Anant Kartik Mithal
Department of Computer Science
University of Oregon
Eugene, OR 97403-1202
akm@cs.uoregon.edu
Pointing devices have become very important for HCI and their design
needs to move beyond iterative engineering approaches towards methods
guided by models that describe how pointing devices are used. This
thesis aims to extend psychologists' models of manual pointing to
pointing devices, as a step towards providing human factors
engineers with a basis for pointing device design.
Fitts' law, pointing devices, mouse, isometric
joystick, modeling, design, psychomotor models.
The rise in popularity of pointing devices has seen with it an
increase in the types of devices available. For example, in the 5
months since the last CHI conference (CHI '94), isometric joysticks
have become the rage in PC-compatible notebooks while Apple has
introduced a 'trackpad' in their computers. In addition, 3-D pointing
devices based on technologies such as gyroscopes have become
available. While these are exciting developments, designers of
pointing devices lack models of computer mediated pointing to guide
their efforts. They are therefore forced to follow an engineering
cycle of building the device, testing it, and then re-engineering the
device based on the test results. The performance of a new design
cannot be predicted, but must be built and tested. The goal of this
dissertation is to develop a model that describes cursor movement as a
function of time, and can help to analyze and design pointing devices.
Prior research on pointing devices has focused on gross pointing times
and adherence to Fitts' law [4]. Fitts' law does not help us predict
the performance of new devices because it is a post-hoc descriptive
measure of performance. Thus, while Fitts' law analyses allow us to
make generalizations about relative performance [2], much of pointing
device design involves more subtle changes. For example, the
designers of the IBM TrackPoint II used notions such as "a 'solid'
feel", and a "low slope of the [transfer] function at low speeds" [6]
to guide their design. It is not possible to predict whether such
ideas will improve pointing speed without actually testing prototypes.
For example, tests on accelerated 'power mice' showed no significant
difference between them and normal mice [3]. This approach has the
drawback that the result might be sub-optimal in terms of speed and
accuracy. What is needed is a systematic, principled approach, based
on a model that describes how people point with pointing devices.
While such models describing pointing devices do not exist, there are
psychomotor models of movement that describe how manual movement
occurs. These models can predict Fitts' law [5] and need to be
extended to pointing devices.
All psychomotor models of movement assume that a single pointing
action is made up of a sequence of smaller submovements. One class of
(unitary) models assumes that all the submovements are similar in
terms of accuracy and speed, while the other class of (dichotomous)
models assumes that the first submovement is different from the
subsequent sub-movements. Empirical testing has rejected the unitary
models in favor of the dichotomous models, and the model that has
shown the most promise is the Stochastic Optimized Submovement Model
(SOS Model) [5]. It assumes that an initial ballistic movement to the
target is followed by smaller feedback controlled movements. It also
assumes that the faster movements are made, the more errors occur,
which cause more corrective movements. Users try to minimize the
number of corrective move- ments, but maximize total speed. These
assumptions are used to derive a number of predictions about the move-
ment, including Fitts' law. Preliminary research has shown the model
holds promise in describing mouse movement [7], but this awaits
further study, as well as an extension to other pointing devices.
Thus, there are two gaps in our understanding of pointing
devices. First, the applicability of psychomotor models to
pointing devices is unclear. This is partly because there has
not been much research on modeling pointing devices and
partly because there are many types of pointing devices [1].
Therefore, we do not know if models applicable to one
device will be applicable to others. Second, we do not
know how to use these models in the design of pointing
devices.
This dissertation aims to fill these gaps by building on psychomotor
models of movement and extending them to pointing devices. Formally
stated, the research questions are:
The aim is to extend the SOS Model by studying it for two very
different pointing devices namely, a mouse and an isometric joystick.
If the model holds for these two devices, then the likelihood that it
holds for other pointing devices increases. The mouse was selected
because it is the baseline device in many studies, and a common device
can be used to extend the results from multiple studies [2]. The mouse
is an isotonic device, i.e., when users point with a mouse their limbs
change position. The mouse also employs a simple transfer function
converting input (mouse displacement) into output (cursor
displacement).
The isometric devices do not change shape, and so differ from isotonic
devices in that they do not provide any kinesthetic velocity or
position feedback to the user, which mice do. In addition, isometric
joysticks are typically designed as velocity controlled devices, where
the input force on the joystick controls the velocity of the cursor on
the screen. These characteristics make the mouse and isometric
joystick very different from one another, so a model that describes
movement with both devices is likely to also describe movement with
other devices.
FIGURE 1
A diagram of the screen of the proposed
experimental setup. A is the amplitude, or the distance of
the home square from the center of the target ribbon, W is
the width of the target ribbon.
This thesis is based on a series of two experiments. The first
experiment, illustrated in Figure 1, will measure displacement as a
function of time as subjects perform a simple pointing task. The data
will be parsed into the component submovements, and this data will be
used to test the predictions of the SOS model, such as the relative
speed and accuracy of the first submovement compared to subsequent
submovements, the number of submovements as a function of distance and
width, and the total movement time as a function of distance and
width.
The second part of the study will use the knowledge of movement
characteristics to redesign the joystick. This can be done by suitable
changes to its transfer function. One possible modification is that
if, as the SOS Model suggests, there are two distinct phases of
movement, then the isometric joystick could use different transfer
functions for each phase of the movement, such as velocity control for
the first phase, and position control for the second phase. While
this might sound counter-intuitive, such a device might in fact have
better overall performance. Once the modifications to the transfer
function of the isometric joystick have been made, a second
experiment will be carried out to compare the performance of the
original and modified joysticks.
At the present time, the software for the first experiment is being
tested, and the experimental protocol has been approved by the human
subjects review board. I expect to have recruited subjects and started
the first experiment by the beginning of November, and completed the
analysis by mid December. By April, I expect to be close to the
completion of the second phase of the study.
Abstract
Keywords:
Introduction
PREVIOUS RESEARCH
PSYCHOMOTOR MODELS OF MOVEMENT
THESIS PROPOSAL
CURRENT AND EXPECTED STATE OF RESEARCH
