Abstract
Objective
The current study examined whether differences in the branding and description or mode of training materials influence drivers’ understanding and expectations of a partial driving automation system.
Background
How technology is described might influence consumers’ understanding and expectations, even if all information is accurate.
Method
Ninety drivers received training about a real partial driving automation system with a fictitious name. Participants were randomly assigned to a branding condition (system named AutonoDrive, training emphasized capabilities; or system named DriveAssist, training emphasized limitations) and training mode (quick-start brochure; video; or in-person demonstration). No safety-critical information was withheld nor deliberately misleading information provided. After training, participants drove a vehicle equipped with the system. Associations of drivers’ expectations with branding condition and training mode were assessed using between-subjects comparisons of questionnaire responses obtained pre- and post-drive.
Results
Immediately after training, those who received information emphasizing the system’s capabilities had greater expectations of the system’s function and crash avoidance capability in a variety of driving scenarios, including many in which the system would not work, as well as greater willingness to utilize the system’s workload reduction benefits to take more risks. Most but not all differences persisted after driving the vehicle. Expectations about collision avoidance differed by training mode pre-drive but not post-drive.
Conclusion
Training that emphasizes a partial driving automation system’s capabilities and downplays its limitations can foster overconfidence.
Application
Accuracy of technical information does not guarantee understanding; training should provide a balanced view of a system’s limitations as well as capabilities.
Keywords: mental models, vehicle automation, surface transportation, driver behavior, training evaluation
Introduction
Advanced driver assistance systems (ADAS) and driving automation technologies allow vehicles to perform some aspects of the driving task without direct driver input. These encompass a range of capabilities—from warning the driver about a potential rear-end collision, to controlling lane position and following distance from the vehicle ahead for an extended period—and significantly change the way drivers interact with their vehicles. Safe use of these systems depends on the driver adequately understanding the systems’ capabilities and limitations.
Research has shown that drivers are often not aware of the capabilities and limitations of new vehicle technologies (Jenness et al., 2008; Braitman et al, 2010; McDonald et al., 2017), and often receive no formal, structured training prior to use (Lubkowski et al., 2021). Marketing materials, social media, and other information sources may negatively influence proper understanding of such systems (e.g., Abraham et al., 2017a; Beggiato & Krems, 2013; Dixon, 2020; Seppelt & Victor, 2020). This is concerning because drivers with deficient understanding of ADAS or automation technologies may exhibit poor and potentially unsafe performance when the system behaves in ways that they do not expect or understand (Gaspar et al., 2020).
Feature names and their associated branding may have the potential to mislead drivers regarding the extent to which the operator versus the system is responsible for safe operation (AAA, 2019; Abraham et al., 2017b; Teoh, 2020). This is particularly concerning due to well-documented cognitive biases that dictate that first impressions form quickly and can be difficult to overcome. Numerous studies have documented an anchoring effect, in which people tend to form an initial judgment based on the first information they receive, even if it is of dubious relevance (e.g., Tversky & Kahneman, 1974). Relatedly, people often give excessive weight to their initial judgments, failing to revise them adequately in response to conflicting information obtained later (e.g., Bruner & Potter, 1964; Jones et al., 1968; Nickerson, 1998), as processing information that conflicts with an initial hypothesis requires more cognitive effort than processing information that reinforces it (Einhorn & Hogarth, 1981). Thus, naming and branding that emphasize a system’s capabilities may lead drivers to form a positive initial judgment that the system is highly capable. They might have difficulty updating their expectations later upon encountering new information about the system’s limitations, which seems to conflict with their initial impression and requires them to assimilate negative information.
Research has shown that providing drivers with training on ADAS or partial driving automation can improve drivers’ comprehension, confidence, and use of these technologies (Abraham et al., 2017b; Forster et al., 2019; Manser et al., 2019; McDonald et al., 2017). However, the relative benefits of different training modes and approaches are unclear (Manser et al., 2019; McDonald et al., 2017). Research also suggests that drivers differ in terms of their preferred methods for learning about ADAS, and that drivers who learn in their preferred method report improved comprehension and use of these technologies (Abraham et al., 2018).
The purpose of the current study was to investigate how differences in the branding of a previously unfamiliar partial driving automation feature, as well as the mode of training, influenced drivers’ initial understanding and expectations of the feature. We also investigated whether any effects observed immediately after training persisted after brief exposure to the feature in real-world on-road driving.
Method
Design
Participants were provided brief training about an SAE Level 2 (hereafter L2) partial driving automation feature in one of two randomly assigned branding conditions and one of three training modes. After training, participants drove a vehicle equipped with the feature that they had learned about. Questionnaires were used to measure participants’ understanding of and expectations about the system after training, both before and after driving the vehicle. In the resulting experimental design, branding condition and training mode were between-subjects variables; pre-drive versus post-drive was a within-subjects variable. This research was approved by the Institutional Review Board of Westat. Informed consent was obtained from each participant.
Setting and Participants
Participants (n = 90) were recruited from the Washington, DC metropolitan area using local advertisements. Participants were required to be 20–70 years old, possess a valid driver’s license for at least 3 years, drive at least 3 days per week, be able to read and write in English, have no major moving violations and no more than a small number of minor moving violations in the past 5 years, and report having no previous experience driving a vehicle equipped with an L2 feature. Participants were provided a $100 cash incentive upon study completion.
Materials
Vehicle
The study vehicle was a 2018 Cadillac CT6 with an L2 feature called Super Cruise (Cadillac, 2017). When activated, Super Cruise maintains the vehicle’s speed, lane position, and distance from a lead vehicle without requiring continuous driver input. It can be used only on limited access, divided highways subject to geofencing. Participants were told that the L2 feature was a new system being studied; all Cadillac, General Motors, and Super Cruise names and logos were obscured or removed from the vehicle.
Training Materials
Participants received training on the L2 feature in one of three modes: (1) print quick-start brochure, (2) video, or (3) in-person demonstration. The quick-start brochures were full-color, glossy booklets designed to resemble the original version produced by Cadillac (Cadillac, 2017); however, the information was adapted to remove all information about the make and model of the vehicle as well as all information unrelated to the L2 feature. The video showed a driver operating the vehicle, turning on the L2 feature, and using the L2 feature; inset still images showed close-ups of important buttons and icons. The video included a running narration by a male speaker that paralleled the text of the quick-start brochure. For the in-person demonstration, a member of the project team trained the participant by providing a scripted overview of the L2 feature and demonstrating its use while driving on real roads with the study participant seated in the front passenger seat. The same researcher conducted all in-person demonstrations to ensure consistency across participants; this researcher was not involved in any subsequent data collection. Images of the quick-start brochures are provided in Supplemental Web Appendices 1 and 2; the video and in-person demonstration were adapted from the quick-start brochure.
Two versions of the materials were developed for each training mode: one in which the L2 feature was named AutonoDrive, and another in which the feature was named DriveAssist. AutonoDrive materials emphasized the L2 feature’s capabilities and benefits; DriveAssist materials emphasized the feature’s limitations and the driver’s responsibilities. For example, where the AutonoDrive material states, “Now AutonoDrive is in control and you can relax,” the DriveAssist material states, “Now DriveAssist is helping you to steer and maintain speed. You must continue to monitor the system and the road ahead. Do not engage in distracting tasks such as reading or using your mobile device.” Despite the differences in emphasis, neither version omitted any important safety information nor presented any false information about the capabilities or limitations of the system.
Questionnaires
Participants completed three questionnaires. The initial questionnaire (Supplemental Web Appendix 3), administered before participants viewed any training materials or drove the vehicle, collected information about participant demographics (Section A) as well as the Prosocial and Aggressive Driving Inventory (PADI) (Harris et al., 2014) (Section B), Brief Sensation Seeking Scale (BSSS) (Hoyle et al., 2002) (Section C), and Affinity for Technology Interaction (ATI) scale (Franke et al., 2018) (Section D), as well as several miscellaneous questions about driving behavior not included in the current study. Demographic information as well as these scales were used in analyses as covariates.
The post-training questionnaire (Supplemental Web Appendix 4) was administered immediately after participants received brief training about the L2 feature but before they drove the vehicle. All items used in the current study were 5-point Likert-type items. Items central to the current study included 18 items that assessed whether participants expected the L2 feature to control the vehicle’s speed and lane position in a variety of scenarios (Question 1), 8 items that assessed whether participants expected the L2 feature to automatically take action to avoid a collision without any action by the driver (Question 3), and 11 items that assessed participants’ willingness to engage in various driving behaviors while using the L2 feature versus without the L2 feature (Question 12). Items related to expectations regarding control of speed and lane position and items regarding collision avoidance used response options of definitely would not (1), probably would not (2), I have no idea (3), probably would (4), and definitely would (5). (Note: participants were instructed to make their best guess and use “I have no idea” only as a last resort.) Questions regarding willingness to perform behaviors used response options of much less willing (1), somewhat less willing (2), no difference (3), somewhat more willing (4), and much more willing (5).
The post-drive questionnaire (Supplemental Web Appendix 5) included the same items as the post-training questionnaire to assess changes relative to immediately after training (before driving), as well as several miscellaneous items to solicit general reactions to the L2 feature.
Procedure
Study sessions were conducted from August through October 2019. Upon arrival for their session, participants read and signed an informed consent form and completed the initial questionnaire.
Participants were then trained according to their randomly assigned branding condition and training mode. For the quick-start brochure, participants were handed the brochure and instructed to read it in its entirety. Participants generally took 5 to 10 minutes to read the quick-start brochure, though reading times were not recorded for individual participants. For the video condition, participants were seated at a computer monitor and the researcher played the video. The DriveAssist training video had a duration of 7:36 and the AutonoDrive video had a duration of 6:43. (The longer duration of the DriveAssist video was attributable mainly to presenting the limitations of the system more slowly and with greater emphasis than was done in the AutonoDrive video.) Participants were not permitted to rewind the video or watch it more than once. For the in-person demonstration, a different researcher escorted the participant to the study vehicle and conducted the demonstration. The in-person demonstration included approximately 3 minutes of in-vehicle training with the vehicle parked and approximately 6 minutes of on-road use of the L2 feature by the researcher. After the demonstration was completed, the demonstrator escorted the participant back to the original researcher. After completion of the assigned training, participants completed the post-training questionnaire.
Participants then drove the vehicle on a predetermined route accompanied by a researcher in the front passenger seat. All on-road driving sessions were conducted in dry weather and in daylight conditions, in the summer and early fall, between 10 a.m. and 4 p.m. to avoid sun glare and rush hour traffic. Traffic conditions were free flowing in all sessions and were generally light. For the drive, the researcher gave the participant navigational directions, instructed the participant to activate the L2 feature at a predetermined location, and otherwise gave minimal instruction only as needed to maintain safety. Participants were expected to use the L2 feature for approximately 31 miles of limited access highway driving from the point at which the experimenter instructed them to activate it until they reached the end of the study route. Participants generally remained in L2 mode for the entire duration of the official study route unless the system unexpectedly required the driver to retake control for a moment or if the driver chose to retake manual control briefly (e.g., to change lanes to pass another vehicle or accommodate merging traffic). After reaching the formal end of the study route, participants were given the option to continue using the L2 feature or resume ordinary manual driving for the remaining 1.7 miles until exiting the highway (at which point resuming ordinary manual driving was required). Upon completing the drive, participants completed the final questionnaire and then were paid their incentive and debriefed.
Variables
Dependent Variables
The main dependent variables were 8 composite measures of drivers’ expectations about the performance of the L2 feature, derived through principal component analysis (PCA) of the previously mentioned items from the post-training questionnaire, to reduce the number of dependent variables from 37 individual items to a more concise number of factors. A separate PCA was performed for each battery of items. The number of principal components for each set of items was determined using parallel analysis with 1,000 repetitions (Horn, 1965; Dinno, 2009). Principal components were obtained by computing the mean of all items that loaded onto each principal component with absolute loadings ≥0.2 and no other loadings of similar magnitude after oblique oblimin rotation. Seven items had no absolute loadings ≥0.2 or had similar loadings on multiple components and thus were not included in the final components. Final components are summarized below; individual items and their factor loadings are provided in the Appendix.
• General Function: Expects system to control vehicle speed and lane position in general types of conditions (4 items, e.g., heavy rain).
• Specific Capabilities: Expects system to control vehicle speed and lane position in specific driving scenarios (4 items, e.g., stop-and-go traffic).
• Nonexistent Features: Expects system to perform tasks it cannot perform (3 items, e.g., automatically change lanes).
• Environmental Changes: Expects system to respond to environmental changes (2 items, e.g., changes in speed limit).
• Collision Avoidance Possible: Expects system to avoid a collision in specific scenarios in which collision avoidance is possible given the actual capabilities of the vehicle and system (4 items, e.g., lead vehicle braking hard).
• Collision Avoidance Impossible: Expects system to avoid a collision in specific scenarios in which collision avoidance is not possible based on its designed capability (4 items, e.g., construction worker standing in lane).
• Workload reduction: More willing to take various actions while driving with system engaged versus without system (6 items, e.g., talk on handheld cell phone).
• Impaired Driving: More willing to drive while impaired with system engaged versus without system (3 items, e.g., drive after 3 alcoholic drinks).
Independent Variables
The main independent variables were branding condition (AutonoDrive or DriveAssist) and training mode (quick-start brochure; video; in-person demonstration). Pre-drive versus post-drive (hereafter drive) was an additional independent variable.
Covariates
Covariates were participant age, gender, highest level of education completed, and scores on the PADI, BSSS, and ATI scales.
Statistical Analysis
The associations of each dependent variable with branding condition, training mode, and drive were estimated using a generalized least squares random effects model that included condition, mode, drive, all two- and three-way interactions, and the above-mentioned covariates. Age was modeled using continuous age in years and its square to allow nonlinearity. Education was modeled as categorical with categories of less than bachelor’s degree, bachelor’s degree, and more than bachelor’s degree. The fitted regression models were then used to estimate the covariate-adjusted mean and standard error of each dependent variable, for each branding condition and training mode, averaged over all participants. These were then used to calculate the statistical significance of differences of means between conditions and between modes both pre-drive and post-drive. Statistical significance was assessed at the ⍺=0.05 confidence level using two-tailed tests. No statistical adjustments were made for multiple comparisons; implications are discussed in the Discussion. All analysis was performed using Stata version 17.0 (StataCorp LLC, College Station, Texas).
Results
Participant Characteristics
Table 1 shows the characteristics of study participants in relation to branding condition and training mode. Participant characteristics differed little between groups, with the following minor exceptions. Mean age was 4 years younger among participants assigned to the in-person demonstration versus the other two training modes. Relatively more men were assigned to the DriveAssist condition and to the brochure mode. The proportion with at least a bachelor’s degree was higher among those assigned to the AutonoDrive condition and the video training mode and was lowest among those assigned to the brochure mode. Scores on the aggressive component of the PADI were slightly higher for the participants assigned to the AutonoDrive condition (suggesting greater proclivity for aggressive driving) than the DriveAssist condition. Participants assigned to the brochure had slightly lower scores on the BSSS (suggesting less tendency toward sensation seeking) than participants assigned to other training modes. Participants assigned to the AutonoDrive condition had slightly higher scores on the ATI (suggesting higher affinity for interacting with new technology) than did those assigned to DriveAssist.
TABLE 1:
Participant Characteristics in Relation to Branding Condition and Training Mode
| Branding Condition | Training Mode | All (n = 90) | ||||
|---|---|---|---|---|---|---|
| DriveAssist (n = 45) | AutonoDrive (n = 45) | Brochure (n = 30) | Video (n = 30) | Demo (n = 30) | ||
| Age, years | ||||||
| Mean (SD) | 47.2 (13.7) | 46.8 (15.2) | 48.4 (14.0) | 48.2 (14.5) | 44.3 (14.9) | 47.0 (14.4) |
| Gender | ||||||
| % Male | 48.9 | 42.2 | 50.0 | 43.3 | 43.3 | 45.6 |
| Education | ||||||
| % BA or higher | 68.9 | 77.8 | 60.0 | 86.7 | 73.3 | 72.2 |
| Prosocial and aggressive driving inventory | ||||||
| Prosocial, Mean (SD) | 5.66 (0.39) | 5.61 (0.47) | 5.63 (0.47) | 5.67 (0.36) | 5.60 (0.47) | 5.63 (0.43) |
| Aggressive, Mean (SD) | 1.67 (0.47) | 1.82 (0.55) | 1.71 (0.47) | 1.73 (0.46) | 1.80 (0.62) | 1.75 (0.51) |
| Brief sensation seeking scale | ||||||
| Mean (SD) | 2.53 (0.70) | 2.47 (0.76) | 2.33 (0.78) | 2.58 (0.76) | 2.59 (0.63) | 2.50 (0.73) |
| Affinity for technology interaction | ||||||
| Mean (SD) | 4.33 (0.92) | 4.44 (0.69) | 4.36 (0.81) | 4.38 (0.78) | 4.42 (0.86) | 4.39 (0.81) |
Initial Effects of Training
Branding Condition
Figure 1 shows mean scores on all factors in relation to branding condition both before and after participants drove the vehicle. Immediately after training (pre-drive), participants assigned to the AutonoDrive branding condition reported significantly higher expectations of Specific Capabilities (p = 0.03), Environmental Changes (p = 0.01), Crash Avoidance Impossible (p = 0.003), Crash Avoidance Possible (p < 0.001), and Workload Reduction (p = 0.01) compared with participants assigned to the DriveAssist condition. Although scores for General Function (p = 0.47), Nonexistent Features (p = 0.17), and Impaired Driving (p = 0.52) did not differ significantly by condition, mean scores on these factors were also nominally higher among participants assigned to the AutonoDrive condition than among those assigned to DriveAssist.
Figure 1.
Pre-drive and post-drive expectations of L2 system in relation to branding condition for each composite factor, adjusted for participant characteristics. Error bars show 95% confidence intervals.
Training Mode
Figure 2 shows mean scores on all factors in relation to training mode both before and after participants drove the vehicle. Participants assigned to the video training mode had lower expectations of Collision Avoidance Possible than did participants assigned to the other two modes (p = 0.02). No other factors measured pre-drive differed significantly in relation to training mode, nor were there any consistent, interpretable patterns with respect to directional differences in relation to mode.
Figure 2.
Pre-drive and post-drive expectations of L2 system in relation to training mode for each composite factor, adjusted for participant characteristics. Error bars show 95% confidence intervals.
Interaction of Branding Condition and Training Mode
The interaction of branding condition with training mode was not statistically significant for any of the factors examined. Although not statistically significant at the ⍺ = 0.05 level, the magnitude of the difference in mean scores of Crash Avoidance Possible between AutonoDrive versus DriveAssist groups was substantially larger among participants trained by video or in-person demonstration than among participants trained by brochure (p = 0.07).
Post-Drive Effects of Training
Branding Condition
After driving the vehicle, scores for Specific Capabilities (p = 0.04), Crash Avoidance Impossible (p = 0.03), and Workload Reduction (p = 0.01) remained significantly higher among participants in the AutonoDrive condition than among those in the DriveAssist condition (Figure 1). In addition, scores for General Function became marginally higher among participants in the AutonoDrive condition than among those in the DriveAssist condition (p = 0.048), whereas their respective pre-drive scores on this factor did not differ significantly. Pre-drive differences in expectations for Environmental Changes and Crash Avoidance Possible narrowed substantially post-drive and no longer differed significantly by condition.
Notably, although not all of the following were statistically significant, mean scores on every factor were higher among participants in the AutonoDrive condition than in the Drive-Assist condition. Additionally, mean scores were higher post-drive than pre-drive for all factors for participants in both conditions.
Training Mode
None of the factors differed significantly in relation to training mode after participants drove the vehicle. Expectations regarding Collision Avoidance Possible increased significantly more post-drive among participants trained by video than among those trained by brochure or in-person demonstration (p < 0.001), erasing a large difference present pre-drive. The post-drive change in expectations for Workload Reduction also differed significantly by mode (p = 0.02), decreasing among those trained by in-person demonstration while increasing slightly among those trained by brochure and video (though neither pre-drive nor post-drive differences in scores varied significantly by mode).
Interaction of Branding Condition and Training Mode
The interaction of branding condition with training mode was not statistically significant for any of the factors examined after participants drove the vehicle. Although not statistically significant at the ⍺ = 0.05 level, the magnitude of the difference in mean scores of Crash Avoidance Impossible between AutonoDrive versus DriveAssist groups was substantially larger among participants trained by in-person demonstration than among participants trained by brochure or video (P = 0.06).
Discussion
Previous research had shown that the names of driver assistance systems could influence drivers’ beliefs and expectations about the capabilities of the system and the responsibilities of the driver, and that training (in general) can help drivers to form stronger initial mental models of vehicle technologies. The current study finds that differences in the overall branding and emphasis of training information can significantly influence drivers’ initial understanding and expectations of a previously unfamiliar partial driving automation system, and that differences persist even after brief experience interacting with the system.
Large, consistent associations were observed between whether participants received training emphasizing the system’s capabilities or its limitations and their corresponding expectations about its performance. Participants who received training emphasizing system capabilities were more likely to correctly expect the system to perform actions that it was capable of performing (e.g., control vehicle speed in stop-and-go traffic; bring the vehicle to a stop if the driver loses consciousness). However, consistent with anchoring (Tversky & Kahneman, 1974) or primacy effects (Jones et al., 1968), they were also more likely to expect the system to possess many capabilities that it did not (e.g., automatic lane changing; adaptation to changes in speed limit; ability to avoid a collision with a car moving into its lane from directly to the side). Although the training given to participants contained no factually incorrect information, there were some notable differences between branding conditions in the extent to which certain limitations were made explicit. For example, while all participants received the same bulleted list of specific actions that the system does not perform, the DriveAssist materials also stated, “DriveAssist is not a collision avoidance system. It will not steer to avoid objects within your lane or engage emergency braking,” whereas the AutonoDrive material omitted this statement. While results were broadly consistent with anchoring and the primacy effect, it is also possible that participants were more able to process this simple, high-level summary statement of system limitations given to the Drive-Assist group than the detailed list of specific limitations given to both groups, or alternatively, that repetition of these points was critical to retention.
Many differences observed in relation to branding condition, albeit not all of them, remained even after participants used the system that they had just learned about while driving an actual vehicle in traffic. In cases where group differences shrank or vanished post-drive, the changes were due mainly to increased expectations among participants who received the training emphasizing limitations, not decreased expectations among those who received training emphasizing capabilities. For example, while the between-group differences in expectations for crash avoidance in scenarios where the vehicle would not actually avoid a crash (“Crash Avoidance Impossible”) shrank somewhat after participants drove the vehicle, the average expectations of both groups increased significantly post-drive. This was unexpected, particularly for participants who received training emphasizing limitations. Beggiato & Krems (2013) found that understanding of adaptive cruise control became more accurate over time with repeated exposure, at least with respect to limitations encountered during use. Notably however, participants in the current study received only a single brief exposure to on-road use of the system and did not encounter any scenarios that would have demonstrated the system’s inability to avoid crashing in the scenarios presented in the questionnaire. Nonetheless, participants’ increased confidence post-drive in the system’s capabilities, even in safety-critical scenarios they did not experience and in which the system would not perform, is concerning. This illustrates the importance of using training to set realistic baseline expectations for drivers learning about a new system. Furthermore, Beggiato et al. (2015) noted that limitations not experienced are forgotten, and thus recommended periodic reminders to aid retention. This might be particularly critical in the case of safety-critical limitations experienced infrequently, such as the system’s inability to avoid a pedestrian standing in the travel lane on a high-speed highway.
While training mode may be an important consideration regarding formation of initial mental models, the current study is unable to provide clear guidance. While some differences in drivers’ expectations were observed in relation to training mode, they were not consistent nor readily interpretable. It is possible that the few differences observed in relation to training mode may have been specific to the implementation in the current study and not indicative of effect of training mode generally. In real-world applications, the effect of training mode may be confounded by non-mode-specific factors. While many drivers may wish to consult the vehicle owner’s manual or quick-start brochure to learn how to use a system, many likely will not (McDonald et al., 2017). While an automobile dealer might be able to ensure that a vehicle purchaser is given an in-person demonstration for a technology, drivers other than the original purchaser are unlikely to receive such a demonstration. Additionally, the quality of information provided in-person at automobile dealerships has been shown to vary considerably (Abraham et al., 2017a).
Several limitations should be noted. Our manipulation of branding condition confounded the system name with its description, thus we cannot determine the relative contributions of the system name versus its description to the overall pattern of results observed. Participants received brief exposure to information about an unfamiliar L2 system under contrived circumstances, for example, those in the video condition were not allowed to rewind it or watch it multiple times. Outside of experimental conditions, some participants might have engaged with the training materials less than they did in the study (or not at all), whereas others might have engaged with them more. This could bias results if such differences varied between branding conditions or training modes. Relatedly, training mode was assigned at random; results might have differed had participants been allowed to choose their preferred mode (c.f. Abraham et al., 2018). Recruitment advertisements might have also influenced participants. Anecdotally, some participants reported that they expected a higher degree of automation than was presented to them. This, however, would not be expected to have a differential impact on participants randomly assigned to one training condition versus another. Although we sought to ensure that all participants experienced similar traffic and environmental conditions during the on-road drive, idiosyncratic events (e.g., interactions with other traffic; unexpected takeover requests) might have influenced post-drive questionnaire responses. However, we would expect such factors to vary randomly between participants, not systematically between groups. A large number of statistical comparisons were made with no adjustment for multiple comparisons; some of the individual statistically significant differences observed may have been due to chance and thus should be treated with appropriate caution. The overall pattern of results, however, would not be materially affected by any plausible number of chance associations. Finally, the questionnaire probed items of substantive interest to the authors (e.g., willingness to drive with versus without the system under a variety of conditions; expectations about when or if the L2 feature would take action to avoid a crash); however, the questionnaire is not represented as a validated measure of mental models per se.
In sum, results suggest that emphasizing the capabilities and benefits of a partial driving automation feature can lead drivers—at least those previously unfamiliar with it—to overestimate its capabilities and overlook or forget important limitations. These inflated expectations are not necessarily tempered by brief experience using the system. These results highlight the importance of ensuring that information provided to consumers is not only technically accurate but that it leaves the consumer with a balanced overall impression, with appropriate emphasis placed on key limitations and the need for driver attention and engagement.
Supplemental Material
Supplemental Material for Driver Expectations of a Partial Driving Automation System in Relation to Branding and Training by Jeremiah Singer, Brian C Tefft, Aaron Benson, James W Jenness, William J Horrey in Human Factors : The Journal of the Human Factors and Ergonomics Society
Supplemental Material for Driver Expectations of a Partial Driving Automation System in Relation to Branding and Training by Jeremiah Singer, Brian C Tefft, Aaron Benson, James W Jenness, William J Horrey in Human Factors : The Journal of the Human Factors and Ergonomics Society
Acknowledgments
This work was supported by the AAA Foundation for Traffic Safety (AAAFTS 4035-51173).
Biography
Jeremiah Singer National Highway Traffic Safety Administration (BA, Research and Experimental Psychology, 2000, Binghamton University). Mr Singer contributed to the work reported in this article while employed by Westat.
Brian C. Tefft AAA Foundation for Traffic Safety (BS, Mechanical Engineering, 2003, Brown University)
Aaron Benson Toxcel LLC (MA, Human Factors, 2017, George Mason University). Mr Benson contributed to the work reported in this article while employed by the AAA Foundation for Traffic Safety.
James W. Jenness Westat (PhD, Biopsychology, 1992, University of Michigan)
William J. Horrey, AAA Foundation for Traffic Safety (PhD, Engineering Psychology, 2005, University of Illinois Urbana-Champaign)
APPENDIX Items Composing Dependent Variables.
TABLE A1:
Items Composing Principal Components Measuring Expectations for Function of L2 System, and Their Corresponding Loadings
| General Function | Specific Capabilities | Nonexistent Features | Environmental Changes | |
|---|---|---|---|---|
| heavy rain | 0.50 | |||
| heavy snow | 0.49 | |||
| darkness | 0.40 | |||
| undivided road | 0.23 | |||
| stop and go traffic due to traffic jam | 0.47 | |||
| driving down long hill | 0.41 | |||
| stop vehicle if driver loses consciousness | 0.40 | |||
| driving through mile-long tunnel | 0.38 | |||
| change lanes to pass slow vehicle | 0.57 | |||
| highway exit ramp | 0.57 | |||
| merge where lane ends | 0.31 | |||
| following truck with 15-foot pole sticking out back | 0.55 | |||
| reduce speed when speed limit drops | 0.54 |
Question: For each situation [above], indicate whether or not you expect [DriveAssist/AutonoDrive] to successfully control vehicle speed and keep the vehicle in its lane without the driver doing anything.” Absolute loadings <0.2 not shown.
TABLE A2:
Items Composing Principal Components Measuring Expectations for Crash Avoidance of L2 System, and Their Corresponding Loadings
| Crash Avoidance Impossible | Crash Avoidance Possible | |
| deer walking toward lane from side of road | 0.55 | |
| worker standing in road | 0.49 | |
| mattress in road | 0.48 | |
| car immediately to side moving into lane | 0.43 | |
| stopped vehicles ahead | 0.62 | |
| vehicle ahead braking hard | 0.57 | |
| approaching motorcycle in lane | 0.42 | |
| slower car cutting in with very small gap | 0.35 |
Question: “For each situation [above], indicate whether or not you expect [DriveAssist/AutonoDrive] to take action (brake and/or steer) and avoid a collision, without the driver doing anything.”
Absolute loadings <0.2 not shown.
TABLE A3:
Items Composing Principal Components Measuring Willingness to Take Advantage of System Benefits, and Their Corresponding Loadings
| Workload Reduction | Impaired Driving | |
| 6-hour drive to other state | 0.42 | -0.31 |
| eat while driving | 0.40 | |
| drive with back or shoulder pain | 0.39 | |
| hands-free cellphone conversation | 0.38 | |
| handheld cellphone conversation | 0.35 | |
| drive with severe headache | 0.34 | |
| drive after medications that warn not to drive | 0.50 | |
| drive after 3 alcoholic drinks | 0.52 | |
| drive drowsy | 0.48 |
Question: “How willing would you be to drive in the following situations while using [DriveAssist/AutonoDrive], compared to driving a vehicle without [DriveAssist/AutonoDrive]?”
Absolute loadings <0.2 not shown.
Footnotes
Supplemental Material: Supplemental material for this article is available online.
Key Points
Training that emphasizes the capabilities of a partially automated driving system can lead drivers to overestimate its capabilities and overlook or forget its limitations, even if the limitations are disclosed.
Drivers who received training that emphasized the limitations of a partially automated driving system and the responsibilities of the driver exhibited fewer incorrect expectations.
Differences between expectations of drivers who received training emphasizing system capabilities versus system limitations often persisted even after brief experience using the system.
Technical accuracy of consumer information and training is necessary but not sufficient. Designers of information and training materials should seek to provide an accurate overall impression of a system’s limitations as well as its capabilities.
References
- AAA (2019). Advanced Driver Assistance Technology Names: AAA’s Recommendation for Common Naming of Advanced Safety Systems. [Google Scholar]
- Abraham H., McAnulty H., Mehler B., Reimer B. (2017. a). Case study of today’s automotive dealerships: Introduction and delivery of advanced driver assistance systems. Transportation Research Record: Journal of the Transportation Research Board, 2660(1), 7–14. [Google Scholar]
- Abraham H., Reimer B., Mehler B. (2018). Learning to use in-vehicle technologies: consumer preferences and effects on understanding. In Proceedings of the Human Factors and Ergonomics Society Annual Meeting (pp. 1589–1593). Los Angeles, CA: Sage Publications. [Google Scholar]
- Abraham H., Seppelt B., Reimer B., Mehler B. (2017. b). What’s in a name: Vehicle technology branding & consumer expectations for automation. In Proceedings of the 9th International Conference on Automotive User Interfaces and Interactive Vehicle Applications. Oldenburg, Germany: AutomotiveUI ‘17. [Google Scholar]
- Beggiato M., Krems J. F. (2013). The evolution of mental model, trust and acceptance of adaptive cruise control in relation to initial information. Transportation Research Part F: Traffic Psychology and Behaviour, 18, 47–57. [Google Scholar]
- Beggiato M., Pereira M., Petzoldt T., Krems J. (2015). Learning and development of trust, acceptance and the mental model of ACC. A longitudinal on-road study. Transportation Research Part F: Traffic Psychology and Behaviour, 35, 75–84. [Google Scholar]
- Braitman K.A., McCartt A.T., Zuby D.S., Singer J. (2010). Volvo and Infiniti drivers’ experiences with select crash avoidance technologies. Traffic Injury Prevention, 11(3), 270–278. [DOI] [PubMed] [Google Scholar]
- Bruner J. S., Potter M. C. (1964). Inference in visual recognition. Science, 144(3617), 424–425. [DOI] [PubMed] [Google Scholar]
- Cadillac (2017). CT6 Super Cruise™ Convenience & Personalization Guide. [Google Scholar]
- Dinno A. (2009). Implementing Horn’s parallel analysis for principal component analysis and factor analysis. The Stata Journal, 9(2), 291–298. [Google Scholar]
- Dixon L. (2020). Autonowashing: The Greenwashing of vehicle automation. Transportation Research Interdisciplinary Perspectives, 5, 100113. 10.1016/j.trip.2020.100113. [DOI] [Google Scholar]
- Einhorn H. J., Hogarth R. M. (1981). Behavioral decision theory: Processes of judgement and choice. Annual Review of Psychology, 32(1), 53–88. [Google Scholar]
- Forster Y., Hergeth S., Naujoks F., Krems J., Keinath A. (2019). User education in automated driving: Owner’s manual and interactive tutorial support mental model formation and human-automation interaction. Information, 10(4), 143. 10.3390/info10040143. [DOI] [Google Scholar]
- Franke T., Attig C., Wessel D. (2018). A Personal Resource for Technology Interaction: Development and Validation of the Affinity for Technology Interaction (ATI) Scale. International Journal of Human–Computer Interaction. 10.1080/10447318.2018.1456150 [DOI] [Google Scholar]
- Gaspar J., Carney C., Shull E., Horrey W. J. (2020). The Impact of Driver’s Mental Models of Advanced Vehicle Technologies on Safety and Performance (Technical Report). Washington, DC: AAA Foundation for Traffic Safety. [Google Scholar]
- Harris P. B., Houston J. M., Vazquez J. A., Smither J. A., Harms A., Dahlke J. A., Schau D. A. (2014). The Prosocial and Aggressive Driving Inventory (PADI): A self-report measure of safe and unsafe driving behaviors. Accident Analysis & Prevention, 72(1), 1–8. 10.1016/j.aap.2014.05.023. [DOI] [PubMed] [Google Scholar]
- Horn J. L. (1965). A rationale and test for the number of factors in factor analysis. Psychometrika, 30(2), 179–185. [DOI] [PubMed] [Google Scholar]
- Hoyle R. H., Stephenson M. T., Palmgreen P., Lorch E. P., Donohew R. L. (2002). Reliability and validity of a brief measure of sensation seeking. Personality and Individual Differences, 32(3), 401–414. [Google Scholar]
- Jenness J.W., Lerner N.D., Mazor S., Osberg J.S., Tefft B.C. (2008). Use of advanced in-vehicle technology by younger and older early adopters. Selected results from five technology surveys (Report no. DOT HS 811004). Washington, DC: National Highway Traffic Safety Administration. [Google Scholar]
- Jones E. E., Rock L., Shaver K. G., Goethals G. R., Ward L. M. (1968). Pattern of performance and ability attribution: An unexpected primacy effect. Journal of Personality and Social Psychology, 10(4), 317–340. [Google Scholar]
- Lubkowski S. D., Lewis B. A., Gawron V. J., Gaydos T. L., Campbell K. C., Kirkpatrick S. A., Cicchino J. B. (2021). Driver trust in and training for advanced driver assistance systems in real-world driving. Transportation Research Part F: Traffic Psychology and Behaviour, 81, 540–556. [Google Scholar]
- Manser M. P., Noble A. M., Machiani S. G., Shortz A., Klauer S. G., Higgins L., Ahmadi A. (2019). Driver Training Research and Guidelines for Automated Vehicle Technology (No. 01-004). U.S. Department of Transportation. [Google Scholar]
- McDonald A., Reyes M., Roe C., McGehee D. (2017). Driver understanding of ADAS and evolving consumer education 25th International Technical Conference on the Enhanced Safety of Vehicles [Conference Proceedings]. Detroit, MI. [Google Scholar]
- Nickerson R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology, 2(2), 175–220. [Google Scholar]
- Seppelt B., Victor T. (2020). Driver’s Mental Model of Vehicle Automation. In Fisher D. L., Horrey W. J., Lee J. D., Regan M. A. (Eds), Handbook of Human Factors for Automated, Connected and Intelligent Vehicles. Boca Raton, FL: CRC Press. [Google Scholar]
- Teoh E. R. (2020). What’s in a name? Drivers’ perceptions of the use of five SAE Level 2 driving automation systems. Journal of Safety Research, 72, 145–151. [DOI] [PubMed] [Google Scholar]
- Tversky A., Kahneman D. (1974). Judgment under Uncertainty: Heuristics and Biases: Biases in judgments reveal some heuristics of thinking under uncertainty. Science, 185(4157), 1124–1131. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplemental Material for Driver Expectations of a Partial Driving Automation System in Relation to Branding and Training by Jeremiah Singer, Brian C Tefft, Aaron Benson, James W Jenness, William J Horrey in Human Factors : The Journal of the Human Factors and Ergonomics Society
Supplemental Material for Driver Expectations of a Partial Driving Automation System in Relation to Branding and Training by Jeremiah Singer, Brian C Tefft, Aaron Benson, James W Jenness, William J Horrey in Human Factors : The Journal of the Human Factors and Ergonomics Society