S. de Groot*J. C. F. de WinterDepartment of BioMechanicalEngineeringFaculty of Mechanical, Materials andMaritime EngineeringTU Delft2628 CD, Delft, The NetherlandsM. MulderControl and Simulation DivisionFaculty of Aerospace EngineeringTU Delft2600 GB, Delft, The NetherlandsP. A. WieringaDepartment of BioMechanicalEngineeringFaculty of Mechanical, Materials andMaritime EngineeringTU Delft2628 CD, Delft, The NetherlandsNonvestibular Motion Cueing in aFixed-Base Driving Simulator:Effects on Driver Braking andCornering PerformanceAbstractMotion platforms can be used to provide vestibular cues in a driving simulator, andhave been shown to reduce driving speed and acceleration. However, motion plat-forms are expensive devices, and alternatives for providing motion cues need to beinvestigated. In independent experiments, the following eight low-cost nonvestibularmotion cueing systems were tested by comparing driver performance to controlgroups driving with the cueing system disengaged: (1) seat belt tensioning system,(2) vibrating steering wheel, (3) motion seat, (4) screeching tire sound, (5) beepingsound, (6) road noise, (7) vibrating seat, and (8) pressure seat. The results showed thatthese systems are benecial in reducing speed and acceleration and that they improvelane-keeping and/or stopping accuracy. The seat belt tensioning system had a particu-larly large inuence on driver braking performance. This system reduced driving speed,increased stopping distance, reduced maximum deceleration, and increased stoppingaccuracy. It is concluded that low-cost nonvestibular motion cueing may be a welcomealternative for improving in-simulator performance so that it better matchesreal-world driving performance.1 IntroductionDriving simulators are broadly used for research, training, and assessment.The effectiveness of a simulator depends to a large extent on its delity, or levelof realism. A distinction can be made between two types of delity: objective(or physical) delity and perceptual (or psychological) delity (Advisory Groupfor Aerospace Research and Development, 1980; Burki-Cohen, Soja, & Long-ridge, 1998). Objective delity is the extent to which the simulator replicatesthe physical characteristics of the simulated vehicle and environment, for exam-ple, in terms of brightness and contrast of visual display, or temporal synchroni-zation of physical motion. Perceptual delityarguably a more valid criterionthan objective delityis dened as the degree to which the operators per-formance and control strategies in the simulator and real vehicle correspond, aswell as the degree to which the operator subjectively perceives the simulator toproduce its real-world counterpart. Researchers have been concerned with gain-ing an in-depth understanding of the relationship between user performancePresence, Vol. 20, No. 2, April 2011, 117142 2011 by the Massachusetts Institute of Technology *Correspondence to s.degroot@tudelft.nl.de Groot et al. 117and perception in virtual environments in comparison toits real world counterpart (e.g., Bella, 2008; Kemeny &Panerai, 2003; Mania, Troscianko, Hawkes, & Chalm-ers, 2003; Nikooyan & Zadpoor, 2009; Shechtman,Classen, Awadzi, & Mann, 2009). This study is also con-cerned with perceptual delity. More precisely, we inves-tigate the means to make the simulator realistic interms of recorded driver performance and subjectiveexperience.A large share of the delity of a driving simulator istraditionally attributed to the implementation of amotion platform and the quality of its motion cues.Motion platforms feed back vehicle acceleration by tilt-ing and translating the driver, and have been shown tobe successful in reducing vehicle speed and accelerationduring many driving tasks. More specically, motionplatforms have been shown to result in lower onset jerkduring braking, lower maximum deceleration duringbraking, lower cornering acceleration, better lateral vehi-cle control, more precise positioning of the vehicle to astopping marker, and a higher subjective realism of thesimulator (Colombet et al., 2008; Brunger-Koch, Briest,& Vollrath, 2006; Greenberg, Artz, & Cathey, 2003;Pinto, Cavallo, Ohlmann, Espie, & Roge, 2004; Siegler,Reymond, Kemeny, & Berthoz, 2001; Reymond,Kemeny, Droulez, & Berthoz, 2001). With the adventof inexpensive high-end outside visual systems, motionplatforms are becoming relatively more expensive. Inmany research applications as well as in commercialdriver training, the implementation of a motion platformcan seldom be justied, as a good motion platform oftencosts much more than a real car (Evans, 2004).Fixed-base (without a motion platform) driving simu-lators avoid the cost issue. The literature shows thatthese nonmoving simulators provide driving perform-ance measures which correlate well with those obtainedin real-world driving, such as driving test performanceand accident-involvement (e.g., Allen, Park, Cook, &Fiorentino, 2009; De Winter et al., 2009; Lee, 2003).Although this relative validity is very encouraging, theabsolute values of driving speed, acceleration, and num-ber of driving errors are generally considerably higher inxed-base simulators than in reality (Green, 2005; Hoff-man & McDowd, 2010; Hurwitz, Knodler, & Dulaski,2005; Boer, Girshik, Yamamura, & Kuge, 2000; Reed &Green, 1999). This impairs the validity of the researchconducted with these devices and likely hampers a xed-base simulators training effectiveness. Whereas normaldriving on the road typically results in sustained accelera-tions of 4 m/s2during cornering or braking, decelera-tions of 6 or 7 m/s2are not uncommon in xed-basesimulators (cf. Siegler et al., 2001). Such deviant per-formance is often attributed to the lack of physicalmotion cueing, leaving the driver with only the visualsystem to perceive his or her locomotion through theenvironment, which in turn results in inferior speed per-ception as compared to reality (Greenberg et al., 2003;Boer et al., 2000). In conclusion, low-cost alternativesfor providing motion cues need to be investigated.1.1Nonvestibular Motion CueingIn the context of aviation, Vaden and Hall (2005),as well as Burki-Cohen, Sparko, and Go (2007) com-mented that a motion chair may offer many of the advan-tages of motion platforms without the disadvantage ofhigh costs. With alternative motion cues, which are not(primarily) aimed at stimulating the vestibular organs,there is a potential to provide fully proportional andsustained acceleration feedback. For ghter aircraft,dynamic seats have been studied extensively and havebeen shown to yield a positive effect on pilot ying per-formance and the reported realism of the simulation(Ashworth, McKissick, & Parrish, 1984; Chung, Perry,& Bengford, 2001; Flach, Riccio, McMillian, & Warren,1986; Martin, 1986; Parrish & Steinmetz, 1983; Rut-ten, 1999; but see Showalter & Parris, 1980, for acounterexample).Riecke, Schulte-Pelkum, Caniard, and Bulthoff(2005) conducted experiments to achieve self-motionsimulation in virtual reality without physically movingthe observer. They investigated the effects of scene con-sistency, minor modications of the projection screen,and multisensory cue integration using seat vibrationsand auditory cues. Riecke et al. showed that the illusionof motion can be facilitated using these modications,that is, without physical movement. Mollenhauer,Romano, and Brumm (2004) studied different types of118 PRESENCE: VOLUME20, NUMBER2motion that were presented by a motion seat in a drivingsimulator. They found that a motion seat without visualcompensation of the seats movements was preferred bythe participants.1.2Research AimThis study aims to investigate whether low-costmotion cueing devices can be used to improve in-simula-tor driving performance. Our main emphasis lies inreducing driving speed and acceleration in order toimprove lane keeping and stopping precision in the sim-ulator, and to obtain more realistic values that are com-parable to those reported in typical on-the-road drivingtasks. We describe eight independent experiments, ineach of which the effects of a low-cost motion cueingsystem were investigated using elementary braking andcornering tasks. In addition, a meta-analysis was con-ducted on the results of the eight experiments, in orderto detect underlying regularities in the individual experi-ments. In the meta-analysis, the moderating role ofdriver experience was investigated as well.2 Experiments2.1ApparatusAll experiments were conducted in a xed-basedriving simulator, called the Dutch Driving Simulator(Green Dino, 2008). The steering wheel, pedals, andgear lever were obtained from a real car and the dash-board, interior, and mirrors were integrated in the pro-jected outside world image, as shown in Figure 1. Steer-ing wheel force feedback was provided through anelectrical motor according to the self-aligning torque ofthe front wheels. The simulator provided a horizontalvisual eld of view of 1808 through three projectors. Thefront view projection had a resolution of 1,024 786pixels; the side views featured a resolution of 800 600pixels. The simulator provided realistic engine and windsound from four speakers in the simulator cabin. Thesimulation ran at a frequency of 100 Hz and the updaterate of the visual projection was always larger than25 Hz.2.2Experimental ProtocolsEight experiments were performed independentlyover the course of four years (20052008). Each of theseexperiments evaluated a particular motion feedback de-vice, and compared driving performance as obtained withthe system enabled versus disabled (i.e., motion on vs.motion off), using either a between-subjects or a within-subjects experimental design. The high similarity betweenthe experimental protocols allows for a joint investigationof the relative effects of each motion system. Comparisonbetween experiments of the absolute effects can only bedone with care, however, because inevitably the experi-mental protocols differed from each other. Table 1 pro-vides relevant details about the experimental protocols.In the between-subjects experiments, the participantswere allocated at random or alternately to either themotion on or off group. The same method was appliedfor the within-subjects experiments, with the additionalproviso that participants who drove with motion on in asession drove with motion off in the following session,and vice versa. In one of the between-subject experi-ments (motion seat), we deviated from the above proce-dure because of the practical difculties of changing themotion conditions. In this experiment, the participantsdrove with the same motion condition on a single day.However, we considered it unlikely that unidentiedsequence effects contaminated the results, because theeffects of the braking experiment were replicated in anFigure 1. Driving simulator as used for the experiments.deGrootetal. 119Table1.ExperimentalProtocolsNumberofparticipantsNumberofwomenExperienced/inexperiencedaWithin-/between-subjectsdesignNumberofsessionsperparticipantNumberofmaneuversbTaskcAutomatedcontrolsd1.Seatbeltsystem200IW210stopsASG2.Vibratingwheel132EW210stopsA3.Motionseat600EB110stopsAG3.Motionseat14turnsCG4.Screechingtires120IW44stopsASG4.Screechingtires44turnsBG5.Beepingsound280IB16stopsASG5.Beepingsound24turnsBG5.Beepingsound114turnsCG6.Roadnoise369IB110stopsASG6.Roadnoise214turnsCG7.Vibratingseat153EW48turnsB8.Pressureseat315IB114turnsCGaExperiencedisdenedasbeinginthepossessionofadriverslicense.bNumberofmaneuversindicateshowmanymaneuversofonedrivingsessionweretakenintoaccountinthestatisticalanalyses.cTaskindicatesthetaskthatparticipantshadtoperform.Threedifferenttaskswereused,asexplainedinthetext.dAutomatedcontrolsindicateswhethersteering(S)and/orgearchanging(G)wasautomatedbythesimulator.120 PRESENCE: VOLUME20, NUMBER2additional within-subjects experiment with the motionseat (De Winter, De Groot, Mulder, & Wieringa, 2007;N 24; data not included in this study).The experiments were composed of one or more ofthree different tasks marked as Task A, Task B, and TaskC. All tasks were conducted without any other trafcpresent in the simulated world. During all experiments,participants had 45 min of driving time in a different vir-tual world, in order to become familiarized with the sim-ulator before the actual experiment started. The speedlimit was indicated by road signs, and participants weregiven no further feedback about their driving speed.During Task A, participants had to drive along astraight road containing many intersections, which weremarked with a stop sign and a stop line. The speed limitvaried from intersection to intersection between 30, 50,and 80 km/hr, as indicated by road signs and 80 km/hr.Braking performance was analyzed during Task A, andthe stops at intersections 211 were included in the anal-ysis per driving session. Task B was a cornering task dur-ing which participants had to turn either left or right atintersections. These intersections had no specic roadmarkers, and without the presence of other vehicles, par-ticipants were free to choose their path and speedthrough the turn. The turn direction was kept constantduring one driving session, thereby essentially driving asquare around the block. For the second session, thesubjects had to perform the same task with the samemotion condition, but in the opposite direction. Turns25 of every driving session were included in the analy-sis. The speed limit during Task B was 50 km/hr. TaskC featured many curves with different radii on a closedtwo-lane lap without intersections, and a speed limit of80 km/hr. Participants were asked to drive around inthis world and follow the road. One lap (7.5 km) of thisworld was in the data analysis, which had fourteen 908curves. Lane width was 5 m.2.3Participants and Instructions toParticipantsAll participants were recruited from the Delft Uni-versity of Technology community, and were almost allundergraduate students at the Mechanical Engineeringfaculty. Participants were not informed about the precisepurpose of the experiment, and instructions were kept toa minimum. Participants were asked to perform a drivingtask or complete a driving lesson. For a number ofexperiments, experienced participants were used, withexperience dened as having a drivers license, whereasother experiments featured inexperienced drivers. Whenexperienced drivers were used, they were asked to driveas they normally would while respecting the trafc laws.When inexperienced drivers were used, they were giventhe minimum information necessary to complete a (sim-plied) driving task. Most experiments were completedwith an automatic gearbox, and some braking experi-ments with automated steering and an automatic gear-box. Table 1 summarizes this information for all experi-ments.2.4Dependent Measures and StatisticalAnalysesTables 2 and 3 summarize the dependent measuresthat were used for the analyses. The measures werelargely based on the work of Siegler et al. (2001), as theyanalyzed motion versus no motion during braking andcornering. For the stopping task (Task A), speed and dis-tance at the onset of braking (Vini and DTTini) are im-portant since they effectively determine how hard thedriver should brake in order to come to a stop at the stopline. The nal stopping position (DTTn) is a measure ofprecision, and the R2time measure describes how driversslowed down, that is, in one movement or in variousstages with multiple brake pedal modulations. Thesefour measures have been shown to be valid and uniquedescriptors of braking performance (De Groot, De Win-ter, Wieringa, & Mulder, 2009). Additionally, the meanspeed over the complete stopping maneuver, the stop-ping consistency (SD DTTn, measured as the SD of alldistances-to-stop-target in the session), the maximumdeceleration (max. dec.), and the initial rate of change ofdeceleration (onset jerk) provide further informationabout braking performance. Note that the braking varia-bles did not include the foot-off-gas-pedal time prior topressing the brake; the onset of braking was dened asdeGrootetal. 121the rst moment of pressing of the brake pedal at a dis-tance of less than 175 m before the intersection.For the cornering tasks (Tasks B and C), the measuresLCE45deg (lane center error at the midpoint of the 908turn) and V45deg (velocity at the midpoint of the 908turn) provide information on the path and speed halfwaythrough the turns. Because only relatively small changesin vehicle paths can be expected, the speed effectivelydetermines the lateral acceleration required to completea turn. We therefore decided not to include the lateralacceleration in the measures, also because of the largevariance of the lateral acceleration during cornering.Siegler et al. (2001) showed that the LCE45deg andV45deg measures were sensitive measures to describecornering behavior, as opposed to the lateral accelerationat the midpoint. Finally, the SDLP (SD of the lateralposition) measure was calculated for the complete turn,which describes the variance of the vehicle path com-pared to the average lateral position. For corners withouta clear centerline (i.e., during Task B), the lateral posi-tion was calculated as compared to a participants per-sonal mean path (resulting in the SDLP measure), asTable 2. Dependent Measures for Task AAbbreviation Unit DescriptionVinim/s Speed at onset of brakingDTTinim Distance to the target line at onset of brakingDTTnm Distance to the target line of the stopping positionR2time Squared correlation coefcient of the speed versus time data from brakeonset to stopMean V m/s Mean speed from 175 m before to 30 m after the stopSD DTTnm Session SD of distance to the target line at stop positionMaximum deceleration m/s2Maximum deceleration during braking for speed greater than5 km/hr (t T s)Onset jerk m/s3Mean initial rate of deceleration (deceleration at t T/2 divided by T/2)NOTE. The measures were calculated for each participant over the 2nd to the 11th stop of the session. t representsthe elapsed time since the onset of braking.Table 3. Dependent Measures for Tasks B and CAbbreviation Unit DescriptionNrDeparturesa# Number of road departuresLCE45deg (L) m Lane center error when halfway through left turnsLCE45deg (R) m Lane center error when halfway through right turnsV45deg (L) m/s Speed when halfway through left turnsV45deg (R) m/s Speed when halfway through right turnsSDLP (L) m SD of the lateral position through left turns (Task C)SDLP (R) m SD of the lateral position through right turns (Task C)SDLP (L) m SD of lateral deviation from own mean path through left turns (Task B)SDLP (R) m SD of lateral deviation from own mean path through right turns (Task B)aAll other measures during Task B were calculated for each participant as the mean of the second to the fth turn ofthe session. For Task C, the dependent measures were calculated as the means of all 908 turns with radii of 1520 m.Data from 10 s prior to 20 s after road departures were removed.122 PRESENCE: VOLUME20, NUMBER2opposed to the mean lateral position with respect to theroads centerline. For these measures, a distinction wasmade between left (L) and right (R) corners, as left andright corners have different radii, lane markings, and laneboundaries, and large differences in behavior betweenleft and right turns can be expected.Apart from the objective measures, a short question-naire addressing the participants subjective experienceof the simulators realism and/or the participants owndriving performance in the simulator was used in eachexperiment. If a questionnaire item contained the wordsrealism or realistic (e.g., The sound in the simulator isrealistic), then this item was included in the calculationof a subjective realism score. For some experiments, onlyone question was used to calculate a participants score.For other experiments, the average of ve questions wasused. All questionnaires (except that of the screechingtires experiment in which only open questions wereasked, and the vibrating seat in which binary answerscould be given) contained 4-point, 5-point, or 10-pointLikert items. Except for the open and binary questions,the Likert scale data were transformed to a percentageranging from 0 to 100 for the 4 to 10 entry possibilitiesand treated as interval scale data during further analysis.The dependent measure and questionnaire data weretested for the effects of the experimental conditionsusing either a paired or an independent Students t-test,depending on whether the experiment was of within-subjects or between-subjects design. Note that the t-testis a powerful and robust alternative to nonparametrictests for Likert-type questionnaire data (De Winter &Dodou, 2010). In the meta-analysis we used a combinedindependent-paired t-test (Looney & Jones, 2003).Two-tailed p-values were calculated, as well as the 95%condence interval for the true difference of populationmeans. Cohens d (i.e., the mean standardized differ-ence) was used as an effect size measure; this standar-dized effect size allows for comparisons across studiesand across the dependent measures.2.5The Eight Systems Under EvaluationEight motion cueing systems, all aimed at reducingdriving speed and acceleration, were developed and eval-uated. The systems were: (1) seat belt tensioning system,(2) vibrating steering wheel, (3) motion seat,(4) screeching tire sound, (5) beeping sound, (6) roadnoise, (7) vibrating seat, and (8) pressure seat. The rsttwo systems provided feedback of the cars longitudinalacceleration; systems (35) fed back both longitudinaland lateral accelerations; system (6) provided feedbackon the cars velocity, and systems (7) and (8) providedfeedback of the cars lateral acceleration. The characteris-tics of all systems and experiments will be brieydescribed below.2.5.1Seat Belt Tensioning System(Longitudinal Acceleration Cueing). The feedbackcue was a tension force in the seat belt proportional tothe deceleration of the vehicle. Seat belt tensioning haspreviously been used in aircraft simulation for lateral andvertical acceleration cueing by tensioning as well as later-ally scrubbing the seat belt over the pilots lap(Heintzman, 1996). In the present study, the tensionforce of a standard seat belt over the drivers leftshoulder increased linearly from 0 to 150 N for decelera-tions between 0 and 5 m/s2. For decelerations largerthan 5 m/s2, the force remained at 150 N. These forcesettings were based on the results of a just-noticeable-difference experiment (N 8, data not shown), whichshowed that participants had trouble discriminating aforce increase of 25 N, while a reduction of 25 N wasnoticed clearly. Increases of 50 N and above were clearlynoticed by all participants. The maximum force of150 N was determined by the experimenters as the maxi-mal force which could still be considered comfortable.To provide some sense of force magnitude, 150 N isapproximately the force required to support the uppertorso (assumed mass of 30 kg) during a 0.5 g decelera-tion. During crashes, forces exceeding 5,000 N canresult (Gavelin, Lindquist, & Oldenburg, 2007). Theseat belt tension was generated by a moment-inducingmotor with a reduction gearbox, as shown in Figure2(a). The seat belt was fed through a sleeve in a cylinderon the back of the seat, and when a moment was gener-ated by the motor, the belt was gripped by the cylinder,pulling the belt tightly over the shoulder and chest ofthe driver. This effect is illustrated by Figures 2(b) anddeGrootetal. 1232(c). When no moment was exerted, the seat belt couldbe operated in regular fashion. The experiment was runas a within-subjects design, with 20 participants.2.5.2Vibrating Steering Wheel (LongitudinalAcceleration Cueing). Vibrating elements are rela-tively inexpensive and easy to implement. Transferringinformation through vibrations has already been success-fully applied during driving, for example, for presentingspatial warning signals (Ho, Reed, & Spence, 2006),lane departure warnings (Suzuki & Jansson, 2003), andalso for navigation (Van Erp, 2005). Drivers have beenshown to respond quickly and intuitively to vibrations(Suzuki & Jansson, 2003). During ight simulation,vibrations are used to present cues such as stall and stallonset, high speed buffet, landing gear extension, enginevibrations, turbulence, and runway roughness(Heintzman, 1996). In the present study, vibrationswere provided by a bass speaker attached to the steeringwheel, as shown in Figure 3. A low frequency (75 Hz)sample was played. The volume of the sample increasedwith increasing deceleration of the car, according to thegraph shown in Figure 4. The experimenters set up thevibrations so that the vibrations were detectable fromsmall deceleration onward, and set the upper limit of thevolume so that the speaker did not produce audible orresonating sounds. The experiment was run with 13 par-ticipants as a within-subjects design.2.5.3Motion Seat (Longitudinal and LateralAcceleration Cueing). An investigation by Mollenha-uer et al. (2004) showed that a motion seat had a posi-tive effect on driving performance and the subjective ex-perience of realism, irrespective of the motion tuningFigure 2. (a) Seat belt tensioning system on the back of the seat. From right to left, the motor, gearbox, coupling, and sleevedcylinder can be seen. The original belt tension device is also visible below the sleeved cylinder. (b) No tensioning moment appliedby motor. (c) Tensioning moment applied, and belt pulled by the sleeved cylinder.Figure 3. Bass speaker as tted on the steering wheel without thesteering wheel cover.124 PRESENCE: VOLUME20, NUMBER2parameters. Burki-Cohen et al. (2007) showed that asimulator with a motion seat may be useful for trainingairline pilots. In ghter jet simulation, motion seats, suchas the ALCOGS (Advanced Low Cost G-cueing System,Heintzman, 1996; see also Flach et al., 1986; Martin,1986) have been used to feed back information aboutthe state of the aircraft. In these two latter studies it wasshown that a moving seat-pan can be used to provide rollor roll velocity information, which both improved per-formance up to the same level as full body motion cue-ing. Other experiments showed that dynamic seatsimprove performance in compensatory and pursuit track-ing in ghter aircraft simulators (Ashworth et al., 1984;Rutten, 1999). In our experiment, we used the motionseat shown in Figure 5 (Frex Japan Trading, 2008) dur-ing a braking and a cornering task. During braking, theseat tilted forward proportionally with increasing decel-eration up to an angle of 4.78 at 7.7 m/s2. At higherdecelerations, the seat angle remained constant at 4.78.Two conditions were tested during the braking task: onand off. During the cornering Task C, two motion cue-ing algorithms were tested versus the no motion condi-tion (off). During the normal seat movement, alsoknown as the engineering approach (Eng), the driversbody is tilted outward in the turns, and a lateral accelera-tion of 8.3 m/s2corresponded to a seat inclination of6.28. The second motion condition was exactly the op-posite of Eng, and called the fun ride approach (Fun);the drivers body was tilted inward in the turns. Thesemotion conditions were inspired by a paper of Von derHeyde and Riecke (2001), in which they hypothesizedthat the fun ride approach would yield higher pleasureratings and lower realism ratings than the engineeringapproach, and no motion would be worse on all meas-ures.The maximum rotation rate of the seat was 100 deg/s,and the visual scene did not compensate for the seatmotions. This means that the seat moved with respect tothe simulator cabin, and the visual projection as well asthe pedals and steering wheel did not move at all. Theexperiment was a between-subjects design using 60 par-ticipants. Two tasks had to be performed by the partici-pants, rstly Task A, followed by Task C, always in thisorder.2.5.4Screeching Tire Sound (Longitudinaland Lateral Acceleration Cueing). Vehicle sound canincrease the overall sensation of speed (Davis & Green,1995). Not all simulators generate screeching tiresounds when driving near the performance limit of thetires. Considering the fact that people in simulators oftendrive unrealistically fast, providing screeching tire soundsseems a relevant cue to enhance driver awareness. Thescreeching sound was generated when the accelerationFigure 4. Volume regulation for increasing deceleration during thevibrating steering wheel experiment.Figure 5. Motion seat positioned in the Dutch driving simulator. Notethe two linear actuators on the back of the seat.deGrootetal. 125of the simulated car exceeded a friction ellipse (Milliken& Milliken, 1995) with a semimajor axis of 8.0 m/s2inthe longitudinal direction and a semiminor axis of 7.2m/s2in the lateral direction. The sound volume of thescreeching sound increased proportionally for increasingacceleration beyond the ellipse, with maximum volumeat the absolute limit of the car (9.4 m/s2in the longitu-dinal direction and 8.5 m/s2in the lateral direction).The sound itself had a high pitch (frequency contentbetween 1.0 and 1.8 kHz), based on a real world on-board sample of screeching tires. The sound volume levelwas about equal to the engine and wind noise of the sim-ulator near the ellipse, and could increase to the domi-nant sound audible near the acceleration limit. Thesound was presented nondirectionally through fourspeakers in the four corners of the simulator cabin. Thisexperiment was a within-subjects design with 12 partici-pants. Participants rst had to complete four sessions ofTask A, during which the system was alternatively turnedon and off, and subsequently had to perform Task B,with a similar protocol.2.5.5Beeping Sound (Longitudinal andLateral Acceleration Cueing) Similar to the screech-ing tire sound experiment described above, this experi-ment aimed at feeding back vehicle acceleration throughthe auditory modality. The supposed advantage of theless realistic beeping sound feedback with respect to thescreeching tire sound is that it also feeds back accelera-tion during normal driving, instead of only near theacceleration limit. Here, a beeping sound was used witha tone of 1.0 kHz and beep duration of 0.2 s. The beep-ing frequency depended on the magnitude of thevehicles total acceleration vector, calculated as thesquare root of the sum of the squared longitudinal andlateral accelerations. For accelerations from 1 m/s2up to7 m/s2, a beeping frequency starting at 0.5 Hz andincreasing up to 7 Hz, according to the graph shown inFigure 6, was presented. A dead zone between 0 and1 m/s2was implemented to avoid beeping when thevehicle was standing still or driving with constant veloc-ity. The function shown in Figure 6 was implemented tomaximize beeping frequency differences between accel-erations of 2 and 5 m/s2while having smooth transi-tions between frequencies. The sound volume level wasdesigned so that it was clearly audible over the alreadypresent engine and wind noise. The sound was presentedthrough four speakers placed in the four turns of thesimulator cabin. This experiment was set up as abetween-subjects design with 28 participants, and partic-ipants had to perform Tasks A, B, and C consecutively.During Task A and Task B, the on group drove with thesystem enabled and the off group drove with the systemdisabled, while the system was disabled for both groupsduring Task C, to test for a short-term transfer of train-ing effect.2.5.6Road Noise (Vehicle SpeedCueing). Vibrations in the bottom of the seat were usedto simulate the effects of road noise. This system fedback vehicle speed, and not acceleration as all other sys-tems in this paper do. The vibrations were provided by26 small tactile display elements as found in mobilephones, which were equally distributed underneath thedrivers legs (13 per side, as shown by Figures 7[a] and7[b]). The tactile elements were controlled for speeds of0 km/hr up to 90 km/hr. From 0 km/hr, the excitationvoltage of the display elements was increased as a quad-ratic function of speed up to the maximum excitationvoltage at 90 km/hr, from which the voltage remainedconstant with increasing speed. With increasing voltage,the rotational speed of the motor increases, and theeccentric mass causes an increase in vibration amplitude.Participants performed Task A, followed by two sessionsof Task C. This experiment was a between-subjectsdesign, using 36 participants. Two groups were made;one group drove all but the nal session with road noise,Figure 6. Beeping frequency as a function of total vehicle acceleration.126 PRESENCE: VOLUME20, NUMBER2and the other group always drove without road noisefeedback. During the nal session (Task C) all partici-pants drove without feedback and the vibration groupwas tested for a short-term transfer of training effect.2.5.7Vibrating Seat (Lateral AccelerationCueing). In order to provide a sense of lateral accelera-tion to the driver, vibrations depending on vehicle lateralacceleration were provided by tactile display elements onthe bottom of the seat, with the same hardware as wasused for the speed-dependent road noise describedabove. When driving through right turns, elements inthe left part of the seat vibrated and vice versa. For lateralaccelerations above 1 m/s2, four elements (labeled 1 inFigure 7[a]) vibrated on one side. Above 3 m/s2, threeadditional elements (labeled 2 in Figure 7[a]) alsostarted vibrating, and above 5 m/s2all elements (labeled1, 2, and 3 in Figure 7[a]) under the outside leg in theturn would vibrate. This provided participants with astepwise proportional cueing. This experiment was awithin-subjects design using 15 participants. Participantshad to complete four sessions of Task B, in which thesystem was alternately enabled and disabled.2.5.8Pressure Seat (Lateral AccelerationCueing). When driving through a turn in a real car, thedriver is pressed to the side of the seat, and as a resultfeels a force on the side of his or her back. In order tosimulate these reaction forces in a qualitatively correctfashion, the seat of the simulator was equipped with twopneumatic cylinders. For ying, similar pneumatic seatshave been used, for example, the NASA Langleydynamic seat (Heintzman, 1996). The general idea ofthese systems is that strong stimulation of selected parts(small surface, high pressure) of the body provides con-siderable feedback, while the reaction forces (which areequally large because the participant remains in the sameposition) are distributed over a larger part of the body(larger surface, lower pressure), providing less feedback.Here, pneumatic pressure was regulated proportional tothe lateral acceleration. The actuators were tted withmetal plates, on the right and left side of the seat back asFigure 7. (a) Seat cover with 26 vibration elements. The elements are positioned under the drivers left and right legs. (b) Seatcover as used in the simulator. The two most centrally placed actuators were not used.deGrootetal. 127shown in Figures 8(a) and 8(b). For left turns, the plateon the right side was pressurized and the plate on the leftside remained unpressurized. For right turns, this situa-tion was reversed. The maximal force of approximately200 N was chosen by the experimenters as being thelargest force which could still be considered comfortablewhile driving. This experiment was a between-subjectdesign with 31 participants who had to complete Task Cin one driving session.3 ResultsThe results of Task A (the braking task) are shownin Table 4; the results of Tasks B and C (cornering tasks)are shown in Table 5.3.1Seat Belt Tensioning System(Longitudinal Acceleration Cueing)The seat belt tensioning system had a large effecton braking performance. As a result of the system, partic-ipants initiated braking at a greater distance from theintersection (p .013, d 0.69) and at lower speeds(p < .001, d 0.55). They also stopped closer to thestop line (p .016, d 0.53) and slowed down with asmoother deceleration prole as measured by a higherR2time score (p .026, d 0.71). Additionally, themean speed over the complete maneuver was lower(p .005, d 0.46), as were the maximum decelera-tions (p < .001, d 0.87) and onset jerk (p < .001, d 0.99). The stopping consistency was also better with thesystem enabled (p .009, d 0.52). Participantsreported to have noticed the seat belt very well (mean 4.6 on a questionnaire item running from 1 [poorlynoticed] to 5 [strongly noticed], 95% condence interval4.324.88).3.2Vibrating Steering Wheel(Longitudinal Acceleration Cueing)The vibrating steering wheel had a relatively smalleffect on braking performance. The only signicanteffect was that the vibrations resulted in a lower brakeonset jerk, which is the derivative of acceleration at thestart of the braking maneuver (p .013, d 0.54). It isinteresting to note, however, that other performancemeasures changed in the expected direction, althoughnot signicantly. For example, the speed at brakeinitiation was reduced (p .267, d 0.24), thesmoothness of the deceleration prole was improvedFigure 8. (a) Schematic picture of the pressure seat. Note the placement and orientation of the two pneumatic actuators.The right actuator is active in left turns and vice versa. (b) Pressure system tted in the back of the simulators seat, with thetwo metal plates tted to the pneumatic cylinders.128 PRESENCE: VOLUME20, NUMBER2Table 4. Means of the Dependent Measures for the Braking ExperimentsOff On p d 95% low 95% high1. (A) Seat belt system within (npairs 20)Vini (m/s) 17.02 15.84