Much research on driver attention, including evaluations of in-car equipment, at least implicitly assumes that attention is where the gaze is. Research on the dynamics of visual attention, however, suggests that drivers may use peripheral vision and that they learn its use over time, depending on the task demands and eccentricity. To investigate effects of task load and position on lane keeping, 11 novices and 16 experienced drivers were asked to drive along a straight road using only peripheral vision for lane keeping while doing another task foveally. The task varied in position and in mental load, with two difficulty levels in each of two different tasks. In the visual attention tasks, position had a clearly different effect on lane-keeping performance among novices and the experienced, as measured by the distance covered before crossing a lane boundary. Novices' performance deteriorated with the foveal task at near periphery at the speedometer level, whereas the performance of experienced drivers dropped only when the foveal task was down in the middle console. The result supports the hypothesis that novices need foveal vision at first for lane keeping but, with increasing practice, learn to manage with more peripheral vision. In the arithmetic tasks, however, no consistent experience-dependent task position effects occurred. Different results obtained for different tasks imply that when evaluating in-car facilities, the task characteristics and the respective resource allocation needed should be taken into account.
Requests for reprints should be sent to Heikki Summala, Department of Psychology, Traffic Research Unit, P.O. Box 11, 00014 University of Helsinki, Finland; firstname.lastname@example.org.
Visual information is essential for a driver to maintain lateral and longitudinal control and to avoid hitting obstacles, other vehicles, cyclists, and pedestrians. Beginners are taught that they should keep their hands on the wheel and their eyes on the road while also scanning adjacent lanes and the rearview mirrors. If the driver does not look at the road, people assume he or she is taking a risk. However, with practice, drivers' task becomes overlearned and easy, and, except under certain difficult conditions, they have "spare capacity" available (Brown & Poulton, 1961; Harms, 1991; N„„t„nen & Summala, 1976). Consequently, they tend to share time and attentional resources among driving itself and other tasks inside and outside the car (Antin, Dingus, Hulse, & Wierwille, 1990; Hughes & Cole, 1986; Senders, Kristofferson, Levinson, Dietrich, & Ward, 1967; Summala, 1994).
New technology has brought more equipment into the car-- among other things, mobile offices--to compete ever more with the driving task for drivers' visual and attentional resources (Fairclough, Ashby, & Parkes, 1993; Rockwell, 1988; Rumar, 1988). Much emphasis is being put on evaluating the safety and usability of these devices. Thus design criteria have been proposed, based mainly on the glance length and frequency required to perform certain tasks (e.g., Wierwille, Antin, Dingus, & Hulse, 1988; Zwahlen, Adams, & DeBald, 1988). It may not be sufficient, however, to consider time sharing as such--for example, by measuring glance length at the road and at in-car tasks, respectively. Although time-off-the-road is a measure of high face validity, simple eye-movement analysis and calculation of glance times leaves open the crucial question of where attention is focused when the driver is fixating at an in-car task (e.g., Luoma, 1988; Stapleton, Ward, & Parkes, 1993).
Recent research on the dynamics of visual attention shows that people can shift attention quickly within fixations, including the periphery (Engel, 1971; Kr"se & Julesz, 1988; Saarinen, 1993a; Saarinen & Julesz, 1991; Sagi & Julesz, 1986). Thus drivers may cope with dual-task settings by dividing attention serially within the visual field, as during a glance (a series of fixations) at an in-car task. It is even possible that experienced drivers use automated parallel routines based on the peripheral view.
An old tenet in the traffic safety community states that drivers learn to use peripheral vision in lane keeping, whereas beginners need focal vision for it; this was hypothesized by Mourant and Rockwell (1972) on the basis of their eye movement studies. They showed that novices' fixations scatter more and also cover lane boundaries, whereas with increasing practice, drivers' fixations focus more at the vanishing point on the horizon, at a greater distance from the vehicle. This hypothesis has not been confirmed, however (see Miltenburg & Kuiken, 1991).
Nevertheless, there are obvious limits in the use of peripheral vision. Hella (1987) showed that although drivers are able to control lateral position while reading a display mounted in the parafoveal field of vision, at the lower edge of the windshield, or at speedometer level, the lateral scatter starts growing when the display is mounted at the top of the windshield or down in the middle console. In the case of obstacle avoidance, and even movement detection, thresholds and response latencies also grow with the eccentricity of stimuli because of the inferior performance of peripheral vision (McKee & Nakayama, 1984; Saarinen, 1993b; Virsu, N„s„nen, & Osmoviita, 1987) and because of the eye and head movements required at large eccentricities (Bartlett, Bartz, & Wait, 1962; Sanders, 1963).
People's attention capacity is both limited and task dependent. The useful visual field gets narrower with increasing mental load (e.g., Ball & Owsley, 1991; Ikeda & Takeuchi, 1975; Miura, 1990; Sanders, 1970), and attention sharing in a dual-task paradigm is dependent on the type of resources required for each task (Burrus, Dewing, & Hancock, 1994; Matthews & Sparkes, 1993; N„„t„nen, 1992; Wickens, 1984). In studying driver performance during in-car tasks, therefore, one should take into account both the position (eccentricity, or the visual angle from the normal sight axis) and the mental load of the task, the latter consisting of the type and magnitude of the resources needed.
This study measures the effects of the in-car task's mental load and eccentricity on lane-keeping in real-life settings, with an emphasis on driver experience. Rather than permitting participants to share time freely between the road and the in-car task, we asked them to focus on the in-car task continuously in order to measure lane-keeping performance based on peripheral vision only. In this "forced peripheral driving" paradigm (see Bhise and Rockwell, 1971) different task positions enforced various levels of eccentricities.
The traffic research unit's instrumented car (a European compact 1988 Lada Samara) equipped with three video cameras (Panasonic WV-CD2) was used in this study. One camera recorded the road scene in front of the car and two were installed to record the driver's face: one in front of the driver and the other on the central line of the car to record glances away from the in-car task. The three pictures were mixed into the same videoscreen, overlaid with VGA graphics with digital data from the sensors in the car and controls, and stored on a videotape. The use of controls, speed, and three accelerations (x, y, z) were also stored in a computer file at 15 Hz.
For the in-car tasks, a seven-segment LED display (size 7 13 mm, horizontally centered in a grey box 100 50 mm, 8 mm from the upper edge) was used, controlled by an independent personal computer. Three positions were used (see Figure 1). The first was just above the dashboard, 8 cm down from the average driver's eye position and at an eccentricity of approximately 7 straight down from the sight axis between the eyes and vanishing point on the horizon. The second was at the level of the speedometer, 25 cm down and 8 cm right from the average eye position, at an eccentricity of 23 approximately on the 162th meridian. The third was on the midconsole of the car, where the radio and other accessories are installed, 34 cm down and 36 cm right from the average eye position and at an eccentricity of approximately 38 on the 133th meridian.
In-Car (Foveal) Tasks
Four in-car tasks were used in different trials. Each task was performed for random digits (1-9) that were presented on the LED display and required an oral response (see Table 1). In the two attentional tasks, participants either had to name aloud each digit presented at intervals of one s (easy task) or to name all the fours in a random series of digits presented four times per second (difficult). In the two arithmetic tasks, participants either had to add a constant (three) to the digit shown (easy) or to add the last-shown digit to the former one (difficult). In the arithmetic tasks, the interstimulus interval was set to 2 s. Therefore, both task types required viewing of the display, but the latter required additional mental operations.
Participating were 27 conscripts from the Helsinki Anti-Aircraft Regiment, aged 18 to 22. All had recently applied for but not taken special training as military truck drivers. Participants were divided into two groups according to their driving experience. The 11 novices had a total self-reported driving experience of less than 5000 km (median 1500 km; one participant later claimed to have driven 6000 km). The 16 experienced drivers had a total driving experience of at least 30 000 km (median 50 000 km, range 30 000-120 000 km; one participant claimed to have driven 400 000 km, which may be exaggerated). The novice group was a little older than the experienced group, with a mean age of 20.3 years (SD = 0.87 years) versus 19.8 years (SD = 0.85 years). The mean license ages were 0.6 and 1.6 years, respectively.
The vision of all participants met the licensing requirements (minimum visual acuity 0.6, minimum visual field 140 ), with neither group differing from the other. The groups did not differ in average eye height when sitting in the driver's seat.
Experiments were conducted on a straight stretch of paved road in a closed military area. The 10-cm, white, continuous lane boundaries were painted for this experiment to provide a lane width of 3 m. Participants' task was to accelerate from stop to 30 km/h within 50 m and to drive a segment of 210 m using only peripheral vision for lane-keeping while doing the other task foveally.
Each participant performed the run 12 times, doing each of the four tasks at each of the three display positions. The order of tasks and positions was balanced across participants.
The instruction for each task, with a corresponding example, was given in the stationary car. After having accelerated to a steady 30 km/h, the participant was prompted by the experimenter at the beginning of the test segment to shift his gaze to the display and to start responding to the stimuli according to the task. He was instructed to keep his gaze on the display, keep the car in its lane, and keep its speed constant while doing the task until the experimenter, at the end of the test segment, prompted him to shift his gaze back to the road.
Before the experiment, each participant had done 30 km of highway driving on two-lane arterials and minor roads to adapt to the car. No practice run of the test itself was done, however.
Video recordings of the points of lane crossings and occasional glances at the road were analyzed for each run. The primary measure of lane-keeping performance was the proportion of the distance (out of 210 m) that the participant could drive before crossing a lane boundary. This was the reason that we did not run a control condition with no in-car task; successful lane keeping was self-evident, and participants' ability to keep within the edge lines without an auxiliary task had also been checked on the highway route.
The average distance cleared (as a percentage of the whole track) ranged from 46.1% to 98.3% among novices (mean 84.3%, SD17.0%) and from 59.6% to 100% among experienced drivers (mean 91.8%, SD 12.1%), indicating substantial overlap between the two groups in lane-keeping performance across all the tasks and positions. The two groups did not show different learning during the task; repeated-measures multivariate analysis of variance, or MANOVA; Trial Order Experience Group interaction; successive trials averaged in groups of four trials.
Figure 2 shows the average distance cleared by task and display position for novice and experienced drivers separately. Lane-keeping performance appears to decline with the increasing eccentricity of the display. This decline differs between novices and experienced participants. In a MANOVA with repeated measures, the main effect of display position, Rao's f-approximation and the Position Driving Experience interaction showed statistically significant effects on the distance cleared. Neither task, driving experience, nor their interaction showed any significant effects on lane keeping.
The experience and position effects occurred only in the attention tasks, however. The two attention tasks taken together showed a marked position effect and a Position Experience Interaction; neither task, driving experience, nor their interaction showed any significant effects on lane keeping. The paired comparisons of the differences between positions showed that reduction in lane-keeping performance from Position 1 to Position 2 (from windshield to speedometer) was consistently larger among novices than among experienced participants, both for Task 1 and Task 2. Instead, for the two arithmetic tasks, neither position, task, driving experience, nor their interactions had any significant effects on the distance cleared. The check for paired-position comparisons showed no group difference between Positions 1 and 2 in either task.
The results therefore showed that in the attentional tasks, lane-keeping performance declined with eccentricity. Among novices this impairment occurred at a smaller eccentricity (with the display at the speedometer level) than it did among experienced participants.
Number of Glances
Despite having been instructed to keep their gaze on the display, participants did not fully succeed but, as some put it, were "forced" to glance at the road sometimes.
Figure 3 shows that the average number of glances for the test segment consistently increased with task eccentricity for both novices and experienced drivers. The position and task main effects were clearly significant, but neither the experience effect nor any interaction were significant. (Note that in separate MANOVAs, the degrees of freedom differ somewhat because of the lack of observations.)
The analysis of first glances at the road coincided well with the average number of glances over the entire test segment. The MANOVA with repeated measures showed a significant position and task effect on the distance cleared before the first glance at the road, but neither the main effect of experience nor any interaction was significant. A significant Position Experience interaction appeared instead when considering the first of two different "failures" in the peripheral lane keeping--that is, either the first glance at the road or crossing a lane boundary-- supporting the earlier analysis of lane performance.
Secondary Task Performance
Unfortunately, we could not measure the secondary task performance because of the failure to store the random stimulus series shown to participants. The experimenter recorded only the total number of fours and the missed responses for Attention Task 2.
Figure 4 shows an increasing trend in the percentage of missed items with increasing eccentricity; however, because of large variance, neither the position effect nor the main effect of driving experience reached statistical significance. It is important to note, however, that the increasing trend of missed items with eccentricity was similar in both groups: F(2, 24) = 0.12 for the Position Experience interaction, p > .8, in contrast with the Position Experience interaction in lane-keeping performance.
The mean speed computed over the entire test segment and over all the tasks was 33.7 km/h (SD = 2.60), for novices and 32.2 km/h (SD = 2.69), for experienced drivers. The repeated- measures MANOVA showed neither significant main effects--for experience--nor interactions of experimental variables.
The attention tasks showed a consistent interaction between task position and driving experience. Novices' lane-keeping performance was already deteriorating with the task at the level of the speedometer (23 from the normal sight axis), whereas experienced drivers' performance was impaired only when the task was low on the middle console (38 from the normal sight axis). The lack of corresponding interaction in glance frequency as well as in secondary task performance (available only for Task 2) indicates that the result is not attributable to a different time-sharing strategy or a different trade-off between driving and other tasks in the two groups. The fact that the experience effect was almost the same in the two attention tasks, although they differed substantially in task load, further suggests that the superior "peripheral" lane-keeping ability in our experienced participants was based on an automated, parallel routine rather than on serial attention sharing in the visual field.
The result is in accordance with the hypothesis of Mourant and Rockwell (1972), based on their eye movement registrations, that novices first need foveal vision for lane keeping but, with increasing practice, learn to manage with more peripheral vision. These data suggest that a prominent change appears between the first 1500 and 50 000 km of driving.
Note, however, that the experience effect was not consistent at the smallest eccentricity of 7 , though this position is closest to that normally kept in driving and therefore most practiced. That position appeared to be easy enough for our novice participants as well. Actually, the display location was just above the lower edge of the windshield, and, bearing in mind that novice drivers especially may fixate somewhat lower than at the vanishing point in the horizon at eye height, Position 1 could provide an even smaller deviance from the normal gaze direction.
It was the second position, at the speedometer level of 23 from the sight axis, that showed the largest difference between the groups, indicating that the more efficient use of peripheral vision among experienced drivers is extended to cover a somewhat larger area. Problems arose among experienced drivers when the display was located down in the midconsole, confirming the results of Hella (1987); in that position the difference between novices and experienced drivers actually decreased. It can be assumed that this is largely attributable to a ceiling effect: The entire track was only 210 m, and the task lasted only 25 s at 30 km/h, so it could be possible to cover a long distance in a straight stretch even without visual information.
Considering the experience effects, however, it is not sufficient to speak only of learning to use peripheral vision rather than foveal vision in lane keeping. Practice at the wheel means improved use of visual (as well as haptic and kinesthetic) information, including the more efficient use of different cues, whether they be based on foveal or peripheral information. Thus in lane keeping, Smiley, Reid, and Fraser (1980) found an early change in the steering control strategy among beginners during the first days of driving, and Riemersma (1987) found that experienced drivers use lateral speed as an effective visual cue for straight-road driving but that inexperienced drivers do not. Therefore, more efficient cues learned with practice make the use of peripheral vision easier either because the task gets generally more simple or because the cues are better suited to the periphery.
Although a distinction is traditionally made between focal and ambient processing in the visual system (Leibowitz & Owens, 1977; Leibowitz, Post, Brandt, & Dichgans, 1982) and between a serial focal attention task and a feature-gradient task (for perceiving salient boundaries and singularities across the entire visual scene; see Braun & Sagi, 1990), the present results indicate that the use of peripheral vision in driving can be learned through practice (see Engel, 1971; Johnson & Leibowitz, 1975). What is more, practice is presumably required for parallel, ambient processing. Conversely, these results support the view that driving is learned by driving: The first 30 000 km- -the minimum for our experienced group--equals several hundred hours at the wheel (Summala, 1987).
The present results showed that task position (eccentricity) effects largely depended on the task, so that they occurred in the attention but not in arithmetic tasks. At first sight this appears to be in line with the multiple-resource theory (Wickens, 1980, 1984): The arithmetic tasks allow use of spare visual capacity in the lane-keeping task. However, it should be noted that more time was available in the arithmetic tasks than in the attention tasks, and this alone could explain the result.
This study showed that keeping the task load constant appears to be a problem in this kind of experiment. The easy arithmetic task (add a constant to the digit shown) might have approached the easy attention task in having a low resource demand for many participants, whereas the other arithmetic task was difficult enough to make participants in both groups drop out of their task for a while in order to allocate plenty of attentional resources to the road.
We saw clear differences in lane-keeping performance between the groups, but only in certain well-defined conditions. In the experiments allowing free time sharing, however, novices and experienced drivers differed neither in the average glance times at the in-car tasks nor in the control of the lateral position (as a function of the time of gaze off the road or of oncoming traffic; Nieminen & Summala, 1994).
Even more generally, drivers show consistent time-sharing patterns (Rockwell, 1988, Wierwille, 1993). Recall that these experiments have typically been made under supervision, and when maximum performance is not being measured, the bias caused by the presence of the experimenter may limit glances at in-car tasks to below those in normal driving. To estimate safety problems of time sharing veridically, unobtrusive data on time sharing in normal, unsupervised conditions are urgently required.
Note that learning an automated (more peripheral) routine of keeping the car in a lane, which probably coincides with (or incorporates) automatization of steering control, means a profound change in the entire task. Parallel lane keeping releases resources for obstacle detection and other driving subtasks (see, e.g., Summala, 1995), and although overt behavior may remain the same--at least in a supervised experiment--an experienced driver with his or her expertise is actually performing another task and is doing it more safely.
The development in lane-keeping performance found in this study coincides well with a substantial improvement in terms of accident involvement during the first 50 000 km (Summala, 1987). Novices' inferior performance challenges the design and safety standards of both traditional and new in-car devices. Rather than waiting for intelligent driver support, with its many technical and behavioral problems (Piersma, 1993; Schumann, Godthelp, Farber, & Wontorra, 1993; Summala, 1994), further research is also needed to reveal the more efficient mechanisms that drivers appear to learn by driving, and how learning of these mechanisms could be accelerated in driver curricula.
Heikki Summala is professor of traffic psychology at the University of Helsinki, Finland. He received his PhD in experimental psychology from the University of Helsinki.
Tapio Nieminen is a graduate student in psychology at the University of Helsinki.
Maaret Punto is a graduate student in psychology at the University of Helsinki.
[Extracted from HUMAN FACTORS: THE JOURNAL OF THE HUMAN FACTORS AND
ERGONOMICS SOCIETY, Volume 38, Number 3, September 1996. Copyright 1996 by
the Human Factors and Ergonomics Society, P.O. Box 1369, Santa Monica, CA
90406-1369 USA; 310/394-1811, fax 310/394-2410, email@example.com,