[Top] [Prev] [Next] [Bottom]
Chapter 6
Usability Analysis of 3D Rotation Techniques
6.1 Introduction
One of the main points I would like to make in this dissertation is that passive haptic issues are important: facile virtual manipulation requires studying the feel of the interface, and not just the look of the interface. Although I believe my main contributions are related to the two-handed issue, I certainly had to consider other issues as I designed the interface for the neurosurgery application. One of the first major design decisions I faced was whether or not an approach based on 3D input was appropriate for the neurosurgery application at all. Visualizing and manipulating volumetric data sets with dexterity is important for this application, but the literature often seems to suggest that 3D input may have some shortcomings in this regard. My design experience suggested that, with appropriate design, this is not necessarily true: 3D input devices can be useful for both dextrous and fast virtual manipulation. The main contribution of the experiment presented in this chapter is to formalize this suggestion for the task of orienting virtual objects.
A more general motivation may also be appropriate here. With the rapid migration of fast, cheap 3D graphics to the PC platform [170], and with 3D graphics beginning to make their presence felt on the web through standards such as VRML [23], applications which incorporate 3D manipulation and 3D object viewing will become increasingly prevalent. In particular, orienting a virtual object to a desired view is a fundamental task, since viewing or inspecting an object is often a precursor to further manipulation.
Since a 1988 study by Chen which introduced the Virtual Sphere [40], I am not aware of any quantitative data comparing performance of currently proposed techniques for specifying 3D orientation. For example, there is no formal user study to compare the Virtual Sphere and the Arcball proposed by Shoemake [153][154], nor is it known what advantages (if any) direct orientation input using a 3D input device might offer for 3D orientation tasks. Chen's study does not include detailed observations of user expectations or the common difficulties encountered. Given that computer users typically have no experience with interactive computer-based 3D rotation tasks, a description of the usability problems that novice users may encounter when first exposed to these techniques forms a useful contribution.
The high cost of 3D input devices has traditionally limited their use to research or niche market applications such as head-mounted virtual reality systems, high-end animation software, or medical visualization. Free-space 3D input devices are still expensive compared to the mouse, but with the recent introduction of PC-based devices [137][5] priced near $1000, these devices are more affordable and practical now than they have ever been, and a growing number of interface designers will have the opportunity to explore the possibilities of free-space 3D input.
The intent of this study is not to argue that any one device or technique is "best." Each interface device or input technique will excel for some tasks and languish for others [31]; the most appropriate device for an application depends on the context of tasks to be supported and the intended users. The goal here is to collect some solid performance data for an experimental rotation matching task, so that informed design decisions can be made, and to collect some qualitative observations that will help to illustrate some strengths and weaknesses of each technique.
6.2 Overview
This formal user study evaluates interactive 3D rotation techniques including the mouse-driven Virtual Sphere and Arcball techniques (which I will shortly explain in more detail), as well as direct 3D orientation input techniques based on magnetic orientation sensors. The data suggest that when performing an orientation matching task, users can take advantage of the integrated degrees of freedom of 3D orientation input to complete the task up to 36% faster, without necessarily sacrificing any statistically detectable degree of accuracy.
I report detailed observations of user expectations and common usability problems when first encountering the techniques. My qualitative observations also suggest some design issues for 3D input devices. For example, the physical form-factors of the 3D input device had a marked effect on the subjective user acceptance (but not on quantitative task performance) for otherwise identical input sensors. The device should afford some tactile cues, so the user can feel its orientation without looking at it. In the absence of such cues, some test users were not able to form a clear concept of how to use the device.
6.3 Interaction techniques
My analysis includes the following interactive 3D rotation techniques:
Virtual Sphere: This mouse-driven 2D interface simulates a physical trackball. The virtual object is shown on the screen, and when the user clicks and drags on the virtual object, the computer interprets these drags as tugging on the simulated trackball. The virtual object rotates correspondingly. To provide the third rotational degree of freedom, a circle is drawn around the object (figure 6.1), and when the user clicks and drags in the area outside of the circle, rotation is constrained to be about the axis perpendicular to the computer screen. Hereafter, this outside area of the circle will be referred to simply as "the outside."
Arcball: This interface is similar to the Virtual Sphere, but it is based upon a more mathematically rigorous quaternion [152][155] implementation. It does not suffer from problems with gimbal lock or noisy data, and its implementation affords easy addition of constraint modes. Some designers consider the Arcball to be the best known 2D technique for 3D rotation. Shoemake has performed an informal comparison [153], but no quantitative data currently exist which compare the Arcball and the Virtual Sphere.
The Virtual Sphere and Arcball both require the user to achieve some orientations by composing multiple rotations, since only two of the three possible rotational degrees of freedom can be accessed at any one time.
Figure 6.1 Screen snapshot of the experiment software.
3D Ball: The user rotates a two-inch diameter plastic sphere (fig. 6.2) instrumented with a magnetic tracker to manipulate the virtual object. The magnetic tracker simultaneously provides all three rotational degrees of freedom, so in principle the user never has to mentally compose rotations with this interface. However, it is not clear if users can employ coupled rotation axes effectively [40], nor is it clear if the multiple degrees of freedom result in faster, but possibly less accurate, input of orientation data.
Figure 6.2 The 3D Ball input device.
The 3D Ball always acted as an absolute rotation controller: the orientation of the object being manipulated always matched the orientation of the 3D Ball. With the addition of a clutching mechanism (for engaging and disengaging the ball from a virtual object), it would be possible to use the 3D Ball as a relative rotation controller, by performing "ratcheting" movements [179]. Some other ball-shaped 3D input devices with integrated clutch buttons have used this technique [41][137][164]. The 3D Ball used here did not include any integrated buttons; I will return to the issue of integrated control buttons in section 6.9 of this chapter.
For the purposes of this experiment, the design of the device was kept as simple as possible, to avoid introducing secondary variables (such as absolute versus relative rotation). The basic task of rotating the object with integrated degrees-of-freedom is the main issue for this study.
Tracker: The Tracker interface, which also uses a magnetic orientation sensor, is identical to the 3D Ball in all regards except the physical packaging (fig. 6.3). This is the default form for the input device as shipped by the manufacturer [137], and as such represents the only way to use the device without designing or purchasing an alternative housing [47].
Figure 6.3 The Tracker 3D input device.
The Tracker has an unusual and unfamiliar shape. In our virtual reality lab [131], we have noted that with practice, experts can become quite proficient with the Tracker despite its awkward shape. It is not clear how well novice users will be able to adapt to its design.
6.4 Hypotheses
This study investigates the following specific hypotheses:
H1: Users can effectively use coupled rotation axes, and integrated control of all three degrees-of-freedom for rotation will provide significantly faster input of orientation data.
A study by Jacob [91] suggests that multiple degree-of-freedom input will be most appropriate when users think of a task's control parameters as integral attributes; I propose that 3D rotation matching is one such task. Most people are not good at mentally composing rotations, so when attempting to perform complex rotations, the separated 2D+1D control required by the Arcball and Virtual Sphere techniques should reflect this.
H2: Three-dimensional input is often assumed to be fast but inaccurate. I hypothesize that, at least for a 3D orientation matching task, 3D input provides fast orientation input without necessarily sacrificing any accuracy.
H3: The physical shape (or affordances) of the 3D input device can be an important design consideration in itself.
This hypothesis arose from my previous work on the props interface. The props take advantage of natural affordances (as discussed by Norman [126]), which can help users to know what to do just by inspecting or grasping an object or input device. The 3D Ball and Tracker used in this experiment are more general-purpose 3D input devices, yet nonetheless each communicates natural affordances which will implicitly channel user behavior; I intend to explore these issues in the analysis.
H4: The Arcball includes several apparent improvements over the Virtual sphere. As such, the Arcball should outperform the Virtual sphere in terms of task performance, user acceptance, or both.
6.5 The Experiment
6.5.1 Task
Test users performed an orientation matching task based on the task employed by Chen [40]. The goal here is not to reproduce Chen's results, but rather to extend the set of interactive 3D rotation techniques that have been formally evaluated with a common task.
A static view of a solid-rendered 3D model of a house, at a randomly generated orientation [4], was shown on the left side of the screen (figure 6.1). Test users attempted to manipulate a second view of the house on the right-hand side of the screen to match the random orientation. Each side of the house was colored uniquely to facilitate the matching. A circle was always drawn around both images of the house to assist matching, even though the circle only was strictly necessary for the Arcball and Virtual Sphere techniques.
When test users felt that the orientations matched, they clicked a footpedal to end the trial. After each trial, performance was rated as "Excellent!" (shortest-arc rotation less than 5.7 degrees), "Good Match!" (less than 7.6 degrees), or "Not good enough, try harder next time." I chose a footpedal to end the trials, rather than the spacebar used by Chen [40]. This kept the desk surface open for manipulation and it also allowed test users to use both hands (if desired) to manipulate the 3D input devices. The keyboard was removed from the desk during all experimental conditions.
Participants were given as little instruction as possible with each controller. With the Arcball and Virtual Sphere, the experiment began with a practice exercise to encourage the test user to click and drag within the circle (for two degree-of-freedom rotation) as well as outside the circle (for the third degree of freedom). When presenting the 3D Ball or Tracker devices, I placed the device on the table in front of the test user, mentioned that "It is not a mouse, it works differently," and suggested "Try to discover how to use it."
6.5.2 Experimental design
A within-subjects latin square design was used to control for order of presentation effects. Test users tried all four interfaces in a single session lasting about 1 hour. Test users performed matches for 15 unique orientations with each interface, but only the last 10 of these were included in the data analysis, to avoid any transient initial learning effects. There was a short break between conditions, during which I interviewed the test user about his or her impressions of the interface technique.
Dependent variables were Time to completion and Accuracy of the match. Accuracy was measured by the shortest-arc rotation between the final user-specified rotation and the ideal matching orientation.
6.5.3 Test users
Twenty-four unpaid test users (12 male, 12 female, all right-handed, mean age 19.1 years) were recruited from the University of Virginia psychology department's subject pool. All test users had experience with the mouse, while none had any experience with 3D input devices. Two test users had previously tried an interface similar to the Virtual Sphere, in the context of a molecule visualization application.
6.6 Results
Figure 6.4 shows the mean completion times and accuracies which test users achieved. The 3D Ball was 36% faster than the 2D techniques and the Tracker was 33% faster. There was little variation in the mean accuracies obtained.
Figure 6.4 Mean times (top) and accuracies (bottom)
Comparing to Chen's results for the Virtual Sphere, test users in the present study had longer times (Chen reported a mean of 17.5 seconds for complex rotations [40], while this study found 27.7 seconds), but in this study test users were more accurate (Chen reported 8 degrees of error1, while here test users achieved 6 degrees of error).
These discrepancies are probably primarily due to the differing test user populations: Chen used all males, some graduate students, and some students with experience in 3D graphics systems. This study includes females, all test users were undergraduates, and only two test users had any experience with interactive 3D rotation tasks. Another, probably less important, difference between the studies is that Chen's test users also performed more trials (27 trials vs. 15 trials in this study) so they may have become more proficient with the experimental task.
6.6.1 Statistical analysis
I performed an analysis of variance with repeated measures on the within-subjects factor of interface used, with task completion time and accuracy as dependent measures. The interface used was a highly significant factor for completion time (F(3,69) = 37.89, p <.0001) but not for accuracy (F(3,69) = 0.92, p >.4, n.s.).
Comparisons for completion time (fig. 6.5) revealed that the 3D interfaces were significantly faster than the 2D interfaces, but there was no significant difference between the 3D Ball versus the Tracker, nor was there any significant difference between the Arcball versus the Virtual Sphere.
These results strongly supports hypothesis H1, suggesting that users can perform an orientation matching task significantly faster when the rotation is presented as three integrated degrees of freedom.
Figure 6.5 Interface comparisons for completion time.
Comparisons for accuracy confirmed that there were no significant differences between any of the interfaces. This supports H2, suggesting that any accuracy differences between the interfaces are nonexistent or too small to detect with N=24 test users.
The analysis also revealed that the between-subjects factor of Sex was significant for completion time, as were the Interface 5 Sex and Interface 5 Order interactions for both completion time and accuracy (table 6.6). This indicates a need to investigate possible biases in the data due to the Sex or Order factors.
Figure 6.6 Significant between-subjects factors and interactions.
6.6.2 Separate analysis for males and females
This study was not designed to analyze sex differences, yet the results suggest that Sex was a significant factor. This is consistent with the sex differences literature, which has found an advantage for males on some tasks which involve mental rotation [74]. To ensure that Sex was not a distorting factor in the final study, I performed separate analyses with the N=12 male and N=12 female test users (fig. 6.7). For completion time, the results of the separate analysis were similar to those obtained in the combined analysis, suggesting that Sex was not a distorting factor.
Figure 6.7 Means for each interface technique by sex.
For accuracy, there was a relatively small, but significant, effect for females only (fig. 6.8). Females were about one degree more accurate (fig. 6.7) when using the mouse-based techniques vs. the 3D Ball, but not vs. the Tracker. This suggests a minor qualification to H2, that 3D input (at least with the 3D Ball) may be slightly less accurate than 2D input for females but not for males.
Figure 6.8 Significant accuracy device comparisons for females only.
6.6.3 Between-subjects analysis
I also performed a between-subjects analysis using only the data from the first interface that each test user tried. Thus, the 24 test users were divided into 4 groups with 6 users each, resulting in an analysis which eliminates the Order 5 Interface effects (fig. 6.6) as possible factors, but which also results in an overall less sensitive analysis.
Figure 6.9 Means obtained for the first interface tried.
The accuracy data suggested that test users were less accurate on the first interface tried for all interfaces, except for the Arcball (figs. 6.9, 6.10). I can offer no explanation for this result, but the within-subjects analysis strongly suggests that any such initial accuracy differences between conditions evened out as the test user became more experienced with the task. Nonetheless, even though there is only data for N=6 test users in each group, the between-subjects analysis suggests the following qualification to H2: The 3D Ball may be less accurate than the Arcball when users are first exposed to the orientation matching task.
Figure 6.10 Significant effects for between-subjects analysis of accuracy.
6.7 Qualitative results
Figure 6.11 shows a histogram of the subjective ranks which test users assigned to each interface. The Arcball ranks suggested a generally positive reaction, but this trend was not significantly better than the ranks assigned to the other techniques. The 3D Ball was most commonly selected as the preferred technique for the experimental task, and nobody rated it as the worst technique. Figure 6.12 shows a pie chart illustrating how many subjects chose each interface as their "most preferred technique."
Figure 6.11 Histogram of subjective ranks for each interface.
Figure 6.12 Pie chart showing distribution of votes for the favorite technique.
Statistical analysis confirmed that the 3D Ball had significantly higher ratings than the mouse-based 2D techniques or the Tracker 3D input technique (fig. 6.13).This provides strong evidence in favor of H3, that the physical form factors of a 3D input device can be an important design consideration. The awkwardness of the Tracker resulted in poor subjective impressions, despite the relatively high task performance which test users were able to achieve with it. Yet the exact same input sensor packaged as a 3D Ball resulted in significantly more favorable reactions.
Figure 6.13 Statistical comparison of subjective ranks.
6.7.1 2D techniques: the Arcball and Virtual Sphere
The Arcball & Virtual Sphere shared many qualities, so I will discuss their similarities before contrasting the techniques. The techniques were generally well accepted, with many users commenting that the techniques were "pretty easy" or that they "worked really well once you learned the inside and outside of the circle."
The most common problems related to the modal distinction between the "inside" and the "outside" of the circle. During practice, I was careful to have users try both dragging inside and outside of the circle. But users see the circle as a target, and not as a border distinguishing two separate modes, so users would frequently attempt to click on the circle itself, and thus mistakenly switch between the inside and outside behaviors. Similarly, when attempting a large single-axis rotation using the outside of the circle, users would mistakenly come inside the circle, and would be surprised when the initial "outside" behavior changed to the "inside" behavior.
Several test users avoided using the outside, and would sometimes become unsure of what to do next if all the rotations matched except the rotation about the third axis perpendicular to the screen. A previous pilot study2 had revealed that even though this third rotation axis was available, it was a hidden feature of the interface which very few users would discover on their own. In the final study reported here, during initial practice I had each user try dragging the mouse in the outside area of the circle, yet some test users still chose to ignore this feature.
Many test users were uncertain about where to click and drag with the mouse, and once they started to drag, they were reluctant to stop. This resulted in the impression that using the mouse was "not as smooth as [the 3D techniques]- you have to start and stop a lot to click the mouse button." Some test users hesitated to make a large movement which would disturb the progress made so far. As one user commented, "When you get close, you can screw it up -- I didn't want to mess up what I already had." Thus, test users sometimes seemed to be unsure of what effect their actions would have, and as a result they would try to plan their motions carefully. This is not a behavior which I observed with the 3D techniques.
6.7.2 Comparison of Arcball and Virtual Sphere
In theory, there are two primary distinctions between the behavior of Arcball and that of the Virtual Sphere: the Arcball avoids "hysteresis effects" and the Arcball uses half-length arcs. Hysteresis occurs when "closed loops of mouse motion may not produce closed loops of rotation" [153]. This means that it may not be possible to "undo" a sequence of drags by reversing their order. Half-length arcs are a property of how rotations combine, and result in a fixed C:D (Control:Display) ratio which is free of hysteresis. For example, with half-length arcs, a sweep across Arcball's inner circle moves the virtual object 360 degrees. The same sweep would only move the Virtual Sphere 180 degrees. This rotational C:D ratio is fixed by the mathematics underlying Arcball and cannot be changed without reintroducing hysteresis.
Some users did not notice these differences between the Arcball and Virtual Sphere techniques; as one test user put it, "I felt like I was doing the same thing again." Nonetheless, there was a general preference (16/24 participants) for the Arcball over the Virtual Sphere. A typical reaction was: "The Arcball is a little more responsive, and it gives you more control." Test users who preferred the Virtual Sphere often commented that they liked its slower rate of movement. The Virtual Sphere's slow movement suggested that the mouse was "pushing on a specific spot," whereas "Arcball just rotated around in different axes."
The Arcball also displays feedback arcs to illustrate how mouse movement affects rotation. These feedback arcs did not seem to be a significant advantage. The feedback arcs were often ignored or even regarded as annoying, although at least a few test users thought they helped at first. The exception was feedback arcs for the outside of the circle (for rotating about the axis perpendicular to the screen), which test users did find to be helpful.
6.7.3 3D Ball
Overall, test users had very positive reactions to the 3D ball technique. Typical comments were "this makes you have total control," "it was like holding the object," and "you could just turn it around instead of repeatedly clicking [with the mouse]."
Unlike the mouse-based techniques, the 3D manipulation techniques had to consider the physical form of the input device, as none of the test users had previous experience with 3D input. The 3D Ball's form-factors help to convey a clear message: balls are for rolling, spinning, or turning. However, this didn't always assist learning of the device. Several test users were initially convinced that the 3D Ball should be used by rolling it on the desk surface, rather than by picking it up. This "rolling" strategy was especially problematic for orientations which required the cord to point directly downward.
Although the ball shape conveys a clear message, its smooth, spherical aspect was sometimes problematic, because the surface offered few tactile landmarks or handles. This seemed to prevent some test users from forming a clear concept of the device. One test user commented that "I don't think I ever figured out how to use it-- I just wiggled it around in my fingers until I found it [the matching rotation]." The smooth surface was also somewhat slippery, making it more difficult for some test users with small hands to tumble the ball.
Most test users found the 3D Ball's cord annoying, as it would sometimes get in the way or become tangled. Some users also felt that they weren't as precise with the 3D Ball as with the mouse techniques, and some preferred using the mouse because they were used to it.
6.8 Tracker
The Tracker's physical form-factors do not convey any clear message. Test users were typically quite confused when first encountering the Tracker ("this is a very weird thing"), but they were able to adapt after several trials. One test user explained that "at first I thought it had this really dumb shape, but it turned out to be easier to hold than I thought it would be."
Several test users initially tried to use the tracker by sliding it on the desk, and users often experimented with alternative grips for holding the device in the air. Users commented that they were "unsure how to use it," that it was "an odd shape for manipulation," or that the device seemed to be too small to grasp effectively. Many test users used both hands to hold the tracker, but this did not seem to be by choice. As one test user indicated, "you almost had to use two hands" to manipulate the tracker effectively. With the 3D Ball, most test users employed one hand only.
The hindering effect of the cord was the most common verbal complaint. The weight of the cord is comparable to the weight of the tracker itself, which makes it difficult to rotate the device about its center of mass, and results in some awkward postures.
The irregular shape of the tracker caused much confusion, but it also conferred some advantages. The mounting flanges for the tracker (figure 6.3) served as handles, and gave some tactile information as to the orientation of the device. One user commented that the "ball was a big smooth object, but now I have some handles and landmarks." Three test users who struggled with the 3D Ball performed quite well with the Tracker: the tactile cues afforded by the device seemed to be essential for these individuals.3
6.9 Discussion
The qualitative observations from this formal study, as well as prior informal observations from implementing several variants of 3D input devices for orientation [80][164], suggest some general design parameters which can influence how users will employ these devices. These are not well-formulated principles for design, but rather some issues intended to demonstrate how a design can implicitly channel user behavior to differing styles of interaction.
Affordances: The input device needs to suggest that it serves to orient virtual objects, while at the same time suggesting that it should be picked up, and not used on the desk surface. In my work with the neurosurgical interface props, the doll's head serves this purpose well: the doll's head just can't be manipulated effectively on the desk surface, and so strongly encourages being picked up. The 3D Ball presented in this experiment did not suggest this well enough, leading some test users to initially use it by rolling it on the desk surface.
Tactile Cues: For absolute rotation control, the device should not necessarily be completely symmetric and should have clear tactile information which indicates preferred orientation and helps to hold the device securely. In this regard, a ball-shaped device has some shortcomings. For relative rotation control, however, this issue needs to be explored further. The symmetric ball shape might be advantageous in this case, since any landmarks on the device would not correspond to the virtual object, and could be misleading. This is an example of how a drawback for one style of input may be a virtue for another.
Grasp: Many 3D input device designs encourage the user to hold the device against the palm at a fixed orientation. This is known as the power grasp [113] because this hand posture emphasizes strength and security of the grip. By contrast, the precision grasp involves the pads and tips of the fingers and so emphasizes dexterity and free tumbling of the input device. This issue was formally analyzed by Zhai for a 6 DOF docking task, where he found that using the fine muscle groups emphasized in the precision grasp results in significantly faster performance [187].
The design of a 3D input device can influence which style of grasp users will choose. Integrating buttons with the input device encourages users to adopt a power grasp, holding the device in the palm, while the fingertips maintain contact with the buttons. The power grasp may require relative rotation control to avoid awkward postures, since the biomechanical constraints of the hand and arm can limit the range of rotation. Also, the muscle tension required to press or hold an integrated switch can interfere with the user's ability to manipulate the input device. Using a footpedal to separate the button from the device can be an effective alternative [80], and can help to encourage use of the precision grasp.
Device acquisition time: The results reported here suggest that the mouse-based techniques are slower for the experimental rotation matching task. Nonetheless, in a work routine that uses the mouse frequently, the time to switch over to a 3D device, and then back to the mouse, might become a dominating factor. This could result in longer times for users to accomplish their overall goals, even though individual virtual object rotation tasks could be performed more quickly. This illustrates why one cannot conclude from this study that "3D input is best," but that one must also consider the context of surrounding tasks and the intended users.
It may be possible to eliminate acquisition times in some cases by using a 3D device in the nonpreferred hand. This offers the possibility of working with a 3D device and a mouse at the same time [107], or with a pair of 3D devices simultaneously [80][141][149].
6.10 Conclusion
Revisiting the original hypotheses, this study provides clear evidence that test users were able to take advantage of the integrated control of 3D orientation input to perform a rotation matching task more quickly than with 2D input techniques, despite years of prior experience with the mouse. More importantly, with possible slight qualifications for female test users or during initial exposure to the task, there were no statistically reliable accuracy discrepancies between any of the input techniques, demonstrating that 3D orientation input is fast without necessarily sacrificing accuracy.
The quantitative data did not lend any support to the hypothesis that the physical shape of the 3D input device can be an important design consideration, but the subjective ranks which test users assigned to the interfaces spoke definitively on this point. Users reported diametrically opposed impressions of input sensors which differed only in their physical housing.
The study did not provide evidence that the Arcball performs any better than the Virtual Sphere. Test users tended to prefer the Arcball over the Virtual Sphere, but this advantage was not significant statistically. These non-results do confirm, however, that the Arcball's possibly confusing concept of half-length arcs (as described in section 6.7.2 on page 124) does not cause any obvious usability problems [153].
This study has focused on manipulative techniques to achieve 3D rotation, but of course direct manipulation (whether 2D or 3D) is not the only approach to orienting and viewing objects. Even with a good interface, specification of orientation can be a difficult task for some users, so it makes sense to obviate the task when appropriate.
For example, the Sketch system [186] (fig. 2.3 on page 17) provides a casual, inexact interface and a cleverly chosen set of heuristics which often allows the system to select views of objects based on fast, relatively imprecise 2D mouse clicks. The Jack system [135] takes a similar approach by automatically generating unobstructed views of selected objects. Additional work is needed to understand what classes of user tasks are appropriate for each technique, and to determine how to best mix the two styles in the same interface.
I have attempted to provide a careful analysis of performance in terms of time and accuracy and to provide detailed observations of usability problems one might expect for novice test users performing tasks similar to the experimental rotation matching task. I have also attempted to synthesize specific qualitative observations from this study and my previous experience with designing 3D input devices for rotation control to suggest some tentative design parameters which can influence how users will employ these devices. These contributions should prove useful to the growing number of interface designers who will incorporate 3D rotation control techniques into their applications.
1
Chen reported accuracy in terms of the sum of squared errors between the user's rotation matrix and the rotation matrix to match. Chen indicates a squared error of 0.04 [40], which converts to 8.1 degrees.
2
I would like to acknowledge the contribution of Joe Tullio, who implemented the interfaces, ran all of the pilot studies, and assisted with the final study in the course of his senior thesis.
3
These are the three test users who rated the Tracker as their favorite technique (fig. 6.12).
[Top] [Prev] [Next] [Bottom]
Copyright © 1996, Ken Hinckley. All rights
reserved.