Drivers' Adaptation to Adaptive Cruise Control - University of Washington

1468

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 13, NO. 3, SEPTEMBER 2012

[12] A. Amditis, E. Bertolazzi, M. Bimpas, F. Biral, P. Bosetti, M. Da Lio, L. Danielsson, A. Gallione, H. Lind, A. Saroldi, and A. Sjögren, “A holistic approach to the integration of safety applications: The insafes subproject within the european framework programme 6 integrating project prevent,” IEEE Trans. Intell. Transp. Syst., vol. 11, no. 3, pp. 554–566, Sep. 2010. [13] P. Needham, “Collision prevention: The role of an accident data recorder (ADR),” in Proc. Int. ADAS Conf., 2001, pp. 48–51, IEE Conf. Publ. No. 483. [14] J. Dean, “Extremely mobile devices,” PopSci, Sep. 2011. [Online]. Available: http://www.popsci.com/cars/article/2011-08/extremely-mobiledevices [15] J. Eriksson, L. Girod, B. Hull, R. Newton, S. Madden, and H. Balakrishnan, “The pothole patrol: Using a mobile sensor network for road surface monitoring,” in Proc. 6th Int. MobiSys Conf., New York, 2008, pp. 29–39. [16] J. Wang, J. Cho, S. Lee, and T. Ma, “Real time services for future cloud computing enabled vehicle networks,” in Proc. Int. WCSP Conf., Nov. 2011, pp. 1–5. [17] Y. Zhang, W. Lin, and Y.-K. Chin, “A pattern-recognition approach for driving skill characterization,” IEEE Trans. Intell. Transp. Syst., vol. 11, no. 4, pp. 905–916, Dec. 2010. [18] BOSCH, Bmi150—Digital, Triaxial Acceleration Sensor-Data Sheet, Oct. 2011. [Online]. Available: http://catalog.gaw.ru/project/download. php?id=20526 [19] S. Reddy, M. Mun, J. Burke, D. Estrin, M. Hansen, and M. Srivastava, “Using mobile phones to determine transportation modes,” ACM Trans. Sensor Netw., vol. 6, no. 2, pp. 13-1–13-27, Feb. 2010. [20] J. Kwapisz, G. Weiss, and S. Moore, “Cell phone-based biometric identification,” in Proc. 4th IEEE Int. Conf. BTAS, Sep. 2010, pp. 1–7.

Drivers’ Adaptation to Adaptive Cruise Control: Examination of Automatic and Manual Braking Huimin Xiong and Linda Ng Boyle

Abstract—Drivers may adapt to the automatic braking control feature available on adaptive cruise control (ACC) in ways unintended by designers. This study examines drivers’ adaptation using a conceptual model of adaptive behavior developed and examined quantitatively using logistic regression techniques. Data for this model come from a field operational test on the use of an advanced collision avoidance system, which integrated forward collision warning and ACC functions. A sample of “closing” events was extracted from a subset of these ACC data. The logistic regression model predicted the drivers’ likelihood to intervene (i.e., manually brake) whenever ACC began braking or slowing down the vehicle. The results indicate that several factors influence drivers’ response, including the environment, selected gap setting, speed, and drivers’ age. Safety consequences and the design of future ACC systems based on drivers’ adaptation to these factors are discussed. Index Terms—Adaptive cruise control (ACC), driver behavior.

Manuscript received June 9, 2011; revised November 21, 2011; accepted February 20, 2012. Date of publication April 27, 2012; date of current version August 28, 2012. This work was supported by the U.S. National Science Foundation under Grant IIS 1027609. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the U.S. National Science Foundation. The Associate Editor for this paper was H. Jula. H. Xiong is with the Department of Industrial and Systems Engineering, College of Engineering, University of Washington, Seattle, WA 98195 USA (e-mail: [email protected]). L. N. Boyle is with the Department of Industrial and Systems Engineering and the Department of Civil and Environmental Engineering, College of Engineering, University of Washington, Seattle, WA 98195 USA (e-mail: linda@ uw.edu). Digital Object Identifier 10.1109/TITS.2012.2192730

I. I NTRODUCTION A. Background Adaptive cruise control (ACC) is an enhanced version of conventional cruise control, which can detect a lead vehicle and allows a driver to follow it at a preset time headway and speed by controlling the engine and braking capability [1], [2]. It automates two components of the driving task, namely, operational control of longitudinal time headway and speed, such that ACC can regulate the speed of a vehicle when a slower lead vehicle is present [3], [4]. Manufacturers typically market ACC systems as convenience systems rather than safety systems. However, ACC has been also incorporated into collision avoidance and brake assist systems as part of integrated advanced safety systems. Some of the benefits of ACC described in the literature have included better distance keeping [5], with thresholds typically ranging from 0.9 to 2.5 s [6], and reduced necessity to monitor the external surroundings or manually accelerate or brake [7]. The reduction in required mental and physical resources makes driving less effortful and reduces driver stress and human errors [8], [9]. However, if the reduction gets too high, it may actually affect the ability of the driver to maintain awareness of situations. ACC systems do have maximum deceleration rates and do not regulate speeds based on stationary vehicles or objects [10]. Furthermore, spacing and velocity errors can occur as shown by Gao and Xiao [11]. ACC cannot properly function when a lead vehicle enters a curved road [12]. That is, the ACC radar will not detect a lead vehicle with high road curvature even when in close proximity to the lead vehicle. Hence, the ACC vehicle will accelerate back to the original cruise speed. When these limitations are coupled with a low level of situation awareness, potential hazards may increase rather than decrease. This was also observed by Rudin-Brown and Parker [12], who found that drivers’ response time to a hazard detection task increased with ACC. Many studies on ACC have typically focused on the potential innovations of the system such as full start–stop control [10], [13] or the overall effect it may have on traffic systems [6], [14]. Both the human driver and the ACC system control the driving task. Hence, driver intervention can always override the ACC control by activating the brake or the accelerator or switching ACC on or off [15], [16]. The gap with the lead vehicle may also decrease whenever the driver deactivates ACC by pressing on the brake, as shown by Pauwelussen and Feenstra [17]. Therefore, additional studies on the safety implications and human interaction with ACC are needed. Because the likelihood of a collision is rare within a study period, a potential surrogate measure (or proxy) for safety, with respect to ACC, is a “closing” event. A “closing” event is defined as the moment that the ACC’s automatic braking control is activated until any braking (or deceleration) ceases, regardless of whether a driver intervenes or not. The braking (or deceleration) could be based on either automatic braking from ACC (no driver’s intervention) or the driver depressing the brake (driver intervenes). This study focuses on drivers’ adaptive responses to ACC during these “closing” events. Studies have shown that adaptation can greatly influence a driver’s response to safety-based systems, and this adaptation may result in unintended safety consequences. For example, Evans and Gerrish [18] showed that antilock braking systems significantly reduced front impacts but increased the number of rear impacts. Similarly, ACC may not be appropriate for dense driving conditions where driver assistance is most needed. Additionally, drivers who become dependent on the feedback provided by ACC may actually become complacent and less aware of their surroundings, resulting in a negative impact on safety.

1524-9050/$31.00 © 2012 IEEE


1469

the influence of initiating and mediating factors on drivers’ response during first exposure to ACC. II. M ETHOD A. Data Set

Fig. 1.

Factors that influence operator’s adaptive behavior.

B. Conceptual Model of Adaptive Behavior Adaptive behavior is defined, in the context of this paper, as the behavior exhibited when one adapts in unintended ways to systems that are designed to improve safe operation. Drivers’ initial use of the system may not necessarily relate to their long-term use, and they may ultimately change their behaviors in ways that the initial designer had not anticipated. This compensation to technology can positively and negatively influence safety. For ACC systems, potentially positive safety consequences include fewer lane changes, more consistent speeds, and longer headway times [19]–[21]. Negative consequences can include decreased awareness of surroundings, increased driver distractions, decreased ability to respond to safety-critical situations [8], [12], [22], a higher likelihood of rear-end collisions [23], and induced traffic congestion [24]. The goal of this study is to quantify the factors that influence drivers’ initial adaption to ACC, which can impact safety, and to gain insights for the design of future in-vehicle systems. Based on previous studies [25]–[28], a model of an operator’s adaptive behavior was developed that includes the relationship between initiating and mediating factors and the influences of each toward the operator’s behavior (see Fig. 1). Initiating factors are based on the driver’s direct interaction with the system, the environmental cues, and the actual risk at the moment of ACC use. For example, an actual risk can relate to the following distance of the ACC vehicle, which becomes shorter as a lead vehicle brakes. Coupling the response of a safety-based system (ACC) with environmental conditions (i.e., roadway, weather, and traffic) provide feedback that drivers can use to respond to the next safetycritical situation. These factors have immediate adapting effects on the driver. Mediating factors are more subjective but may actually have a greater influence on the drivers’ long-term behavior. These factors emerge from exposure to a system in conjunction with perceived risks (e.g., gap and speed settings), motivational factors (e.g., willingness to use ACC), attitudes/biases (e.g., driving styles, trust in ACC, and overall system use), experiences (e.g., exposure to ACC), and user limitations (e.g., physical, cognitive, and medical). The interaction of these factors will influence the final response of the operator in ways that may not have been incorporated into the system design. Therefore, the first step in assessing driver adaptation is to understand

A field operation test (FOT) was conducted from 2003 to 2004 on an advanced collision avoidance system by the University of Michigan Transportation Research Institute (UMTRI) [21]. The system was composed of two components, namely, ACC and forward collision warning. This FOT provided a useful data set for this study because it was one of the first studies conducted on early users of an ACC system and included naturalistic driving. The original study included 96 drivers and ten vehicles. Of this original study sample, 66 drivers used ACC and were included in the sample of this analysis. There were three age groups with an equal number of males and females: younger aged (20–30 years old), middle-aged (40–50 years old), and older aged (60–70 years old) drivers. The details of the study can be found in [29]. However, some of the interesting results that are relevant to this study are presented here. The results indicated that ACC was widely accepted by drivers. Possible benefits observed during ACC driving included a significant reduction in short (less than 1 s) time headways and reduced overtaking behavior. In the United States, there are approximately 70 vehicle models with ACC as a standard or optional feature, as of 2010. Therefore, with the growing prevalence of ACC, understanding the effect of adaptive behavior on the safe usage of this system is critical. Drivers in this study were able to select a gap setting (time headway) from six intervals. The six intervals ranged from 1 to 2 s with 0.2-s increments (e.g., gap setting 1 would correspond to a 1-s headway, and setting 2 corresponds to a 1.2-s headway). The ACC system was capable of providing warnings for braking events with a maximum value of 0.3 g. There were a total of 3924 “closing” events extracted from the 64 drivers (two drivers were excluded as they had no “closing” events) with an average of 61 events per driver (SD = 68.83). For each driving event, information was recorded on drivers’ response (manual or automatic braking), speed, gap setting, range, weather condition, road type, lighting condition, and traffic condition. Traffic conditions were predefined by UMTRI as heavy, moderate, or light congestion based on a combination of vehicle speed for the road type and traffic density; a specific definition for each is available in [29]. This database also included drivers’ demographic information, and this was also used in the analysis. B. Data Analysis Based on the model of adaptive behavior (see Fig. 1), the initiating factors, or those factors that directly impact the immediate decision, are defined in this study as the distance to the lead vehicle (range). This factor is also used as a measure of the actual risk. Other factors such as weather condition, road type (highway or nonhighway), lighting condition (day or night), and traffic condition (heavy, moderate, or light congestion) are used to describe the environment. The combination of the environment, actual risk, and ACC response then provides feedback to the driver, which influences the drivers’ response for the next safety-critical situation. These initiating factors are then considered along with (or adjusted for) the mediating factors that measure perceived risk (the selected gap setting and vehicle speed), experience defined as the number of sequential “closing” events (which range from 1 to 408 events), and user factors (age and gender) (see Table I). When a “closing” event occurs, there are several decisions that can be chosen by the driver: to allow ACC to automatically brake, to

1470


TABLE I P REDICTOR VARIABLES TO I DENTIFY D RIVER A DAPTATION

TABLE II R ESPONSES D URING C LOSING E VENTS BY G AP S ETTINGS

manually brake for the vehicle in front, to move to another lane, or even to accelerate. The drivers’ response related to the “closing” event is used as an indicator of how drivers behave and initially adapt to the ACC system. In this data set, there were no situations in which the drivers accelerated (whether inadvertently or intentionally is not known) or move to another lane. The drivers responded by either manually braking or allowing ACC to automatically brake. These responses can be considered as a binary outcome with the probability of occurrence predicted using a logistic regression model. There are times when the ACC system begins to brake automatically but is immediately followed by the driver also braking manually—these events were coded as a manual braking event since the driver did intervene. Each driver’s responses were recorded at multiple points in time and were therefore correlated. The most widely used approach for a repeated ordinal response is the generalized estimating equation (GEE) method, which has been applied in many domains [30], [31]. The GEE method was used to build the logistic regression model in R statistical software, using the “geepack” package [32], [33]. The parameter estimates for the logistic regression model (or regression coefficients) show the magnitude of each variable on the likelihood of the response: A positive number indicates that the driver is more likely to manually intervene, and a negative number indicates that the driver is less likely to intervene. Each parameter estimate is evaluated using Wald statistics with significance assessed at p < 0.05. The model is based on a logarithmic distribution, and as such, the estimates are exponentially transformed prior to interpretation.

TABLE III N UMBER OF C LOSING E VENTS BY ROAD T YPE AND T RAFFIC C ONDITIONS

III. R ESULTS A. Predict Likelihood of Drivers’ Interventions (Manually Brake) The number of “closing” events was greatest for the older drivers (1603 events), followed by younger (1327 events) and middle-aged drivers (994 events). There were more events for males (n = 2410, 61.4%) than females (n = 1514, 38.6%). Of the six gap settings, there were not as many observations for gap setting 2 (1.2-s headway), gap setting 4 (1.6-s headway), and gap setting 5 (1.8-s headway) (see Table II). Hence, the data were recoded into three groups of gap settings and used as the explanatory variable in the forthcoming logistic regression model: short (gap settings 1 and 2), medium (gap settings 3 and 4), and long (gap settings 5 and 6). There were 1574, 1296, and 1054 observations for gap settings short, medium, and long, respectively. The results showed a significant association between gap settings and drivers’ responses. There were more occurrences of drivers’ intervention in the longer gap setting (χ2 (2) = 19.6, p < 0.001; see Table II).

Fig. 2.

Frequency of closing events by gap setting and road type.

Other explanatory variables included in the regression model were road type, speed, and age group. Variables that were considered but removed due to insignificant effects included range, exposure to “closing” events with ACC, weather condition, lighting condition, traffic condition, and gender. The relationship among traffic congestion level, road type, and drivers’ response was further explored with respect to number of “closing” events. There were 73 events (out of 3871 observations) or 1.89% of ACC “closing” events triggered in heavy congestion, 1682 events (43.4%) in light congestion, and 2116 events (54.7%) in moderate congestion. There were 53 events recorded as unknown traffic condition. Controlling for the traffic condition level, a significantly higher frequency of drivers’ intervening in nonhighway situations was observed when compared with highways (χ2 (4) = 227.1, p < 0.001). Controlling for road type, more drivers intervened in heavy congestion when compared with the other congestion levels (χ2 (6) = 27.0, p < 0.001; see Table III). As observed in Fig. 2, the road type had an impact on gap selection. Drivers were more likely to use the longer gap setting in nonhighway situations and the shorter gap setting on highways. Speed was another factor that highly influenced gap settings. The mean speeds for the short, medium, and long gap settings were 30.5, 27.9, and 25 m/s, respectively.


1471

TABLE IV L IKELIHOOD OF M ANUAL B RAKING D URING C LOSING E VENTS

Fig. 3. Examples of closing events (represented by the dotted line) for (left) a conflict and (right) a near crash (based on [25]). TABLE V D RIVER R ESPONSE BASED ON C RITICALITY OF C LOSING E VENTS

The outcomes of the logistic regression model (see Table IV) show the regression coefficient for highway roads as −0.94. The negative estimate indicates that drivers are less likely to intervene on highways. More specifically, they are 2.6 times (1/e∧ (−0.94) = 2.6) less likely to intervene on the highway when compared with the nonhighway condition. The higher the vehicle speed, the less likely the driver is to intervene. As noted earlier, both of these factors impacted gap selection. Hence, when these factors are included in the model (i.e., their effect is accounted for in the regression model), drivers are shown to intervene more with short gap settings (or shorter time headway settings) than with medium and long gap settings. The relative likelihood of intervention for short gap settings is 1.4 times higher (1/e∧ (−0.33)) than that for long gap settings. Additionally middle-aged drivers are 1.5 (e∧ 0.41) times more likely to intervene compared with younger aged drivers. B. Safety Consequence As previously discussed, studies have shown that fewer lane changes and more consistent speeds can be viewed as positive safety outcomes with ACC use. In this study, drivers did not change lanes or accelerate as often with ACC, but only responded by either manually braking or allowing ACC to automatically brake during a “closing” event. Hence, a negative safety outcome was not apparent. Drivers’ decision making in braking has been related to timeto-collision [34]. By examining the last-second braking behavior of expert drivers, Wada et al. [35] developed a control method for a braking-assistance system. To further examine drivers’ intervention related to safety-critical events, these events were classified into three categories, namely, low-risk (no prompt intervention is needed to avoid a crash), conflict, or near-crash events. Conflict and near-crash events were determined by identifying situations where the ACC vehicle exceeded defined levels based on time-to-collision and range rate with a lead vehicle. These classifications were based on results of test track studies conducted by the General Motors–Ford Crash Avoidance Metrics Partnership to characterize drivers’ last-second braking and steering performance [29]. For example, a “closing” event is classified as a near crash whenever it crosses the near-crash boundary, regardless of the starting or ending point (see right side of Fig. 3). There were 606 events not defined and were therefore coded as unknown events.

At low-risk situations, the frequency of automatic braking was 13 times higher than that of manual braking (see Table V). During conflicts, the frequency of automatic braking was similar to that of manual braking. However, the ratio of automatic to manual braking frequency was much lower in the near-crash event situations when compared with the low-risk and conflict situations. Hence, the ACC system may help avoid a crash in the majority of the low-risk events (92.9%), in more than half of the conflict events (57.8%), and in less than half of the near-crash events (44.1%). IV. D ISCUSSION The goal of this study was to examine drivers’ adaptation to new technology. Although the ACC system examined in this study has substantially evolved since its inception, this study does provide insights on drivers’ behavior when first exposed to a novel in-vehicle technology. We quantified drivers’ responses as they closed in on a lead vehicle while ACC was engaged. The findings showed that drivers would do one of two things: either allow ACC to take control or override ACC by manually braking. Drivers’ intervention played an important role in conflict and near-crash situations. The drivers’ responses reflected high acceptance of ACC from an objective perspective that complemented the survey results in the original study. The conceptual model of adaptive behavior provides a framework for the design and analysis of safety-based systems that also account for operators’ responses from continuous feedback. In this study, several initiating and mediating factors (road type, gap setting, speed, and age) were observed to influence drivers’ initial (or short-term) adaptation to the ACC system. The findings suggest that ACC may be more effective if the algorithms (for gap settings and warnings) were designed to recognize individual differences and changing environmental conditions. Vahidi and Eskandarian indicated that two major goals of ACC included maintaining a safe headway and increasing driving comfort [36]. Compared with manual braking, ACC automatic braking is usually less abrupt with a smoother transition [37]. The model showed that, when road type and speed were accounted for, the use of ACC automatic braking was more frequent with longer headway settings, confirming the two major goals of ACC. Drivers that maintain

1472


longer headway settings may be more comfortable allowing ACC to brake automatically, a finding that may relate to Hoedemaeker’s study that showed that close following distances can be stressful for drivers [38]. ACC systems can be overridden by driver intervention. In this study, drivers were more likely to intervene (brake manually) with no lane change when ACC was on. This could be considered a positive outcome [20], [21]. Unfortunately, drivers were less likely to intervene on roads that typically have higher posted speeds (when ACC is most likely to be engaged). Although this may be expected, a negative safety consequence may result in safety-critical situations since drivers may not response quickly enough. In this case, the initiating (ACC feedback) and mediating (speed) factors may actually negate any positive effects. Hence, recognizing how drivers’ response may change over time is an important factor for safety. Future research will need to consider whether drivers can appropriately respond regardless of the speed selection with ACC engaged. This study was based on data from a FOT that was conducted over a relatively short period of time (three weeks of ACC use). Hence, there are limitations with regard to capturing severe near-crash data. Future studies may be able to evaluate this effect using a more controlled setting, such as a simulator or test track. Some mediating factors such as users’ motivation and attitudes/biases were also not available in the original data set and were therefore not included in the logistic regression model. It would be also useful to test the predictive nature of the regression model using more recent data. Future studies should also examine the usefulness of this model to predict outcomes for moderate and long-term use of ACC. Finally, this study only considered drivers’ response to a lead vehicle, and data were not captured on trailing vehicles. In some cases, maintaining the gap with a lead vehicle may actually decrease the gap with the following vehicle, creating an even greater safety-critical situation. The behavioral outcomes from this study provide insights for the introduction of other novel systems placed inside the vehicle. One of the goals of this study was to establish a conceptual model of adaptive behavior in the driving domain and identify factors that influence drivers’ adaptation and overall safety. A general finding from this study is that the design of safety-based systems needs to account for changing range and headways, as well as road type, traveling speed, and individual user differences. This study captures some insights on factors related to ACC use based on naturalistic data, but this is only a first step. The findings demonstrate a need for future research on driver adaptation over prolonged use. Additional research is needed to understand how continual exposure to ACC and other such technology can impact drivers’ behavior and how drivers’ adaptation to ACC during safety-critical events can be taken into consideration in future systems. ACKNOWLEDGMENT The authors would like to thank Dr. J. Sayer (UMTRI), the IEEE reviewers and Editor for their insightful comments, and the researchers at the Human Factors and Statistical Modeling Laboratory for technical editing. R EFERENCES [1] P. Szuszman, “Introduction to the workshop problem: Adaptive cruise control (ACC),” in Proc. 5th Meeting U.S. Softw. Syst. Safety Work. Group, Anaheim, CA, Apr. 12–14, 2005. [2] R. Bishop, Intelligent Vehicle Technology and Trends. Boston, MA: Artech House, 2005. [3] W. Wang, W. Zhang, and H. Bubb, “Car-following safety algorithms based on adaptive cruise control strategies,” in Proc. 5th Int. Symp. Intell. Syst. lnf., Subotica, Serbia, Aug. 24–25, 2007, pp. 135–140.

[4] R. Rajamani, Vehicles Dynamics and Control. New York: SpringerVerlag, 2006. [5] M. Minderhoud and P. H. L. Bovy, “Impact of intelligent cruise control strategies and equipment rate on road capacity,” in Proc. 5th Annu. World Congr. ITS, Seoul, Korea, 1998. [6] A. Kesting, M. Treiber, M. Schönhof, and D. Helbing, “Extending adaptive cruise control to adaptive driving strategies,” J. Transp. Res. Board, vol. 2000, pp. 16–24, 2007. [7] M. Young and N. A. Stanton, “Taking the load off: Investigations of how adaptive cruise control affects mental workload,” Ergonomics, vol. 47, no. 9, pp. 1014–1035, Jul. 2004. [8] M. Hoedemaeker and K. Brookhuis, “Behavioural adaptation to driving with an adaptive cruise control (ACC),” Transp. Res. F, Traffic Psychol. Behav., vol. 1, no. 2, pp. 95–106, Dec. 1998. [9] N. A. Stanton and P. Marsden, “From fly-by-wire to drive-by-wire: Safety implications of automation in vehicles,” Safety Sci., vol. 24, no. 1, pp. 35–49, Oct. 1996. [10] J. E. Naranjo, M. A. Sotelo, C. Gonzalez, R. Garcia, and T. de Pedro, “Using fuzzy logic in automated vehicle control,” IEEE Intell. Syst., vol. 22, no. 1, pp. 36–45, Jan./Feb. 2007. [11] L. Xiao and F. Gao, “Practical string stability of platoon of adaptive cruise control vehicles,” IEEE Trans. Intell. Transp. Syst., vol. 12, no. 4, pp. 1184–1194, Dec. 2011. [12] C. M. Rudin-Brown and H. Parker, “Behavioral adaptation to adaptive cruise control (ACC): Implications for preventive strategies,” Transp. Res. F, Traffic Psychol. Behav., vol. 7, no. 2, pp. 59–76, Mar. 2004. [13] P. Venhovens, K. Naab, and B. Adiprasito, “Stop and go cruise control,” Int. J. Automotive Technol., vol. 1, no. 2, pp. 61–69, 2000. [14] L. C. Davis, “Effect of adaptive cruise control systems on mixed traffic flow near an on-ramp,” Phys. A, Stat. Mech. Appl., vol. 379, no. 1, pp. 274–290, Jun. 2007. [15] G. Marsden, M. McDonald, and M. Brackstone, “Towards an understanding of adaptive cruise control,” Transp. Res. C, Emerg. Technol., vol. 9, no. 1, pp. 33–51, Feb. 2001. [16] N. Wu, F. Chu, S. Mammar, and M. Zhou, “Petri net modeling of the cooperation behavior of a driver and a copilot in an advanced driving assistance system,” IEEE Trans. Intell. Transp. Syst., vol. 12, no. 4, pp. 977–989, Dec. 2011. [17] J. Pauwelussen and P. J. Feenstra, “Driver behavior analysis during ACC activation and deactivation in a real traffic environment,” IEEE Trans. Intell. Transp. Syst., vol. 11, no. 2, pp. 329–338, Jun. 2010. [18] L. Evans and P. H. Gerrish, “Antilock brakes and risk of front and rear impact in two-vehicle crashes,” Accid. Anal. Prev., vol. 28, no. 3, pp. 315–323, May 1996. [19] N. A. Stanton, M. Young, and B. McCaulder, “Drive-by-wire: The case of driver workload and reclaiming control with adaptive cruise control,” Safety Sci., vol. 27, no. 2/3, pp. 149–159, Nov./Dec. 1997. [20] T. Chira-Chavala and S. M. Yoo, “Potential safety benefits of intelligent cruise control systems,” Accid. Anal. Prev., vol. 26, no. 2, pp. 135–146, Apr. 1994. [21] R. D. Ervin, J. Sayer, D. LeBlanc, S. Bogard, M. Mefford, M. Hagan, Z. Bareket, and C. Winkler, “Automotive collision avoidance system field operational test methodology and results,” Univ. Michigan Transp. Res. Inst., Washington, DC, Dept. Transp., HS 809 900, Aug. 2005. [22] R. Ma and D. Kaber, “Situation awareness and workload in driving while using adaptive cruise control and a cell phone,” Int. J. Ind. Ergonom., vol. 35, no. 10, pp. 939–953, Oct. 2005. [23] A. Touran, M. A. Brackstone, and M. McDonald, “A collision model for safety evaluation of autonomous intelligent cruise control,” Accid. Anal. Prev., vol. 31, no. 5, pp. 567–578, Sep. 1999. [24] B. S. Kerner, “Control of spatiotemporal congested traffic patterns at highway bottlenecks,” in Proc. IEEE Intell. Transp. Syst., Vienna, Austria, Sep. 13–16, 2005, pp. 109–114. [25] L. N. Boyle and F. Mannering, “Impact of traveler advisory systems on driving speed: Some new evidence,” Transp. Res. C, Emerg. Technol., vol. 12, no. 1, pp. 57–72, Feb. 2004. [26] R. Fuller, “Towards a general theory of driver behaviour,” Accid. Anal. Prev., vol. 37, no. 3, pp. 461–472, May 2005. [27] J. Hedlund, “Risky business: Safety regulations, risk compensation, and individual behavior,” Injury Prev., vol. 6, no. 2, pp. 82–90, Jun. 2000. [28] J. A. Michon, “Explanatory pitfalls and rule-based models,” Accid. Anal. Prev., vol. 21, no. 4, pp. 341–353, Aug. 1989. [29] W. G. Najm, M. D. Stearns, H. Howarth, J. Koopmann, and J. Hitz, “Evaluation of an automotive rear-end collision avoidance system,”


[30]

[31] [32]

[33]

Nat. Highway Traffic Safety Admin., Washington, DC, Dept. Transp., HS 810 569, Apr. 2006. M. G. Kenward, E. Lesaffre, and G. Molenberghs, “An application of maximum likelihood and generalized estimating equations to the analysis of ordinal data from a longitudinal study with cases missing at random,” Biometrics, vol. 50, no. 4, pp. 945–953, Dec. 1994. S. R. Lipsitz, K. Kim, and L. Zhao, “Analysis of repeated categorical data using generalized estimating equations,” Stat. Med., vol. 13, no. 11, pp. 1149–1163, Jun. 1994. N. R. Parsons, M. L. Costa, J. Achten, and N. Stallard, “Repeated measures proportional odds logistic regression analysis of ordinal score data in the statistical software package R,” Comput. Stat. Data Anal., vol. 53, no. 3, pp. 632–641, Jan. 2009. U. Halekoh, S. Hjsgaard, and J. Yan, “The package geepack for generalized estimating equations,” J. Stat. Softw., vol. 15, no. 2, pp. 1–11, 2006.

1473

[34] A. R. A. V. D. Horst, “A time-based analysis of road user behaviour in normal and critical encounters,” Ph.D. dissertation, Delft Univ. Technol., Delft, The Netherlands, 1990. [35] T. Wada, S. Doi, N. Tsuru, K. Isaji, and H. Kaneko, “Characterization of expert drivers’ last-second braking and its application to a collision avoidance system,” IEEE Trans. Intell. Transp. Syst., vol. 11, no. 2, pp. 413–422, Jun. 2010. [36] A. Vahidi and A. Eskandarian, “Research advances in intelligent collision avoidance and adaptive cruise control,” IEEE Trans. Intell. Transp. Syst., vol. 4, no. 3, pp. 143–153, Sep. 2003. [37] P. Ioannou and M. Stefanovic, “Evaluation of ACC vehicles in mixed traffic: Lane change effects and sensitivity analysis,” IEEE Trans. Intell. Transp. Syst., vol. 6, no. 1, pp. 79–89, Mar. 2005. [38] M. Hoedemaeker, “Driving with intelligent vehicles: Driving behavior with ACC and the acceptance by individual drivers,” in Proc. IEEE Intell. Transp. Syst., Dearborn, MI, Oct. 1–3, 2000, pp. 506–509.