewer drivers (18%) considered that the deceleration of the conventional AEB was appropriate on snowy compared to dry roads (73%, χ² = 32.29, df = 1, P < .001). In addition, significantly fewer drivers (36%) considered the AEB useful on snowy compared to dry roads (88%, χ² = 21.18, df = 1, P < .001). Differences between gender and age groups in the evaluation of automated control strategy of the conventional AEB did not reach a statistical level of significance.

Drivers' safety and trust in the conventional AEB in conditions of high and reduced friction

The results of univariate analysis of variance showed a significant effect of the road friction on drivers' safety, F(1,85) = 220.10, P < .0001, η² = 0.72, and trust in the conventional AEB, F(1,85) = 92.46, P < .0001, η² = 0.52. As the total scores in Table 1 show, the ratings of safety and trust were lower when using the conventional AEB on snowy compared to dry roads. A significant difference between the gender groups emerged when compared on the issues of safety, F(1,85) = 7.49, P < .008, η² = 0.08, with lower ratings for safety among female compared to male drivers (Table 1). The distribution of safety ratings across age and gender groups is illustrated in Figure 1. Gender differences in trust were not of statistical significance. Differences among the age groups in safety and trust were also not of significance.

Table 1. Drivers' subjective ratings of safety and trust in the conventional and adaptive AEB.^a

AEB type	Conventional AEB		Conventional AEB		Adaptive AEB
Tire–road grip	High friction (µ = 1)		Reduced friction (µ = 0.3)		Reduced friction (µ = 0.3)
Descriptive data	M	SD	M	SD	M	SD
Safety ratings
Male drivers	5.19	1.21	2.25	1.77	5.21	1.05
Female drivers	4.36	1.78	1.70	1.28	4.49	1.46
Total	4.78	1.57	1.98	1.56	4.85	1.31
Trust in automation
Male drivers	4.06	1.73	2.27	1.70	4.60	1.33
Female drivers	3.83	1.75	2.26	1.69	4.21	1.57
Total	3.95	1.73	2.26	1.69	4.41	1.46

^a The scale ranged from 1 (very low) to 6 (very high).

Figure 1. Mean safety ratings for the conventional AEB on dry and snowy roads. Error bars show standard deviations.

Evaluation of the adaptive and conventional AEB in conditions of reduced tire–road friction

In snowy road conditions, significantly fewer drivers (34%) considered that the adaptive AEB reacted too late compared to the conventional AEB (99%, χ² = 30.50, df = 1, P < .001). Significantly more drivers (88%) considered that the deceleration of the adaptive AEB was appropriate compared to the conventional AEB (18%, χ² = 44.45, df = 1, P < .001). In addition, significantly more drivers (92%) considered the adaptive AEB useful in snowy road conditions compared to the conventional AEB (36%, χ² = 23.21, df = 1, P < .001). Differences between gender and age groups in the evaluation of automated control strategy of the adaptive and conventional AEB did not reach statistical significance.

Drivers' safety and trust in the conventional and adaptive AEB in conditions of reduced friction

The results of univariate analysis of variance showed a significant effect of AEB type on drivers' safety, F(1,85) = 231.07, P < .0001, η² = 0.73, and trust in automation, F(1,85) = 152.42, P < .0001, η² = 0.64. As the total scores in Table 1 show, the ratings of safety and trust were higher when using the adaptive AEB compared to the conventional AEB on snowy roads. A significant difference between genders emerged in the comparison of safety, F(1,85) = 8.45, P < .005, η² = 0.09, with lower ratings for safety among females than males (Table 1). The distribution of safety ratings across age and gender groups is illustrated in Figure 2. Gender differences in trust and age differences in safety and trust were not of statistical significance.

Figure 2. Mean safety ratings for the conventional AEB and adaptive AEB on snowy roads with reduced friction. Error bars show standard deviations.

Discussion

Compared to dry conditions, winter precipitation such as snow is associated with a 19% increase in traffic crashes and a 13% increase in injuries (Black and Mote 2015). Adaption to the road friction is not legally required for conventional AEB systems, but it could significantly improve their potential to prevent collisions. Human factors such as occupant's trust and safety in automated functions at different road conditions have to be considered for the design of adaptive functions at SAE level 3 and above. The aim of this study was to assess drivers' evaluations of the automated braking actions of a conventional and an AEB adapted to road friction with 96 drivers in a driving simulator and to specify areas of improvement. Although automated driving systems could have a higher potential for preventing accidents on roads with reduced friction (e.g., wet, snowy, icy surfaces), the currently available driver assistance systems are designed to function effectively only in conditions of high friction (e.g., dry surfaces).

The first hypothesis stating that drivers will notice that the automated braking strategy of the conventional AEB is not appropriate on snowy roads with reduced friction was confirmed. More drivers considered that the conventional AEB applied the brakes too late and too weakly on snowy roads (reduced friction) compared to on dry roads (high friction). Fewer drivers considered the conventional AEB useful on snowy roads compared to dry roads. Drivers trusted the conventional AEB less and felt less safe with the conventional AEB on snowy roads compared to dry roads. Therefore, the second research hypothesis was also confirmed.

The adaptive AEB evaluated in this study considered the road friction and applied the brakes earlier, at 1.9 s TTC. On snowy roads, more drivers considered that the adaptive AEB was useful compared to the conventional AEB. In addition, fewer drivers considered that the adaptive AEB applied the brakes too late and too weak on snowy roads compared to the conventional AEB. Drivers trusted the adaptive AEB more and felt safer with the adaptive AEB than with the conventional AEB on snowy roads. The third and fourth research hypotheses were thus confirmed. The conventional AEB failed to adapt its braking strategy to snowy roads with reduced friction by braking earlier, as human drivers generally tend to do (Kilpeläinen and Summala 2007). In contrast, the new adaptive AEB decelerated earlier than the conventional AEB on snowy roads and could better meet drivers' expectations.

This study shows no significant age and gender differences in the evaluation of either the conventional or the adaptive AEB in terms of usefulness and control strategy, despite reports of age and gender differences in the estimation of time and distance to collision (Koglbauer 2015; Koglbauer et al. 2015), driving/ braking strategy (Kusano et al. 2015; Montgomery et al. 2014), and in the frequency and severity of traffic accidents (Statistisches Bundesamt (Federal Statistical Office of Germany) 2016; Oltedal and Rundmo 2006; Pulido et al. 2016; Rhodes and Pivik 2011; Scott-Parker and Oviedo-Trespalacios 2017; Tavris et al. 2001). However, female drivers felt less safe than male drivers when observing the braking reactions of the conventional AEB and adaptive AEB.

If the AEB is intended to contribute to reducing the number and the severity of crashes, then its braking strategy should be adapted to the road friction and aim for collision avoidance in all road conditions. Given that the conventional AEB's reaction is later than that of human drivers in an emergency situation of an imminent crash, its contribution to preventing accidents and reducing crash severity could be further improved. A high priority in the AEB design process was given to the reduction of false positive reactions (e.g., unnecessarily decelerating the vehicle or decelerating too early on dry surfaces). It is assumed that drivers would deactivate or refuse to purchase automated systems with frequent false positives. However, both the results of this study and the European roadmap policy toward automated driving (European Road Transport Research Advisory Council 2015) indicate that future driving strategies need to be calculated that consider the dynamic interaction of the traffic participants in different environmental conditions (e.g., different coefficients of friction).

More research is necessary to understand drivers' processing of anticipated friction for redesign of the AEB. Research findings suggest that human drivers use visual cues indicating reduced friction (e.g., snowy surface) and adapt their driving strategy before the kinesthetic cues signalize the reduced friction (Öberg 1978; Wallman 1997). Lex et al. (2017) showed that drivers mainly rely on visual cues as well as the vehicle's response when estimating the road conditions, being able to distinguish various categories of the road friction: Dry, wet, icy, and snowy. Notwithstanding future challenges for the improvement of adaptive automated systems, this study highlights the potential of the AEB to prevent collisions and meet driver expectations by including the friction in the automated control algorithm. Taking into account that higher levels of automation will release the driver from her or his current duty to monitor the environment including road conditions, the importance of ongoing research in estimation of road friction with respect to accuracy, reliability, and robustness is emphasized from the human factors point of view presented in this study.

In summary, past AEB systems have been frequently investigated for their potential benefit in accident avoidance and collision severity mitigation. However, in order to avoid false-positive braking interventions, the braking strategy of conventional AEB systems is designed for late braking assuming high friction. This resulted in low TTC values for braking initiation. Adaptation to road friction will be crucial for future automated systems. This article addresses human factors that are relevant for the design of these functions. In order to investigate the human–machine interaction of AEB systems, we conducted a driver simulator study to compare conventional AEB systems and AEB systems adapted to the road condition. The experimental evaluation with 96 drivers shows that the reported potential benefits of AEB can be further improved by including road friction in the automated braking algorithm in an adaptive AEB braking strategy. The study focused on human factors in automated driving and showed a significant increase in drivers' perceived safety, trust, and usefulness of the AEB. Female drivers felt less safe than male drivers when observing the AEB, but gender differences in the evaluation of the AEB control strategy did not reach statistical significance. Different levels of perceived safety and trust in the AEB can affect traffic injury prevention. If drivers perceive the AEB as not useful or not safe or trustworthy, they may not purchase it for their car or they may deactivate it. Thus, they would decide against a safety system that may not be effective in avoiding a vehicle collision but could diminish the impact forces and occupant injury in case of a crash. We expect that awareness about both the perceived and objective limitations of the nonadaptive AEB on roads with reduced friction will motivate the development and implementation of an adaptive AEB in due time. An adaptive AEB will have greater potential to prevent traffic crashes and injuries.

Additional information

Funding

This work was financially supported by the Austrian Federal Ministry for Transportation, Innovation and Technology, the Austrian Research Promotion Agency, FEMtech Program “Talents,” Grant No. 3413253. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the article.