18-02-2011, 02:48 PM
ieee pdf.pdf (Size: 1.72 MB / Downloads: 89)
Driver Behavior Analysis During ACC Activationand Deactivation in a Real Traffic Environment
Abstract—
For the development of a traffic-simulation modelto estimate the effect of adaptive cruise control (ACC) systemson traffic safety, throughput, and environment, data of a fieldoperational test (FOT) were analyzed, in which vehicles wereequipped with ACC and lane-departure warning (LDW) systems.The objective of this paper is to use this FOT to investigate thedeactivation or (re)activation of the ACC on driver behavior in areal traffic environment. By taking these results into account inthe traffic simulation environment, a more realistic evaluation ofthe impact of ACC on safety, throughput, and environment canbe achieved. Some of the conclusions that were found are thatafter the participants deactivated the ACC by pressing the brakepedal, the gap with the lead vehicle was decreased. Resuming theACC by activating the system or by releasing the throttle afteroverruling the system resulted in a larger gap between participantand lead vehicle than an overruled ACC or the ACC turnedoff. The participants overruled the ACC by pressing the throttlemainly to overtake the lead vehicle.Index Terms—Adaptive control, driver behavior, driverperformance..
INTRODUCTION
THE CARS of tomorrow will increasingly be equipped withadvanced driver-assistance systems (ADAS) to support thedriver in the driving task to improve traffic safety, throughput,and/or the environment. One of these ADAS is the adaptivecruise control (ACC), which is also known as advanced orintelligent cruise control (CC) [1], [2]. The ACC could bedefined as an extension of the CC and maintains, besides acertain set speed, a certain set distance with respect to the leadvehicle. To keep this set distance, the ACC is able to accelerateand decelerate the vehicle. The acceleration and deceleration ofthe system are limited because of comfort and legal reasons.This means that the driver has to intervene if the system isnot able to achieve the required needs. The ACC is mainlya comfort system that takes over the car-following task. Thedriver remains responsible for steering and collision avoidance.If the deceleration of the ACC is not sufficient to avoid acollision, then the ACC warns the driver with a warning sound. In addition to considering the ACC only as a support systemfor the individual driver’s comfort, it could also be beneficial atan aggregated level, i.e., traffic safety, throughput, and environment.Because the number of ACCs in real traffic is limited anddifficult to determine, traffic simulation models are often usedto investigate the impact of the ACC on traffic behavior.Traffic simulations have shown that the ACC could improvetraffic safety because of better distance keeping [3]–[7]. Thiscould result in less traffic accidents and might therefore alsobe an improvement of the traffic throughput. According to [8],full penetration of the ACC may reduce rear-end collisionsby up to 17%. Furthermore, it is expected that the ACC willreduce the environmental pollution because of smoother accelerations[9]–[11]. However, many of these ACC models incurrent traffic simulations only provide insight on the effectof the ACC system for different penetration levels under theassumption that the ACC system remains active under almostall circumstances. In particular, according to [12]–[15], thecircumstances in which the driver has to take over control, i.e.,switching between automation and manual, could have a largeeffect on traffic safety and traffic throughput.To improve the validity of traffic simulation model results,these circumstances should be accounted for by extending thecurrent driver behavior model, e.g., allowing for realistic deactivatingand activating ACC. Furthermore, a better understandingof the use of the ACC system by the driver is a requirementto improve this system to be more acceptable and complementaryto the driver’s own intended actions, as emphasized in[16]–[19].In this paper, we analyzed real traffic data to investigatereal driver behavior and the drivers’ motives for the transitionbetween ACC activation and deactivation. The results of thisanalysis will then be used for the improvement of a driverbehavior model for driving with and without ACC, whichwas described in more detail in [6] and implemented on theITS Modeler platform [4], [7]. This ITS Modeler platformis a traffic simulation model that was developed as a trafficenvironment platform in which ADAS and intelligent cooperativevehicle–infrastructure systems can be modeled, tested,and evaluated to determine their impact on traffic throughput,safety, and environment.For the analysis of real driver behavior and the drivers’motives for the transition between ACC activation and deactivation,real traffic data were used. These real traffic datawere collected during the Dutch full-traffic field operationaltest (FOT) “The Assisted driver,” which was similar to theFOT that was conducted in the U.S. [20], [21]. The Dutch fulltrafficFOT was part of the “Roads to the Future” project in which 19 participants drove with an instrumented vehicle forsix months [22], [23]. The vehicles were equipped with an ACCand a lane-departure warning (LDW) system. During the firstmonth, they drove with the systems turned off. This was calledthe before measurement or reference measurement period. Thesecond month was the transition period, in which participantshad the time to get used to the system. The last four months areknown as the after measurement period, with the participantsbeing asked to use the systems, but it was not obligated. Fromeach vehicle, several driving measures were collected, e.g., thespeed and acceleration of the vehicles, their location, the statusof ACC and LDW, and the distance and speed with respect tothe lead vehicle.These full-traffic data were also analyzed in two previousstudies. The conclusions of these studies were, besides theoutcome of the study described in this paper, also used toimprove the already developed driver-behavior model in theITS Modeler, which was mentioned earlier, to obtain a morerealistic traffic simulation. A few important conclusions ofthese studies are therefore repeated here. The first study was theanalysis within the “Roads to the Future” project, which consideredthe effects of ACC and LDW on traffic safety, throughput,and environment, and when these systems were active [22],[23]. Some of the conclusions were that the ACC was oftenused during free driving and busy traffic (i.e., speed between 70and 90 km/h on the highway) and was hardly used during slowtraffic. When the ACC was active, it was concluded that the timeheadway (THW) in car-following situations, which is definedas the time separation between the fronts of two successivevehicles passing the same point on a road, was around 0.2 shigher than without the ACC. In addition, the variation in THWwas smaller with an active ACC. Furthermore, it was concludedthat the results found for the ACC were not influenced by theLDW. The second study [24] investigated the effect of ACC onbehavioral adaptation, which was focused on traffic throughput.According to this study, the ACC was often deactivated underdense traffic conditions. It turned out that drivers tend to overrulethe ACC and make it inactive under densities ranging from20 to 40 vehicles/km/lane.In this paper, we investigate the real driver behavior, and thedrivers’ motives for the transition between ACC activation anddeactivation were described. In Section II, the research methodis discussed, which considered the specifications of the ACCthat was used, the data set of independent variables that wereconsidered, the dependent driver-behavior measures, and thepossible driver motives for ACC state transitions. The results ofthe analyzed driver behavior measures and the drivers’ motivesare described in Section III. Section IV draws the final conclusions,which should be implemented in the driver-behaviormodel.II. METHODFor the analysis of driver behavior during the activation anddeactivation of the ACC, the differences in driver-behaviormeasures were analyzed just before and just after the ACCstate transition. Furthermore, to understand why the driversmainly activate and deactivate the ACC, the drivers’ motives for activation or deactivation of the ACC system in the Dutchfull-traffic operational test were investigated.A. ACC SpecificationsThe ACC system was limited to a maximum deceleration of3m/s2 and a minimum speed of 30 km/h. The THWs that couldbe set were 1.0, 1.4, 1.8, 2.2, 2.6, and 3.0 s. The ACC could bein four different states:State 0: ACC off (by on/off switch);State
1: ACC active;State
2: ACC inactive;State
3: ACC active and accelerate
(i.e., ACC active anddriver presses the accelerator pedal).The ACC was activated (State 1) and turned off (State 0)by an on–off switch (see Fig. 1). The driver could overrule theACC in two ways. One way was by pressing the brake pedal.This would deactivate the ACC (State 2). The driver could againmanually resume the ACC system by the reset switch. The otherway to overrule the ACC was by pressing the gas pedal for extravehicle acceleration (State 3). The ACC would still be activeand return to the set THW and speed after the gas pedal wasreleased.B. Data SetThe transitions that were considered only include the transitionsthat were caused by driver actions and do not includetransitions that could be caused by ACC system faults. The transitionfrom ACC off (State 0) to the other states only occurredsporadically and was therefore not further considered. Thetransitions that were considered were referred to as follows:A. ACC Inactive (State 2) → ACC Active (State 1) bypressing reset switch;B. ACC Active (State 1) → ACC Inactive (State 2) bypressing brake pedal;C. ACC and Accelerate (State 3) → ACC Inactive (State 2)by pressing brake pedal after accelerating;D. ACC Active (State 1) → ACC and Accelerate (State 3)by pressing accelerator pedal;E. ACC and Accelerate (State 3) → ACC Active (State 1)by releasing accelerator pedal after accelerating.Fig. 1 shows the considered ACC state transitions and howthe transition is executed. In this paper, we are mainly interestedin the circumstances in which the driver switched from oneACC state to the other and the effects that were shown ondriver behavior. We consider driver behavior as the primarytask performance, i.e., car-following and speed-regulation taskperformance, just before and just after the transition. The driverbehaviormeasures in the period 10 s before and 10 s after thesetransitions were taken into account. All transitions between theACC states, which are represented by A, B, C, D, and E inFig. 1, were considered for all 19 participants of the full-trafficFOT. Because the ACC was most often used in the speed rangebetween 70 and 130 km/h, only the data set for this speed rangewas considered. C. Driver-Behavior MeasuresTo consider the driver behavior at the previously discussedtransition points, two types of longitudinal control metrics wereanalyzed: 1) car following and 2) speed regulation. For car following,the THW and the relative speed (Vrel=Vcar −Vlead)divided by the car speed (Vrel/Vcar) were analyzed, which isalso referred to as the normalized closing speed (NCS). TheNCS was considered instead of, for example, time-to-collision(TTC), because TTC depends on the THWthat was already considered.A positive NCS indicates that the participant approachesthe lead vehicle. For speed regulation, the speed anddeceleration of the subject vehicle were considered.The means and standard deviations of the car-following andspeed-regulation measures were calculated because the meansand standard deviations in the traffic simulation model for theACC state transitions in the ITS Modeler should correspond tothose means and standard deviations calculated from real trafficdata.Furthermore, for traffic safety, it was interesting to considerthe events in which the driver drove “dangerously” close to thelead vehicle, approaching the lead vehicle fast or deceleratingvery hard. Therefore, the minimum THW and maximum NCSfor car-following behavior and the minimum speeds and maximumdecelerations for speed regulation were analyzed in moredetail, as well as how often these behavior measures occurred,i.e., the number of samples. The extreme values of each of thepreviously described measures that occurred during a periodof 10 s before the ACC state transition were collected andcompared with the extreme values that occurred in a periodof 10 s after the transition. The results were represented in one figure with the behavior measures on the x- and y-axes inthe horizontal plane and the number of samples on the verticalz-axis, as shown in the two plots on top of Fig. 2. For a betteroverview of the results, the figure was viewed from the top witha legend that indicates the number of samples, as shown in thetwo plots on the bottom of Fig. 2.The number of samples was represented in a percentagewith respect to the maximum number of times that the driverbehavior measures occurred. This means that the number ofsamples of 100% indicates which driver-behavior measuresoccurred most often, i.e., the maximum number of samples. Forexample, a number of samples of 50% means that these driverbehaviormeasures occurred half as often as the maximumnumber of samples. For the comparison between the driverbehaviormeasures before and after the transition, it becomesimmediately clear if the maximum number of samples shifted