刘恩泽-Sensitivity of the lane change test as a measure of in-vehicle system demand.pdf

VictoriaT) isbrotont sdiswithrsioet al., 20technologiesfromcations of in-vehicle systems. Several such methods have beendeveloped, including the visual occlusion technique (Chiang et al.,2004; Gelau et al., 2009; Noy et al., 2004; Senders et al., 1967)and the peripheral detection task (Harms and Patten, 2003; Olssonand Burns, 2000,p. 8; van Winsum et al., 1999).et al., 2000). These distraction types, particularly visual andcognitive distraction, have been shown to impair different aspectsof driving performance, with lateral control and event detectionmetricsbeingparticularlysensitivetodifferentformsofdistraction.For instance, visual load has been shown to increase lane keepingvariation (e.g., Greenberg et al., 2003; Zwahlen et al., 1988).Moderate levels of cognitive load, in contrast, have been shown tohave little effect on lane keeping performance and can even lead tomore precise lateral control (Brookhuis et al.,1991; Engstrom et al.,* Corresponding author. Tel.: 613 9905 1258; fax: 613 9905 4363.Contents lists availableApplied ErgonomicApplied Ergonomics 42 (2011) 611e618E-mail address: .au (K.L. Young).policy initiatives worldwide (Collet et al., 2009; Wittmann et al.,2006). An important goal in the design of these systems is toensure that their use while driving does not interfere with thedrivingtaskandundulycompromisesafety.Thisisachallengeinanarea where the introduction of technology is largely commercial,rather than safety driven. The realisation of this goal is dependantupon the provision of widely accepted and scientically robustmethods for informing the design and assessing the safety impli-the differential effects of various types of distraction on drivingbehaviour.1.1. Driver distractionDriver distraction is commonly described as comprising a rangeof different, but not mutually exclusive, elements; for example,visual, cognitive, auditory and biomechanical (physical) (Ranney1. IntroductionDriver distraction is acknowledgedimportant road safety issue (Reganpotential for in-vehicle and portableinformation, communication, entertainmentassistance systems, to distract driversdegrade performance has been the subject0003-6870/$ e see front matter C211 2010 Elsevier Ltddoi:10.1016/j.apergo.2010.06.020sensitive to the different levels of task demand. The other LCT metrics examined were less sensitive totask differences. A major factor inuencing the sensitivity of at least some of the LCT metrics could bethe type of lane change instructions given to participants. The provision of clear and explicit lanechange instructions and further renement of its metrics will be essential for increasing the utility ofthe LCT as an evaluation tool.C211 2010 Elsevier Ltd and The Ergonomics Society. All rights ernationally as an08). In particular, the, including,and advanced driverthe driving task andof intense research andAs evidenced by its current development into an ISO standard(ISO, 2009), another candidate methodology that shows promise inthis area, is the Lane Change Test (LCT; Mattes, 2003; Mattes andHalln, 2008). To be useful as an evaluation tool, however, theLCT must be valid (i.e., it measures what it claims to measure) andreliable (i.e., the results obtained are consistent across adminis-trations), as well as having high sensitivity. The focus of this paperis on the LCT methods sensitivity. That is, its ability to distinguishevent detection metrics were less able to discriminate between tasks. The mean deviation and laneexcursion measures were able to distinguish between the visual and cognitive tasks, but were lessIn-vehicle information systemsEvaluation methodstasks. A number of the lateral control metrics were found to be sensitive to task differences, but theSensitivity of the lane change test as a measurKristie L. Young*, Michael G. Lenn, Amy R. WilliamsonMonash University Accident Research Centre, Building 70, Monash University, Clayton,article infoArticle history:Received 29 October 2009Accepted 18 June 2010Keywords:Driver distractionLane change testabstractThe Lane Change Test (LCperformance degradationreliability, for such a testindividual tasks. The curredetection parameters toInformation System tasksyears) completed a PC vejournal homepage: www.elsand The Ergonomics Society. All rightse of in-vehicle system demand3800, Australiaone of the growing number of methods developed to quantify drivingught about by the use of in-vehicle devices. Beyond its validity andbe of practical use, it must also be sensitive to the varied demands oftudy evaluated the ability of several recent LCT lateral control and eventcriminate between visual-manual and cognitive surrogate In-Vehicledifferent levels of demand. Twenty-seven participants (mean age 24.4n of the LCT while performing visual search and math problem solvingat ScienceD/locate/apergoreserved.2005; Jamson and Merat, 2005). Further, both cognitive and visualtasks can impair event detection (Klauer et al., 2006), but cognitivedistraction can also impair drivers ability to respond to eventsquickly and adequately (Consiglio et al., 2003; Recarte and Nunes,2003; Strayer et al., 2003).Given its current development into an ISO standard and itsincreasing use in distraction research, it is important that the LCT iscapable of measuring and distinguishing these diverse distractioneffects. This study therefore aimed to evaluate the sensitivity ofa range of LCT metrics in being able to distinguish between visual-manual and cognitive tasks with different levels of demand.1.2. The lane change testThe LCT is a PC-based driving simulation that is designed toquantitatively measure the level of degradation in driving perfor-mance induced by the simultaneous performance of a secondarytask. It has been widely used to assess driving performance withconcurrent use of a range of in-vehicle information systems (IVIS)which provide information that supports primary driving tasks(e.g., navigation), as well as Advanced Driver Assistance Systems(ADAS) that directly support the primary driving task (Burns et al.,2005, pp. 1980e1983; Mntyl et al., 2009).A numberof studieshave focusedonvalidatingthe LCT. Manyoftheseearlyvalidationtestswerecarriedoutaspartof theAdvancedDriver Attention Metrics (ADAM) project, in which the LCT wasdeveloped, and suggest that the LCT is avalid, reliable and sensitivemeasure (see Mattes and Halln, 2008). Subsequent researchdemonstrated that the LCT could discriminate between secondarytasks with different workload levels (Burns et al., 2005, pp.1980e1983),withdriversdemonstratingagreaterdeviationinlanechange path when performing a complex versus simple navigationtask while driving.More recently, work has continued on the LCT to expand itsdiagnostic power by proposing new performance metrics (Mattesand Halln, 2008). Given the complexity and multifaceted natureof distraction, it is important foranyevaluation method tomeasuremultiple aspects of the driving task in order to draw conclusionsabout the safety effects of in-vehicle devices.A number of studies have examined the sensitivity of several oftheLCTmetricsin beingabletodistinguishbetweendifferenttypesof distraction (Bruyas et al., 2008; Engstrm and Markkula, 2007;Harbluk et al., 2009,24e30 p.). Engstrm and Markkula (2007)have examined the sensitivity of two new LCT metrics e pathcontrol (high-pass ltered SDLP) and sign detection/recognition(Percent correct lane; PCL) e to distinguish visual and cognitivetasks. Results revealed that the two types of distraction eachimpaired LCT performancedifferently. The visual, but notcognitive,tasks led to reduced path control, while the cognitive, but notvisual, tasks affected detection and sign recognition and responses.Bruyas et al. (2008) found that the adapted mean deviation score,ratio of correct lane changes and Lane Change Initiation (LCI)metrics were capable of differentiating some visual-manual andauditory tasks, but not others. Finally, in order to take into accounttask duration, Harbluk et al. (2009,24e30 p.) examined the LCTmean deviation per average task by dividing the mean deviationscore by the number of task completed per run. They found thatthis adapted measure was better able than the original meandeviation score to discriminate between navigation tasks withdifferent levels of complexity.These studies demonstrate that at least some of the proposedLCT performance metrics are sensitive to the disparate effects ofdifferent forms of distraction. However, there is still a need todetermine if other LCT metrics, that are increasingly being used byK.L. Young et al. / Applied Ergonomics612researchers and policy and system developers to draw conclusionsaboutthe safetyanddesignaspectsofIVISsystems(e.g.,MaciejandVollrath, 2009), are also sensitive to task differences. This studyextends the ndings of the previous research discussed above byexamining the sensitivity of a range of new and recently proposedLCT lateral control and event detection parameters (lane keepingvariation between signs, percent correct lane changes, number oflane excursions, LCI, and mean steering angle) in being able todistinguish between visual-manual and cognitive tasks withdifferent levels of demand. Thendings of this studycan be used toinform decisions regarding which LCT metrics are suitable for useand whichones mayneed furtherrenement. It will alsoadd tothegrowing number of studies aimed at establishing the psychometricproperties of the LCT as part of its development into an ISOstandard.2. Method2.1. DesignThis study used a repeated-measures design, with one inde-pendent variable, task condition, which had ve levels: a baseline(no secondary task) condition and four secondary task conditions:visual easy, visual hard, cognitive easy and cognitive hard. Partici-pants completed the four secondary task conditions while drivinga PC version of the LCT. Further details of the secondary taskconditions are contained in Section 2.3.2. This combination ofsecondary task conditions ensured that it was possible to examinethe ability of the LCT to distinguish between different levels ofdemand as well as different types of distraction.2.2. ParticipantsTwenty-seven drivers who held a valid drivers license partici-pated the study. Sixteen of the participants were male and 11 werefemale and the mean age for the group was 24.4 years (SD 3.0;range 21e31 years). All participants held a valid full driverslicense, apart from one who held a probationary license, which isissued for the rst four years of solo driving and contains certainpassenger, mobile phone and vehicle power restrictions. The meanage at which their solo (probationary) license was obtained was19.3 years(SD2.6),andthe averagetime spentdrivingeachweekwas 7.3 h (SD 6.6).Participants were recruited through campus notice boards andnewsletters, the Monash Careers Website and the local newspaper(WaverleyLeader).EthicsapprovalforthestudywasgrantedbytheMonash University Standing Committee on Ethics in ResearchInvolving Humans (SCERH). Participants were reimbursed for theirtime and travel expenses.2.3. Materials2.3.1. Driving taskDriving performance was measured using the LCT (version 1.2;Mattes, 2003). The LCT is a simple driving simulation consisting ofa 3000 m straight, three-lane road. Speed is limited to 60 km/h bythe system, which the test participants were asked to maintainthroughout the drive. No other trafc is present on the road. Thedrivers are instructed to change lanes via 18 signs that appear oneach side of the road every 150 m, on average. The signs are blankuntil 40 m before the sign, at which point the lane change infor-mation is given (Fig. 1a). Participants were instructed to changeslanes as soon and as quickly as possible after they see the infor-mation appear on the sign. Participants were not required to have42 (2011) 611e618completed their lane change before they reached the sign.withTheLCT wasrun on adesktopPC.Theset-up of the testfollowedthat set out in the Draft ISO standard (ISO, 2009). The visual scenewas presented on a single 1900LCD monitor. Control of the simu-lation was achieved through a Logitech MOMO force-feedbackgaming steering wheel with accelerator and brake foot pedals.Participants sat in a height adjustable chair, with the chair andpedals positioned to approximate that of a real vehicle.2.3.2. Secondary taskTwo surrogates visual-manual and cognitive IVIS tasks wereused as the secondary tasks, each with two levels of difculty.Articial or surrogate in-vehicle information system (IVIS) taskswere used so that the level of task demand could be systematicallymanipulated.2.3.3. Visual-manual taskThe Surrogate Reference Task (SuRT v.2.1) was employed as thevisual-manual task. The task requires participants to search fora larger target circle among visually similar, smaller distractercircles (visual demand), and to select, using the keyboard arrowkeys, the portion of screen containing the target (manual demand).Task difculty (easy, hard) was manipulated by varying the size ofthe distracter circles compared to the target circle, and byincreasingthenumberofscreensegmentsinwhichthetargetcouldappear (easy 2 regions, hard 6 regions). For the easy condition,thetargetcirclewastwicethesizeofthedistractercircles,whileforFig. 1. Primary and secondary tasks used. Panel A (left) shows the LCT driving scenecondition).K.L. Young et al. / Applied Ergonomicsthe hard condition the target circle was 15% larger than the dis-tracters. An example of the SuRT display is contained in Fig.1b. Thevisual task was semi self-paced, whereby the participants couldtake as much time as needed to make their selection, but thesoftwarecontrolledwhenthenextstimuliwaspresented.TheSuRThas been used widely in IVIS research (e.g., Bruyas et al., 2008;Rognin et al., 2007) and has been found to be a valid measure ofIVIS task demand (Wynn and Richardson, 2008).The visual secondary task was presented on a laptop with 1400screen which was located on a table to the left of the driver, within30C14(horizontal and vertical) of their normal eld of view andwithin easy reach.2.3.4. Cognitive taskThe cognitive task comprised a math problem solving task,involving basic addition, again with two levels of difculty e easyand hard. Random numbers were read aloud to the participant(through a headset) using DirectRTsoftware. For the easy level, theparticipant was asked to add 5 to the number and say theirresponse out loud.Forthe difcult level, participantswererequiredto add 7. The task was semi self-paced, whereby participants weregiven as much time as needed to respond to each problem (self-paced), but the system presented the next problem immediatelyafter the previous response was given (system-paced). This taskwas piloted extensively prior the experiment to ensure that therewas an adequate degree of difcultly and differentiation amongstthe easy and hard levels.2.4. ProcedureOn arrival at the session, participants completed a demographicquestionnaire.TheywerethengivenaverbalexplanationoftheLCTand secondary tasks, followed by static (no LCT) practice andbaseline trials of the visual and cognitive secondary tasks. Partici-pants completed 1e2 practice drives on the LCT and then the sixtrial runs: baseline, visual easy, visual hard, cognitive easy, cogni-tive hard and a nal baseline trial. Before each dual-task drive,participants were instructed to “concentrate your attention ondriving safely, but do not ignore the secondary tasks”.The order in which the visual and cognitive secondary taskswere presented was counter-balanced across participants toaccommodate any practice effects.2.5. Data analysisThe LCT driving performance examined can be broadly cat-lane change sign (arrow). Panel B (right) shows the SuRT task visual display (easy42 (2011) 611e618 613egorised into lateral vehicle control measures and event detectionmeasures:2.5.1. Lateral control measures. Mean deviation. Drivers mean lateral deviation score wascompared to the LCT normative model that is automaticallycalculated by the LCT analysis software. The normative modelrepresents an ideal lane change path. Deviation scores werecalculated for each run over the entirelength of the drive (standardmean deviation score), as well as for the straight sections betweenlanechanges(10mafterchangeto10mbeforenextchange),whichprovides an additional measure of lateral control with the effect ofthe lane changes removed. Instances where the participants hadmade an incorrect lane change were excluded from the straightsection analysis as they over-inated the deviation score.. Mean steering angle. The mean steering angle in grads(equivalent to1/400of a full circle) was calculated for each entireLCT run. This measure was calculated automatically using thestandard LCT analysis software.. Lane excursions. Thenumberoflaneexcursionsmadeduringeach run was calculated by examining the lateral position traceschematicsproducedbytheLCT.Alaneexcursionwasdenedas anyinstance where the LCT actual deviation trace moved outside of thecorrect lane of travel. Lane excursions were examined for the entiredrive,includingduringlanechangemanoeuvresandstraightsections.2.5.2. Event detection measures. Lane change initiation (LCI). LCI represents the difference(in metres) between the distance at which the lane change infor-mation appears on a sign and the distance at which the driveractually initiated the lane change. Thus, LCI represents an eventdetection measure.entire drive differed signicantly across the task conditions (F(4,100) 6.53, p.05).Duetoacorrupteddatale,meandeviation and LCI datawere not able to be calculated for one driver.3. Results3.1. Lateral control measures3.1.1. Mean deviationThe mean deviation scores for the entire drive and the straightlinesegmentsarepresentedinFig.2.Themeandeviationacrossthe1.81.92Baseline Cog Ea