Driver trust in five driver assistance technologies following real-world use in four production vehicles

ABSTRACT

Objectives: Information about drivers' experiences with driver assistance technologies in real driving conditions is sparse. This study characterized driver interactions with forward collision warning, adaptive cruise control, active lane keeping, side-view assist, and lane departure warning systems following real-world use.

Methods: Fifty-four Insurance Institute for Highway Safety employees participated and drove a 2016 Toyota Prius, 2016 Honda Civic, 2017 Audi Q7, or 2016 Infiniti QX60 for up to several weeks. Participants reported mileage and warnings from the technologies in an online daily-use survey. Participants reported their level of agreement with five statements regarding trust in an online post-use survey. Responses were averaged to create a composite measure of trust ranging from −2 (strongly disagree) to +2 (strongly agree) for each technology. Mixed-effect regression models were constructed to compare trust among technologies and separately among the study vehicles. Participants' free-response answers about what they liked least about each system were coded and examined.

Results: Participants reported driving 33,584 miles during 4 months of data collection. At least one forward collision warning was reported in 26% of the 354 daily reports. The proportion of daily reports indicating a forward collision warning was much larger for the Honda (70%) than for the Audi (18%), Infiniti (15%), and Toyota (10%).

Trust was highest for side-view assist (0.98) and lowest for active lane keeping (0.20). Trust in side-view assist was significantly higher than trust in active lane keeping and lane departure warning (0.53). Trust in active lane keeping was significantly lower than trust in adaptive cruise control (0.67) and forward collision warning (0.71). Trust in adaptive cruise control was higher for the Audi (0.72) and Toyota (0.75) compared with the Honda (0.30), and significantly higher for the Infiniti (0.93). Trust in Infiniti's side-view assist (0.58) was significantly lower than trust in Audi (1.17) and Honda (1.23) systems. Coding of answers to free-response questions showed that more than 80% of complaints about Honda's adaptive cruise control were about the way it functioned and/or performed. Infiniti's side-view assist was the only one with complaints mentioning circumstances where it was used. Trust in forward collision warning, lane departure warning, and active lane keeping was not significantly different among vehicles.

Conclusions: Driver trust varied among driver assistance technologies, and trust in adaptive cruise control and side-view assist differed among vehicles. Trust may affect real-world use of driver assistance technologies and limit the opportunity for the systems to provide their intended benefits.

KEYWORDS:

• Trust
• driver assistance systems
• crash avoidance
• advanced vehicle technology

Introduction

Driver assistance technologies are becoming increasingly prevalent in the vehicle fleet. Front crash prevention, side-view assist, and lane departure warning are available as standard or optional features on more than half of 2017 model year vehicles (Highway Loss Data Institute 2016). Driving automation technologies such as adaptive cruise control and active lane keeping also are available to consumers. Driver assistance technologies have the potential to prevent many crashes (Jermakian 2011) and already are demonstrating their efficacy in some cases (Cicchino 2017; Sternlund et al. 2017). However, the ability of some systems to reach their full promise depends on how drivers use and understand them.

Systems cannot be effective if drivers do not use them. Observed and self-reported evidence demonstrates that drivers do not always keep assistance technologies turned on and that use varies considerably by system type (Braitman et al. 2010; Cicchino and McCartt 2015; Eichelberger and McCartt 2014; 2016; Flannagan et al. 2016; Reagan and McCartt 2016). For instance, Reagan and McCartt (2016) reported that even though nearly all Honda vehicles with forward collision warning observed at service centers had that system turned on, lane departure warning was activated in only 32% of vehicles.

Drivers' trust in vehicle systems likely affects whether they choose to use them (Parasuraman and Riley 1997). Studies of prototype driver assistance technologies tested in simulators have found that features such as intervention timing and warning reliability can affect drivers' trust in the systems (e.g., Abe and Richardson 2006; Lees and Lee 2007; Rajaonah et al. 2006), but the performance of production systems in real traffic can differ substantially from that of a prototype in a simulator. Data from a European field operational test have provided some preliminary evidence of how trust in different collision avoidance systems used in real-world conditions differs, by system type (Sanchez et al. 2012), but there is a dearth of evidence on how real-world vehicle performance affects trust and how trust varies among different implementations of the same types of systems.

The goal of the current study was to examine driver trust in driver assistance technologies in production vehicles following real-world use. Trust was compared among different types of systems (forward collision warning, adaptive cruise control, side-view assist, lane departure warning, active lane keeping) and among different implementations of the same types of systems. The attributes drivers liked least about the systems were analyzed to gauge why some systems were trusted more than others or why trust varied among vehicles.

Method

Participants

Participants were employees of the Insurance Institute for Highway Safety, an independent, nonprofit scientific and educational organization based in the United States that is supported by automobile insurers and insurance associations. Employees were recruited from both the Ruckersville and Arlington office locations in Virginia. Ruckersville is a rural community just outside of Charlottesville, Virginia, the nearest major city. Arlington is an urban community located in the greater Washington, DC, metropolitan area.

Each employee received an electronic invitation to voluntarily participate in the study. Fifty-four (26 males, 28 females) of the 108 employees contacted volunteered to participate. Participants were 26–70 years old with a mean age of 42 years. More participants were recruited from the Ruckersville location than from the Arlington location, but the sample demographics were similar (Table 1). Each participant had a valid driver's license, and no incentives were offered to participate.

Table 1. Sample demographics overall and by location.

	Age				Gender
Location	Mean	SD	Minimum	Maximum	M	F
Arlington	41.6	8.8	28	70	10	12
Ruckersville	41.7	10.5	25	64	20	12
Overall	41.7	9.8	25	70	30	24

Vehicles

Participants drove up to four different vehicles: a 2017 Audi Q7 large four-door SUV (sports utility vehicle), a 2016 Honda Civic small four-door sedan, a 2016 Toyota Prius small four-door hatchback, and a 2016 Infiniti QX60 large four-door SUV. Each vehicle was equipped with various advanced driver assistance and crash avoidance systems, but only the systems examined in the current study are discussed.

Each vehicle was equipped with a front crash prevention system that included forward collision warning and automatic emergency braking, a lane departure warning system, and an adaptive cruise control system that could bring the vehicle to a complete stop. The Audi and Honda were equipped with active lane keeping that applied torque to the steering wheel to assist with lane maintenance above 40 mph (Audi) and 45 mph (Honda). The Audi also had a traffic jam assist feature that provided active lane keeping and car following at slower speeds.

The Infiniti, Audi, and Honda each were equipped with side-view assist. The Audi and Infiniti systems notified the driver of a vehicle present or quickly approaching in an adjacent lane using a warning light in the A-pillar (Infiniti) or side mirror housing (Audi). The Infiniti system also warned audibly when a vehicle was detected and the turn signal was activated. The Honda system provided a rear-facing camera image of the lane on the passenger side of the vehicle. The image was displayed when the right turn indicator was activated or when the system was manually activated by the driver.

Forward collision warning, lane departure warning, and side-view assist communicated audible, visual, and/or haptic warning information to the driver. Table 2 summarizes the warning modalities for these systems in each vehicle. Detailed descriptions of the warning interfaces and strategies used to communicate information to the driver are beyond the scope of this study.

Table 2. Warning modality for the forward collision warning, lane departure warning, and side-view assist systems in each vehicle.

Vehicle	Forward collision warning	Lane departure warning	Side-view assist
Honda Civic	A, V	V, H	V
Audi Q7	A, V, H	V, H	V
Infiniti QX60	A, V	A, V	A,V
Toyota Prius	A, V	A, V	—

Note. A, audible warning. V, visual warning. H, haptic warning. Prevention mechanisms are not described.

Procedure

Each participant signed a vehicle use policy document that outlined the appropriate use of study vehicles (e.g., adherence to traffic laws, refrain from cell phone use and from operating under the influence). By signing the policy, participants also agreed to keep the technologies on and use them as often as possible when driving the vehicles. Participants were permitted to modify settings for each technology if desired.

Study personnel contacted participants to schedule time periods where a study vehicle could be used as a personal vehicle with the goal of having every participant drive at least one of the four vehicles. Before receiving a vehicle, each participant was familiarized with the vehicle technologies. First, participants were shown the locations of various sensors, cameras, and displays on the exterior and interior of the vehicle, and the purpose of each was described. Next, participants sat in the driver seat and adjusted the seat and vehicle mirrors to their liking. Then the researcher described the purpose, function, and operation of each technology. The researcher also described the limitations of the technology (e.g., adaptive cruise control failing to detect stationary vehicles) and emphasized the importance of remaining engaged in the driving task and not overly relying on the technology.

Participants were shown how to activate, deactivate, and adjust the settings for each technology, and how these actions corresponded with changes to the status and information displays in the instrument panel. The activation status of the technologies when the vehicle was started varied. Some technologies were active when the vehicle started, some had to be activated by the driver, and some maintained their last activation setting on startup. When reviewing each technology, the researcher made sure that each technology setting and status was returned to a similar state before the vehicle was taken by the participant. All available warning modes were activated and warning sensitivity settings were typically set to the middle setting. The earliest warning or intervention setting was selected when only two sensitivity settings were available. The vehicle owner's manual was stored in each vehicle as a reference for participants.

Next, participants completed a short familiarization drive with the research assistant in the front passenger seat. The purpose of the drive was to give participants an opportunity to use and experience most of the systems equipped on the vehicle, including adaptive cruise control, lane departure warning, side-view assist, and active lane keeping. These familiarization drives at both office locations included slower speed local roads and higher speed highways or interstates. The researcher instructed participants to use various technologies and perform maneuvers that would result in warnings or interventions from the technologies. For example, after merging onto a highway, participants were instructed to set the adaptive cruise control and adjust the gap and speed settings when there was a vehicle ahead. The researcher also instructed participants to depart a lane in a controlled manner to experience the lane departure warning. Forward collision warning and automatic emergency braking were not intentionally demonstrated during the familiarization drive, but any system warnings or interventions that occurred were used as an instructional opportunity. In total, the familiarization drive lasted about 30 min.

Participants were instructed to use the study vehicle as their personal vehicle while it was in their possession. Participants were instructed to complete an online daily-use survey each day they used the vehicle and then a post-use survey after returning the vehicle. Participants were aware that the purpose of the study was to collect information about their experiences and interactions with the technologies equipped on the study vehicles, and were encouraged to be as descriptive as possible when completing the surveys.

Survey instruments

The daily-use survey collected information about the total number of miles driven, driving conditions (e.g., proportion of driving time spent on high-speed roads, proportion of driving time with precipitation of any kind), and safety-relevant events such as near-crashes, warnings or interventions from the crash avoidance systems, and unfavorable or unexpected interactions with each technology. Participants were asked to indicate whether they received 0, 1, or 2 or more forward collision warnings when driving the vehicle that day. Additional questions asked participants to report the circumstances when they received a warning, whether they understood the warning, whether they found it useful, and their opinions about the warning characteristics. Similar questions were administered about the lane departure warning and side-view assist systems. Only information about the total number of miles driven and days with a forward collision warning was analyzed.

The post-use survey collected more detailed information about participants' experiences, changes in driving behavior, trust in each technology, ease of use of each technology, and whether they understood and what they thought about the various interfaces and settings. Five statements were used to assess drivers' trust in each technology by having participants indicate their level of agreement or disagreement with a statement using a 5-point Likert scale with the response choices “strongly disagree,” “disagree,” “neutral,” “agree,” and “strongly agree.” The statements were selected from a 12-item scale developed by Jian et al. (2000) to assess a general propensity to trust an automated system. The statements were:

1.	I am suspicious of the [system name]'s intent, action, or outputs.
2.	The [system name]'s actions will have a harmful or injurious outcome.
3.	The [system name] is dependable.
4.	The [system name] is reliable.
5.	I can trust the [system name].

Responses to each statement were assigned numerical values ranging from −2 (strongly disagree) to 2 (strongly agree). The first and second statements were reverse coded so that positive scores reflected more trust in the technology and negative scores indicated less trust.

Participants were asked to describe what they liked most and least about each technology in a free-response format. The objective of these questions was to gather more detailed information about participant's experiences with the technologies and gain insight into their responses to the five statements of trust. The other post-use survey questions were outside the scope of the current study and were not analyzed.

Data analysis

Responses to the^{Table 3} five statements about trust were combined to create a single measure of trust for each technology. Cronbach's alpha was computed to assess whether the five statements reliably measured the same construct. A Cronbach's alpha value greater than or equal to 0.80 is considered adequate for research tools (Streiner 2003).

Mixed-effect regression models were constructed to compare trust among technologies and trust in each of the five technologies separately among the study vehicles. Depending on the model, trust was modeled with the fixed effect of technology (adaptive cruise control, forward collision warning, lane departure warning, active lane keeping, side-view assist) or study vehicle (Honda Civic, Audi Q7, Infiniti QX60, Toyota Prius). Driver was included as a random effect in every model to account for within-subject variance from repeated observations. An exchangeable correlation structure was assumed for the repeated measurements from each driver in each model. Type 3 tests were performed to determine if the set of variables that made up the fixed effect of technology or study vehicle was statistically significant at the 0.05 level. Least square means estimates and 95% confidence intervals were computed to examine pairwise differences among technologies or study vehicles. Pairwise comparisons were adjusted using the Tukey method to control for alpha inflation. Modeling was performed using the PROC MIXED procedure in SAS 9.4.

Responses to free-response questions in the post-use survey about what participants liked most and least about each technology were coded to provide insight into the specific technology attributes that might have influenced trust. Two study authors (DK and LK) coded whether each response mentioned the functionality and/or performance of the technology, the user interface, circumstances in which the technology operated, or provided no information (Table 3). Responses that were neutral, were not directed at the correct technology, or were inconsistent with expressing like or dislike as posed in the question were not coded in any of the four categories. More than one attribute could be coded for each response.

Table 3. Operational definitions of attributes that participant's liked or disliked about a given technology.

Attribute	Operational definition	Example comments about adaptive cruise control
Functionality/performance	Describes specific function(s) or range of functions that the technology performed or how the technology performed it	“It tended to slow the vehicle down later than I would as the driver.”
User interface	Mentions how easy or difficult it was to use or understand the technology; system controls, displays, and/or settings were referenced; comments about the way information was communicated (e.g., warning modalities) or interactions with the system (e.g., fighting against input).	“The distance control was set to the farthest distance by default.”
Circumstance	Indicates specific condition(s) or situation(s) where the technology was or would be used	“Not for city or congested roads.”
None	No stated likes or dislikes or no response	“I did not experience any negatives.”

Results

Participants used the study vehicles 80 total times. Twenty-nine participants used one vehicle, 24 participants used two vehicles, and one participant used three vehicles. Participants used study vehicles for different durations, ranging from a single trip to multiple weeks. Three vehicle uses were for a single trip, 49 uses were for 1–5 days, and 28 uses were for more than 5 days. Participants reported driving 33,584 total miles across 354 total daily reports during the 4 months of data collection from March 19 to July 25, 2016; hence, the average number of daily reports per 1,000 miles of reported travel was 10.6. Total reported mileage was greatest for the Audi (14,488 miles), followed by the Infiniti (9,152 miles), Honda (5,411 miles), and Toyota (4,534 miles). Daily mileage ranged from 3 to 872 miles. Based on responses to the daily-use reports, on average 57% of participants' driving time was spent on freeways or interstates, 92% was during daylight hours, 83% was during dry conditions, 95% was during clear visibility, and 79% was during free-flowing traffic conditions. Driving conditions reported in the daily-use reports were similar across vehicles.

Daily reports were reviewed to examine warning frequency among vehicles. At least one forward collision warning was indicated in 26% of all daily reports. Seventy percent of the 69 Honda daily reports indicated a forward collision warning. This was a much larger proportion than indicated for the other vehicles—only 18% of the 114 Audi daily reports, 15% of the 93 Infiniti daily reports, and 10% of the 78 Toyota daily reports indicated at least one forward collision warning.

Trust in technology

The internal consistency among the five statements used to create a trust score was acceptable for each technology (Cronbach's α > 0.87). Driver trust among the five technologies ranged from −1.8 to 2.0 with a mean of 0.67 (SD = 0.80). The range in trust scores was largest for adaptive cruise control (−1.8 to 2.0), followed by active lane keeping (−1.2 to 2.0), lane departure warning (−1.2 to 2.0), forward collision warning (−1 to 2.0), and side-view assist (−0.4 to 2.0).

A mixed-effect regression analysis indicated that trust was significantly different among technologies (F(4,279) = 7.78, p < 0.001). Trust was highest for side-view assist and lowest for active lane keeping (Figure 1). Trust in side-view assist was significantly greater than trust in active lane keeping (p < 0.001) and lane departure warning (p < 0.05). Trust in active lane keeping was significantly lower than trust not only in side-view assist but also in adaptive cruise control (p < 0.001) and forward collision warning (p < 0.01). The difference in trust between active lane keeping and lane departure warning approached significance (p = 0.06).

Figure 1. Least squares means estimate for driver trust in various driver assistance and crash avoidance technologies. Error bars indicate 95% confidence intervals.

Next, trust in each technology was compared among study vehicles. The effect of study vehicle on driver trust in adaptive cruise control approached statistical significance (F(3,75) = 2.66, p = 0.05). Driver trust in Honda's adaptive cruise control was substantially lower than trust in Infiniti's system (Table 4).

Table 4. Least squares means estimate and 95% confidence intervals for driver trust in various technologies by study vehicle.

Vehicle	Adaptive cruise control	Forward collision warning	Lane departure warning	Side-view assist	Active lane keeping
Honda Civic	0.30 [–0.05,0.65]	0.54 [0.22, 0.86]	0.62 [0.25, 0.99]	1.23 [0.89, 1.56]	0.42 [–0.18, 1.02]
Audi Q7	0.72 [0.40, 1.04]	0.91 [0.55, 1.27]	0.55 [0.18, 0.92]	1.17 [0.91, 1.43]	0.14 [–0.39, 0.68]
Infiniti QX60	0.93 [0.54, 1.32]	0.77 [0.34, 1.21]	0.33 [–0.05, 0.71]	0.58 [0.27, 0.89]	—
Toyota Prius	0.75 [0.35, 1.14]	0.99 [0.45,1.54]	0.83 [0.42, 1.23]	—	—

Note. — indicates the technology was not equipped.

Driver trust in side-view assist was significantly different among the Honda, Audi, and Infiniti systems (F(2,56) = 7.69, p < 0.01). Trust in Infiniti's side-view assist was significantly lower than trust in Honda and Audi systems (p < 0.01). Trust in forward collision warning, trust in lane departure warning, and trust in active lane keeping were not significantly different among vehicles.

Free-response questions

Cohen's kappa (Cohen 1960) was computed for each attribute category and technology combination to assess the interrater reliability between coders. Kappa values were above 0.60 for every attribute and technology combination, indicating substantial agreement or better between coders. The two coders reviewed and reconciled the remaining coding discrepancies, and the attributes that participants liked least about each technology were analyzed.

Complaints about each system varied (Table 5). About two-thirds of participants complained about the functionality and/or performance of adaptive cruise control. Half complained about functionality and/or performance of lane departure warning. Participants frequently complained about the functionality and/or performance of active lane keeping, but complaints about the user interface were most common. In contrast, fewer than half of participants complained about forward collision warning or side-view assist.

Table 5. Percentage of complaints reported in free-response questions about the attributes of each technology overall and by study vehicle.

	Attribute
Technology and study vehicle	Functionality/performance	User interface	Circumstance	None
Adaptive cruise control
Honda Civic (n = 21)	81	5	52	14
Audi Q7 (n = 26)	69	23	35	4
Infiniti QX60 (n = 18)	56	28	44	17
Toyota Prius (n = 15)	60	33	53	13
Overall	68	21	45	11
Forward collision warning
Honda Civic (n = 21)	62	33	10	19
Audi Q7 (n = 26)	15	15	0	65
Infiniti QX60 (n = 18)	11	22	6	61
Toyota Prius (n = 15)	0	7	0	93
Overall	24	20	4	58
Lane departure warning
Honda Civic (n = 21)	62	38	0	38
Audi Q7 (n = 26)	46	23	12	46
Infiniti QX60 (n = 18)	61	56	22	0
Toyota Prius (n = 15)	27	40	7	47
Overall	50	38	10	34
Side-view assist
Honda Civic (n = 21)	33	24	0	52
Audi Q7 (n = 26)	27	12	0	65
Infiniti QX60 (n = 18)	39	33	22	28
Overall	32	22	6	51
Active lane keeping
Honda Civic (n = 21)	43	33	33	33
Audi Q7 (n = 26)	58	73	39	0
Overall	51	55	36	15

Note. More than one attribute may be coded for each response.

Complaints about adaptive cruise control and side-view assist in specific vehicles provided some insight into why driver trust in these systems varied. More than 80% of complaints about Honda's adaptive cruise control mentioned the functionality and/or performance of the system. Many participants said Honda's adaptive cruise control made late and harsh changes to vehicle speed; similar comments were not made about the other vehicles' systems. More than half of participants who drove the Honda and nearly two-thirds who drove the Audi did not complain about side-view assist. Complaints were more common for Infiniti's side-view assist, especially about the circumstances where the system was used. A common complaint about Infiniti's side-view assist was that it gave false alerts when passing guardrails, static roadside objects (e.g., mailboxes, poles), and roadside terrain.

Discussion

The objective of the current study was to compare perceptions of trust in various driver assistance technologies and in implementation of the same system among different vehicles after the technology was used on actual roads. Drivers do not always use production driver assistance technologies (Braitman et al. 2010; Cicchino and McCartt 2015; Eichelberger and McCartt 2014; 2016; Flannagan et al. 2016; Reagan and McCartt 2016), and lack of trust may be one reason for nonuse. In this study, trust varied both among different types of technology and among different implementations of the same technology when used in real driving conditions. Participants trusted side-view assist most and active lane keeping least. Among system implementations, participants trusted Honda's adaptive cruise control less than Infiniti's system, and trusted Infiniti's side-view assist less than Honda's and Audi's systems.

Trust in automation and other technology is a strong predictor of its use (Muir and Moray 1996; Parasuraman and Riley 1997). Participants' higher trust in side-view assist than in other systems in the current study mirrors the large proportion of owners who report they always keep these systems turned on and would want them in their next vehicle (Braitman et al. 2010; Cicchino and McCartt 2015). Yet trust differed by implementation even for this system that was most trusted overall. Participants trusted Infiniti's side-view assist least and reported disliking certain situations where it was used, mainly because of false alerts. This is similar to findings from studies of prototype forward collision warning and lane departure warning, where driver trust was diminished by false alerts from less reliable systems (e.g., Lees and Lee 2007; Rudin-Brown and Noy 2002). Infiniti's side-view assist differed from the other vehicles' systems in that it provided an auditory alarm when a vehicle in the blind spot was detected and the turn signal was activated. Infiniti's side-view assist may have issued more false warnings than Audi's system, or Infiniti's auditory warning may have drawn more attention to false alerts than occurred with Audi's visual-only warning.

Past research has shown that drivers find lane departure warning annoying (Braitman et al. 2010; Cicchino and McCartt 2015; Eichelberger and McCartt 2014; 2016), and the current study found that they also do not strongly trust this system either. Annoyance and lukewarm feelings of trust might explain why lane departure warning systems are deactivated more often than other warning systems (Flannagan et al. 2016; Reagan and McCartt 2016). Active lane keeping presents an opportunity to be a more widely used alternative than lane departure warning for keeping drivers in their lanes because this system does not issue warnings that drivers may find annoying. However, drivers in the current study trusted active lane keeping the least, which may undermine real-world use by consumers (e.g., Muir and Moray 1996; Parasuraman and Riley 1997). Additional research is needed to compare consumers' use of active lane keeping systems with lane departure warning systems and the reasons for disuse (e.g., annoyance, trust). Observing vehicles brought in for service at dealerships (Reagan and McCartt 2016) and exploring telematics-based data from vehicles equipped with these and other advanced vehicle technologies (Flannagan et al. 2016) are two promising methods for performing this research.

Specific system characteristics also can influence trust. For example, participants' trust in Honda's forward collision warning was not rated significantly lower than trust in the same systems on the other vehicles, but it was rated the lowest. Participants received far more warnings from Honda's forward collision warning than from the other vehicles' systems. Drivers reported disliking that Honda's system issued warnings too early. Abe and Richardson (2005; 2006) found that drivers have greater trust in prototype systems that issue warnings earlier rather than later (Abe and Richardson 2005; 2006), but have less tolerance for early warnings issued in more demanding driving conditions. Drivers may have trusted Honda's forward collision warning system less if it warned more frequently than other systems in demanding situations such as in heavier traffic. Drivers also complained that Honda's system was too sensitive and provided warnings unexpectedly and in situations where it was not needed. Drivers also may have trusted Honda's system less if they thought it did not reliably indicate a true collision threat (Parasuraman and Riley 1997).

Trust may be especially important to ensure drivers' correct use of and interactions with driving automation technologies. Participants had moderate levels of trust in the two driving automation technologies examined in the current study: adaptive cruise control and active lane keeping. Participants frequently complained about adaptive cruise control's functionality and performance (e.g., saying that changes to vehicle speed were harsh and late). This is similar to what has been reported in survey and focus-group studies of owners of vehicles with adaptive cruise control (de Winter et al. 2017; Larsson 2012; Strand et al. 2011). These complaints were most frequent for Honda's system, which also was the least trusted adaptive cruise control. More than half of participants complained about the functionality/performance and user interface of active lane keeping. This included complaints about inconsistent recognition and tracking of lane markings and that steering inputs from the system were inappropriate or discomforting. As driving automation technologies become increasingly available in production vehicles, they will need to function in ways drivers expect and have intuitive interfaces that encourage driver engagement, ensure appropriate reactions to the systems, and provide understanding of system disengagement.

A limitation of the current study is that participants were employees of a nonprofit highway safety research organization in the United States. The experiences and opinions of the vehicle technologies reported in this study may not reflect those from the general driving population in the United States and may not reflect the opinions of drivers in other countries. Another limitation is that most participants only experienced the technologies for a short time. Trust is a dynamic process that changes over time, and can increase as drivers develop a better understanding of a technology's capability and learn to cope its limitations (Beggiato and Krems 2013; Lee and See 2004; Muir and Moray 1996). However, this process occurs quickly. Beggiato et al. (2015) found that trust in adaptive cruise control developed rapidly in a group of drivers with no prior knowledge of or experience with adaptive cruise control and stabilized after about 3.5 h of driving with the system. In this study, almost every vehicle was used for at least 1 day with over half of vehicles being used for more than 5 days, so many participants' perceptions of trust in each technology may have been well developed. Finally, the findings are based on self-reports, so caution must be used when interpreting reported miles traveled and forward collision warnings.

In conclusion, drivers did not exhibit strong trust in any driver assistance technology, and trust varied by system type and system implementation following use in real-world driving conditions. Drivers may elect not to use systems they do not trust, thereby limiting the opportunities for systems to provide their intended benefits. This study demonstrates that existing driver assistance technologies can be improved to impart greater trust among drivers and increase confidence during use.

Acknowledgments

The authors thank Tiera Matthew and Andrew Reagan for assisting with data collection. They also thank their colleagues at the Insurance Institute for Highway Safety for reviewing and commenting on earlier drafts of this article.

Funding

This work was supported by the Insurance Institute for Highway Safety.