As discussed in the introduction, the virtual environments based driving simulator should create a realistic driving experience for the user. This includes an immersive virtual environment and a realistic simulation of the driving task. The realism of the simulation involves the perception of the user and the response of the simulation in driving tasks. Immersion and the perception of simulator realism are primarily subjective phenomena, while the response of the simulator can be measured quantitatively and compared with driving task studies conducted in other simulators and in real vehicles.
Accordingly, the simulator was used to conduct two studies which involved quantitative measures of driver performance and a qualitative assessment of the driving experience. The studies analyzed the effects of three variables on the perceived realism of the immersive driving environment and the performance of driving tasks: the shading algorithm used to render the image, the display device, and visual cues in the form of roadside delineation poles.
The first study examined the effects of the display variables in a simple road-following task. Subjects drove through a series of road courses, and data were collected on the lateral lane position and speed of the driver's vehicle. The subjects then answered a questionnaire designed to elicit a qualitative assessment of the effects of these display variables on the immersive quality of the simulation. Subjects filled out the questionnaire after they had driven the simulator. The second study concentrated on the effects of different levels of roadside delineation on speed estimation. Here, the subjects were not asked to drive the simulator, but rather to be "passengers" in the driver's seat of a moving vehicle. The subjects estimated the vehicle velocity.
The first variable, rendering type, compared flat shaded images to wireframe rendering of the same road environment. This variable addresses the role of texture in drivers' visual perception and in the perceived realism of the virtual environment. Some studies suggest that visual motion cues depend more on the size and shape of visual elements than on their texture (Grunwald and Kohn, 1993; Groeger and Brown, 1988). If this is true, the shading algorithm should have no significant effect on the performance of driving tasks.
The effect of shading on the perceived realism of the virtual environment depends on the relative importance of visual quality relative to interactivity in inducing a sense of immersion. Intuitively, a simulation with more "real" (i.e. shaded) graphics should seem more real, and induce a stronger sense of immersion and simulator realism. The role of shading has important implications for virtual environment development. A wireframe image should require less computing power to render. A simulator with wireframe graphics can run at a faster update rate, or be used on a less-powerful (and less-expensive) computer. In the system used for these studies, the simulator ran about one frame per second faster with wireframe graphics than with flat shaded images.
To test the effect of the display device, subjects viewed the virtual road environment using two different display systems: an HMD with head tracking and a color monitor without head tracking. The HMD, as described in Chapter 2, had a 60° horizontal field of view, compared with 30° for the monitor. In addition, the coupling of the display in the HMD to head motion affords the ability to scan the visual environment in almost all directions. The HMD resolution of 430x133 pixels was less than the 640x480 pixel resolution of the monitor. The wide field of view and visual scanning ability of the HMD and head tracker should create a stronger sense of immersion than the monitor, and may lead to better performance in driving tasks.
The third variable, road side delineation, evaluated the effects of peripheral visual cues on the performance of both driving tasks, road-following and speed estimation. Landwehr (1991) suggested that delineation poles be used to improve the ability of drivers to maintain speed and lane position, particularly on dangerous stretches of road. Blaauw (1985) demonstrated the usefulness of roadside delineation systems in improving driving performance at night.
20 volunteer subjects drove the simulator through a simulated road course. The road model was a 1.88 mile (3000 m) flat two-lane road, consisting of straight sections and an equal number of left-hand and right-hand 90-degree curves. The model included curves of two radii: 200 ft (61.0 m) and 300 ft (91.4 m). The road was 24 feet (7.3 m) wide, divided into two 12-foot (3.7 m) lanes by a solid white line, and bounded on both sides by a 10-foot (3.0 m) wide shoulder. Figure 3-1 shows a schematic representation of the 200-ft and 300-ft curves, illustrating the two lanes, centerline and shoulders.
Data were not collected in the first 3/4 mile of the course, which allowed the user to become familiar with the simulator and the course. The portion of the course on which data were collected was 1.1 miles (1811 m) long, and consisted of seven 400-ft long straight sections, four 300-ft radius curves (two left-hand and two right hand), and four 200-ft radius curves (two left-hand and two right hand). Table 3-1 shows how the length of this portion of the course was distributed among the three sections of road.
| Section | Straight | 300-ft curves | 200-ft curves |
|---|---|---|---|
| Length | 2800 ft (853 m) | 1885 ft (575 m) | 1256.6 ft (383 m) |
The study examined three factors, as follows:
A 2x2x3 factor mixed factorial design was used, with one between-subjects factor (rendering) and two within-subjects factors (display and delineation). Each subject drove through the course six times, under the following display and delineation conditions:
The subjects were divided into two groups of ten subjects each. One group saw flat shaded rendering of the six courses, which the other group saw wireframe rendered images. The vehicle interior model was always displayed to the user, and always using flat shaded rendering. Figure 3-2 shows shaded and wireframe examples of the graphic image seen by the subjects. The examples were recorded for the HMD condition, on the course with poles 10 feet from the road edge. Each subject drove through three courses using one display type, followed by three courses with the second display type. The display type and delineation conditions were presented to the subjects in a random order.
Each subject was asked to sit in the seat, and adjust it to a comfortable position. Before beginning the study, the subject was asked to drive through the course once, to gain familiarity with the system. For this training run, the subject used the display type which would be presented first in the study, and drove through one of the courses with poles along the sides of the road.
After the training run, the subjects drove through the six test courses. They were instructed to drive in the right lane of the road, maintaining a speed as close as possible to 40 mph. In the runs with the head-mounted display and head tracking, the display was removed from the subject's head after each course. The vehicle position and speed were recorded each simulation cycle, and written to a data file. The data were only collected in the last mile of each course, giving the subjects additional opportunity to train.
In each simulation run, the position of the vehicle in world coordinates and its speed were recorded to a data file every simulation cycle. After all six runs for each subject were complete, the six data files were passed through a post-processing program, which computed the values used for data analysis.
The position of the vehicle was first converted into a tracking deviation value. The tracking deviation was defined as the lateral distance of the centerline of the vehicle from the center of the right lane. This value was used to separate the data into two groups, corresponding to driving in and out of the lane. Out of lane driving was defined as a deviation greater than 3.5 feet, or half the lane width (6 ft) minus half the width of the vehicle (approx. 2.5 ft).
For the in-lane data, the program calculated the mean and variance of the two dependent variables, tracking deviation and speed. The mean and variance were calculated for the different road types within each of the six courses: straight sections, 200-ft curves, and 300-ft curves. For the out of lane data, each time the vehicle left the lane was considered one tracking error. The number of such errors and their duration, in simulation cycles, was recorded. The total number of errors and their average were calculated for the three road types identified above.
The questionnaires filled out by the subjects were composed of 18 questions. Each subject filled out two questionnaires, one after each set of three courses. The questionnaires formed a 2x2 mixed factorial design, which evaluated the effects of rendering (between-subjects) and display type (within-subjects) on the qualitative perceptions of the subjects.
The first five questions directly asked the subjects to evaluate the realism of the driving simulation, and in some cases to compare it to real driving. The rest of the questions addressed the immersive nature of the virtual environment. These questions were taken from a more detailed questionnaire developed by Psotka (1994) to evaluate the depth of immersion created by a virtual environment. Each question had five possible answers, which could be translated to a five-point scale. The text of the questionnaire appears in Appendix B.
The procedure used for this study is based on speed estimation studies performed by Salvatore (1969). This study used a modified version of the driving simulator program. The vehicle controls were disabled, and the program simulated constant-speed driving along a straight road. The road course for all runs was a 1/2 mile long straight two-lane road, with the same dimensions as the road model used for Study 1.
The study examined two factors: four levels of road side delineation at three speeds: 25, 40, and 55 mph. The road models used for this study used the 12-foot poles from Study 1, again spaced 80 feet apart along both sides of the road. The four delineation levels were:
The horizontal poles were placed 5 feet above the road surface. Flat shaded images and the head-mounted display were used for the entire study.
Each subject viewed 12 sets of three runs each. In each set, the vehicle traveled three times down a road at one of the delineation levels, once at each of the three speeds. Six versions of this set of three could be run, each with the speeds presented in a different order. The 12 sets comprised three sets at each of the four delineation levels. The delineation levels were presented in a random order.
As in Study 1, each subject was asked to sit in the seat, and adjust it to a comfortable position. After the purpose of the study was explained, the subject was asked to drive through a road course once, at various speeds, to get acquainted with the system and the appearance of the virtual road environment. The training run used a flat-shaded course from Study 1, with poles 10 feet from the edge of the road. The subject used the head-mounted display, which was removed after the training run.
After the training run, the head-mounted display was placed back on the subject's head. The speed courses were presented one by one. The subject was asked to estimate the vehicle speed for each trip down the 1/2-mile road. If the subject did not volunteer an answer, he/she was asked for an estimate after about 5 seconds of travel. The estimates were recorded by hand. The subject wore the head-mounted display throughout the study. For analysis, the speed estimates were converted to error values, by subtracting the actual speed from each estimate.