Geometric field of view manipulations affect perceived speed in driving simulators

Transcription

1 Geometric field of view manipulations affect perceived speed in driving simulators C. Diels A. M. Parkes Transport Research Laboratory, Crowthorne House, RG40 3GA, Wokingham, UK subm. 6 th October 2009 approv. after rev. 25 th March 2010 Abstract In driving simulators and other virtual reality systems, visual speed is frequently underestimated leading to speed overproduction. This, in turn, may compromise the validity of human behaviour in these environments. The aim of this study was to investigate the feasibility of improving speed perception in a driving simulator by manipulating the Geometric Field Of View (GFOV) of the virtual scene within the projected Field Of View (FOV) of the participant. 16 experienced drivers were asked to produce four target speeds (20, 30, 50, and 70mph) with the speedometer covered. Each target speed was produced under four GFOV/FOV ratios (.83:1, 1:1, 1.17:1, 1.33:1) twice whereby a GFOV/FOV ratio of 1:1 refers to the standard simulator configuration displaying geometrically correct optic flow. Results showed that in the standard configuration, visual speed was consistently underestimated resulting in speed overproduction of 10% on average. The smaller GFOV/FOV ratio of.83:1 led to even greater overproduction, whereas the larger GFOV/FOV ratios reduced the error in speed production. Based on a linear regression, the optimum GFOV/FOV ratio was determined at 1.22:1. It was concluded that manipulation of the GFOV provides a subtle technique to improve the perception and production of speed within simulated and virtual environments. Keywords speed production, driving simulator, geometric field of view, validity 1. Introduction Perceptual judgments are often different in virtual and simulated environments than in the real world [1]. Despite optic flow being veridically displayed, a common observation in driving simulators and other Virtual Reality (VR) systems is that visual speed is underestimated [2-7]. This may compromise the validity of results obtained in driving simulators. For example, when participants compensate for the misperception of speed and subsequently drive too fast, their actions and consequences of their actions may be different and unexpected. Furthermore, in the context of driver training, a misperception of speed can create dangerous situations when trainees transfer their driving behaviour to the real driving situation [7]. Thus, a better perception of speed improves the quality of the simulation. Previous research has shown that the perception of speed is affected by such factors as spatial frequency [8,9], contrast [8,10], declination angle [11], and the Field Of View (FOV) [7,12]

2 Fig. 1 - Schematic drawing showing the computer graphics Geometric Field Of View (not drawn to scale). The manipulation of this variable magnifies or minifies an image Although in theory manipulation of these variables could improve the perception of speed in driving simulators, due to technical limitations or requirements of the application this may not be desirable, applicable, or feasible. It is suggested here that manipulation of the Geometric Field Of View (GFOV) may provide a feasible alternative. The GFOV is the FOV of the virtual scene (see Figure 1). The effect of changing the GFOV, while holding the size of the FOV (viewport) constant, is to minify (wide angle view) or magnify (telephoto view) the displayed image. That is, with a GFOV greater than the FOV, the visible scene is larger than in the real world (see Figure 3 for an illustration). In effect, changing the GFOV alters the geometric structure of the virtual scene. Of particular relevance is that it alters the amount of visual information (object density) in the visual periphery, as well as the proximity of objects to the observer. It has long been known that edge rate information presented to the peripheral visual field has a powerful effect on an observer s perception of speed with perceived speed being positively related to edge rate [13-15]. Edge rate can be defined as the rate at which local discontinuities cross a fixed point of reference in the observer s field of view [16]. A driving simulator study investigating the influence of roadside delineation poles on perceived speed showed that the farther away the poles were from the road, the smaller the perceived speed was [17]. These findings are also in line with the fact that driving down a narrow road or through a tunnel is often accompanied by an exaggerated sense of speed whereas situations with reduced edge rate information, such as open motorways are associated with lower perceived speeds [18,19]. A wider GFOV would therefore be expected to lead to the perception of travelling at a higher velocity. As a corollary, individuals would be expected to produce lower velocities with an increase in GFOV. Indeed, recent studies have provided support for this contention. In a study investigating the perception of speed, participants were shown to perceive the velocity of oncoming vehicles to be faster with increasing GFOV [20]. In another study using a fixed-base driving simulator with a horizontal FOV of 45, the production of vehicle velocity was investigated [21]. Participants were asked to produce two target velocities (30 and 60 mph) under different GFOV settings (25, 55, and 85 horizontal). As expected, the results showed that mean velocity production decreased with increased GFOV

3 It is noteworthy that in the latter study a highly significant interaction effect was found between target velocity and GFOV. In the 60 mph condition, participants were most accurate in velocity production when the GFOV was intermediate at 55, whereas in the 30mph condition, the accuracy of participants production of 30mph velocities increased as the GFOV became larger, reaching its maximum at the 85 GFOV. Thus, if an experiment were to be designed such that vehicles travelled around 60mph, a 55 GFOV would be best to have drivers feel that they were really going 60mph; Similarly, a 85 GFOV would be best if travelling around 30mph [21]. Of course, such an interaction effect would severely limit the usability of GFOV manipulation as a technique to improve speed perception, particularly in simulator scenarios involving variable speeds. An additional consideration is that manipulation of the GFOV should not compromise the simulator s face validity and presence. Presence has been defined as the psychological perception of being in or existing in the virtual environment in which one is immersed [22]. It can be distinguished from real world presence as the extent to which participants believe they are somewhere different than their actual physical location while experiencing a computer generated simulation. In the context of driving simulation, the concept of presence is of particular relevance to ensure participants drive as natural as possible. Numerous factors have been identified that influence presence, including the ease of interaction, level of control, pictorial realism, exposure duration, and system factors such as the FOV, sound, and head tracking [23]. Of course, the selected GFOV may particularly affect pictorial realism. Large deviations from the natural depiction of scenes may hinder engagement in a virtual environment. Thus, manipulation of the GFOV should be subtle enough to go unnoticed but at the same time affect the observer s perception of speed. Returning to the abovementioned study [21], it appears unlikely that these manipulations were left unnoticed considering the extreme discrepancies between FOV and GFOVs employed. However, this issue was not addressed in their study. The aim of the present study was to investigate the effect of GFOV manipulation on speed production in the TRL driving simulator. The current study differed from the abovementioned study into the effect of the GFOV [21] in several ways. First, a wide range of target speeds (20, 30, 50, 70mph) were assessed using different environments (rural and urban). Four different GFOV settings were evaluated, one of which corresponded to a simulator configuration displaying geometrically correct optic flow (i.e. no minification/magnification) for reference. Additionally, subtle GFOV manipulations were employed to investigate the feasibility of affecting speed perception unnoticeably in order to warrant presence and face validity. Finally, the current study examined for the first time the feasibility of GFOV manipulation in a wide FOV simulator. A wide GFOV is normally adopted to compensate for a restricted FOV. Here, the starting point is the veridical situation of a full FOV stimulation with no modification of the virtual environment. Therefore, current results have implications for more advanced driving simulators with a wide FOV of 180º or above. 2. Method 2.1. Participants 16 Participants (12 male, 4 female) with normal or corrected-to-normal vision participated in the study. All participants had a valid driving licence, used their car daily, and had a minimum annual mileage of They were paid and were naïve as to the purpose of the experiment

4 Fig. 2 - TRL driving simulator 2.2. Driving simulator The advanced TRL driving simulator (Figure 2) comprised a Honda Civic family hatchback car with a five-speed manual gearbox. The driving environment was projected at a resolution of pixels per channel onto three forward screens giving a 210 horizontal forward field of view and a rear screen providing a 60 rear field of view. Its engine and major mechanical systems have been replaced by an electric motion system that drives rams attached to the axles underneath each wheel. These impart limited motion in three axes (heave, pitch, and roll) and provide the driver with an approximation of the acceleration forces and vibrations that would be experienced when driving a real vehicle. A stereo sound system provided simulated engine, road, and traffic sounds at approximately 75dB(A). The driving simulation was generated by the SCANeR II software (OKTAL, Toulouse, France), and the driving performance data were recorded at a frequency of 20 Hz throughout each participant s drive. An intercom facility allowed for communication between the vehicle and the control room Virtual Environment An urban and rural environment was used (Figure 3). The rural environment consisted of a 12km section of a single-carriage way and included some gentle curves requiring the driver to maintain active steering and speed control. The roadway surroundings were generic with fields and trees. The urban environment consisted of a 12km section of a single carriageway modeled after London. There was no other traffic present in either environment Experimental manipulation The experimental manipulation included four GFOV settings. Instead of absolute values, the ratio of GFOV angle to FOV angle is used here to describe the minification/magnification factor applied to the virtual environment

5 GFOV / FOV ratio.83:1 1:1 1.17:1 1.33:1 Rural environment Urban environment Fig. 3 - Screenshots of the rural and urban environment in each of the GFOV settings Tab. 1 - Simulator configurations indicating the Geometrical Field Of View (GFOV), Field Of View (FOV), and GFOV/FOV ratio GFOV FOV GFOV/FOV ratio 50º 60º.83:1 60º 60º 1:1 70º 60º 1.17:1 80º 60º 1.33:1 For illustrative purposes, Table 1 describes the four GFOV settings with reference to the FOV of only the centre forward screen of the driving simulator which subtends an angle of 60º horizontally by 40º vertically (note that the changes in GFOV settings applied to all three forward screens which created a 210º FOV). A GFOV smaller/larger than the FOV creates a magnification/minification factor whereby less/more of the virtual environment becomes visible. The ratio of 1:1 represents the standard simulator configuration where the FOV equals the GFOV displaying the optic flow geometrically correct. Again, note that the actual (horizontal) FOV of the simulator (210º) remained unchanged throughout the study. Figure 3 shows screenshots of the rural and urban environment with the four GFOV/FOV ratios Design and procedure Participants were asked to produce four target speeds with the speedometer covered. In the rural environment, target speeds were 30 and 70mph; in the urban environment, target speeds were 20 and 50mph. Training: In the first training phase, participants were asked to drive at the lower target speed either in the urban (20mph) or rural (30mph) environment for one minute. During this phase the speedometer was visible and participants were asked to keep their speed constant. With the speedometer visible participants could compare and adjust their driving speed. After the first minute, participants were instructed to accelerate up to the higher target speed of 50 (urban) or 70 (rural)

6 Tab. 2 - Overview of experimental runs Rural environment GFOV 1 GFOV 2 GFOV 3 GFOV Urban environment GFOV 1 GFOV 2 GFOV 3 GFOV Again, they were instructed to keep their speed constant when the target speed was reached. Following the drive at the higher speeds for one minute, participants were instructed to decelerate to the previous lower target speed. Finally, participants were again asked to accelerate to the higher target speed, keeping the speed constant for another minute before bringing the vehicle to a halt. This was the end of the training phase and the experiment was commenced in the same environment as the training took place. Experimental runs: In the experimental runs, the instrument panel was covered throughout, obscuring the vehicle speedometer and engine tachometer. Half the participants started in the rural environment, the other half in the urban environment. They were asked to accelerate up to one of four target speeds, depending on the environment. As soon as they felt they were travelling at the target speed, they were asked to flash the headlights, and verbally inform the experimenter they had reached the target speed. Participants were then asked to accelerate or decelerate to the next target speed. Again, as soon as they felt they were travelling at the next target speed, they had to flash the headlights and inform the experimenter they had reached the target speed. After this, they were asked to bring the vehicle to a halt and this was the end of the first run. To control for speed adaptation [24]-[26], for each combination of GFOV and target speeds, participants performed two runs starting with either the lower or higher target speed. Note that for the statistical analysis, the average of the two runs was used. The same procedure was repeated with a different GFOV setting. In each environment, participants performed 8 runs. This was followed by a brief break after which the second training phase took place following the same procedure but in the second environment and associated target speeds. The order of the target speed presented conditions was randomised across participants and runs. Table 2 summarises the experimental runs. After the experimental runs, participants were asked to fill out a brief questionnaire. 3. Results Figure 4 (left) shows the mean produced speed for each of the four target speeds as a function of the ratio of the GOV and FOV. Note that the target speeds 20, 50 and 30, 70mph were produced in the urban and rural environment, respectively. Overall, it can be seen that with increased GFOV/FOV ratio participants reduced their speed. Furthermore, with a geometrically correct optic-flow (GFOV/FOV = 1:1) participants tended to underestimate their speed resulting in speed overproduction. In contrast, the highest ratio (1.33:1) led participants to overestimate their own speed resulting in underproduction of speed. These effects were found to be consistent across the different target speeds and associated environments

7 Tab. 3 - Percentage speed production error as a function of target speed and GFOV / FOV ratio GFOV / FOV ratio Target.83:1 1:1 1.17:1 1.33:1 Total % 13.0% 2.2% -5.0% 7.9% % 12.9% 4.9% -3.8% 8.9% % 8.1% 1.3% -6.3% 6.0% % 7.3% 1.4% -6.9% 4.3% Total 19.9% 10.3% 2.5% -5.5% 6.8% Produced speed (mph) Mean ratio Produced / Target Speed :1 1:1 1.17:1 1.33:1 GFOV / FOV :1 1:1 1.17:1 1.33:1 GFOV / FOV Fig. 4 - Mean produced speed (left) and mean ratio of produced and target speed (right) for the four target speeds (20, 30, 50, 70mph) as a function of the ratio of the Geometrical Field Of View (GFOV) and Field Of View. Error bars indicate the Standard Error (SE) of the mean For statistical analysis, the raw data (mph) were transformed to ratios simply by dividing each value by the particular target speed the participant was attempting to produce. The reason for this transformation is that there is an inherently large variation in the results that is a function of target speed (e.g. 70mph = 2.3 * 30mph). Figure 4 (right) shows the mean ratio of the produced and target speed for each target speed. A ratio of 1.1, for example, indicates a 10% speed production error. The percentage speed production errors are shown in Table 3. Across target speeds, the percentage speed production error was smallest with a ratio of 1.17:1. Using the transformed data, a three factor repeated measure analysis of variance was performed with the factors being GFOV (4 levels), Target Speed (4 levels), and Environment (2 levels). Mauchly s test indicated that the assumption of sphericity had been violated for the main effect of GFOV (Chi 2 (5) = 12.09, p <.05), and interaction effect of GFOV Target Speed (Chi 2 (44) = 85.35, p <.001). Therefore, degrees of freedom were corrected using Greenhouse-Geisser estimates of within-subjects effect. There was a significant main effect of GFOV on produced speed (F(1.197, 29.52) = 65.30; p <.05). Pairwise comparisons (Least Significant Difference) showed each comparison to be statistically significant (p <.05 in each case)

8 .83:1 1:1 1.17:1 1.33:1 Fig. 5 Ratio of produced and target speed plotted versus the ratio of GFOV and FOV. The line gives the best linear fit No other main and interaction effects (Target Speed Environment) were found to be statistically significant. To further explore the relationship between produced speed and GFOV, a linear regression analysis was performed on the basis of the ratio of produced speed and target speed. In Figure 5 the ratio of produced and target speed is plotted versus the ratio of GFOV and FOV and shows the linear regression with 95% mean prediction interval (y= (GFOV/FOV)). As indicated by the R 2 value, 31% of the variance was explained by the GFOV/FOV ratio. The optimum GFOV range was subsequently determined on the basis of the point where the lower and upper bounds of the 95% mean prediction interval crosses the horizontal line representing no speed production error (i.e. produced/target speed ratio = 1). This area lies between the left and rightmost vertical dotted lines in Figure 5. The lower and upper bound of the 95% mean prediction interval correspond to a GFOV/FOV ratio of 1.18:1 and 1.26:1, respectively. GFOV/FOV ratio corresponding to the point where the regression line crosses the produced / target speed ratio of 1 was 1.22:1. Finally, participants were asked to indicate whether they had noticed any difference with regard to the simulator settings, and if so, what these differences were. 5 out of 16 participants indicated they had noticed a difference in simulator setting. However, none of these participants had noted that the visual scene had changed and instead referred to changes in engine yield. 4. Discussion The current results are in accordance with previous findings that geometrically correct optic flow appears to be too slow during simulated self-motion in Virtual Environments including driving simulators [2]-[7]

9 Under the standard simulator configuration (GFOV/FOV ratio = 1:1) participants consistently overestimated their speed and, as a consequence, drove on average 10% faster than the target speed. The speed production error in the current study was considerably smaller than reported by others [21] whereby drivers overestimated the production of 30mph on average by no less than 20mph, or 67%. In the current study, when collapsed across GFOVs, drivers were found to overestimate the production of 30mph by 8.9%. This large difference in speed production error can, at least partly, explained by the difference in FOV. In the abovementioned study [21], the simulator had a horizontal FOV of 45º, whereas the FOV of the simulator in this study was 210º. A reduced FOV has consistently been shown to lead to lower subjective speeds [12,27,28] and the lack of peripheral stimulation and/or lamellar flow due to a limited FOV has often been referred to as the underlying cause of speed underestimation [1,2,7]. The current results however indicate that underestimation also occurs with a large FOV display (and subsequent presence of lamellar flow) and suggests that the misperception of speed in simulated and virtual environment cannot be solely explained by a lack of peripheral stimulation and/or lamellar flow. Of course, self-motion, and by extension speed, is specified by several sources of information including not only optic flow, but also acoustic flow, proprioception, and vestibular inputs [29]. The full complement of cues is rarely modeled in a virtual environment. In the TRL driving simulator, a high frequency vibration platform and sound system provide proprioceptive and acoustic cues with a relatively high level of fidelity. Vestibular inputs, on the other hand, are almost completely absent. Although the limited fidelity of vestibular cues is often considered an important limitation within driving simulators [1], it should be noted that the relevance of vestibular cues is highly dependent on the specific driving task performed. The vestibular system responds to accelerations only, is unable to signal constant velocity motion, and is less effective in signaling low frequency motion [30]. From this it follows that for driving tasks that require participants to drive at a constant speed, as in the current study, the importance of vestibular self-motion cues are subordinate to visual self-motion cues. The finding that speed is nevertheless underestimated might suggest that the visual selfmotion cues presented in a driving simulator are sufficiently different from the real world to warrant misperceptions of self-motion. Although linear perspective, texture mapping, and lighting are provided by most simulators, motion parallax and stereopsis are generally absent. It is however not self-evident how the latter two would contribute to the perception of speed. The main aim of the present study was to investigate manipulation of the GFOV as a possible compensatory measure of the misperception of speed typically observed in driving simulators. As expected, increased GFOV/FOV ratio consistently led participants to perceive the vehicle as travelling at a higher speed, resulting in a reduction in produced speed. The highest ratio (1.33:1) led participants to overestimate their own speed resulting in underproduction rather than overproduction of speed. An important finding was that, contrary to previous findings [21], no interaction effects for target speed, environment, and GFOV were observed. This has important implications for the exploitation of manipulation of the GFOV in driving simulators

10 It suggests that a single adjusted GFOV setting can be employed under variable speed conditions and different simulated or virtual environments. Based on linear regression analysis, the optimum GFOV/FOV ratio was determined at 1.22:1. As mentioned in the introduction, ideally, manipulation of simulator settings should not affect presence and face validity. When participants were asked afterwards whether they had noted any changes in the simulator setup in the experiment, none of the participants reported having noticed any changes in the visual environment. This indicates that although the modification of the GFOV may lead to changes in the perception of speed, participants remained unaware of the changes in the visual environment. This suggests that manipulation of the GFOV can be subtly implemented without compromising presence and face validity. To evaluate the effects of GFOV manipulation on behavioural validity, future research might benefit from the assessment of the effects of GFOV manipulation on drivers interaction with traffic and events. Adjustment of the GFOV can be expected to change drivers speed choice which, in turn, can be expected to affect their actions and consequences of their actions. In addition, it should be noted that the optimum GFOV/FOV ratio was determined using a wide FOV display (210 ). Considering that speed perception is heavily influenced by the FOV [20], it would be of interest to further explore the relationship between the FOV and optimum GFOV/FOV ratio as this may differ for different FOVs. Investigations into this relationship would allow for the creation of a design curve that can be employed across simulators and other VR setups employing different FOVs. 5. Conclusions Despite optic flow being veridically displayed on a large FOV (210º) display, visual speed was consistently underestimated resulting in speed overproduction of 10% on average. Manipulation of the GFOV provides a subtle technique to improve the perception of speed within driving simulators and other virtual environments. Absence of interaction effects between GFOV, target speed, and virtual environment indicates a single optimum GFOV setting can be employed. The optimum GFOV/FOV ratio was determined at 1.22:1. Manipulation of the GFOV was not consciously detected by observers thereby warranting presence and face validity. References 1. A. Kemeny, F. Panerai, Evaluating perception in driving simulation experiments, Trends in Cognitive Science, vol. 7(1), pp , T. Banton, J. Stefanucci, F. Durgin, A. Fass, D. Proffitt, The perception of walking speed in a virtual environment, Presence-Teleoperators and Virtual Environments, vol. 14(4), pp , G.J. Blaauw, Driving experience and task demands in simulator and instrumented car: a validation study, Human Factors, vol. 24, pp , J. Ostlund, L. Nilsson, J. Tornros, A. Forsman, Effects of cognitive and visual load in real and simulated driving, VTI report 533A, P. Pretto, A. Chatziastros, The role of scene contrast and optic flow on driving speed, in Proc Eleventh International Conference Vision in Vehicles, 2006, pp

11 6. J. Törnros, Driving behavior in a real and a simulated road tunnel-a validation study, Accid Anal Prev, vol. 30(4), pp , H.A.H.C. Van Veen, H.K. Distler, S.J. Braun, H.H. Bulthoff, Navigating through a virtual city: Using virtual reality technology to study human action and perception, Future Generation Computer Systems, vol. 14(3 4), pp , H. Distler, Psychophysical experiments in virtual environments, in Proc. Virtual Reality World 96, Miinchen, H.C. Diener, E.R. Wist, J. Dichgans, T. Brandt, The spatial frequency effect on perceived velocity, Vision Res,. vol. 16, pp , R. J., Snowden, N., Stimpson, R. A., Ruddle, Speed perception fogs up as visibility drops, Nature, vol. 392, pp. 450, F. Panerai, Speed and safety distance control in truck driving: comparison of simulation and real-world environment, in Proc. Driving Simulation Conf. DSC 2000, N. Osaka, Speed estimation through restricted visual field during driving in day and night: nasotemporal hemifield differences, in A. G. Gale et al. (Eds.), Vision in Vehicles II (Amsterdam: Elsevier), J.J. Gibson, The Ecological Approach to Visual Perception. Houghton Mifflin, Boston, MA., D.N. Lee, Visual information during locomotion. In MacLeod, R., Pick, H. (Eds.), Perception: Essays in Honor of James J. Gibson. Cornell University Press, Ithaca, NY, pp , D.L. Warren, Speed zoning and control. In Synthesis of Safety Research Related to Traffic Control and Roadway Elements, vol. 2. Report No. FHWA-TS Federal Highway Administration, Washington, D.C., ch. 17, J.F. Larish, Sources of optical information useful for perception of speed of rectilinear self-motion, J. Exp. Psych.: Human Perception and Performance, vol. 16, pp , O.H. Levine, R.R. Mourant, Effect of visual display parameters on driving performance in a virtual environments driving simulator, in Proc. Human Factors and Ergonomics Society 40th Annual Meeting, pp l 140, D.B. Fambro, D.S. Turner, R.O. Rogness, Operational and safety effects of driving on paved shoulders in Texas. Report No. FHWA-TX-81/ F, Texas Transportation Institute. Texas A&M University, College Station, TX, A. Smiley, Driver characteristics and safety, in Proc. Human Factors and Highway Safety Conference, Washington, D.C., 1997, pp C. Adetiloye, Q. Wu, R.R. Mourant, Perception of optical flow and geometric field of view, in Proc. 32nd International Conference on Computer Graphics and Interactive Techniques (SIGGRAPH 2005), R.R. Mourant, N. Ahmad, B.K. Jaeger, Y. Lin, Optic flow and geometric field of view in a driving simulator display, Displays, vol. 28, pp , C. Heeter, Being there: the subjective experience of presence, Presence-Teleoperators and Virtual Environments, vol. 1(2), pp , W. Sadowski, K.M. Stanney, Presence in Virtual Environments, In K. M. Stanney (Ed.), Handbook of virtual environments: design, implementation, and applications, pp Mahwah, New Jersey: Lawrence Erlbaum Associates, F. Schmidt, J. Tiffin, Distortion of Drivers Estimates of Automobile Speed as a Function of Speed Adaptation, Journal of Applied Psychology, vol. 53(6), pp , A. Irving, The Perceptual Problems of the Driver in Proc. First International Conference on Driving Behaviour, Zurich, Suisse, G.G. Denton, The Influence of Adaptation on Subjective Velocity for an Observer in Simulated Rectilinear Motion, Ergonomics, vol. 19(4), pp , S. Salvatore, The estimation of vehicular velocity as a function of visual stimulation, Human Factors, vol. 10, pp.27 32,

12 28. A.H. Jamson, Image characteristics and their effect on driving simulator validity, in Proc., 1st International Driving Symposium on Human Factors in Driver Assessment, Training and Vehicle Design, Aspen, Colorado, 2001, pp A. Berthoz, The brain's sense of movement, (G. Weiss, Trans.): Harvard University Press, I.P. Howard, The vestibular system, in K. R. Boff, L. Kaufman & J. P. Thomas (Eds.), Handbook of Perception and Human Performance, vol. 1. Sensory processes and perception, New York: John Wiley, pp ,