!

"#$%&'$-The EU funded RTD project ~HAPTEX addresses

the challenge of developing a Virtual Reality (VR) system for the
realistic and accurate rendering of the physical interactions of
humans with textiles, through the real-time generation of
artificial visual and haptic stimuli. This challenge concerns the
development of both the SW and the HW components of the VR
system, as well as it implies a substantial advancement in the
understanding of the mechanisms underlying the human haptic
perception of fine physical properties like those of textiles.
This paper reports some important details relating to the
technical implementation of the developed HW and SW
components with special emphasis on the issues related to their
integration into a single VR system. Furthermore some
preliminary results relating to the functional tests carried out on
the integrated system are also reported.


()*+, .+%/#-Force and Tactile Rendering, Haptic Interface,
Tactile Actuators, Virtual Reality System.

I. INTRODUCTION
EXTILES are deIormable objects characterized by very Iine
surIace and bulk physical properties, indicated with terms
such as stiIIness, smoothness, soItness, Iullness, crispness,
thickness, weight, etc. Taken as a whole they constitute the so
called Fabric Hand (deIined in |1|) oI a speciIic Iabric, which
is the basis Ior assessing its quality in relation to a given use
(Ior example Ior realizing a man`s winter suit). These
properties can be well distinguished and quantitatively
evaluated by the human haptic sensorial system (related to
sense oI touch), with an important contribution given by the
sense oI sight. There is experimental evidence that the highly
sophisticated mechanoreceptors located in the human skin
have a predominant role in the evaluation, even iI the signals
generated by these sensors are combined in the brain with
those generated by the kinesthetic sensors located in the
physiological articulations and in the muscles. For example,
when gently stroking the Iingertip on a Iabric to evaluate its
smoothness, the kinesthetic sensors give to the brain

* PERCRO Laboratory, Scuola Superiore Sant`Anna, Pisa, Italy Iontana,
marcheschi, tarri, Iabio, bergamasco}sssup.it
# WelIenlab, Institute oI Man-Machine Communication, Leibniz Universitt
Hannover, boettcher, allerkamp, Iew}welIenlab.de
Biomedical Physics Group, School oI Physics, University oI Exeter
alan.c.brady, j.qu, i.r.summers}exeter.ac.uk
inIormation about the Iingertip speed and the global Iorce
exerted on the Iabric while the mechanoreceptors sense the
small local Iluctuation oI the tangential Iorce due to Iriction.
The EU Iunded RTD project 'HAPTEX addresses the
challenge oI developing a VR system intended Ior the visual-
haptic realistic and accurate rendering oI the complex physical
interactions arising during the manipulation oI textiles. Due to
the limitations oI the present technology, since the beginning
it has been decided to Iocus the system simulation capability
on the interactions that can be attained using only two
Iingertips: the ones oI the index and the thumb (see Figure 1).

Figure 1 - Reference scenario of the HAPTEX project.
Taking into account the above considerations, the reIerence
conIiguration Ior the development oI the device responsible
Ior generating the artiIicial mechanical stimuli to be delivered
on the Iingertips(named the Whole Haptic Interface, WHI),
has been conceived as the combination oI two independent
Iorce-controlled manipulators (Force Feedback Device, FFD),
and two arrays oI independently actuated pins (Tactile
Actuator, TA). Each FFD is able to track the movements oI
the Iingertip and to convey the global Iorce oI arbitrary
direction on it, and each TA mounted on the end-eIIector oI
the corresponding manipulator is able to deliver to the surIace
oI the Iingertip skin speciIied spatial and temporal patterns,
(see Figure 2).
Integrating Force and Tactile Rendering
Into a Single VR System
Marco Fontana *, Simone Marcheschi*, Federico Tarri *, Fabio Salsedo *, Massimo Bergamasco *,
Dennis Allerkamp #, Guido Bttcher #, Franz-Erich Wolter #,
Alan C. Brady , Jianguo Qu , Ian R. Summers
T
2007 International Conference on Cyberworlds
0-7695-3005-2/07 $25.00 2007 IEEE
DOI 10.1109/CW.2007.37
277


Figure 2 - Reference configuration of the Whole Haptic
Interface .
II. DESCRIPTION OF THE SYSTEM
A. Svstem Architecture
Achieving a convincing virtual textile simulation requires a
good compromise to be reached between the need oI accuracy
in the material representation and the need oI speed Ior
obtaining visually realistic simulation Irame rates compatible
with real-time perception. These Iactors have to be considered
both in the visual and the haptic Iields. However, the graphics
rendering loop has diIIerent requirements compared to the
haptic rendering loop in terms oI reIresh Irequencies. While in
graphics a reIresh rate oI 30 Ips is quite acceptable, in haptics
a response Irequency oI 300-1000 Hz is needed to ensure
accurate interaction. A dedicated structure has thereIore been
deIined Ior adapting the diIIerent Irame rates required by the
mechanical simulation and the haptic rendering computations.
Two separate computation threads were implemented: The
Iirst is a low-Irequency thread running a complex large-scale
simulation oI the whole cloth surIace achieving quantitative
accuracy oI the diIIicult nonlinear anisotropic behavior oI
cloth in real-time. This simulation has to use nonlinear strain-
stress Iunctions related to the cloth being simulated. An
eIIicient non-linear minimization method computing spline
Iunctions approximating the aIorementioned non-linear strain-
stress curves is presented in |20|. The second thread is a high-
Irequency thread Ior computing the local data necessary Ior
haptic rendering and Ior accurately sending haptic Iorces back
to the mechanical simulation.
Textile
Simulation
Thread
Textile
Simulation
Thread
Local
Simulation
Thread
Local
Simulation
Thread
Force
Feedback
Thread
Force
Feedback
Thread
running at
60Hz
running at
300Hz
running at
1kHz
bordering geometry
Local
Geometry
bordering geometry bordering geometry
Local
Geometry
force generation Force
Extrapolation
Force
Feedback
Device
Force
Feedback
Device
forces positions
Motion
Estimation
movements
Visual
Display
Visual
Display
global mesh adaption
Fingertip
Model
contact
formulation
deform refine
Tactile
Rendering
Thread
Tactile
Rendering
Thread
Tactile
Array
Tactile
Array
contact area
force distribution, velocity
amplitudes

Figure 3 - Different threads within the HAPTEX system
Although the threads allow parallelizing oI the computation,
we also need some synchronization oI the threads. ThereIore,
the synchronization works as Iollows. In the initial stage all
threads are running at their dedicated update rates. The Iorce
Ieedback thread is constantly Ietching new positions Irom the
Iorce Ieedback device. These positions are processed to
predict the user`s motion and to estimate the next position. At
the same time the (global) textile simulation thread is
computing the deIormations caused only by gravity, whereas
the local simulation thread waits Ior any new local geometries
to be simulated. At each simulation step oI the global thread
the local thread receives Iingertip dimensions, the current and
predicted position. The global thread analyses its underlying
mesh with respect to potential collisions with the Iingertip Ior
the next time step. These regions including their physical
states are sent to the local thread in order to be geometrically
reIined and inserted into the local simulation. AIterwards both
simulation threads continue to run according to their data.
With the newly added local mesh, the local thread checks iI
any collision has taken place in between a local simulation
time step. In case oI a contact the occurring deIormation oI the
local part oI the textile is computed according to the Iingertip
model being used. The Iorces at the Iingertip generated during
the contact are sent to the Iorce Ieedback thread.
The contact area estimated by the contact model is transmitted
to the tactile renderer whereby Ior each pin a contact Iorce is
computed. According to the deIined positions oI the pins on
the Iinger the local velocity is also provided.
The schematic in Figure 3 shows the separation oI the
computational tasks into the diIIerent threads.
B. Force Renderer
The Iorce-Ieedback renderer is responsible Ior the modeling oI
the interaction between the Iingertip and the Iabric. This
implies the computation oI Iorces occurring at the contact,
considering the physical properties oI the objects involved. In
the haptics literature there exist several approaches to render
contact Iorces. In |17| an eIIective point-based rendering
algorithm was Iirstly introduced and constantly improved by
others, i.e. |19||18|.Ruspini et al. |15||16|extended the
algorithm to support contacts oI arbitrary shapes. In contrast to
the well known proxy method a recent approach (see |14|)
suggests to compute the contact Iorces by solving the
Signorini contact problem employing Iinite elements.
Although the latter method models the deIormation oI the
Iingertip at the contact appropriately, it is demanding oI
precious computation time because it requires iteratively
solving non-linear equation systems. A good compromise
between accuracy and computational eIIort is oIIered by a
penetration- or penalty-Iorce based method. The penetration-
or penalty-Iorce-model computes the Iorce as a result oI the
contact proportional to the penetration depth or intersecting
volume. The depth is given as the length oI the vector deIining
the shortest translation oI the colliding bodies to a touch
situation. For computational reasons we use the penetration
depth Ior Iorce calculations.
The mechanical equivalent to the aIorementioned situation can
be described by a spring attached to both bodies enIorcing a
repulsion in case oI a collision. The Iorces being applied on
both bodies are computed by the length oI the depth vector
and diIIer only in the direction. II no additional external Iorces
are applied, then aIter several simulation steps the bodies will
reach a Iorce equilibrium as depicted in Figure 4.
278

d
d
finge rt ip
s phe re
x
d x
v
n
x
v
n+ 1
F
C
n
F
C
n+ 1
node s of t e xt ile
d
d
node s of t e xt ile
finge rt ip
s phe re
x
d x
v
x
v
n
n+ 1
F
C
F
C
n
n+ 1
d
d
finge rt ip
s phe re
x
d x
v
n
x
v
n+ 1
F
C
n
F
C
n+ 1
node s of t e xt ile
d
d
node s of t e xt ile
finge rt ip
s phe re
x
d x
v
x
v
n
n+ 1
F
C
F
C
n
n+ 1

Figure 4 - Initial state (left) and final state (right) of the
penetration algorithm
At this stage the bodies are still intersecting. In this strategy
the Iingertip is modeled as a rigid sphere being the Iirst
contact body. Due to its shape, a Iast collision test with the
textile can be made. The Iingertip collides with a textile
particle n with position
v
n
x iI
d d v
n
d x x < holds.
Both Iingers are in collision under the Iollowing condition:
d d d
d x x 2
1 0
< .
To distinguish between static and kinetic Iriction we separate
the Iorce resultant
r
v
F into tangential
T
v
f and normal
N
v
f
components applied to a textile particle v. The Iorce
r
v
F is the
sum oI the internal Iorce
J
v
F , penetration Iorce
C
v
F and
Iriction Iorce.
II both Iorces satisIy the stick condition equation Ior static
Iriction as seen below.
N
v S
t
v
f f <
Then we have
v
N
v
r
v
T
v
r
v v
N
v
n f F f
F n f
=
=

In this state the particles are not moving relatively to the
Iingertip. The position is set according to the Iingertip
movement. Forces oI the remaining Iree particles oI the textile
have to be recomputed. Otherwise, iI the stick condition is not
met, the kinetic Iriction is added to the particle.
The Iorce sent to the Iorce Ieedback device is determined by
the reaction Iorce oI the contacting particles and the second
Iingertip
D
F . This leads to the Iollowing equation
rel
rel
v
N
v
v v rel
rel N
v S
C
v
D d
v
v
F
v
v
f F F F
K
K
S
s

=
where v indexes the particles being in contact with the
Iingertip and v
S
and v
K
deIining the particles respective to
their Iriction state.

C. Tactile Renderer
The tactile renderer is based on vibrations that play an
important role in the tactile exploration oI Iine surIaces. To
produce appropriate excitation patterns we use an array oI
vibrating contactor pins as described in Section II-E.
The tactile renderer generates 24 drive signals Ior the
individual contactors oI a single tactile stimulator, on the basis
oI the Iollowing inputs:
small scale description oI the textile surIace: a single
'tile oI the textile weave pattern, speciIied as a
pseudo-topology at 0.1 mm resolution over the 2D
surIace oI the tile (derived Irom Kawabata
measurements oI roughness and/or Iriction);
large-scale description oI the textile surIace: a
representation oI the non-uniIormity oI the textile
surIace, speciIied as a pseudo-amplitude at 1 mm
resolution over the 2D surIace oI the 200 mm 200
mm sample oI virtual textile (derived Irom optical
imaging);
position and orientation oI the Iinger pad on the
virtual textile;
speed and direction oI the movement oI the Iinger
pad over the virtual textile.

Figure 5 - The tactile renderer. Input and ouput data are
specified in 25 ms timesteps.
Figure 5 shows an overview oI the process using the input data
are used to speciIy the output drive signals. Taking into
account oI the direction oI the movement, a spatial-Irequency
spectrum is calculated Irom the pseudo topology oI the small-
scale description oI the virtual surIace. InIormation on the
speed oI movement oI the Iinger pad is used to convert spatial-
Irequency components into temporal-Irequency components.
The resulting temporal- Irequency spectrum is reduced to only
two amplitudes, A
40
and A
320
, by applying appropriate
bandpass Iilter Iunctions (see |21|), corresponding to the 40-
Hz and 320-Hz channels. Amplitudes Ior the 40- Hz
279

component in the drive signals Ior each oI the 24 channels are
obtained Irom A
40
by weighting according to data Irom the
large-scale description oI the virtual surIace, Ior the 24
locations on the Iinger where the contactors oI the tactile
stimulator are positioned. Similarly, amplitudes Ior the 320-Hz
component in the drive signals Ior each oI the 24 channels are
obtained Irom A
320
by weighting according to data Irom the
large-scale description oI the virtual surIace.

D. Force Feedback Device
In the scope oI the HAPTEX project the Force Feedback
Device (FFD) is responsible Ior tracking the global
movements oI the user`s index and thumb Iingertips,
delivering controlled Iorces on them as evaluated by the Force
Renderer Module and holding the Tactile Actuators. In order
to allow early integration oI the diIIerent components oI the
HAPTEX system, it has been decided to develop two diIIerent
types oI FFD, the Iirst one derived Irom an existing device
(named 'FFD in conIiguration A) and the second one
designed Irom scratch (named 'FFD in conIiguration B).
Although the two conIigurations use very diIIerent basic
solutions and hardware implementations, they have quite
similar Iunctionalities, even iI the related perIormances are
substantially diIIerent. In both cases, one oI the main
requirements in their development has been the achievement
oI a highly accurate Iorce Ieedback. This requirement derives
directly Irom the stated goal oI allowing the discrimination oI
the Iine mechanical properties oI the textiles. From the
technical point oI view this implies the accurate generation
and control oI Iorces that can be oI the order oI Iew grams
(0.01 N), as they arise during the natural manipulation oI
textiles.
To address this challenge, a new explicit Iorce control has
been developed making use oI purpose-designed highly
sensitive multi-component Iorce sensors, placed directly in
contact with the user`s Iingertips.
The Iollowing paragraphs describe in detail the two
conIigurations and report some experimental data relating to
their perIormances.
1) FFD in Configuration A
The FFD in conIiguration A has been derived Irom the GRAB
system (a detailed description is reported in |2|), developed by
the PERCRO Laboratory in the Iramework oI the
homonymous European RTD Project. The system is composed
oI two identical robotic arms, each having 6 DOF oI which the
Iirst 3 translational DOF are instrumented and actuated, while
the remaining 3 orientational DoF are only passive. Each
robotic arm is Iunctionally equivalent to the well known
Phantom |3| haptic interIace but has larger workspace.


Figure 6 - Picture of the sensor-equipped gimbal.
With respect to the original GRAB system, the FFD in
conIiguration A uses new gimbals (see Figure 6) mounted on
the robotic arm, each equipped with 3 rotational position
sensors (encoders) Ior the measurement oI the Iingertip
orientation and with a 3 component highly sensitive Iorce
sensor speciIically conceived Ior the application, and an
explicit Iorce control making use oI the said Iorce sensor.
Explicit Iorce control algorithm consists oI an inner velocity
loop-outer Iorce loop scheme (see Figure 7). The mechanical
model oI the device and the criteria Ior the dimensioning oI
the controller are described in detail in |5| and in. |6| Due to
the presence oI elasticity in the transmission oI the Iorce Irom
the actuator to the moving mass, the bandwidth oI the Iorce
controller has been limited to 3 Hz in order to ensure the
stability oI the system.
d F
f K Kv V1 F (s)
(s) F
V2
m F
s
Xm
Velocity Loop
(s) F
f
Fu
Force Loop Plant Force Loop Plant

Figure 7 - Scheme of the force control
Furthermore, investigations on the attachments oI the device
with the Iingertips have been carried out. Indeed, during the
natural handling with the textiles the interaction Iorces can be
oI the order oI Iew grams. This implies that the attachment oI
the Iingertip with the device has to be so to not hinder the high
sensitivity oI the human mechanoreceptors located in the skin.
From this point oI view the thimble-like attachment,
commonly used by the existing haptic interIace, is not well
suited because it produces a pre-stress oI the skin reducing the
sensitivity oI the mechanoreceptors. In order to address this
issue, diIIerent alternative solutions Ior the Iingertip
attachment have been investigated (a bare plate, a plate with
elastic strip, a plate with eccentric thimble). As expected the
experimental test demonstrated that the human sensitivity is at
best when the bare plate is used, even iI this solution is the
worst Ior the transmission oI the torques required Ior the
orientation oI the gimbal.
280

Finally a Iield test investigating the system capability to track
the Iinger motion with minimal resistance Iorce has been
carried out. The subjects were asked to move their Iingers at
constant velocity and the system was set to display no Iorces.
The maximum module oI the acquired resistant Iorce was
about one tenth oI a Newton (10 grams Iorce), as it can be
observed in Figure 8.
2) FFD in configuration B. development of the hand
exoskeleton
The FFD in conIiguration B consists oI a Hand Exoskeleton
(HE) expressly conceived Ior the accurate generation oI light
Iorces.
0 2 4 6 8 10 12 14 16
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
Time (s)
F
o
r
c
e

(
N
)

Figure 8 - Plot of the resistant force versus time during
finger tracking at constant velocity
Several works (Ior example |4|, |7| and |8| can be Iound in the
literature addressing the development oI HEs.
According to their type oI Iunctionality, the existing HE can
be grouped in two diIIerent categories,:
- Multi-phalanx HEs: they can generate Iorces on each
phalanx oI the Iinger along a Iixed direction with respect to
the phalanx (e.g. normal to the phalanx);
- Fingertip HEs: they can generate Iorces only on the (the
Iingertips along arbitrary direction in 3 D space.
Considering the application addressed by the HAPTEX
project, the second type oI Iunctionality has been selected.
A scheme oI quasi-anthropomorphic kinematics has been
selected Ior the implementation oI the Iinger exoskeleton. This
solution allows exploiting the beneIits oI anthropomorphic
kinematics, like the maximum ratio oI the available over the
needed workspaces and minimum encumbrance oI the
linkages, while avoiding at the same time the singularity that
would occurs when the Iinger is completely extended.
In Figure 9 a CAD model oI the HE is shown. It can be
noticed that the encumbrance oI the device has been mainly
located in the dorsal side with the aim oI allowing the
complete closing oI the hand. This has been achieved through
the use oI Remote Centre oI Rotation Mechanisms that
implement rotational joints having the axes located outside the
linkages.
The whole mechanism has 4 Degrees OI Freedom (DoF), even
iI it is actuated with only three motors, thanks to the coupling
oI the last DoF (end-eIIector Joint) with the previous one. The
coupling is acceptable because also in the human hand the last
phalanx can be rarely moved independently Irom the middle
phalanx during natural movements.
The HE is equipped with three electrical motors with low
speed reduction ratios (1:14). The actuators are placed at the
base oI the Iinger exoskeleton in proximity oI the dorsal side
oI the palm. The joints are actuated through steel cables
working in tension.
For the position sensing common incremental encoders
located on the axis oI the motors have been used while Ior the
Iorce sensing a compact highly sensitive 3 component Iorce
sensor, placed directly in contact with the user`s Iingertip, has
been expressly conceived Ior the application.
The purposely developed electronics Ior the sensor acquisition
and the driving oI the motors have been located inside the
motor box. The communication with the control PC takes
place through standard RS-232 serial cable.
The device is capable oI exerting a continuous Iorce on the
Iingertips oI 5N with a resolution oI 0.005 N.
At present, several parts oI the device have been realized and
other ones are being manuIactured.
A Iurther development will be the integration oI the dorsal
tactile array described in Section III on the dorsal side oI the
last phalanx oI the exoskeleton.

Figure 9 - CAD model of the Hand Exoskeleton.
E. Tactile Actuator
Two designs oI stimulator array have been developed as the
tactile component oI the HAPTEX system, hereaIter reIerred
to as conIigurations 'A and 'B. Both conIigurations use
identical actuators and have the same contactor layout, and
each has been designed to be used with the FFD.
The conIiguration 'A actuator has been designed to display a
2D tactile pattern on a single Iingertip. In contrast, the type
'B actuator (pictured in Iigure 10) is suitable Ior use in a pair,
displaying tactile inIormation to both the Iinger and thumb in
a grasp conIiguration.
In an ideal device, the Iingertip would be completely covered
with contactors. The spacing oI these would be around 1 per
square millimeter, matching the spatial acuity oI the
mechanoreceptors in the skin |9|. In reality, the number oI
actuators is limited by the space available and the complexity
oI the drive electronics and the wiring. A pin spacing oI 2 mm
has been used, as it is believed that Ior many tactile tasks this
is indistinguishable Irom a 1 mm spacing |10|. 24 pins are
arranged in 64 grid, with Iixed pins in interstitial positions to
localise the stimulation sites on the skin, and to transIer Iorces
Irom the FFD to the Iinger. The height oI the active pins is set
such that they indent slightly into the skin when the Iinger pad
is gently pressed into the Iixed pins and the surround.
281

Both designs are driven by piezo-electric bimorphs, chosen Ior
their simplicity, ease oI assembly and low cost. The
dimensions oI the bimorphs were selected by numerical
modelling and experiment.
The Iree-length oI the bimorphs was chosen such that their
resonant Irequency (100 Hz, rising to 120Hz when loaded by
the skin) lies between the two working Irequencies oI the
renderer (40 Hz and 320 Hz), and such that the Iirst anti-
resonance (450 Hz) is not too close to the upper working
Irequency. The response is designed such that the roll-oII
above resonance counteracts the increase in sensitivity oI the
human Iingertip |11|. This balances the perceived intensity at
a given drive signal, and reduces the dynamic range required
Irom the drive electronics.
The width oI each bimorph is a compromise must be struck
between the amount oI Iorce the actuator could deliver and the
overall size oI the device. A pair oI commercially available
2.1 mm wide actuators (APC International), arranged in
parallel, was chosen. This gave a width oI 4.2 mm and a static
blocked Iorce oI 0.5 mN/volt.
The design oI the type 'A is a logical development oI earlier
systems |12|,|13|, with the piezoelectric actuators located
below the palmar surIace oI the Iingertip. It is suitable Ior use
with a single Iinger in a 2D environment, and has been used to
develop the drive electronics and soItware, and Ior testing the
tactile renderer.


Figure 10 - Early prototype of the configuration ~B,
showing the two different orientations of the piezo-beams
The design oI the type 'B actuator TA was constrained by the
need to move the actuators out Irom underneath the Iingertip
to reduce the depth oI the device below the Iingertip, and so
allow Ior a grasp conIiguration. In addition, the device had to
accommodate and integrate with the Iorce sensor Ior the FFD,
and to Iit the FFD mechanism.
In the type 'A conIiguration, the Iingertip rests on a planar
surIace. The contactor pins are arranged to be perpendicular to
this plane. The type 'B conIiguration uses a curved contact
plate which Iollows the shape oI the Iinger pad. Contactor pins
are approximately normal to the contact plate (illustrated in
Figure 11), with diIIerent orientations oI the drive mechanism
Ior diIIerent rows oI pins. As a result, some oI the drive links
have to Iollow relatively complex paths. Subjective tests oI
this conIiguration have shown that the perIormance is good.

Figure 11 - This end on sketch of the fingertip shows the
orientation of the contactor pins and the direction in which
they are driven.
III. SYSTEM INTEGRATION
A. Mechanical Integration of TA with FFD
The integration oI the FFD with the TA requires the resolution
oI non-trivial problems that have implications Ior the most
appropriate mutual mechanical arrangement oI the diIIerent
parts constituting the two devices. This arrangement should be
so to allow, on one hand, the delivery to the Iingertip oI the
mechanical stimulations that each device can generate and on
the other, to comply with the speciIic capabilities and
requirements oI the devices. In particular the TA pins can
produce dynamic indentation oI the skin, but the Iorces
(normal to the surIace) that they can deliver are relatively
small (in the range oI Iew grams).

Figure 12 - Scheme of the force transmission between
fingertip and plate
Furthermore they cannot withstand lateral Iorces acting on the
pin because these Iorces would produce their deIlection. In
short the mechanical arrangement has to be so to prevent the
loading oI the pins with relatively high Iorces. Vice versa the
FFD can deliver to the skin relatively high Iorce in every
direction, but in order to guarantee a high accurate Iorce
Ieedback these Iorces must be sensed by the 3 component
Iorce sensor (see scheme in Figure 12).

Figure 13 - Mechanical Integration of FFD and TA

282

The FFD in conIiguration A and the TA in conIiguration B
have been successIully integrated in order to realize the
development level 4 (DL4) oI the HAPTEX system (see
Figure 13).
B. Integration of the Force and Tactile Renderer
For the computation oI the Iorce-Ieedback the haptic renderer
already requires among others the contact point, the velocity
oI the Iingertip relative to the Iabric and the normal Iorce (cI.
|20|). Instead oI a single contact point the tactile renderer
requires a contact area to decide which contactor pins have to
be activated. Also a single normal Iorce is not suIIicient Ior a
convincing tactile simulation. Rather a distribution oI the
normal Iorce over the whole contact area is needed by the
tactile renderer.
For each contactor pin, the geometry oI the tactile actuator
determines the point on the Iingertip where the stimulation
caused by that pin occurs. These points have been included in
the model oI the Iingertip. Employing the contact model the
position (in local coordinates oI the Iabric) and the normal
Iorce at each point is computed. The velocity is computed by
applying a discrete diIIerential operator.
The tactile renderer is called every 25 ms, i.e. it runs in a 40
Hz loop. However, the contact data is computed by the haptic
renderer every millisecond and the data needs to be resampled.
To avoid aliasing the data is passed through a digital
antialiasing Iilter that suppresses Irequency components above
20 Hz.
In the stand-alone version oI the tactile renderer the Iabric was
assumed to be planar. Furthermore the Iinger was assumed to
always being in contact with the plane oI the Iabric (see |9|).
However, in the integrated case the Iabric may have Iolds and
wrinkles, resulting in a more complex contact geometry. The
Iingertip deIorms due to the contact pressure, also inIluence
the shape oI the contact area. As a consequence not all pins oI
the tactile actuator should be activated, only those inside the
contact area. The shape oI each Iinger`s contact area is
provided by the Iorce-Ieedback renderer. Note that a contact
area may not exist, i.e. in that case the Iinger does not touch
the Iabric.
While moving a Iingertip over a rough surIace we experience
a tactile sensation. Its intensity also depends on the Iorce with
which the Iinger is pressed against the Iabric. For relatively
small Iorces a linear dependency between the Iorce and the
sensation seems to be reasonable. ThereIore the amplitudes
computed by the tactile renderer are multiplied by k
F
F
beIore they are transmitted to the tactile actuator. F denotes
the Iorce between the Iinger and the Iabric and k
F
is a constant
relating the Iorce to the intensity. Previous to the integration
there was no possibility to assess the contact pressure between
the Iingertip and the Iabric. ThereIore a constant Iorce normal
to the Iabric was assumed implicitly by the tactile renderer.

IV. PRELIMINARY TESTS
Extensive tests have been perIormed to assess the mechanical
and electrical disturbance induced by the vibration and the
electrical noise generated by the TA on the measure oI the
interaction Iorce, because oI its potential negative impact on
the accuracy oI the Iorce Ieedback. The tests have been
perIormed activating only one speciIic pin oI the TA array at a
time and acquiring the resulting measured Iorce signal. The
Irequency spectrum oI the signal has been then evaluated
using the Discrete Fourier TransIormation (see Figure 14).
The disturbances produced by 4 diIIerent pins, having
diIIerent locations in the TA array and mechanical coupling
conditions (Iree-moving or in contact with the Iorce plate
hole) have been investigated. The tests evidenced that the
disturbance produced by the vibration is predominant with
respect to the electrical noise and that Iree-moving pins
produce about 10 times less disturbance than produced by pins
in contact. The magnitude oI the disturbance on the excitation
Irequency (40Hz and 320 Hz) is about 200 times greater Ior
the Iree-moving pin with respect to the base noise than when
the pin is not activated. Furthermore a system test has been
also perIormed in order to assess the global consequence oI
the induced noise on the accuracy oI the Iorce Ieedback. No
meaningIul eIIects have been detected due to the low
bandwidth oI the Iorce Ieedback (about 3 Hz).

Figure 14 - Typical Discrete Fourier Transform of the
disturbance produced by one TA pin vibrating at 40Hz on
the force measure.
V. CONCLUSIONS AND FUTURE WORK
In this paper we have reported the preliminary work
perIormed towards the realization oI a system able to simulate
the visual and haptic interaction with Iabrics.
The manipulation oI Iabrics is a very complex perceptual
experience that is associated with a lot oI stimuli that are very
hard to simulate and to reproduce virtually.
The system that has been realized integrates two kinds oI
stimulation devices: a Tactile Array (TA) Ior the tactile
stimulation oI the Iingertip and a Force Feedback Device
(FFD) Ior delivering the interaction Iorces. In this work we
Iocused on the issue oI the integration oI these two devices
Irom the soItware and the hardware point oI views.
From the soItware point oI view the main issue is to generate
simultaneously the signals to be provided to the FFD and the
TA complying with real time speciIications. To achieve this
two separate soItware components called the Force Feedback
Renderer and the Tactile Renderer, were realized and
integrated.
283

For what concerns the hardware, the real challenges consist oI
realizing a device able to convey both the delicate and Iine
interaction Iorces that are involved with the manipulation oI
textiles and deliver the associated tactile stimulus. It has been
shown that the integration oI these two devices can raise
several issues like the electrical and mechanical cross-talk
between the two Ieedback devices and the issue oI how to
deliver the Iorce to the user`s Iingertip whilst not transmitting
it to the pins oI the array. This has lead to the design oI a
purposely developed haptic-tactile integrated device.
The preliminary tests on the integrated system have shown
that the global perIormances oI the device are acceptable and
the major issues associated with the integration have been
successIully overcome.
In the near Iuture we will set up the complete system Ior two
Iinger interaction and perIorm several tests.
The next step will be the realization oI a portable wearable
FFD and to integrate it with the tactile array.
ACKNOWLEDGMENT
The project HAPtic sensing oI virtual TEXtiles (HAPTEX)
is a research project Iunded under the Sixth Framework
Programme (FP6) oI the European Union (Contract No. IST-
6549). The Iunding is provided by the Future and Emerging
Technologies (FET) Programme, which is part oI the
InIormation Society Technologies (IST) programme and
Iocuses on novel and emerging scientiIic ideas. Its mission is
to promote research that is oI a long-term nature or involves
particularly high risks, compensated by the potential oI a
signiIicant societal or industrial impact.
REFERENCES
|1| H. Behery, EIIect oI Mechanical and Physical Properties on Fabric
Hand, Publisher: Taylor & Francis Ltd, 2005.
|2| Avizzano C. A., Marcheschi S., Angerilli M., Fontana M., Bergamasco
M.: 'A Multi-Finger Haptic InterIace Ior Visually Impaired People. In
Proceedings oI the 2003 IEEE International Workshop on Robot and
Human Interactive Communication, Vol. 31, (2003), 165- 170.
|3| Thomas H. Massie and J. K. Salisbury., 'The PHANTOM Haptic
InterIace: A Device Ior Probing Virtual Objects Proceedings oI the
ASME Winter Annual Meeting, Symposium on Haptic InterIaces Ior
Virtual Environment and Teleoperator Systems, Chicago, IL, Nov. 1994.
|4| Bouzit M., Popescyu G., Burdea G., Bioan R.: 'The Rutger Master II-nd
Force Feedback Glove, In Proc. IEEE Vr. Haptic Sysmposium, Vol.7
(March 2002) 256-263.
|5| Eppinger S., Seering, W.: 'Understanding bandwidth limitations in robot
Iorce control. In Proc.Robotics and Automation ConIerence, 1987
IEEE International ConIerence on , Vol.4 (1987), 904- 909.
|6| Volpe R., Khosla P., 'A theoretical and experimental investigation oI
explicit Iorce control strategies Ior manipulators. In IEEE Transactions
on Automatic Control, vol.38, 11 (1993), 1634-1650.
|7| Wege, A.; Kondak, K.; Hommel, G., Mechanical design and motion
control oI a hand exoskeleton Ior rehabilitation Mechatronics and
Automation, 2005 IEEE International ConIerence, Vol.1, Iss., 29 July-1
Aug. 2005, Pages: 155- 159 Vol. 1
|8| Virtual Technologies Inc., Cyber Grasp User Guide, 1999..
|9| K.O. Johnson, T. Yoshioke, F. Vega-Bermudez, 'Tactile Iunctions oI
mechanoreceptive aIIerents innervating the hand, J. Clin.
Neurophysiol., vol. 17, 2000, pp. 539558.

|10| I.R. Summers, C.M. Chanter, A.L. Southall and A.C. Brady, 'Results
Irom a Tactile Array on the Fingertip in Proc. Eurohaptics 01,
Birmingham, 2001, pp. 2628
|11| R.T. Verillo, A .J. Fraioli, R. L. Smith, 'Sensation magnitude oI
vibrotactile stimuli, Perception & Psychophysics, vol. 6, pp. 366372,
1969.
|12| I.R. Summers, C.M. Chanter, 'A broadband tactile array on the
Iingertip, J. Acoust. Soc. Am., vol. 112, pp. 21182126, 2002.
|13| I.R. Summers, A.C. Brady, M. Syed, C.M. Chanter, 'Design oI Array
Stimulators Ior Synthetic Tactile Sensations, in Proc. World
Haptics`05, Pisa, 2005, pp. 586587.
|14| Duriez C. et al.: Realistic Haptic Rendering oI Interacting DeIormable
Objects in Virtual Environments, IEEE Transactions on Visualization
and Computer Graphics (2006), pp. 36-47
|15| Ruspini D.C. et al.: Haptic interaction in virtual environments, Proc.
IEEE/RSJ int. conI. on Intel. Robots & Systems (1997), Vol.3
|16| Ruspini D.C. et al. : The haptic display oI complex grapical
environments, Proc. 24th annual. ConI. on Computer Graphics and
Interactive Techniques (1997)
|17| Zilles C.B., Salisbury J.K.: A constraint-based god-object method Ior
haptic display, Proc. IEE/RSJ Int. conI. on Intel. Robots & Systems
(1995), Vol.3, pp. 146-151
|18| Ho C.H., Basdogan C., Srinivasan M.A.: EIIicient point-based rendering
techniques Ior haptic display oI virtual objects, Presence: Teleoperators
and Virtual Environments (1999), Vol. 8
|19| Morgenbesser, H., and Srinivasan, M. Force shading Ior haptic shape
perception. In Proceedings oI the ASME Dynamics Systems and Control
Division (1996), vol. 58, pp. 407--412.
|20| Volino P. et al.: From measured physical parameters to the haptic Ieeling
oI Iabric, Proc. OI HAPTEX`05
|21| Dennis Allerkamp, Guido Bttcher, Franz-Erich Wolter, Alan C. Brady,
Jianguo Qu, and Ian R. Summers. A vibrotactile approach to tactile
rendering. Vis. Comput., 23(2):97108, 2007


284