CHI 97 Electronic Publications: Late-Breaking/Short Talks
Object Manipulation in Virtual Environments: Human Bias, Consistency and Individual Differences
Yanqing Wang
School of Kinesiology
Simon Fraser University
Burnaby, BC V5A 1S6
Canada
+1 604 291 5794
wangy@move.kines.sfu.ca
Christine L. MacKenzie
School of Kinesiology
Simon Fraser University
Burnaby, BC V5A 1S6
Canada
+1 604 291 3004
cmackenz@cs.sfu.ca
Valerie A. Summers
Dept. of Computer Science
University of British Columbia
Vancouver, BC V6T 1Z4
Canada
+1 604 291 5794
summers@cs.ubc.ca
ABSTRACT
This paper investigates human bias, consistency and individual
differences when performing object manipulation in a virtual environment. Eight
subjects were asked to manipulate a wooden cube to match a 3-D graphic target
cube presented in 3 locations and 2 orientations. There were two visual
conditions for the experiment: the subject performed the tasks with or without
vision of the hand and the wooden cube. The constant errors of object
translation and orientation suggested specific human biases. In terms of the
variable errors, visual feedback appeared to be more critical for object
transportation than object orientation. It was also found that individual
differences were more pronounced in human bias than in consistency during
object manipulation. These results suggest tolerance for human bias and
variability should be accommodated in human-computer interface design.
Keywords
Interface design, object manipulation, virtual environment, human performance
© 1997 Copyright on this material is held by the authors.
INTRODUCTION
Object manipulation is an activity we perform every day. The promise of
direct manipulation, one of the most important user interface styles, is to
transfer object manipulation skills developed in the physical world into
human-computer interaction environments. However, evidence shows that humans
may make systematic and random errors in prehension movements including object
manipulation under certain task conditions (Soeching and Flanders, 1993; Jacob
et al., 1996; Fikes, 1995). Such errors will occur in human-computer
interaction as well. The purpose of this study is to examine human bias,
consistency and individual differences when manipulating an object in a virtual
environment, and therefore to provide useful information for future computer
interface design.
EXPERIMENTAL DESIGN
Manipulation tasks were designed that required both object
transportation and orientation, under different visual feedback conditions. A
stereoscopic, head-coupled graphical display, and a half-silvered mirror were
used to create a virtual environment (The Virtual Hand Lab). Eight students
were asked to manipulate a small wooden cube so that it occupied the same
position and orientation as a graphic target cube. The target cube was located
on the table top, 30 mm, 100 mm or 200 mm from the starting position, and
rotated by 22.5 or 45 degrees from the frontal plane of the subjects. The tasks
were performed in one of two conditions: without vision (only the graphical
objects were visible), or with vision (the subject's hand and the physical cube
were also visible). An OPTOTRAK monitored 3-D positions of 3 infrared markers
placed on the cube. Dependent measures were errors in location matching and
orientation matching between the manipulated cube and the target cube. The
constant error indicated human bias in object manipulation, and the variable
error reflected the consistency of control on a trial-by-trial basis.
RESULTS
Constant Errors
As shown in Figure 1, there was a highly significant bias to the right
of the target (positive Y direction), in both visual conditions, p < .01.
Further, the bias differed between the two conditions (p < .01), 2.6 mm
without vision of the hand and the wooden cube, and 12.9 mm with vision. There
was a trend (6 out of 8 subjects) to overshoot the target, but the bias in the
X direction was not statistically significant for either visual condition.
The subjects significantly under-rotated (p < .05) the angle required to
match the target orientation, as shown in Figure 2. The average under-rotation
was 2.5 degrees in the "with vision" condition and increased to 6.6 degrees
when no visual feedback was available (p < .01).
Variable Errors
For both X and Y axes, subjects were less consistent without vision of
the hand and the cube than with vision (p < .01). With vision, the average X
error was 1.5 mm and the average Y error was 1.2 mm, while without vision the X
error increased to 11.6 mm and the Y error increased to 5.7 mm.
It is worth noting that even though the angular variable error increase
from the vision to the no vision condition was statistically significant (p
< .01), this increase was rather small, less than 1 degree (Figure 2).
Individual Differences
Figures 3 and 4 demonstrate the individual differences in bias and consistency
of object manipulation, in terms of standard deviation for the eight subjects.
In general, individual differences were greater in bias than in consistency
during object manipulation, Deprivation of visual feedback information of the
hand and the wooden cube increased the performance variation among the
individuals. In the no vision condition, the individual transportation bias
ranged from -15.6 mm (target was undershot) to 43.7 mm (target was overshot)
along the X axis, and from -0.8 mm (cube to left of target) to 25.2 mm (cube to
right of target) along the Y axis. The individual rotation bias ranged from
-9.0 degrees (under-rotation) to 1.7 degrees (over-rotation).
Figure 1. Displacement errors (mm) in X and Y directions. CE =
Constant Error and VE = Variable Error
Figure 2. Angular errors (degrees) on XY plane
Figure 3. Between subject standard deviation of displacement
errors (mm)
Figure 4. Between subject standard deviation of angular errors
(degrees)
DISCUSSIONS AND CONCLUSIONS
The magnitude of the human bias in object manipulation is larger in the
no vision condition, but the trend of the bias is very consistent whether or
not visual feedback of the hand and the cube are present. Users tend to
overshoot the target, but under-rotate the target orientation.
Consistency decreases with "impoverished" visual feedback in the virtual
environment. However, the change in consistency over different visual
conditions is not as dramatic for object orientation as for object
transportation. When visual feedback of the manipulator (hand) and manipulated
object are unavailable in a virtual environment, individual differences in bias
are particularly large. This implies that direct manipulation interfaces in
virtual environments will be more effective should they utilize a calibration
mechanism which accounts for individual biases. The error values presented in
Figure 1 and 2 may serve as rule of thumb for determining the tolerance bounds
for matching or selection tasks.
REFERENCES
1. Fikes, T.G. Spatial and temporal characteristics of feedforward reaching.
The 36th Annual Meeting of the Psychonomic Society, Los Angeles CA, 1995.
2. Jacob, R.J.K, Sibert, L.E. McFlarlane, D.C. and Mullen, M.P.Jr. Integrality
and separability of input devices. ACM Trans. on Computer-Human Interaction,
Vol. 1, No. 1, 2-36, 1994.
3. Soeching J. S. and Flanders, M. Parallel, interdependent channels for
location and orientation in sensorimotor transformations for reaching and
grasping. Journal of Neurophysiology, Vol. 70, No. 3, 1137-1150, 1993.
CHI 97 Electronic Publications: Late-Breaking/Short Talks
