44696988 Automation of the Driving Task

8/8/2019 44696988 Automation of the Driving Task

1/65


2/65

............................Ministerie vat1 Verkeer en W aterstaatinfarmatlo en DocumentatiePostbus 209012500 EX Oer1 klaagTel. U/O-3517086 / Fax. 070-3516430I

Automation of the driving taskFinal report

R-98-9Torn Heijer, SWOV Institute for Road Safety ResearchKarel Brookhuis, Centre for Environmen tal and Traffic PsychologyWim 'van Winsum, TN O Human Factors Research InstituteLies Iluynstee, Transport Research Centre of the Dutch M inisby of Transport and Public WorksLeidschendam, 1998SWOV Institute for Road Safety Research, The Netherlands


3/65

Report documentationNumber:Title:Subtile:Author(s):

Research manager:Project number SWO V:Project code client:Client:

Keywords:

Contents of the project:

Number of pages:Price:Published by:

R-98-9Automation of the driving taskFinal reportTom Heijer, SWOV institute for Road Safety ResearchKarel B rookhuis, Centre for Environmental and Traffic PsychoiogyWim van Winsum, TNO Huma n Factors Research InstituteLies Duynstee, Transport Research Centre of the Dutc h Mhistry ofTransport and Public WorksSiem Oppe54.5 12HVVL 97.501This research was nded by the Dutch Ministry of Transport andPub ic Works.Driver information, safety, traffic, driving (veh), telecommunication,data processing, simulation, stress (psychol), evaluation (assessment),design (overall design), psychology.The fast development of al1sorts of telematic devices to support orpartially substitute a drivers tasks has also led to some concern for thepossible detrimental effects that such devices rnay have on the safety ofdriving. This report summarises the final results of a three year projectaimed a t the development of criteria to assess the effects on road safetyof various applications of Advanced Traffic T elematics (ATT systems)intended to support the driver in different aspec ts of the driving task.66 PP.Dfl. 5,-SWO V, Leidschendam, 19 98

S W O V Institute for Road Safety ResearchP.O. Box 10902260 BB LeidschendamThe N etherlandsTelephone 3 1703209323Telefax 31703201261


4/65

SummaryThe fast development of al1 sorts of teiematic devices to support or partialiysubstitute a drivers tasks has also led to some concern for the possibledetrimental effects that such devices may have on the safety of driving.A three year research project has been carried out to develop criteria andprocedures that can be used to assess the nature and exten t of thesedetrimental effects. A further aim was, to make the assessm ent procedure assimple as possible, so that it can be used by reiatively non-expert users.Moreover, these procedures aim to limit the num ber of otherwise necessary(expensive) field tests.A genera1 conciusion is, that current knowledge on this subject is not yetsufficient to provide a comprehensive set of checks and this has resulted inthe foliowing comprom ise for the testing procedure:- A checklist for safety characteristicsof telematic devices is proposedbased upon known safety effects of task load changes: overload andunderload: this checklist can be employed by non-scientists to provide afirst screening of unsafe characteristics.- A iaboratory test has been developed that uses an ordinary PersonalComputer. The test emulates a simplified driving task and canaccommodate a finctional simulacrum of a telematic device. The user isalso provided with a set of criteria to produce an assessment of the safetyeffects. This test can also be used by non-experts.Furthermore, recognising that in the current state of affairs field testing wil1still often be necessary, an attempt was made to formulate guidelines andcriteria for the setup of these tests.

3


5/65

Contents1.2.3.3.1.3.1.1.3.1.2.3.1.3.3.1.4.3.2.3.2.1.3.2.2.3.2.3.3.3.3.3.1.3.3.2.3.3.3.3.3.4.3.4.3.4.1.3.4.2.3.4.3.3.4.4.4.4.1.4.1.1.4.1.2.4.1.3.4.2.4.2.1.4.2.2.4.3.4.4.4.5.5.5.1.5.1.1.5.1.2.5.1.3.5.2.5.2.1.5.2.2.5.2.3.6.6.1.6.2.6.2.1.

Genera1 introductionSetup of he projectResults of the literature study and experimentsThe literature sudy by TNO-HFRI nd COVIntroductionOveroadUnderloadCounterproductive behaviour adaptationExperimental results by TNO-TMIntroductionMethodMajor resultsExperimental results by COVIntroductionDevelopment of a prototype intelligent speed adaptorMethodResults and conclusionsResults and conclusions with respect to guidelines, criteriaA fiamework for developing safety related guidelines, criteriaSafety assessment and performance guidelinesProcedural guidelines for safety testingProduct guidelinesCriteria or experimental testngSummary of resultsRecommendations for the general design of the experimentsRecommended parametersSafety assessment by expert opinionBuilding blocks for studying overloadAssumptionsAssessing safety effectsBuilding blocks for studying underloadBuilding biocks for studying counterproductive adaptationSelecting participants and driving situationsLabora tory testngLaboratory Test Results by TNO-HFRIIntroductionMethodResults and conclusionsLaboratory Test Results by CO VIntroductionMethodResults and conclusionsA safefy checklistIntroductionA safety checklist based on taskload considerationsMental taskload

78

10101011141618181919192020212222232427283030303031313132343637393939404141414242434343435


6/65

6.2.2.6.2.3.6.3.6.4.6.4.1.6.4.2.6.4.3.7.7.1.7.2.7.3.7.4.7.4.1.7.4.2.7.4.3.7.5.7.5.1.7.5.2.7.5.3.7.5.4.LiteratureGlossary

Visual taskloadPhysical task loadChecking system interferenceCounterproductive adaptationIntroductionResults fiom the expert m eetingFurther resultsConclusions and recornmendationsGenera1 conclusionsConclusions regarding criteria for field testingConclusions regarding th e construction of a checklistConclusions regarding the lab ora toy PC-testThe results o f the validation test by TNO-HFRiThe results of the validation test by CO VComparison of the resultsRecommendationsStep 1 a safety checklistStep 2: a PC estStep 3: criteria for field testingFurther recommendations

44444546464647494949495050505151525252535465

6


7/65

1. Genera1 introductionThis report summarises the final resuits of a three year project commissionedby the Transport Research Centre of the Dutch M inistry of Transport andPublic Works (TR C) to the following research institutes:- TN O Human Factors Research Institute (TNO-HFRI);- Centre for Environmental and T r a c c Psychology (COV);- SWOV Institute for Road Safety Research (main contractor for this

project).The project is aimed at the deveiopment of criteria to assess the effects onroad safety of various applications of Advanced Traffic Telematics (ATTsystems) intended to suppo rt the driver in diffe rent aspects of the drivingtask. Such A TT system s are being deveioped (or are already on the market)e,g . to provide up to da te route information, to maintain a constant speed andheadway, to adapt the m aximum speed t o the local limit or to preventcollisions.Although many of those support systems are intended to make driving easieror safer, they can also interfere with or modi@ the drivers tasks in such away that safety is impaired. This leads to the conclusion that acceptability ofATT systems should be determined by a carefbl consideration of both theintended beneficia1 and the unintended detrimental effects on safety beforeany appiication is given a green light by the government. Preferably, sucha consideration shouid be conducted by way of standardised procedures an dcriteria, but these d o not yet exist. This project has been initiated to provideat least a preliminary set of guidelines and methods to identifi potentiaisafety hazards that single or m ultiple applications of these A T T systems mayproduce.

7


8/65

2. Setup of the projectThe project was initiated to develop an efficient procedure to assess theacceptabiliy (for reasons of safety) of marketab ie in-car telematic devices.Efficient in this respect means, that as much as possible o f the assessmentprocedures can be carried out quickly and cheaply behind the desk and thatmore expensive and tim e consuming field testing will only be used to ',resolve remaining questions or doubts.Ideally, the e nd result of this project would b e a cornprehensive checklistthat can be appiied to al1som of features of ATT applications and that wil1immediateiy render a verdict in terms of safe or unsafe. Furthermore, thischecklist must be formulated in such a way that it can b e applied by(relative) non-experts like policy makers. In this way, the participation ofexpensive experts or researchers could be avoided and the evaluation can bemade very quickly.Acknowledging that such a procedure is, as yet, unfeasible, we haveattempted to develop a procedure in which the work of evaluating safetyeffects can be shared between policy makers and expertshesea rchers in amore or iess efficient way. This procedure then consists of the followingsteps:1. Preliminary screening of a AT T device by a non-expert with the aid of achecklist: this checklist will only rarely result in an absolute verdict butwill mostly generate points of attention.2. Possible extension of the screening by non experts, employing a simplelaboratory test that uses a desktop PC and a limited amount of additional

equipment: this form of testing can be applied to certain points ofattention generated by the checklist. Such a test will, for the time being,only apply to a limited number of types of ATT applications.3. If the previous two steps do not produce a conclusive result because theattention points that remain unresolved seem too severe or numerous, afield test will be indicated to render a fina l verdict. This test, or rather atest series, is set up around well documented testing procedures and useswell defined testing parameters, criteria and assessment protocols.These steps have been worked out in a research programme that wasexecuted in the past three years. The project itself contained the followingp hases:. An inventory of existing knowledge on the subjec t by means of aliterature study.+ Execution of preliminary experiments on well known A TT systems.

The results of these first two phases haven been described in chapter 4 ofthis report.. The formuiation of practica1 criteria and procedures necessary to wellbased experimental safety tests: chapter 5 contains the results of this partof the work.. The first formulation of an evalua tion method suitabie for use outside aiaboratory consisting o f :- a preliminary checklist described in chapter 7;- a limited experimental testing method suitable for a desktop PC asdescribed in chapter 6.

+ An evaluation of the PC test u sing the results of the preliminaryexperiments; the results of this part can also be found in chapter 6 .

8


9/65

Finally, in chapter 8 the conclusions per phase are summarised as weli as hegeneral conclusions and recommendations.This programme was not the only research concem ing safety effects of A T Tunder way dwing this three year period: in parallel other, more specificprojects have been carried out by some of these same research institutes.The results of one of these projects, the N I S project primarily aimed atRDS-TMC, have also contributed to this project, specifically to the criteriafor testing, the checklist and the PC est.

9


10/65

3. Results of the literature study and experiments

3.1. The literature study by TNO-HFRI and COVThe discussion in this paragraph is based on the report Safes, efects of in-vehicle information systems of Verwey, Brookhuis & Janssen (1 996).

3.1.1. IntroductionA growing number of traffic safety studies shows that hum an error is amajor con tributing factor in traffic accidents. For instance, Trea t et al.(1977) showed that difficulties with perception, attention, distraction, etc.are important causes in 30 to 50 percent of traffic accidents.Countermeasures have to be devised and introduced to prevent behaviourcontributing to these accidents, without eliciting undesirable side-effects.Modern electronic systems in- and outside the car, also indicated byteiematics appiications, could be such countermeasures but their usage m aybe accompanied by side-effects such as distraction, overioad, insufficientattention for the driving task, and counterproductive adaptation asaconsequence o f (misieading) feelings of safety evoked by these measures.So, an important question is how can one decide w hether a specificteiematics appiication affects tr a fi c safety.A literature study was performed to delineate how a standardised testmethodology for assessing safety effects of in-vehicle information systems,also indicated by telematics appiications, should be designed. Since it isvirtually impossible to cover al1 safety aspects that te iematics appiicationscould have in the broadest sense, a limitation to the effect of applications onthe driving task in a narrow sense has been chosen in this report. An ana lysishas been carried out of how the use of in-vehicle telematics applicationscould affect driving performance of individual drivers,A major reason to develop driver support systems is the reduction of trafficaccidents. One of the probiems with these systems is that it is very difficuit,if not impossible , to forecast the savings of death and disability that mightresult from the introduction of such systems. Although there is an urgentneed to know what the effects are of introducing a specific system before itenters the market, no data exist on w hich estimates of the risks caused bythese systems can be based. However, the effect of individual systemscanbe studied on a low level, i.e. safety aspects of the driving task per se.Hence, each individual teiematics application should be cubjected to a testfor behaviourai safety effects before marketing in order to pinpointunwanted side-effects at the behavioural level. However, in order todetermine w hether a system has unacceptable side-effects, criteria must bedeveloped for what exactiy constitutes unacceptabie.Even though the need for a general, preferably standardised methodologyfor assessing safety effects of telematics applications has been generallyrecognised, there have only been a few indecisive attempts to come to amethodology at a level on which empirica1 studies can be easiiy based.Knowledge about how actual safety effects should be assessed is onlybeginning to emerge from a handful of DRWE iI projects. On e of the10


11/65

reasons is that a lot of empirica1 research has been devoted to thecomparison of different versions of a group o f telematics applications inorder to improve the hum an-machine interface rather than to absolute safetyeffects. However, a t some point in time it still has to be ciarified whether aspecific telematics application should be allowed on the market or not andthe project reported here was specifically aimed to provide instruments forsuch a clarification. To that end, the study focused on the safety effects ofthree mechanism s: workload and distraction, reduced attention, and counter-productive behavioural adaptation which wil1 now be treated briefly.

3.1.2. OverloadIn driving, overload refers to the situation that the driver is unabie to processal1 relevant information for executing the driving task. This results inincreased error rates and later detection of other traffic participants and ,hence, to reduced traffic safety (e.g., Rumar, 1990). Th e role of overload inaccident causation is supported by studies showing that high levels ofworkload on specific routes were associated with the probability of tr af icaccidents on those routes (MacDonald, 1979; Taylor, 1964) and the findingthat drivers with less information processing capacity are more likely tohave accidents (Lim & Dewar, 1988).Whereas overload may occur while no telematics application is being used,the introduction of such app lications makes overload more likely becauseinteraction with these app lications involves a task additionai to normaldriving.A common model of the driving task assumes that driving tasks can becategorised int0 tasks at three levels (Allen, Lunenfeld & Alexander, 1971;King & Lunenfeld, 1971;Michon, 1985). The first level, the control level, isconcemed with elementary vehicle handling nctions iike lane-keeping andhandling of the controls which allows one to follow the road and to keep thevehicle on the road. Time constants at this level are usually below onesecond and the tasks at this level usually cause little mental workload.The manoeuvring level deals with reactions to events in the trafficenvironment. These reactions have to d o with interactions with o ther trafficlike overtaking, intersection negotiation and th e like. Time constants arenormally between one an d ten seconds and m ental workload is usuallyhigher than tha t associated with the control level. Th e strategie level regardschoice of transport modality, route planning, and route following. That is,how drivers choose their destination, the route and the m odality of travel.Time constants related to the processes at the strategic level are typicallymore than ten seconds and workload is generally high.In general, mental overioad may be caused by tasks at the manoeuvring andstrategic level, especially for tasks with low time constants. Visual workloadmay be high for control tasks too. So, if one considers the occurrence ofoverload caused by telematics applications, the type of task tha t is likeiy tobe carried ou t and the type of workload it incurs may be important toconsider before workload is actually measured.The major methodologies for assessing driver workload are summ arisedwith an emphasis on the techniques sensitive to various types ofworkload atthe Same time.

11


12/65

3.1.2.1. The secondary task techniqueThe secondary task technque involves the addition o f a low priority task tothe primary - here, driving - task. It is assumed that an increase in workloadassociated with the primary task is indicated by performance reduction ofthe secondary task. Depending on the type of secondary task, the load onseparate resources can be assessed. Applying a task which has manycomm on characteristics with driving wou ld provide a genera1 indication fordriver workload. With the utilisation of this technique, Venvey (1 99 3q b)could show in which driving situations overload ismore probable. However,the secondary task method see ms less suited for performing safetyassessments of telematics applications because it is likely to affect drivingperformance in any case (e.g., Noy, 1987). This m akes it hard to directlyassess safety effects fiom driving performance.

3.1.2.2. Physiologica l measuresA number ofphysiologcal measures is also used for assessing workload.Some me asures a re primarily sensitive to load on specific processingresources w hereas others index overall mental workload. The m ost commonones in driving research are various measures based on heart rate. Lesscommon nowadays are responses in skin conductance (SCR) and eye blinkfrequency.The advantage of physiological measures is that they affect the driving taskonly in minor ways and, once the electrodes are a ttached, are fairlyunobtrusive. Their disadvantage is that they usually have a limited temporalresolution (Verwey & Veltman, 1995). Therefore, these measures appe arprimarily useful when overload occurs for periods of at least a few minutes.Another disadvantage is that analysis of physiological d ata is fairlylabourious because these data are often contaminated by other physiologicalsignals and noise. One promising index of workload for assessing effects oftelematics applications on driver workload is eye movement registration.Eye movements can be registered by way of electrodes located round theeye, which is why eye movements are often categorised under physiologicalmeasures, but the more common procedure nowadays is to use externalregistration by way of on e or more video cameras.A number of studies have measured eye movements in terms of giancefiequency (the number of glances toward a display) and glance duration (thetime the driver looks at a display). Wierwille (1993) found that glanceduration towards an in-car display will generally not exceed about 1.5 s .If more time is required glance frequency increases. Glance duration wouldbe associated with the time required for chunking parts of the display (e.g.reading words) while glance frequency would indicate overall complexity ofthe display. Zwahlen et al. (1 988) state that more than three glance times of1 s each are unacceptable. The se studies give a general idea of what isacceptable in terms of giance times and frequencies.However, it remains hard to conclude any thing about safety if it is not clearwhen, in w hich situations, drivers will actually look at the display. Forexample, glances of only half a second may be fatal in complex drivingsituations whereas much longer glance times are harmless on straight roads.Therefore, safety evaluations should also consider when telematicsapplications increase workload and whether the increase will actually affectsafety. For example, future w ork might focus on the possibility to comparerecorded glances times toward in-vehicle displays with glance times that aretheoretically allowed (given current headway, lane position etc.).12


13/65

A m ethod fiequ ently applied to assess driver workload is simply asking theparticipants in a study how ioaded they are (subjective measures).Reiativelysimple rating lisis have initially been w ed which involved only a singlerating, that is a one-dimensional list. in the last decade the NASA-TLX(Hart & Staveland, 19 88) and the SW AT (e.g., Reid & Nygren, 1988) havebeen used m ore frequently. These techniques involve several subscales andwould give more accurate indications for workload than univariate lists.A disadvantage of subjective ratings is that they invoive an additionai taskand, as such, are iess reiiabie when workioad exceeds working mem orylimitations (Yeh & Wickens, 1988).This pieads for the use of more simpieone-dimensional iists. In fact, recent studies found that one-dimensional listsgive as reliable results as he more complex SWAT and TLX (Hendy,Hamilton & Landry, 1993; Verwey & Veltman, 1995). Care should be takenthat it is always ciear to the participants to what periods th e estimates referand whether the estimates shouid refer to peaks or to averages (Verwey &Veltman, 1995). Ano ther disadvantage of subjective iists is that subjectivescores may be affected by opinions on system characteristics. With respectto safety evaiuations it should be noted that subjective rating iists areprobably sensitive to workioad in generai rather than to specific aspects(resources) of workioad (Verwey & Veltman, 1995)

3.1.2.3. Performance on some drving subtasksPe$ormance on ome driving subrasks can be regarded as indicators fordriver workload because these hav e no d irect implications for safety (e.g.,steering wheei frequency) w hereas others are more d irectiy associated withsafety (e.g., headw ay). As drivers are usuaily able to distinguish the safetyof various subtasks, they wil1 give safety related tasks a higher priority attimes of elevated workload. Hence, high priority aspects of driving are lesslikeiy to be affected by increased levels of workload than iower prioritytasks. Consequentiy, the effect o f interacting with a teiematics appiicationwil1 affect low priority subtasks earlier and more often than high prioritysubtasks.For exampie, Verwey (1 991 a) found that giances at the interior mirrorreduced with increased task ioad. Obviousiy, this has no direct consequencesfor safety but it might still affect safety in that the driver is less aware ofwhat is going on around the car. Th is leads to the inference that reducedperformance of high priority parts of the driving task are ciear indicationsthat safety is affected but this wiii occur rather infiequently, whereasdeterioration of iow priority task performance is more likeiy and, hence, areasonabie indicator for driver overioad, but a worse indicator for reducedlevels of safety.W ith respect to evaiuating safety effects of telematics appiications, Zaidei( i 991) proposed to assess the quaiity of driving in a real driving study inwhich an expert observer, most likely a driving instructor, gives detaiiedjudgem ents on the quaiity of driving with and without the teiematicsapp lication.Such a method is presented in De Gier (1 980). However, objections againstthis method have been raised. The major p robiem is that subjective ratingsare sensitive to the individual raters opinions and to whether or no t the rateris abie to register aii relevant information.As there are many ways to assess (driver) workload, it is not always obviouswhich ones shou id be used in a specific experiment. There appear to be afew criteria. First, the techn ique shouid be sensitive to reiatively short and

13


14/65

acute changes in workioad. Second, the measures should not affect drivingbehaviour as, n fact, aspects of this behaviour should say also somethingabout safety effects. Third, the measure shouid be sensitive to generaiworkioad and show overioad when oniy one processing resource isoverioaded. Fourth, the method shouid preferably not be too labourious.Given the large variety of techniques available for assessing workload anddriving performance, it is not at ali obvious how actual safety effects ofteiematics applications due to information overioad can be assessed.The reiation between driving performance, workload and safety is a com plexone because im portant parts of driving wil1 usually not suffer and are,therefore, not sensitive indicators for reduced safety. S till, in practice theymay sometimes be affected and cause accidents anyway. If it can be shownthat driver workioad does not increase at aliwhen one is using a teiematicsapplication o ne may infer that safety wil1 not be affected by driver overioad.But if there is an increase in driver workioad it is not clear what level ofincrease can be considered acceptable.The view that workioad may not increase at al1 is too simple because manytelematics applications may increase workload in some situations and reduceworkload in other situations. Aiso, a minor increase needs not affect tr a fi csafety, especialiy when the telematics application does not presentinformation that is distracting and if message presentation happens insituations in which the driver is not heavily loaded. In order to assess safetyeffects, one m ight analyse to what extent anticipatory behaviour (includingiooking out) is affected. Before workload is assessed, it is important todetermine in which driving situations and during what type o f tasks,interaction with telematics applications is iikely to occur so that one canhypothesise which type of overioad can be expected at which mom ents intime. The re is a need to cross-vaiidate popular objec tive workioad anddriving performance measures with safety assessmen ts of experts in order todetermine which objective measures are most sensitive to safety effects andwhat their absolute threshoid va he sa re fiom a safety point of view. To findout how reiiable subjective rater opinions are, inter-rater reliability (ie., theconsensus of various raters) shouid be assessed too.

3 .1 .3 . UnderloadDriver underload is defined a s indicating the situation that the driver getsint0 a state of limited attention to driving, due to either d iverted attention(e.g., no specific driving task dem ands) or deactivation (e.g., the driverdozes off). Driver state is not some unitary, fixed phenomenon, not evenwithin an individuai. It varies with time-of-day, age, subjective feelings(mood), but also with time-on-task and al1 kinds of externai influences, suchas tra fi c environment and situational task-ioad, alcohol and (medicinai)drugs.Certain telematics appiications are developed to support the driver, forinstance, by taking over parts of the driving task whereafter iess attention toaspects of the driving t a k is strictly needed. O ne approach in the field ofsafety assescment is locating the optimum with respect to drivingperformance, and then detecting signs of deviations from the optimum (cf.Wiener et al., 1984), anaiogous to the concept of arousai, as expressed withthe inverted-U-hypothesis. This hypothesis states that there is a level ofarousal, or activation, that yields the highest level of performance. Idealiy,the driver acts on top of this performance curve (cf. Wiener, 1987),gainingmaximum results. Deviations from the top (Wieners optimum) could be14


15/65

caused by a decrease in arousal according to the inverted-U-hypothesis ordiminished workload, or task demand, according to W en er s analogue.The concomitant phenomena can be registered better from the individuaisphysiology, being the precursor of behavioural decremenis, than fiom therelevant behavioural parameters themseives.Changes in driver state are reflected in changes in relevant physiologicalparameters such as electroencephalogram (EEG), electrocardiogram (ECG),galvanic skin response (GSR), Electro myogram (EM G) (see Brookhuis &De Waard, 1993). In order to estimate the contribution of the changes indriver state to accident causation, one would want to be able to continuouslymeasure these physiological signals in real life driving. However, it is highiyunlikeiy that measuring physiological parameters while driving a m otor-vehicle wil1 be acceptable ever. Therefore, measuring physiologicalparameters of underload can only be used for assessing safety effects ofvarious telematics applications in experimentai conditions.After prolonged driving, a drivers activation tends to diminish rapidly, suchas can be measured by m eans of spectra1 analysis of the E EG . Alcohol andmedicinal drugs mostly aggravate these effects (Brookhuis et al., 1986)although this is not necessarily the case, depending on dose and type ofdrug. Petit et al. (1990) dem onstrated a relationship between the handling ofthe steering wheel and the occurrence of alpha waves in the EEG. T he stateof vigilance, as indicated by the power in the alpha range, was found tocorrelate highly with specially deveioped steering wheel functions in mostcases.De Waard & Brookhuis (1991) als0 found effects on steering wheelbehaviour with time-on-task (1 50 minutes of c ontinuous driving).The standard deviation of the steering wheel movements increased and thenumber of steering wheel reversals per minute decreased, both highlysignificant. Subjects activation, measured by a relative energy parameter[(alpha+theta)/beta], gradually diminished fiom the start of the experiment.With respect to the assessment of safety effects of telematics applications, itis important that underioad may wel1 be caused by the presence of telematicsapplications that take over (part of) the driving task. Until telematicsapplications are capable of reducing the driving task fully to a meresupervisory task, introduction of teiematics applications taking over part ofthe driving task must be considered very carefully and maybe evenreluctantly.In contrast to the matter of overload, where the relationship betweenobjective measures of driving performance and safety isyet largeiyunresolved, the D RIVE I V 1004project (DREAM ) cross-validated physio-logica1 and performance data (Thomas et al., 1989). Because safety effectsof underload are fairly ciearly related to performance data there is less needfor a further cross-validation with subjective opinions on safety. Safetyeffecis of underload are largeiy confined to reduced lane keepingperformance and increased reaction times to events in the traffic environ-ment and as such fairly easy to measure.On the basis of a literature study and experimental research D e Waard &Brookhuis (1 991) concluded that the com bined measurement o f the driversphysiology and behaviour shouid allow the deveiopment o f a m onitoring

15


16/65

I

device on the basis of unobtrusive ehicle parameters alone. The driveractions, as measured fiom vehicle pam ete rs, that seemed most promisingtime-to-line crossing, reactions to lead vehicles, as measured by reactiontime an d time-to-collision, and p erhaps speed management. Furthermore,subjective measures of, for instance, psychological impairment, effort andacceptance should be taken int0 account.

were aspects of steering wheel movements,; variability in lateral position,Similar to the problem that high w rkload and driving performancemeasures have not been linked to s fety measures directly yet, there is noAlthough for some driving parameters a f in t attempt has been made toestablish safety criteria, a complete picture of the relationship is still faraway. In fact, studies in various DRIVE projects have enabled US to relate afew driving parameters to physiological measures and based on that alimited series of absolute and relative criteria have been proposed. Thesecriteria facilitate the evaluation of underload effects introduced by telematicsapplications. However, based on ddiving performance alone it is hard topredict when and where things go wrong in the absolute sense.

clear relationship between safety an.d measures of underioad in al1 respects.

3.1.4. Counterproductive behaviour adaptationCounterproductive behavioural adaptation is the phenomenon that driversstart behaving in riskier ways exactly because they are supported by asafety-raising device. It may thus m anifest itself, first of all, in thebehavioural param eters that are typically used in driver behaviour stud ies. Ina wider sense counterproductive a d p a t i o n may also be taken to comprisedrops in attention, or a reduced level of genera1 alertness that is inducedbecause the device acts as a guardian angel. In a still more genera1 sensecounterproductive adaptation may qccur at the strategic level of the drivingtask, that is, with respect to conditions under which trips are undertaken atal1 or with respect to changes in overall mobility (mileage).The effect of counterproductive adsiptation, when it occurs, is to make thenet safety benefit less than wouid have been expected on the basis of theeffect of the dev ice by itself. As a ryle of thumb counterproductiveadaptation can be expected to occui in the fastest possible way for devicesthat are clea rly intended to raise safety, that are conspicuousiy presentwithin the vehic le, and that act relatively frequently, i.e., under conditionsthat a re not necessarily highiy criticai. It wil1 take longer for drivers tocompensate, if at ai] , for safety features that are not immediately makingtheir presence clear, since the feedback loop here is more indirect anddiffuse.A probiem that is typicai for countemroductive adaptation effects is how toassess what the net effect of a (safety-raising) device is, that is, how acounterproductive adaptation effect should be subtracted fiom the devicesoriginal-beneficial-effect.The latter, the so-caiied engineering estimate ofa devices expected safety effect, is the accident reduction that would beachieved if 1O0 percent of the relevbnt population had t he device and if thatpopulation showed no behavioural adaptation to the new situation (Janssen& Van der Horst, 1992).The basic notion in making an enginee ring estimate is that expected safetybenefits are given as an extrapoiation or an implication of a rather16


17/65

straightforward engineering calculation. By doing so, physical changes tothe system are considered without initially addressing possible induced userchange. For example, if design changes to some roadside device would becalculated by engineering methods to reduce the probability of a driver deathon impa ct by 1O percent, then the engineering estimate is that a 1O percentreduction in driver dea ths Gom collisions with the mod ified device wil1occur, The m ost common way to obtain an engineering estimate is indeedfrom accident data. A device that, a ccording to accident statistics, causes xpercent of deaths is expected to yield a safety return of x percent upon itsremoval. Alternatively, if the absence o f a device would lead to y percent ofal1deaths, then the im plementation of that device would be expec ted toreduce deaths by those same y percent.In other cases an engineering estimate can be made on the basis oflaboratory results, e.g., for hardware devices tested unde r crash conditionsrepresentative of those occu rring in reality. In still other cases theengineering estimate can be no more than the expectation of a beneficia1safety effect, or an order of magnitude thereof. T he estimation of (ne t) safetyeffects can never be better than the enginee ring estimate permits. Tha t is,each and every safety measure nee ds an estimate of its effect per se when itsimplementation is being considered, and against which the effect that isuitimately realised must be evaluated. It should be considered to be theobligation of those proposing a devices implementation to support thesafety claim that is being made by a hard engineering estimate originatingfiom either accident or laboratory studies. In case th e device is not directlysafety-directed, but aims to support other driver functions (like navigating),the engineering estimate is naturally close to O.Another factor that should be taken int0 account when assessing the netsafety benefit of any device is the devices use rate within the population.For safety-directed measures which for their effectiveness rely on theacceptance of the popuiation there i s the complicating and com plex issue ofselective recruitment, meaning that the use rate per se as well as the effectthat is achieved are affected by self-selective processes in the population.The hypothesis is that those who opt for some device differ from those whodo not in respects that are essential to its effectiveness. Usefui quantitativeexpressions describing the implications of self-selective processes fordriving behaviour as well as for resultant accident involvement rates havebeen derived by Evans ( 1 987a,b).It should be noted that counterproductive adaptation may a h o manifest itselfin ways that are not restricted to driving behaviour per se . There are severalways in which this may occur. T he availability of a device that reduces riskper kilometre driven may thus lead to any or al1of the following:1. The participation in traffic of segments of the general population that didformerly not dare to d o so because they considered the risks on the roadtoo high.2. An increase in mileage driven (overall Vehicle Miles Travelled), bothbecause of:- a shift in modal split (from other modes of transport to the automobile);- a d irect increase per vehicle.3. An increase in mileage driven under more unfavourable conditionsandor under decreased levels of personal fitness.

17


18/65

Al1 these effects wil1 have as a conse quence that the exposure to risk wil1increase, which by itself wil1 generate more acc idents even when the risk perkilometre has been reduce d.It is presentiy not possible, given the sta te of rafic safety science, toanticipate upon the occ urrence of these effects in a quantitative form. It is,nevertheless, unwise not to consider the possibility that this type ofconsequences couid occur. The rule of thumb must be that, if the averageautomobile driver can imagine a way of getting a mobility profit out of asafety-raising device, the average experimenter should surely be capable ofdoing so and must adm it that this can be a c onsequence.

3.2. Experimental results by TNO-TMThe discussion in this paragraph is based on the foilowing reports:- Venvey, W. (1 996). Evaluating safety ef ec ts o f in-vehicle informationsystems; A detailed researchproposal.- Venvey, W. (1 996). Evaluating safety e fec ts o f in-vehicle informationsystems; Testing the me thod.- Venvey, W. (1996). Evaluating safety ef ec ts ofin-veh cle informationsystems (IVIS); A jeld experiment wth tra flc c ongestion informationsystems (RDS-TMC) and preliminary guideiines or IVIS.

3.2.1. IntroductionIn recent years, there has been a considerabie boost of research anddevelopment of modem technology in road transport. From the early starton, many people have expressed their concern that this technology, knownas in-vehicle information systems (WIS), advanced transport telematics(AT T) or intelligent transport systems (ITS), might jeopardise traffic safetyrather than that it would improve safety as is claimed by others (e.g.,Hancock & Parasuraman, 1992; Parkes & Ross, 19 9 ) .One reason for this concern is the distraction and overload the driver may beconfronted with. An ordinary driver may weli be able to perform additionaltasks whiie driving on a quiet motonvay. However, in dense city traffic andwhile negotiating complex intersections behaviour may become unsafe.With attention attracting information in demanding driving situations, thereis the danger that drivers are not able to ignore the message entirely. Whenmessages are not conspicuous, drivers may choose to attend to theinformation because they think they can handle it, which is not necessarilythe case. Finally, even in quie t driving situations, drivers may take risks bygiving too much attention to the WIS.Many studies have shown effects of N I S on driving performance. Theseeffects suggest that safety is affected as well. However, given the difficultiesto assess safety in experimental situations in a reliabie way, there is stilllimited proof for negative safety effects. There is an urgent need to deveiopguidelines and standards for the design of the m an-machine interface ofWIS based on safety research.The experiment aim ed at testing safety effects of three major types of driver-system interaction with a specific WIS, that is a system giving on-line trafficinformation, and relating the characteristics of human-machine interface tothese safety effects. The results of the study and guidelines from theliterature are used to propose preliminary guidelines for the driver-vehicleinteraction with N I S systerns and to propose a fiam ework for the18


19/65

development of guidelines aimed a t preventing negative safety effects of theworkload and distraction caused by use of WIS.3.2.2. Method

Twelve experienced drivers drove a route through the city of Amersfoort. Ata set of predetermined driving situations, which included right turns,intersections and straight driving, congestion information was presented bymeans of either a m ap display or a speech message, or subjects were .instructed to program a filter on an RD S-TMC system. This filter determineswhich particular subset of al1 available RDS-TMC nform ation wil1 bepassed to the user (e.g. the information for a particular road). Th e taskinvolves quite elaborate manipulation of buttons on the receiver and readinga display.Driving performance and looking behaviour in these situations wereanalysed in terms of their safety ramifications by comparison with absolutesafety limits obtained fiom the literature (Verwey, 1996), safety judgm entsby an experienced driving instructor, and comparison with a controlcondition.

3.2.3 . Major resultsTh e driving instructor opinions show ed that, acro ss al1 situations and typesof judgm ents, driving safety was significantly impaired by each of the threeW IS tasks. In the filter programming task these effec ts were in large partcaused by poor course keeping and br akin dde cel erathg (mainly looking inadvance). M ore detailed analyses showed that the safety reductionsconcerned looking, course keeping, and braking (mainly anticipa tion) whenturning right, course keeping when approaching general rule intersections,braking (mainly anticipation) when approaching priority intersections, andcourse keeping when driving straight. Steering wheel frequency increased atthe straight urban sections with filter programm ing but not with map an dspeech.Performing the N I S tasks in this study did not affect the ratings with respectto adapt ing speed to other traffic (straight driving), distance to headingtraffic (straight ahead), anticipation in generai (straight ahead, general ruleand yield inte rsections), braking and dece leration (straight ahead), givingpriority (general rule and yield intersections), watching priority traffic(general rule and yield intersections), and cou rse keeping (yield inter-sections). Neither were any effects found on the objective measures lookingbehaviour as scored from video (right turns,general rule intersections), thehypothetical occurrences of high decelerations (right turns, yield inter-sections), exceeding the critical minimum T TI during each approach of anintersection, steering frequency (map and speech conditions), the proportionhigh dece lerations (al1 situations), and the standard deviation of speed andTLC (straight driving).

3.3. Experim ental results by COVTh e discussion in this paragraph is based on th e report:- Brookhuis, K.A., Waard, D. de (1996).Limiting speed through

telematics; Towards an Intelligent Speed Adap tor (ISA).

19


20/65

3 . 3 . 1 . IntroduetionThrough recent legislation in the Netherlands, the m aximum driving speed isrestricted by a speed lim iter in the heavier types of lorries and coaches.The effect of these devices on fbel consumption, noise, air pollution,wearing of the tires and trafc safety is expected to be mainly positive (e.g.,Almqvist, 1991;Van der Mede, 1992; Wilbers, 1992).The obvious restriction of the speed limiter as mandatory now, is that it oniyprevents driving above the maxim um allowed driving speed of heavy goodshaulage vehicles, and is independent of local limit in a specific roadenvironment.An intelligent speed adapter (ISA ) takes int0 account local restrictions, andadjusts the maxim um driving speed to the posted maximum speed. When itcomes to restriction of driving speed ofprivat e vehicles, the use o fintelligent speed limiters is to be preferred due to further differentiation ofspeed limits for private cars as compared to heavy goods vehicles.A non-intelligent speed limiter is set at the maxim um allowed driving speedfor m otonvays (1 20M),hile the m ajority of a speed limiting systemssafety benefits can be attained on A-class roads (limit 80 km/h) and inbuilt-up area s (50 km/h).In general, a standard speed limiter is an inrusive system that restricts speedcontrol, i.e. the device sets the maximum possible driving speed.An intelligent speed limiter is able to set this maximum speed in accordancewith local posted legal limits. A less intrusive device is a system thatprovides the driver with feedback about local limits, for instance, on the gaspedal. An active gas pedal increases the counterforce if the driver is drivingtoo f a t (Godthelp & Schum ann, 1991). In principle, such a speed limiterleaves the driver in control, while the feedback provided in case of a speedviolation is highly compelling. Moreover, the feedback is provided in thetactile modality, i.e. the sam e m odality through which action has to beundertaken to observe the rules again.Feedback could also be presented in the visual modality, e.g. a warning lightor message in the dashboard, or aud itory, an acoustic signa1 or vocalmessage. On the one hand this type of feedbac k seems less intrusive than thefeedback an active gas pedal provides because these warnings can easily beignored. On the other hand , it might be that the social effect of being warnedin the presence of other passengers is a more severe chastisement andtherefore less preferred. A nyway, acceptance of the feedback type systemscan be expected to be higher than of a strict, standard speed limiter, becausebehaviour is less restrained.Observation of behaviour at the ievel of driver reactions to these systems isof primary importance. However, apart from individual reactions, interactionwith other traf fc that is not equipped with speed limiters is also important.In such a mixed tra ffc situation, cars with a speed limiter could easilyannoy drivers of cars that are not restricted and vice versa (Almqvist et al.,1991,Persson et al., 1993). These type of interactions deserve at least someattention in behavioural studies as well.

3.3.2. Development of aprototype intelligent speed a dapto rAn effort is now undertaken in the Netherlands to dev elop a prototypeintelligent speed adaptor tha t leaves the driver in control. For a start, this20


21/65

resuited in the deveiopment of a continuous feedback display in closeproximity of the speedometer indica ting the curren t speed limit, quite similarto the CAROSI system (Nilsson and Beriin, 1992). Centra1 part of theCAR OSI (CAr ROadside SIgnaiiing) system is the instrument panel, whichincludes not only standard displays such as he speedometer, but alsoconains sections on which roadside information is displayed. Amongstthese is the posted speed limit, which is dispiayed below the speedometer.Major adva ntage of giving feedback by displaying the speed limit inside thecar is that this information remains continuousiy visible instead of onlybeing visibie at the mom ent a s i p is passed. This might reduce speedingbecause o f genera1 unawareness of the limit, which is not uncommon in theNetherlands (e.g., Steyvers et al., 1992; De W aard et al., 1995).A special version o f the latter type o f feedback display is deveioped forimplementation in the C OV experimental test-vehicle. Whenever th e speedlimit is exceeded the colour in which the speed limit is displayed changesfrom green (normalheutral) to amber, or yeiiow, (waming). In case thespeed limit is exceeded by 1O%, the coiour changes fiom amber to red(violation), and then an additional, auditory warning message is issued (seealso De W aard et al., 1994; De Waard & Brookhuis, 1995a,b). The systemsare integrated in the existing DETE R system (see De W aard & Brookhuis,1995a), which is developed as an open system to integrate driver monitoringand feedback (sub)systems. In the present experiment this set-up is tested,letting subjects drive the CO V vehicie with and without the feedbacksystems.Additionally, an active gas peda is tested as a medium for haptic feedbackin case of speed limit violations, exceeding by 1O%, in the COV drivingsimulator with the same subjects, in a cross-over design. Al1 modes of feed-back are studied to effects on behaviour, m ental workload and acceptance.

3.3.3. MethodTwenty-four subjec ts were paid for their participation in the test on effectsof feedback concerning speed restrictions and violations in the institutesinstrumented test vehicle and driv ing simulator.Half of the subjects drove an instrumented test vehicle over a fixed route andthen performed a simulator test, half of the subjects vice versa. Each of thetest-rides consisted of two parts, first the baseline measurement, then after ashort break, either the test ride with feedback or the control ride. Half of thesubjec ts received feedback, half w ere in the controi condition.The test rides in the instrumented test vehicle were in normai traffic, undervarious conditions. Subjects were guided by sampled vocal route guidancemessages that w ere triggered by the investigator for reasons of propertiming. They were led over a varied route that inciuded sections of motor-ways, A-roads and built-up areas, with speed restrictions of 50,70, 0, 1O0and 120 kmh.After each of the (four) test rides, subjects were requested to completequestionnaires concerning perceived workioad and subjective drivingquaiity. At the end of the who ie test, subjects compieted a generaiquestionnaire again, asking for their ideas with respect to speed restrictingsystems again.

21


22/65

3.3.4. Results and conclusionsAlthough the num ber of detected violations during the second (feedback)series of trials is lower in the experimental group, this effect did not attainstatistical significance. The difference in the number of speed violationsbetween the Wo est facilities was significant, drivers more frequentlyviolated the speed Iimit on-the-road. The exten t to which the speed limit wasexceeded was higher in the simulator. During the feedback-trial the extent towhich the limit was exceeded was on average lower (with the exception ofthe second on-the-road trail for the control group).A new parameter, the proportion of time violating the limit, does notdiscretely sum up the number of times the limit is exceeded, but reflects thetime the driver is not complying. Two parameters were determined, theproportion of time driving abo ve the limit, i.e. the time the display was orwould have been ambe r or red, and the proportion time driving above thelimit + 1O%, i.e. the time the display was or would have been red and anauditory message was or would have been icsued. The would have been-condition is for the control group, the experimental group actually receivedthe described feedback. As much as 20 to 25% drivers were speeding in thestrict juridica i sense. Between 5 and 10%of the time they are driving fasterthan the speed limit plus a 10% margin. The effect of the feedback system ishighly significant for the latter parameter.From the acceptance data it followed that acceptance very much dependsupon feedback system, continuous feedback (on display) was accepted bestof al1 means of feedback by far. The ratings for the continuous visualfeedback were unusually high and can maybe even considered as (highly)appreciated.An new, unexpected effect of the compound feedback was found, asignificant reduction in speed variation. One of the reasons is the earliermentioned use of the amber to stay in the m argin of limit to limit+lO%.The implication of this finding is that less variation in driving speed couldhelp to harmonise traffic, which is one of the candidate toois to reduce thenumber of accidents (see aiso Brookhuis & Brown, 1992).N o effects on workload were found in this study, again contrary to the firsttw o experiments as mentioned. However, in the iatter studies the (slight)effects were marginally significant, while in the present data the (slight)effects demonstrated in either of the two measures of mentai load did notattain significance. The implication of these findings, in line with Venvey,Brookhuis & Janssen ( 1 996), is that before implementing teiematicssystems, in principle workload effects shouid be measured, just to be sure ,but the type of systems tested so far are not implying alarming effects.

3.4. Results an d conclusions with respect to guidelines a nd c riteriaThe discussion in this paragraph is based on the following report:- Verwey, (1 996a-d). Evaluating safety effects of in-vehicle informationsystems (IVIS); A field experiment with trafic congestion informationsystems (RDS-TMC) and preiiminary guidelines fo r IVIS.Venveij ( 1 996c,d) proposed several guidelines for IVIS. Guidelines andstandards for IVIS are still under development. The currentiy existing and22


23/65

3.4.1

directly relevant stand ards for in-vehicle system design come largely fromthe office and manufacturing environment, such as human computerinteraction but efforts are mom entarily undertaken to develop guidelines andstandards for IVIS within the I S 0 fiamework (TC/SC13/WG8/9).Guidelines can be categorised as procedural guidelines, product guidelinesand performance guidelines (Shenvood-Jones, 1990; Parkes, 1995).Procedural guidelines indicate how usability and safety should be assessed,performance guideiines specify acceptable user performance levels(maximum looking time) and product guidelines speciy the physical aspectsof the system (e.g. eye-display distance). Basically, a manufacturer needsproduct guidelines. However, Parkes (1995) argues that standards andguidelines for N I S should be expressed as performance standards.This makes the guidelines product independent and takes the interactionbetween various IVIS into account. This implies that standardised methodsare to be developed which allows the translation of performance standardsint0 product guidelines to which manufacturers can adhere.Because safety should be the ultimate criterion for WIS interface design,guidelines for N I S should be tested against safety criteria. Furthermore,effect of IVIS on the traffic flow should be taken int0 account. So,guidelines for N I S should involve procedural guidelines indicating howsafety effects can be assessed, performance gu idelines, indicating whichperformance levels indicate unsafe situations, and for certain types of WIS,product guidelines indicating which system characteristics are likely toinvoke unsafe behaviour. Th is procedure can be carried out for traffic floweffects as well.in the following section, a fiamework is presented which allows determiningwhich types o f guidelines are required for preventing effects of WIS ontr af fc safety. This leads to a limited set of procedural and performanceguidelines and criteria. Also ergonomically oriented product guidelines wil1be discussed. Th e actual product guidelines themselves are presented inVerweij (1 996).It should be emphasised that the guidelines presented here are preliminary.They are derived from the experimental results and from existing guidelinesand s tandard s which have partly been tested with respect to traffic safetyeffects. If the text refers to the present experim ent the experimentsdiscussed in paragraph 4.2 is implied. The guidelines hold for overload anddistraction situations. That is, the possibility that the WIS may also havepositive safety effects, such as with anti-collision systems and detection ofdriver drowsiness, is not considered.

A JTamework or developing safety related guidelines and criteriaTh e development of guidelines for preventing negative safety effectsrequires a theoretica1 analysis of the driving task and the WIS task.The distinction between control and maneuver tasks can be extended intopart tasks for which individual guidelines should be proposed. Then therelevant control tasks are course keeping and speed con trol. Relevantmaneuver tasks are car following, intersection negotiation, and obstacledetection. The control task speed control refers to the task of keeping thevehicle on the road . This is especially relevant with respect to curves. Speed

23


24/65

control in interaction with other vehicles are considered maneuver tasks(e.g., car following).For quantitative criteria, values may diffe r for various types of road. Forexample, criteria for glance times based on course keeping criteria maydiffer for freeways and residentiai areas because of the differences in lanewidth and speed. Furthermore, the consequences of not obeying thesecriteria wil1 differ for differen t roads. Safety effects will be much greaterwhen obstacle detection is affected in residential areas than at freewayswhere obstacles are rare.Finally, criteria will have to be developed for different types of human-machine interaction. With respect to WIS, these criteria should indicate thedegree of glance time for visually demanding displays, mental workload forcomplex messages, manual complex ity in system control tasks, and theexten t that interaction with the system is paced by the driver or the system.Future work might also take effects of haptic and kinesthetic informationint0 account.

3.4.2. Safety assessment and performance guidelinesThis section proposes preiiminary performance guidelines which are mostcruciai for safety evaluation. Later sections will present proceduralguidelines wh ich wil1 have to assure that the app ropriate m ethods are beingused. Finally, product guidelines will be discussed.

3.4.2.1. Visual messagesThe safety effect of visually presented m essages lies quite obviously in thefact that looking at a display interferes with looking at the road environment.Even though peripheral detection of objects outside the car is possible w henglancing at an in-car display, this possibility should not be taken tooseriously.Guidelines to prevent driver overload by visuai information presentationshould preferably be expressed in terms of total glance time, time ofindividual glances, and glance frequency. Even though peopie have someiiberty in choosing the fiequency/duration ratio when extracting informationfrom a single display, it appea rs that the total glance time is relativelyconstant ac ross driving situations. This suggests that total iooking time is areasonable measure for expressing visual workload criteria.Visual in-vehicle displays should not require more than three glances of 1 san it should be possible to acquire useful chunks of information in at leastone second. Messages should always be driver paced in that the informationwill remain to be presented for a relatively long time (e.g., 1 min) and thedriver is free to decide when to look. For the display of complexinformation, the possibility should be considered to present information inrelatively simpie portions which are presented only when requested by thedriver. Only with heads-up displays longer glance times are allowed becausecourse keeping and car following can then be carried out with peripheraivision.

24


25/65

The following issues still need to be resolved:Do drivers adapt headway safely when they are looking at an in-vehicledisplay in a car-following situation?Do they take following r af ic int0 account when braking in front of anintersection or wil1 the probability on rear-end collisions increase due tosudden hard braking?

3 -4.2.2. Mentally demanding messagesOn the assumption that drivers should not fully attend to other tasks forlonger periods of time even though their eyes are directed at the roadway,auditory (speech) messages should not last longer than a certain period oftime unless the messages do not require full attention, such as with highlyfamiliar messages or messages that do not provide crucial information.For the time being, it is proposed that loading messages of more than fiveseconds should be segmented and the driver should ind icate when eachsegment is to be presented. A repeat function of the last message should beavailable and be easily activated . Obviously, it is necessary to determinehow long highly loading speech messages may take before safety is affected.Notice that repetition and segmentation are characteristics which are usuallyinherent in normal (and informal) phone conversations.in case ofconversations requiring much attention or more forma1 telephoneconversations, in which the driver might hesitate to interrupt and ask forinformation again, safety might be affected because the driving task getsinsufficient attention (cf. Briem & Hedman, 1995; McKnight & McKnight,1993; Parkes, 1991). Similarly, when drivers listen closely to the normaltraffic messages the safey of d riving reduces (Akerboom, 1989). In otherwords, it has been demonstrated that driving performance and safety areaffected with tasks that are legally allowed now and a test for safety effectsshould be able to show this,

3.4.2.3. Manual system controlManual system control may affect driving safety due to visual, mental, andmanual demands of the task. Task analysis should estimate which of thesedemands are most detrimental to safety. For now, the following guidelinescan be presented. Controls requiring visual feedback during their use, suchas controls which are sm all, close together, or which function is visuallyindicated, should be avoided. The need to look at the movement is allowedonly when reaching for a control. Hand support and tactile cues shouldfacilitate control without looking at the m ovements. The direction of . .movement of a control should take account of the location and orientation ofthe driver relative to the control. Critica1 and frequently used controls shouldbe close to the predominate position of the hands and should be relativelybig. Make errors difficult but be forgiving. Al1 controls should be in easyreach of the driver. Sing le handed operation should always be possible.Remming the hand to the steering wheel should be possible immediately.

3.4.2.4. Sensory modality andpacingVisualOY auditory information?The present results show that visual as wel1 as complex auditory messagesmight reduce safety. This indicates that the choice between visual andauditory information presentation is one that requires ca refi l consideration.

25


26/65

Auditory signals may be used to inform the driver that important visualinformation is being presented. This signal shouid not be such that the driverstartles and the nature of the visuai message should not force the driver tolook immediately which would change the task in a system-paced task.Also, auditory information is recommended for signals of acoustic origin,for warning signals, to draw attention to visual indicators, when informationmust be presented independently of the orientation of the head, and whenvision is limited or impossible. Tonal signals can be used when the messageis extremely simple, the signal designates a point in time, the message callsfor immediate action, speech s i p a l s are overburdening the driver, andconditions are unfavourable for speech messages (e.g., h igh noise levels).Speech can be used when flexibility of communication is necessary, rapidtwo-way exchanges of information are necessary, and the message dealswith a time-related activity.PacingEven though the aspect of pacing has been mentioned a few times, thischaracteristic can be considered of primary importance. Drivers are able toperform complex tasks and process complex information as long as they areable to indicate when they are able to do so. This m akes them responsiblefor negative safety effects. Therefore, the system should be designed suchthat pacing, though determ ined largely by the driver, is still limited by thesystem. So, the N I S should not present more information than can beprocessed in a limited time (visual 1 s, speech 5 s) and the rate at which nextchunks are presented should be limited with sufficient inter-chunk intervaisto allow the driver to redirect attention to the drivin g task.

3.4.2.5. Testing safety and performance guidelinesThis section presented guidelines and criteria for the hum an-machineinterface of N I S . The se guidelines were presented with respect to the visual,mental, and manual demands of the driver-NIS interaction. Visual demandsshould be limited to three or iess glances of up to 1 s on the average. Anyvisual information shouid be presented sufficiently long so that the driverhac ample time to scan the display at a moment that the driving task allowsdisplay scanning. Smart display design should prevent individual glances ofmore than 1 s.Since instrumented vehicle studies are time and money consuming, there isthe need to test new WIS with respect to their safety effects in a relativelysimple setting. The visual workioad associated with N I S displays couldpossibly be assessed by a laboratory test in which su bjects watch the WISdisplay for succecsive 1 s intervals. The duration of these intervals aresystem controlled but the onset of each glance shou ld be controlled by thesubject. Visual occlusion can be created w ith spectacles or another deviceoccluding part of the visual scene. To mimic the temporal and mentaldernands of course keeping, subjects shouid also pe rf om a tracking taskwith time characteristics similar to those found in real course keeping. Inthis setup, visual workload is equated to the number of 1 s intervals giventhat tracking performance was acceptable. The NIS display is consideredsafe when no more than three 1 s glances were required for understandingthe information.Mentally demanding speech messages should not last longer than a fixedperiod of time. Only when speech m essages are not very ioading, forexample, when they are familiar or redundant, they may be ionger. Loading26


27/65

messages that last longer than five s should be segmented and the drivershould indicate explicitly whether the next or the last segment should bepresented next.The safety effect of speech messages should be determined in a laboratorysetting too. Such a test should assess both the duration and the mentalworkload of the message. A possible metric for mental demand of speechmessages can be developed with the Continuous Memory Task (CMT)technique (Boer & Joma, 1987).Then the N I S safety test would involveresponding to speech messages of the N I S while at the sam e time countingthe num ber of target letters that are displayed visually. Higher w orkload andlonger N I S message durations will reduce CM T accuracy. Given the lack ofdata on safety effects of mentally demanding messages, driving studiesshould indicate which levels of mental demand are acceptable. The speechmessages used in the present experiment might be used as a firstapproximation of unacceptable mental workload. Similar tasks of varyingcomplexity can be used to determine safety as a function of mentalworkload.The present results clearly show that manually controlling a system mightaffect safety. The ca use may lie primarily in the visual, the mental, a nd themanual workload of the interaction. For visual and mental w orkload, theabove criteria should be used: n o more than three 1 s glances and no highmental dema nds for longer than a certain period of time. Operatingcomplex ity should be limited too in order to limit the visual and mentalworkload caused by movement control. Again, driver workload should betested in a laboratory setting in which interference with a secondary taskindicates the workload of the interaction.Now, this secondary task should be sensitive to al1 types of w orkloadassociated with manual system control. For example, subjects wil1 have toperform a tracking task with timing characteristics which are com parable toiane keeping (indicating visual and manual workload). Certain simplediscrete actions will have be made in response to stimuli which are presentedat various locations (indicating mental workload). The timing and locationof some of these stimuli can be anticipated, others cannot. Next, the effectsof a couple of different manual control tasks should be tested with this taskas wel1 as in a safety study in real traffic in order to determine the relation-ship between workload and safety effects. Then, the de gree of performancereduction on this laboratory task wil1 indicate the degree of safety reductionby controlling the NIS.Finally, no matter the type of task the driver has to perform with an IVIS,explicit attention should be given to the timing of the dr iver -NISinteraction. The driver should always be able to determine when he or sh e isable to pay attention to the IVIS and should always be able to interrupt anongoing interaction. Furthermore, interactions that require much attentionmay never last longer than a certain period of time (visual: 1 s, speech: 5 sfor the time being) until the driver ind icates that a next part of the task canbe performed . For visual displays and manual control this implies that thestate of the system does not automatically change. Speech m essages shouldbe of limited length and repeatable. Al1 WIS should comply with theseguidelines on pacing

3 A.3. Procedura l guidelines o r safety testingEventually, each N I S should be tested on its safety implications byassessing the safety effects caused by its support to the driver (positivesafety effects) and its effects due to workload and distraction, and its effects

27


28/65

of exposure (negative safety effects). It is not possibie to give concreteprocedures for safety effects due to driver support and changes in exposurefor ai1N I S as these effects are largely dependent on the system at hand.For exampie, with respect to safety effects caused by changed route choice ,potentiai users may be interrogated how they wouid ada pt their route choicegiven certain types of W IS messages.Effects can also be established in fleet-studies but this is usualiy not feasible.If negative safety effects can be expected due to improper route choice,system design may be changed such that they p i d e drivers more alongroads that are known to be relatively safe. Also, when certain roads areknown to be used m ore frequently due to som e types ofWIS, measures maybe taken to increase the safety of those roads.Safety effects due to workload and distraction, which was the topic of thepresent report, can probably be tested by laboratory tests. With respect tovisual workioad, the guidelines are sufficientiy ciear to allow laboratoryassessment of safety effects although m ore detailed anaiysis might still testthe 3 times 1 s rule in more detail. Laboratory assessment of men taiworkioad and the workioad involved in manual system control is not yetpossible. The iaboratory procedures to investigate these types of workloadrequire frther detail.Also, there is still the need to link these results to safety effects in actuaidriving. Tasks with various experimenter-controlled levels of mental andmanual workload shouid be carried out both in an instrumented vehicie andin a laboratory setting. The safety effects obtained in real driving wil1indicate which are the criteria for acceptable leveis of performance in thelaboratory tests. It should be stressed that, as long as many of the criteria areunknown, assessm ent of these safety effects shouid be done both by way ofby (subjective) expert opinion and by measuremen t of objectivecharacteristics. Once these criteria are known, safety effects of N I S can bedetermined in relatively simple laboratory tasks.

3.4.4. Product guidelinesIn order to function appropriately, an N I S shouid fulfill elementaryergonomics criteria. The product guidelines are derived from variousproposals in th e literature (Boff & Lincoln, 1988; Galer & Simmonds, 1984;Green et al., 1995; Sherwood-Jones, 1990). These guidelines do notguarantee safety but are aimed at coming to optima1 designs for individuaihuman-machine interaction. As such, they can be used for improving N I Swhich are iikely to affect safety. The gu ideiines are meant for inform ationwhich is not directly alerting. That is, the messages are assumed to conveyinformation that need not be perceived quickly by the driver. For alertinginformation which require immediate driver responses, more conspicuousmessages may be allowed.The product guidelines should follow som e genera1 principles: It should bepossible to turn off ai1WIS at any time with a single action in order toinhibit al1 potentially distracting messages. B e consistent across varioussubsystems or tasks and across time with respect to system characteristicsthat recur such as coior coding and data entry. For exampie, use the samesequence of actions across subsystems for entering information and forpresenting similar types of information. Compiy with peopies expectations.For example, increase volume and any another vaiue by rotating clockwise.Minimise what the driver has to remember du ring interaction with the N I S .Operations that occur most often or have the greatest impact on driving28


29/65

safety should be the easiest to carry out. For example, setting a destinationshould be easier than recalibrating a system. Also, present only informationuseful to the driver; always consider whether the driver needs theinformation that is presented.Two types of ergonomics criteria can be distinguished. Criteria that wil1have a d irect effect on a safety and performance evaluation and those that donot have such immediate effects because they are m eant for specialconditions such as low or high luminance or high noise conditions (drivingfast). Characteristicsof the N I S that directly affec t safety and perform ancewill af fect the measurements in the safety and performance test described inthe previous section. No further test of these criteria is needed. Yet, they canbe considered for improving the IVIS. The ergonom ics criteria that aremeant to improve the N I S under specific conditions, require carefu lconsideration in a safety evaluation too. For each NIS, it should bedemonstrated that that system can be used under the various cond itions thatoccur in driving such as different light and noise conditions.The list of ergonom ics criteria and product guidelines (Verweij, 1996) formsa core list. On the basis of this knowledge, each N I S should be tested onwhether they will still convey their information and perform the ir task underadverse light and noise conditions. This will have to prevent that thedemands of the IVIS increase under such conditions.

29


30/65

4.

4.1.

4.1.1.

4.1.2.

Criteria for experimental testingTh e discussion in this paragraph is based on the previous chapter and on thereport:- Verwey, W., Brookhuis, K.A. & Janssen, W.H., (1996).Safefy efects ofin-vehicle information systems ( I V I , .In the present chapter it is discussed what type of experimental study isrequired for assessing safety effects caused by the use of telematicsapplications. It d escribes a series of general guidelines, or building b locks,of how potential overload, underload, or counterproductive adaptationeffects should be investigated in relation to safety.

Summary of resultsCriteria for experimental tests o f safety basically consist of three parts:recommendations for th e general design of the experiments;recommended parameters to be observed;recommended procedures for interpretation of the results.The results on these items are summarised below.

Recommendations fo r the general design of the experimentsTo get an overall safety assessment it seems useful to use a representativesample of the driver population in the sense that the number of older,middle-aged, and younger/inexperienced drivers should be proportional tothe distance driven by these three groups. Since the resulting interindividualnoise might reduce the reliability of measured effects, a more labourious butbetter method than having a representative sample is to have three groups ofdrivers and weigh their contributions according to their relativecontributions to the annual distance'driven .After obtaining some experience with the operation of the A TT device to betested, these subjects should drive in a number of crowded road situationsfor 45-60 minutes.Furtherm ore it is recommended that safety assessment wil1 be based uponthree phenomena: task overload, task underload and cou nterproductivebehavioural adapatation. This assessment requires at least the observation ofthe following parameters.

Recommended parametersThe following list of measurable parameters is based upon a numb er of sub-tasks and the task load of the driving task.o lateral position handling represented by:- SDLP (standard deviation o f lateral position);- steering w heel handling;- time-to-line crossing;- the proportion under the distribution of lateral positions of the vehicle

that is actually outside the lane.

30


31/65

4.1.3

4.2.

4.2. i

headway control represented by:value of time headway;time-to-coliision;the proportion under the distribution of headways to preceding vehiclesthat is below some criticai value, notabiy 1 O s (short headways) or0.5 s (critically short headways).speed management represented by:actual speed;characteristics of the speed- an d acceleration distribution: standarddeviation of speed;excess of speed limit;pedai use;the 50th and 85th or 95th percentile of the driving speeds and 50th and85th or 95th percentile of acellerations and decellerations may be used tocharacterise the distributions of these parameters. The changes in thesevalues provide sufficient insight in the way these distributions areaffected (usually shified) as a result of task interference.genera1 driving behaviour like:running redyellow lights;reaction time to important events;reaction time to e vents of secondary importance;iane exceedence.task load represented by:seif reported taskload (acco rding to scaies Iike RSME, NASA-TLX);heart rate variability;mirror usage;duration and frequency of giances at display;steering wheel fiequency;

Safes ) assessment by expert opinionThe parameters listed above can be observed and compared to, more or less,established criteria. However, individual differences that are hard to assesswil1 still provide considerabie diflkult ies in interpretation of these objectivemeasures. It is therefore strongiy recommended to include a subjectiveassessment of the safety of driver behaviour by an experienced drivinginstructor. So far the consistency and comprehensiveness of this subjectiverating has been proven superior to the interpretation objective measures.In section 5.2.2, under the hea ding expert opinions, references areprovided for the type of variables and assessment techniques that are to beused by such experts.

Building blo& for studying overload

AssumptionsThe method evoiving from the building biocks in this section involves anumber of assumptions. First, the participants in the study are assumed tobehave as they wouid do in normal driving. This is a very strong assumption,considering that subjects are driving an instrumented vehicie undersurveillance, even if it is in real traffic. For driving on a closed circuit andfor tests in driving simulators, this argument holds a forteriori. Therefore,

31


32/65

care should be taken that driver performance and behaviour are a s normal aspossible.To make the situation more realistic, participants wil1 be o ffered bonuses toincrease time pressure and to prevent over-cautious driving behaviour. Also,it can be stated that system performance is the topic o f study rather thandriving behaviour.Furthermore, it is assumed that the participants are famiiiar with thetelematics application at hand. Basically, their performance and workload ,should more or less have reached an asymptotic level when the experimentstarts. This is more problematic as he application is more complex becauseit takes more time to train the participants. General guide lines in this respectcan not be given. With respect to route guidance systems, Za id el (l9 91 )suggests a period of four hours. To check whether there is sufficientpractice, the study should involve a pilot study in which repeatedmeasurements show how long it takes the participants to get used to thetelematics application.In general, drivers should not exceed certain performance boundaries,should not undertake unsafe driving actions, and should respond sufficientlyfast to predictable and unpredictable events.

4.2.2. Assessing safety effectsThe major issue is how one can decide whether a telematics applicationjeopardises traffic safety or not. For example, even if it is shown thatworkload increases this need not necessarily affect safety as the workloadincremen t may occur at places where workload is low anyway. The re is aneed for cross-validation between objective measures and subjectivemeasures as given by driving experts such as driving instructors. That is, astudy should be performed in which drivers follow a prescribed route withand without a telematics application.The objective measures summarised in paragraph 5.1 should be employedwhich are related to aspects of driving safety and workload and which arelikely to be affected by the use of a telematics application. Expert driversshould rate driving performance on a series of safety related aspects. Bothsubjective and objective m easures should be assessed per driving situation.This procedure wil1 allow cross-validation of the subjec tive and objectivemeasures so that acceptabie boundary values for the various measures can beestablished. F urthermore, to check the reliability of the expert opinions,inter-rater variability should be determ ined. If the inter-rater variabiiity ishigh the ratings are of little use. In order to get an idea of how the safetyeffects relate to the interaction with conventional systems (radio, cassette,heating system ), the study migh t have interaction with these systems as acontrol condition (apart from a non-interaction control condition).The driving task invo lves a series of subtasks differing in their relationshipwith traffic safety.Subtasks of driving that are closely related to safety should not be affectedin any case. These are:General driving behaviourLane exceedence frequency, running red/yellow traffic lights, and reactiontime to important events should not be affected at all. Reaction time toevents of secondary importance should not change m ore than a certainpercentage.

32


33/65

Speed managementSpeed should never increase and should not drop more than a certainpercentage. Brookhuis (1 99 ) found that drivers show speed drops whenevertheir attention is diverted se riously, which is demonstrated, for example, incase of telephoning while driving. How much change is allowed will dependon the driving situation and the degree that it relates to the opinions ofexperts. M unden (1967) asserted that speed should not change more thanone standard deviation fiom the mean for else accident probability willincrease too.Furthermore, deceleration should not change more than a certain percentage.Lateral position handlingSteering frequency and Time-to-Line Crossing should not c hange more thana certain percentage.Headway controlTTC should remain above the 1.5 s criterion. The m inimum followingdistance should not change more than a certain percentage.Task loadAlso, drivers should never look at a display for such a long time that there isthe (theoretical!) possibility of hitting other traffic participants or leaving theroad. Usage of mirrors should also not be affected more than a certainmargin.Expert opinionsThe expert observers should provide detailed opinions about the quality ofdriving. Z ai de l( l9 9 1) provides a list of driving quality variables. Use canalso be made of a standardised observa tion techn ique for assessingerroneous and unsafe behaviour (e.g., Galsterer et al., 1990; Risser &Brandsttter, 1985). Evaluations should be aggregated and averaged acrossthe units of evaluation (street, kilometre of road, junctio n, etc.) so thatdifferent trips can be compared. To prevent experts from knowing whetherthe driver is or is not interacting with a telematics application, the irjudgements should preferably be made from video so that they do not knowwhether a telematics application is being used.Genera1 designOften the effect of a telematics application is Iikely to be smaller than thedifferences between individuals so that within-subject designs are required.Afier sufficient experience with the telematics application, subjects shoulddrive in crowded road situations, preferably in their own car, for 45-.60minutes.Notice that it is important to give participants in an experimenta l studyrelatively much freedom so that they can compensate for the occ urrence ofoverload. For example, if headway increases when interacting with atelematics application this indicates an increased level of workload but notnecessariy unsafety. In this case, eye movement analyses may indicatewhether the headway increase is su fic ie nt for compensating for the visualattention paid to a display.In short, assessment of driver overload should not affect drivingperformance. Also, the measures should have a high temporal resolution andallow one to determine

Documents

44696988 Automation of the Driving Task