BROOKS, JOHNELL O. PhD; TYRRELL, RICHARD A. PhD; FRANK, TALISSA A. MS
Despite the widely prevalent challenges to safe driving that exist, it is a simple fact that the vast majority of driving is incident-free. Some researchers have pointed out that the average U.S. driver would have to drive an enormous distance (∼100,000,000 km) before falling victim to a fatal crash.1,2 This suggests that driving is largely an exceptionally safe behavior and that drivers acquire and possess key skills that support safe driving. It seems clear that some of these skills have a strong visual component. For example, one critically important skill is steering – the ability to coordinate the visual contact with the roadway with the motor demands involved in maintaining the vehicle within its intended lane. But researchers have yet to develop a complete answer to what appears to be a fundamental question: What visual mechanisms support steering? As a step towards answering this question, we report two experiments that explore the nature of the visual information that supports steering. We used a driving simulator to quantify drivers' steering abilities while the drivers were confronted with severe visual challenges – extreme optical blur, extremely low luminance, and extremely restricted visual fields. We now briefly review research that is relevant to the effects that these challenges have on driving performance.
Higgins, Wood, and Tait3 asked young participants to drive on a closed-road circuit while wearing blurring lenses such that high contrast acuity ranged from 20/20 (6/6) to 20/200 (6/60). As one might expect, increasing levels of optical blur resulted in decreases in the ability to read road signs. Blur also increased the number of low contrast foam hazards that were hit. But blur did not impair the drivers' steering abilities to the same extent; drivers did not hit more cones during a maneuvering task, and their ability to perform a gap perception task was unaffected. Further, drivers did not slow their speed enough to compensate for the reduction in acuity. Similarly, Owens and Tyrrell4 used extreme amounts of optical blur (up to + 10 D) in a driving simulator and found that despite severe degradations of visual acuity, participants' ability to maintain their lane position was relatively unaffected. Although the authors concluded that a high resolution image does not appear to be necessary for accurate steering, their driving simulator had an unrealistically low visual fidelity, as its display consisted only of white posts that flowed down a black screen.
Several researchers have investigated the extent to which steering performance is affected by the presence, location, contrast, and luminance of lane delineators. In an on-road experiment that compared the responses of drivers on rural roads to responses to roads with various lane delineation approaches, Steyvers and deWaard5 found that the presence of either an edge line or a center line resulted in increased speeds (relative to when no delineation was present) and that the presence of a center line minimized drivers' variability in lane position. This suggests that increasing the visual saliency of the lane's boundaries affects steering performance but can result in drivers increasing their speed. The latter conclusion was also reached by Kallberg,6 who studied the effects of retroreflective lane markings on driver behavior. Kallberg studied 20 pairs of roadways, where each pair had been matched in terms of road width, curvature, hilliness, speed limit, and traffic density. In each pair, Kallberg positioned retroreflective lane markings every 60 m along one of the roadways. Kallberg found that although there was little effect of the lane markings on relatively fast roadways (relatively straight roads with 100 km/h speed limits), on slower and curvier roadways (80 km/h speed limits) the lane markings resulted in increased nighttime speeds and an increased number of nighttime injury crashes.
While both Steyvers and deWaard5 and Kallberg6 manipulated the presence or absence of lane markings, McKnight, McKnight, and Tippetts7 conducted a simulator study in which the appearance of the lane markings was manipulated. The authors tracked four steering-related variables (including the number of lane excursions) as participants drove along simulated roadways that featured lane markings that varied in contrast and width. McKnight et al. found steering performance to be quite robust; the contrast of the lane markings had little effect on steering behavior until the contrast was at its lowest level. Owens and Tyrrell4 also found steering performance to be robust to low luminance challenges. They positioned neutral density (ND) filters over the eyes of young drivers in order to reduce the luminance of their stimuli (30 cd/m2, 1.0 cd/m2, 0.03 cd/m2, and 0.003 cd/m2). Here, luminance had a significant effect on steering performance only when comparing the highest luminance to the lowest luminance condition.
Other researchers have examined the effect of restrictions to drivers' field of view. McKnight, Shinar, and Hilburn8 found no significant difference in any steering-related variables when comparing monocular truck drivers to age-matched binocular truck drivers in a series of on-road driving tasks. In a study of 10,000 drivers, Johnson and Keltner9 found that drivers with a field loss in one eye were not significantly different from full-field binocular drivers in either crash rates or conviction rates but that drivers with a field loss in both eyes had significantly elevated crash rates and conviction rates. Wood and Troutbeck10,11 conducted two experiments in which modified swimming goggles were used to restrict field size while participants completed a number of driving tasks on a closed-road circuit. In their first study, each driver experienced binocular field sizes of 20°, 40°, a monocular field of 130°, and a binocular field of 160°. Reducing the field size resulted in significant performance declines, including a reduction in detecting on-road obstacles (cardboard boxes), a reduced ability to navigate through a slalom course, a reduction in the ability to detect peripheral objects (road signs and pedestrians), a decrease in the ability to maintain a steady position within the lane, and a speed decrease. In their second study, Wood and Troutbeck used a binocular field of 90°, a monocular field of 105°, and a binocular field of 150° and found a similar pattern of results. Here, however, an in-vehicle task was added – the ability to respond quickly to LEDs positioned on the windshield at various eccentricities. Reaction times were significantly longer both when the field size was reduced and when the LED stimulus was at a greater eccentricity. In neither study was there a reliable difference between large monocular fields and large binocular fields in any steering-related variables. Owens and Tyrrell4 used an even more dramatic reduction in field size – a binocular field of only 1.8° – and found a dramatic and significant reduction in steering performance.
The above studies suggest that steering abilities are sensitive to some visual challenges (reduced field size) but are more robust to others (blur, low luminance, low contrast delineators). The two experiments presented here address this issue directly by asking participants to steer along a curvy road while experiencing unusually strong experimentally induced visual challenges. In experiment 1, participants experienced a luminance manipulation (luminance as low as 0.003 cd/m2), and a blur manipulation (maximum of + 10 D of optical blur) and a visual field manipulation (minimum field size of 1.7°). In experiment 2, a new group of ten young participants drove while exposed to seven visual field sizes (1.7 to 150°).
Ten university students (M = 21.2 years) with a minimum of four years of driving experience (M = 5.6 years) received extra credit in a psychology course or $25 for participating in this study. None of the participants had previous experience with the driving simulator or reported any history of visual problems. Participants wearing glasses were excluded from participating due to interference with the experimental devices; participants wearing contact lenses were permitted to participate. All procedures were consistent with the Declaration of Helsinki and were approved by the Clemson University Institutional Review Board.
Participants drove a fixed-base driving simulator (DriveSafety, Inc.) that is built around a four-door sedan (Mitsubishi Galant). An integrated system that consists of six networked computers runs the simulation, gathers input from the driver, and displays the driving scene using four visual channels. Each visual channel consists of a digital projector (1024 × 768 resolution) and a projection screen. The center of each projection screen is positioned 2.65 m from the driver's eyes, and the image projected by each visual channel subtends 50° horizontal by 40° vertical. Together, three of the visual channels presented a 150° forward view that was used in the present study. A fourth (rear) visual channel was not used. Participants interacted with the simulator using the sedan's original steering wheel and pedals. Although neither the steering wheel nor the pedals provided force feedback, the participants were given considerable practice with the simulator and all participants became both comfortable with and competent at using the simulator prior to data collection (see below).
The simulated roadway environment used in the present study consisted exclusively of a flat but unusually curvy two-lane rural road (one lane in each direction) with a lane width of 3 meters. Fig. 1 presents a partial view of a typical scene. The curvy road was designed to challenge the participants' steering capabilities even in optimal viewing conditions. To prevent familiarization with the road, the pattern of curves differed between scenarios but all curves were of equal radius. The participants never encountered other vehicular traffic but did periodically encounter stationary pedestrians standing on either shoulder of the road. Six pedestrians had been quasi-randomly placed (3 on the right and 3 on the left) on the shoulder of the road in all scenarios. The pedestrians (three males and three females) were never visible to the driver from the straightaway that was the beginning of each trial. In order to increase the drivers' workload, drivers were asked to announce the presence of all pedestrians.
During the driving trials, the participants were instructed to maintain a speed of 55 miles per hour (89 km/hr). In all practice and experimental trials, a digital display showing the simulated speed of the vehicle (miles per hour) was projected on the center (front) screen at a height just above the vehicle's hood. Because in some conditions it was impossible for the driver to read this display (as a result of the experimental manipulations), an experimenter read the speed aloud every 20 s throughout all trials. Drivers could request to have their speed read more frequently if desired. If a driver's speed was lower than 80 km/h (50 mph) or > 97 km/h (60 mph), an experimenter asked the driver to adjust the speed accordingly.
Luminance was manipulated by positioning neutral density filters in front of both of the driver's eyes. The filters were mounted on welding goggles that allowed a 116° horizontal by 33° vertical field of view. The four luminance conditions used neutral density filters that reduced the scene luminance by 0.0 (control), 1.2, 2.7, and 3.7 log units. With these filters in place, the average luminance of the roadway's dividing lines (the maximum luminance in the scene) was 16.7, 1.0, 0.03 and 0.003 cd/m2 (1.2, 0.0, -1.5 and -2.5 log cd/m2, respectively). These values were chosen to cover a broad range of luminances from photopic to scotopic, with luminance values in the intermediate conditions typical of those encountered during the dimmer periods of civil twilight.12 Luminance was always tested with maximum image clarity (i.e., no blurring lenses).
Image clarity was manipulated by positioning positive lenses in front of both of the driver's eyes. Trial lenses of 0 (control), +1, +2, +5 and + 10 diopter (D) were mounted in trial frames. Blinders were placed on the sides of the trial frames to prevent the use of peripheral (unblurred) vision. Image clarity was always tested at the highest luminance (no neutral density filters).
In the tunnel vision condition, the participant's field of view was restricted by positioning a cone that had been mounted to a modified eye patch over the driver's left eye. The cone, which was constructed of heavy paper and sheet metal, severely restricted the driver's visual field to a circular field with a diameter of 1.7°. The driver's right eye was occluded with an eye patch. The cone was used without neutral density filters or blurring lenses.
High contrast Bailey-Lovie visual acuity charts13,14 were used to measure visual acuity. Acuity, which was always measured binocularly, was quantified using the logarithm of the minimum angle of resolution (logMAR). Two different forms of the chart were used to reduce learning effects.
Each participant experienced two 90 to 120 min sessions that were split across two different days. Each participant's two sessions were separated by less than seven days. On the first day, participants were introduced to the simulator via several practice trials. During an initial two-minute practice trial participants became acquainted with the simulator by driving along a straight road. All subsequent practice and experimental trials lasted five minutes. The second practice trial introduced the participant to a curvy road of the type that was used in the remainder of the experiment. Drivers were informed that they would never encounter other vehicles and that there were no intersections, stop signs, or stop lights in the simulated world. The third practice trial was similar to the second trial but included pedestrians. The fourth (and final) practice trial was similar to the third but included training in recovering from gross steering errors. Here, steering errors were induced by the experimenter, who periodically steered the vehicle off of the road. The participants could repeat each practice trial as many times as necessary until he or she felt comfortable with the driving simulator. Before all practice and experimental trials drivers were reminded that they had three goals: 1) to keep the vehicle centered in their lane, 2) to maintain a speed of 89 km/hr (55 mph), and 3) to detect and announce the presence of all pedestrians (e.g., “pedestrian left!”). After each trial vision testing was conducted and the participant was given a short break. The first session ended by testing the five blur conditions in a random order.
The second session began with refresher training during which the participants experienced a minimum of three five-minute practice trials on a curvy road. Participants then dark adapted for 30 min, during which the participant sat blindfolded in the driver's seat while talking informally with the experimenters. Once dark adapted, the participant experienced the four luminance trials. In order to optimize dark adaptation and minimize inconvenience to the participants, the four luminance conditions were tested from dimmest to brightest. The tunnel vision condition was tested last.
Simulator data were sampled at 5 Hz during all trials, with the first 30 s of each trial (during which the participant was accelerating) being ignored. As a result of the powerful manipulations, the curvature of the simulated roadway, and the relatively high forward speed, a sizable percentage of the participants' trials were spent outside their designated lane. Because the simulator does not collect lane position data when the vehicle is on the outside shoulder of the roadway or when it is in the opposing lane, the primary measure of steering performance reported here is the percentage of time (out of 270 s) during which the entire vehicle was positioned within the correct lane. We also recorded the number of excursions outside the correct lane (an excursion was recorded when any portion of the vehicle extended beyond the lane for at least 1 s). Although additional steering variables – standard deviation of the vehicle's lateral position and the vehicle's mean lateral speed – are also reported here, interpretation of these variables is more difficult since they are based only on those moments when the vehicle was entirely within the correct lane.
Acuity and driving data are presented in Table 1. Figs. 2 and 3 present the time-in-lane and acuity data as a function of blur and luminance, respectively. To facilitate the reader's ability to compare the effects of the manipulations on the different dependent variables, the time-in-lane and acuity data in Figs. 2 and 3 were transformed into relative values by dividing each mean by the corresponding mean with the largest value for that manipulation. Thus for both the blur and the luminance manipulations, the mean with the highest performance was given a score of 100% and conditions associated with poorer levels of performance received correspondingly lower percentages. (Prior to these transformations acuity scores were converted from logMAR values into their decimal acuity equivalent in order that better acuities correspond with higher numeric values.) These transformations were done only for the purpose of presenting the data in Figs. 2 and 3; all statistical tests were conducted on untransformed data.
As expected, both the blur and the luminance manipulations induced dramatic decrements to the participants' visual acuity. Mean logMAR acuity values ranged from 0.01 (equivalent to a Snellen ratio 20/20 or 6/6) in optimal viewing conditions to 1.51 (20/647 or 6/197) when experiencing maximum blur and to 0.95 (20/178 or 6/54) when experiencing the minimum luminance condition. Repeated measures ANOVAs revealed that the effect of blur on acuity was significant, F(4, 36) = 223.2, p < 0.001, and that blur accounted for 94.1% of the variability in acuity values (ω2 = 0.941). A separate ANOVA revealed that the effect of luminance on acuity was also significant, F(3,27) = 136.8, p < 0.001, and that luminance accounted for 94.6% of the variability in acuity. As expected, visual acuity in the tunnel vision condition was nearly at optimal levels (0.06; 20/23 or 6/7) and was not significantly different from baseline levels, t(9) = 0.5, p = 0.66, ω2 = 0.168.
The blur and the luminance manipulations induced relatively small but significant decrements to the participants' ability to keep the vehicle within their lane. The percentage of time spent entirely within the lane ranged from 95% in the 0 D blur condition down to 83% with minimum luminance and to 88% with + 5 D blur. An ANOVA revealed that the effect of blur on steering performance (as measured by time-in-lane) was significant, F(4,36) = 4.1, p = 0.007, and that blur accounted for 27.6% of the variability in performance. Similar patterns were present in the other measures of steering performance (Table 1). Separate ANOVAs revealed that the effect of blur significantly affected the number of lane excursions, F(4,36) = 4.3, p = 0.006, and lateral speed, F(4,36) = 14.3, p < 0.001, but not the standard deviation of the vehicle's lateral position, F(4,36) = 1.2, p > 0.05.
Another ANOVA revealed that the effect of luminance on steering performance (time-in-lane) was also significant, F(3,27) = 3.4, p = 0.031, and that luminance accounted for 23.8% of the variability in performance (similar patterns were again seen in the other measures of steering performance). Separate ANOVAs revealed that the effect of luminance significantly affected the number of lane excursions, F(3,27) = 5.1, p = 0.006, the standard deviation of the vehicle's lateral position, F(3,27) = 13.5, p < 0.001, and lateral speed, F(3,27) = 40.0, p < 0.001. Interpreting the effect of luminance on steering is not straight-forward, however. To prevent the need for participants to spend 30 min dark adapting prior to every luminance trial, the luminance trials were always tested in order from minimum to maximum luminance. Thus the effect of luminance is confounded with the effect of time and it is therefore difficult to determine how much of the increase in steering performance is a result of increasing luminance and how much is due to practice. However, the data from the blur trials, when the order of the trials was randomized, revealed that there was not a significant practice effect from the first trial to the last trial, F(4,36) = 1.30, p = 0.29. Thus it seems reasonable to infer that the majority of the effect measured during the luminance trials was largely a result of luminance, not practice. Also, by this time in the experiment the participants already had considerable practice with the driving simulator (since luminance trials were tested in the participants' second day of data collection and both days had begun with multiple practice trials). It is also important to note that the total increase in steering performance across the luminance trials was only modest (from 83.3% to 87.8% of the trial spent within the correct lane) despite the 3.7 log unit increase in luminance.
Unlike the other manipulations, tunnel vision severely degraded steering performance. On average, drivers were within the correct lane during only 24% of the tunnel vision trial; a t-test revealed that this effect was significant, t(9) = 18.7, p < 0.001, and that the effect of tunnel vision accounted for 98.4% of the variability in performance. Separate analyses revealed that tunnel vision also significantly affected the number of lane excursions, t(9) = 4.4, p = 0.002, standard deviation of the vehicle's lateral position, t(9) = 9.4, p < 0.001, and lateral speed, t(9) = 7.6, p < 0.001.
In order to increase driver workload, participants were instructed to verbally report the presence of the stationary pedestrians who had been positioned along the shoulders of the roadway. Detection performance was relatively robust to blur levels of up to + 2D, when a mean of 98% of the pedestrians were detected. At higher levels of blur, however, pedestrian detection performance was degraded, with a mean of only 55% of the pedestrians detected with maximum blur. An ANOVA revealed that blur significantly reduced their ability to perform this task, F(4,36) = 28.3, p < 0.001 and that blur accounted for 76.9% of the variability in performance. Similarly, pedestrian detection was relatively robust to moderate luminance challenges. At the two highest luminance levels all participants detected 100% of the pedestrians, while at the two lowest luminance levels participants detected a mean of only 90% and 55% of the pedestrians, respectively. The effect of luminance on detection performance was significant, F(3,27) = 20.7, p < 0.001, and accounted for 71.7% of the variability in performance. A mean of only 17% of the pedestrians were detected in the tunnel vision condition, t(9) = 23.7, p < 0.001, an effect that accounted for 99.0% of the variability in performance.
Participants had been instructed to maintain a speed of 89 km / hr (55 mph) during all trials. They were largely successful; mean speeds ranged from a minimum of 79.2 km/hr (49.2 mph) in the tunnel vision trial to a maximum of 89.5 km/hr (55.6 mph) in the highest luminance trial and in the + 1 D blur trial.
Ten new university students (M = 18.5 years) with a minimum of two years of driving experience (M = 3.1 years) received extra credit in a psychology course in exchange for participating in this study. None of the participants had previous experience with the driving simulator and none wore glasses or reported any history of visual problems. The participants' binocular acuity ranged from 20/33 (6/10) to 20/13 (6/4) and averaged 20/17 (6/5). All procedures were consistent with the Declaration of Helsinki and were approved by the Clemson University Institutional Review Board.
Because the same simulator described in experiment 1 was used for experiment 2, we describe only changes from the methodology used in experiment 1.
Rather than having an experimenter read the speed aloud every 20 s, an automated procedure was used. Whenever the speed was below 80 km/hr (50 mph), the simulator presented audio clips telling the participant to “increase your speed.” Whenever the speed exceeded 97 km/hr (60 mph), an audio clip instructed drivers to “slow down.”
Relative to experiment 1, an additional training element was used during the first and second training trials of experiment 2. In order to compensate for the lack of kinesthetic and auditory feedback when one leaves the lane, in these practice trials the drivers were encouraged to move around within the lane to find the lane's boundaries. When either edge of the vehicle extended beyond an edge line, either towards the shoulder or the oncoming lane, a large red “out of lane” message appeared on the center screen. This message disappeared when the vehicle re-entered the lane.
Ten, rather than six, stationary pedestrians were positioned quasi-randomly on the shoulder of the road (5 on the right and 5 on the left) in all trials.
Field of view was manipulated by having drivers use either a full binocular field, full monocular field, or circular monocular fields of 46, 23, 11, 3.4 or 1.7°. For the monocular conditions drivers always used their left eye; the right eye was occluded with an eye patch. Visual fields were restricted by mounting aluminum cylinders to trial frames; blinders placed on the sides of the frames prevented the use of peripheral vision. All trials were tested at maximum luminance (maximum scene luminance = 16.7 cd/m2).
The experiment took place in a single session that lasted 90 to 120 min. Each participant was introduced to the simulator via a two-minute training trial that used a straight road, the “out of lane” indicator, and pedestrians. Each subsequent training and experimental trial lasted five minutes. The second practice trial used a curvy road, the “out-of-lane” indicator, and pedestrians. The third practice trial used a curvy road, pedestrians, and an experimenter who would periodically steer the vehicle off the road, thus giving drivers an opportunity to learn to recover from gross steering errors. The auditory speed indicators were introduced in the fourth and final training trial. During this trial the participants were instructed to modulate their speed in order to get familiar with the auditory feedback.
The participants could repeat each practice trial as many times as necessary until he or she felt comfortable with the driving simulator. Before each subsequent trial the drivers were reminded of their three goals: 1) to keep the vehicle centered in their lane, 2) to maintain a speed of 89 km/h (55 mph), and 3) to detect and announce the presence of all pedestrians. All practice trials and experimental trials were be followed by a short break. Once the driver was both comfortable with and proficient at using the simulator, the seven field of view conditions were tested in a random order.
The simulator data were again sampled at 5 Hz, the first 30 s of each 5-min trial were ignored, and the primary measure of steering performance is the percentage of time (out of 270 s) during which the entire vehicle was within the designated lane.
Steering and pedestrian detection data are presented in Table 2; time-in-lane data are also presented in Fig. 4.
The percentage of time spent entirely within the desired lane ranged from 98.7% with a full binocular field to 28.8% with the smallest monocular field of view (1.7°). A repeated measures ANOVA revealed that the effect of field of view was significant, F(6,54) = 135.0, p < 0.001, and that field of view accounted for 93.9% of the variability in performance. Separate ANOVAs revealed that the effect of field of view significantly affected the number of lane excursions, F(6,54) = 28.2, p < 0.001, the standard deviation of the vehicle's lateral position, F(6,54) = 50.0, p < 0.001, and lateral speed, F(6,54) = 48.2, p < 0.001.
Participants had again been instructed to maintain a speed of 89 km/hr (55 mph) during all trials. Mean speeds ranged from a minimum of 82.4 km/hr (51.2 mph) in the 1.7° condition to a maximum of 90.9 km/hr (56.5 mph) in the 11° condition.
At all field sizes of 11° and above, the participants detected at least 90% of the pedestrians. But only 57% and 12% of the pedestrians were detected at the 3.4° and 1.7° conditions, respectively. An ANOVA revealed that the effect of field size was significant, F(6,54) = 145.8, p < 0.001, and that the effect of field size accounted for 94.3% of the variability in detection performance.
Two experiments investigated the extent to which the ability to steer a vehicle depends on having a clear, large, and high luminance visual scene. Visually healthy young participants drove at approximately 89 km/h (55 mph) along a curvy simulated roadway while experiencing severe blur, luminance, or field size challenges. In general, steering performance was remarkably robust to the blur and luminance challenges. Despite the fact that during the blur manipulation drivers in experiment 1 experienced up to ten diopters of blur, which caused acuity to drop to 20/647 (6/197), steering performance dropped only modestly, from spending a maximum of 95% of the no blur trial in the lane down to a minimum of 88% of the + 5 D trial in the lane. With double the blur (+10 D), steering performance was no worse, with drivers spending 90% of the trial within the lane. Similarly, despite the large (3.7 log unit) reduction in luminance (from 16.7 cd/m2 down to 0.003 cd/m2) that caused acuity to drop to 20/178 (6/54), steering performance again dropped only modestly, from 88% of the trial spent within the intended lane at maximum luminance down to 83% at minimum luminance. Calculation of effect sizes, which describe the proportion of the variability in steering performance that can be explained by the manipulations, confirmed that both the blur and luminance manipulations had relatively small effects on steering performance (accounting for 27.6% and 23.8% of the variability, respectively) but had very large effects on visual acuity (accounting for 94.1% and 94.6% of the variability, respectively). Although visual acuity is highly sensitive to blur and to low luminance images, visually guided steering is remarkably robust to these same challenges.
In contrast, steering performance (but not acuity) is quite sensitive to large reductions in visual field. When drivers in experiment 1 looked through a cone that limited their field to 1.7° (monocular) they spent on average only 24% of the trial within their intended lane despite the fact that their acuity was nearly at optimal levels (20/23; 6/7). Experiment 2 elaborated on this effect by testing seven field sizes ranging from a full binocular field down to a 1.7° monocular field. Results from this study revealed that steering performance tended to be strong (i.e., above 80% in lane) until monocular fields were reduced below 11° but dropped precipitously when field size was reduced further. Effect size calculations confirmed that the visual field manipulation had a large effect on steering performance, accounting for 98.4% and 93.9% of the variability in experiments 1 and 2, respectively.
In general, the present findings are consistent with earlier studies of the dependence of steering on blur,3 luminance,5–7 and field size.8–11 Because it used a high fidelity driving simulator, however, the present study could incorporate more severe visual challenges than these earlier studies. Thus the present data provide a stronger test of the extent to which steering is robust to blur, low luminance, and restricted visual fields.
While the present data document the effects of severe visual challenges on the steering performance of young visually healthy drivers, the experiments were not designed to quantify the effects of true visual field defects or pathology-induced acuity decrements on real-world driving performance. The visually healthy drivers in the present studies experienced a sudden onset of the visual challenges and did not have months or years to develop compensatory adjustments or strategies. Similarly, the scenario used in the driving simulation was designed to challenge the drivers' steering capacities (the roadway was unusually curvy and speed was kept relatively high), but it was not designed to simulate normal driving conditions. For example, our participants were not required to scan the whole driving scene – or to use their periphery – to detect hazards (they knew that no other traffic was present and that there were no intersections, stop signs, or stop lights) and therefore were able to concentrate primarily on the steering task. Importantly, the stationary pedestrians were the only objects that appeared along the road shoulders. The pedestrian detection task was used to increase the participants' workload, and while performance on this task is informative regarding the effects of the severe visual manipulations, the pedestrian detection task was not intended to reflect real world conditions.28
The present study was similar to that of Owens and Tyrrell4 both in terms of its methodological approach and in terms of the severity of the visual challenges that were used. Unlike the present study, however, the Owens and Tyrrell study relied upon a low fidelity simulator in which participants rotated a small handheld knob to adjust the flow of a field of white vertical posts that simulated a constant forward speed. Participants in the present study, however, used a steering wheel (albeit without force feedback) to maintain their lane position within a larger and richer display while working to maintain their own speed (using brake and accelerator pedals) and while engaging in a pedestrian detection task. Despite the methodological differences between these studies, however, their findings are quite consistent. In both sets of studies steering was found to be sensitive to a reduced field size but not to blur and luminance challenges. Acuity, on the other hand, was sensitive to blur and luminance challenges but not to reduced field size. This pattern suggests that the visual mechanisms that support steering are largely independent of those that support the ability to resolve fine detail.
The notion that acuity and steering are supported by separate visual mechanisms is not new. As far back as 1967 researchers began to appreciate a neural distinction between two different visual systems, one that supports the ability to recognize details and objects – focal vision – and one that supports visually guided body movements – ambient vision.15–19 Since the late 1960s, considerable support has emerged for this neural distinction.20–26 In 1977, Leibowitz and Owens27 became the first to connect the idea of a neurological distinction to problems of night driving. They hypothesized that drivers are unaware of the “selective degradation” of focal visual functions (such as their ability to recognize objects and hazards at night) in part because ambient visual functions (including steering) are preserved at low luminances. In this view, drivers get little feedback that their visual abilities are degraded at night because they do not experience difficulties steering their vehicles and because many objects in the night environment are engineered to be easily seen (e.g., retroreflective road signs). Thus it is only when we encounter an inconspicuous and unexpected object (such as a darkly clad pedestrian) that we begin, too late, to realize our own visual limitations. The present experiments, together with those of Owens and Tyrrell,4 provide empirical support for the hypothesis that steering abilities are remarkably robust to low luminance. The present data are also consistent with a steering model put forth by Wilkie and Wann wherein various sources of directional information are weighed differently as luminance decreases.29–30
The finding that unlike visual acuity, the ability to steer is relatively insensitive to luminance supports a key element of Leibowitz and Owens' selective degradation hypothesis. We are currently testing the other element of the hypothesis – that drivers fail to appreciate the extent to which their ability to see objects is degraded at night.31 Future research should also explore the steering capabilities of other populations of drivers, including elderly drivers and those with visual pathologies and in the face of even more severe visual challenges. It would also be useful to continue exploring whether other visually based driving skills (e.g., gap perception, following behavior) are as robust to visual challenges. Future researchers should also track how steering performance is affected by haptic feedback from the steering wheel and how drivers' gaze patterns are affected by reductions in acuity and visual field size.32–34
The authors are grateful for the contributions of Robert Isenhower, Nathan Klein and Jordan Addison for their assistance with data collection, Fred Switzer and Mike Ashmore for programming assistance, and for the helpful comments of two anonymous reviewers. Johnell Brooks was supported by a Dwight D. Eisenhower Graduate Transportation Fellowship. Talissa Frank is currently at SA Technologies, Marietta, GA.
Johnell O. Brooks,
418 Brackett Hall,
Dept. of Psychology,
Clemson, SC 29634-1355;
1.Owens DA, Helmers G, Sivak M. Intelligent vehicle highway systems - a call for user-centered design. Ergonomics 1993;36:363–69.
2.Sivak M. How common sense fails us on the road: contribution of bounded rationality to the annual worldwide toll on one million traffic fatalities. Transport Res F: Traf Psychol Behav 2002;5:259–69.
3.Higgins KE, Wood J, Tait A. Vision and driving: selective effect of optical blur on different driving tasks. Hum Factors 1998;40:224–32.
4.Owens DA, Tyrrell RA. Effects of luminance, blur, and age on nighttime visual guidance: a test of the selective degradation hypothesis. J Exp Psychol-Appl 1999;5:115–28.
5.Steyvers FJ, de Waard, D. Road-edge delineation in rural areas: effects on driving behaviour. Ergonomics 2000;43:223–38.
6.Kallberg, VP. Reflector Posts – Signs of Danger? Transportation Research Record No. 1403. Washington, DC: Transportation Research Board; 1993.
7.McKnight AS, McKnight AJ, Tippetts AS. The effect of lane line width and contrast upon lanekeeping. Accid Anal Prev 1998;30:617–24.
8.McKnight AJ, Shinar D, Hilburn B. The visual and driving performance of monocular and binocular heavy-duty truck drivers. Accid Anal Prev 1991;23:225–37.
9.Johnson CA, Keltner JL. Incidence of visual field loss in 20,000 eyes and its relationship to driving performance. Arch Ophthalmol 1983;101:371–5.
10.Wood JM, Troutbeck R. Effect of restriction of the binocular visual field on driving performance. Ophthal Physiol Opt 1992;12:291–8.
11.Wood JM, Troutbeck R. Effect of visual impairment on driving. Hum Factors 1994;36:476–87.
12.Owens DA, Francis EL, Leibowitz HW. Visibility distance with headlights: a functional approach (SAE Technical Paper Series, No. 890684). Warrendale, PA: Society of Automotive Engineers; 1989.
13.Bailey IL, Lovie JE. New design principles for visual acuity letter charts. Am J Optom Physiol Opt 1976;53:740–5.
14.Ferris FL III, Kassoff, A, Bresnick GH, Bailey I. New visual acuity charts for clinical research. Am J Ophthalmol 1982;94:91–6.
15.Held R. Dissociation of visual function by deprivation and rearrangement. Psychol Forsch 1968;31:338–48.
16.Held R, Ingle D, Schneider G, Trevarthen C. Locating and identifying: two modes of visual processing. A symposium. Psychol Forsch 1967;31:42–43.
17.Ingle D. Two visual mechanisms underlying the behavior of fish. Psychol Forsch 1967;31:44–51.
18.Schneider GE. Contrasting visuomotor functions of tectum and cortex in the golden hamster. Psychol Forsch 1967;31:52–62.
19.Trevarthen CB. Two mechanisms of vision in primates. Psychol Forsch 1968;31:299–348.
20.Boussaoud D, di Pellegrino, G, Wise SP. Frontal lobe mechanisms subserving vision-for-action versus vision-for-perception. Behav Brain Res 1995;72:1–15.
21.Goodale MA, Humphrey GK. The objects of action and perception. Cognition 1998;67:181–207.
22.Goodale MA, Milner AD. Separate visual pathways for perception and action. Trends Neurosci 1992;15:20–5.
23.Livingstone MS, Hubel DH. Psychophysical evidence for separate channels for the perception of form, color, movement, and depth. J Neurosci 1987;7:3416–68.
24.Mishkin M, Unterleider LG, Macko KA. Object vision and spatial vision: two cortical pathways. Trends Neurosci 1983;6:414–17.
25.Stoerig P. Varieties of vision: from blind responses to conscious recognition. Trends Neurosci 1996;19:401–6.
26.Ungerleider LG, Mishkin M. Two cortical systems. In: Ingle, DJ, Goodale MA, Mansfield RJW, eds. Analysis of Visual Behavior. Cambridge, MA: MIT Press;1982:549-86.
27.Leibowitz HW, Owens DA. Nighttime driving accidents and selective visual degradation. Science 1977;197:422–3.
28.Wood JM, Tyrrell RA, Carberry TP. Limitations in drivers' ability to recognize pedestrians at night. Hum Factors 2005;47:in press.
29.Wilkie RM, Wann JP. Driving as night falls: the contribution of retinal flow and visual direction to the control of steering. Curr Biol 2002;12:2014–7.
30.Wilkie R, Wann J. Controlling steering and judging heading: retinal flow, visual direction, and extraretinal information. J Exp Psychol Hum Percept Perform 2003;29:363–78.
31.Brooks JO, Tyrrell RA, Wood JM, Stephens BR, Stavrou EP. Comparing estimated and actual visual acuity at high and low luminance. Presentation at the Vision Sciences Society Annual Meeting, Sarasota, Florida, May 6-11, 2005.
32.Land MF. Predictable eye-head coordination during driving. Nature 1992;359:318–20.
33.Land MF, Lee DN. Where we look when we steer. Nature 1994;369:742–4.
34.Land M, Horwood J. Which parts of the road guide steering? Nature 1995;377:339–40.