Upload
dennis-allerkamp
View
214
Download
1
Embed Size (px)
Citation preview
Cognitive Systems Monographs
Volume 10
Editors: Rüdiger Dillmann · Yoshihiko Nakamura · Stefan Schaal · David Vernon
Dennis Allerkamp
Tactile Perception of
Textiles in a Virtual-Reality
System
ABC
Rüdiger Dillmann, University of Karlsruhe, Faculty of Informatics, Institute of Anthropomatics,
Humanoids and Intelligence Systems Laboratories, Kaiserstr. 12, 76131 Karlsruhe, Germany
Yoshihiko Nakamura, Tokyo University Fac. Engineering, Dept. Mechano-Informatics, 7-3-1 Hongo,
Bukyo-ku Tokyo, 113-8656, Japan
Stefan Schaal, University of Southern California, Department Computer Science, Computational Learn-
ing & Motor Control Lab., Los Angeles, CA 90089-2905, USA
David Vernon, Khalifa University Department of Computer Engineering, PO Box 573, Sharjah, United
Arab Emirates
Author
Dipl.-Math. Dennis Allerkamp
Gottfried Wilhelm Leibniz Universität HannoverInstitut für Mensch-Maschine-KommunikationFachgebiet Graphische DatenverarbeitungWelfengarten 130167 HannoverGermanyE-mail: [email protected]
ISBN 978-3-642-13973-4 e-ISBN 978-3-642-13974-1
DOI 10.1007/978-3-642-13974-1
Cognitive Systems Monographs ISSN 1867-4925
Library of Congress Control Number: 2010929202
c© 2010 Springer-Verlag Berlin Heidelberg
This work is subject to copyright. All rights are reserved, whether the whole or part of the material isconcerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publicationor parts thereof is permitted only under the provisions of the German Copyright Law of September 9,1965, in its current version, and permission for use must always be obtained from Springer. Violationsare liable for prosecution under the German Copyright Law.
The use of general descriptive names, registered names, trademarks, etc. in this publication does notimply, even in the absence of a specific statement, that such names are exempt from the relevant protectivelaws and regulations and therefore free for general use.
Typeset & Cover Design: Scientific Publishing Services Pvt. Ltd., Chennai, India.
Printed on acid-free paper
5 4 3 2 1 0
springer.com
Preface
This monograph originates from my work on the HAPTEX project. In De-cember 2004 Prof. Franz-Erich Wolter, the head of the Institute of Man–Machine Communication of the Leibniz Universitat Hannover, offered methe opportunity to participate in that EU funded project. Being a mathe-matician I had only very little experience in the field of haptic simulationin those days, but Prof. Wolter trusted in my ability to become acquaintedwith new fields of research in a very short time. I am still thankful for theconfidence he has shown me.
Since then I indeed learned and found out a lot. With this monograph Itry to pass on the knowledge I gained. Having a reader in mind who—likeme at the beginning of the project—has no background in psychophysics,neurophysiology or textile engineering I will provide the necessary basics. Theskilled reader may safely skip these parts. Nevertheless I presume some basicknowledge in mathematics. I hope that this thesis might help a newcomer todiscover the fascinating field of tactile simulation.
This work would not have been possible without the funding of the project“HAPtic sensing of virtual TEXtiles” (HAPTEX) under the Sixth Frame-work Programme (FP6) of the European Union (Contract No. IST-6549).The funding is provided by the Future and Emerging Technologies (FET)Programme, which is part of the Information Society Technologies (IST)programme and focuses on novel and emerging scientific ideas. For invent-ing this project and accordingly requesting the funding I thank Prof. Na-dia Magnenat-Thalmann, Dr. Harriet Meinander, Prof. Franz-Erich Wolter,Dr. Ian Summers and P.Eng. Fabio Salsedo.
Special thanks go to my thesis advisor Prof. Franz-Erich Wolter for hisguidance and support throughout my work on my PhD thesis which hasgreatly benefited from all his feedback. I hope I inherited some of his goodsense for interesting and profound research.
My research was influenced a lot by Dr. Ian Summers who originallyworked on the subject of tactile simulation. He not only provided several
VI Preface
tactile displays for the experiments but also shared a lot of his knowledgeand experience. I really appreciate I was able to work with him.
I also thank Prof. Nadia Magnenat-Thalmann for the coordination of theHAPTEX project. She always applied very high standards to our researchwork. This was sometimes very stressful but the project would not have beennearly as successful as without her leadership. I appreciate her willingness tobe a member of my examination committee.
I really enjoyed the interdisciplinary collaboration within the HAPTEXproject. I thank all the team members for the interesting discussions and thepositive working atmosphere. I would like to give a special mention to GuidoBottcher; we worked closely together on the haptic rendering software andcomplemented each other very well. It was a pleasure to work with him.
I had the pleasure of working with some very talented students whocontributed to my research: I thank (in alphabetical order) Steffen Blume,Daniel Glockner, Tjard Kobberling, Karin Matzke and Natalya Obydenna. Ialso thank Steffen Blume, Wiebke Frey and Iris Lieske for proofreading themanuscript and for the helpful discussions. The experiments conducted formy research were only possible because numerous persons voluntarily partic-ipated as probands. I thank all of them for their support. Finally, I reallyappreciate the moral support of my colleagues, friends and especially of myfamily. Thank you!
Hannover, March 2010 Dennis Allerkamp
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Human Perception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.1 Introduction to Psychophysics . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Just Noticeable Differences and Weber’s Law . . . . . . . 62.1.2 Weber-Fechner Law and Stevens’ Power Law . . . . . . . . 72.1.3 Measuring Psychometric Distances . . . . . . . . . . . . . . . . 82.1.4 Statistical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.1.5 Signal Detection Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 102.1.6 Multidimensional Scaling . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 The Human Nervous System . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.2.1 Neurons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.2.2 Neural Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172.2.3 Sensory Reception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Human Tactile Perception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.3.1 Mechanoreceptive Afferents . . . . . . . . . . . . . . . . . . . . . . . 182.3.2 Perception of Form and Texture . . . . . . . . . . . . . . . . . . . 192.3.3 Perception of Fine Textures . . . . . . . . . . . . . . . . . . . . . . . 202.3.4 Perceptional Model of Tactile Simulation . . . . . . . . . . . 20
2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3 Devices for Tactile Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 253.1 Survey of Existing Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1.1 Electromagnetic Displays . . . . . . . . . . . . . . . . . . . . . . . . . 263.1.2 Pneumatic Displays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273.1.3 Displays with Shape Memory Alloys . . . . . . . . . . . . . . . 273.1.4 Piezoelectric Displays . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
VIII Contents
3.1.5 Other Actuator Mechanisms . . . . . . . . . . . . . . . . . . . . . . 283.2 The Tactile Displays Used . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.1 The Piezoelectric Effect . . . . . . . . . . . . . . . . . . . . . . . . . . 293.2.2 Mechanical Behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303.2.3 Geometrical Configurations . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 Drive Electronics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333.3.1 USB Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343.3.2 Data Bus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373.3.3 Variable Voltage Supply . . . . . . . . . . . . . . . . . . . . . . . . . . 393.3.4 Digital–Analogue Conversion . . . . . . . . . . . . . . . . . . . . . 393.3.5 Amplification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4 Force-Feedback Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423.4.1 The Haptic Loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433.4.2 Modeling the Contact Forces . . . . . . . . . . . . . . . . . . . . . . 433.4.3 Different Degrees of Freedom . . . . . . . . . . . . . . . . . . . . . 44
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4 Generation of Virtual Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 494.1 Sample Fabrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.1.1 Fibres, Yarn and Fabrics . . . . . . . . . . . . . . . . . . . . . . . . . 504.1.2 Selection of Sample Fabrics . . . . . . . . . . . . . . . . . . . . . . . 52
4.2 Kawabata Evaluation System for Fabrics . . . . . . . . . . . . . . . . . 534.2.1 KES-F Roughness Test . . . . . . . . . . . . . . . . . . . . . . . . . . . 544.2.2 KES-F Friction Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.3 Spatial Frequency Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 564.3.1 Selection of Appropriate Sections . . . . . . . . . . . . . . . . . . 574.3.2 Discrete Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . 584.3.3 Two-Dimensional Composition . . . . . . . . . . . . . . . . . . . . 60
4.4 The Correlation–Restoration Algorithm . . . . . . . . . . . . . . . . . . 614.4.1 Data Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 624.4.2 Two-Dimensional Composition . . . . . . . . . . . . . . . . . . . . 63
4.5 Surface Reconstruction from an Optical Surface Scan . . . . . . 644.5.1 Symmetry Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 654.5.2 Shape from Shading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.6 Artificial Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 684.6.1 Brownian Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5 Tactile Rendering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 755.1 Rendering Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5.1.1 Position Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 775.2 Experiment with a Force-Feedback Device . . . . . . . . . . . . . . . . 80
5.2.1 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Contents IX
5.2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 815.3 Vibrotactile Rendering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.3.1 Computation of Resulting Vibrations . . . . . . . . . . . . . . 855.3.2 Decomposition of Vibrations into Base
Frequencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 865.3.3 First Results with Brownian Surfaces . . . . . . . . . . . . . . 885.3.4 Further Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.4 Integration with a Force-Feedback Device . . . . . . . . . . . . . . . . . 915.4.1 Physical Simulation and Haptic Rendering . . . . . . . . . . 935.4.2 Integrated Interface Hardware . . . . . . . . . . . . . . . . . . . . 945.4.3 Software Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 955.4.4 Evaluation of the Integrated System . . . . . . . . . . . . . . . 96
5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6 Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
A Fabrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
Acronyms
ACM autocorrelation matrixCRA correlation–restoration algorithmDAC digital–analogue converterDFT discrete Fourier transformDIP dual in-line packageDOF degrees of freedomFFT fast Fourier transformfps frames per secondIC integrated circuitIPC inter-process communicationjnd just noticeable differenceKES-F Kawabata evaluation system for fabricslpi lines per inchMDS multidimensional scalingOPA operational amplifierPC Pacinian mechanoreceptive afferentRA rapidly adapting mechanoreceptive afferentrms root mean squareSA1 slowly adapting type 1 mechanoreceptive afferentSA2 slowly adapting type 2 mechanoreceptive afferentSFM statistical feature matrixSFS shape from shadingUSB Universal Serial BusVR virtual reality
Chapter 1
Introduction
Virtual Reality (VR) has a lot of applications ranging from entertainment to mechan-
ical design and medical training. VR systems are often used in training situations
where training in a real environment would be inappropriate and possibly even dan-
gerous. Pilots, for example, often practise on a flight simulator before flying with
a new type of aeroplane. Interestingly, flight simulators are also sold as games for
personal computers. Today, computer games are probably the most common VR
systems.
VR systems can be categorised by the modalities they support. In today’s systems
the modalities of seeing and hearing are the most commonly employed as these are
also the modalities in which we as human beings mostly exchange information.
They require least effort in terms of energy transfer, the corresponding sensory re-
ceptors are concentrated in the retina and the cochlea and can be excited remotely
with light and sound waves respectively.
In contrast to seeing and hearing the creation of appropriate haptic stimuli de-
mands very sophisticated hardware. Firstly, the skin with its size of 1.5–2 m2 is a
very large organ. Therefore most haptic devices focus on a rather small part of the
human body—usually the fingertip. Secondly, forces cannot be transmitted contact-
free with current technology. Thus haptic devices always need direct contact to the
parts of the skin where the forces are applied. Thirdly, the amount of energy is rel-
atively high compared to other modalities, e.g. if one wants to simulate the lifting
of an object with a mass of 500 g the haptic device has to create a force of ap-
proximately 5 N. All these properties make haptic simulation a complex task still
presenting a lot of problems awaiting a good solution.
Today’s VR systems provide haptic interaction with virtual objects only indi-
rectly via tools like a thimble or a pen-like probe or more specialised tools like, for
example, the yoke of a plane, the handle of a scalpel or the handles of a pair of scis-
sors. These tools usually appear twice in a VR system: as end effector transmitting
the forces from the haptic device and as a virtual representation of the tool in the
simulated virtual environment. The VR system is responsible of matching the phys-
ical state of the end effector and its virtual representation. Therefore, these tools can
be seen as link between the real and the virtual world. However, this rather simple
solution has two drawbacks. Firstly, special end effectors have to be designed for
D. Allerkamp: Tactile Perception of Textiles in a Virtual-Reality System, COSMOS 10, pp. 1–4.
springerlink.com c© Springer-Verlag Berlin Heidelberg 2010
2 1 Introduction
different applications, e.g. an end effector imitating the handle of a scalpel is prob-
ably useless for flight training of aviators. Secondly, it is not possible to directly
touch the virtual objects with the hand, which reduces the possible realism of a VR
system.
In order to simulate direct touch the parts of the skin that are in contact with
virtual objects have to be appropriately stimulated with a tactile display. Although
there exists a large variety of tactile displays only very little research has been done
on the tactile simulation of real objects. The latter is the topic of the work at hand.
In order to further narrow the research question, the virtual objects to be simulated
have been restricted to textiles and only one of the many different types of tactile
displays was investigated.
This work was part of the EU funded HAPTEX project, which aimed at develop-
ing a VR system for the visual and haptic presentation of textiles (cf. [4, 1, 2]). The
project was coordinated by the MIRALab at the University of Geneva which also
contributed the physical based simulation system of the fabrics. The Biomedical
Physics Group at the University of Exeter developed the tactile stimulator hard-
ware and was responsible for the multimodal integration and validation. The PER-
CRO Laboratory at the Scuola Superiore di Studi Universitari e di Perfezionamento
Sant’Anna in Pisa developed the force-feedback hardware which also carried the
tactile stimulator hardware. The SmartWearLab at the Tampere University of Tech-
nology provided a selection of fabrics together with measurements appropriate for
the project. And the Welfenlab at the Leibniz Universitat Hannover developed the
haptic rendering software computing appropriate signals for the force feedback and
the tactile devices.
The target scenario of the project is depicted in Fig. 1.1. The virtual cloth is
attached to a fixed stand. The user can touch, squeeze, rub and stretch the fabric
with the thumb and index finger, feeling appropriate tactile stimulation at the finger-
tips. Reaching the target scenario is a very challenging task, because force-feedback
and tactile simulation have to be integrated, posing more than only mechanical
Fig. 1.1 HAPTEX final
scenario from [1] (courtesy
of PERCRO Laboratory)
1 Introduction 3
difficulties. Therefore, an intermediate scenario has been defined as a test bed for
the tactile rendering. Here the fabric is completely fixed on a rigid support. The user
can tactually explore the fabric’s surface with one fingertip. Force feedback is only
available to define the plane of the surface.
This research has been conducted at the Welfenlab of the Leibniz Universitat
Hannover. The workgroup can draw upon many years of experience in the field of
haptics in VR environments. Already in the Nineties a cooperation with the Virtual
Reality Research Laboratory at the Volkswagen AG in Wolfsburg existed. From this
cooperation the doctoral theses [3] and [5] emerged.
Figure 1.2 shows the structure of this monograph. Chapter 1 provides an introduc-
tion to the problem tackled in this work and describes the context of each chapter.
Chapters 2, 3 and 4 build the foundation for the following chapters.
In Chap. 2 human perception is described from two different angles: in psy-
chophysics perception is regarded as a black box and reactions to stimuli are
!"#$%&'()*$&+,-.$/+*
!"#$%&'01/2.-22/+*'"*,'3-$4++5
!"#$%&'67".$/4%'8%*,%&/*9
!"#$%&':;-<"*'=%&.%#$/+*
!"#$%&'>?%*%&"$/+*'+@A/&$-"4'B-&@".%2
!"#$%&'C1%D/.%2'@+&
7".$/4%'B/<-4"$/+*
Fig. 1.2 Structure of this monograph
4 1 Introduction
investigated whereas in neurophysiology the functionality of single neurons is re-
garded. The former provides methods for the evaluation of the tactile renderers de-
veloped within this work and the latter provides more insight into human perception
and leads to a theoretical model that builds the theoretical frame of the research
described in this book. Finally, tactile perception is described in more detail.
In Chap. 3 an overview of existing tactile displays is given and the display used
within this work is presented. In order to generate appropriate electric signals for the
tactile display an electronics module had to be developed, which is also described in
the chapter. Since also force-feedback devices are used in Chap. 5 an introduction
to this class of devices is given.
The generation of virtual surfaces is the topic of Chap. 4. As the virtual surfaces
are intended to be representations of real fabrics the production and measurement
of fabrics is treated first. Then three different methods for the generation of virtual
representations from real samples are presented. The chapter concludes with the
generation of artificial virtual surfaces that do not have a real counterpart.
After all these preparations the main topic of this monograph, the tactile ren-
dering of virtual surfaces, is presented in Chap. 5. Several different approaches have
been investigated within this work which share a common software framework. This
framework manages the virtual fabrics positioned on a virtual workspace and con-
trols the device tracking the position of the tactile display. After the description of
the framework several experiments on tactile rendering and also the integration with
a force-feedback device are described in the chapter.
Chapter 6 summarises the results of this work and suggests directions for future
research on tactile rendering.
References
[1] Fontana, M., Marcheschi, S., Tarri, F., Salsedo, F., Bergamasco, M., Allerkamp, D.,
Bottcher, G., Wolter, F.E., Brady, A.C., Qu, J., Summers, I.R.: Integrating force and
tactile rendering into a single VR system. In: 2007 International Conference on Cyber-
worlds, pp. 277–284. IEEE Computer Society, Los Alamitos (2007)
[2] Magnenat-Thalmann, N., Volino, P., Bonanni, U., Summers, I.R., Bergamasco, M.,
Salsedo, F., Wolter, F.E.: From physics-based simulation to the touching of textiles: The
HAPTEX project. International Journal of Virtual Reality 6(3), 35–44 (2007)
[3] Rabatje, R.: Virtuelle Montage- und Demontageuntersuchungen mit Hilfe von flexiblen
Bauteilen unter Verwendung geeigneter Kraftruckkopplungsgerate. PhD thesis, Welfen-
lab, Universitat Hannover (2002)
[4] Salsedo, F., Fontana, M., Tarri, F., Ruffaldi, E., Bergamasco, M., Magnenat-Thalmann,
N., Volino, P., Bonanni, U., Brady, A.C., Summers, I.R., Qu, J., Allerkamp, D., Bottcher,
G., Wolter, F.E., Makinen, M., Meinander, H.: Architectural design of the HAPTEX sys-
tem. In: Haptex 2005 – Workshop on Haptic and Tactile Perception of Deformable Ob-
jects, Welfenlab, Universitat Hannover, pp. 1–7 (2005)
[5] Schulze, M.: Von computergraphischen zu haptischen Texturen. Virtual Reality fur den
Entwicklungsbereich Design/Styling in der Automobilindustrie. PhD thesis, Welfenlab,
Universitat Hannover (2005)
Chapter 2
Human Perception
Perception is the process of attaining awareness or understanding of sensory infor-
mation. Together with our motor skills it is nothing less than our gate to the reality
outside our psyche. As the scope of this monograph is the simulation of a certain
facet of reality, namely the tactile properties of textiles, a good insight into human
perception is a necessary prerequisite. The intention of this chapter is to cover the
aspects of human perception that are important for this task.
In Sect. 2.1 an introduction to psychophysics is given. It provides theories and
methods for the measurement of a subjects internal response to physical stimuli.
Section 2.2 describes the biological background of our perception. It covers neurons,
the structure of our nervous system and different methods for the measurement of
neural activity. After these rather broad treatments of perception Sect. 2.3 focuses
on tactile perception. It justifies the way the virtual textiles are rendered in Chap. 5.
2.1 Introduction to Psychophysics
Psychophysics is a discipline of psychology that establishes a relationship between
a physical stimulus and its sensation (percept). Like in physics attainment of knowl-
edge is achieved by finding theoretical models describing experimentally gained
data.
The term “psychophysics” was coined by Gustav Theodor Fechner in his work
“Elemente der Psychophysik” released in 1860. Fechner states that bodily and con-
scious facts are different sides of one reality and that a mathematical relation can be
established between them: In order that the intensity of a sensation may increase in
arithmetical progression, the stimulus must increase in geometrical progression.
The latter is known as the Weber-Fechner law and, together with Stevens’ power
law, will be described in more detail in Sect. 2.1.2.
D. Allerkamp: Tactile Perception of Textiles in a Virtual-Reality System, COSMOS 10, pp. 5–23.
springerlink.com c© Springer-Verlag Berlin Heidelberg 2010
6 2 Human Perception
2.1.1 Just Noticeable Differences and Weber’s Law
A just noticeable difference (jnd) is the smallest difference in the intensity of two
stimuli that is still detectable by subjects. Its measurement leads to a psychometric
function as shown in Fig. 2.1. Such a function usually resembles a sigmoid function
with the percentage of the correctly noticed differences displayed on the ordinate
and the difference in the intensity of the stimuli on the abscissa. The jnd is by defi-
nition the point where the function crosses 50%.
0
20
40
60
80
100
0 5 10 15 20
co
rre
ctly n
otice
d d
iffe
ren
ce
s in
%
difference in intensity
psychometric curvetheoretical curve
Fig. 2.1 An example of a psychometric function
A special case of a jnd is the absolute threshold which is the smallest detectable
level of a stimulus.
In 1834 Ernst Heinrich Weber studied jnds between stimuli and found the fol-
lowing relation, today known as Weber’s law:
ΔR = cwR (2.1)
where ΔR denotes the jnd to a given stimulus intensity R. The constant cw depends
on the kind of stimulus.
2.1 Introduction to Psychophysics 7
2.1.2 Weber-Fechner Law and Stevens’ Power Law
Weber’s law only relates jnds to stimulus intensities. It is not a psychophysical scale
E(R) relating the intensity of a stimulus R to a value E indicating the intensity of
the corresponding percept. E(R) is assumed to be a continuous function. Fechner
constructed such a psychophysical scale based on Weber’s law. Its construction is
described below.
We define the following sequence of stimuli, where R0 denotes the absolute
threshold:
R1 = R0 + ΔR0 = R0 + cwR0 = (1 + cw)R0
R2 = R1 + ΔR1 = (1 + cw)2R0
...
Rn = (1 + cw)nR0
Assuming that the difference in perceived intensity is the same for all jnds we have
E(Ri+1)− E(Ri) = E(R + ΔR)− E(R) =: cp. (2.2)
We claim E(R0) = 0, i.e. the perceived intensity is zero at the absolute threshold,
and we arbitrarily set cp = 1. Now we can replace Rn by R(E) and use (2.2) to
construct the inverse function of the psychophysical scale:
R(E) = (1 + cw)ER0 (2.3)
By inverting this function and setting c = 1ln(1+cw) we obtain the Weber-Fechner law:
E = c lnR
R0
(2.4)
The Weber-Fechner law is not only a theoretical construct. It is also part of our
everyday life. A well-known example is the keyboard of a piano. An octave is always
perceived as the same interval and therefore there are always 12 semitones in an
octave. However, from the lower to the higher pitch the corresponding frequency
is always doubled. In other words: As the keys represent the sensation increasing
in arithmetical progression, the corresponding frequencies increase in geometrical
progression. (Most of today’s keyboards are well temperamented, so the situation is
a little more complicated, but this is definitely out of the scope of this work.)
One hundred years after Fechner constructed his famous psychophysical scale
the psychologist Stanley Smith Stevens designed a different method for the con-
struction of such scales: Subjects were presented a reference stimulus which was
assigned a number value reflecting the perceived intensity. The subjects then had
to rate the intensity of successive stimuli with respect to the reference stimulus, i.e.
if the reference had an intensity value of for instance 100 another stimulus with only
8 2 Human Perception
half the intensity of the reference should be rated with an intensity value of 50 by
the subject. According to Stevens the results could be well approximated by
E = kRb (2.5)
where E is again the perceived intensity and R is the intensity of the stimulus. The
exponent b depends on the type of stimulation and k is a proportionality constant
depending on the units used. Equation (2.5) is known as Stevens’ Power Law.
For some modalities, e.g. the intensity of light, the Power Law resembles the
Weber-Fechner Law, while for others, e.g. electric shocks, Stevens’ Power Law
turns out to be superior. Stevens’ law also proved of value because it is applica-
ble to nearly every dimension of physical perception. Psychologists used it for a
variety of scales, e.g. tone pitch, length, beauty or weight of crimes.
Human perception is a very complicated process. Approximating it with simple
formulas such as the Weber-Fechner Law or Stevens’ Power Law appears to be very
naive. But their applicability to a lot of practical examples proves their usefulness.
Perhaps they might be compared to Newton’s laws of motion which later turned
out to be too simplistic to accurately reflect reality. However, with some limitations
these laws are still applicable and are indeed also applied in practice.
2.1.3 Measuring Psychometric Distances
In Sect. 2.1.1 the concept of jnds is introduced. How can the ability of a subject to
detect a difference be measured? The subject could of course simply be asked but
this is prone to response bias: the systematic tendency of a subject to answer in a
certain way that has nothing to do with sensory factors. This biased behaviour can
be measured with so called catch trials and then used to offset the original measured
data. See Sect. 2.1.5 for a detailed discussion. The result is the theoretical probability
that an arbitrary subject is able to detect the given difference. As shown in Sect. 2.1.1
these probability values are used to produce the psychometric function.
According to [4] we distinguish between six standard discrimination methods:
2-Alternative Forced-Choice method (2-AFC) Two different objects are presented
to the subject, who has to select the object with the strongest (or weakest) sensory
characteristic.
3-Alternative Forced-Choice method (3-AFC) Three objects are presented to the
subject, with two of them being identical. The subject has to select the object
with the strongest (or weakest) sensory characteristic.
Duo-Trio method Three objects are presented to the subject, with two of them
being identical. One of the objects is labeled as the “control” sample. The subject
has to decide which of the remaining two objects matches the control object. In
this configuration it is possible that the odd object is chosen as control sample.
Triangular method Three objects are presented to the subject, with two of them
being identical. The subject then has to tell, which one of the three objects is
different from the two other ones.
2.1 Introduction to Psychophysics 9
A–Not A method The subject is familiarised with two different objects labeled as
“A” and “Not A”. One of these objects is presented to the subject, who has to
decide whether it is “A” or “Not A”.
Same–Different method Two objects, which can either be identical or different,
are presented to the subject, who has to decide whether they differ or not.
For our experiments we mostly used the triangular method, also referred to as the
odd-man-out forced-choice procedure. Here the subject either really detects the
difference or just randomly draws an object. In the latter case the subject might
accidentally draw the right object with a probability of 13.
In our experiments we observed that subjects thinking they were just randomly
drawing an object still performed better than the theoretical probability suggests.
This might indicate that rather than measuring the conscious ability to detect dif-
ferences, the odd-man-out procedure also measures the sub-conscious ability. How-
ever, this assumes that subjects really tried to intuitively draw an object when not
being able to consciously detect the different object. Some subjects tended to sys-
tematically draw for instance the first object, so the results were in danger of be-
ing biased by the way subjects “randomly” drew objects. Therefore we generally
advised all subjects to intuitively draw the object when being unsure rather than
systematically choose one.
2.1.4 Statistical Analysis
For a given pair of stimuli the discrimination experiment is usually repeated several
times. Let us assume the experiment is repeated n times, then the subjects will detect
a difference in k of n times. Such an experiment is called Bernoulli experiment and
the probability of k is given by a binomial distribution
Pp(k) = bn,p(k) :=
(
n
k
)
pk(1 − p)n−k. (2.6)
The letter p denotes the probability that a subject detects the difference or chooses
the right object, respectively.
For given n and k we want to estimate the probability p, such that Lk(p) := Pp(k)becomes maximal:
Lk(p) := sup{Lk(p) : 0 ≤ p ≤ 1} (2.7)
p(k) is called maximum likelihood estimator. For a Bernoulli experiment we find
p(k) =k
n. (2.8)
Depending on the number of experiments n we can have more or less confidence
in the estimated probability p. For large n the error |p − p|is probably small. We
10 2 Human Perception
can define a confidence interval C(k) for which we can assume that p ∈C(k) with a
certain probability, called confidence level 1 − α:
Pp(p ∈C(k)) ≥ 1 − α. (2.9)
For many domains α = 0.05 is considered to be sufficiently low. This definition is
not unique as there are various intervals fulfilling (2.9), a trivial example is C(k) =[0,1]. For Bernoulli experiments a table of confidence intervals for confidence levels
0.95 and 0.99 that is suitable for our needs is provided in [5].
Related to confidence intervals is the concept of statistical hypothesis testing.
Here we want to decide if an estimated parameter belongs to a previously defined
interval. For our experiments we mostly want to know whether the subjects just
guessed, denoted by p = p0. This is called the null hypothesis. If we can falsify
the null hypothesis we can assume the alternative hypothesis p > p0 which means
that the subjects sometimes were able to detect a difference. As it is not possible
to measure the probability p directly we have to rely on a Bernoulli experiment
again. Assuming that the null hypothesis holds, we can explain the outcome k of
the experiment with the binomial distribution bn,p0(k). We now define t to be the
smallest value that fulfills
Pp0(x ≥ t) =
n
∑x=t
bn,p0(x) ≤ α. (2.10)
The value t is called critical value to the significance level α . Like for the confidence
intervals α = 0.05 is usually considered to be sufficiently low. If k ≥ t we may reject
the null hypothesis and assume the alternative hypothesis.
We are interested in the probability pd that a subject detects the difference of a
given pair of stimuli. This is not equal to the probability pm that a subject chooses the
right object in an odd-man-out experiment, because, as already stated in Sect. 2.1.3,
when the subject does not detect the difference he or she still chooses the right object
with a probability of 13. Given an estimation pm of pm we can estimate pd with
pd = max{0, pm −1
2(1− pm)}. (2.11)
2.1.5 Signal Detection Theory
In Sect. 2.1.3 experiments for the measurement of psychometric distances were in-
troduced. The subsequent statistical discussion in Sect. 2.1.4 only dealt with the
probability that a subject detects a difference between two stimuli. However, we are
more directly interested in the perceived distance between two stimuli. The Weber-
Fechner law described in Sect. 2.1.2 assigned a perceived distance value of cp to a
discrimination probability of 50% in (2.2). Under some assumptions it is possible
to assign a distance value to every discrimination probability. Signal-detection the-
ory provides a well-established framework for this kind of problems. Only a short
2.1 Introduction to Psychophysics 11
introduction to this theory can be given here. The interested reader is referred to
[19].
In a classical signal-detection experiment an observer listening to white noise has
to decide whether a faint tone (the signal) is added to the noise or not. There are four
cases: there is no signal and the observer does not report a signal (correct rejection),
there is no signal and the observer reports a signal (false alarm), there is a signal
and the observer does not report a signal (miss) or there is a signal and the observer
reports a signal (hit). The hit rate h is defined as
h =number of hits
number of signal trials(2.12)
and the false alarm rate f is defined as
f =number of false alarms
number of noise trials. (2.13)
In the same way we can define the miss rate and the correct rejection rate, but these
do not carry any additional information.
As stated in Sect. 2.1.3 such an experiment is vulnerable to bias, so the hit rate
alone does not suffice to describe the result of the experiment. But also together with
the false alarm rate the situation does not become immediately evident. It would be
much better to have a single value describing the detectability of the signal and
another value describing the bias.
Signal-detection theory makes three assumptions:
• The internal response of the observer to a stimulus can be represented by a single
number.
• The internal response is subject to random variation.
• The choice is based on a simple decision criterion applied to the magnitude of
the internal response.
Figure 2.2 shows an example of these assumptions. The internal response is repre-
sented as a value on the abscissa. It is a random variable drawn from one of two
probability distributions, each belonging to one of the two stimuli. The decision
criterion is described by λ . If the internal response is larger than λ the signal is
detected otherwise not.
A detection model has to specify the form of the probability distributions. In the
equal-variance Gaussian model both distributions are assumed to be normal distri-
butions with a variance of one. The mean of one distribution is set to zero, the mean
of the other to d′. With these assumptions λ and d′ can be estimated from f and h:
λ = −Z( f ) (2.14)
d′ = Z(h)−Z( f ). (2.15)
Z denotes the inverse of the cumulative normal distribution. A derivation of (2.14)
and (2.15) is given in [19]. λ and d′ describe the result of the experiment much
12 2 Human Perception
0
No Yes
dλ
Stimulus 1 Stimulus 2
internal response
Fig. 2.2 Decision based on probability distributions for two different stimuli
clearer than f and h: d′ represents the detectability of the signal whereas λ repre-
sents the bias.
The detectability d′ can not be influenced by the observer, but λ is freely cho-
sen. One can assume that an unbiased observer chooses a criterion such that the
likelihood of a correct answer is maximised (see Chap. 9 of [19]). For many of our
discrimination experiments the triangular method was used, for which [8] provides
a table assigning a d′ to a given probability. In such an experiment the internal re-
sponse to two of the objects is drawn from one probability distribution, because
these objects create similar stimuli. The remaining object creates a different stimu-
lus and so the corresponding internal response is drawn from another distribution.
Craven assumes that the observer chooses the object maximising the likelihood that
the remaining objects are drawn from a single distribution. He shows that this rule
is equivalent to simply choosing the object which is farthest from the mean of all
objects. Because of the difficulty of analytically expressing the relationship between
probability and d′ the table was generated by the Monte Carlo method. Figure 2.3
depicts Craven’s result.
2.1.6 Multidimensional Scaling
So far the considered scales related only one-dimensional stimuli and their subjec-
tive correlates, which were also expressed with only one dimension. The scale was
thus expressed as a function E : IR → IR. Nevertheless, in many stimulus domains
the dimensions are not known. Often even the number of relevant dimensions is not
clear. Here, a versatile tool called multidimensional scaling (MDS) turns out to be
very useful. Given measurements of similarity among pairs of objects, MDS repre-
sents these measurements as distances between points of an embedding multidimen-
sional space. Depending on the dimension of this space the distances between the
2.1 Introduction to Psychophysics 13
0
1
2
3
4
5
6
7
0 0.2 0.4 0.6 0.8 1
dis
tan
ce
probability
Fig. 2.3 Distances d′ for given probabilities according to [8]
points can be consistent—to a greater or lesser extend—with the measured similari-
ties. The overall error is called stress and decreases with an increase of dimensions.
It vanishes for a dimensionality m ≥ n− 1 where n is the number of objects. How
the positions of these points are computed is not described here. As we only use the
MDS, we do not bother with such details. The interested reader is referred to [6].
The following example might be enlightening: In [10] 14 colours differing only
in their wavelength but not in their brightness or saturation were used. Each of all
possible pairs of different colours was projected onto a screen and 31 subjects were
asked to rate the similarity on a scale from 0 (no similarity) to 4 (identical). The
averaged and normalised ratings are summarised in Table 2.1. The first row and
column show the wavelengths of the colours in nanometers.
The same data is depicted in Fig. 2.4, where MDS was used to express proxim-
ities as geometrical distances. The result reminds of the CIE chromaticity diagram
shown in Fig. 2.5. This similarity is quite interesting, because the CIE diagram has
been constructed with a very different, more complicated experiment (for details see
Chap. 13.2.2 of [11]).
This example not only shows how MDS works in principle but also illustrates a
case where a one-dimensional physical entity is represented in a two-dimensional
perceptual space. The colour circle becomes a one-dimensional manifold embedded
in this perceptual space.
14 2 Human Perception
Table 2.1 Similarity matrix of colours from [10]
nm 434 445 465 472 490 504 537 555 584 600 610 628 651 674
434
445 .86
465 .42 .50
472 .42 .44 .81
490 .18 .22 .47 .54
504 .06 .09 .17 .25 .61
537 .07 .07 .10 .10 .31 .62
555 .04 .07 .08 .09 .26 .45 .73
584 .02 .02 .02 .02 .07 .14 .22 .33
600 .07 .04 .01 .01 .02 .08 .14 .19 .58
610 .09 .07 .02 .00 .02 .02 .05 .04 .37 .74
628 .12 .11 .01 .01 .01 .02 .02 .03 .27 .50 .76
651 .13 .13 .05 .02 .02 .02 .02 .02 .20 .41 .62 .85
674 .16 .14 .03 .04 .00 .01 .00 .02 .23 .28 .55 .68 .76
Fig. 2.4 MDS representa-
tion for color proximities in
Table 2.1 (from [6])
The dimensionality m of the embedding space has not been chosen arbitrarily. As
stated above setting m >= n−1 = 13 would be an uninteresting choice, because the
colours then can be perfectly represented in these spaces. We are more interested
in the minimal number of dimensions that still results in an acceptable error. In
this example the stress of the 1D, 2D and 3D MDS representations is 0.272, 0.023,
0.018, respectively. So two dimensions are clearly preferable to only one dimension.
But higher dimensions do not have much of an effect, because the stress of the 2D
solution is so close to zero already.
2.2 The Human Nervous System 15
Fig. 2.5 CIE 1931 xy chro-
maticity diagram (Source:
Wikimedia Foundation)
2.2 The Human Nervous System
In order to understand human perception it is necessary to gain some insight into the
human nervous system as this is the system that processes all information acting on
the sensory organs. It performs three tasks: sensory input, integration of information
and motor output. The content of this section is mainly based on [7].
2.2.1 Neurons
The basic unit of the nervous system is the neuron. Although the functionality of a
single neuron is relatively simple, the interaction of several million neurons—every
cubic centimetre of the human brain contains more than 50 million—is able to fulfil
very complex tasks. The complexity is caused by the large number of intercon-
nections, every neuron communicates with thousands others. Together the neurons
control all aspects of our perception and motor activity. They allow us to learn, to
remember and to consciously perceive our environment and ourselves.
Neurons are electrically excitable cells that are composed of a cell body (called
soma), dendrites, and an axon (see Fig. 2.6). The dendrites collect signals from other
cells and forward these information as an electrical signal to the soma. Information
sent by the neuron is transmitted by its axon. These are usually much longer than
the dendrites, some more than a metre. At its end the axon branches out into several
terminals which release neurotransmitters to transfer information to other cells. The
contact where this information is transmitted is called synapse.
While a neuron is in a resting state it has an electrical potential of −70 mV com-
pared to the outside of the cell. This so called membrane potential can be changed by
16 2 Human Perception
Fig. 2.6 Signal propagation
in neurons (Source: U.S.
National Institute on Aging)
outside influences, usually by the neurotransmitters from other cells (see Fig. 2.7).
If the membrane potential reaches a threshold potential of usually −50 to −55 mV,
an action potential of about +35 mV is triggered. After this spike the membrane
potential returns to its resting potential of about −70 mV.
Fig. 2.7 A triggered action
potential
2.2 The Human Nervous System 17
The amplitude of the action potential is independent of the intensity of the stim-
uli. The action potential can thus be seen as a binary information: it is either there
or not. The intensity is rather coded by the frequency of the occurrences of the ac-
tion potentials, typically denoted by impulses per second. The maximal frequency is
given by the length of the refractory period, i.e. the time until the neuron recovered
its resting potential after an action potential. The action potential is transmitted to
other parts of the nervous system by the axon, which enables signal speeds of up to
150 m s−1.
2.2.2 Neural Integration
A single neuron can receive information from other neurons via thousands of
synapses. Each synapse influences the neuron either in an inhibitory way (decrease
of the membrane potential) or in an excitatory way (increase of the membrane po-
tential). Depending on the location of the synapse it has more or less effect on the
membrane potential. The effect of a synapse also decays some time after an impulse.
There are two kinds of summation: spatial summation means that the simulta-
neous effects of several synapses on different places on the neuron are combined,
whereas temporal summation means that the effects of several impulses of a single
synapse are combined. The latter only occurs if the pause between two successive
impulses is sufficiently short, i.e. the effect of the impulse does not fully decay.
2.2.3 Sensory Reception
Neurons communicate with each other via neurotransmitters. But how is informa-
tion coming from outside the nervous system translated into a code that can be
understood there? There must be some transducers that react to physical stimuli
and accordingly excite neurons. And indeed, for every perceivable kind of physi-
cal stimulus a sensory transducer exists. Such a transducer emits neurotransmitters
when excited by a stimulus it is reactive to. This signal is then forwarded by neu-
rons called afferent neurons into the central nervous system. There are also some
transducers like for example the pain receptors that are themselves neurons that can
directly send action potentials via their axons into the central nervous system.
The information is processed in the central nervous system by intraneurons and
often results in the excitation of efferent neurons, also called motor or effector neu-
rons, which then excite the muscles they are connected to. A very simple example of
this process is the patellar reflex, which is sometimes tested by physicians. When the
patellar tendon is tapped just below the knee, the lower leg automatically, i.e. with-
out intentional control, kicks forward. The tap excites elongation receptors inside
the quadriceps muscle which is connected to the patellar tendon. Afferent neurons
connected to the elongation receptors forward this information to intraneurons in-
side the spinal cord. These excite efferent neurons controlling the quadriceps muscle
which then contracts and in this way lifts the lower leg. In the same time the effer-
ent neurons controlling the opposite muscle are inhibited to make this movement
possible.
18 2 Human Perception
The elongation receptors inside the muscles, also called muscle spindles, are ex-
amples of mechanoreceptors. This type of receptors respond to mechanical stress
or mechanical strain. The function of the muscle spindles is to send propriocep-
tive information, i.e. information on the relative position of neighbouring parts of
the body, to the central nervous system. Another kind of mechanoreceptors are hair
cells, which can be found in the cochlea of the inner ear and in the vestibular system.
The latter is also known as balance system and provides information on accelera-
tion and the direction of gravity. The cutaneous mechanoreceptors are responsible
for touch perception and are described in detail in Sect. 2.3.
Nocioceptors are free dendrites that send signals to the central nervous system
that cause pain in response to a possibly damaging stimulus. Different types of no-
cioceptors react to extreme heat, strong pressure or chemical substances like his-
tamine and several acids. Pain is one of the most important perceptions, because it
can result in a defensive reaction that prevents damage to the body.
Another kind of receptors are the thermoreceptors which are part of the system
that regulates the temperature of the body. These receptors are not fully understood
yet.
Chemoreceptors detect chemical stimuli in their environment. These can either
be very specialised to a certain substance or can react to a broad class of stimuli.
The olfactory and gustatory systems use chemoreceptors to detect smell and taste,
respectively. There are also chemoreceptors in the human body, that control blood
parameters like the concentration of oxygen and carbon dioxide.
Photoreceptors respond to electromagnetic radiation with wavelengths of about
400–700 nm. This is the range of light visibile to the human eye. Colour is per-
ceived by three types of photoreceptors that respond to different parts of the visual
spectrum.
2.3 Human Tactile Perception
Tactile perception is the sensation of pressure, vibration and temperature via recep-
tors of the skin. Together with proprioception, which delivers information on the
position of the joints and the tension inside of the muscles and the tendons, it be-
longs to the haptic perception.
2.3.1 Mechanoreceptive Afferents
The glabrous skin of the human hand is innervated by four types of mechanorecep-
tive afferents, which are responsible for the perception of pressure and vibration.
They are presented in Table 2.2. They can be categorised by receptor type, rate of
adaptation, size of the receptive field and by their function. In this context adap-
tation is the change over time in responsiveness of a sensory system to a constant
stimulus. The receptive field of an afferent is the area of the skin where stimuli have
an effect on its firing rate. The functions of the afferents are summarised according
to [15, 14].
2.3 Human Tactile Perception 19
Table 2.2 Characteristics of mechanoreceptive afferents
Afferent Receptor type Adaptation Receptive
field
Function
SA1 Merkel discs slow small form and texture
SA2 Ruffini’s corpuscles slow large hand conformation and forces
acting on the hand
RA Meissner corpuscles rapid small minute, low-frequency skin
motion and feedback required
for grip control
PC Pacinian corpuscles rapid large detection of distant events by
vibrations
The receptive fields of the slowly adapting type 1 (SA1) and rapidly adapting
(RA) afferents are very small in the fingertip, about 3 mm2, whereas the fields of the
slowly adapting type 2 (SA2) afferents are large and even larger for Pacinian (PC)
afferents. The latter spread over the whole finger or half of the palm.
SA1 sensors measure the indentation depth of a mechanical stimulus to the skin
and continually send impulses to the central nervous system during the whole dura-
tion of the stimulus. Having small receptive fields they can collect information on
form and texture of surfaces.
Similar to the SA1 sensors the SA2 sensors continually send impulses during the
whole duration of the stimulus. But their receptive fields are much larger and they
mainly react to shear forces inside the skin.
RA sensors react to the change of a stimulus and thus belong to the differential
receptors. Their impulse frequency depends on the velocity of the indentation.
PC sensors measure the acceleration of a pressure stimulus and thus are espe-
cially appropriate for the detection of vibration. In [15, 14] it is therefore concluded
that PC sensors are mainly responsible for the detection of remote events via vibra-
tions. They are most sensitive to vibrations between 100 and 300 Hz.
2.3.2 Perception of Form and Texture
Independent of the functionality of single receptors the sense of touch can also be
investigated with psychophysical experiments. MDS was used in [12] to find tactile
properties characterising surfaces. The sample surfaces employed in the experiment
could be represented as points in a three-dimensional metric space.
The relation between mechanical deformation of the skin and the firing rate of
the SA1 sensors was analysed in [9] by means of a finite element model of the
fingertip. Measurements on neural activity recorded on monkeys were used to match
the model. A model based on continuum mechanics for the prediction of the SA1
and RA sensors is presented in [18].
20 2 Human Perception
2.3.3 Perception of Fine Textures
SA1 receptors cannot be the only receptors responsible for the perception of rough-
ness, because humans are able to discriminate very fine textures with structure sizes
well below the resolution of the SA1 receptors.
Experiments described in [13, 1, 3, 2] show that these fine textures are perceived
by the PC receptors. While being rubbed over a surface the skin is oscillated in a
way characteristic of the surface’s texture, which enables the discrimination of fine
surface textures.
The importance of the fingerprints for the perception of fine textures is investi-
gated in [17]. Two sensors imitating the geometrical and mechanical properties of
the human fingertip were developed. Only one sensor surface was patterned with
ridges mimicking the fingerprints. Comparing the frequency responses of the sen-
sors, the sensor with ridges shows a reduced damping in a frequency range that
corresponds to the sensitivity range of the PC sensors at a typical exploration ve-
locity of 12–15 cm s−1. The authors conclude that the fingerprints play an important
role in the perception of fine surface textures.
2.3.4 Perceptional Model of Tactile Simulation
With the information given in this chapter a theoretical framework for the following
chapters can be defined. Figure 2.8 outlines such a model, which was inspired by
a systems theoretical model of human perception as proposed in [16]. On the left
side of the figure the steps of the information processing taking place when directly
touching a surface are depicted. The right side equals the left side in most parts, but
the surface is palpated indirectly via the combination of a tactile sensor, renderer and
display. Because of the latter, which can be seen as a filter of the signals, relevant
information is lost and the stimuli and thus also all following steps of perception
differ on both sides.
In this model direct and indirect touch can be compared on every level of human
perception:
• On the level of the stimuli an objective measurement with appropriate sensors is
possible. These sensors should correspond to the human fingertip in their geo-
metrical and mechanical properties.
• Neural activity of single afferent nerves can also be measured. This, however, is
only possible with invasive methods.
• The sensations emerging from neural activity can be investigated with appropri-
ate methods of psychophysics as explained earlier in this chapter.
• On the level of symbolic perception psychophysical experiments are also
possible.
Because of redundancies on the levels of information processing there exist subsets
of the stimuli whose elements cause the same percept. This implies a partition into
equivalence classes of the set of stimuli.
2.3 Human Tactile Perception 21
!"#$%!&#'()"*
#+ ,-*'./!%.!/0'-)
%!)&!%!&#'()"*
#!)#"0'-)#
)!$%"*".0'1'0+
#0' $*'
203
4'
.'203
"'203
#'203
&203
4&
.&203
"&203
#&203
#$%5".!/%-/!%0'!#
%!"*#$%5".!
#!)#-%
%!)&!%!%
&'#/*"+&'%!.060-$.7
!
"
!
"
8
8
9:; 9:;
#!)#-% #!)#-%
8/%-,! /%-,!
Fig. 2.8 A model of the contact mechanical interaction of the fingertip and the further pro-
cessing of the nervous system
22 2 Human Perception
2.4 Conclusion
In this chapter an introduction to psychophysics is given. It starts with the basic
laws of psychophysics and continues with methods for measuring psychometric dis-
tances. With a statistical analysis of the collected data and a detection model from
signal detection theory it is possible to gain a measure of subjective similarity. Given
a set of stimuli and psychometric distances for all pairs within this set, multidimen-
sional scaling represents stimuli as points of a multidimensional space. The latter
might be referred to as perceptional space of the given class of stimuli. The second
section of this chapter describes the human nervous system. This section is neces-
sary to understand the third section, which enlarges on human tactile perception. In
order to embed the research on tactile simulation into a theoretical framework a per-
ceptional model was developed within this work which is described in Sect. 2.3.4.
With the information given in this chapter the reader should be prepared to fully
understand the simulation systems and the experiments described in Chap. 5.
References
[1] Bensmaıa, S.J., Hollins, M.: The vibrations of texture. Somatosensory & Motor Re-
search 20(1), 33–43 (2003)
[2] Bensmaıa, S.J., Hollins, M.: Pacinian representations of fine surface texture. Perception
& Psychophysics 67(5), 842–854 (2005)
[3] Bensmaıa, S.J., Hollins, M., Yau, J.: Vibrotactile intensity and frequency information
in the pacinian system: A psychophysical model. Perception & Psychophysics 67(5),
828–841 (2005)
[4] Bi, J.: Sensory Discrimination Tests and Measurements: Statistical Principles, Proce-
dures, and Tables. Blackwell, Malden (2006)
[5] Blyth, C.R., Still, H.A.: Binomial confidence intervals. Journal of the American Statis-
tical Association 78(381), 108–116 (1983)
[6] Borg, I., Groenen, P.J.F.: Modern Multidimensional Scaling: Theory and Applications,
2nd edn. Springer, Heidelberg (2005)
[7] Campbell, N.A., Reece, J.B.: Biology, 8th edn. Pearson/Cummings (2008)
[8] Craven, B.J.: A table of d′ for M-alternative odd-man-out forced-choice procedures.
Perception & Psychophysics 51(4), 379–385 (1992)
[9] Dandekar, K., Raju, B.I., Srinivasan, M.A.: 3-d finite-element models of human and
monkey fingertips to investigate the mechanics of tactile sense. Journal of Biomechani-
cal Engineering 125(5), 682–691 (2003)
[10] Ekman, G.: Dimensions of color vision. Journal of Psychology 38, 467–474 (1954)
[11] Foley, J.D., van Dam, A., Feiner, S.K., Hughes, J.F.: Computer Graphics: Principles and
Practice, 2nd edn. Addison-Wesley, Reading (1990)
[12] Hollins, M., Faldowski, R., Rao, S., Young, F.: Perceptual dimensions of tactile surface
texture: A multidimensional scaling analysis. Perception & Psychophysics 54(6), 697–
705 (1993)
[13] Hollins, M., Bensmaıa, S.J., Washburn, S.: Vibrotactile adaption impairs discrimina-
tion of fine, but not coarse, textures. Somatosensory & Motor Research 18(4), 253–262
(2001)
References 23
[14] Johnson, K.O.: The roles and functions of cutaneous mechanoreceptors. Current Opin-
ion in Neurobiology 11(4), 455–461 (2001)
[15] Johnson, K.O., Yoshioka, T., Vega-Bermudez, F.: Tactile functions of mechanoreceptive
afferents innvervating the hand. Journal of Clinical Neurophysiology 17(6), 539–558
(2000)
[16] Kammermeier, P., Buss, M., Schmidt, G.: A systems theoretical model for human
perception in multimodal presence systems. IEEE/AMSE Transactions on Mechatron-
ics 6(3), 234–244 (2001)
[17] Scheibert, J., Leurent, S., Prevost, A., Debregeas, G.: The role of fingerprints in the
coding of tactile information probed with a biomimetric sensor. Science 323(5920),
1503–1506 (2009)
[18] Sripati, A.P., Bensmaıa, S.J., Johnson, K.O.: A continuum mechanical model of
mechanoreceptive afferent responses to indented spatial patterns. Journal of Neurophys-
iology 95, 3852–3864 (2006)
[19] Wickens, T.D.: Elementary Signal Detection Theory, Oxford (2002)
Chapter 3
Devices for Tactile Simulation
Nearly every office or home computer is capable of graphics or sound output in a
quality that, because of the limits of human perception, hardly needs to be improved.
The haptic impression however is limited to the contours of the keyboard and the
mouse.
In some special fields force-feedback devices are used, but these only simulate
mechanical contact with virtual objects via special tools (e.g. like a thimble) and
are therefore not applicable for more general uses. It would be better if the contact
could be simulated without such tools limiting the perceptional capabilities.
For a convicing simulation of direct contact with the skin a single force is not
sufficient. Instead the skin has to be deformed in a more realistic way by a so called
tactile display. In Sect. 3.1 existing tactile displays are evaluated according to their
usefulness for virtual reality applications.
The overview of existing tactile displays is followed by Sect. 3.2, which describes
the tactile displays used within this work in more detail.
The tactile signals calculated on a computer need to be transmitted to the tac-
tile display. An electronic circuit that receives tactile data from the computer and
converts these to analogue signals for the tactile display is presented in Sect. 3.3.
The combination of a tactile display with a force-feedback device is described
in a later chapter. Therefore, Sect. 3.4 provides a short introduction to the haptic
simulation with force-feedback devices.
3.1 Survey of Existing Devices
As there exist a lot of very different tactile displays it would not make sense to
describe all of them in this book. The main interest of this work is the applicability
of existing displays for virtual-reality simulations. This restriction already excludes
many displays that were only designed for the transmission of information, e.g. for
systems that help blind persons to navigate around obstacles (cf. [12]).
D. Allerkamp: Tactile Perception of Textiles in a Virtual-Reality System, COSMOS 10, pp. 25–47.
springerlink.com c© Springer-Verlag Berlin Heidelberg 2010
26 3 Devices for Tactile Simulation
If a user of a virtual-reality simulation wants to touch a virtual object he or she
will probably use the hand to explore the object. A restriction of the survey to tac-
tile displays for the hand is therefore only natural. As the fingertips are the most
sensitive parts of the human hand most displays are designed for that part of the
skin.
There are two completely different approaches to establish the connection be-
tween the finger and the display: the finger is either permanently connected to the
display or not. The latter allows for an active exploration of the simulated surface,
i.e. the user can move the fingers over the display to asses the properties of the
virtual surface. An active exploration with a display tied to a finger is only possi-
ble if the display follows the movements of the finger. This approach seems to be
more difficult to realise, but it is also more flexible. If the display is attached to a
force-feedback device as demonstrated, for example, in [13] it can simulate the sur-
face of curved objects in a three-dimensional space while the force-feedback device
delivers information on the elasto-mechanical properties of the object. Because of
its greater flexibility this approach is preferable for virtual-reality systems and this
survey is restricted to this kind of displays.
The most common approach in the development of tactile displays is based on
a dense array of stimulator pins. The fingertip is stimulated by the movement of
individual pins which are usually moved normally to the skin’s surface. Different
tactile stimuli are created by different patterns of pin movements.
In the following the large variety of different actuator principles is shown by pre-
senting a selection of tactile displays. However, this is not a comprehensive survey
which would go beyond the scope of this chapter.
3.1.1 Electromagnetic Displays
In [31], [26] and [18] electromotors are used to control the heights of the contact
pins of the tactile displays.
In [31] each pin is lifted by a small rotating lever. The presented prototype con-
tains 36 independently controlled pins with a maximal deflection of 2 mm. The pins
are positioned with an interspacing of 2 mm and are covered by a sheet of rubber
that functions as a spatial low-pass filter. The system is relatively large and reaches
a bandwidth of about 25 Hz.
In [26] 4096 pins are positioned on an area of 200 mm times 170 mm. The dis-
tance between adjacent pins is about 3 mm. The pins are lifted by elevating screws,
which allow for deflections of a few mm. Because of the transmission ratio of the
screws large forces can be generated. However, an update of the heights of the pins
might need as much as 15 seconds.
In [18] a linear actuator is used to adjust the heights of the display pins. In this
way contact forces as high as several Newtons at a maximal deflection of 2.5 mm
are reached. The prototype offers 400 pins on an area of 1 cm2, but it is very large
because of the large number of linear actuators.
3.1 Survey of Existing Devices 27
3.1.2 Pneumatic Displays
A pneumatic approach is pursued in [11]. The 16 pins can be separately controlled
with forces up to 2 N at a maximum height of 3.5 mm. Vibrations in the range of 20–
300 Hz are also possible. The pins are positioned with an interspacing of 1.75 mm
and can additionally be moved laterally to the skin with a pneumatic muscle.
In [23] the pressure inside 25 chambers made of silicone is controlled allowing
for a maximum extension of the chambers of 0.7 mm normal to the skin. With this
approach a bandwidth of only 5 Hz is reached and the distance between adjacent
contact points is about 2.5 mm. Because of the pneumatic actuation additional un-
wanted vibrations occur.
3.1.3 Displays with Shape Memory Alloys
Shape memory alloys change their shape depending on the temperature. In [19],
[32] and [30] wires made of these alloys are used to lift the pins of the dis-
plays. The approaches used differ in the transfer of the movement to the pins. All
presented solutions have problems with hysteresis and a high power dissipation.
The latter is mainly transformed into heat which requires active cooling for some
displays.
In [19] 24 pins are lifted by levers. In this way high amplitudes and high contact
forces can be created, but the bandwidth of about 10 Hz is relatively low.
In [30] 64 pins are pushed upwards by springs while the actuator wires pull the
pins downwards when contracting. The prototype only reached a bandwidth of about
1–3 Hz.
In [32] 10 pins are positioned in a single row. Each pin is attached to the middle
of an actuator wire. When the wire contracts it pushes the pin upwards. This display
allows a bandwidth of up to 30 Hz.
3.1.4 Piezoelectric Displays
Piezoelectric materials deform proportional to an electric voltage. This principle
is used in [15, 14] to control the vibration of 50 contact pins. The prototype only
allows a single vibration frequency of 250 Hz, the amplitude can be adjusted from 5
to 57 µm.
Amplitudes up to 0.7 mm can be generated by the display presented in [22]. The
bandwidth of the 48 contact pins is, however, limited to 20 Hz. The spatial resolution
is about 8 mm2 per pin.
A relatively high resolution of 1 mm2 per contact pin is described in [27]. The
presented display can independently control 100 pins from 20 to 400 Hz with an
amplitude of up to 50 µm.
28 3 Devices for Tactile Simulation
In [24] a different approach is taken: the 64 contact pins are not moved normally
but laterally to the skin’s surface.
3.1.5 Other Actuator Mechanisms
Electroactive polymers change their form under an electric voltage. In [20] an ionic
electroactive polymer is used for a tactile display with 55 actuators. It has a band-
width of 100 Hz but only works in water.
A rather unconventional solution is described in [8]. The skin is stimulated with
an air jet which creates a perception not comparable to the other tactile displays.
The prototype only offers one contact point.
As the skin is conductive, forces act on it in an electric field. This effect is used in
[29] to create friction forces. In this approach the finger moves over the display, i.e.
the area of the simulated surface is the same as the area of the display. The system
is sensitive to changes of the skin’s moisture.
In [16, 25, 21] the forces occuring in an electric field are used for miniature
actuators. Capacitor plates are linked by an elastic isolator which is contracted when
a voltage is applied. Because of their small size the actuators can be arranged in a
grid.
Electrocutaneous displays stimulate the receptors of the skin with electric im-
pulses. These displays can be very simple and compact, e.g. they could be inte-
grated in a glove. Also the electrodes can be arranged in a dense grid. However, the
variable electric resistance of the skin requires an adjustment of the current which
makes it sometimes impossible to avoid pain. In [17] a display with 49 electrodes is
presented.
3.2 The Tactile Displays Used
The Biomedical Physics Group at the University of Exeter kindly provided several
tactile displays for the experiments described in this work. All displays are devel-
oped further from the display described in [27] and also use piezoelectric bimorphs
for the transformation of electrical signals into movements of the stimulator pins.
The displays mainly differ in their geometrical configuration.
In the colour model of computer graphics a colour consisting of a whole spec-
trum of light is substituted with the combination of only three colours creating
the same sensory impression. This model works because the human eye only pro-
vides three channels for the perception of colour. Virtually all computer screens
and televisions are based on this principle. The success of this model suggests that
a similar model for tactile perception might also be successful. The displays de-
veloped by the Biomedical Physics Group follow such a suggestion by Bernstein
et al. (cf. [9]). Like the colour model two different tactile channels of the human
skin are stimulated with sinewaves at two different frequencies. For the stimulation
3.2 The Tactile Displays Used 29
of the two types of mechanoreceptors a superposition of sinewaves at 40 and 320 Hz
was chosen.
3.2.1 The Piezoelectric Effect
If a piezoelectric crystal is deformed under mechanical pressure a voltage is induced.
This effect is known as the piezoelectric effect. Figure 3.1 shows this effect on
a schematic drawing of a quartz. The displacement of the negative and positive
ions results in a displacement of the centres of the positive and negative charges
and thus in a voltage. The inverse effect is also possible: If a voltage is applied to
the piezoelectric crystal, forces act on the positive and negative ions in opposite
directions, which result in a deformation of the crystal.
F
F
Q−
Q+
++
+
−
− −
+ +
+
Q−
Q+
−
−−
Fig. 3.1 Schematic drawing of the piezoelectric effect of a quartz
The inverse piezoelectric effect can be used to move the stimulator pins of a tac-
tile display. Piezoelectric crystals do not provide sufficient displacement at moder-
ate voltages, so either the voltages used in the system have to be increased or some
form of mechanical leverage is necessary. For the displays used in this work the
latter solution has been chosen. The piezoelectric material is used in a bimorph con-
figuration. Figure 3.2 shows a piezoelectric bimorph that is connected to a voltage
source such that one layer contracts while the other layer expands. As both layers
are bonded the whole bimorph bends when a voltage is applied. One side of the bi-
morph is clamped to make use of the deflection and to control the force at the other
end of the bimorph.
Readily available bimorphs were acquired from APC International Ltd., USA.
With these bimorph cantilevers of thickness 0.6 mm and a free length of 32 mm
displacement amplitudes of maximal 100 µm are reached.
30 3 Devices for Tactile Simulation
Fig. 3.2 Schematic drawing of a piezoelectric bimorph
3.2.2 Mechanical Behaviour
Electromechanical systems like the piezoelectric bimorphs usually do not transform
electrical signals into mechanical movements independent of the signal frequen-
cies. Therefore, special care has to be taken that signals in the intended frequency
range are transformed in an unbiased way. In [28] and [7] a mathematical model
of the piezoelectric drive system and the mechanical load presented by the skin is
described. The model assumes a linear behaviour of the piezoelectric material and
the skin load, which is reasonable for the relatively small amplitudes of less than
100 µm. With this assumption the system can be represented by a simple mass-on-
a-spring model as depicted in Fig. 3.3. This model predicts the resonance and the
frequency responses for different lengths of piezoelectric drive elements as depicted
in Fig. 3.4.
Fig. 3.3 Mass-on-a-spring
model of the piezoelectric
material and the skin load
piezoelectric
drive system
finger
In Fig. 3.5 the response of the system with and without skin load measured with
a miniature accelerometer is shown. From this data the effect of the skin can be esti-
mated: Significant stiffness of about 100 N m−1 and resistance of about 0.1 N s m−1
3.2 The Tactile Displays Used 31
Fig. 3.4 Predicted fre-
quency response for dif-
ferent length of the piezo-
electric drive element from
[28]
is added to the system. The additional effective mass of about 10−5 kg is negligible.
As can be seen in the figure the resonant frequency of the system is increased by the
load of the finger while the narrowness of the resonance (the Q-factor) is reduced.
Fig. 3.5 Measured fre-
quency response of the
piezoelectric drive element
with and without the me-
chanical load presented by
the skin from [28]
As the tactile stimuli only consist of superpositions of 40 and 320 Hz sinewaves,
bimorphs have been chosen which have a principal resonance that lies between the
two operating frequencies of 40 and 320 Hz.
3.2.3 Geometrical Configurations
For the experiments conducted within this work three tactile displays mainly differ-
ing in their geometrical configuration were used. Figure 3.6 shows these displays
together with a depiction of the arrangement of their stimulator pins as seen from
the direction of the finger. Each circle represents a contactor pin. Not all pins are
32 3 Devices for Tactile Simulation
2 mm
2 mm
2 mm
Fig. 3.6 The tactile displays used within this work
driven by an actuator. The moving pins are represented by filled circles. The other
circles represent stationary pins.
The first display has been developed for a previous study in which five of these
displays were used to stimulate the digits of one hand. Compared to the other dis-
plays it is relatively large.
The second display is quite similar to the first. But instead of a 5-by-5 array of
contactor pins it provides only 24 pins in a 6-by-4 configuration. The size of the
display is significantly reduced compared to the first display. To achieve this the
size of the bimorphs also had to be reduced.
The third display is intended for the integration with a force-feedback device as
described in Chap. 5. The bimorphs are positioned such that they do not lie beneath
the contactor surface. This allows thumb and fingers to be brought close together in
a system with more than one tactile display per hand. This display provides a curved
contactor surface matching the shape of the fingertip.
3.3 Drive Electronics 33
3.3 Drive Electronics
The tactile displays used within this work are operated with input signals at voltages
of up to 40 V rms. As the tactile signals are calculated on a digital computer an
electronic component converting the digital data to appropriate analogue signals had
to be developed. The analogue parts of the electronics have been designed following
suggestions by the Biomedical Physics Group at the University of Exeter, which also
provided the tactile displays used. The development and testing of some components
is presented in [10].
Nowadays the most common interface to peripheral hardware is the Universal Se-
rial Bus (USB). Nearly every modern personal computer provides at least one USB
connector. Therefore, it was decided that the drive electronics should be controlled
via USB.
Fig. 3.7 Schematic of the
drive electronics
DACOPA
Data b
us
OPA DAC
USB controller
Figure 3.7 shows a block diagram of the drive electronics. A USB controller
receives tactile data for all stimulator pins from the host computer. After some pro-
cessing of the data as described below it distributes the signals to the different chan-
nels via a data bus. The bus is designed for the unidirectional transmission of data
from the USB controller to the digital–analogue converters. To avoid noise from
ground loops the bus is galvanically isolated from the USB controller via optical
34 3 Devices for Tactile Simulation
couplers. In each channel a digital–analogue converter (DAC) receives the data for
a single stimulator pin and outputs an analogue voltage ranging from −Vref to +Vref.
The reference voltage Vref for the DAC is set by a variable voltage supply to a value
from 0 to 2.5 V for all channels. In this way the global gain of the system can be set.
An operational amplifier (OPA) is used to amplify the analogue signal to voltages
appropriate for the tactile display.
Figure 3.8 shows the developed drive electronics for a tactile display with 24
pins. The electronic board on the bottom of the figure hosts the USB controller, the
optical couplers and the variable voltage supply. Each of the six boards above hosts
a DAC for four channels, four amplifications circuits and a circuit for the address
logic of the data bus. The board on the right just provides connectors for the other
boards.
Fig. 3.8 The drive electron-
ics for one finger
3.3.1 USB Controller
The USB controller communicates with the host computer. It receives, processes
and forwards data to the DACs. An integrated circuit (IC) that provides all necessary
features is the CY7C64713 produced by the Cypress Semiconductor Corporation. It
includes a USB transceiver and an Intel 8051 microprocessor with 16 kBytes of on-
chip random access memory (RAM). The software running on the microprocessor
can be uploaded via USB, which makes the IC especially attractive for development
purposes.
The design of the supporting circuit is taken from [6]. Only a few capacitors
stabilising the voltage of the power supply and an oscillator with a resonance of
24 MHz are needed to run the IC. The IC can be directly connected to the USB.
Although the IC can also be directly connected to the power supply of the USB
it was decided to include a fixed-voltage regulator reducing the 5 V of the USB to
3.3 V preferred by the IC.
3.3 Drive Electronics 35
Fig. 3.9 The board hosting the USB controller and the variable voltage supply
Figure 3.9 shows the board hosting the USB controller. On the left third of the
board the CY7C64713 IC, the oscillator, the USB connector and the fixed-voltage
regulator can be seen. The black connector provides access to most ports of the IC
for debugging purposes.
The software running on the USB controller is called firmware. The firmware
processes the data sent from the host computer. For the tactile simulation of fabrics
only the relevant properties have to be modeled and displayed. Human performance
in tactile perception and the reduced abilities of the tactile display build the frame in
which tactile signals make sense. The display used within this work creates vibro-
tactile stimuli consisting of a superposition of only two sinewaves (see Sect. 3.2).
Therefore, instead of transmitting the whole tactile signal via USB only the time-
varying amplitudes of the two base signals have to be transmitted. This results not
only in a reduced amount of data to be transferred via USB but also in a reduced
amount of computation time on the host computer running the tactile renderer, be-
cause the actual waveform of the signal is computed by the firmware running on the
USB controller.
It should be noted that the restriction to only two frequencies is only implemented
in the software part of the system. The designed hardware is still capable of produc-
ing more complex signals.
The implementation of the firmware is able to serve at least 24 channels with a
sample rate of at least 5 kHz. It stores a table defining the wave form of the signal.
With some optimisations of the assembler code, the firmware reached a sample rate
of 10 kHz for 24 channels.
The output signal is created from a periodic function f with period length 2π .
Let ωi be the angular frequencies of the signals with indices i = 1;2 and let an,i(t)be the time dependent amplitudes. Then the output signal for channel n is computed
as
gn (t) = an,1(t) f (ω1t)+ an,2 (t) f (ω2t) . (3.1)
36 3 Devices for Tactile Simulation
The angular frequencies are defined with constants wi and a basic angular fre-
quency ω0:
ωi = wiω0 (3.2)
From the periodicity of the function f follows
f (τi) = f (wiω0t) (3.3)
τi := wiω0t −2πk = wi(ω0t −2πk
wi
) k ∈ Z, (3.4)
such that 0 ≤ τi < 2π .
For the digital processing the function f and the amplitudes an,i have to be dis-
crete in time and value. Let Δ t denote the time between two samples. Then the
function f can be represented as an array of length l. The value of f at time m is
found at position si,m = l2π τi (0 ≤ sm < l) within the array.
The output values are computed as follows:
gn,m = an,1 f [s1,m]+ an,2 f [s2,m] with m :=⌊ t
Δ t
⌋
(3.5)
Let ω0 := 2πlΔ t
be the basic angular frequency. Then
si,m+1 = (si,m + wi) (mod l) (3.6)
follows from (3.4).
The host computer transmits data to the USB controller with a well defined pro-
tocol. The first byte of a data packet determines the command to be performed.
Necessary data follow dependant on the command. The protocol is summarised in
Table 3.1.
The first and second commands are used to define the array f holding the values
of the waveform of the tactile signals. The command “create new table” destroys
a possibly already existing table and allocates memory on the USB controller for a
new table of the given length. As the table might be larger than the maximum size of
a data packet, the command “transmit part of the table” allows for the transmission
of only a part of the table.
The next command sets the number of channels, i.e. the number of stimulator
pins to be controlled by the system. This value has also an impact on the sample
rate of the generated tactile signals.
The command “transmit amplitudes” is frequently issued by the tactile renderer
running on the host computer. With this command the amplitudes an,i are set for
each channel and signal frequency.
The two following commands are used to start and stop the output of the USB
controller. With the command “start output” the increments wi for the two signal
frequencies can be configured.
3.3 Drive Electronics 37
Table 3.1 Protocol for the communication with the USB controller
Command Byte Value
create new table 0 0
1 length
transmit part of the table 0 1
1 offset
2 length
3– data
set number of channels 0 2
1 number of channels
transmit amplitudes 0 3
1 signal number (0 or 1)
2– amplitude data for the channels (descending order)
start output 0 4
1 reserved
2 reserved
3 increment for signal 1
4 increment for signal 2
stop output 0 5
set global gain 0 6
1 gain factor
The command “set global gain” is used to set the gain factor of the whole sys-
tem. The transmitted value is forwarded to the variable voltage supply that sets the
reference voltages −Vre f and +Vre f for the DACs.
3.3.2 Data Bus
The signals created by the USB controller are unidirectionally transmitted to DACs
via a bus system which makes the system scalable in the number of contactors sup-
ported.
The data bus consists of 8 data lines, 8 address lines and 2 control lines, namely
“write” and “load”. The availability of a valid date and address is signalled with a
high state on the control line “write” as depicted in Fig. 3.10. The data transmitted
in this way is expected to be buffered at the location specified by the address lines.
After all buffers have been updated a low state on the control line “load” signals
that the data should be used, which guarantees that all DACs update their output
simultaneously. The timing requirements of the bus are determined by the DACs
used.
38 3 Devices for Tactile Simulation
data
address
write
load
D1 D2 D3 D4
A1 A2 A3 A4
Fig. 3.10 Digital timing diagram of the data bus
As there are 8 lines for the data 256 different voltages are available to define
the output signals of the DACs. The bus also provides 256 different addresses of
which the 4 highest are reserved for the variable voltage supply, which means that
the system could be extended to drive up to 252 stimulator pins.
On the upper right of Fig. 3.11 the address logic circuit can be seen. It consists of
a 6 Bit dual in-line package (DIP) switch used to configure the address range of the
board and a comparator IC. The comparator IC compares the address range defined
by the switch with the highest 6 address lines and the “write” signal with a high
state to signal the transfer of data into the buffer. The two lowest address lines are
directly connected to the address ports of the four-channel DAC.
Fig. 3.11 The board hosting the DAC and the OPAs
The data bus galvanically isolates the USB controller and thus also the host com-
puter from the rest of the system to avoid problems with noise. It also amplifies
the signals from the USB controller. Both aims are reached with the circuit depicted
3.3 Drive Electronics 39
in Fig. 3.12. An optical coupler galvanically isolates both parts of the system. The
6N137 IC described in [5] features the necessary characteristics. Since the optical
coupler inverts the signal and since the output currency of the CY7C64713 is not
sufficient to drive the LEDs inside the optical couplers, an inverter is inserted be-
tween the output ports of the CY7C64713 and the input of the optical couplers. The
use of optical couplers has the additional advantage that a different voltage may be
used as ground voltage as needed for the DACs. The 18 optical couplers can be seen
in the middle of Fig. 3.9.
Fig. 3.12 The circuit gal-
vanically isolating the USB
controller from the rest of
the system
inV
outV
GND1 GND2
VCC
3.3.3 Variable Voltage Supply
The variable voltage supply is used to adjust the global gain of the system by setting
the reference voltages for the DACs −Vre f and +Vre f . Its circuit is placed on the
same board as the USB controller (see the right third of Fig. 3.9). Like the DACs
for the channels it is also connected to the data bus and thus can be configured by
the USB controller. A comparator compares the address on the data bus with a fixed
address range from 252 to 255. If data is available for one of these addresses it is
transfered to a DAC. In this circuit the AD558 IC is used for the digital–analogue
conversion (see [1]). Dependent on the digital input the DAC sets its output to a
voltage from 0 V to 2.55 V. This output is inverted twice by the circuit depicted in
Fig. 3.13, which is implemented by the L272 IC (see [2]).
3.3.4 Digital–Analogue Conversion
The DAC receives data from the bus, which is kept in a buffer until the “load” sig-
nal indicates the output of the value as the new voltage. For the digital analogue
conversion the MAX5100 IC described in [3], which provides four independent
channels, was used. For the selection of the channel the two lowest address lines
of the data bus were directly connected to the IC. The channels of the MAX5100
40 3 Devices for Tactile Simulation
Vin
Vref
Vref
−
10k
10k
10k
10k
Fig. 3.13 Amplification stages of the variable voltage supply
are already buffered. It provides a low-active input that initiates the transfer of the
buffered value to the analogue output. The MAX5100 can be seen in the middle of
the right half of Fig. 3.11.
Fig. 3.14 Passive, first order
low-pass RC filter
υout
υin
R
C
To reduce imaging effects caused by the discrete quantisation of the output values
of the DAC a passive, first order low-pass RC filter as depicted in Fig. 3.14 was
used. The cutoff frequency fc, i.e. by definition the frequency at which the signal is
damped by 3 dB, can be computed with
fc =1
2πRC. (3.7)
The filter used consists of a 33 kΩ resistor and a 4.7 nF capacitor, which results in
a cutoff frequency of roughly 1 kHz. The values of the resistor and capacitor were
chosen such that the sampling rate of the signals is well above and their maximum
frequency content is well below the cutoff frequency.
3.3 Drive Electronics 41
3.3.5 Amplification
High-voltage operational amplifiers (OPA) are used to amplify the output of the
DACs to the levels required by the bimorphs. The OPA551 described in [4] was
used as amplifier in this circuit. The implemented circuit is able to amplify the out-
put of the DACs, which have a maximum output of 2.5 V, to 60 V. To avoid high
voltages with respect to the ground voltage the 2.5 V signal is first amplified by a
non-inverting amplifier as depicted in Fig. 3.15.
Fig. 3.15 Non-inverting
amplifier with an adjustable
gain factor between 11 and
13.5
outυ
inυ
200k
50k
20k
The exact amplification factor can be adjusted with a potentiometer. The output
voltage from the first amplification stage is inverted with respect to the ground volt-
age by an inverting amplifier with a fixed amplification factor of −1 as depicted in
Fig. 3.16. Between the two resulting voltages the signal can in theory reach voltages
of up to 60 V.
Fig. 3.16 Inverting ampli-
fier with gain factor -1
υin
υout
20k
20k
Figure 3.17 shows the output of a single channel recorded with an oscilloscope.
The signal is a superposition of two sinewaves with frequencies 40 and 320 Hz.
42 3 Devices for Tactile Simulation
Fig. 3.17 The output of a channel
3.4 Force-Feedback Devices
Force-feedback devices are much more common than tactile displays. While the
latter are still rather experimental, a number of force-feedback devices is already
commercially available. Figure 3.18 shows such a device.
Fig. 3.18 PHANToM Omni
haptic device by SensAble
Technologies
Force-feedback devices only allow for a haptic interaction with virtual objects
via a special tool. Such a tool could be a pen-like probe, a thimble or more special
tools like for example the scalpel of a surgeon. The simulation of direct touch is
not possible. Instead the user of the system touches the real representation of the
tool, often called end effector. The force-feedback device tracks the position of the
3.4 Force-Feedback Devices 43
end effector and exerts forces on it. The end effector also has a representation in the
virtual scene that follows the position of the real end effector.
3.4.1 The Haptic Loop
In the real world bodies never overlap. When real bodies touch, forces act on these
bodies that prevent them from penetrating each other. The magnitude of the forces
depends on the stiffness of the bodies. The same behaviour is simulated in the virtual
scene. If the virtual end effector collides with another object in the virtual scene, the
occuring forces are computed and the force acting on the end effector is sent to the
force-feedback device. This uses its motors to create a similar force on the real end
effector. This haptic loop is depicted in Fig. 3.19.
Fig. 3.19 The haptic loop
!"#$% $$&'(#)&$*+#$
#!,,$#-$&.-!/01(,.!2$"(-!"
2/34+#(5.4+105(-+!,! .*+"-0(5.4#$,$
2!4+-+!,.! .$,&%$ $#-!"
!"#$.(#-+,6.!,.$,&%$ $#-!"
As the physical simulation of the virtual scene is implemented as a computer
program, the physical state of the scene can only be updated in discrete moments in
time. This, however, poses a problem: since the user decides on the position of the
end effector between two discrete moments, the user can move it to a position that
would require extremely high forces. This is particularly a problem if the stiffnesses
of the virtual objects are high. To prevent these high forces the system has to be
updated relatively often. Usually haptic systems are updated at 1000 Hz. Compared
to computer graphics, where frame rates of 25 frames per second (fps) are usually
sufficient, the relatively higher frame rate of force-feedback systems poses higher
requirements to the computation of the physical simulation.
3.4.2 Modeling the Contact Forces
As the force acting on the end effector has to be computed in a very short time
usually a simple model of the contact situation is used. The probably most simple
and therefore widely used contact model is the spring–damper model: if the position
44 3 Devices for Tactile Simulation
Fig. 3.20 The spring–
damper contact model
surface of the
virtual object
position at
the surface
position at
time t i−1
position at
time t i
of the end effector lies inside a virtual object a simple spring is attached to the end
effector and to a point on the surface of the object as depicted in Fig. 3.20.
At time frame ti−1 the end effector is still outside the object, but because of its
movement it is already inside the object at time frame ti. To counteract this physi-
cally impossible situation the spring forces the end effector out of the object. The
spring has a resting length of zero and a stiffness that is determined by the stiff-
nesses of the virtual object and the end effector. To avoid oscillation of the system a
damper is added to the spring.
3.4.3 Different Degrees of Freedom
A body in three-dimensional space has a three-dimensional position and a three-di-
mensional orientation. Altogether it has six degrees of freedom (DOF). However,
only some force-feedback devices fully provide six DOF. Many devices are able to
track only the position and not the orientation of their end effector. These devices
can only exert a three-dimensional force but no torque forces. One of these devices
is the PHANToM Premium 1.0 haptic device by SensAble Technologies that has
been used for an experiment described in Chap. 5.
There also exist some force-feedback devices that are able to track all six DOF,
but which cannot exert torque forces. Such a device was kindly provided by the
PERCRO laboratory at the Scuola Superiore Sant’Anna in Pisa for the integration
of the tactile display described in Sect. 3.2 with a force-feedback device as described
in Chap. 5.
3.5 Conclusion
A lot of different tactile displays have been created by various research groups, but
not one of these devices can create all tactile impressions necessary for a convinc-
ing tactile simulation. However, some displays follow interesting approaches that
are worth to be further investigated. The display used in this work is such a display,
References 45
because it is based on the interesting idea of substituting a whole spectrum of fre-
quencies with a signal consisting of only very few frequencies intended to cause an
equivalent sensation. Such a scheme is already successfully utilised for the render-
ing of colours and might also be promising for tactile rendering.
For the use of the displays with a tactile renderer running on a computer an elec-
tronic component had to be created to translate the digital signals into voltages suit-
able for the tactile display. This component has been designed following the tactile
two-channel model, but is flexible enough to create other waveforms as well. Four
modules of this drive electronics have been created and are in use at the MIRALab
at the University of Geneva, PERCRO at the Scuola Superiore Sant’Anna in Pisa,
the Biomedical Physics Group at the University of Exeter and the Institute of Man–
Machine Communication at the Leibniz University of Hannover.
A tactile display alone does not suffice to create a satisfactory simulation of direct
touch to the full extend, because it does not provide the forces necessary for the
creation of kinaesthetic sensory impressions. Therefore, it is sensible to combine a
tactile display with a force-feedback device as described in Chap. 5. The principle
function of these devices is described in Sect. 3.4.
References
[1] AD558. Analog Devices, Inc., One Technology Way, Norwood, MA, USA (1987)
[2] L272. SGS-THOMSON Microelectronics (1995)
[3] MAX5100. Maxim Integrated Products, 120 San Gabriel Drive, Sunnyvale, CA, USA
(1999)
[4] OPA551, OPA552. Burr-Brown Corporation, 6730 S. Tucson Blvd., Tucson, AZ 85706,
USA (1999)
[5] 6N137. Fairchild Semiconductor Corporation (2005)
[6] CY7C64713/14. Cypress Semiconductor Corporation, 198 Champion Court, San Jose,
CA, USA (2005)
[7] Allerkamp, D., Bottcher, G., Wolter, F.E., Brady, A.C., Qu, J., Summers, I.R.: A vibro-
tactile approach to tactile rendering. The Visual Computer 23(2), 97–108 (2007)
[8] Asamura, N., Tomokori, N., Shinoda, H.: A tactile feeling display based on selective
stimulation to skin receptors. In: Proceedings of the Virtual Reality Annual International
Symposium, pp. 36–42. IEEE Computer Society, Los Alamitos (1998)
[9] Bernstein, L.E., Eberhardt, S.P., Demorest, M.E.: Single-channel vibrotactile supple-
ments to visual perception of intonation and stress. The Journal of the Acoustical Soci-
ety of America 85(1), 397–405 (1989)
[10] Blume, S.: Konstruktion eines Gerates zur Simulation von Oberflachenstrukturen fur
den Einsatz in VR-Umgebungen. Student research paper, Welfenlab, Universitat Han-
nover (2006)
[11] Caldwell, D.G., Tsagarakis, N., Giesler, C.: An integrated tactile/shear feedback array
for stimulation of finger mechanoreceptor. In: Proceedings of IEEE International Con-
ference on Robotics and Automation, vol. 1, pp. 287–292 (1999)
[12] Cardin, S., Thalmann, D., Vexo, F.: Wearable obstacle detection system for visually
impaired people. In: Haptex 2005 – Workshop on Haptic and Tactile Perception of De-
formable Objects, Welfenlab, Universitat Hannover, pp. 50–55 (2005)
46 3 Devices for Tactile Simulation
[13] Fontana, M., Marcheschi, S., Tarri, F., Salsedo, F., Bergamasco, M., Allerkamp, D.,
Bottcher, G., Wolter, F.E., Brady, A.C., Qu, J., Summers, I.R.: Integrating force and
tactile rendering into a single VR system. In: International Conference on Cyberworlds,
pp. 277–284. IEEE Computer Society, Los Alamitos (2007)
[14] Ikei, Y., Shiratori, M.: Textureexplorer: a tactile and force display for virtual textures.
In: Proceedings of the 10th Symposium on Haptic Interfaces for Virtual Environment
and Teleoperator Systems, pp. 327–334 (2002)
[15] Ikei, Y., Wakamatsu, K., Fukuda, S.: Vibratory tactile display of image-based textures.
IEEE Computer Graphics and Applications 17(6), 53–61 (1997)
[16] Jungmann, M., Schlaak, H.F. (2002) Miniaturised electrostatic tactile display with high
structural compliance. In: Proceedings of Eurohaptics (2002)
[17] Kaczmarek, K.A., Tyler, M.E., Bach-y-Rita, P.: Pattern identification on a fingertip-
scanned electrotactile display. In: Proceedings of the 19th Annual International Confer-
ence of the IEEE Engineering in Medicine and Biology Society, vol. 4, pp. 1694–1696
(1997)
[18] Killebrew, J.H., Bensmaıa, S.J., Dammann, J.F., Denchev, P., Hsiao, S.S., Craig, J.C.,
Johnson, K.O.: A dense array stimulator to generate arbitrary spatio-temporal tactile
stimuli. Journal of Neuroscience Methods 161(1), 62–74 (2007)
[19] Kontarinis, D.A., Son, J.S., Peine, W., Howe, R.D.: A tactile shape sensing and display
system for teleoperated manipulation. In: IEEE International Conference on Robotics
and Automation, vol. 1, pp. 641–646 (1995)
[20] Konyo, M., Tadokoro, S., Takamori, T.: Artificial tactile feel display using soft gel ac-
tuators. In: IEEE International Conference on Robotics and Automation, vol. 4, pp.
3416–3421 (2000)
[21] Matysek, M., Lotz, P., Schlaak, H.F.: Braille display with dielectric polymer actuator. In:
Conference proceedings: Actuator. In: 10th International Conference on New Actuators
& 4th International Exhibition on Smart Actuators and Drive Systems, pp. 997–1000
(2006)
[22] Maucher, T., Meier, K., Schemmel, J.: An interactive tactile graphics display. In: Sixth
International Symposium on Signal Processing and its Applications, vol. 1, pp. 190–193
(2001)
[23] Moy, G., Wagner, C., Fearing, R.S.: A compliant tactile display for teletaction. In: IEEE
International Conference on Robotics and Automation, vol. 4, pp. 3409–3415 (2000)
[24] Pasquero, J., Hayward, V.: STReSS: A practical tactile display system with one mil-
limeter spatial resolution and 700 Hz refresh rate. In: Proceedings of Eurohaptics 2003,
pp. 94–110 (2003)
[25] Schlaak, H.F., Jungmann, M., Matysek, M., Lotz, P.: Novel multilayer electrostatic solid
state actuators with elastic dielectric. In: Bar-Cohen, Y. (ed.) Proceedings of the SPIE.,
Presented at the Society of Photo-Optical Instrumentation Engineers (SPIE) Confer-
ence, vol. 5759, pp. 121–133 (2005)
[26] Shinohara, M., Shimizu, Y., Mochizuki, A.: Three-dimensional tactile display for the
blind. IEEE Transactions on Rehabilitation Engineering 6(3), 249–256 (1998)
[27] Summers, I.R., Chanter, C.M.: A broadband tactile array on the fingertip. The Journal
of the Acoustical Society of America 112(5), 2118–2126 (2002)
[28] Summers, I.R., Brady, A.C., Syed, M.K., Chanter, C.M.: Design of array stimulators
for synthetic tactile sensations. In: Proceedings of the First Joint Eurohaptics Confer-
ence and Symposium on Haptic Interfaces for Virtual Environment and Teleoperator
Systems, pp. 586–587. IEEE Computer Society, Los Alamitos (2005)
References 47
[29] Tang, H., Beebe, D.J.: A microfabricated electrostatic haptic display for persons with
visual impairments. IEEE Transactions on Rehabilitation Engineering 6(3), 241–248
(1998)
[30] Taylor, P.M., Moser, A., Creed, A.: The design and control of a tactile display based on
shape memory alloys. In: IEEE International Conference on Robotics and Automation,
vol. 2, pp. 1318–1323 (1997)
[31] Wagner, C.R., Lederman, S.J., Howe, R.D.: A tactile shape display using RC servomo-
tors. In: Proceedings of the 10th Symposium on Haptic Interfaces for Virtual Environ-
ment and Teleoperator Systems, pp. 354–355. IEEE Computer Society, Los Alamitos
(2002)
[32] Wellman, P.S., Peine, W.J., Favalora, G., Howe, R.D.: Mechanical design and control
of a high-bandwidth shape memory alloy tactile display. In: Experimental Robotics V,
pp. 56–66. Springer, Heidelberg (1998)
Chapter 4
Generation of Virtual Surfaces
To achieve a realistic tactile simulation of fabrics a display, a tactile renderer and
virtual representations of the fabrics’ surface properties are needed. The two former
requisites are covered in other chapters, the latter is the scope of this chapter.
As this research aims at a realistic simulation of real fabrics a set of fabric sam-
ples has been acquired. These fabrics were used to test the methods for the genera-
tion of virtual surfaces and to conduct experiments for the evaluation of the tactile
renderers. In Sect. 4.1 these sample fabrics are described.
The first step in the generation of virtual surfaces from real surfaces is always the
measurement of the real fabrics’ properties. A standard for the measurement of fab-
rics has been proposed by Kawabata in [11]. Although this standard was orignially
intended to assess the subjective hand properties of fabrics, it was used within this
work to measure the properties relevant for the generation of virtual surfaces and is
therefore described in Sect. 4.2.
There exist many different ways to model the surface properties of fabrics. In this
work only a few methods could be tested. These are described in Sects. 4.3, 4.4 and
4.5. All these methods have in common that a fabric’s surface is represented as a
two-dimensional property map P(u,v). However, the description of these properties
depend on the model of the fabric. The generation of virtual fabrics is not a real-time
operation and may even be manually supported.
Virtual fabrics generated from real fabrics are not always suitable to test the tac-
tile renderers. Artificial surfaces having well-defined properties were used instead.
These surfaces are described in Sect. 4.6.
4.1 Sample Fabrics
The systems for tactile simulation developed in this work are not only theoretical
constructs but aim at practical applications in the future. Therefore their usefulness
has to be properly evaluated. In Chap. 2 the basis for appropriate evaluation meth-
ods is presented. This section is about the selection of fabric samples useful for
the evaluation of the whole tactile simulation system and for testing the different
algorithms.
D. Allerkamp: Tactile Perception of Textiles in a Virtual-Reality System, COSMOS 10, pp. 49–74.
springerlink.com c© Springer-Verlag Berlin Heidelberg 2010
50 4 Generation of Virtual Surfaces
Using fabrics for the evaluation is handy, because there exists a large variety of
fabrics with different surface properties. It is possible to test the simulation system
with samples having rather extreme properties, e.g. extremely rough or extremely
smooth. Nevertheless it is also possible to test the system with samples that are
not very different in order to investigate the ability of the system to simulate only
very little differences. However, fabrics are just a test case, the systems could be
generalised to a larger class of surfaces later.
4.1.1 Fibres, Yarn and Fabrics
Fabrics are usually made of yarns that are themselves made of fibres. There are also
some fabrics that are not made of yarns. These are called “non-woven” fabrics and
will be treated later in this section.
Fibres are much longer than thick, i.e. the ratio length/diameter is very high. For
example the length of a cotton fibre is about 2000 times greater than its diameter.
This property of fibres makes spinning of yarns possible, because a large number of
long, fine fibers grip each other when twisted in a thread.
Figure 4.1 shows how fibres can be categorised (cf. [19]). Natural fibres either
stem from plants like cotton, flax or jute or from animals like wool or silk. The
manufactured fibres can either be created from natural or synthetic material. While
natural fibres are always staple fibres, i.e. fibres with a length, there are some syn-
thetic fibres that are continuous filaments. The latter are of indefinite length and run
continuously throughout the length of the yarn without a break.
The choice of the fibre material already determines a large quantity of a fab-
ric’s properties. These properties can also be altered by applying special finishing
processes. Fibres differ for example in their fineness, absorption of moisture, static
electricity, tensile strength, extensibility, elastic recovery, stiffness, friction and their
visual appearance. Fabrics are often made of blends to profit from the properties of
the individual fibre types.
For the production of yarn the fibres are usually first carded and sometimes even
combed to align the fibres in one direction. After that the fibres are spun to form a
yarn, which is called “single yarn”. If two or more single yarns are twisted together
the result is called “folded yarn” and if two or more folded yarns are twisted together
they form a “cabled yarn”.
Fibres and yarns have a rather one-dimensional shape. Several mechanisms like
for example weaving, knitting or felting exist to obtain a two-dimensional fabric
from fibres or yarn, respectively. Fabrics are woven by interlacing two sets of threads
to each other. These two sets are called warp and weft. The warp threads are already
fixed to the loom while the weft threads run orthogonally to the warp threads. The
interlacing pattern, known as the weave, is obtained by lifting some of the warp
threads such that the weft thread currently interlaced runs under these warp threads
and over the others. Figure 4.2a shows such a pattern where every second warp
thread was lifted alternately. This pattern is called plain weave. It is the simplest
and most frequently used weave (cf. [19]).
4.1 Sample Fabrics 51
!"#$!%&'()$*+
,*%%#%-+* !""!# $%&' ()"*
.$-"*(/ +!!% ,&-./0*121/3&456*.*7 8-%9/03)%"-:&"*;/!./<-%;7
0(/*$!% =4>*4"!4
0!/#1!2"#$*3&'()$*+
!"#$!%&.-%45*$ ?-43!4* @!;&% A.-&3*"&"* =3*"&"* BC!3*%%
64/"7*"(2&.-%45*$ D!%C&6-;* D!%C*4"*. E%&4"&#* 111
8"7*$+ &.>!#/F->.* G%&44/F->.*
Fig. 4.1 Categorisation of different fibre types
Weaves can be described on squared paper as shown in Fig. 4.2b. Columns cor-
respond to warp threads while rows correspond to weft threads. Thus, each square
represents a junction of a warp and a weft thread. It is filled when, looking at the
fabric’s face side, the warp thread lies over the weft thread.
Knitted fabrics consist of a series of interlinked loops of yarn. Two main cate-
gories of machine knitting exist: weft knitting and warp knitting. Figure 4.3 shows
examples of these categories. A weft knitted fabric could be made from a single
yarn. Loops are formed in succession by needles with hooks at their tops. The hook
holds the loop. When the yarn arrives again it pulls the yarn through the existing
loop forming a new loop in this way. Warp knitted fabrics are produced from a set
of warp yarns. Loops are formed simultaneously by needles with hooks. To interlink
the parallel threads, the threads are moved sideways at intervals. Unlike weft knitted
fabrics warp knitted fabrics do not ladder, i.e. they cannot be unravelled.
There exist several different definitions of “non-woven” fabrics (cf. [22]). In this
work fabrics that are not produced by weaving or knitting are called “non-woven”
fabrics. Examples for non-woven fabrics are leather, which is made from animal
skin, and felt, which is directly made from fibres, not yarns.
52 4 Generation of Virtual Surfaces
(a) Plain weave fabric (b) Weave diagram of plain weave
Fig. 4.2 Plain weave fabric with its weave diagram
(a) Face of plain weft knitting (b) Back of warp-knitted half-tricot
Fig. 4.3 Examples of knitted fabrics
4.1.2 Selection of Sample Fabrics
A set of altogether 54 different fabric samples has been kindly provided by the Smart
Wear Lab at the Tampere University of Technology. These fabrics are presented in
[21] and the specification of all 54 samples is also tabulated in Annex A.
The samples were acquired in three subsets. The first set of 32 samples was cho-
sen such that the fabrics largely differed in many properties. It was expected to result
in a very wide range of different virtual fabrics. The samples were differently fab-
ricated (woven, knitted and non-woven), and they largely differed in their weight,
thickness and in their fibre contents (cf. [14]).
4.2 Kawabata Evaluation System for Fabrics 53
The second set consists of 10 samples. These were taken from a narrower col-
lection of men’s woollen or wool blend woven suit fabrics, weft and warp knitted
blended or synthetic fabrics containing elasthane (cf. [21]).
The third set of samples consists of 12 samples of which 6 are non-laminated
synthetic warp knits used for car seats. The other samples consist of the originally
non-laminated fabrics laminated to foam.
4.2 Kawabata Evaluation System for Fabrics
The subjective hand properties of fabrics are called “fabric hand”. These properties
are assessed when a real fabric is touched and manipulated with the hands. Several
more precise but slightly differing definitions exist (cf. [1]), but are not of greater
interest for this work. The objective measurement of fabric hand preoccupied tex-
tile researchers for decades (cf. [4]). Today several measurement standards exist.
The probably most well-known are the KES-F system (cf. [11]) developed by Sueo
Kawabata and the FAST system (cf. [15]) developed by the research organisation
CSIRO.
The KES-F system consists of standardised testers for the bending, compression,
tensile, shear, surface roughness and surface friction properties of fabrics. These
testers measure the properties in weft and warp direction and for the surface prop-
erties on the face and back sides of the fabrics. From the data recorded by each
tester characteristic values are calculated. Table 4.1 summarises these characteristic
values together with their physical units. Tensile, bending, shearing and compres-
sion are elasto-mechanical properties while surface properties, weight and thickness
are other physical properties related to the elasto-mechanical properties. The KES-F
system is described in more detail in [11].
The KES-F standard uses the unusual unit gram-force (gf) which is defined
as the magnitude of the force exerted on one gram of mass by standard gravity
(9.80665 m s−2). Thus, 1 gf is equal to 0.00980665N.
Although the KES-F system is only intended to produce the aforementioned char-
acteristic values, the testers could be used to obtain the whole measurements. To be
able to further process the recorded data, a system for the conversion of the analogue
measurements to digital data files was used as described in [14]. As the KES-F sys-
tem is not intended for such a modification special care had to be taken to avoid
problems with noise.
As the surface measurements are the most important for this work they are de-
scribed in more detail in this section. The surface properties were measured on both
sides (right and back side) of most of the 54 fabric samples. For each side there are
four measurement directions: weft and warp each in forward and backward direc-
tion. As most of the measurements were repeated there were altogether 611 mea-
surement files available for the experiments. However, some measurements had to
be discarded, because some fabrics were too difficult to measure, e.g. for some very
stretchable fabrics material accumulated in front of the sensor.
54 4 Generation of Virtual Surfaces
Table 4.1 The characteristic values of the KES-F properties (cf. [11])
Parameter Definition Unit
Tensile LT Linearity -
WT Tensile energy gf·cm/cm2
RT Resilience %
EM Max. extension %
Bending B Bending rigidity gf·cm2/cm
2HB Hysteresis gf·cm/cm
Shearing G Shear stiffness gf/cm·◦
2HG Hysteresis at 0.5◦ gf/cm
2HG5 Hysteresis at 5◦ gf/cm
Compression LC Linearity -
WC Compressional energy gf·cm/cm2
RC Resilience %
Surface MIU Coefficient of friction -
MMD Mean deviation of MIU -
SMD Geometrical roughness µm
Weight W Weight/unit area mg/cm2
Thickness T Thickness at 0.5 gf/cm2 mm
4.2.1 KES-F Roughness Test
The KES-F roughness tester measures a one-dimensional height profile along a seg-
ment of length 2 cm. Therefore, the fabric is moved under the sensor at 1 mm s−1
for 20 s as depicted in Fig. 4.4. The tension of the fabric is kept at 20 gf cm−1 along
the measurement direction.
Fig. 4.4 Schematic drawing of the KES-F roughness tester
4.2 Kawabata Evaluation System for Fabrics 55
(a) Photograph of the contactor
20 gf/cm
1 mm/s
10 gf
(b) Front and side view of the contactor
Fig. 4.5 The contactor of the KES-F roughness tester
The contactor of the sensor is a steel piano wire of diameter 0.5 mm. In the stan-
dard version it is bent as shown in Fig. 4.5. The sensor is pressed onto the fabric
with a contact force of 10 gf by a spring with stiffness 25 gf mm−1 and the fabric is
supported by a smooth steel plate. The data recorded by the sensor is digitised and
saved in a measurement file for further processing.
For the computation of the characteristic SMD value by the tester (see Table 4.1)
the analogue signal is first passed through a high-pass filter with transfer function
G( jω) =( jω)2
( jω)2 + 2( jω)ξ ωn + ω2n
, (4.1)
where ωn = 2π rad/s, ξ = 0.6 and ω is the angular frequency. This filter allows
signals of wave length smaller than 1 mm to pass. The SMD value is the mean
deviation of the surface thickness in µm and is calculated as
SMD =1
X
X∫
0
|T (x)− T |dx (4.2)
where X = 2 cm, x is the position of the contactor on the fabric, T (x) is the thickness
of the fabric at position x and T is the mean value of T .
The calculation of the SMD value as done by the KES-F roughness tester has
also been implemented for the digital data to test the consistency of the digital data
and the result from the tester.
4.2.2 KES-F Friction Test
The surface friction is measured with the same device as the surface roughness, but
with a different sensor. It consists of ten piano wires resulting in a larger contact area
56 4 Generation of Virtual Surfaces
20 gf/cm
1 mm/s
50 gf
Fig. 4.6 Front and side view of the KES-F friction contactor
as shown in Fig. 4.6. The contactor is pressed onto the fabric with the compressional
force of 50 gf by a dead weight. The speed at which the fabric is moved under the
sensor is the same as with the roughness measurement.
The MIU and MMD (see Table 4.1) values are calculated by the tester. As with
the SMD value the measured signal first passes the high-pass filter. MIU is the mean
value of the coefficient of friction and MMD is its mean deviation. The definitions
of both values are
MIU =1
X
X∫
0
µ(x)dx (4.3)
MMD =1
X
X∫
0
|µ(x)− µ|dx (4.4)
where X = 2 cm, x is the position of the contactor on the fabric, µ(x) is the frictional
force at position x divided by the compressional force and µ is the mean value of µ .
4.3 Spatial Frequency Analysis
The raw KES-F measurements are not adequate for use with the tactile renderer.
They have to be converted to a two-dimensional property map P(u,v), where the
map P depends on the requirements of the tactile rendering algorithm. One class of
these algorithms translates the spatial frequency content of the surfaces into appro-
priate temporal frequencies. The preparation of the surface measurements for these
algorithms is described in this section.
The surface model is computed in two steps as depicted in Fig. 4.7. At first the
fast Fourier transform (FFT) is used to obtain the spatial frequency content of the
Small Scale
Surface Model
Fast Fourier
Transform
Kawabata
Surface Profile
Measurements
Two−Dimensional
Composition
Fig. 4.7 The sequence of the spatial frequency analysis
4.3 Spatial Frequency Analysis 57
measurements. The resulting spectra are combined to a two-dimensional property
map.
4.3.1 Selection of Appropriate Sections
For the KES-F surface measurement the sensor is moved about 3 cm in forward di-
rection and then about the same length in backward direction. During this procedure
the height in µm for the roughness measurement and the friction force in gf divided
by 50 gf for the friction measurement, respectively, are recorded at a sample rate of
about 1000 Hz. For both measurements also the position on the fabric measured in
cm is recorded. Before any further processing can take place two sections of length
2 cm are selected. The first section is taken from the part of the measurement with
positive sensor speed (from 0.5 to 2.5 cm) and the second section is taken from the
part with negative sensor speed (from 2.5 to 0.5 cm). The selected sections are suf-
ficiently far away from the starting point at 0 cm, the turning point at 3 cm and the
end point at 0 cm to avoid unwanted side effects that might occur in the proximity of
these points, for example due to the change in velocity. Figure 4.8 shows an example
of a selection of appropriate sections.
−20
−15
−10
−5
0
5
10
15
0 10000 20000 30000 40000 50000 60000
Fig. 4.8 Selection of appropriate sections
A linear regression analysis of the position measurements showed that a length
of 2 cm corresponds to about 18,756 samples for all measurements. This result con-
flicts with the sensor speed of 1 mm s−1 specified in the KES-F and the sample rate
of about 1000 Hz. However, as the KES-F tester is regularly calibrated it was as-
sumed that it correctly measured the position and the sample rate was actually only
58 4 Generation of Virtual Surfaces
about 938 Hz. Therefore, a common length of 18,756 samples, corresponding to
2 cm, was used for all sections.
4.3.2 Discrete Fourier Transform
In Chap. 5 the surface models developed in this chapter are used to compute vi-
brations that occur when the finger moves over the fabric. In the approach that is
prepared in this section the spatial frequency spectrum of the surface is directly
transformed into the temporal frequency spectrum of the vibrations. As the mea-
surements are discrete signals their frequency content is computed with the discrete
Fourier transform (DFT). Here, only a very short introduction to the DFT is given.
In [6] and [3] the topic is treated more comprehensively.
The DFT detects harmonic components of a discrete signal fi with i = 0, . . . ,n−1
by comparing the signal with unit-waves of given frequency:
Fk =n−1
∑i=0
fi cos
(
2πki
n
)
− jn−1
∑i=0
fi sin
(
2πki
n
)
. (4.5)
Here j is the imaginary number and therefore the Fk with k = 0, . . . ,n − 1 are
complex numbers. The absolute magnitude |Fk| =√
ℜ(Fk)2 + ℑ(Fk)2 describes
the magnitude of the k-th frequency component while the angular measure ∠Fk =
arctan(
ℑ(Fk)ℜ(Fk)
)
describes its phase.
If the DFT was directly implemented as shown in (4.5), each computation of
the n different frequency components would involve a summation of 2n different
multiplications. So the complexity would be O(n2). However, this complexity can be
reduced: a FFT is an implementation of the DFT that has a complexity of O(n logn).The DFT assumes the signal to be periodic with a period corresponding to the
length of the signal. However, this is rarely the case and discontinuities are intro-
duced at the start and the end of the signal adding erroneous frequency content to
its spectrum. To avoid this problem “windows” are used that are multiplied with the
original signal and cause it to continuously vanish at the start and the end. Let zi
be the discrete signal and wi be a window of the same length, i.e. i = 0, . . . ,n− 1.
The resulting signal is then defined as fi = ziwi. Windows commonly in use are the
trivial rectangular window
wi = 1, (4.6)
the Blackman window
wi = 0.42−0.5cos
(
2πi
n−1
)
+ 0.08cos
(
4πi
n−1
)
(4.7)
and the Hamming window
wi = 0.54−0.46cos
(
2π i
n−1
)
. (4.8)
4.3 Spatial Frequency Analysis 59
0
0.2
0.4
0.6
0.8
1
0 100 200 300 400 500 600 0
0.2
0.4
0.6
0.8
1
0 100 200 300 400 500 600
-1
-0.5
0
0.5
1
0 100 200 300 400 500 600-1
-0.5
0
0.5
1
0 100 200 300 400 500 600
0
50
100
150
200
250
1 10 100 0
20
40
60
80
100
120
140
160
1 10 100
Fig. 4.9 A rectangular (left column) and a Hamming window (right column) and their effect
on a signal in time (second row) and frequency domain (third row)
Figure 4.9 shows the rectangular and the Hamming windows and their effect on a
signal.
In order to obtain a spatial frequency spectrum at different points on the fabric’s
surface a Hamming window of length 4096, which corresponds to 4.37 mm, was
used at different offsets of the measurement. The window was shifted by 938 sam-
ples or about 1 mm and the FFT was used to compute the spectra at the different
offsets. In this way 16 different spectra were computed for a measurement with a
length of 2 cm.
Most fabrics are created such that there is a repetition in their surface pattern
(cf. Sect. 4.1). Therefore, an ideal fabric would result in a periodic measurement
which could be described with only one discrete frequency spectrum. However, due
to irregularities in the yarns and in the weaves the measurements are usually not
60 4 Generation of Virtual Surfaces
perfectly periodic. As these irregularities affect the tactile properties of fabrics it is
important to include them in the surface model. This is the reason why the frequency
spectra are computed at different positions.
4.3.3 Two-Dimensional Composition
Up to now only the one-dimensional KES-F surface measurements were regarded
in this section. However, these represent only a small part of the two-dimensional
surface and are not sufficient to define the whole surface. But with some additional
assumptions a reasonable surface model can be deduced.
Most fabrics are created such that there is a repetition in their surface pattern (cf.
Sect. 4.1). But due to irregularities in the yarns and in the weaves this repetition is
not always clearly recognisable in the measurements.
However, because of the randomness of these irregularities periodicities can often
still be recognised in the measurements in weft and warp direction with lengths teand ta, respectively. It is assumed in the following that te ≤ 16mm and ta ≤ 16mm.
As there are 16 different spectra per measurement, te and ta are defined as
te := maxk∈IN, kte≤16mm
kte, (4.9)
ta := maxk∈IN, kta≤16mm
kta. (4.10)
If there is no recognisable periodicity observable at all, te and ta are both set to
16 mm.
The exact positions of the measurements on the fabric are not known. As the
measurements are orthogonal to each other it is assumed that the measurement in
weft direction lies on the u-axis of the fabric’s coordinate system whereas the warp
measurement lies on the v-axis. For points on the u-axis or v-axis, i.e. if v = 0 or
u = 0, the corresponding spectra can be found with
ie = ⌊u (mod te)⌋ (4.11)
for v = 0 and with
ia = ⌊v (mod ta)⌋ (4.12)
for u = 0, respectively, where ie and ia denote the indices of the corresponding
spectra. Here, the operator ⌊·⌋ is defined as
⌊x⌋ := maxi∈Z, i≤ x
0.001m
i. (4.13)
Figure 4.10 shows a grey-scale image of a fabric’s surface. As indicated by the green
line the surface has a pattern with a certain angle. The surface structure along the
weft and warp direction, respectively, has to be shifted according to that angle, as
depicted by the red lines, to be consistent with each other.
4.4 The Correlation–Restoration Algorithm 61
Fig. 4.10 Example of a
surface having a pattern
with a certain angle
Let α be the angle between the u-axis and the direction of the surface pattern.
Then a point (u,v) on the fabric’s surface can be projected to the axes with
u =u−v
tanα, (4.14)
v =v−u tanα (4.15)
and after substitution of u,v with u, v in (4.11) and (4.12) the indices of the spectra
can be computed with
ie =⌊(
u− v(tanα)−1)
(mod te)⌋
, (4.16)
ia =⌊(v−u tanα) (mod ta)⌋ . (4.17)
These two equations are also depicted in Fig. 4.11.
4.4 The Correlation–Restoration Algorithm
In [9, 10] an algorithm is presented, which also uses the KES-F surface measure-
ments to create a surface model. In contrast to the algorithm presented in the previ-
ous section, this algorithm constructs a two-dimensional height profile of arbitrary
size from the one-dimensional measurements. In [9, 10] friction and sound is re-
garded as well, but for this research only the algorithm for the construction of the
height profile was reimplemented (see [12]). As depicted in Fig. 4.12 the algorithm
reduces the data from the KES-F surface tester to its essentials first and then con-
structs a two-dimensional surface from the remaining data.
62 4 Generation of Virtual Surfaces
Fig. 4.11 A point on the
fabric’s surface is projected
into the measurements
ta
te1.
2.
2.
1.
warp
α
weft
Two−Dimensional
Composition
Small Scale
Surface Model
Kawabata
Surface Profile
Measurements
Data Reduction
Fig. 4.12 Construction of a two-dimensional height profile from the KES-F surface
measurements
4.4.1 Data Reduction
The correlation–restoration algorithm (CRA) is used to decompose a signal into
its strongest harmonic components and a rest, which is modeled as noise. It uses
autocorrelation to emphasise the strongest harmonics and is therefore more stable
in the presence of noise than the usual FFT alone. The algorithm is useful as a
lossy compression technique as the components are computed in the order of their
relevance. The latter holds paticularly for periodic functions. The remaining noise
gives information about the error of the approximation, so maximum error values
can be specified.
Usually only a FFT is used to compute the frequency content of a given signal.
The main idea of the CRA is to compute the autocorrelation of the signal before
applying the FFT. According to [9] the advantage of this procedure is a greater
stability in the presence of noise.
For a discrete signal zi, with i = 0, . . . ,n−1, the autocorrelation is defined as
Rzz(p) =1
n− p
n−p−1
∑i=0
zizi+p (4.18)
4.4 The Correlation–Restoration Algorithm 63
for p = 0, . . . ,n− 1, i.e. the autocorrelation itself is a discrete signal of the same
length. The strongest harmonic component of the original signal is found by apply-
ing a FFT to its autocorrelation.
In this way, however, only the frequency ω j of the harmonic component can be
determined. To also obtain its amplitude and phase the correlation of the original
signal with a unit wave si = cos(2πω jΔxi) with i = 0, . . . ,n−1 is computed. Here,
Δx is the position difference between two consecutive sample points. This unit wave
has the same frequency as the harmonic component found before, its amplitude is
one and its phase is zero. The correlation of two discrete signals is defined as
Rzs(p) =1
n
n−1
∑i=0
zisi+p (4.19)
for p = 0, . . . ,n− 1. It is proven in [9] that the correlation Rzs(p) is a harmonic
function with only one single frequency ω j and that the amplitude A j and the phase
φ j can be obtained with
A j = limn→∞
Rzs(p1peak)− lim
n→∞Rzs(p1
valley), (4.20)
φ j = 2πω jΔxp1peak = 2πω jΔxp1
valley + π . (4.21)
It is of course impossible to have n → ∞ in the numerical implementation of the
algorithm, but if N − p is large the error is negligible.
After amplitude, frequency and phase of a harmonic component have been de-
termined, the component is subtracted from the original signal and the procedure is
repeated until there are not any clearly recognisable harmonic components left in
the signal. The remaining signal is assumed to be noise.
4.4.2 Two-Dimensional Composition
Like before the one-dimensional KES-F surface measurements have to be composed
to create a two-dimensional surface. Similarly to Sect. 4.3.3 the angle α between
the fabric weaving pattern and the weft direction is used in [9, 10] to define a two-
dimensional height profile:
Z(x,y) =ne
∑i=1
Aei cos(2πωe
i (x + y tanα)+ φ ei )
+na
∑i=1
Aai cos(2πωa
i (y− x tanα)+ φai )
(4.22)
where −45◦ < α ≤ 45◦. The height profile defined in this way is a superposition of
cosine waves that linearly propagate in two perpendicular directions. These waves
are defined by their amplitudes Aei and Aa
i , their projected frequencies ωei and ωa
i
and their projected phases φ ei and φa
i .
64 4 Generation of Virtual Surfaces
A frequency component wei appears in the spectrum of the KES-F measurement in
weft direction and multiplied with tanα also in the measurement in warp direction.
Similarly wai appears in the warp measurement and multiplied with tanα also in the
weft measurement. Therefore, these pairs of harmonic components in weft and warp
direction have to be identified and modeled with only one corresponding harmonic
component in (4.22).
It is claimed in [10] that over 70% of all fabrics can be modeled with such a
simple harmonic function. This is probably true for the large-scale structure of the
fabrics but questionable for small structure sizes. However, for the indirect haptic
interaction via a stylus as investigated in [9, 10] the large-scale structure is probably
sufficient to model a fabric. This model is further investigated in Chap. 5.
4.5 Surface Reconstruction from an Optical Surface Scan
An optical surface scan of a fabric sample is easy to obtain and can be used to
reconstruct its height profile. This method called “Shape from Shading” is tested in
[18] and [12] for the application in haptic virtual-reality environments. Results are
presented in [17].
Small Scale
Surface Model
Optical Surface
Scan
Finding Tilings Shape From Shading
Fig. 4.13 Surface Reconstruction from an Optical Surface Scan
A method that reconstructs a height function from a grey-value image is de-
scribed in this section (see Fig. 4.13). The first step in this reconstruction process is
the detection of symmetry in the grey-value image. The term “symmetry” is defined
with regard to a set P ⊆ E 2 (E 2 denoting the Euclidean plane) as a bijective map
α : E 2 → E 2 with α(P) = P . Every symmetry of E 2 is a translation, rotation,
reflection or glide reflection. The symmetries of most fabrics are combinations of
two nonparallel translations. These translations define a basic element called “tile”
or “primitive” that is repeated in the translational directions (see Fig. 4.14). An
algorithm for finding such a tile was implemented based on [7] at the Institute of
Man–Machine Communication at the Leibniz University of Hannover. As the sym-
metries of real surfaces are not perfect in general, stochastic methods are used to
find an idealised tiling.
Since the tile is repeated on the grey value image the average of these repetitions
is used to create a new synthetic image of the surface, that way being cleaned from
noise. This image in turn is used as input for a shape-from-shading algorithm (see
[20]) that computes an appropriate height profile. See Fig. 4.15 for an example. The
tile can be used to describe fabrics of arbitrary size with relatively few data.
4.5 Surface Reconstruction from an Optical Surface Scan 65
Fig. 4.14 Tile with transla-
tions t1 and t2
t1
t 2
Fig. 4.15 Photograph of a fabric and the corresponding height profile
4.5.1 Symmetry Detection
In order to extrapolate the fabric sample beyond its borders the smallest tile has to
be found, which is equivalent to the detection of the isometry group. An algorithm
to obtain this tile using stochastic methods is described in [18] and in more detail in
[16].
To detect symmetrical portions of a texture, its picture has to be segmented into
repeating sections. The average of all these sections is called primitive and can provide
synthetic textures of arbitrary size. Our implementation of the symmetry detection
employs ideas originating from [7] and [13]. The first step of the identification of
symmetries within textures is the creation of the autocorrelation matrix (ACM), which
describes the translations of parts of the texture onto other parts of the same texture.
A statistical feature matrix (SFM) of a discrete grey-value image t under the
operation αα1...αn with an arbitrary feature f t is defined as
66 4 Generation of Virtual Surfaces
SFMf tαα1...αn
(t) = ( f t(t,αα1...αn(t)))αα1 ...αn. (4.23)
In this context an operation αα1...αn is a translation in the Euclidean plane E 2. Let
p,q,r ∈ E 2. Then a translation τ from p to q is defined as
τp,q(r) = r +(q− p). (4.24)
The translation may not leave the region of the grey-value image. Furthermore, to
achieve reasonable results each translation needs to affect at least one fourth of the
original image.
The correlation ρ was utilised as feature f t. It is defined as
ρ(t,t(s)) :=Cov(t,t(s))
√
Var(t)Var(t(s))(4.25)
where Cov() denotes an estimator for the covariance and Var() for the variance.
The ACM is an SFM with translation τp,q as operation αα1...αn and correlation ρas feature f t. From the ACM a grid of primitives can be computed with the following
steps:
1. identification of the maxima of the ACM
2. computation of the main directions of the maxima
3. computation of the distances between the lines of the grid
The correlation describes the amount of congruence of two parts of the texture be-
ing compared. Large elements of the ACM correspond to a large congruence (see
Fig. 4.16). Therefore, the maxima of the ACM are of interest.
Fig. 4.16 The grey value image of a texture (left) and its ACM (right), bright pixels in the
ACM correspond to large values
As not all maximal values are adequate for the computation of the grid, criteria
for the selection of the maxima have to be defined. Because of the large amount
of different textures an automatic determination of these criteria is not possible.
Therefore, a minimum distance R between two adjacent maxima and a threshold S
4.5 Surface Reconstruction from an Optical Surface Scan 67
for the value of the maximum have to be set. Let ACM(i, j) denote the value of the
autocorrelation matrix at position (i, j). It is a maximum if and only if
S < ACM(i, j) (4.26)
and for all ACM(p,q) with i− R2≤ p ≤ i + R
2and j − R
2≤ q ≤ j + R
2and ¬(i =
p∧ j = q)
ACM(p,q) < ACM(i, j). (4.27)
Given the maxima of the ACM the main directions of the grid can be computed.
Therefore the Hough transformation recognising straight lines and circles is used.
Straight lines in the Euclidean plane can be uniquely described by
d = xcosα + ysinα (4.28)
where α is the angle between the normal vector and the x-axis and d is the distance
to the origin.
For the computation of the main directions the angle α runs in small steps from 0
to π . The size of the steps depends on the size of the image. Each point P(x,y) in the
set of the maxima can be seen as point on a straight line with angle α . The distance
d can be computed with (4.28). Points resulting in similar distances are considered
to be on the same line.
The two angles with most matching distances are selected as main directions. We
denote them by q1 and q2. To get the length of the vector with direction q1, all max-
ima of the ACM are projected in direction q2 onto a straight line with direction q1.
The projected position of each maximum can be understood as a normal distribution
with given variance σ . The maximum of the superposition of all these distributions
is taken as optimal distance for the vector with direction q1 (see [16]). The length
of the vector with direction q2 is computed in the same way, with q1 and q2 being
interchanged.
In Fig. 4.17 the intermediate steps are depicted: First the maxima of the ACM are
extracted. Then the main directions are found (here represented by two lines). In the
last step the lengths of the spanning vectors are computed. Here, the length of the
vectors is represented by the shape of the primitive. The primitive is computed by
averaging values from the original picture. In this way the primitive is cleaned from
noise. The picture on the bottom right shows the synthetically generated image of
the texture. It is created by repetition of the primitive.
4.5.2 Shape from Shading
Shape-from-Shading (SFS) algorithms compute a surface profile of an object from
one or more two-dimensional images. Pioneer work in this field was done by Horn
in [8].
68 4 Generation of Virtual Surfaces
Fig. 4.17 Maxima of the ACM, the directions represented by two lines, the primitive and the
synthetically generated image (from top left to bottom right)
The algorithm used here was developed by Tsai and Shah in [20]. It extracts the
information of the height from only one picture. In [23] the algorithm of Tsai and
Shah is recommended for images with little noise and a source of light being close to
the object. These two criteria are fulfilled in our case, since the symmetry detection
cleans the image from noise and the textures were digitised with a flat bed scanner,
so that the source of light was indeed very close to the texture.
4.6 Artificial Surfaces
For testing purposes artificial surfaces were generated. Figure 4.18 shows some ex-
amples, where the grey values represent different heights. These surfaces were used
to validate the functioning of the tactile renderer. Synchrony of the visual and the
tactile display was tested with bars of variable width.
It is shown in [2] that there exists a strong correlation between the fractal di-
mension of a haptic texture and the impression of roughness perceived by tactually
exploring it with a pen-like probe. Therefore, surfaces with different fractal dimen-
sions were used for experiments.
4.6 Artificial Surfaces 69
Fig. 4.18 Artificially generated surfaces with bars (left and middle) and sinusoidal Surfaces
(right)
4.6.1 Brownian Surfaces
Brownian surfaces are randomly generated surfaces with a defined fractal dimen-
sion. Before the construction of these surfaces is described, one of the various ways
to define fractal dimension, the Hausdorff dimension, is presented at first.
Let F ⊆ IRn, F �= /0 with d(x,y) denoting the Euclidean distance in IRn. Then
diam(F) := sup{d(x,y)|x,y ∈ F} is called the diameter of F .
Let δ ∈ IR+ and {Ui}i∈IN be a countable set of subsets of IRn with diam(Ui) ≤ δ .
Then {Ui}i∈IN is called a δ -cover of F if F ⊂⋃∞
i=1 Ui.
Let s ∈ IR+. For every δ ∈ IR+ let
Hs
δ (F) := inf
{
∞
∑i=1
(diam(Ui))s|{Ui}i∈IN δ -cover of F
}
. (4.29)
Then the s-dimensional Hausdorff measure is defined as
Hs := lim
δ→0H
sδ (F). (4.30)
Hausdorff proved the existence of a number DF fulfilling
Hs(F) =
{
∞ for s < DF
0 for s > DF .(4.31)
Now the Hausdorff dimension dimH(F) can be defined as
dimH(F) := inf{s|H s(F) = 0}
=sup{s|H s(F) = ∞}.(4.32)
The Hausdorff dimension dimH(F) has some basic properties:
• If F ⊂ IRn, then dimH(F) ≤ n.
• If F ⊂ G, then dimH(F) ≤ dimH(G).• If F is countable, then dimH(F) = 0.
70 4 Generation of Virtual Surfaces
A Brownian surface XH : IR2 → IR with index H (0 ≤ H ≤ 1) is a stochastic process
with:
1. XH(0,0) = 0.
2. XH(x,y) is continuous.
3. Let σ ,τ ≥ 0. Then the increments XH(x + σ ,y + τ)−XH(x,y) are normally dis-
tributed with expectation 0 and variance (σ2 + τ2)H
=: ρ , i.e.
P(XH(x + σ ,y + τ)−XH(x,y) ≤ z) = (2π)−12 (ρ)−
12
z∫
−∞
exp
(
−r2
2ρ
)
dr.
XH(x,y) itself is also normally distributed with expectation 0 and variance (x2+ y2)H
.
Figure 4.19 shows some examples of Brownian surfaces with different indices.
It is shown in [5] that a Brownian surface XH(x,y) with index H has Hausdorff
and box dimension dimH XH(x,y) = dimB XH(x,y) = 3−H. In the following these
dimensions are shortly refered to as fractal dimension dimXH(x,y).
Fig. 4.19 Brownian surfaces with index 0.1 (red), 0.5 (green) and 0.9 (blue)
A Brownian surface with (x,y) ∈ [0,1]× [0,1] can be constructed with the fol-
lowing algorithm (see [18]):
1. Set XH(0,0) = 0 and randomly draw XH(0,1), XH(1,0) and XH(1,1) from a nor-
mal distribution with expectation 0 and variance 1.
2. Iterate over (i, j) and randomly draw XH(k2−i, l2− j) from a normal distribution
with expectation
4.6 Artificial Surfaces 71
1
4
[
XH
(
(k−1)2−i,(l −1)2− j)
+ XH
(
(k−1)2−i,(l + 1)2− j)
+XH
(
(k + 1)2−i,(l −1)2− j)
+ XH
(
(k + 1)2−i,(l + 1)2− j)
]
and variance
2(−i− j)H.
The variables k, l, i and j have to satisfy
0 ≤ k ≤ 2i, i ≥ 1, k odd and
0 ≤ l ≤ 2 j, j ≥ 1, l odd.
Because of the continuity of XH(x,y) the resulting surface is completely determined.
The correlation between fractal dimension and impression of roughness was in-
vestigated in [2] using a single-point contact with the virtual surface. The PHAN-
ToM device was used to simulate the feel of surfaces with different indices H. The
test subjects could use a pen-like probe to explore the surfaces. To permit compa-
rability the smoothest and the roughest surface were presented and they were given
fixed roughness values of 1 and 20. The test persons were asked to rate randomly
chosen surfaces inside this range.
Figure 4.20 shows some results of this experiment. As one can see there exists
a strong correlation between the fractal dimension of a haptic texture XH(X ,Y ),i.e. dimXH(x,y) = 3 −H and the impression of roughness received by tactually
exploring it.
0
2
4
6
8
10
12
14
16
18
20
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
test person 1test person 2test person 3test person 4test person 5
Fig. 4.20 Results of the experiment with the indices H of the Brownian surfaces on the ab-
scissa and the impression of roughness on the ordinate (with 20 corresponding to the roughest
and 1 to the smoothest surface)
72 4 Generation of Virtual Surfaces
4.7 Conclusion
Two different types of surfaces are investigated in this chapter: artificial surfaces
with well-defined properties described in Sect. 4.6 and real surfaces of 54 different
sample fabrics which are described in Sect. 4.1. The aim of this chapter is the gener-
ation of virtual surfaces that are useful for the tactile simulation systems described
in Chap. 5. In the case of artificial surfaces only the computation of height profiles is
regarded (see Sect. 4.6), but in the case of real surfaces the surface properties have
to be measured and digitised before any further processing can take place. Three
different sources of information are used in this chapter: information on the produc-
tion process of fabrics described in Sect. 4.1, the one-dimensional KES-F surface
measurements as described in Sect. 4.2 and a two-dimensional optical surface scan
obtained with a standard flat bed scanner. From these sources of information either
descriptions of the spatial frequency content at different positions on the surface or
two-dimensional height profiles are computed with different methods. These meth-
ods are described in Sects. 4.3, 4.4 and 4.5 and are indicated with different types of
arrows in Fig. 4.21.
!"#$"%&'()*+),-.-/,#),#
/,)01$2), $/,"%34506
2)" +()2),#
()"% +('"-)
"(#$'$-$"% +('"-)
$,'/(2"#$/,/,&'"7($-"#$/,
!(/-)
#8/01$2), $/,"%/!#$-"%
+('"-)& -",
#8/01$2), $/,"%9)$:9#&!(/'$%)
Fig. 4.21 Four different methods for the generation of surface models indicated by the differ-
ent types of arrows
References 73
The method described in Sect. 4.3 is the only one described in this chapter that
does not create a height profile. Instead it describes a surface in terms of its local
spatial frequency content. In contrast to a height profile this is a description conve-
nient for the tactile renderer, because it computes the tactile signals for the tactile
display in the frequency domain anyway as described in Chap. 5.
Both methods described in Sects. 4.3 and 4.4 analyse the frequency content of the
KES-F surface measurements. While the spectra are adopted without any changes
in Sect. 4.3, the method in Sect. 4.4 attempts to separate the true harmonic com-
ponents from the noise. The latter is necessary because a two-dimensional height
profile is constructed from the two perpendicular one-dimensional KES-F surface
measurements under the assumption that the whole surface can be modeled with
simple harmonic functions. As in Sect. 4.3 additional information on the production
process of fabrics has to be integrated for the two-dimensional composition. There-
fore, both methods can only be applied to regular fabrics or very similar surfaces.
The methods fail with more complicated fabrics like Jacquard fabrics.
The method described in Sect. 4.5 does not require specialised hardware like
the KES-F testers but works with a standard flat bed scanner. Because the mea-
surement is already two-dimensional it does not require the assumptions needed
for the two-dimensional composition carried out by the other methods. However,
it is assumed that a tiling of the regarded surface can be found in order to be able
to clean it from noise and to extrapolate the height profile beyond the borders of
the image taken by the flat bed scanner. Furthermore, the method does only work
with monochrome surfaces, because otherwise the darker colours are recognised as
shades by the shape-from-shading algorithm.
References
[1] Behery, H.M. (ed.): Effect of mechanical and physical properties on fabric hand. Wood-
head Publishing Limited (2005)
[2] Bergmann, M., Herbst, I., von Wieding, R., Wolter, F.E.: Haptical rendering of rough
surfaces using their fractal dimension. In: Proceedings of the First PHANToM Users
Research Symposium, German Cancer Research Center, Heidelberg, Germany, pp. 9–
12 (1999)
[3] Bloomfield, P.: Probability and Statistics, 2nd edn. Fourier Analysis of Time Series.
John Wiley and Sons, Chichester (2000)
[4] Boos, A.D.: Concepts and understanding of fabric hand, ch. 2, pp. 11–44. Woodhead
Publishing Limited (2005)
[5] Falconer, K.J.: Fractal geometry: mathematical foundations and applications. Wiley,
Chichester (1990)
[6] Grover, D., Deller, J.: Digital signal processing and the microcontroller. Prentice-Hall,
Englewood Cliffs (1998)
[7] Handley, C.: The analysis and reconstruction of repetitive textures. In: Computer Graph-
ics International 1998, pp. 273–276. IEEE Computer Society, Los Alamitos (1998)
[8] Horn, B.K.P.: Shape from shading: A method for obtaining the shape of a smooth
opaque object from one view. PhD thesis, Department of Electrical Engineering, Mas-
sachusetts Institute of Technology (1970)
74 4 Generation of Virtual Surfaces
[9] Huang, G.: Feel the “fabric” via the phantom. PhD thesis, University of Pennsylvania
(2002)
[10] Huang, G., Metaxas, D., Govindaraj, M.: Feel the “fabric”: an audio-haptic interface.
In: Proceedings of the 2003 ACM SIGGRAPH/Eurographics Symposium on Computer
animation, Eurographics Association, pp. 52–61 (2003)
[11] Kawabata, S.: The standardization and analysis of hand evaluation. Tech. rep., The Hand
Evaluation and Standardization Commitee, The Textile Machinery Society of Japan.
Osaka Science and Technology Center Bld., 8-4, Utsubo-1-chome, Nishi-ku, OSAKA
550 Japan (1980)
[12] Kobberling, T.: Verschiedene Methoden zur Generierung virtueller taktiler
Oberflachenstrukturen. Master’s thesis, Welfenlab, Universitat Hannover (2006)
[13] Kruse, S.M.: Ein Beitrag zur Segmentierung von Bildszenen im Kontext der objektori-
entierten Bilddatencodierung. Wiss.- und Technik-Verlag, Berlin (1997)
[14] Makinen, M., Meinander, H., Luible, C., Magnenat-Thalmann, N.: Influence of phys-
ical parameters on fabric hand. In: Haptex 2005 – Workshop on Haptic and Tactile
Perception of Deformable Objects, Welfenlab, Universitat Hannover, pp. 8–16 (2005)
[15] Minazio, P.G.: Fast – fabric assurance by simple testing. International Journal of Cloth-
ing Science and Technology 7(2/3), 43–48 (1995)
[16] Peinecke, N.: Symmetriefindung in periodischen Texturen. Student research paper,
Welfenlab, Universitat Hannover (1999)
[17] Peinecke, N., Allerkamp, D., Wolter, F.E.: Generating tactile textures using periodicity
analysis. In: 2007 International Conference on Cyberworlds, pp. 308–313. IEEE Com-
puter Society, Los Alamitos (2007)
[18] Schulze, M.: Von computergraphischen zu haptischen Texturen. Virtual Reality fur den
Entwicklungsbereich Design/Styling in der Automobilindustrie. PhD thesis, Welfenlab,
Universitat Hannover (2005)
[19] Taylor, M.A.: Technology of Textile Properties, 3rd edn. Forbes Publications, London
(1990)
[20] Tsai, P.S., Shah, M.: Shape from shading using linear approximation. Image and Vision
Computing Journal 12(8), 487–498 (1994)
[21] Varheenmaa, M., Meinander, H.: Mechanical properties as a base for haptic sensing of
virtual fabrics. In: Proceedings of the Autex 2007 Conference, Tampere, Finland (2007)
[22] Vaughn, E.A.: An introduction to textile manufactoring — principles, products and pro-
cesses. Tech. rep., School of Textiles, Fiber and Polymer Science, Clemson University,
Clemson, South Carolina, USA (1996)
[23] Zhang, R., Tsai, P.S., Cryer, J.E., Shah, M.: Analysis of shape from shading techniques.
In: IEEE Computer Vision and Pattern Recognition 1994 Conference, pp. 377–384.
IEEE Computer Society, Los Alamitos (1994)
Chapter 5
Tactile Rendering
While moving a fingertip over a fine surface we experience a sensation that gives us
an idea of its properties. A satisfactory simulation of this feeling is still an unsolved
problem, although a lot of different tactile displays have been developed. In most
attempts to display real surfaces only a few features, like e.g. ridges and grooves,
are simulated. A system for the display of real surfaces is presented in [8], but
the computation of the tactile signals is only based on a photograph of the surface
texture and is not further motivated with respect to its physical correctness. In this
chapter rendering strategies based on vibrations, which play an important role in the
tactile exploration of fine surfaces, are described. To produce appropriate excitation
patterns the tactile displays described in Sect. 3.2 are used.
The tactile renderers described in this chapter are all based on the idea of using a
superposition of vibrations at 40 and 320 Hz to stimulate the SA1 and PC channel,
respectively. These frequencies have been chosen because the two tactile channels
are especially sensitive in the respective frequency regions and the fact that one fre-
quency is a multiple of the other is quite handy for the generation of the vibrations.
This approach has the advantage that only amplitude data have to be transmitted to
the electronic component driving the tactile display. But it is not immediately clear
how the amplitudes have to be computed from a given surface model. Several rea-
sonable approaches exist and only some of these could be investigated within this
work.
All renderers have in common that the position of the finger on the surface has
to be tracked, that appropriate commands have to be sent to the tactile display via
USB, that a possibility to select the fabric to be rendered has to be offered and that
also a visual feedback has to be given to the user. Therefore a common framework
was implemented which is described in Sect. 5.1.
When the first experiment was conducted a tactile display was not available yet.
In order to test the experimental methods presented in Sect. 2.1 with the methods for
the generation of virtual surfaces described in Sects. 4.4 and 4.5 a force-feedback
device was used instead. The experimental setup and the results are described in
Sect. 5.2.
D. Allerkamp: Tactile Perception of Textiles in a Virtual-Reality System, COSMOS 10, pp. 75–100.
springerlink.com c© Springer-Verlag Berlin Heidelberg 2010
76 5 Tactile Rendering
The tactile renderer presented in Sect. 5.3 extracts the two amplitudes for the
40 and 320 Hz vibrations from the frequency spectrum of the time-varying height
along the trajectory of the finger on the rendered surface. For each tactile channel
a band-pass filter is used to extract the relevant parts of the frequency spectrum. A
single amplitude is obtained via integration.
The tactile simulation system was integrated with a force-feedback system which
was developed within the HAPTEX project. The involved software components and
their integration as well as the integrated hardware are described in Sect. 5.4. The
results of a subjective evaluation of the system is also described.
5.1 Rendering Framework
A common framework was developed for the different versions of the tactile render-
ing algorithms, which is described in this section. It consists of four threads executed
in parallel as shown in Fig. 5.1. Two threads are responsible for acquiring the input
data needed by the two other threads. The integrity of the data shared between the
threads is guaranteed by using mutex objects.
!"#$%&'($!)"*
(!"&+%, ($!)"*
,)(-.!+($!)"*
("&(%/)!),*)!%, ($!)"*
-.!+'#"&)
#.'%(%.,.01("&(%/)*%'#/"2
34156 78156
#/"&)'10"9!%&'.,1-.!+'#"&)
')('1#.'%(%.,1.0("&(%/)1*%'#/"2
*!"-10"9!%&'
*!"-1("&(%/)*%'#/"2
!),*)!
!),*)!
Fig. 5.1 The four threads of the tactile rendering framework
The network thread listens on a network port for user requests and accordingly
places fabrics on a virtual workspace. In this way an experiment supervisor is able
to set up the workspace for the subject from a remote computer.
The tracking thread continuously receives positional data from a device track-
ing the position of the tactile display. For this work a standard computer mouse, a
5.1 Rendering Framework 77
graphics tablet or a force-feedback device were used as tracking device. Figure 5.2
shows the tactile display connected to a computer mouse used as tracking device.
The whole platform is supported by roller bearings that allow it to be moved on the
table in any direction.
Fig. 5.2 The tactile display connected to a computer mouse used as tracking device
The function implementing the rendering algorithm is constantly called by the
tactile rendering thread every 25 ms, which corresponds to 40 Hz. This function
computes the new amplitudes for the tactile display based on the positional data
from the tracking device and the configuration of the virtual workspace.
The graphics thread shows the workspace and the position of the tactile display
on the subject’s computer screen (see Fig. 5.3). The screen is constantly updated at
25 Hz.
5.1.1 Position Tracking
Different devices were used to track the position of the tactile display. The accuracy
of the position measurements is mainly influenced by the spatial resolution of the
device, usually specified in lines per inch (lpi), and by the sample rate at which the
position is measured.
Apart from the position the tactile rendering algorithms also need to know the
velocity of the display. A Kalman filter not only reduces noise but also can compute
the velocities corresponding to the measured positions (cf. [24]). It is based on the
78 5 Tactile Rendering
Fig. 5.3 Graphical presentation of the workspace and the position of the tactile display pins
(depicted as green boxes)
idea of estimating the state x ∈ IRn of a discrete-time controlled process by predict-
ing the state of the next time step and by correcting the predicted state using the
corresponding measurement of the process as depicted in Fig. 5.4.
!"#$%&'()#*&+#'!,)!-./
0#(12+#"#.)%&'()#
*,-++#,)!-./
Fig. 5.4 Kalman filter as a predictor–corrector method
The discrete-time controlled process mentioned above is described by
xk = Axk−1 + Buk + wk−1. (5.1)
The matrix A ∈ IRn×n relates the state at time step k− 1 to the state at time step k.
For the position tracking the state is chosen to be
x =
⎛
⎝
p
v
a
⎞
⎠ (5.2)
5.1 Rendering Framework 79
where p, v and a denote the position, the velocity and the acceleration, respectively.
The used model assumes the acceleration a to be approximately constant and thus
the matrix A is chosen to be
A :=
⎛
⎝
1 Δ t 12Δ t2
0 1 Δ t
0 0 1
⎞
⎠ (5.3)
where Δ t denotes the time between two time steps. The term Buk describes a driving
force, which is not existent in the model used for the position tracking. The random
variable wk−1 represents the process noise, which is assumed to be white, with a
normal probability distribution, a mean of zero and a covariance matrix Q. The
latter is defined as
Q :=
⎛
⎝
14Δ t4 1
2Δ t3 1
2Δ t2
12Δ t3 Δ t2 Δ t
12Δ t2 Δ t 1
⎞
⎠var(a). (5.4)
In this way the influence of the user on the position of the tactile display is modeled
as noise of the acceleration. The variance of the acceleration has been experimen-
tally estimated as 90 m s−2.
A measurement zk ∈ IRm is related to the state with
zk = Hxk + vk. (5.5)
As only the position is measured H has the form
H :=(
1 0 0)
. (5.6)
The random variable vk represents the measured noise. Like the process noise it
is assumed to be white, with a normal distribution, a mean of zero and covariance
matrix R. For the position tracking R is a 1×1 matrix and has been estimated as
R = (0.001m). (5.7)
In our model (see Fig. 5.5) every contactor of the tactile array is assigned a two-
dimensional device coordinate
(
xi
yi
)
. Let
(
0u(t)0v(t)
)
denote the position of the origin
of the device coordinate system on the fabrics surface (trajectory) and φ(t) the angle
between the x-axis and the u-axis at the time t, then the position of the contactors
on the fabric’s surface can be computed with
(
ui(t)vi(t)
)
=
(
0u(t)0v(t)
)
+
(
cosφ(t) −sinφ(t)sinφ(t) cosφ(t)
)(
xi
yi
)
(5.8)
Using the property map P(u,v) from Chap. 4 we can now define the time dependent
property function Pi(t) := P(ui(t),vi(t)) along the trajectory.
80 5 Tactile Rendering
Fig. 5.5 Tactile contactors
on the fabric’s surface
u
v
x
y
φ
5.2 Experiment with a Force-Feedback Device
In Chap. 4 different methods for the generation of virtual surfaces are presented. In
order to compare these methods using the evaluation methods described in Chap. 2 a
first series of experiments was conducted. The main intention of these experiments
was not to compare the virtual surface models but to test whether the evaluation
methods are useful for such a comparison. The experiments presented in this section
are also described in [9] and in [14].
When this first series of experiments was conducted a tactile display was not yet
available then. However, this was not a real drawback as the intention was to try the
evaluation methods first and not to already evaluate the tactile simulation systems.
It was decided to simulate the surface textures with a force-feedback device as there
already existed some experience in this area (cf. [16] and [13]). However, it should
be noted that the pen-like probe used for the experiments is not a natural tool to
assess the properties of fabrics. Therefore, this series of experiments is unsuitable
for the evaluation of the usefulness of the methods for the generation of virtual
surface models for tactile simulation systems.
5.2.1 Method
Two methods are compared in this section: The method described in Sect. 4.4, here
shortly refered to as the “Kawabata method”, and the method described in Sect. 4.5,
here simply called “optical method”. Both methods create a height profile which is
then used to render the information using a PHANToM Premium 1.0 haptic device
as described in Sect. 3.4.
Since we wanted to find out more about discriminability, we chose a subset of
relatively similar fabrics. Furthermore, for our experiments the haptic properties
of the fabrics had to be visible in order to be reconstructable from an optical scan.
5.2 Experiment with a Force-Feedback Device 81
These criteria resulted in a test set of the cloth samples 8, 9, 18, 22, 23, 28, 29 (see
Annex A).
To compare both methods with each other, and to compare them with real fabrics,
a setup similar to [18] was chosen. In a triangular method three objects were pre-
sented to the subject, with two of them being identical. The subject had to tell, which
one of the three objects was different from the two other ones (cf. Sect. 2.1.3). After
that, the subject had to judge, if it was “easy”, “normal” or “hard” to make a deci-
sion. Since each odd-man-out experiment incorporates two different textiles, each
subject had to answer n(n−1)/2 questions for n samples.
In order to understand which properties of the fabrics are important for the dis-
crimination task an MDS procedure was employed (see Sect. 2.1.6). To gather the
perceptual dimensions subjective distances between the fabrics had to be found. In
Sect. 2.1.5 the computation of subjective distances is explained. Here a little differ-
ent approach was used: each answer was weighted with 1, 0.5 or 0.1 if the test was
considered “easy”, “normal” or “hard”, respectively. If the answer was wrong, i.e.
the subject did not identify the odd sample, the answer received weight 0. Note that
the choice of weights is more or less arbitrary here. Obviously we had to choose
a non-zero value for the “hard” value but this could have been a lesser value than
0.1 also. Aside from that choosing other values results in a slight transformation
of the resulting spaces. But the overall qualitative distances should be kept as long
as the “normal” value is located near the middle between “hard” and “easy”. The
subjective distance is then defined as the sum of all weights. The distances between
all samples build up a complete distance graph which is then embedded into a low-
dimensional space. This projection is of course subject to an error which depends
on the dimension of the embedding space. The dimension is chosen as maximal in a
sense that higher dimensions do not reduce the error significantly (see Sect. 2.1.6).
Twelve subjects (4 female, 8 male) participated in the experiments. They were
aged between 20 and 64 years (on average 32 years). None of the subjects used
a force-feedback device or participated in a similar experiment before. During the
tests the subjects could neither see the fabric nor hear the noises generated by the
pen or the device, so that visual and acoustic impressions could not interfere with
the tactile sensation.
5.2.2 Results
In the first experiment the subjects could freely move a real pen (resembling the
interface pen of the force-feedback device) over a sample of real fabric. Figure 5.6
shows an MDS plot of the subjective distances gathered from the first experiment. In
contrast to Fig. 5.6 Fig. 5.7 shows the fabrics in an objective parameter space (with
values taken from [9] and [12]). As one can see the relative positions of the fabrics
(emphasised by triangles) are quite similar in both figures, with the exception of
Sample 29. Thus the first dimension of Fig. 5.6 probably resembles the roughness
of the fabric as also stated in [9]. The second dimension probably corresponds to the
compressional resilience.
82 5 Tactile Rendering
-6
-4
-2
0
2
4
6
-6 -4 -2 0 2 4 6 8
dim
en
sio
n 2
dimension 1
8
9
18
22
23
28
29
Fig. 5.6 MDS plot of the first experiment
In the second experiment the subject could explore a virtual sample of fabric
generated from the optical method (cf. Sect. 4.5) rendered by the force-feedback
device. Just like in the first experiment the size of the virtual samples was 2cm2.
Figure 5.8 shows an MDS plot of the subjective distances gathered from the second
experiment. Note that the MDS method has reduced the data set to one dimension
here. Similar to one of the dimensions in the first experiment the one dimension that
is left after reducing gives a good approximation of the textures roughness. Just like
in the real world set up the samples 8 and 28 are very close to each other, resembling
the real impression. Since compressional resilience was not simulated by the system
this dimension does not appear as in the first experiment.
The third experiment was identical to the second, except that the haptic sensation
was generated using the Kawabata measurements (cf. Sect. 4.4). Figure 5.9 shows
an MDS plot of the subjective distances gathered from the third experiment. Just like
the optical method the Kawabata method gives a good discrimination of the fabrics’
roughness. There is only one exception concerning the samples 8 and 28, that were
very similar in the first two experiments but are placed at distant positions by the
Kawabata measurements. This could possibly be caused by interpolation problems
of the very fine grained surfaces of these fabrics.
During these odd-man-out tests the subjects were asked if they found the dis-
criminability of the samples “hard”, “normal” or “easy”. In Fig. 5.10 one can see
the accumulated results of these questions. One observes that it seems to be much
easier and more accurate to tell the different fabrics apart in the real setup. The two
5.2 Experiment with a Force-Feedback Device 83
30
40
50
60
70
80
90
100
2 4 6 8 10 12 14 16
co
mp
ressio
na
l re
sili
en
ce
roughness
8
28
18
22
9
23
29
Fig. 5.7 Roughness and compressional resilience of the fabrics
-8 -6 -4 -2 0 2 4 6
89 182223 2829
Fig. 5.8 MDS plot of the second experiment
-8 -6 -4 -2 0 2 4 6 8
89 182223 2829
Fig. 5.9 MDS plot of the third experiment
virtual methods perform relatively similar with no significant drawbacks on both
sides.
In Fig. 5.11 the errors of the subjects are shown for each fabric. It can be seen
that the optical method performs comparable to the Kawabata method with the only,
already known, exception of Sample 8. One can see in the figure that the errors for
the samples 18, 28 and 8 are very low for the experiment with real fabrics. This is
probably due to their relatively high distance in the second perceptual dimension
mentioned earlier in this section (see Fig. 5.6).
84 5 Tactile Rendering
0
10
20
30
40
50
60
wrong hard normal easy
an
sw
ers
in
%real
opticalKawabata
Fig. 5.10 Classification of discriminability
0
5
10
15
20
25
30
29 23 9 22 18 28 8
err
or
in %
realoptical
Kawabata
Fig. 5.11 Errors of the subjects
5.3 Vibrotactile Rendering 85
5.3 Vibrotactile Rendering
Basically, a tactile renderer needs two ingredients to work properly: a computer
representation of a fabric and the trajectory of the fingertip on the fabric’s surface.
As described in Sect. 2.3 vibrotaction plays an important role in the perception of
fine surface textures. Therefore, we compute the vibrations occurring in the finger-
tip while moving along the trajectory. These are decomposed into only two basic
frequencies intended to directly stimulate the “Pacinian” (PC) and “non-Pacinian”
(SA1) receptors. We describe this process later in this section. The used stimulator
hardware is described in Sect. 3.2. This approach has already been presented in [1].
5.3.1 Computation of Resulting Vibrations
For each contactor we need to compute the vibrations occurring due to the move-
ment of the fingertip over the surface. This computation depends on the model gen-
erated in Chap. 4. If we represent the surface as a height function as in Sects. 4.4
and 4.5, Pi(t) describes the height under the i-th contactor as a function of time. As a
start we act on the simple assumption that the vibration at the contactor position on
the finger directly corresponds to Pi(t). We want to express a vibration as a function
F assigning to each frequency ω a corresponding amplitude F(ω), which is done
by Fourier transforming Pi(t). Since we need to render the device signals for the
next 25 ms, this time segment becomes our window for the Fourier transform. We
smoothed the boundaries of this segment with a Blackman window.
If the surface is represented as spatial frequency spectra Fe(k) and Fa(k) in weft
and warp direction respectively as computed in Sect. 4.3 a corresponding temporal
frequency spectrum F(ω) has to be computed. When the finger moves over a surface
a spatial frequency k and a temporal frequency ω are related by
ω = vk (5.9)
where v is the velocity of the finger on the surface. This results in the following
conversion
Fe(ω) = Fe
(
ω
ve
)
Fa(ω) = Fa
(
ω
va
) (5.10)
where ve and va denote the component of the velocity in weft and warp direction,
respectively. Assuming that the vibrations in the two orthogonal directions can be
simply combined by addition we obtain
F(ω) = Fe(ω)+ Fa(ω) = Fe
(
ω
ve
)
+ Fa
(
ω
va
)
. (5.11)
86 5 Tactile Rendering
5.3.2 Decomposition of Vibrations into Base Frequencies
The previous steps result in a frequency spectrum F(ω), which now has to be de-
composed into amplitudes at only two base frequencies that will then be reproduced
on the tactile stimulator. We use 40 and 320 Hz as base frequencies. The function
mapping the spectrum to these two amplitudes has to ensure that the resulting tac-
tile sensation is very similar to the sensation which would result from reproducing
the whole spectrum on the tactile stimulator. This approach is analogous to the ap-
proach used for multicolour video displays which produce almost arbitrary colours
as mixture of only three fundamental colours.
Similar to the colour model of computer graphics (e.g. see [4]) we assume the
existence of two functions H40 and H320 such that the two amplitudes for 40 and
320 Hz can be computed as
a40 =
∫
H40(ω)F(ω)dω
a320 =
∫
H320(ω)F(ω)dω .(5.12)
This hypothesis still has to be verified with appropriate experiments. Nevertheless,
in the rest of this section we will show the reasonableness of this approach and we
will even derive first approximations of the functions H40 and H320.
With Ip(F) and Inp(F) we denote the stimulation intensity caused by a vibra-
tion with spectrum function F at the Pacinian and non-Pacinian channel, respec-
tively. Note that these intensity functions are subjective and thus hard to define. Our
suggestion for an approximation of these functions is
Ip(F) =
∫
1
Tp(ω)F(ω)dω
Inp(F) =
∫
1
Tnp(ω)F(ω)dω
(5.13)
Since the channels react depending on the frequency (e.g. the Pacinian channel
is most sensitive around 300 Hz while the non-Pacinian channel is most sensitive
around 50 Hz) we use 1Tp(ω)
and 1Tnp(ω)
as frequency dependent damping factors.
The assignment of a linear factor to each frequency seems to be a reasonable ap-
proximation suggested by some curves of equal sensation magnitude plotted as a
function of stimulus frequency (see [20]). Furthermore, the approach of using an
integral is supported by [19, 1], where mixtures of 40 Hz and 320 Hz sinewaves re-
sulting in a constant subjective intensity have been measured. The results of these
measurements can be explained with a linear model as taken as a basis for (5.13).
In [6] thresholds of the tactile channels as functions of stimulus frequency were
measured (see Fig. 5.12 for approximating curves). These thresholds refer to the
smallest perceivable amplitude of a stimulus with only a single frequency. Along
these measured curves the intensity is considered to be constant, making these
5.3 Vibrotactile Rendering 87
0.01
0.1
1
10
100
1000
10000
1 10 100
Dis
pla
cem
ent
Frequency (Hz)
T_pT_np
Fig. 5.12 The functions Tp and Tnp
functions suitable as frequency dependent damping functions Tp(ω) and Tnp(ω)that normalise the intensity to 1 at threshold level.
Let
Sω,α(ω) =
{
α for ω = ω
0 else(5.14)
denote a simple sinusoidal vibration with frequency ω and amplitude α . Our goal
is to replace a given vibration F by two vibrations S40,a40and S320,a320
such that
Ip(F) = Ip(S40,a40)+Ip(S320,a320
)
Inp(F) = Inp(S40,a40)+Inp(S320,a320
)(5.15)
These equations are equal to a two-dimensional linear equation system with (5.13):
∫
F(ω)
Tp(ω)dω =
a40
Tp(40)+
a320
Tp(320)∫
F(ω)
Tnp(ω)dω =
a40
Tnp(40)+
a320
Tnp(320)
(5.16)
The solutions of this system have the form
a40 = B11
∫
F(ω)
Tp(ω)dω + B12
∫
F(ω)
Tnp(ω)dω
a320 = B21
∫
F(ω)
Tp(ω)dω + B22
∫
F(ω)
Tnp(ω)dω
(5.17)
88 5 Tactile Rendering
Setting
H40(ω) :=B11
Tp(ω)+
B12
Tnp(ω)
H320(ω) :=B21
Tp(ω)+
B22
Tnp(ω)
(5.18)
we obtain (5.12). H40 and H320 are depicted in Fig. 5.13.
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
0 100 200 300 400 500 600
Frequency (Hz)
H_40H_320
Fig. 5.13 The functions H40 and H320
5.3.3 First Results with Brownian Surfaces
We used Brownian surfaces (see Sect. 4.6) to test our vibrotactile rendering strat-
egy. Since the system was not yet fully operable for active exploration of a surface in
real time, we used a passive presentation, i.e. stimuli were presented to the station-
ary fingertip computed for the case of the virtual surface moving over the fingertip
at a constant speed of 8 cm s−1. As expected from the single-point-contact experi-
ment, different Brownian surfaces could be distinguished on the basis of roughness
corresponding to the fractal dimension. However, there was little or no sensation of
movement over the fingertip, presumably because individual surface features were
not sufficiently identifiable to allow them to be tracked as they moved over the array.
In the case of active exploration envisaged for the fully operable system, the sensa-
tion of movement would be available from kinaesthetic and visual cues, and so the
results would presumably be more convincing.
5.3 Vibrotactile Rendering 89
5.3.4 Further Evaluation
The vibrotactile simulation system was evaluated using the method already de-
scribed and tested in Sect. 5.2. In order to allow for the comparability of the results
the same set of fabric samples 8, 9, 18, 22, 23, 28, 29 (see Annex A) was used. Again
the subjects had to tell which one of three presented fabrics was different from the
two other ones (cf. Sect. 2.1.3) and had to judge whether it was “easy”, “normal” or
“hard” to make a decision. Each subject performed all comparisons twice in random
order.
Two female and six male subjects aged between 22 and 32 years participated in
the experiment. All subjects used the tactile display with the right hand although
one subject stated to be left-handed. Two subjects used the tactile display before,
the other subjects had no experience with such systems. The simulation system did
not display the visual appearance of the fabrics and subjects wore ear protectors. So
visual and acoustic impressions could not interfere with the tactile sensation.
In the prior experiment (see Sect. 5.2) each answer was weighted with 1, 0.5 or
0.1 if the test was considered “easy”, “normal” or “hard”, respectively. If the answer
was wrong, i.e. the subject did not identify the odd sample, the answer received
weight 0. The subjective distance was then defined as the sum of all weights. This
time subjects were asked to rate the weights for “normal” and “hard” where the
weight for “easy” was predefined as 1 and a weight of 0 was considered to represent
no distinguishablity. Interestingly almost all subjects chose weights very similar to
the rather arbitrarily chosen weights of the prior experiments. The mean values of
the answers are 0.51 for “normal” and 0.14 for “hard”. Therefore, for the sake of
simplicity weights were chosen as in the prior experiments.
The subjective distances between all samples build up a complete distance graph
which is then embedded into a low-dimensional space (see Sect. 2.1.6). Figure 5.14
shows such an MDS plot of the subjective distances gathered from the experiment.
As the underlying surface model of the tactile simulation system is mainly based
on the roughness of the fabrics, the KES-F geometrical roughness values in weft and
warp direction (cf. Sect. 4.2) have been chosen to position the fabrics in an objective
parameter space. Figure 5.15 shows this space. In contrast to the prior experiments
the relative positions of the fabrics are not very similar in Fig. 5.14 and Fig. 5.15.
However, some similarities are still recognisable as depicted by the connecting lines
in both figures.
The visible deformation of the MDS plot might be caused by the design of the
experiment. While the small differences between fabrics were rated by the subjects
in a relatively detailed way this was not the case for the obvious differences which
virtually always received a weight of 1. Therefore, the maximum subjective distance
between fabrics was limited and the distance between very different fabrics was
underestimated.
During these odd-man-out tests the subjects were asked if they found the dis-
criminability of the samples “hard”, “normal” or “easy”. In Fig. 5.16 one can
see the results of these questions. As can be seen the answers of the subjects are
90 5 Tactile Rendering
-8
-6
-4
-2
0
2
4
6
8
-12 -10 -8 -6 -4 -2 0 2 4 6
dim
en
sio
n 2
dimension 1
8
9
18
22
23
28
29
Fig. 5.14 MDS plot of the experiment
2
4
6
8
10
12
14
16
18
0 2 4 6 8 10 12 14
rou
gh
ne
ss w
arp
roughness weft
8
9
18
22
23
28
29
Fig. 5.15 Roughness of the fabrics in weft and warp direction
5.4 Integration with a Force-Feedback Device 91
0
10
20
30
40
50
60
wrong hard normal easy
an
sw
ers
in
%
Fig. 5.16 Classification of discriminability
comparable to the results from the prior experiment with real fabrics and better than
with the force-feedback simulation of the fabrics (cf. Fig. 5.10).
In Fig. 5.17 the errors of the subjects are shown for each fabric. These are com-
parable to the errors of the prior experiments shown in Fig. 5.11.
5.4 Integration with a Force-Feedback Device
The simulation systems regarded until now only provide tactile feedback. With these
systems only planar, not curved, surfaces can be simulated. Furthermore, mechani-
cal properties like elasticity cannot be presented. In order to create a more compre-
hensive simulation system the combination of a tactile display with a force-feedback
device providing the large-scale forces is necessary.
Such an integrated system was developed within the EU-funded HAPTEX project
(cf. [15, 5, 11]). It simulates fabrics that are square shaped with a side length of
20 cm. The user can select a fabric from the property database which is then sim-
ulated hanging from a stand. The user can touch the virtual fabric with thumb and
index finger. The fabric can be squeezed, stretched, rubbed and lifted.
The simulated scene is quite challenging, because the physical simulation of the
fabric, the force-feedback renderer and the tactile renderer have to be integrated
in a single simulation system. In order to reduce complexity the above mentioned
subsystems were implemented and tested separately before they were integrated.
92 5 Tactile Rendering
0
5
10
15
20
25
30
29 23 9 22 18 28 8
err
or
in %
Fig. 5.17 Errors of the subjects
amplitudes
Textile
Simulation
Thread
Local
Simulation
Thread
Force
Feedback
Thread
running at60Hz running at
1kHz
running at
1kHz
bordering geometry LocalGeometry
force generation ForceExtrapolation
Force
Feedback
Device
forces positions
MotionEstimation
movements
Visual
Display
global mesh adaption
FingertipModel
contact
formulation
deformrefine
Tactile
Rendering
Thread
Tactile
Array
contact area
force distribution, velocity
Fig. 5.18 The structure of the integrated software system from [3]
Figure 5.18 shows the structure of the integrated system. In one thread the whole
fabric is simulated by a simulation software module kindly provided by the MI-
RALab at the University of Geneva. As this simulation module is relatively accurate
it does not provide an update rate sufficiently large for the integration with a force-
feedback device. Therefore, another simulation module has been introduced as in-
termediate layer between the simulation of the whole fabric and the force-feedback
5.4 Integration with a Force-Feedback Device 93
thread. This module reaches the necessary update rate but is less accurate and only
simulates the part of the fabric that surrounds the contact point between the user’s
virtual finger and the fabric. This part of the fabric is called “local geometry” in
the following. The force renderer manages the whole communication between the
physical simulation and the force-feedback devices. Incoming local geometries are
processed and resulting forces are transferred to the force-feedback devices. Geom-
etry updates are sent back to the physical simulation. The tactile renderer controls
the tactile devices. The amplitudes generated from the fingertips’ movement over
the surface are sent to the drive electronics at an update rate of 40 Hz. To enable
the simulation system to work without the tactile renderer a loose coupling between
both was chosen: the tactile renderer is implemented as a separate application, data
is exchanged via inter-process communication (IPC).
5.4.1 Physical Simulation and Haptic Rendering
The simulation software used for the integrated system is described in [22, 21, 23].
Here only a very simplified model is presented in order to explain the fundamentals
of the simulation of fabrics. Although the mass is quite continuously distributed
on a fabric its simulation model consists of a finite set of discrete points called
“particles” each having a position xi and a mass mi. These points are kept together
by internal forces acting on adjacent masses that reflect the real tensile behaviour of
the simulated fabric. Although these forces are usually non-linear they are simply
depicted as springs in Fig. 5.19. In addition to the internal forces external forces, like
for example gravity, act on the particles. Usually also damping forces are included
in the model. The sum of all these forces acting on a single particle is denoted by
Fig. 5.19 A simple model
for the simulation of fabrics1m
m
m m m
m m m
m m m m
mmmm
2 3 4
5 6 7 8
9 10 11 12
13 14 15 16
94 5 Tactile Rendering
Fi(x1,x2, . . . , x1, x2, . . .). With Newton’s second law of motion we obtain the system
of differential equations
mixi = Fi(x1,x2, . . . , x1, x2, . . .). (5.19)
This system of ordinary differential equations can be numerically solved with stan-
dard algorithms.
The physical simulation used within the HAPTEX project considers stretch/
tensile, shear, bending and damping forces. Furthermore, the contact simulation
considers friction forces. As for the tactile rendering the relevant physical data have
been acquired with the KES-F system.
In Fig. 5.20 the haptic loop is depicted. During the interaction with the fabric the
position of the user’s fingers and the shape of the fabric change. Both is considered
by the contact model that computes the forces affecting the finger and the fabric.
Here a problem becomes evident: to create a convincing illusion and to avoid stabil-
ity problems the system needs to react to the user’s motion within one millisecond
which is not possible involving the slower simulation of the fabric. However, note
that the single loop depicted in Fig. 5.20 can also be seen as two loops: one between
the user and the contact model and the other between the simulation and the con-
tact model. This view led to the solution described in [3]. A dual-layer approach is
employed there to allow the two loops to run at different speeds.
Fig. 5.20 The haptic loop
with the physical simulation
forc
eposition
shape forc
e
contact model
FF Device
textile simulation
5.4.2 Integrated Interface Hardware
The GRAB device is a force-feedback device consisting of two identical and in-
dependent robotic manipulators (see Fig. 5.21), each having the base link fixed to
the desktop and the end effector (contact part) attached to the palmar surface of
the user’s thumb or index fingertip. Each manipulator measures the absolute posi-
tion and orientation of the contact part. These are used by the haptic renderer to
compute an appropriate force. The manipulator is able to generate this force on the
contact part within a workspace of 400 mm in width, 300 mm in height and 200 mm
5.4 Integration with a Force-Feedback Device 95
Fig. 5.21 The final demonstrator of the HAPTEX Project
in depth. The workspaces of both manipulators overlap. Force errors are limited in
a range of about 10 g (0.1 N). The device can exert forces up to 20 N. The GRAB
device used for the HAPTEX project is described in [2]. It was kindly provided by
the PERCRO laboratory at the Scuola Superiore Sant’Anna in Pisa.
The end effectors of the GRAB device have been replaced with the tactile dis-
plays described in Sect. 3.2. These displays are designed such that the size of the
mechanism between the finger and thumb is minimised. Still, the spacing between
the digits caused by the tactile mechanisms is about 10 mm. As the maximum stiff-
ness between the finger and thumb is limited because of instabilities in the force
control loop the force-feedback renderer enforces an additional separation of 6 mm
to avoid collisions between the mechanisms (cf. [3] and [17]).
5.4.3 Software Integration
The tactile renderer needs the following contact data to work properly:
• contact area between the fingertip and the fabric (in local coordinates of the
fabric)
• velocity of the fingertip relative to the fabric (in local coordinates of the fabric)
• distribution of normal forces inside the contact area
96 5 Tactile Rendering
For the computation of the force feedback the haptic renderer already requires
among others the contact point, the velocity of the fingertip relative to the fabric and
the normal force (cf. [22, 3]). Instead of a single contact point the tactile renderer
requires a contact area to decide which contactor pins have to be activated. Also
a single normal force is not sufficient for a convincing tactile simulation. Rather a
distribution of the normal force over the whole contact area is needed by the tactile
renderer.
For each contactor pin the geometry of the tactile actuator determines the point on
the fingertip where the stimulation caused by that pin occurs. These points have been
included in the model of the fingertip. Employing the contact model the position (in
local coordinates of the fabric) and the normal force at each point is computed. The
velocity is computed by applying a discrete differential 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 we need
to resample that data. To avoid aliasing the data is passed through an antialiasing
filter that suppresses frequency components above 20 Hz (cf. [7]).
In the stand-alone version of the tactile renderer the fabric was assumed to be
planar. Furthermore the finger was assumed to always being in contact with the
plane of the fabric. However, in the integrated case the fabric may have folds and
wrinkles resulting in a more complex contact geometry. Due to the contact pressure
the fingertip deforms which also influences the shape of the contact area. As a con-
sequence not all pins of the tactile actuator should be activated, only those inside
the contact area. The shape of each finger’s contact area is provided by the force-
feedback renderer. Note that a contact area may not exist, i.e. in that case the finger
does not touch the fabric.
While moving a fingertip over a rough surface we experience a tactile sensation.
Its intensity also depends on the force with which the finger is pressed against the
fabric. For relatively small forces a linear dependency between the force and the
sensation seems to be reasonable. Therefore the amplitudes computed by the tactile
renderer are multiplied by kFF before they are transmitted to the tactile actuator. F
denotes the force between the finger and the fabric and kF is a constant relating the
force to the intensity. Previous to the integration there was no possibility to assess
the contact pressure between the fingertip and the fabric. Therefore a constant force
normal to the fabric was assumed implicitly by the tactile renderer.
5.4.4 Evaluation of the Integrated System
The goal of the HAPTEX project was the development of a system for the realistic
haptic simulation of real fabrics. In order to evaluate the system four subjects were
asked to rate the tensile, roughness, friction and bending properties of five fabrics.
Two subjects (A, B) rated the properties of real fabrics whereas two other subjects
(C, D) rated the same properties of the corresponding virtual fabrics simulated by the
system. For each property a single manipulation for its assessment was defined (cf.
[17] and [10]). The interaction with the fabrics was restricted to these manipulations
for the experiments. As using the simulation system is very difficult without vision
5.4 Integration with a Force-Feedback Device 97
Fig. 5.22 Reference samples 1 (top row) and 5 (bottom row) for tensile (left) and bending
(right)
a coarse wireframe model of the fabric was shown as a compromise. Consequently
the tensile and bending tests were performed with vision on the real textiles as well.
The fabric samples 33, 34, 36, 37 and 39 were used for the experiment (see
Annex A). The subjects were asked to rate the fabrics’ properties in the range 1
to 5. For each property two other fabrics, called reference 1 and reference 5, were
provided to define the scale. Figure 5.22 shows the reference samples for the tensile
and bending properties. Samples 39 and 10 were used as reference 1 and reference
5 for tensile stiffness, 24 and 10 for roughness, 24 and 16 for friction, and 21 and
32 for bending. Each fabric was presented twice in a sequence of 10 samples to
subjects C and D.
The data obtained by the experiment were analysed by calculating the correla-
tion of two successive runs of the same subject (“repeatability”), the correlation of
the results of C and D (“consistency”) and the correlation of the mean ratings of A
98 5 Tactile Rendering
Table 5.1 The correlation results of the evaluation (cf. [17])
repeatability consistency realism
C D
tensile 0.99 0.97 0.91 0.96
roughness 0.91 0.95 0.93 0.90
friction 0.66 0.67 0.76 0.93
bending 0.30 0.07 0.61 0.00
and D with the mean ratings of C and D (“realism”). Table 5.1 summarises the re-
sults of the evaluation. Tensile and roughness gave the best results with a correlation
coefficient of at least 0.9 for repeatability, consistency and realism. Friction had a
realism coefficient of 0.93 although its repeatability correlation coefficient was only
average with about 0.66. This relatively high realism coefficient might be caused by
the use of the mean ratings as averaging has the effect of reducing the amount of
random variation. Unfortunately bending showed no real correlation, giving coeffi-
cients as low as 0.07 for successive runs with the same subject. The reason for the
latter was that with the haptic feedback being unable to represent the small forces
resulting from bending, the user had to rely on what one could see to rate the bend-
ing property. However, rating the bending property by visual assessment was most
likely impaired by the coarseness of the mesh not giving enough hints to the user.
For complete results see [17]. The results of the evaluation are also presented in [3].
5.5 Conclusion
Only few works exist on the tactile simulation of real surface textures. The only
system intended for the rendering of arbitrary surface textures that is comparable to
the systems presented in this chapter is described in [8]. However, this system is not
aimed at a realistic simulation of the surfaces.
In this chapter several experiments are described that were conducted to explore
the problem from different points of views. Therefore, a common software frame-
work was developed which manages a virtual workspace where the virtual fabrics
are presented, which processes the date from the tracking device and which trans-
mits the amplitude data to the tactile display.
The experiment described in Sect. 5.2 was conducted as a pilot study. Its purpose
was to test the experimental methods from Sect. 2.1. Instead of a tactile display a
force-feedback device with a pen-like end effector was used to display the virtual
surfaces since there already existed some experience in the simulation of surface
textures with force-feedback devices (cf. [16] and [13]). The results of the exper-
iment suggest that subjects mainly regarded compressional resilience and surface
roughness when distinguishing real fabrics with a pen-like probe. The results fur-
ther suggest that from these two properties only the roughness of the surfaces was
References 99
properly produced by the simulation system. Although the system accurately ren-
dered the height profile of the two-dimensional surfaces, i.e. the space of all possi-
ble surfaces having a lot of dimensions, the perceptual space seems to consist of the
one-dimensional roughness only.
As the pen-like probe used for the experiments is not a natural tool to assess
the properties of fabrics it is an interesting question whether the perceptual space
has more dimensions when a tactile display is used. In Sect. 5.3 a tactile rendering
system based on vibrations is presented. It is important to note that the rendering
algorithm proposed in the section is only one of many possible rendering schemes.
However, the renderer is based on current knowledge on human tactile perception
and only a few reasonable assumptions had to be presumed. An evaluation of the
system showed that subjects were able to distinguish relatively similar fabrics when
rendered by the system. Statements of the subjects suggest that their decision was
mainly based on the roughness of the fabrics in weft and warp direction. However,
the latter assumption was not clearly supported by the results of the evaluation.
In Sect. 5.4 the integration of the tactile simulation system with a force-feedback
system is presented. In such an integrated system many different subsystems have
to be integrated: a physical simulation of the fabric, a force-feedback renderer with
an appropriate force-feedback device and a tactile renderer with a tactile display. As
shown by a subjective evaluation the integration effort resulted in a system that is
capable of not only simulating the surface roughness of fabrics but also their tensile
and friction properties.
References
[1] Allerkamp, D., Bottcher, G., Wolter, F.E., Brady, A.C., Qu, J., Summers, I.R.: A vibro-
tactile approach to tactile rendering. The Visual Computer 23(2), 97–108 (2007)
[2] Bergamasco, M., Salsedo, F., Fontana, M., Tarri, F., Avizzano, C.A., Frisoli, A., Ruf-
faldi, E., Marcheschi, S.: High performance haptic device for force rendering in textile
exploration. The Visual Computer 23(4), 247–256 (2007)
[3] Bottcher, G., Allerkamp, D., Glockner, D., Wolter, F.E.: Haptic two-finger contact with
textiles. The Visual Computer 24(10), 911–922 (2008)
[4] Foley, J.D., van Dam, A., Feiner, S.K., Hughes, J.F.: Computer Graphics: Principles and
Practice, 2nd edn. Addison-Wesley, Reading (1990)
[5] Fontana, M., Marcheschi, S., Tarri, F., Salsedo, F., Bergamasco, M., Allerkamp, D.,
Bottcher, G., Wolter, F.E., Brady, A.C., Qu, J., Summers, I.R.: Integrating force and
tactile rendering into a single VR system. In: 2007 International Conference on Cyber-
worlds, pp. 277–284. IEEE Computer Society, Los Alamitos (2007)
[6] Gescheider, G.A., Bolanowski, S.J., Pope, J.V., Verillo, R.T.: A four-channel analysis
of the tactile sensitivity on the fingertip: frequency selectivity, spatial summation, and
temporal summation. Somatosensory & Motor Research 19(2), 114–124 (2002)
[7] Grover, D., Deller, J.: Digital signal processing and the microcontroller. Prentice-Hall,
Englewood Cliffs (1998)
[8] Ikei, Y., Wakamatsu, K., Fukuda, S.: Vibratory tactile display of image-based textures.
IEEE Computer Graphics and Applications 17(6), 53–61 (1997)
100 5 Tactile Rendering
[9] Kobberling, T.: Verschiedene Methoden zur Generierung virtueller taktiler
Oberflachenstrukturen. Master’s thesis, Welfenlab, Universitat Hannover (2006)
[10] Luible, C., Varheenmaa, M., Magnenat-Thalmann, N., Meinander, H.: Subjective fab-
ric evaluation. In: 2007 International Conference on Cyberworlds, pp. 285–291. IEEE
Computer Society, Los Alamitos (2007)
[11] Magnenat-Thalmann, N., Volino, P., Bonanni, U., Summers, I.R., Bergamasco, M.,
Salsedo, F., Wolter, F.E.: From physics-based simulation to the touching of textiles:
The HAPTEX project. International Journal of Virtual Reality 6(3), 35–44 (2007)
[12] Makinen, M., Meinander, H., Luible, C., Magnenat-Thalmann, N.: Influence of phys-
ical parameters on fabric hand. In: Haptex 2005 – Workshop on Haptic and Tactile
Perception of Deformable Objects, Welfenlab, Universitat Hannover, pp. 8–16 (2005)
[13] Meyer-Spradow, J.: Ein mathematisches Echtzeit-Modell taktiler Merkmale von
Oberflachen. Diploma thesis, Welfenlab, Universitat Hannover (2005)
[14] Peinecke, N., Allerkamp, D., Wolter, F.E.: Generating tactile textures using periodicity
analysis. In: 2007 International Conference on Cyberworlds, pp. 308–313. IEEE Com-
puter Society, Los Alamitos (2007)
[15] Salsedo, F., Fontana, M., Tarri, F., Ruffaldi, E., Bergamasco, M., Magnenat-Thalmann,
N., Volino, P., Bonanni, U., Brady, A.C., Summers, I.R., Qu, J., Allerkamp, D., Bottcher,
G., Wolter, F.E., Makinen, M., Meinander, H.: Architectural design of the HAPTEX
system. In: Haptex 2005 – Workshop on Haptic and Tactile Perception of Deformable
Objects, Welfenlab, Universitat Hannover, pp. 1–7 (2005)
[16] Schulze, M.: Von computergraphischen zu haptischen Texturen. Virtual Reality fur den
Entwicklungsbereich Design/Styling in der Automobilindustrie. PhD thesis, Welfenlab,
Universitat Hannover (2005)
[17] Summers, I.R.: Final demonstrator and final integration report. Tech. rep., HAPTEX –
HAPtic sensing of virtual TEXtiles (2008),
http://haptex.miralab.unige.ch
[18] Summers, I.R., Irwin, R.J.: Tactile discrimination of paper. In: Haptex 2005 – Work-
shop on Haptic and Tactile Perception of Deformable Objects, Welfenlab, Universitat
Hannover, pp. 43–49 (2005)
[19] Summers, I.R., Brady, A.C., Syed, M.K., Chanter, C.M.: Design of array stimulators
for synthetic tactile sensations. In: Proceedings of the First Joint Eurohaptics Confer-
ence and Symposium on Haptic Interfaces for Virtual Environment and Teleoperator
Systems, pp. 586–587. IEEE Computer Society, Los Alamitos (2005)
[20] Verrillo, R.T., Fraioli, A.J., Smith, R.L.: Sensation magnitude of vibrotactile stimuli.
Perception & Psychophysics 33, 379–387 (1969)
[21] Volino, P., Magnenat-Thalmann, N.: Accurate anisotropic bending stiffness on particle
grids. In: 2007 International Conference on Cyberworlds, pp. 300–307. IEEE Computer
Society, Los Alamitos (2007)
[22] Volino, P., Davy, P., Bonanni, U., Magnenat-Thalmann, N., Bottcher, G., Allerkamp,
D., Wolter, F.E.: From measured physical parameters to the haptic feeling of fabric.
In: Haptex 2005 – Workshop on Haptic and Tactile Perception of Deformable Objects,
Welfenlab, Universitat Hannover, pp. 17–29 (2005)
[23] Volino, P., Davy, P., Bonanni, U., Luible, C., Magnenat-Thalmann, N., Makinen, M.,
Meinander, H.: From measured physical parameters to the haptic feeling of fabric. The
Visual Computer 23(2), 133–142 (2007)
[24] Welch, G., Bishop, G.: An introduction to the Kalman filter. Tech. rep., University of
North Carolina, Chapel Hill (2001)
Chapter 6
Summary and Outlook
The research described in this monograph is the first attempt to develop a realistic
tactile simulation system for real existing surfaces. Although there exist many differ-
ent tactile displays developed by various research groups, most systems are designed
to convey information to visually impaired persons (e.g. [3] or [2]). Other systems
are limited to the simulation of simple geometrical surface features like ridges and
grooves (e.g. [4]). However, unlike the work at hand none of these systems aims at
a comprehensive and realistic tactile simulation of real existing surfaces.
This work originated from the EU-funded HAPTEX project and thus inherited
some of its prerequisites: the class of the surfaces to be simulated by the system
is limited to fabrics and only one type of tactile displays was available for the ex-
periments. The former is only a minor restriction as there exists a large variety of
different fabrics—certainly enough to argue that the system is in principle able to
simulate arbitrary surfaces. The latter prerequisite is a necessity since the resources
of the research project were limited and the development of a tactile display is rel-
atively cost-intensive. Therefore, it is reasonable to conduct an in-depth analysis of
only a single display. Moreover, the display used within this work follows an in-
teresting approach. Instead of simply reproducing the geometry of a surface it is
intended for the direct stimulation of two different kinds of tactile receptors with
superimposed vibrations of two different frequencies.
Before a tactile rendering system could be implemented as presented in Chap. 5
a number of preparations had to be made which are described in Chap. 2, Chap. 3
and Chap. 4. In order to systematically tackle the problem it was necessary to gain
insight into many different disciplines: psychophysics, neurophysiology, electronic
circuit design, textile engineering and digital signal processing. For each discipline
a short introduction is given in this monograph or at least a reference to an introduc-
tory book.
In Chap. 2 three important preparations are made. Firstly, it presents the current
knowledge on human tactile perception which is used later to motivate the tactile
rendering algorithms presented in this book. Therefore, it is necessary to introduce
psychophysics and neurophysiology first. Secondly, a perceptional model of tactile
simulation is developed that builds a theoretical frame for this work. And thirdly,
D. Allerkamp: Tactile Perception of Textiles in a Virtual-Reality System, COSMOS 10, pp. 101–103.
springerlink.com c© Springer-Verlag Berlin Heidelberg 2010
102 6 Summary and Outlook
psychophysical methods for the evaluation of the tactile simulation systems are pre-
sented. These are used later to analyse the developed simulation systems.
The tactile displays used within this work are presented in Chap. 3. In order to
use these displays the development of an electronic component was necessary. This
component translates the digital amplitude data from the tactile rendering algorithm
into appropriate electronic signals which are then transformed into mechanical vi-
brations by the display. The design of this electronic component is described in the
chapter. As explained above the choice of the tactile display was inherited from the
HAPTEX project. Nevertheless, the chapter still gives an overview on other existing
displays. Finally, force-feedback systems are introduced as these are also used later.
A tactile rendering system transforms measured surface data into appropriate tac-
tile signals. For a realistic simulation of active touch the system needs to react to the
movement of the user’s finger, i.e. the system needs to compute the amplitudes for the
tactile display in real-time according to the movement of the finger. However, there
is no need to use the measured data directly. The surface data can be prepared pre-
viously for the real-time part. The latter generation of virtual surfaces is the topic of
Chap. 4. Three different methods for the generation of virtual surfaces are described.
One of these uses an optical surface scan as input data whereas both other methods
rely on the KES-F surface measurements. As all these methods more or less make
use of surface properties specific to fabrics an introduction to the production of fab-
rics is also given. Furthermore, the selection and measurement of sample fabrics is
described. A comprehensive list of all sample fabrics can be found in Annex A.
After all these preparations a tactile simulation system can be developed which
is described in Chap. 5. For different tactile rendering algorithms a common frame-
work was created which tracks the position of the tactile display, i.e. the position of
the user’s finger on the virtual fabrics. The tactile rendering algorithm presented in
the chapter is based on vibrations occurring while moving a fingertip over a rough
surface. The system attempts to reproduce these vibrations with a tactile display
such that a similar sensation is created. An evaluation of the system showed that
subjects were able to distinguish relatively similar fabrics when displayed by the
tactile simulator. The tactile simulation system was combined with a force-feedback
system. An evaluation showed that subjects recognised the tensile, roughness and
friction properties of the sample fabrics.
Psychophysical experiments described in [1] suggest that the tactile properties of
surfaces build a three-dimensional perceptual space (cf. Sect. 2.3). The number of
perceptual dimensions is further reduced when the surface is rendered by a tactile
simulation system. The dimensionality of these perceptual spaces stands in contrast
to the dimensionality of the space of all possible virtual surfaces which is much
higher. This provokes the question whether the virtual surface models presented in
Chap. 4 are still too sophisticated. However, it should be noted that the experiments
conducted do not reveal the realism of the simulation. It might be possible that only
a few perceptual dimensions are sufficient to distinguish the properties of a fabric
but many more are needed for a realistic simulation.
How the realism of the simulation could be evaluated is not answered in this work
and could be the subject of further research. The experiments showed that subjects
References 103
were able to rate the roughness of the sample fabrics but this does not mean rough-
ness was realistically rendered. It is possible that the stimuli causing a real sensation
of roughness were substituted by other stimuli which gave enough information on
the roughness but did not feel realistic. However, asking subjects to rate the real-
ism of the simulation would probably rather reveal the subjects’ sympathy for the
experimenter than anything else. So this is also not a solution. In this work the real
situation and its simulation were only compared on the level of sensory impressions
(cf. Sect. 2.3.4). A comparison on the level of neural activity or of physical stimuli
could be more suitable for the evaluation of realism. Measuring the neural activity
of single afferent neurons could probably deliver interesting insight into classes of
stimuli resulting in equivalent sensations. However, the invasive methods needed in
this case raise ethical problems.
The displays used within this work follow an interesting approach. Recreating
the vibrations occuring while moving a fingertip over a rough surface might result
in an appropriate stimulation of the PC sensors, however, it is questionable whether
the same holds for the SA1 sensors. The developed system only creates vibrations
when the fingertip is moving. If the finger stops its movement the vibrations vanish.
This behaviour does not reflect reality as surface features can still be distinguished
in the real case without movement of the fingertip. Representing this static structure
of the surface by vibrations is not a good solution as the vibrations do not create
a static sensation but are always recognised as vibrations. A display recreating the
height profile under the fingertip intended for the stimulation of the SA1 sensors
together with vibrating elements for the stimulation of the PC sensors might create
a more realistic sensation. The development of such a display could be an interesting
research project for the future.
In this monograph a system is described that simulates the tactile properties of
fabrics. Although it is not perfectly realistic the sensations caused are relatively
convincing and subjects were able to distinguish different fabrics and correctly rate
their roughness. However, this is just a small step towards a completely convincing
simulation of direct touch and a lot of research needs still to be done in this area.
References
[1] Hollins, M., Faldowski, R., Rao, S., Young, F.: Perceptual dimensions of tactile surface
texture: A multidimensional scaling analysis. Perception & Psychophysics 54(6), 697–
705 (1993)
[2] Petit, G., Dufresne, A., Levesque, V., Hayward, V., Trudeau, N.: Refreshable tactile
graphics applied to schoolbook illustrations for students with visual impairment. In: Pro-
ceedings of the 10th international ACM SIGACCESS conference on Computers and ac-
cessibility, pp. 89–96. ACM, New York (2008)
[3] Vidal-Verdu, F., Hafez, M.: Graphical tactile displays for visually-impaired people. IEEE
Transactions on Neural Systems and Rehabilitation Engineering 15(1), 119–130 (2007)
[4] Wang, Q., Hayward, V.: Tactile synthesis and perceptual inverse problems seen from the
view point of contact mechanics. ACM Transactions on Applied Perception 5(2), 1–19
(2008)
Appendix A
Fabrics
In this appendix all fabrics that were available for this work are described (cf. [1, 2]).
The samples have been kindly provided by SmartWearLab at the Tampere Univer-
sity of Technology. The abbreviations of the contents are summarised in Table A.1.
An image of the fabric has been included true to scale where available. The selection
of the samples and their measurements is described in Chap. 4.
Table A.1 Abbreviations of the fabrics’ fibre contents.
AF Other fibres CV Viscose PAN Polyacryl
CA Acetat EL Elastane PES Polyester
CLY Lyocell JU Jute PU Polyurethane
CMD Modal LI Linen SE Silk
CO Cotton PA Polyamide WO Wool
106
AF
abrics
# Image Description Content Structure Weight Thickness
% g/m2 mm
1 Denim 100 CO twill 380 1.6
2 Shirt cotton 100 CO combined twill 120 0.61
3 Cord 100 CO velveteen 330 1.76
4 Linen 100 LI plain weave 250 1.09
5 Gabardine 100 WO twill 175 0.55
Continued on next page.
AF
abrics
107
Continued from previous page.
# Image Description Content Structure Weight Thickness
% g/m2 mm
6 Crepe 100 WO plain weave 145 0.93
7 Silk 100 SE plain weave 15 0.1
8 Natural silk
(bourette)
100 SE plain weave 150 0.8
9 Wild silk (dupion) 100 SE plain weave 80 0.44
10 Jute 100 JU plain weave 300 1.44
Continued on next page.
108
AF
abrics
Continued from previous page.
# Image Description Content Structure Weight Thickness
% g/m2 mm
11 Flannel 80/20 WO/PES twill 290 1.53
12 Denim 62/35/3 PES/CO/EL twill 275 1.13
13 Plaid 35/35/30 PES/AF/WO twill 270 1.14
14 Tweed 66/14/10/10
AF/WO/PES/CMD
combined twill 270 3.9
15 Velvet 92/8 CO/CMD velvet 300 1.88
Continued on next page.
AF
abrics
109
Continued from previous page.
# Image Description Content Structure Weight Thickness
% g/m2 mm
16 Lurex knit 70/30 PES/PA held stitch knit 215 2.94
17 Crepe-jersey 85/15 PES/EL rib knit 135 0.73
18 Motorcyclist. coated 72/28 PA/PU plain weave 90 0.39
19 Woven easy care 65/35 PES/CO twill 180 0.57
20 Warp knitted velour 90/10 PA/EL warp knit velour 235 1.56
Continued on next page.
110
AF
abrics
Continued from previous page.
# Image Description Content Structure Weight Thickness
% g/m2 mm
21 Weft knitted plain 98/2 CLY/EL single jersey 172 1.21
22 Taffeta 100 CA plain weave 125 0.33
23 Crepe 100 PES plain weave 85 0.25
24 Satin 100 PES satin 125 0.3
25 Felt 100 PES nonwoven 155 1.25
Continued on next page.
AF
abrics
111
Continued from previous page.
# Image Description Content Structure Weight Thickness
% g/m2 mm
26 Organza 100 PES plain weave 25 0.16
27 Fleece 100 PES weft knit 250 3.99
28 Woven upholstery 100 PES woven jacquard 600 2.38
29 Woven leisure 100 PES plain weave 90 0.2
30 Tulle 100 PA warp knitted tulle 10 0.3
Continued on next page.
112
AF
abrics
Continued from previous page.
# Image Description Content Structure Weight Thickness
% g/m2 mm
31 Warp knitted tricot-
satin
100 PA warp knitted tricot-
satin
100 0.4
32 Leather 100 leather 815 1.68
33 Men’s woven suit
fabric (plain)
60/38/3 WO/PES/EL plain weave 195 0.57
34 Men’s woven suit
fabric (herringbone)
100 WO broken twill 232 0.83
35 Men’s woven over-
coat fabric
80/20 WO/PA modified plain weave 324 2.64
Continued on next page.
AF
abrics
113
Continued from previous page.
# Image Description Content Structure Weight Thickness
% g/m2 mm
36 Men’s woven over-
coat fabric (twill)
59/25/11/5
CO/PAN/WO/PES
twill 460 3.23
37 Woven outdoor
leisurewear fabric
100 PES plain weave 98 0.26
38 Weft knitted jersey
fabric
48/48/4 CO/CMD/EL single jersey 208 1.09
39 Weft knitted terry
fabric
55/45 CV/PES weft knit terry 288 1.69
40 Warp knitted jersey-
based fabric
100 PES warp knit jersey-
based
154 0.51
Continued on next page.
114
AF
abrics
Continued from previous page.
# Image Description Content Structure Weight Thickness
% g/m2 mm
41 Warp knitted mesh
fabric
100 PES warp knit mesh 128 0.51
42 Brushed warp knitted
fabric
100 PES brushed warp knit 215 0.98
43 warp knit for car
seats
89/11 PES/EL pile warp knit 278 1.22
44 warp knit for car
seats
100 PES flattened pile warp
knit
168 1.1
45 warp knit for car
seats
100 PES raised pile warp knit 147 1.6
Continued on next page.
AF
abrics
115
Continued from previous page.
# Image Description Content Structure Weight Thickness
% g/m2 mm
46 warp knit for car
seats
100 PA raised pile warp knit 98 1.55
47 warp knit for car
seats lamin
Face fabric: 100 PES (sam-
ple 42) Foam: E-35 PF (35
kg/m3 polyether) Backing:
100 PES (sample 52)
laminated brushed
warp knit
386 4.09
48 warp knit for car
seats lamin
Face fabric: 89 PES/ 11 EL
(sample 43) Foam: E-35 PF
(35 kg/m3 polyether) Back-
ing: 100 PA (sample 53)
laminated pile warp
knit with
438 4.19
49 warp knit for car
seats lamin
Face fabric: 100 PES (sam-
ple 44) Foam: E-35 PF (35
kg/m3 polyether). Backing:
100 PES (sample 52)
laminated flattened
pile warp knit
335 4.11
50 warp knit for car
seats lamin
Face fabric: 100 PES (sam-
ple 45) Foam: L30PS (30
kg/m3 polyester) Backing:
100 PES
laminated raised pile
warp knit
332 3.54
Continued on next page.
116
AF
abrics
Continued from previous page.
# Image Description Content Structure Weight Thickness
% g/m2 mm
51 warp knit for car
seats lamin
Face fabric: 100 PA (sam-
ple 46) Foam: E-25 TX (25
kg/m3 polyether)
laminated raised pile
warp knit
170 4.03
52 backing 100 PES warp knit 53 0.42
53 backing 100 PA interlock 121 0.81
54 warp knit for car
seats lamin
Face fabric: 89 PES/ 11 EL
(sample 43. different colour)
Foam: L30PS (30 kg/m3
polyester)
laminated pile warp
knit
287 3.35
References 117
References
[1] Makinen, M., Meinander, H., Luible, C., Magnenat-Thalmann, N.: Influence of physical
parameters on fabric hand. In: Haptex 2005 – Workshop on Haptic and Tactile Perception
of Deformable Objects, Welfenlab, Universitat Hannover, pp. 8–16 (2005)
[2] Varheenmaa, M., Meinander, H.: Mechanical properties as a base for haptic sensing of
virtual fabrics. In: Proceedings of the Autex 2007 Conference, Tampere, Finland (2007)
Index
absolute threshold, 6
afferent neuron, 17
alternative hypothesis, 10
amplitude, 58
artificial surface, 68
autocorrelation, 62
autocorrelation matrix (ACM), 66
Bernoulli experiment, 9
bimorph, 29
binomial distribution, 9
Brownian surface, 70
confidence interval, 10
confidence level, 10
consistency, 98
correlation, 63, 66
correlation–restoration algorithm (CRA), 62
Craven’s distances, 12
data bus, 37
digital–analogue converter (DAC), 39
discrete Fourier transform (DFT), 58
discrimination methods, 8
drive electronics, 33
dual-layer approach, 94
efferent neuron, 17
end effector, 43
evaluation, 96
fabric, 50
fabric hand, 53
fabric samples, 52
false alarm rate, 11
fast Fourier transform (FFT), 58
fibre, 50
firmware, 35
force-feedback device, 42
frequency response, 30
geometrical roughness (SMD), 55
gram-force (gf), 53
HAPTEX project, 91
haptic loop, 43
haptic rendering, 94
Hausdorff dimension, 69
hit rate, 11
hypothesis testing, 10
intraneuron, 17
inverse piezoelectric effect, 29
just noticeable difference (jnd), 6
Kalman filter, 77
Kawabata evaluation system for fabrics
(KES-F), 53
KES-F friction tester, 55
KES-F roughness tester, 54
knitting, 51
low-pass RC filter, 40
maximum likelihood estimator, 9
mechanoreceptive afferent, 18
Pacinian (PC), 19
rapidly adapting (RA), 19
120 Index
slowly adapting type 1 (SA1), 19
slowly adapting type 2 (SA2), 19
mechanoreceptor, 18
model of human perception, 20
multidimensional scaling (MDS), 12
neural activity, 20
neuron, 15
non-woven, 51
null hypothesis, 10
odd-man-out forced-choice procedure, 9
operational amplifier (OPA), 41
phase, 58
physical simulation, 93
piezoelectric bimorph, 29
piezoelectric effect, 29
primitive, 64
psychometric function, 6
psychophysical scale, 7
psychophysics, 5
Q-factor, 31
realism, 98
repeatability, 98
response bias, 8
sensations, 20
shape from shading (SFS), 67
signal-detection theory, 10
spatial summation, 17
spring–damper model, 43
statistical feature matrix (SFM), 66
Stevens’ Power Law, 8
stimuli, 20
symbolic perception, 20
symmetry, 64
tactile colours, 86
tactile display, 25
tactile perception, 18
tactile rendering, 77
temporal summation, 17
tile, 64
tracking device, 77
translation, 66
triangular method, 9
USB controller, 34
variable voltage supply, 39
vibrotactile rendering, 85
virtual workspace, 76
warp, 50
weaving, 50
Weber’s law, 6
Weber-Fechner law, 7
weft, 50
window, 58
Blackman, 58
Hamming, 58
yarn, 50
Cognitive Systems Monographs
Edited by R. Dillmann, Y. Nakamura, S. Schaal and D. Vernon
Vol. 1: Arena, P.; Patanè, L. (Eds.):
Spatial Temporal Patterns for
Action-Oriented Perception
in Roving Robots
425 p. 2009 [978-3-540-88463-7]
Vol. 2: Ivancevic, T.T.; Jovanovic, B.;
Djukic, S.; Djukic, M.; Markovic, S.:
Complex Sports Biodynamics
326 p. 2009 [978-3-540-89970-9]
Vol. 3: Magnani, L.:
Abductive Cognition
534 p. 2009 [978-3-642-03630-9]
Vol. 4: Azad, P.:
Visual Perception for Manipulation
and Imitation in Humanoid Robots
270 p. 2009 [978-3-642-04228-7]
Vol. 5: de de Aguiar, E.:
Animation and Performance Capture
Using Digitized Models
168 p. 2009 [978-3-642-10315-5]
Vol. 6: Ritter, H.; Sagerer, G.;
Dillmann, R.; Buss, M.:
Human Centered Robot Systems
216 p. 2009 [978-3-642-10402-2]
Vol. 7: Levi, P.; Kernbach, S. (Eds.):
Symbiotic Multi-Robot Organisms
467 p. 2010 [978-3-642-11691-9]
Vol. 8: Christensen, H.I.;
Kruijff, G.-J.M.; Wyatt, J.L. (Eds.):
Cognitive Systems
491 p. 2010 [978-3-642-11693-3]
Vol. 9: Hamann, H.:
Space-Time Continuous Models of Swarm
Robotic Systems
147 p. 2010 [978-3-642-13376-3]
Vol. 10: Allerkamp, D.:
Tactile Perception of Textiles in a
Virtual-Reality System
120 p. 2010 [978-3-642-13973-4]