Proceedings of AVI’04 , pp. 23 – 31.

10/6/2003 3:09 PM
Stitching: Pen Gestures that Span Multiple Displays
Ken Hinckley, Gonzalo Ramos, Francois Guimbretiere, Patrick Baudisch, Marc Smith
Microsoft Research, One Microsoft Way, Redmond, WA 98052
{kenh, baudisch, masmith},,
Stitching is a new interaction technique that allows users to
combine pen-operated mobile devices with wireless
networking by using pen gestures that span multiple
displays. To stitch, a user starts moving the pen on one
screen, crosses over the bezel, and finishes the stroke on the
screen of a nearby device. Properties of each half of the
user’s pen stroke are observed by the two separate devices,
synchronized via wireless network communication, and
recognized as a unitary act performed by one user, thus
binding together the devices.
We identify the general requirements of stitching and
describe a prototype photo sharing application that uses
stitching to allow users to copy images from one tablet to
another that is nearby, expand an image across multiple
screens, establish a persistent shared workspace, or use one
tablet to present images that a user selects from another
tablet. Preliminary usability testing suggests that users find
it compelling to have a straightforward means to combine
the resources of multiple mobile devices. We also discuss
design issues that arise from proxemics, the sociological
implications of users collaborating in close quarters.
entry of network addresses and the geometry of displays.
We propose stitching as a new interaction metaphor that
uses commonplace pen input capabilities to establish and
manage serendipitous connections between pen-operated
mobile devices. A stitching gesture consists of a continuous
pen motion that starts on one mobile device, continues past
the bezel of the screen, and ends on the screen of another
device (Fig. 1). Such gestures serve as the basis for a
flexible architecture that allows users to dynamically bind
together mobile devices. Stitching can be implemented on a
variety of pen-operated devices, does not conflict with
existing uses for the pen, and provides a versatile
framework that can accommodate future extensions.
path taken
by the pen
Author Keywords
pen computing, mobile devices, distributed systems,
proxemics, gestural input, synchronous gestures
ACM Classification Keywords
H.5.2 Information Interfaces and Presentation: Input
Much recent discussion has focused on the promise of the
wireless internet, yet there has been relatively little work on
techniques that help users of mobile devices to collaborate
with others and to share information with other persons
[17][18][22]. For example, when attempting to copy a file
between mobile devices, it remains difficult for users to
name a device to connect to, specify how to connect, or
indicate what information to share.
While several previous systems foster collaboration
between ubiquitous devices [15][19][21][23], they may
require special hardware such as overhead cameras or
unique ID tags on each device, or they may require manual
Submitted to CHI 2004
Fig. 1 Here, a user gives some photos to another user by
dragging them from the top tablet to the bottom tablet.
From the user’s perspective stitching seems like a single
cognitive chunk [3], but the gesture actually specifies a
number of properties of the connection:
• It selects which devices participate in the connection.
• We can phrase together selection of operands (e.g., a
file to copy) and commands (how to combine the
devices) with the stitching gesture itself.
• By fitting a line to the pen coordinates as they leave
one device and enter another, we can calculate the
approximate spatial relationship between the two
devices. This allows us to place graphics or provide
feedback that appears to span the displays (Fig. 1).
10/6/2003 3:09 PM
We describe a prototype photo sharing application for the
Tablet PC that supports operations such as copying images
from one tablet to another that is nearby, establishing a
persistent shared workspace for collaboration, expanding an
image across multiple screens, or using one tablet to display
a slideshow of images that a user selects from another
tablet. Usability testing suggests that users readily grasp
stitching, and find it compelling to have a straightforward
means to perform cross-device operations. We also
observed that sociological issues of co-located
collaboration raise several design issues. We found it is
important to support a range of device configurations, from
intimate combinations of devices in direct contact with one
another, to sharing information while maintaining social
distance. The latter requires support of stitching between
devices that are nearby one another, but not touching.
Pick and Drop [18] enables users to pick (copy) an item
from one screen and drop (paste) it onto the screen of
another nearby device. Stitching also supports copying
items between co-located devices, but unlike Pick and
Drop, our technique does not require a unique ID on the
pen; rather, stitching uses the timing and dynamics of the
gesture to pair up pen strokes observed by the devices.
The Pebbles project [17] explores multi-machine user
interfaces that spread computing and user interfaces across
multiple devices. Stitching contributes a versatile
interaction paradigm that could be used to dynamically
form some of the configurations envisioned by Pebbles.
ConnecTables [22] and Smart-Its Friends [14] both form
connections between mobile devices. Smart-Its Friends
connects two devices if they are held together and shaken.
ConnecTables are wheeled tables with mounted LCD
displays that can be rolled together so that the top edges of
two LCD’s meet, forming a collaborative workspace. Each
LCD senses the presence of the other one using radiofrequency identification (RFID) tags. These systems require
special hardware, and support only one type of connection.
Synchronous gestures [13] are distributed patterns of
activity that take on a new meaning when they occur
together in time, or in a specific sequence in time. For
example, the synchronous gesture of bumping two mobile
devices together can connect the devices in various ways
[12][13]. Stitching represents a new type of synchronous
gesture. Bumping provides a means to connect devices that
lack pen input, but stitching is a natural extension of pen
computing that can leverage pen techniques for selecting
commands, operands, and indirect objects [5][10][16].
Co-located collaboration involves users working in close
physical proximity to one another. Proxemics is the study
of how people use the invisible bubble of space that
surrounds an individual [11]. Appropriate social distances
vary between cultures, but violations of intimate personal
space (within approximately 45cm) may produce
discomfort or physical withdrawal to maintain a
comfortable social distance [7][8]; touching is particularly
unwelcome in non-contact cultures [11]. Our experience
with stitching suggests that the proxemics of co-located
collaboration may yield critical insights for designers.
We define a stitch as a gesture, spanning two or more
devices, which establishes a communication infrastructure
or otherwise combines the resources of multiple computers.
In order to provide a flexible and potentially extensible
facility that can support a number of different ways of
combining devices, stitching addresses the following
central design questions:
1. How is a connection established? A user must name the
devices that are involved in a multi-machine operation, and
the system needs to provide feedback to the user(s) of those
devices that a connection has been established.
2. What type of connection is required? The user needs
to be able to choose among several possible ways to
combine the devices. Does the user want to copy a file from
one device to another? Establish a persistent shared
workspace for collaboration? Expand an image across
multiple screens? These all represent multi-device
commands that transcend the barriers between devices.
3. What information is shared? Multi-device commands
may require operands, such as which file to copy to another
computer. Users need mechanisms to select one or more
objects as part of a stitching gesture.
4. How do users share physical space? Proxemics
suggests that the arrangement of spaces can influence
communication; as Hall writes, “what is desirable is
flexibility... so that there is a variety of spaces, and people
can be involved or not, as the occasion and mood demand”
([11], p 110). Interaction techniques that form impromptu
associations between mobile devices should likewise
support the range from users who know each other well and
want to work closely together, to users who are strangers
and want to exchange files while keeping their distance.
5. What is the spatial relationship between the devices?
Several previous systems support features, such as
combining the screens of two devices, that require
knowledge of where one display is relative to another [15]
[19]. Stitching uses the information provided by the pen to
infer the spatial relationship between devices. This also
allows us to provide graphical feedback for multi-device
operations that appears to span devices, as seen in Fig. 1.
6. How do stitching gestures coexist with traditional pen
interactions? Stitching gestures must coexist with existing
uses for the pen including widget interactions, inking,
character entry, and naturally occurring human-human
communicative gestures (such as waving the pen near the
device while discussing the contents of the screen).
The above design questions suggest that stitching
represents a new class of interaction techniques that could
be implemented in a variety of ways. In the remainder of
this paper, we discuss the general concept of stitching in
10/6/2003 3:09 PM
reference to a proof-of-concept photo sharing application
called StitchMaster. With digital photography becoming
widespread, sharing photos with others is a task of interest
to many persons. Also, many of the semantics that we
wanted to explore with stitching, such as expanding an
image across multiple screens or copying objects from one
screen to another, represent useful and compelling
operations for digital photographs. To begin, each user
launches StitchMaster on his own tablet, which displays
that user’s photo collection as a set of thumbnail images.
1. Establishing a Connection
Stitching requires devices that can sense the same pen; the
user names the devices to connect by moving the pen across
them. Since there is a natural order implied by the gesture,
stitching also establishes which machine is the sender of
information, and which machine is the receiver. Some
connection techniques are inherently bidirectional
[9][14][22] and do not naturally provide this information.
Each participating device sends its pen events to a stitching
server, which may be hosted on a machine in the
environment to offload computation from the mobile
devices. The stitching server synchronizes time between the
devices [13] and looks for matching pen traces; when a
match is found, the server sends a stitching event that
informs the two devices of each other’s network address.
Each participating device must know the network address
of the server, but this is the only address needed to
bootstrap the system. In the future, we may instead find this
address via service lookup mechanisms, or by using
wireless signal strengths to locate a nearby server [1].
Stitching Recognition
The stitching server recognizes a stitch by looking at the
patterns of pen activity from each pair of participating
devices. We define an envelope as the time interval during
which the pen is in range of the screen and is moving at a
speed above a predetermined threshold. The stitching server
then looks for two consecutive envelopes from a pair of
devices that match a specific pattern:
• The first envelope must end near the first screen’s
border and last longer than dTmin1 (250 milliseconds).
• The second envelope must start near the second
screen’s border, and last longer than dTmin2 (=100ms).
• The second envelope must start after the first envelope,
but it may occur no longer than dTmax (=3.0s) after the
first envelope. This time interval is sufficiently long to
support stitching between tablets that are within arm’s
reach (a maximum of about 75 cm).
We found these criteria suffice to recognize intentionally
executed stitching gestures, but just as importantly, false
positives were not a problem: incidental pen motions from
two users concurrently using pens on their mobile devices
rarely satisfy these criteria. The main limitation of our
stitching recognition algorithm is that it is difficult to
perform a stitch that starts close to an edge of the screen;
the system does not receive sufficient samples of the
changing pen location, before it leaves proximity, to be
certain of whether the user is lifting the pen from the
screen, or whether the user is sliding the pen towards
another device to perform a stitching gesture. Also, our
current implementation limits stitches to pairs of devices.
User Feedback for Stitching-
A stitch is recognized as soon as the first 100 milliseconds
of the second envelope have been observed by the stitching
server; it does not wait for the user to finish the motion.
Performing this eager recognition allows us to provide
users with feedback of the stitching gesture as soon as
possible after the user has entered the second screen.
Feedback for a successful stitch consists of a short chirp
sound as soon as eager recognition takes place. If the
stitching gesture includes any operands, then the system
shows a semi-transparent blue shadow on the screen in the
shape of the selected photos (Fig. 2a). Upon completion of
the stitching gesture, the system may also provide
additional feedback. For example, for a copy or move
operation, StitchMaster shows an animated semitransparent
cone that appears to whisk files from one machine to the
other (Fig. 1). This provides clear feedback of where the
files came from, and where they were copied to (Fig. 2b).
Fig. 2 Feedback for remote copy. (a) Shadows appear on
the other device, then (b) when the user drops the photos, a
cone connects them to their origin on the other device.
2. Specifying Connection Type: Multi-Device Commands
Multi-device commands supported by StitchMaster include
copying or moving photographs, establishing a persistent
shared work space, expanding an image across multiple
displays, or entering a presentation mode known as the
gallery (described below). StitchMaster presents these
options in a pie menu. There are two basic design choices
for where the command selection can occur:
• Local menus: Users choose the command (e.g. Copy)
on their local screen, and then stitch to indicate the
remote device that is involved.
• Remote menus: Users stitch to another device, and then
a menu appears on the remote device providing options
for how to combine the devices.
StitchMaster implements remote menus, which allows us to
limit the visibility of multi-device operations to situations
where they are known to be applicable; we did not want to
complicate the single-device user experience with options
for multi-device operations. Remote menus appear at the
end of a stitching gesture when the user holds the pen still
for 0.5 seconds. To provide feedback that a menu is a
remote menu, StitchMaster shows a transparent blue cone
that connects the remote menu back to the display where
the stitching gesture originated (Fig. 3).
10/6/2003 3:09 PM
For some stitching gestures, StitchMaster assigns a default
operation, eliminating the need to use the menus. For
example, when stitching with a selected photo (that is,
stitching using an operand as described in the next section),
by default the selected photograph is moved to the other
screen. We chose not to make copy the default since we
found during pilot studies that users would repeatedly copy
files back and forth while trying out stitching, quickly
creating cluttered screens for themselves.
Phrasing works well, but we observed that users sometimes
become focused on the selection step, and momentarily
forget about stitching. Therefore, we do not require that
stitching follow selection in a single uninterrupted gestural
phrase. A stitching gesture that starts over a selection also
includes that object as an operand, but after 3 seconds, the
selection cools and will no longer be treated as the operand
for a stitching gesture. The highlights around selected
photos turn blue once the selection has cooled. This
approach prevents old, stale selections from mistakenly
being interpreted as the operand to a stitching gesture.
Fig. 3
A remote menu shows a link between screens.
Example Multi-Device Command: The Gallery
The Gallery (Fig. 4) allows one user to give a presentation
of selected photos to another user. To start the Gallery, the
presenter selects an image to start with, stitches to the other
screen, and chooses Gallery from the remote menu. The
other tablet then displays a full-screen view of the selected
image, while the presenter’s tablet displays thumbnails of
all of his photos. The presenter can click on any thumbnail
to change the image that is displayed on the other tablet.
Fig. 4 Gallery: The right tablet displays a full-screen view
of an image that the presenter selects on the left tablet.
The Gallery changes the roles of the devices. Instead of two
identical devices, we now have one tablet for interaction,
while the other primarily serves as a display. If users
separate the devices, but keep the Gallery running, the
presenter’s tablet becomes a private view, while the other
tablet represents a public view of selected information.
3. Specifying What to Share: Stitching with Operands
StitchMaster supports tapping on a single photo to select it,
or drawing a lasso to select multiple photos. StitchMaster
outlines the selected photos in orange and scales them to be
slightly larger than the others (Fig. 5). Users can select a
photo and then perform a stitching gesture to another
device all in one gestural phrase [3][5][10]. The user makes
a selection, and then lifts the pen slightly so that the pen is
no longer in contact with the screen, but is still within
tracking range of the Tablet PC screen. The user then
stitches to the other display, and the selection is treated as
the operand of the stitching gesture.
Fig. 5 (a) Multiple selection using the lasso gesture.
(b) Selected photos scale up and highlight in orange.
4. Sharing Physical Space
With varying social and cultural conventions, individual
preferences, and changing needs depending on the task,
users need flexible ways to share physical space when
combining devices. Hall distinguishes two distances within
arm’s reach, intimate and personal, with social and public
distances beyond that [11]. StitchMaster includes support
for intimate, personal, and social distances.
Intimate spaces support tight collaboration between friends
or colleagues who may need to work together on a large
document. For example, StitchMaster supports placing two
tablets together and then expanding an image to fill both
screens. The displays act as tiles of the same virtual space.
This style is also well suited for a single user wishing to
extend his workspace with additional devices.
Personal spaces. Users can stitch together tablets that are
separated by up to about 75 cm. This allows a space created
by stitching to act as a whole, yet each user maintains his or
her own personal space. For example, StitchMaster allows
users to create a persistent shared workspace by making a
“simple stitch” from one screen to another without any
operands. A vignette that appears to surround the two
screens turns red to give users ongoing feedback that the
two machines are connected. Either user has veto power
over the connection and can close the workspace by
choosing Disconnect from a menu.
Social spaces. Once users join a shared workspace, they
can further separate their devices, yet still work together.
For example, a user can employ the transporter to give
photos to the other user, even if that user is no longer
within arm’s reach. The user drags a photo to the edge of
the screen, and dwells with the pen. After a brief pause,
during which we display an animation of a collapsing blue
square, the photo is transported to the other device. This
10/6/2003 3:09 PM
pause is necessary to separate transporting a photo from
normal drag operations; the collapsing blue square gives the
user feedback that the picture is about be transported.
the user follows an arcing path while stitching. We then
calculate PB as the displacement of p1 along the edge of
Device2’s screen by offset pixels.
Orientation of spaces. StitchMaster only supports stitching
between tablets that are at the same physical orientation, so
users must sit shoulder-to-shoulder. However, research
suggests that communication patterns change when persons
sit face-to-face, shoulder-to-shoulder, or at 90 degree
angles to one another [11][20]. We expect it is technically
feasible and may be valuable to extend stitching to support
pen gestures that span tablets in any of these orientations.
Using this approach, our system can transform points from
one device’s coordinate system to the other, thus allowing
the presentation of graphics that appear to span the devices.
5. Calculating the Spatial Relationship between Devices
To infer the spatial relationship between devices, stitching
fits a line equation to the coordinates traversed by the pen
on each screen. Of course, users do not move the pen in
perfectly straight lines, but users do tend to move in arcs
that can be locally approximated by a straight line.
When the stitching server detects a stitch from Device1 to
Device2, it records a small window of samples as the pen
leaves one screen and enters another, yielding p0 (the exit
point of the first pen trace), p10 (the entry point for the
second pen trace), p11 (the point at which the stitch was
recognized), and α0 (the angle of motion at p0); see Fig. 6.
Direction of Stitching
Device #1
Fig. 6
} offset
first half of
half of
Device #2
α = (α0 + α1) / 2
Fitting a line to the user’s pen gesture.
Due to the sampling rate of the pen, the first and last pen
locations reported by the tablet may fall up to 3-4 cm from
the edge of the screen. We found that calculating the width
of the screen bezel or the extent of any empty space
between the devices by using the time interval between the
last and first observed samples may lead to inaccurate
distance estimates. For this reason, we initialize the
device’s bezel thickness as a fixed constant, and then ignore
any empty space that may be present between the devices.
We estimate the intersection of the stitching gesture with
the edge of each screen, yielding the points PA and p1. PA
is the intersection of the screen edge of Device1 with the
line that passes through p0 at an angle α0; p1 is the
intersection of the second screen edge with line that passes
through p10 and p11 at angle α1. If the line between PA
and p1 has angle α, the offset between the two screens is
then tan(α) times the bezel width. We estimate α as the
average of α0 and α1, which seems to work well, even if
6. Coexistence of Stitching with Traditional Interactions
Stitching must allow users to establish connections between
devices without interfering with existing uses for the pen.
Input states supported by pens [4] include tracking (moving
the pen in close proximity to the screen, causing the
tracking symbol to move), dragging (moving the pen in
contact with the screen, causing an action such as dragging
an object or leaving an ink trail), and out-of-range (the pen
is not in the physical tracking range of the screen).
Stitching can be implemented using the dragging state, or
using the tracking state. StitchMaster implements options to
use either style of stitching, or both can be supported
simultaneously (this is the default).
Stitching in the Dragging State
Since traditional GUI interactions occur in the dragging
state, performing stitching by dragging could conflict with
them. For example, when stitching via dragging, the first
device cannot be sure whether to interpret a pen stroke as a
drag until the second device recognizes the completion of
the stitching gesture. To circumvent this problem and allow
stitching via dragging to coexist with other dragging
operations, we use speculative execution [2]: StitchMaster
initially assumes all pen strokes are intended as drags. If the
stitching server then reports a stitch, StitchMaster undoes
the drag and instead treats the gesture as part of a stitch.
During preliminary user testing, we found that users can
easily make a stroke while keeping the pen in contact with
the screen, but when stitching to another display, the screen
bezel gets in the way. This makes it hard for users to make
a quick, fluid pen motion across the bezel while bearing
down on the pen. Instead, users must drag the pen to the
edge of the first screen, lift the pen to jump the screen
bezel, and then complete the stitching gesture by pushing
the pen back into contact with the second device’s screen.
Stitching in the Tracking State
Stitching from the pen’s Tracking state represents a more
advanced skill than dragging, as it requires moving the pen
while keeping the tip within 2 cm of the screen surface to
prevent it from entering the out-of-range state. However,
stitching by moving the pen just above the surface of the
screen (with the base of the hand resting on the screen)
allows the user to make a continuous, quick, and fluid
movement that is not interrupted by the physical “speed
bump” of the screen bezel. Another advantage of stitching
in the tracking state is that it avoids the need for a
speculative execution scheme: stitching gestures occupy a
separate layer that rests on top of GUI interactions.
10/6/2003 3:09 PM
The main drawback of implementing stitching in the
tracking state is that currently available personal digital
assistants (PDA’s) do not support tracking, so future
extensions of stitching to PDA’s would have to use the
dragging state. Another potential problem is that users may
use the pen to gesture while talking about the contents of
the screen with a colleague, potentially resulting in a falsepositive recognition of a stitching gesture. We designed our
stitching recognition with this issue in mind, so false
positives are rare, but no recognition scheme is foolproof.
We conducted a usability study of StitchMaster to identify
usability issues and user concerns with stitching. Our
primary goal was to verify if users could effectively use
stitching gestures to perform multi-device operations.
Participants: 13 participants were recruited from the
general public through Microsoft’s usability pool. As the
study required pairs of participants, the experimenter
assumed the role of the “collaborating” participant for the
13th participant. None of the paired participants knew each
other prior to the study. Collaborating pairs were of the
following genders: 1 pair female-female, 3 pairs malefemale, and 2 pairs male-male. Four of the participants had
previously used a Tablet PC; an additional 6 participants
had previously used some other type of pen-based device.
Materials: We ran the study on Toshiba Portege 3500
Tablet PC’s with built-in 802.11 wireless networking.
These devices measure 28.5 x 23cm with 25 x 18.7cm
screens. Users employed the tablets in the slate (flat) mode.
Procedure: The participants sat shoulder-to-shoulder on the
same side of a table; the experimenter sat at the opposite
side of the table. Each participant was provided with a
Tablet PC running StitchMaster. In a 2-3 minute practice
session, the participants learned basic pen operations such
as selecting images and dragging images on the screen. The
experimenter then explained how to use features of
StitchMaster, and allowed participants to try them out oneby-one, but did not show users what to do. Each session
lasted approximately one hour.
The experimenter first asked participants to “connect the
devices by making a pen stroke across the devices” but did
not show participants how to do this. With this instruction,
all 13 participants, on their first or second try, created the
persistent shared workspace by stitching with no operands.
All but two users made their first attempt at stitching by
moving the pen in contact with the screen. Participants
expressed no clear preference for performing stitching in
the tracking state versus the dragging state; both seemed to
work well. All participants at some point during the study
used stitching in both manners, and users often would mix
styles within the same gesture: for example, a user would
perform the first half of a stitch by dragging, but then jump
the bezel and complete the stitch from the tracking state.
The experimenter next explained how to move files
between devices by stitching with operands. Originally,
StitchMaster required the user to place the pen in contact
with the other screen at the end of the stitching gesture to
drop a photo. On our first day of testing, we found this was
problematic for users who mixed the dragging and tracking
styles of stitching. Users repeatedly moved the pen away
from the screen when they wanted to drop a photo, rather
than touching the screen.
We fixed this problem by having the software drop the
photo if the user lifted the pen at the end of the stitch. When
asked on the first day if “Sometimes I made mistakes when
moving items” subjects tended to agree, with an average
Likert scale response of 5.5 out of 7, but on the second day
the average response improved to 4.0 (neither agree nor
disagree); one user commented that “it was nice to drag
items to the other screen without having to touch it.”
While moving a photo to the other screen, participants
sometimes would pause too long, which caused the remote
menu to appear. Increasing the time-out for the remote
menu reduced the number of subsequent cases of accidental
activation, but did not completely eliminate this problem.
As we had anticipated, participants sometimes failed to
stitch if they started too close to the edge of the screen.
Participants wanted feedback of how far from the edge of
the screen they had to start stitching in order to be
successful. Adding 1-2cm margins would make this
limitation visible, and prevent users from leaving photos at
the extreme edges of the screen where this problem arises.
The only instances of false positive recognition of stitches
that we observed occurred if users failed to successfully
stitch, and then returned the pen to their screen to try again.
Without realizing it, users often returned the pen to their
original screen while remaining in the tracking state, and
this was sometimes recognized as a stitch.
Overall, users were enthusiastic about the concept of
stitching as embodied by StitchMaster. When asked if “I
would use this software if it were available” the average
response was 6.7 out of 7. However, one area of concern
for many participants was security and privacy. For
example, participants wanted to know if “Once connected,
can a person take my other stuff?” or if there was a “lockout for security and privacy.” Currently, there is not.
The Proxemics of Co-located Collaboration
Our usability testing led us to two primary “lessons
learned” in relation to proxemics:
Do Not Require Contact. We began testing sessions by
instructing users to “put your tablets together.” Although
many users followed these directions, some users seemed
hesitant to place their tablet in direct contact with that of
the other user. In 3 of the 7 sessions, participants placed
their tablets together, but asked “Do they have to be right
next to one another?” When the experimenter replied that
they did not, subjects moved them approximately 15 to 40
10/6/2003 3:09 PM
cm apart. Clearly, stitching must support gestures between
tablets that are not immediately adjacent to one another.
Fortunately, we had anticipated this in our design, so
stitching worked well for these participants.
However, this does not mean that intimate spaces, with the
devices close to or in contact with one another, are not
useful. It depends on the users and what they are trying to
accomplish. When asked at the end of the study if
“Combing the screens of multiple Tablet PCs was a
compelling capability,” the average response was 6.8
(agree) out of 7. Users commented that they liked “the
ability to split the view, so there are no two faces trying to
peek at only one screen” and that the “wide screen would
be nice for collaboration, like for two people working on
the same spreadsheet.” Thus, although participants worked
with a stranger during the study, they seemed to envision
other contexts where close, joint work would be valuable.
Establish and relax. Users want to establish a connection
via stitching, but then relax the increasing social tension by
quickly exiting the personal space of the other user. In our
study, when one user reached over with the stylus, the other
user would often lean back to make the intrusion into
personal space less acute. Many subjects made short
stitching gestures that only extended 3-5 cm onto the other
user’s screen, and some users held the pen near the top, like
a pointing stick, rather than holding it at its tip, like a
writing instrument. Users may have adopted these
behaviors in an effort to minimize intrusions into the other
user’s personal space. Similarly, the transporter, which
allows users to share files without repeatedly reaching into
the other user’s personal space, was popular with test users.
Although participants successfully used remote menus to
choose how to combine devices, this perspective does offer
an argument against the use of remote menus, which
require the user to perform command selections while
reaching onto another user’s display. To avoid this, one
approach we are experimenting with allows the user to
stitch, return to his local screen, and then select the multidevice command to execute. Further usability testing will
be required to see if users prefer this approach.
Security and Privacy
Security was one area of concern for some test users. Since
only nearby persons can connect to a device, stitching does
offer some inherent security measures. Social rules are at
play, and because of the physical nature of the gesture,
users who violate these rules by reaching onto a user’s
screen without permission are likely to be noticed. Test
users often verbalized their intent to stitch to another user’s
screen; for example, one user commented “here’s a care
package for you” before moving files to the other user’s
screen. Nonetheless, users in an untrustworthy environment
may wish to “lock out” stitching gestures, accept stitches
only from devices which have previously been granted
permission, or require a password.
Alternative Hardware for Stitching
One of the strengths of stitching is that it leverages widely
available pen-operated mobile devices, but nonetheless
future hardware enhancements may offer ways to improve
our current implementation of stitching.
Unique ID. Stitching works well without a unique ID on the
pen, but if pen ID’s become widely available, the ID could
be used to boost the certainty that two separately observed
pen traces represent a single pen gesture performed by one
user. Whether or not a pen ID is available, recognizing the
requirements for a versatile interaction paradigm for
combining multiple mobile devices, and providing these via
the aspects of stitching outlined in this paper, are the key
contributions of our work.
Tracking beyond the screen boundary. We found that it is
difficult for users to start a stitch from the extreme edges of
the screen. If the tablet could continue to sense the pen
location 1-2 cm beyond the edge of the screen, it might be
possible to eliminate this problem.
Standardized Pens. The pen of one mobile device may not
necessarily work on that of another manufacturer. If pens
become standardized, they could work on any device.
Alternatively, if all pen computers included touchscreens,
users could use their fingers to make stitching gestures.
Multiple Pens. In our system, users cannot perform a
stitching gesture to a tablet while the other user is already
using a pen on that tablet, because current Tablet PC’s can
sense only one pen at a time.
Multi-Device Stitching
We have recently extended our stitching system
architecture to support formation of sets of up to 16
devices, but StitchMaster currently only supports formation
of pairs of tablets. The stitching server adds a device to a
connected set if the user stitches between a connected
device and a new, disconnected device. We also plan to
experiment with long stitches that traverse a series of
devices, connecting them all in one continuous gesture.
Stitching with Other Types of Devices
PDA’s. Currently available PDA’s cannot sense the
tracking state. We have not yet ported stitching to PDA’s,
but since stitching can use the dragging state, we expect it
is feasible to support stitching on PDA’s. Including PDA’s
in our system may allow interesting new applications. For
example, we have considered two designs that use PDA’s
to alter the proxemics of stitching (Fig. 7). Porches use
PDA’s as public areas for receiving visitors; each tablet has
its own porch. To give a file to someone else, a user moves
it onto the other user’s “porch” via stitching, or to offer a
file for taking the user leaves it on his own porch. The other
user can then take the file from a public porch into the more
closely held main screen area. This reduces the need for
each user to violate the personal space of the other user.
The candy dish places a single PDA in the no-man’s-land
between two other devices. Each user may then place files
into the PDA via stitching, or take files that have been left
10/6/2003 3:09 PM
there by the other user. Again, the users would not have to
reach into each other’s personal space to share files.
Tablet 1
Tablet 2
The Candy Dish
Tablet 1
Tablet 2
Fig. 7 Changing the proxemics of file sharing by using
tablet computers and PDA’s together.
Large Displays. It may be possible to support stitching
from small devices onto a large-format pen-operated
display. Because of the size disparity, the small device may
occlude part of the large display, and stitching gestures may
leave the edge of a small device but enter the large display
almost anywhere. Since our current recognition algorithm
looks for stitches that cross the edges of the screens, we
would have to adapt our recognition policies. To avoid
false-positives, it might become necessary to use a pen with
a unique ID capability or to consider further features of the
pen motion, including:
• The direction of travel or the curvature of the arc that
the pen makes as it exits one screen and enters another.
• The velocity of the pen.
• The pen tilt angles (azimuth and elevation).
Alternatively, we could avoid recognition by requiring the
user to explicitly signal stitching gestures. For example, the
user could select a menu command such as Stitch to
Another Device… before starting a stitch, or the user could
hold down the barrel button on the pen while stitching.
We believe that the true untapped potential of the emerging
wireless network lies in dynamic peer-to-peer coordination
between proximal devices. Stitching provides an example
of this perspective by offering users a versatile means to
dynamically bind together pen-operated devices. We have
provided some examples of multi-device commands, such
as copying photos between devices, expanding an image
across displays, creating a shared workspace, or using the
gallery to project selected photos on another user’s display.
It is our hope that by identifying interaction requirements
and usability issues for this new class of distributed pen
interfaces, our work may spark further exploration of new
applications, capabilities, and interaction techniques that
foster communication, sharing, and collaboration between
users of mobile devices, and empower users with new ways
to combine the capabilities of multiple mobile devices.
1. Bahl, P., Padmanabhan, V., RADAR: An In-Building RFBased User Location and Tracking System, IEEE Computer
and Communications Societies (INFOCOM 2000), 775-784.
Bartlett, J.F., Rock 'n' Scroll Is Here to Stay. IEEE Computer
Graphics and Applications, 2000(May/June 2000): p. 40-45.
Buxton, W., Chunking and Phrasing and the Design of
Human-Computer Dialogues, IFIP Information Processing
`86: Amsterdam: North Holland Publishers, 475-480.
Buxton, W., A three-state model of graphical input, Proc.
INTERACT'90, Amsterdam: Elsevier Science, 449-456.
Buxton, W., Fiume, E., Hill, R., Lee, A., Woo, C.,
Continuous hand-gesture driven input, Proceedings of
Graphics Interface '83, 1983, 191-195.
Deasy, C.M., Lasswell, T.E., Designing Places for People: A
Handbook on Human Behavior for Architects, Designers,
and Facility Managers. 1985, New York: Watson-Guptill.
Efran, M., Cheyne, J., Affective concomitants of the invasion
of shared space: Behavioural, physiological, and verbal
indicators. J. Personality & Social Psych., 1974. 29: 219-226.
Felipe, N., Sommer, R., Invasions of personal space. J.
Personality and Social Psychology, 1966. 11: p. 93-97.
Gorbet, M., Orth, M., Ishii, H., Triangles: Tangible Interface
for Manipulation and Exploration of Digital Information
Topography, CHI'98, 49-56.
Guimbretiere, F., Winograd, T., FlowMenu: Combining
Command, Text, and Data Entry, UIST 2000, 213-216.
Hall, E.T., The Hidden Dimension. 1966, New York:
Hinckley, K., Distributed and Local Sensing Techniques for
Face-to-Face Collaboration, ICMI-PUI'03, to appear.
Hinckley, K., Synchronous Gestures for Multiple Users and
Computers, UIST'03, to appear.
Holmquist, L., Mattern, F., Schiele, B., Alahuhta, P., Beigl,
M., Gellersen, H., Smart-Its Friends: A Technique for Users
to Easily Establish Connections between Smart Artefacts,
Ubicomp 2001: Springer-Verlag, 116-122.
Johanson, B., Hutchins, G., Winograd, T., Stone, M.,
PointRight: experience with flexible input redirection in
interactive workspaces, UIST 2002, 227-234.
Kurtenbach, G., Buxton, W., Issues in Combining Marking
and Direct Manipulation Techniques, UIST'91, 137-144.
Myers, B.A., Using Hand-Held Devices and PCs Together.
Communications of the ACM, 2001. 44(11): p. 34 - 41.
Rekimoto, J., Pick-and-Drop: A Direct Manipulation Technique for Multiple Computer Environments, UIST'97, 31-39.
Rekimoto, J., Saitoh, M., Augmented Surfaces: A Spatially
Continuous Work Space for Hybrid Computing
Environments, CHI'99.
Rodden, T., Rogers, Y., Halloran, J., Taylor, I., Designing
novel interactional workspaces to support face-to-face
consultations, CHI 2003.
Streitz, N.A., Geißler, J., Holmer, T., Konomi, S., MüllerTomfelde, C., Reischl, W., Rexroth, P., P. Seitz, R.,
Steinmetz, i-LAND: An interactive Landscape for Creativity
and Innovation, CHI'99, 120-127.
P. Tandler, T. Prante, C. Müller-Tomfelde, N. Streitz, R.
Steinmetz, Connectables: dynamic coupling of displays for
the flexible creation of shared workspaces, UIST'01, 11-20.
Weiser, M., The Computer for the 21st Century. Scientific
American, 1991(September): p. 94-104.