|
|
Usability tests comparing three different virtual environment (VE) interface designs indicate that an immersive VE is more usable than two non-immersive VEs for a task with search and navigation components. Three interface designs were tried in a counterbalanced within-subjects procedure with ten randomly-ordered trials for each interface design. One of the interface designs used a head-tracked, stereoscopic head-mounted display. The other two interface designs used hand-tracking and were non-immersive -- the visual display appeared on a desktop monitor. Results for sixty participants doing the same task with each interface design show faster task completion times with the immersive design.
Virtual environments, evaluation, user studies, immersion.
© 1997 Copyright on this material is held by the authors.
Of the different kinds of systems called virtual environments, a central feature for most is the use of 3D graphics for visual output. Not all systems involve three-dimensional spatial tracking of the user. Toward one side are those that use 3D tracking for input, whether six degree-of-freedom or less, and may use a head-mounted display with head-tracking or gesture input using hand-tracking. Toward the other side are social VEs, some of which use 3D graphics and others are text-based.[2,3,5,6,7]
Given the costs of providing immersion, it is fair to ask about the benefits. Does immersion in a virtual environment increase usability when the task requires search and navigation, specifically egolocomotion?
This study placed three virtual environment interfaces in competition with each other to discover how user performance is affected by various aspects of interface design.[1] Only the human-computer interfaces differed. The underlying VE system code was the same. One comparison this allows is along the immersion dimension: performance with the immersive interface versus the two non-immersive ones.
Each participant in this experiment used three interfaces to perform a simple task ten times in succession with each interface. Since there might be some learning transfer from one interface to the next, participants used the interfaces in six different orders, for a counterbalanced study.
For all three interfaces, the structure of the virtual world was identical and the same task was used. The task includes components of search, navigation over an area with a radius of several walking paces, and navigation to fine-tune the position and orientation of the ego location within a narrow range of several inches and several degrees.
Before a series of trials, the experimenter (E) reads to the participant (S) some instructions on how to use the interface and S demonstrates understanding by testing the controls. Each task begins with S at a fixed starting position in the middle of a green wire-frame floor suspended in a black void. The only wayfinding cues are the lines of the floor grid, and its corners and edges. E clicks a button to create a target object somewhere in the virtual world, within the extent of the floor. S looks around to find the object and uses the interface's navigational affordances to approach the front of the object and identify a visual detail in its structure. The object is long and narrow, to require precision in the final approach. S must see down the length of its interior to complete the task.
After completing a task, S returns to the starting position and repeats the task with a new target object in a different position. Ten preset positions are used in random order. Some are relatively easy to find and approach. Others are more difficult, requiring a more complex sequence of rotations and translations.
The immersive interface, head-tracked walking, affords search and navigation somewhat like that experienced in natural reality: turning the head, walking around to approach the object, moving the torso and head to look into the opening.
The two non-immersive interfaces give substantially different affordances. In one, the puppet interface, S holds a hand-tracking device to simulate holding a puppet or a doll by the head to walk it around in a scaled-down virtual world. In the other design, the flying interface, S operates a metaphorical vehicle rather than a walking model. Holding a hand-tracker with buttons, S presses one button to turn the head and another button to fly around (using velocity control), but can not turn and fly at the same time.
Measurements were taken of the time to complete each of ten trials for each of the three interface designs. The graph below displays three data points for each S, showing the mean trial time in seconds separately for each interface. All three times for a S appear together on the horizontal axis. Data for the immersive design are labelled "helmet," drawn with a solid line. The non-immersive designs are "puppet" with a dashed line and "flying" with a dotted line. The Ss are sorted along the horizontal axis by ascending mean trial time with the helmet interface.
The curve for the helmet interface is almost flat, in contrast to the two non-immersive designs. For all but two Ss, mean trial time is lower for the immersive design than for the other two designs, often by a large factor.
The curves for the non-immersive interfaces show considerable variation, between Ss and within Ss. Half of the Ss are faster with the puppet and half are faster with the flying interface. Some non-immersive times are much larger than the helmet time, while others are close to it.
Of the three methods, head-tracked walking is the most like natural experience.[4] This is true despite some limitations that are likely to reduce performance. The vision system's resolution is lower in the HMD than the desktop monitor. The field of view is relatively narrow. The spatial geometry does not completely register with physical reality. The cable tether used by the tracking and display systems sometimes gets in the user's way, depending on how much each user tends to forget its entangling presence. On a positive note, latency was kept short, since it is crucial to usability. Latency was equal for all three interfaces, as they use the same virtual environment modeling and rendering code.
The other two interfaces require some learning of transfer functions that operate between the user's input and the system's control responses. The puppet interface requires projecting one's viewpoint into the hand and staying aware of the relative rotation between the user's and the puppet's coordinate systems. The flying interface requires learning the operation of the metaphorical vehicle and planning its motion components to effectively complete the task.
The helmet interface was fastest for all but two Ss. The initial hypothesis predicted that the trial times would come out in the order: helmet, puppet, and flying. The results show that order for half of the Ss. The other half are ordered helmet, flying, and puppet.
It is possible that the two non-immersive interfaces are difficult to use because of their particular designs1 and, consequently, an unfavorable comparison with the immersive interface could be expected. Also, they might get faster with practice. However, the data show that many Ss used all three interfaces with nearly equal mean trial times. If the non-immersive interfaces have design flaws, they affect only some users. This study is part of a broader exploration of why users are affected differently by some VE interface characteristics.
The data suggest that users need some skills to operate each of the three interfaces. All of the Ss came prepared with the skills needed for the helmet interface, but not equally ready for the other two.2 A training process might enable the slower participants to be faster with the non-immersive interfaces. Experiments on the amount of improvement from training are planned for the near future.
Clayton Lewis, Lew Harvey, James Wilson, and Greg Carey contributed valuable suggestions concerning several aspects of the study.
Space restrictions do not allow a detailed description here.
2Another part of the study, relating skills to performance, is not discussed here.
|
|