VespR: design and evaluation of a handheld AR device

Eduardo Veas*

Ernst Kruijff

Institute for Computer Graphics and Vision

Graz University of Technology

Figure 1: Two-handed VespR (A), single-handed VespR assisted by non-dominant hand (B),
electronics of small handle (C) and BatPack (D)

ABSTRACT
This paper focuses on the design of devices for handheld spatial
interaction. In particular, it addresses the requirements and
construction of a new platform for interactive AR, described from
an ergonomics stance, prioritizing human factors of spatial
interaction. The result is a multi-configurable platform for spatial
interaction, evaluated in two AR application scenarios. The user
tests validate the design with regards to grip, weight balance and
control allocation, and provide new insights on the human factors
involved in handheld spatial interaction.
CR Categories and Subject Descriptors: I.3.6 [Computer
Graphics]: Methodology and Techniques - Interaction Techniques.
Additional Keywords: Augmented Reality, handheld devices,
mobile computing, 3D user interface, garage interface design
1

INTRODUCTION

Over the years, several hardware platforms have been

introduced to support augmented reality (AR), ranging from
backpack systems to lightweight portable devices. Recently, the
increase in processing power of small devices (PDAs, cell phones)
has brought AR to the handheld realm. These portable devices
excel in form factor, but include a small number of low quality
sensors and limited processing power. While it is imaginable that
future devices will include more sensors (e.g. GPS, inertial
sensors), AR imposes strong requirements on quality and often
requires new, state-of-the-art sensors. Ultra mobile PCs (UMPC)
present an in-between solution combining mobility with increased
processing power while their standard interfaces simplify adding
new sensors.
This paper describes the design and evaluation of a multiconfigurable platform for spatial interaction in handheld AR,
called VespR (Figure 1). VespR is a lightweight construction

*Inffeldgasse 16, 8010 Graz, Austria veas@icg.tugraz.at

IEEE International Symposium on Mixed and Augmented Reality 2008

15 -18 September, Cambridge, UK
978-1-4244-2859-5/08/$25.00 2008 IEEE

built around a UMPC, holding peripherals and additional controls

in an ergonomic way [1]. VespR is specifically meant to support
longer duration and possibly outdoor AR task settings. In
particular, this paper considers the human factors of spatial
interaction with handheld devices in relation to tasks in AR
applications. Only few times the construction of platforms for AR
has been approached with regards to ergonomics [2]. To better
understand the functional requirements, we performed a task
analysis on handheld AR applications, studying the functional
range and associated control aspects. Based on the outcomes, the
development of the device went through several iterative stages:
requirements and ergonomic analysis, functional allocation,
control body linkage, electronics design, and assembly.
Whereas some researchers create I/O devices, most of the
knowledge on these development processes is hardly published.
Within this article, we share some of the ideas that lead both to
device design mistakes and improvements. We have put a focus
on the following device factors, affecting (outdoor) handheld
spatial interaction to a great extent: grip, weight balance and
control allocation. The biggest challenge presented in this paper is
the need to extend and improve the functionality of UMPCs with
additional devices, without reducing ergonomics. The notable
increase of weight must and can be dealt with: as shown, doubling
the weight can still achieve better ergonomics.
A first evaluation using the single-handed VespR (Figure 1B)
showed positive results on fatigue and usage of controllers. The
test allowed users to freely explore an excavation site, applying
several magic lens tools. A second study compared different
VespR configurations with commonly used encasings and
UMPC-only conditions. Through both evaluations we could assert
that using a single handed grip on the UMPC construction is
feasible. We also showed that the UMPC itself performs well in
short, simple (indoor) interaction situations in which no additional
peripherals are needed, but quickly becomes tedious as the
interaction time increases or additional devices are required. Most
importantly, VespR covers the field of more complex, possibly
outdoor AR applications in a highly appropriate way: the twohanded VespR improves interaction and ergonomics of handheld
AR, outperforming other devices for these kinds of applications
by far.

43

Task

Accuracy

Speed

Frequency

Duration

Input

Handedness

Typical
DOF
6DOF
(constr.)

Isometric
isotonic
isotonic
isometric

Viewpoint
manipulation

low

high

long

continuous

single
bimanual

Maneuvering

medium

low to
medium
(fast)
low

medium

short

continuous

single
bimanual
single
bimanual

6DOF
(constr.)
1/2DOF

isotonic
isometric
isometric
isotonic

Move
large datasets
(maps)
System control

low to
medium

low to
medium

vary /
medium

medium
to long

continuous

low

low to
high

vary

short to
medium

discreet
continuous

single
bimanual

1/2DOF
6DOF

isometric
(isotonic)

Visualization
mode change

low

low to
high

vary

short

discreet

single

1/2DOF
(6DOF)

isometric
(isotonic)

Object selection

low to
high

low to
high

vary

short

discreet

single
bimanual

1/2DOF

isometric
(isotonic)

Object
manipulation

medium
to high

low

vary

short to
long

discreet
continuous

single
bimanual

1/2/6DOF
(constr.)

isometric
isotonic

Move lenses

medium
to high

low

vary /
medium

medium
to long

continuous

single
bimanual

1/2/3DOF

isometric
isotonic

Numerical / text
input

low

medium

low

medium

discreet

single
bimanual

1DOF

isometric

Typical control
human motion sensing
(micro)joystick
human motion sensing
(micro)joystick
(micro)joystick
pen / touch
human motion sensing
button / slider
pen / touch
(micro)joystick
human motion sensing
button / switch
pen / touch
(micro)joystick
human motion sensing
Button
pen / touch
(micro)joystick
(micro)joystick
pen / touch
human motion sensing
(micro)joystick
pen / touch
human motion sensing
keyboard
pen / soft-keyboard

Table 1 Overview of typical handheld AR tasks and characteristics defining the interaction requirements (italic tasks are subtasks)

RELATED WORK

Wearable AR systems have been around since the nineties and

make use of mobile computers. Two different directions, each
with different form factors, can be identified: the head-mounted
display direction using laptops, and the handhelds direction. The
first direction ranges from using lighter weight but therefore less
powerful systems, to bulkier ones. These systems are getting out
of date due to the pace miniaturization of hardware proceeds
[3],[4], a trend that can clearly be observed in the appearance of
many PDA and cell phone-based AR solutions. Supported by
increasing processing power and small internal or external
cameras, let alone a large (and growing) user base, these devices
form an interesting platform for predominantly marker-based
AR [5]. These devices excel in form factor for simple
applications: they are small and light, but have limited control
possibilities. A few AR platforms, including the predecessors we
built upon [6-8], focus on extending UMPCs. Some of these
platforms tend to be very specialized, like the force sensing
extension reported in [9]. Most platforms have been developed
from a practical point without consideration of ergonomics or
human factors. One exception, the AR mask by Grasset et al [2]
partly overlaps one of our grip approaches. Alternatively,
devices like the (old) Xybernaut setups separate the processing
unit from the display, lowering the weight to hold in the hands.
This direction has hardly been continued. Separation of units
provides some advantages (lower weight) but also some
disadvantages, including cabling and difficulties to hand over a
device.
3

TASK ANALYSIS

We performed a task analysis on handheld AR applications,

studying the functional range and related control aspects, to
better understand functional requirements on handled AR device
constructions. The majority of handheld AR applications are
related to location-based services (LBS), and may have
considerable overlap with general cell phone or PDA
applications [10]. For interaction, LBS rely on camera or sensor-

44

based solutions, or a combination of both. LBS provide contextbased information delivery, and mostly simplified interaction:
the information is either filtered based on location, or can be
filtered manually by the user. LBS can apply push or pullservices, varying the level of user involvement in information
delivery. A typical example of sensor-based interaction is a map
application on top of which information is layered. Camerabased interaction provides static or dynamic detection of
location to trigger information delivery. It may be associated
with complex applications, like the map example, or just display
web pages. From the AR perspective, it is most relevant that the
camera image can be used to overlay registered data. Most
dynamic camera-based interaction is limited to simple motionbased actions, providing viewpoint-specific content. Location
tracking can be based on optical tracking (using the camera) or
any other location sensing method.
Looking closer at handheld AR interfaces, one can notice a
clear dependency between the system platform and the range
and complexity of interaction. For cell phones, the addition of
sensors is mostly out of the question. For this reason, they make
use of the camera for viewpoint manipulation, and one or two
(integrated) buttons or the micro-joystick for system control and
selection purposes. The tasks afforded by a cell phone are very
restricted: typically, manipulation actions are constrained to 2D,
due to the fact that most controls are small and difficult to use.
Besides, interaction is restricted to short duration tasks.
Applications with reduced interaction requirements and simple
tracking solutions (indoors, low computation vision-based, etc)
are perfect for these unmodified devices. On the other hand,
applications with higher complexity of tracking or interaction
require extending the platform with more accurate sensors.
UMPCs simplify extension through standard interfaces (USB,
Bluetooth), affording other control possibilities and the additions
needed to support outdoor applications.
In our task analysis, based on [11-14], we investigated a range
of handheld AR applications and device setups (real setups and
literature), including [5, 15-18]. Table 1 provides an overview
of representative actions performed with handheld AR setups.

As the table shows, there is hardly any uniformity: many tasks

can be performed in different ways, depending on device
capabilities and interaction style. For example, interaction will
be mainly isometric and 2DOF when applications only access
GPS signals to locate the user, applying built in controls for
other interaction purposes. In contrast, extended UMPC setups
allow for both isometric and isotonic 6DOF interaction modes,
changing the interaction style to a large extent.
Viewpoint manipulation and maneuvering (making fine
movements to obtain a specific viewpoint) also depend on the
interaction style. Dynamic navigation (like moving a map) and
simple movement of data sets can be performed through human
motion sensing, a micro joystick or a pen. Viewpoint
manipulation involves frequent actions, which are mostly lower
accuracy (except maneuvering) and low speed. Only
infrequently, navigation actions will be of higher speed. For
example, a user may just want to quickly recognize a marker for
location purposes.
System control can be performed in many different forms,
heavily depending on the system platform. Predominantly, we
observed desktop-like menu techniques or dedicated buttons.
Some applications even make use of gestures, though these may
pose fatigue on the user. We also observed switch-like control,
to toggle between two visualization modes.
Object selection and manipulation is rather limited in most
applications. The most frequently performed action is the usage
of lenses. Lenses can be either fixed or movable. Movement of
lenses is generally constrained to ease operation. The majority of
applications rely on a micro joystick to move the lens on the
image plane, but in some cases, users can also move a lens
through the surface (3DOF). Rotation was mostly disabled.
Finally, some applications require some way of numerical or text
input, for example to label objects. This only happens seldom,
due to the limited text-input capabilities of most devices.
Outcomes
Several outcomes specifically affect the UMPC construction:

Task performance in outdoor AR is highly dependant on

external devices to the construction. For indoor setups,
depending on tracking and interaction (temporal and
spatial) complexity, the UMPC might suffice.

An increase of functionality does not necessarily lead to

a linear increase of controllers: many tasks may be
controlled with the same controller, depending on the
interaction style. The range of functionality, though, is
limited. The controls in unmodified UMPCs are not
always appropriate enough (number, position, accuracy)
for longer usage sessions.

The kind of interaction performed on the UMPC will

likely combine isometric and isotonic control.

Control can range from 2DOF to 6DOF, and can be

constrained to improve performance.

Duration of interaction is dominated by viewing of data,

which is generally a lower accuracy task and requires the
user to be pointing at a certain spot in the real world. As
can be seen later on, there is an (obvious) direct trade-off
between pose, weight and the task performance in longer
sessions.
4

DESIGN OF THE HANDHELD DEVICE CONSTRUCTION

Based on the results of the task analysis we initiated the design

and construction of a device to support handheld AR interaction.
We followed an iterative process, in which many different
device form variants and controller allocations were analyzed.

During the process, we were supported by several users that

regularly tested and discussed different forms of the
construction. Within this section, we describe different stages of
design and the considerations behind them. Not only do we
focus on the ergonomic form of the case with its specific grips,
we also describe how the functional allocation evolved out of
trying out different possibilities for controllers. Specifically, we
went through the following stages:

Requirements analysis: defining the actual tasks to be

performed with the new construction, and how these tasks
are (or are not) matched by the current device or device
constructions
Ergonomic analysis: investigating the ergonomic
characteristics for the UMPC and its additional housing
(hull / case) by making multiple hull and grip studies.
Functional allocation: matching the functionality on the
controllers.
Control-body linkage design: defining and designing the
device construction consisting of the case and the handles.
Electronic design: choosing and developing the electronic
layout of the construction
Assembly:
refining and complete assembly of the
construction

4.1
Requirement analysis
The device construction presented in this paper evolved out of
requirements from an outdoor AR project for field workers
called Vidente [18]. The application overlays underground
structures over live video images, and makes use of several
tools: an excavation tool (magic lens), an x-ray tool, a filtering
tool (changing visualization modes), a labeling tool, and a
snapshot tool. The tools resemble standard handheld AR tasks as
stated in Table 1, and can be compared to handheld applications
of higher interaction complexity.

Figure 2 Different casing generations holding peripherals

The extended device construction needs to support the full

functional range: high quality AR (accurate registration and
visualization) frequently depends on additional devices like
orientation sensor, GPS, and a higher quality camera. Many
applications can not work without such additional hardware.
However, additional devices put ergonomic constraints on
usage: the higher weight of the construction restricts interaction.
The approach taken most frequently by other developers
integrates sensors in a plastic component attached to the back of
the UMPC (see Figure 2). At the same time, the in-device
controls (buttons, joystick, touch screen) are used for interaction.
Both the controllers and the additional plastic case often limit
usage with the device, especially during longer sessions. The
plastic case is not easy to hold, and the ergonomics of in-device

45

controllers deteriorate over longer time, since the user is strained

by holding up the device. As we will see in the next section, the
latter is caused by the size and position of most controls on the
UMPC that are not always compatible with the pose and grip of
handheld AR usage.

grip. It is ideal at around 76mm, at which it increases the

strength of the wrist, thereby decreasing fatigue [19]. The best
grasp is achieved when users can press the grip firmly into the
palm of the hand, instead of the lower end of the fingers.

Outcomes
Following outcomes directly affect (improve) the performance
of a handheld AR construction:

Integrate external devices without destroying ergonomics

Improve the grip ergonomics

Provide easily accessible and better performing

controllers

Keep weight as limited as possible and /or balanced in a

better way
4.2
Ergonomic analysis
We first looked at how users hold the UMPC, focusing on key
human factors for handheld AR: the pose and grip on the device,
the weight balance affected by the overall pose, and the usage
(muscular movements) to reach and use a controller. We focused
on how the Sony Vaio UMPC (types VGN-UX180P and U70) is
used for AR: both types have very good weight-performance
characteristics and are frequently used by other developers too.
During the analysis, we observed several problems. First, the
user needs to hold the device up high, when observing the
environment at eye sight (pose). Only when objects at the lower
visual field are observed, the device can be lowered, which
changes the weight balance considerably. Holding up the device
causes fatigue that restricts the interaction to a large extent [19].
The usage time gets limited to several minutes at most, and
deteriorates the usage of in-device buttons over time: longer
sessions increase fatigue. This affects fine motor actions: over
time, precise actions will be harder to perform due to increased
fatigue. Biomechanics studies indicate that, depending on the
weight of an object in the hand, users will experience severe
fatigue after latest 3-5 minutes of duration, when an object is
held about 50 cm away from the body, in front of the eyes.
Because of the lever-system biomechanics of the arm (internal
force to balance external force), fatigue will occur in the
shoulder, upper and lower arm. Moving the device down a little
(up to 20cm) can easily double the duration of usage without
experiencing fatigue [19]. Hence, both weight and the way the
device is held (below or from the sides) greatly affects the usage
duration and associated control possibilities. Small displays
have an advantage over large displays. They need to be held
closer to the eyes to observe the data on the small screen: this
vicinity to the body causes less fatigue. Normally, a user will put
down the device shortly to limit fatigue, but with a complex task
(consisting of multiple depending subtasks) this would destroy
the flow of action. During our observations we noted that many
users tried to rest their arms quasi on the hips to balance the
device construction, relieving the arms and thus reducing
fatigue. We tried to support this pose during the design of the
devices, both for one and two handed interaction modes.
Users hold the UMPC in power grip [20] with both hands,
whereas the sides of the device rest in the palms of the hand. The
grip on the device is partly suitable for a power grip, since it can
not be grasped around fully. Hence, the user can only put lower
force on the grip. Putting force on a grip is important to balance
the device and to avoid slipping. Without force, chances are
higher that fatigue will occur. In addition, a slipping grip often
leads to difficulties in using the controls. The strength a user can
put on a grip depends on the opening (fingers to palm) of the

46

Figure 3 Unnatural wrist turn (A), better wrist turn with button
access problems (B)

The UMPC can be held with one hand for a short time, but the
uneven weight distribution will cause the device to tilt to the
opposite side. Balancing the tilt can cause noticeable fatigue: the
fulcrum (balance) of the arms and wrist lever system
biomechanics causes counterbalance muscular activity which
usually pushes the user to apply a two-handed grip. Hence, onscreen interaction, for example with a pen, is tedious and short.
In addition, when external devices are packed to the back of the
UMPC, the construction does not only tilt to one side, but also to
the back during one-handed usage. The balance further restricts
use of controls: accessing the controls at the top side can be
tedious and causes even more fatigue itself, strengthened by the
un-ergonomic turn of the wrist. To use the rather small and flat
controls on the (planar) surface of the device, the user has to
angle the thumb flat on top of the device (Figure 3A). Instead of
using a non-straight, more ergonomic angle (Figure 3B), the
wrists need to be turned towards the device. Slightly bigger
controllers would be beneficial to avoid these access problems.
Outcomes

Pose: the analyzed UMPC constructions forced most

users to take a two-handed grip. Additional devices
disturbed the weight balance and shift, especially during
one-handed grip. Specific poses (hip pose) relieve the
weight to some extent.

Grip: a power-grip is used to balance the weight. Due to

an unnatural wrist angle, the current grip does not afford
fine-motor actions well. A separation of grip and display
could result in a decoupled angle of the display and grip,
possibly relieving the wrists. A better grip (likely a
vertical handle) that allows more force could improve
the ergonomics considerably.

Controllers: the placement and size of the controllers is

not ideal on an UMPC, especially since the user is forced
to make unnatural wrist movements. Bigger and better
placed controllers would be beneficial. Besides, the grip
only allows users to access controllers with the thumb
instead of with any other finger, limiting interaction.
Especially fine motor actions can hardly be supported.
4.3
Functional allocation
The next stage involved matching the functionality of the
target application to controllers. Looking back at Table 1, most
of the functionality can be performed using a limited number of
controllers. Furthermore, most of the interaction being

performed directly on an object is rather more two-dimensional

than three-dimensional: the only real exception is viewpoint
control. As a result, we came up with the following control
allocation plan: the construction should hold several standard 1D
buttons (for system control or visualization mode changes) and
2D controllers to perform most of the other functions. For the
2D control (in particular moving lenses in a constrained way),
we chose micro-joysticks slightly bigger than the UMPC
version. We initially left pen input out as an option, since it did
not seem feasible due to the weight shifting of using the device
in single-handed mode.
During the mapping of functionality, we considered how users
could also use the fingers to access controls, instead of just the
thumb. When mounting controls on a vertical handle, some can
be used well by the thumb, whereas others may be primarily
meant to be controlled by the index finger. Such a hand posture
closely resembles using a pistol or traditional joystick. Using
either one of the controls has specific characteristics. Whereas
the index finger affords finer control due to its high dexterity, it
may also lead to unstable control, since the grasp on the grip
loosens slightly. The grip requires higher force, which is
lowered when the index finger is used to move or press a
controller. The lower force results in a slight misbalance of the
device, which is especially a problem when the construction is
used single-handed (see next section). The thumb provides lower
grained control, but since the fingers push the grip in the palm of
the hand, the misbalance is lower.
Outcomes

Map functions on buttons and micro-joysticks that can be

accessed by thumb and other fingers
4.4
Control-body linkage design
After we defined the boundaries for the grips and the
controllers, we needed to find a construction form which would
hold all the peripherals and handles with controllers. In
addition, we investigated the potential of using a single handed
grip. Though holding less devices, a construction made by
Grasset et al [2] inspired us to take notice of a possible single
grip from below.

Figure 4 wheel mouse (A) and externalized handle (B)

Grip design
As we found out before, users need to hold the construction in
a power grip. The usage of a power grip fits with the mostly
viewpoint oriented usage of handheld AR, in which the device is
moved over the real world to change the viewpoint. To
improve the way users hold the UMPC and the additional
devices, we performed a grip study. Using clay and plastic
parts, we experimented with different forms and observed their
relationship with the placement of controllers. We started

experimenting with abstract grips to discover the actual

consequences of using different grip positions to hold the
UMPC. We used hollow metal and plastic pipes to which we
attached different controllers, using parts from off-the-shelf
hardware. These parts are small and have drivers available,
making them a good choice for explorative approaches [21].
At this stage of the process, we experimented with general
forms of grips, in which different kinds of controls could be
embedded. The main concern was how to fit in all the
electronics. Simultaneous with the placement efforts, we
searched for easily connectable and small controllers, exploring
if normal micro-joysticks and buttons could be usefully replaced
by alternatively functioning ones. Previous handheld AR
interfaces hardly made use of alternative controllers, for which
reason there is hardly a base to compare performance. We
adopted a MIDI interface from Infusion Systems, trying out a
whole range of smaller and larger controllers, including forcebased sensors. Whereas the wireless MIDI interface is rather
small, it needs batteries that will drain over time (thus need to be
replaceable). Alternatively, one can obtain particularly small
USB boards from mice, or small programmable boards. One
consideration was to externalize all the controllers outside the
grip, but we abandoned this idea to keep a level of selfsustenance: the grip must be easily exchangeable, in case the
user requires a control structure with different controllers.
Due to the envisioned two-handed usage of the construction
and the requirement for a vertical handle, we first adopted
designs that placed two handles at the side of the UMPC case.
We also experimented with placing parts at the backside, by
mounting a wheel mouse that could be pressed, and moved to
the left and right (Figure 4A). A second controller simply used
some buttons. Though the buttons worked out fine, the wheel
did not. As Figure 4A clearly shows, the fingers need to be
stretched out rather far, thereby loosing grip on the side of the
device. Alternatively, a micro-joystick controller on top of a
pipe performed fine. We used a gamepad micro-joystick in an
inverted way (Figure 4B), following an idea to use it as sensing
element between two movable parts. Technical difficulties and
grip force limitations of a split-part joystick made us abandon
this idea.

Figure 5 Clay model, hollow plastic/metal mockup and metal

pipe with removable (Velcro) pressure sensor

Throughout the process, we went through multiple studies

taking plastic boxes or foam (resembling devices that need to be
put in), embedding them in an outer case of clay or plastic to
form more precise representations of grips (Figure 5). From
these studies, we obtained several results. First, and not
surprisingly, the grip itself resembles that of similar devices like
a drill or joystick: there is not much to optimize in the form
outside its scale. Second, size turned out to be a problem. Even
with miniaturization of components, one still needs considerable

47

space to put in all the hardware into a grip. Due to the sizes of
the boards and the hidden parts of controllers, the grips had to be
greatly optimized. Finally, the grip is more effective when it
allows the user to grasp around the handle, supporting a more
powerful grip (as previously noted, 76mm from finger to palm).
Some distance between the UMPC and the grip ensures that the
user can really grasp around the grip without being restricted by
the UMPC case.
Configuration of the construction
Having an idea how the handles would look like, we made
mock-ups with foam to find out how the handles could actually
connect to a construction holding all the peripherals. We started
placing handles around the UMPC in different configurations
and analyzed how users would hold the construction, focusing
specifically on the effects the pose would have on limiting
fatigue. The configurations had to comply with several
restrictions. First, the GPS antenna and the external camera had
to be placed at specific locations. Second, the placement
configuration of all devices should be such that weight balance
would be optimal: the construction should not tilt to one side,
leading to user fatigue.
The first device configuration idea was to load all peripherals
below the UMPC like a SLR battery pack, but space turned out
to be a problem. Especially the cables take considerable space
that could not be accommodated, and the camera and GPS
antenna could not be attached easily. We returned to the initial
idea of mounting most peripherals behind the UMPC. The basic
form of the base ended up resembling an L, giving the UMPC
a steady stand.
Ideally, the handles should be attached in the horizontal
weight equilibrium: this equilibrium runs through the middle of
the base, while behind, in front, left and right the weight is about
the same. Calculating the approximate weight of the case and the
peripherals against the weight of the UMPC, we placed the
handles directly behind the back of the UMPC. In addition, we
made some basic arrangements to mount the camera and the
GPS receiver externally. The final form of the construction
resembles a Formula 1 or airplane steering wheel working
well for navigating around objects: the user feels like flying
around objects (see Figure 1A).
Subsequently, we worked on the idea of a single-handed
version. The main idea would be to detach both handles from the
sides. One handle could be mounted below; the second handle
could be put away, or used for freehand (spatial) interaction.
Following the weight distribution factors, we tried to mount the
handle directly below the center of weight. However, due to
limitations in accessing controls at the top of the handle, it had
to be mounted slightly in front of the weight equilibrium. In the
second development stage, we developed a new handle for
single-handed usage that can be mounted directly below the
center of weight.
Outcomes

A steering-wheel-like construction with two handles

detached from the sides provides for an ergonomic way
of holding the UMPC construction.

The handles need to have a suitable power grip; this grip

resembles joystick or drill grips.

Controllers accessible by multiple fingers increase

interaction possibilities.

Off-the-shelf devices form a good source of controllers

to embed in design studies.

48

4.5
Electronic design
Based on the controller allocation plan, we mounted an analog
and a digital joystick, usable by index finger and thumb
respectively. The idea was to map constrained interaction
techniques on both micro-joysticks to control specific axes in
translational task with a dedicated controller (see Figure 6). We
also included 3 thumb-operated buttons, and one that could be
reached by either the index finger or the middle finger. All
controls are mapped to a USB board from an off-the-shelf
joystick. The second handle is a test bed for alternative,
unconventional MIDI controllers, including quasi haptic input
methods relying on touch sensitive Piezo sensors. Initial
experimentation shows that the usage of Piezo based elements
has limitations: in single handed configuration, the force needed
by the fingers to balance the construction prevent fine-grained
control. Thus, only the secondary control unit includes a Piezo
sensor accessible by the thumb, and a bend sensor that can be
used by the index finger. The latter can be used well to adapt the
visualization mode. In addition, but not further discussed in this
article, a wireless camera and laser pointer are mounted in the
joystick for additional tracking and interaction purposes.

Figure 6 Larger handles with the USB and MIDI controllers

After the first evaluation, we created a smaller handle that

could be mounted directly in the weight equilibrium. Inspired by
a previous controller (see Figure 4A), we included a wheel with
middle, left and right clicks. The sideward movement can be
used for directional control. The handle is shorter than the first
handles: in user study 1, we observed that some users grabbed
the longer handle rather low, leading to more tilting than needed.
A shorter handle forces users to grab the device closer to its
weight equilibrium, leading to better weight balance.
Finally, a connection plan was laid out, including the
attachment and de-attachment of handles and the cabling to
connect all peripherals to the UMPC. We adopted a simple
screw mechanism to support multi-configuration, affording both
single and two handed interaction. To get the data out of the
handle, we re-used on-board connectors from input devices,
which can easily be plugged in and out of the handle. All
connectors lead to a modified USB hub. Both the cables and the
HUB save a lot of space in the base in comparison to standard
cables and hardware. The secondary handle does not need
special connectors, since it makes use of MIDI over Bluetooth.
Nonetheless, it needs Li-ion batteries to operate.
Outcomes

Using off-the-shelf hardware, we mounted USB and

MIDI controllers in the different handles and optimized
the cabling.

MIDI controllers provide for alternative input, but usage

is restricted.

4.6
Assembly
The last stage consisted of fine tuning the models in CAD,
adapting the forms and getting the grips right. The models were
smoothened, getting its final form now known as BatPack (the
case holding the peripherals), and the Wings (the handles). We
named the device construction VespR, derived from the Latin
word for Bat. The models were printed using stereolithography,
and layered with a rubber-like material, to improve the grip on
the devices. The rubber is a velvety material, comfortable in the
hands and hygienic (cleanable with water). Its softness prevents
extended usage since it peels a bit, but it is easy to process in
production. The rubber is extremely good for providing a nonslippery grip, a real advantage for the weight imposed on the
hands and fingers.
5

EVALUATION

We performed two user studies to evaluate different aspects of

VespR. The first test was set up as a user attitude evaluation,
exploring a more application specific setup. The second test,
intended as comparative study, evaluated different device setups
in a more abstract task setting. Both tests looked specifically into
pose, grip and controller issues. We hypothesized that at least
the two-handed version of VespR would obtain high ratings for
ergonomic usage. We speculated that the single-handed version
would prove useful, though weight restrictions and associated
fatigue would limit its practicality.
5.1
First evaluation
In the first evaluation, 17 users (16 male and one female, all
with a computer science background, but no AR/MR specialists)
made use of the VespR in single handed mode. The large
handle with the micro-joysticks was mounted below the BatPack
(Figure 1B); the other hand could be used freely. The users
could explore an excavation site, using several magic lens tools,
but did not have a specific task sequence they needed to follow.
Average usage time was about 10 minutes, during which the
evaluators observed users and were open for discussion. After
the practical part, users were asked to fill out a questionnaire
with 13 questions (7 point Likert scale, higher scores are better).
We predominantly focused on the ergonomics of the device
construction and grips, next to the usage of the controls. In
addition, we stated some questions regarding optical issues and
the user interface.
Results
The overall weight of the device was not rated negatively, but
obviously users preferred using a lighter device construction.
Both the weight balance and the grip on the construction were
rated mid range. Nonetheless, most users did not report having
noticeable fatigue, even after using the construction for about 10
minutes. Consequently, the comparative study (section 5.2)
needed to provide better insight on the weight, balance and grip
issues. The placement of the controllers was rated positively,
similar to the control effectivity. The interaction techniques (not
discussed in detail here) also performed well, in the range of
avg. 5.00 and 5.50.
The test showed the weight balance could still be improved.
The device slightly tilts to the back resulting in some fatigue
after extended usage. Better weight balance could be achieved
by placing the handle further to the center of the weight
equilibrium. Searching a better balance, a second smaller handle
was created, with the changed control structure (see section 4,
electronic design).

We also observed that several users held the device in an

unintended but obviously convenient manner: by holding the
BatPack itself, instead of the handle. Five users held the handle
and used the second hand to grab the BatPack below or behind
(see Figure 1B). Another user grabbed the BatPack with both
hands reaching the handle only to control the application. In the
second evaluation, we intended to observe this phenomenon in
more detail, by using the support hand as one condition. The
first impression was that the non-dominant hand was used to
relieve the force on the dominant hand, providing a steadier grip
and avoiding tilting during interaction.

Question

Avg

Stdev

Correlation
with second
study

Overall weight

3,93

1,58

+/-

Weight balance

4,07

1,10

+/-

Ergonomics of grip

4,27

1,16

Grip material

4,80

0,94

Fatigue

5,21

1,58

Placement of controllers

5,07

1,39

+/-

Switch between focal planes

5,20

1,37

Switching nuissance

5,38

1,56

Control effectivity

4,80

1,42

+/-

Table 2 Results of first evaluation. Correlation + means the

rating was higher/better than the second evaluation.

Overall, the results of the test were quite positive, even though
many ratings were mid-range. The explanation, supported by the
second evaluation, is quite simple: single-handed interaction is
tedious, and clearly affects the rating. The weight and associated
fatigue (even when not rated dramatically in test 1) lowers the
overall attitude towards the device.
5.2
Second evaluation
This evaluation compared traditional UMPC setups with the
different configurations that VespR affords. 15 users (12 male,
3 female, 14 right handed, one left handed) participated in the
test. All users had a computer science background, 4 people
were non-specialists in the field of AR/MR. Every condition
took about 5 minutes to perform, totaling the test time to about
40 minutes (including answering the questionnaire). In total, we
used 5 different conditions: UMPC only (535 grams), the old
UMPC device construction shown on Figure 2 (using the buttons
on the UMPC and a small camera, 739 grams, referred to as
UMPC with plastic case), VespR with two large handles
mounted on the sides (totaling 1249 grams, referred to as
handles at side), VespR with one large handle mounted below
(the one with micro-joysticks, 1091 grams, referred to as big
handle below), VespR with the newer, smaller handle mounted
below (1105 grams, referred to as small handle below). The
different weights provided us with insights in the weight balance
and ergonomic factors influencing fatigue.
The test was laid out in two spatial areas: a selection and a
placement area separated by about 5 meters. The selection area
consisted of a poster placed at eye-height with markers over
which different buildings were overlaid. The placement area
consisted of two posters with a city map. One poster was placed
on the wall at eye sight (wall-mode), one on a desk (tablemode). These different modes allowed us to specifically focus
on how pose affects the usage of the different constructions.

49

Within the results analysis we point at some of the differences

between both poses, caused by weight balance and forces on the
hand, wrist and fingers. The users had to place the picked
building on the appropriate site, placing 2 objects (buildings) at
each wall and table placement area. Upon placement, objects
snapped to a specific position close to the final one. The
snapping forced users to perform some translational actions.
During the test, we noted down the specific pose and grip
observations, and watched muscular activity to detect signs of
fatigue. Afterwards, users were asked to answer 13 questions
(total 95 answers), using a 7 point Likert scale questionnaire
(higher is better).
Weight, balance and operation
The overall weight was perceived as very good for both the
UMPC and the UMPC with plastic case. The rating was
followed by a still very good note for the VespR with side
handles, considering that it weights about 500 grams more than
the other constructions. The single-handed versions both scored
the same: quite acceptable, but felt to be on the heavy side.
Actually, both are about 100 gram lighter than the two-handed
version, but the weight mostly leans on the dominant hand.

all configurations are appropriate for one handed viewing

(navigation only tasks). The UMPC-only condition scored well
in both modes, with the plastic case only slightly lower. Still, the
single-handed VespR outperformed them both, with the smaller
handle (wheel stick) performing best in all modes.
Fatigue
The results on fatigue were in line with the weight balance
ratings: the UMPC only, the UMPC with plastic, and the VespR
in two-handed condition performed good to very good, whereas
the single handed version performed worse. Due to the mixed
duration of the tests (5 minutes per configuration and 25 minutes
maximum usage), it is difficult to truly grade the fatigue ratings.
The 25 minutes were not completely continuous: the arms of the
user could relax in between configuration changes, even though
these mostly took just 20 seconds. Nonetheless, we believe that
for normal usage of the device, the ratings are quite
representative, even though making a true duration test would be
an interesting and needed alternative.
Device

Mode Fatigue
avg

Device

Mode

Overall
weight
avg

UMPC
only
UMPC
with
plastic

6.53

Weight
Balance
stdev Avg
0.83

T
W

6.07

1.03

W
Handles
at side
T
Big
W
handle
below
T
Small
W
handle
below
T

5.00

3.73

3.67

1.36

1.91

2.02

6.67

Operate 1
Hand

stdev avg
0.82

2.10

1-handed
view

stdev avg
1.73

5.70

stdev
1.95

6.73

0.59

2.10

1.64

5.60

2.12

6.00

1.20

1.80

1.65

4.70

2.21

6.07

1.10

1.90

1.90

4.90

2.42

6.13

1.30

6.20

1.01

4.07

1.75

4.60

1.64

6.07

1.58

4.27

1.75

4.07

1.67

5.60

1.76

4.00

1.81

4.53

1.46

6.33

1.35

4.07

1.83

3.93

1.39

6.00

1.41

Table 3: Weight, balance, operation ratings. (W = wall, T = table)

As expected, the weight balance was very good for the UMPC
only, independent if used in table or wall-modes. Through its
light weight, the UMPC-plastic case combo was also rated well
in both modes. The two-handed VespR received high ratings,
confirming our work on weight balance. Again, considering that
the two-handed version is considerably heavier than the
traditional light weight UMPC construction, we believe this
result is very good. In line with the first evaluation, the singlehanded configurations performed less good. Users did not notice
a major difference between the two handle placements, event
though the smaller handle is placed far closer to the weight
equilibrium. Possibly the weight was too high for one hand that
a difference could not be clearly noticed, even when the second
hand was used to support the BatPack.
Single-handed operation ratings changed the appreciation of
quality for some devices considerably. Users rated the usage of
both UMPC and UMPC with plastic case as extremely bad for
one handed interaction. Both single-handed VespR scored much
better with acceptable, but not impressive scores. This indicates
that we improved the construction to such extent that singlehanded interaction (control) of applications is better than with
the traditional configurations. Nonetheless, users believed that

50

UMPC
only
UMPC
with
plastic

Regrasp
weight

stdev avg

5.87 1.19

Regrasp
control

Comfort

stdev avg

stdev avg

stdev

6.73

0.46

6.13

1.46

5.33

1.54
1.54

6.00 1.07

6.73

0.46

6.20

1.42

5.33

5.20 1.70

5.67

1.99

5.40

2.23

4.00

1.89

5.47 1.41

5.80

1.86

5.47

2.10

4.07

1.79

5.27 1.49

6.33

1.11

6.47

0.74

6.07

1.39
1.36

W
Handles
at side
T
Big
W
handle
below
T
Small
W
handle
below
T

5.60 1.35

6.40

1.05

6.47

0.74

6.00

4.07 1.79

4.47

1.84

5.07

1.71

4.27

1.94

4.00 1.81

4.53

1.77

5.00

1.73

4.13

1.84

4.00 1.81

4.27

1.91

4.53

2.07

4.60

1.80

3.93 1.62

4.40

1.80

4.40

2.03

4.27

1.53

Table 4: Fatigue factors ratings. (W = wall, T = table)

Results on regrasping showed that at least for short usage

durations and applications with limited button access, the
buttons on the UMPC are usable: the UMPC only and the
UMPC with casing scored very well. This surprised us, since the
grip did not seem very comfortable when observing the users.
The two-handed VespR performed extremely well in both
wall and table mode. Even with the increased weight, no
regrasping of the construction / grips was necessary. In line with
the notes on weight, both single handed grips performed worse.
The increased force on the single hand clearly resulted in
regrasping due to fatigue on the hand holding the joystick, even
when the construction was balanced by the second hand.
With regards to user comfort, the results varied more, clearly
in favor of the two-handed VespR configuration. The UMPC
only still scored reasonably well in both wall and table mode, the
UMPC with plastic case only proved mediocre. The two-handed
VespR received a high score in both modes, supporting our
ergonomic studies and design of the device. As expected, both
the single handed grips performed with acceptable ratings in
both table and wall modes.
Grips and interaction
Investigating the ergonomics of the grips and the interaction
with the controllers, we received some surprising results. None
of the device grips rated extremely well. Surprisingly, people did

not mind holding the UMPC, which was rated about the same as
the big handles we used beside and below the BatPack. The
small handle received a lower score, whereas the UMPC with
the plastic case performed worst. We have the strong impression
users only rated ergonomics of the construction as one instead of
the joystick: hardly without exception, the large joystick
performed extremely well in the two-handed configuration,
while the same joystick received lower ergonomics values in the
single handed configuration.
The material of the joysticks, the velvety rubber, was highly
appreciated by the users (avg. 6.00 / stdev 1.20). The low rating
of the UMPC with plastic could be higher if it had a better
mounting for connecting the UMPC: multiple users reported on
being afraid to drop the UMPC from the casing.
Users found the UMPCs buttons easy to reach. Both the big
and the small handle scored about equally well. Surprisingly,
when asking the users about the placement of the controllers on
the big handles in two-handed configuration, the score was much
higher. Obviously, the weight balance, pose and force on the
hand / wrist have a large effect on how the user reaches and uses
the controllers.
Device

Grip
Mode ergonomics
avg

UMPC
only
UMPC
with
plastic
Handles
at side
Big
handle
below
Small
handle
below

stdev

5,30 1,77

Control
placement

stdev avg

stdev

6.11

1.27

1.18

T
W

4,00 2,16

T
W

3.73

1.10

3.53

1.30

3.67

1.29

6.00

0.88

6.13

0.64

6.20

0.56

5,20 1,82

5.47

1.68

5.00

1.77

4.93

1.79

4,80 1,86

5.33

1.72

4.60

1.64

4.33

1.80

T
W

3.60

5,20 1,82

T
W

Effectivity

avg

Table 5: Grip and controllers ratings. (W = wall, T = table)

We received quite diverse ratings on effectivity. All UMPC

conditions had similar interaction mechanisms: select and drop
objects with a button and move them with the finger mouse. The
two-handed configuration made use of the same mechanisms as
the single handed big handle configuration: objects were
selected by button click and moved in a constrained way using
the micro-joystick. For the wheel mouse, we also constrained the
interaction. Users could select a specific axis using one button
and scroll the wheel to translate the object over this axis.
We observed that most users had problems with the UMPC
only condition due to limitations while translating objects. Since
the micro-joystick is mapped to the mouse pointer, once it hits a
border it does not translate in this direction any further. Thus,
users needed to clutch by tapping on the screen, to move the
mouse pointer back. This obviously limited interaction for some
users. On the other hand, the wheel mouse has a mechanical
disadvantage, since users need to put too much force on the
wheel to click in either direction. The low rating of the wheel
joystick was not only due to the mechanical construction, but
also in relation to the force on the hand in single handed usage.
Without support, the index finger can hardly control the wheel,
but is rather used to balance the construction. At the end, and
most importantly, the two-handed VespR configuration
completely outperformed all others.

DISCUSSION OF EVALUATION RESULTS

The outcomes were in line with our hypothesis that the twohanded VespR is a worthy alternative to the older UMPC
constructions. The second evaluation clearly showed that the
increased weight of the installation can be dealt with in an
ergonomic way: users were very comfortable. It should also be
stated that in simple scenarios, the UMPC without extended
hardware might be useful some of the ratings were more than
acceptable. Particularly if the interaction can be divided in short
subtasks allowing the user to lower the device in between, the
UMPC-only can be effective. However, fine grained interaction
using the UMPC controls is quite tedious. Many users reported
frustration when unable to accurately fine-tune the objects
position (second test) in UMPC only modes. Fine-tuning as
afforded by all VespR configurations was well received.
Although not evaluated in terms of accuracy or duration, finetuning was, in UMPC cases, often aborted by users, while in the
VespR cases it was eagerly performed. The bottom-line is
simple: for complex AR applications, the UMPC-only version
close to useless, since it lacks the necessary quality or
functionality of several devices needed for outdoor usage.
Similarly, some indoor installations require additional tracking
devices that automatically lead to bulkier constructions.
Interaction with the single-handed version is possible, but not
ideal. Both evaluations showed that holding the device singlehanded is not very ergonomic for longer sessions. Leaving the
second hand free for pen-input or real-life communication would
require another approach, particularly for longer sessions. In
single-handed conditions we also noticed some effects described
by Guiard [13]: when placing objects, the non-dominant hand
stabilized the device. The non-dominant hand adjusts the spatial
frame of reference for the dominant hand, supporting it while
performing fine grained actions.
Interestingly enough, users were able to perform fine grained
actions when using the power grip and a thumb based controller.
Furthermore, they were able to ergonomically perform fine
grained actions with their non-dominant hand, which was a big
surprise to us. Currently, the joystick with the index and thumbcontrollers is mounted to the left side of the VespR in the twohanded configuration. Most users, though right-handed, could
control the application perfectly. We do not have an explanation
for this phenomenon, even though we believe that similar
performance can be found in game pads, where users also make
use of the non-dominant hand to perform fine grain interaction.
Finally, comparing the two interaction modes (wall vs. table),
all two handed versions rated higher for interactions in table
mode, while for the single-handed VespR the ratings were
higher in wall mode. In single-handed conditions, the pose
causes the weight balance to tilt to the back when lowering the
device (table mode), deteriorating interaction. Two-handed
setups have the advantage that the interaction mode does not
affect weight balance, while the lower position of the device in
table mode causes less strain [19], as noted in section 4.2.
Take home message

An increase of weight does not necessarily lead to a

decrease of ergonomics: the higher weight can be dealt
with in a suitable way by better balancing weight,
among others by allowing specific poses that relieve
the arms

An ergonomic construction with better controllers and

tracking/optical hardware can considerably improve
interaction in outdoor AR, but may not always be
needed for indoor setups or simpler applications

51

CONCLUSION
VespR significantly advances ergonomics of handheld
devices in such a way that users perform comfortably even with
double the weight of other devices. Extra weight is allocated to
higher quality sensors providing developers the opportunity to
improve registration, presentation, and interaction for AR
applications. VespR has been continuously in use since its
construction by researchers and end-users, for demonstrations
and practical applications.
Even with the trend in miniaturization and integration of
devices, the results of this study are valid for further
developments in the area. Future UMPCs will likely integrate
more or better devices that are currently externalized. Still, AR
will keep putting requirements that cannot be completely
matched, including special tracking requirements. Furthermore,
technological improvements can not ignore human factors if
acceptance is expected in the long run.
We are currently optimizing the VespR by improving the size
and internal organization of the BatPack, and performing further
experiments with the MIDI controllers. In addition, we are
planning additional evaluations of longer duration outdoor
sessions.
ACKNOWLEDGEMENTS
We would like to thank Erick Mendez and Gerhard Schall for
aiding us with the hardware setup material, ideas and the
evaluations at Ubicomp. We also thank Albert Walzer and
Stefan Zedlacher for aiding in the construction of the devices.
This work was partially funded by the Austrian science fund
FWF under grant Y193.
REFERENCES
[1]

Kruijff, E. and E. Veas. Vesp'R: Transforming Handheld

Augmented Reality. In Proceedings of the Sixth IEEE and ACM
International Symposium on Mixed and Augmented Reality

[2]

Microsoft Research.
[10] Jones, M. and G. Marsden, Mobile interaction design. 2006: John
Wiley & Sons Ltd.
[11] Bullinger, H., P. Kern, and M. Braun, Controls, in Handbook of
Human Factors and Ergonomics, G. Salvendy, Editor. 1997, John
Wiley & Sons.
[12] Zhai, S., Human Performance in Six Degree of Freedom Input
Control, PhD Dissertation, Dept. of Computer Science. 1995,
University of Toronto.
[13] Guiard, Y., Symmetric Division of Labor in Human Skilled
Bimanual Action: The Kinematic Chain as a Model. The Journal
of Motor Behaviour, 1987. 19(4): p. 486-517.
[14] Kabbash, P., B. Buxton, and A. Sellen. Two-Handed Input in a
Compound Task. In Proceedings of the 1997 ACM Conference on
Human Factors in Computing Systems. 1997: ACM Press.
[15] Pering, T., Y. Anokwa, and R. Want. Gesture connect: facilitating
tangible interaction with a flick of the wrist. In Proceedings of the
1st international conference on Tangible and embedded
interaction. 2007.
[16] Reitmayr, G. and T. Drummond. Going out: Robust tracking for
outdoor augmented reality. In proceedings of ISMAR 2006. 2006.
[17] Rohs, M. Visual Code Widgets for Marker-based Interaction. In
IWSAWC'05: Proceedings of the 25th IEEE International
Conference on Distributed Computing Systems - Workshops
(ICDCS 2005 Workshops). 2005.
[18] Schall, G., E. Mendez, E. Kruijff, E. Veas, S. Junghanns, B.

Ubiquitous Computing, Special Issue on Mobile Spatial

Feiner, S., B. MacIntyre, T. Hllerer, and A. Webster. A Touring

Interaction, 2008. To appear.

[19] Marras, W., Biomechanics of the Human Body, in Handbook of
Human Factors and Ergonomics, G. Salvendy, Editor. 1997, John
Wiley & Sons.
[20] Zhai, S., P. Milgram, and W. Buxton. The Influence of Muscle

Thomas, B., V. Demczuk, W. Piekarksi, D. Hepworth, and B.

Groups on Performance of Multiple Degree-Of-Freedom Input. In

Gunther. A Wearable Computer System for Augmented Reality to

Proceedings of the 1996 ACM Conference on Human Factors in

Wagner, D. and D. Schmalstieg. Handheld Augmented Reality

Displays. In Proceedings of the 2006 IEEE conference on Virtual
Reality (VR2006). 2006.
Schmalstieg, D. and G. Reitmayr. The World as a User Interface:
Augmented Reality for Ubiquitous Computing. In Proceedings of
the Eurographics Central European Multimedia and Virtual
Reality Conference. 2005.
Schall, G., B. Reitinger, E. Mendez, S. Junghanns, and D.
Schmalstieg. Handheld Geospatial Augmented Reality Using
Urban 3D Models. In Proceedings of the Workshop on Mobile
Spatial Interaction, ACM International Conference on Human
Factors in Computing Systems (CHI07). 2007.

52

Force: Mobile Device Interaction with Force Sensing. 2008,

Multimodal AR Interface: A New Handheld Device for 3D

support Terrestrial Navigation. In Proceedings of ISWC'98. 1998.

[7]

Scott, J., L. Brown, and M. Molloy, I Sense A Disturbance in the

Underground Infrastructure Visualization. In Personal and

1997.

[6]

Augmenred Reality. 2001, Laurence Erlbaum Associates.

[9]

Grasset, R., J. Looser, and M. Billinghurst. A Step Towards a

exploring the Urban Environment. In Proceedings of ISWC'97.

[5]

Augmented Reality, in Fundamentals of Wearable Computing and

Reitinger, and D. Schmalstieg, Handheld Augmented Reality for

Machine: Prototyping 3D Mobile Augmented Reality Systems for

[4]

Rekimoto, J., NaviCam: A Palmtop Device Approach to

(ISMAR 2007). 2007.

Interaction. In Proceedings of ISMAR'05. 2005.

[3]

[8]

Computing Systems (CHI'96). 1996: ACM Press.

[21] Bowman, D., E. Kruijff, J. LaViola, and I. Poupyrev, 3D user
interfaces : theory and practice. 2005: Addison-Wesley.