Design and use of a biomimetic tactile microvibration sensor with human-like sensitivity and its application in texture discrimination using Bayesian exploration

PDF

Download

Open document

Flip pages

Copy asset link

Request this asset

Request accessible transcript

Transcript (if available)

Content DESIGN
AND
USE
OF
A
BIOMIMETIC
TACTILE
MICROVIBRATION
SENSOR

WITH
HUMAN-‐LIKE
SENSITIVITY
AND
ITS
APPLICATION
IN
TEXTURE

DISCRIMINATION
USING
BAYESIAN
EXPLORATION

by

Jeremy
A.
Fishel

A
Dissertation
Presented
to
the

FACULTY
OF
THE
USC
GRADUATE
SCHOOL

UNIVERSITY
OF
SOUTHERN
CALIFORNIA

In
Partial
Fulfillment
of
the

Requirements
for
the
Degree

DOCTOR
OF
PHILOSOPHY

(BIOMEDICAL
ENGINEERING)

August
2012

Copyright
2012

Jeremy
A.
Fishel

ii

The
tactual
properties
of
our
surroundings
do
not
chatter
at
us
like
their
colors;

they
remain
mute
until
we
make
them
speak...

Eye
movements
do
not
create
color
the
way
finger
movements
create
touch

– David
Katz,
The
World
of
Touch
(1925)

iii

To
my
family.

iv

Acknowledgements

Without
the
help
of
the
following
people
my
thesis,
professional
development,

and
sanity
would
not
exist.
These
people
have
been
a
part
of
my
life
both

professionally
and
personally.
They
were
always
there
to
offer
support
and
insight

when
things
became
difficult.
Thank
you
for
supporting
me.

First
and
foremost
I
would
not
have
developed
into
the
scientist
and
researcher
I

am
today
without
the
help
of
my
advisor
and
mentor
Dr.
Loeb.
I
have
been
very

fortunate
to
have
access
to
your
knowledge
and
creativity.
You’ve
had
the
patience

to
guide
me
through
the
early
phases
of
my
studies
to
the
completion
of
my
degree.
I

still
remember
writing
my
first
paper
with
you,
which
was
returned
to
me
with

more
words
in
red
ink
than
I
had
originally
written
in
the
first
place.
It
was
quite

obvious
that
you
could
have
rewritten
entire
manuscripts
in
less
time
than
it
took
to

carefully
(and
thoroughly!)
make
these
edits,
but
you
invested
the
time
to
teach
me,

thank
you!
Hopefully
it
will
now
be
a
thing
of
the
past,
writing
inverted
sentences.

Thank
you
for
your
wisdom,
patience
and
many
creative
discussions
over
the
years.

I
hope
you
have
enjoyed
the
time
as
much
as
I
have.

I
have
also
been
fortunate
to
have
an
outstanding
advisory
committee
to
guide

me
along
the
way.
Dr.
Valero-‐Cuevas,
you
have
been
very
helpful
and
insightful.
I

recall
visiting
your
class
once
where
you
were
teaching
your
students
on
a

v

chalkboard.
You
had
made
the
comment
of
how
amazing
it
was
that
intelligence

could
be
communicated
by
“writing
on
one
rock
with
another
rock.”
That
thought

has
always
stayed
with
me
and
has
helped
me
appreciate
simplicity.
Dr.
Schaal,
it

has
been
an
honor
having
you
on
this
committee
with
your
knowledge
of
robotics

and
many
creative
thoughts.
Dr.
Yen,
you
have
always
been
a
great
person
to
turn
to

when
I
needed
help
and
new
ideas
on
signal
processing.
I
would
really
like
to
extend

the
most
sincere
gratitude
to
you
all
for
mentoring
me
through
the
years
and

providing
feedback
to
make
this
dissertation
the
best
it
could
be.

I
would
also
like
to
thank
the
other
professors
of
the
Biomedical
Engineering
and

other
departments
at
USC
that
I’ve
had
the
chance
to
learn
from
and
work
with:
Dr.

Weiland,
Dr.
D’Argenio,
Dr.
Mel,
Dr.
Meng,
Dr.
Khoo
and
the
rest
of
the
staff.
Last
but

not
least,
Mischal
Disanta,
the
world’s
best
graduate
coordinator,
thank
you
for
all
of

your
help
over
the
years!

I’ve
had
a
rather
unusual
graduate
career.
In
2008,
my
colleagues
and
I
started
a

spinoff
company
called
SynTouch,
which
has
commercialized
the
BioTac
technology.

A
substantial
debt
is
owed
to
my
co-‐workers
at
SynTouch,
without
you
there
would

be
no
BioTac,
no
funding,
and
no
thesis.
Matthew
Borzage,
your
desire
and
ability
to

do
a
job
right
has
helped
to
make
this
company
a
success.
I
do
not
know
where
we

would
be
without
you.
Gary
Lin,
you
have
done
an
amazing
job
on
the
design
and

improvement
of
the
electronics
and
software
of
the
BioTac.
It
has
always
been
a

pleasure
to
work
on
projects
with
you
(now
if
you
too
would
only
graduate
soon,

there
would
be
more
time
for
research
and
development).
Raymond
Peck,
your

vi

expertise
and
knowledge
have
been
invaluable
to
my
training
and
to
this
team.
We

would
not
be
the
same
without
you,
nor
would
we
have
any
BioTacs
to
do
research

with.
I
would
also
like
to
thank
our
past
team
and
newer
team
members
whom

we’ve
been
lucky
to
have
at
SynTouch:
David
Groves,
Tomonori
Yamamoto,
Sagar

Kale,
Blaine
Matulevitch
and
Nicholas
Wettels.

I
would
also
like
to
thank
my
friends
at
the
Medical
Device
and
Development

Facility.
The
laboratory
is
an
evolving
place
where
people
come
and
go
from
all

around
the
world.
I’ve
been
fortunate
to
have
crossed
paths
in
this
laboratory
with

the
most
intelligent
and
diverse
group
of
people
I’ve
ever
met.
When
I
first
arrived,
I

would
have
never
survived
without
the
help
of
the
senior
students
and
post
docs:

Nicholas
Wettels,
Rahul
Kaliki,
Giby
Rapheal,
Marcus
Hauschild,
Dan
Song,
Veronica

Santos
and
Rahman
Davoodi.
As
time
progressed,
new
students
came
along
into
the

PhD
programs,
whom
I
have
gotten
to
know
quite
well
over
the
years.
To
my
good

friend
George
Tsianos,
it
has
always
been
interesting
discussing
our
different

projects
and
watching
your
research
progress
over
the
years.
I
wish
you
the
best
of

luck
with
your
upcoming
defense
and
your
future
career.
Zhe
Su,
it
has
been
exciting

to
work
with
you
on
your
projects
with
the
BioTac
and
I
wish
you
the
best
of
luck
in

medical
school
(robotic
surgery
is
where
it
is
at!).
Jared
Goodner,
I’ve
been
very

impressed
with
your
additions
to
the
spinal
like
regulator,
keep
up
the
good
work

and
best
of
luck
with
your
doctorate!
Shanie
Liyanagamage,
thank
you
again
for

your
thoughtful
feedback
and
edits,
I
wish
you
the
best
of
luck
continuing
this

research!
Also
many
thanks
to
the
students
whom
I
had
the
pleasure
to
supervise:

vii

Blaine
Matulevitch,
Mandy
Lai,
Roman
Sandler,
Mehdi
Roman
and
Dipayon
Roy.

Thank
you
for
keeping
things
fun
and
helping
to
move
the
research
forward.
To
the

many
new
students
in
the
lab
whom
I
have
not
been
able
to
get
to
know
as
well
after

moving
to
SynTouch:
I
wish
you
all
the
best
of
luck,
you
are
in
good
hands
with
Dr.

Loeb.

Most
importantly
of
all
I
would
like
to
thank
my
friends
and
family.
I
am
proud
to

be
the
second
Dr.
Fishel
in
this
family
and
am
thankful
I
had
the
first
one
there
to

help
me
along
the
way.
To
my
grandparents
on
both
sides
of
the
family,
I
have
been

so
lucky
to
have
you
in
my
life
and
I
know
you
would
have
supported
me
in
anything

I’d
chosen.
When
times
were
difficult
you
were
always
there.
Thank
you
for
always

making
me
feel
special.
To
my
aunts,
uncles,
and
cousins,
you
are
an
amazing
family

I
am
very
happy
to
be
a
part
of
it.
To
my
friends,
thank
you
for
always
being

available
to
take
a
break
when
I
desperately
needed
one
from
the
research.

An
extra
special
thanks
to
my
parents
is
necessary.
Thank
you
for
being
such

great
role
models
and
helping
me
when
I
needed
it
the
most.
I
know
I
was
not
the

easiest
child
to
raise
and
much
of
the
gray
hair
you
have
is
my
fault
(although
karma

seems
to
be
getting
even
with
me).
You’ve
been
so
supportive
of
me
my
entire
life

and
were
always
there
to
help
when
I
needed
it.
I
would
not
be
the
person
I
am

today
without
your
love
and
support.
I
love
you
both!

To
my
brother
Jimmy,
you
are
the
best
brother
I
could
have
ever
asked
for.

We’ve
had
some
great
times
over
the
course
of
our
childhood,
and
even
more
so
as

viii

adults.
Even
though
you’re
on
the
other
side
of
the
globe,
you’ve
always
been
close.

We
miss
you,
come
visit
more!

Emily,
my
love.
You
have
been
the
best
thing
that
has
ever
happened
to
me.
I
am

truly
lucky
to
have
found
you.
I
would
not
be
the
person
I
am
today
without
you
in

my
life.
You’ve
been
there
to
listen
to
my
crazy
thoughts
and
ideas,
trying
to

understand
them
when
even
I
haven’t
had
a
grasp
on
them
yet.
You
took
an
interest

in
what
I
did
and
now
probably
know
more
about
the
sense
of
touch
than
anyone
in

the
field
of
criminal
justice
would
ever
need
to
know.
You
took
care
of
me
when

research
got
busy,
and
were
always
there
when
I
needed
to
take
a
break.
You
are
my

soul
mate
and
I
love
you!

P.S.
I
told
you
I’d
graduate
before
our
wedding!

ix

Table
of
Contents

Acknowledgements
.................................................................................................................................
iv

Table
of
Contents
......................................................................................................................................
ix

List
of
Tables
..............................................................................................................................................
xii

List
of
Figures
...........................................................................................................................................
xiii

Abstract
........................................................................................................................................................
xv

Chapter
1:
Introduction
..........................................................................................................................
1

Motivation
.............................................................................................................................................
1

Contributions
.......................................................................................................................................
4

Thesis
Overview
.................................................................................................................................
4

Chapter
2:
Literature
Review
...............................................................................................................
6

Introduction
..........................................................................................................................................
6

Human
Touch
.......................................................................................................................................
7

Physiology
of
Cutaneous
Tactile
Sensing
.........................................................................
9

Object
Identification
by
Touch
............................................................................................
13

Physiology
of
Vibrotactile
Receptors
...............................................................................
14

Biological
Mechanisms
of
Reflexive
Grip
Adjustment
..............................................
16

Biological
Mechanisms
of
Texture
Discrimination
.....................................................
19

Artificial
Touch
..................................................................................................................................
25

Artificial
Tactile
Sensing
of
Microvibrations
.................................................................
25

Dynamic
Tactile
Sensing
Technology
for
Texture
Discrimination
......................
30

The
Role
of
Fingerprints
in
Texture
Discrimination
.................................................
32

Chapter
3:
Preliminary
Findings
.......................................................................................................
35

Preface
..................................................................................................................................................
35

Contributions
of
the
Authors
...............................................................................................
36

Abstract
................................................................................................................................................
36

Introduction
........................................................................................................................................
37

Methods
................................................................................................................................................
39

Designing
a
Robust
Micro-‐Vibration
Sensor
.................................................................
39

Fabrication,
Inflation,
and
Leak
Detection
.....................................................................
41

Data
Acquisition
and
Signal
Processing
..........................................................................
42

x

Response
to
Sliding
Motion
over
a
Controlled
Surface
............................................
43

Results
...................................................................................................................................................
45

Pressure
Sensor
Response
to
Inflation
............................................................................
45

Spectral
Features
of
Contact
and
Slip
...............................................................................
46

Effects
of
Force
and
Velocity
................................................................................................
47

Discussion
............................................................................................................................................
50

Relationship
Between
Fluid
Pressure
and
Volume
....................................................
50

Dynamic
Response
of
the
Sensor
.......................................................................................
50

Applications
.................................................................................................................................
51

Conclusions
.........................................................................................................................................
53

Chapter
4:
A
Robust
Tactile
Sensor
.................................................................................................
54

Preface
..................................................................................................................................................
54

Contributions
of
the
Authors
...............................................................................................
55

Abstract
................................................................................................................................................
56

The
Need
for
Robust
Tactile
Sensors
.......................................................................................
56

The
BioTac
Design
...........................................................................................................................
60

Integrated
Electronics
.............................................................................................................
60

Size
and
Shape
............................................................................................................................
64

Fluidic
Seals
.................................................................................................................................
65

Fabricating
a
BioTac
................................................................................................................
67

Fingerprints
........................................................................................................................................
73

Discussion
............................................................................................................................................
76

Chapter
5:
Sensing
Tactile
Microvibrations
.................................................................................
77

Preface
..................................................................................................................................................
77

Contributions
of
the
Authors
...............................................................................................
77

Acknowledgements
..................................................................................................................
78

Abstract
................................................................................................................................................
78

Introduction
........................................................................................................................................
79

Methods
................................................................................................................................................
82

Physics
of
Fluidic
Vibration
Sensing
.................................................................................
82

Fabrication
of
the
BioTac
......................................................................................................
83

Electronics
Design
and
Noise
Analysis
............................................................................
84

Total
Noise
Calculation
and
Validation
...........................................................................
87

Data
Acquisition
........................................................................................................................
88

Signal-‐to-‐Noise
Estimation
...................................................................................................
88

Comparison
with
Human
Performance
...........................................................................
88

Results
...................................................................................................................................................
92

Vibration
Signals
During
Common
Tasks
.......................................................................
92

Theoretical
vs.
Actual
Noise
.................................................................................................
93

Comparison
with
Human
Performance
...........................................................................
94

Discussion
............................................................................................................................................
96

xi

Future
Work
................................................................................................................................
97

Chapter
6:
Bayesian
Exploration
......................................................................................................
99

Preface
..................................................................................................................................................
99

Contributions
of
the
Authors
...............................................................................................
99

Acknowledgements
...............................................................................................................
100

Abstract:
............................................................................................................................................
100

Introduction
.....................................................................................................................................
101

Material
and
Methods
..................................................................................................................
108

Classification
Theory
and
Strategy
.................................................................................
108

Biomimetic
Tactile
Sensor
.................................................................................................
116

Experimental
Apparatus
.....................................................................................................
119

Analytical
Measures
of
Descriptive
Texture
Properties
.......................................
124

Selection
of
Set
of
Exploratory
Movements
................................................................
134

Classifier
Training
and
Data
Collection
........................................................................
137

Texture
Discrimination
and
Comparison
with
Human
Performance
.............
138

Absolute
Texture
Identification
.......................................................................................
142

Results
................................................................................................................................................
143

Analysis
of
Descriptive
Texture
Properties
................................................................
143

Identifying
the
Most
Useful
Exploratory
Movements
............................................
146

Training
Dataset
.....................................................................................................................
149

Texture
Discrimination
and
Comparison
with
Human
Performance
.............
154

Absolute
Texture
Classification
.......................................................................................
155

Discussion
.........................................................................................................................................
160

Summary
of
Findings
...........................................................................................................
160

Considerations
for
Improving
the
Classifier
..............................................................
163

Considerations
for
Identifying
Objects
by
all
Available
Sensory

Modalities
...............................................................................................................................
165

Chapter
7:
Conclusions
.......................................................................................................................
167

What
is
Necessary
and
Sufficient
for
Sensing
Tactile
Microvibrations?
...............
167

Bayesian
Exploration
...................................................................................................................
169

Nature
to
Inspire
Robotics,
Robotics
to
Understand
Nature
.....................................
172

References
...............................................................................................................................................
174

xii

List
of
Tables

Table
2-‐1:
Human
Cutaneous
and
Proprioceptive
Touch
........................................................
8

Table
2-‐2:
Exploratory
Movement
Control
and
Property
Estimation
..............................
14

Table
4-‐1:
BioTac
Sensory
Modality
Details
................................................................................
64

Table
5-‐1:
Summary
of
Theoretical
Noise
Sources
...................................................................
87

Table
5-‐2:
Spheres
Used
for
Impact
Tests
....................................................................................
91

Table
5-‐3:
Summary
of
Performance
for
Impact
Tests
...........................................................
95

Table
6-‐1:
List
of
117
Textures
Used
in
Study
.........................................................................
123

Table
6-‐2:
Perceived
Properties
of
10
Textures
used
in
Pilot
Study
..............................
135

Table
6-‐3:
Correlation
Matrices
for
Texture
Properties
......................................................
153

Table
6-‐4:
Comparison
of
AB
Discrimination
to
Human
Subjects
...................................
155

Table
6-‐5:
Summary
of
Performance
for
Bayesian
Exploration
......................................
159

xiii

List
of
Figures

Figure
2-‐1:
Mechanoreceptors
Found
in
Human
Skin
.............................................................
12

Figure
2-‐2:
Exploratory
Movements
to
Determine
Properties
by
Touch
.......................
13

Figure
2-‐3:
Frequency
Response
of
Meissner
and
Pacinian
Corpuscles
.........................
16

Figure
3-‐1:
Tactile
Sensor
Conceptual
Design
............................................................................
39

Figure
3-‐2:
Sensing
of
Sliding
Microvibrations
vs.
Control
...................................................
41

Figure
3-‐3:
Procedure
for
Manually-‐Controlled
Sliding
Experiments
..............................
44

Figure
3-‐4:
Pressure
vs.
Inflation
Volume
....................................................................................
45

Figure
3-‐5:
Demonstration
of
Sliding
Vibration
Sensation
...................................................
47

Figure
3-‐6:
Response
to
Variations
in
Force
................................................................................
48

Figure
3-‐7:
Response
to
Variations
in
Velocity
..........................................................................
49

Figure
4-‐1:
Conceptual
Diagram
of
the
BioTac
...........................................................................
59

Figure
4-‐2:
Conceptual
Schematic
of
the
BioTac
Electronics
...............................................
61

Figure
4-‐3:
Fluidic
Seals
of
the
BioTac
...........................................................................................
66

Figure
4-‐4:
Molds
to
Produce
the
Biotac
Core
............................................................................
68

Figure
4-‐5:
Molds
to
Produce
the
BioTac
Skin
............................................................................
70

Figure
4-‐6:
Skin
Installation
Procedure
.........................................................................................
71

Figure
4-‐7:
Assembled
BioTac
...........................................................................................................
72

Figure
4-‐8:
Fingerprint
Pattern
of
the
BioTac
Skin
..................................................................
74

Figure
4-‐9:
Sliding
Vibrations
from
Smooth
and
Fingerprinted
Skin
...............................
75

Figure
4-‐10:
STFTs
of
Smooth
and
Fingerprint
Skin
over
Different
Materials
.............
76

xiv

Figure
5-‐1:
The
BioTac
Conceptual
Diagram
and
Picture
......................................................
81

Figure
5-‐2:
Signal
Conditioning
for
Pressure
Transducer
.....................................................
85

Figure
5-‐3:
Vibration
Signals
During
Common
Tasks
..............................................................
93

Figure
5-‐4:
Frequency
Sensitivity
of
the
BioTac
and
Human
Subjects
............................
94

Figure
5-‐5:
Sample
Signals
Measured
by
the
BioTac
During
Impact
Tests
....................
96

Figure
6-‐1:
The
BioTac
Conceptual
Schematic
and
Picture
with
Fingerprints
..........
117

Figure
6-‐2:
Texture
Exploration
Apparatus
..............................................................................
120

Figure
6-‐3:
Relationship
Between
Normal
Force
and
DC
Pressure
................................
121

Figure
6-‐4:
Typical
Signals
that
Occur
During
an
Exploratory
Movement
..................
131

Figure
6-‐5:
Measures
of
Texture
Properties
at
their
Optimal
Movements
.................
144

Figure
6-‐6:
Spectral
Centroid
as
a
Function
of
Sliding
Velocity
.......................................
145

Figure
6-‐7:
Selection
of
Optimal
Exploratory
Movements
.................................................
148

Figure
6-‐8:
Summary
of
All
Texture
Properties
for
117
Textures
..................................
150

Figure
6-‐9:
Confusion
Probability
Matrices
..............................................................................
152

Figure
6-‐10:
Evolution
of
Probabilities
Through
Bayesian
Exploration
......................
157

xv

Abstract

The
cutaneous
sensing
of
microvibrations
in
human
fingertips
plays
a
central

role
in
the
detection
of
slip-‐related
and
dynamic
information
critical
for
tool
usage,

reflexive
grip
control,
and
the
perception
of
microtextures.
This
is
made
possible
by

the
Pacinian
corpuscle,
a
small
sensory
receptor
in
subcutaneous
tissues
that
is

capable
of
detecting
vibrations
up
to
1000Hz
in
frequency.
These
receptors
are

sensitive
to
vibrations
less
than
a
micrometer
in
amplitude
at
frequencies
around

their
maximal
sensitivity
of
250Hz.
Artificial
systems
seeking
to
provide
human-‐like

dexterity
and
perception
would
benefit
from
similar
sensory
capabilities.

In
this
dissertation,
a
novel
tactile
sensor
capable
of
robustly
sensing
vibrations

with
a
bandwidth
and
sensitivity
that
exceeds
human
performance
is
presented.
The

device,
known
as
the
BioTac,
has
a
biomimetic
structure
that
consists
of
a
rigid

bone-‐like
core
covered
with
an
elastomeric
skin.
The
space
between
the
skin
and

core
is
filled
with
an
incompressible
low-‐viscosity
liquid
that
is
in
contact
with
a

pressure
transducer.
Vibrations
that
originate
on
the
surface
of
the
skin
are

transmitted
as
sound
waves
through
the
liquid
and
are
readily
sensed
by
the

transducer.
The
incompressible
liquid
conducts
these
acoustic
signals
with
little

attenuation,
permitting
the
transducer
to
be
located
in
a
protected
region
inside
the

core
of
the
device,
where
it
is
less
likely
to
get
damaged.
The
addition
of
a

xvi

biologically
inspired
fingerprint-‐like
pattern
on
the
surface
of
the
skin
was
found
to

enhance
vibrations
sensed
by
the
BioTac.
The
BioTac
exceeded
human
capabilities

in
sensitivity
thresholds
to
applied
sinusoidal
vibrations
and
impacts
from
small

spheres.

Biologically
inspired
strategies
to
use
this
tactile
information
for
a
texture

discrimination
task
were
developed.
A
specialized
robot
was
built
to
make
sliding

movements
similar
to
those
humans
make
when
exploring
textures.
The
BioTac
was

slid
over
a
total
of
117
different
textured
surfaces
collected
from
art
supply,
fabric

and
hardware
stores.
Signal
processing
methods
were
developed
to
extract

properties
modeled
after
the
descriptive
language
that
humans
use
when
describing

textures
(rough/smooth,
sticky/slippery,
coarse/fine).
Different
sliding
exploratory

movements
(defined
by
a
combination
of
contact
force
and
sliding
velocity)
were

found
to
be
optimal
for
discriminating
each
of
these
properties.
All
117
textures

were
tested
repeatedly
with
these
three
exploratory
movements
to
collect
a

database
of
prior
experience
similar
to
human
memory.
A
novel
process
of

intelligently
selecting
exploratory
movements
was
developed
to
guide
the

discrimination
task
when
presented
with
an
unknown
texture.
When
exploring
a

texture,
the
Bayesian
exploration
algorithm
selects
the
optimal
movement
to
make

and
the
property
to
measure
based
on
previous
experience
to
disambiguate
the

most-‐probable
candidates.
The
combination
of
biomimetic
hardware
and
software

achieved
performance
that
matched
(and
even
surpassed)
human
capabilities
in

discriminating
and
identifying
textures.

1

Chapter
1:
Introduction

Jeremy
A.
Fishel

Motivation

A
scientist/engineer
is
an
interesting
profession
that
creates
a
conflict
that
has

come
up
many
times
in
my
studies:
do
I
want
to
understand
things
or
do
I
want
to

build
things?
Mixing
the
fields
of
robotics
and
biomedical
engineering
to
make

tactile
sensors
only
makes
matters
worse:
do
I
want
to
understand
how
humans
do

things
with
touch,
or
do
I
want
to
make
robots
that
use
touch
to
do
things
that

humans
can
do?
Over
the
course
of
building
a
biomimetic
tactile
sensor
and

developing
these
applications,
I
have
come
to
an
interesting
realization.
These
two

goals
actually
serve
one
another
quite
well.

Everything
we
know
about
human
touch
and
biological
strategies
can
be
used
to

inspire
advanced
robotic
systems
that
seek
similar
performance
capabilities
as

humans.
This
is
nothing
new
though,
it
is
what
is
known
as
biomimetics
and
has

informed
many
useful
technologies
such
as
airplanes
1

and
Velcro.
However,
what

we
had
not
expected
was
that
these
robotic
systems
can
be
used
to
test
theories
of

human
behavior.
The
interactive
nature
of
touch
has
made
it
quite
difficult
to
study

1

Whether
or
not
airplanes
are
truly
biomimetic
is
a
topic
of
debate
as
they
do
not
incorporate

the
flapping
motion
of
birds.
However,
certainly
the
airfoil
shape
can
be
appreciated
as
a
biologically

inspired
design.

2

human
behavior
using
non-‐invasive
methods,
and
animals
(such
as
the
monkey
who

possesses
similar
physiology)
are
too
difficult
to
train
for
the
many
complex
and

dexterous
tasks
that
utilize
touch.
However,
as
our
technology
advances,
we
can

build
a
suitable
alternative
in
robotics
to
test
these
theories
(Loeb
et
al.
2011).
This

introduces
an
interesting
relationship
between
nature
and
robotics.
We
can
use
our

understanding
of
how
humans
use
touch
to
inspire
the
design
of
robots;
our

findings
from
these
robots
can
further
inspire,
test
and
reinforce
our
theories
of

how
humans
might
use
touch.

At
the
time
of
writing
of
this
thesis,
there
are
few
tactile
sensors
commercially

available
for
robotic
fingers
or
grippers.
Industrial
applications
have
evolved
to

function
without
tactile
sensing,
relying
instead
on
the
technology
for
machine

vision
as
sensory
input.
This
may
account
for
some
of
the
well-‐known
limitations
of

robots,
particularly
when
required
to
handle
objects
whose
identities,
properties

and
orientations
are
uncertain.
Instead,
most
robotic
systems
perform
effectively

only
when
interacting
with
rigid
objects
in
known
locations.

Rothwell
et
al.
(1982)
studied
the
performance
of
a
deafferented
man,
incapable

of
feeling
cutaneous
or
proprioceptive
touch
below
the
elbow.
While
the
subject
had

exceptional
performance
in
a
wide
range
of
prehensile
tasks,
even
reliable
force

production
in
the
fingertip,
without
the
ability
to
feel
he
was
incapable
of

performing
the
simplest
of
tasks
such
as
buttoning
his
own
shirt
button,
holding
a

cup
of
coffee,
or
writing
with
a
pen.
The
lack
of
dexterity
in
the
absence
of
tactile

feedback
is
further
evidenced
in
human
subjects
by
the
inability
to
maintain
grasp

3

on
an
object
when
the
skin
of
the
fingertips
is
anesthetized
(Johansson
&
Westling

1984).
Anyone
who
has
ever
had
numb
fingers
from
the
cold
can
appreciate
how

clumsy
one
can
be
without
the
ability
to
feel.
Robotic
systems
without
tactile

sensing
seem
to
exhibit
similar
limitations.
By
introducing
advanced
tactile
sensors

and
strategies
to
use
this
information
we
can
hope
to
achieve
the
dexterity
and

perception
humans
enjoy
from
tactile
sensitivity.

The
entire
sense
of
touch
is
to
daunting
of
a
topic
to
study
in
a
single
thesis.
In

humans,
there
are
proprioceptors
that
collect
information
on
the
orientation
and

forces
of
the
musculoskeletal
system,
thermoreceptors
that
sense
temperature
and

temperature
changes,
and
mechanoreceptors
to
sense
distribution
of
contact
forces,

skin
stretch
and
vibrations.
The
sensing
of
vibrations,
as
studied
in
this
thesis,
is

only
a
small
subset
of
these
tactile
capabilities.
However,
it
is
a
necessary
sensory

modality
that
gives
rise
to
many
dexterous
and
perceptual
tasks
humans
are
capable

of:
texture
discrimination,
grip
control,
and
tool
usage.
How
to
sense
these

vibrations
and
what
to
do
with
them
is
the
focus
of
this
research.
The
task
of

discriminating
texture
by
touch
is
explored
in
detail.
Solving
this
problem
for
a

robot
required
the
formulation
of
a
general
mathematical
basis
for
the
series
of

decisions
that
robots
or
humans
must
make
when
confronted
with
an
unknown

object
to
be
identified.

4

Contributions

This
thesis
consists
of
the
following
contributions
to
the
field
of
robotics
and

tactile
sensing:

• A comprehensive literature review on human physiology to detect tactile
vibrations and discriminate texture, as well as a review of artificial systems
seeking to replicate these abilities.
• The development of a highly robust tactile sensor (the BioTac) that
surpasses human sensitivity in detecting cutaneous vibrations.
• The development of a novel strategy for exploring objects by touch called
Bayesian exploration, which adaptively and intelligently selects optimal
exploratory movements to disambiguate and discriminate objects.
• The demonstration of performance in a texture discrimination over a large
number (117) of different textures using this tactile sensor and Bayesian
exploration algorithm; results surpassed human capabilities and other
prior art attempting to do the same.
Thesis
Overview

Chapter
2
provides
a
literature
review
of
microvibration
sensing
and
texture

discrimination.
This
review
consists
of
a
summary
of
biological
capabilities

identified
through
physiological
and
psychophysical
studies
and
prior
art
in

artificial
tactile
sensing
applications.

5

Chapter
3
presents
the
preliminary
findings
of
a
novel
micro-‐vibration
sensor

capable
of
sensing
vibrations
correlated
with
slip
and
texture.
This
is
made
possible

with
a
liquid-‐filled
sensor
that
takes
advantage
of
the
excellent
transmission

properties
of
incompressible
fluids
to
achieve
an
unparalleled
sensitivity
and

robustness.

Chapter
4
covers
the
design,
enhancements
and
refinements
of
this
prototype,

which
has
come
to
be
known
as
the
BioTac;
a
multimodal
tactile
sensor
that

includes
all
of
the
biological
cutaneous
tactile
sensing
capabilities
(force,
vibration

and
temperature).
Adding
a
fingerprint-‐like
pattern
to
the
skin
was
demonstrated
to

enhance
vibrations
that
arise
when
sliding
over
a
textured
surface.

Chapter
5
explores
the
sensitivity
of
the
vibration
sensing
modality
of
the

BioTac.
Signal-‐to-‐noise
performance
is
presented
and
the
device’s
sensitivity
is

found
to
surpass
even
human
sensitivity
to
vibrations.

Chapter
6
introduces
a
novel
algorithm
called
Bayesian
exploration
that

intelligently
selects
exploratory
movements
and
properties
to
measure
that

optimize
the
discriminability
between
possible
textures.
Performance
of
Bayesian

exploration
is
found
to
be
superior
when
compared
to
human
subjects
as
well
as

alternative
approaches
from
other
literature
in
artificial
texture
discrimination.

Chapter
7
provides
extrapolations
to
how
other
sensory
modalities
of
touch
can

be
optimized
and
studied,
the
broader
application
of
Bayesian
exploration,
and

thoughts
on
how
robotic
systems
can
teach
us
about
human
touch.

6

Chapter
2:
Literature
Review

Jeremy
A.
Fishel

Introduction

The
development
of
artificial
tactile
sensors
began
to
receive
attention
in
the

1980s
as
the
field
of
robotics
expanded
into
more
complex
tasks
expecting
benefit

from
the
sense
of
touch;
it
still
receives
considerable
attention
to
this
day.
Several

reviews
provide
a
comprehensive
survey
of
progress
(Nicholls
&
Lee
1989;
Lee
&

Nicholls
1999;
Howe
1994;
Dahiya
et
al.
2010).
Collectively,
they
document
the

evolution
of
tactile
sensing
requirements
from
non-‐biomimetic
engineering-‐like

requirements
(high
resolution,
low
hysteresis,
1000:1
dynamic
range,
etc.)(Harmon

1982)
to
requirements
that
begin
to
reflect
human
performance
and
capabilities

(multimodal,
skin-‐like
coverings,
compliance,
etc.)
while
engineering-‐driven

features
such
as
linearity
and
low-‐hysteresis
were
desirable
but
deemed
not
critical

(Dahiya
et
al.
2010).
While
many
of
these
developed
technologies
mention
bio-‐
inspired
design
principles,
the
concept
of
truly
mimicking
the
multimodality
of

human
touch
is
notably
absent.
Instead,
most
researchers
have
aimed
for
high-‐
resolution,
low-‐hysteresis
tactile
pixels
(taxels)
capable
of
sensing
only
distributions

of
static
force
in
the
normal
direction,
behaving
much
like
an
imaging
system

7

analogous
to
vision.
While
such
devices
may
find
their
niche,
this
is
but
a
small

subset
of
human
tactile
capability.
Cutaneous
touch
is
multimodal,
with
specialized

transducers
in
the
skin
to
sense
vibrations,
thermal
gradients
and
skin
stretch
in

addition
to
local,
normal
forces.
Of
particular
importance,
the
sensation
of
vibrations

appears
to
be
important
for
tool
usage
(Brisben
et
al.
1999),
a
defining
feature
of

human
development.
Sensitivity
to
vibrations
is
also
critical
for
slip
detection
for

grip
control,
and
as
perceptual
input
for
texture
discrimination.
This
review
focuses

on
the
often
overlooked,
yet
essential
sensory
modality
of
vibration
and
its

applications.

Human
Touch

Human
skin
contains
many
neural
transducers
whose
structures
are
specialized

to
detect
mechanical
strain
(mechanoreceptors),
thermal
information

(thermoreceptors),
and
pain
(nociceptors,
which
can
be
of
mechanical,
thermal,
or

chemical
nature)
(L.
A.
Jones
&
Lederman
2006).
Collectively,
the
sensory
nerves
in

the
skin
are
referred
to
as
cutaneous
receptors.
The
cutaneous
receptors
combined

with
proprioceptive
receptors
found
in
muscles
and
joints
comprise

somatosensation,
which
gives
rise
to
the
senses
of
touch
(sensation
of
the
skin)
and

kinesthesia
(position
and
motion
of
body
parts).
Summaries
of
the
functions
and

receptors
of
these
two
systems
are
provided
in
Tables
2-‐1
and
2-‐2.

Touch
is
intimately
coupled
with
movement,
unique
from
vision
and
hearing

which
are
for
the
most
part
passive
senses,
which
can
be
appreciated
without
any

8

physical
interaction
with
the
environment.
Movement
is
required
to
produce
tactual

sensations
that
can
be
used
to
perceive
the
environment
by
touch
(action
for

perception).
In
its
most
primitive
form,
it
supports
the
distinction
between
self
and

environment
through
tactile
exploration.
In
more
complex
tasks,
the
ability
to

produce
exploratory
movements
is
essential
for
extracting
object
properties

(Lederman
&
Klatzky
1987).

Function
Neural
Transducers

Cutaneous
Touch

Local
Force
/

Deformation

Merkel
discs,

Ruffini
endings

Vibration

Meissner
corpuscles,

Pacinian
corpuscles

Thermal

Thermoreceptors,

Free
nerve
endings

Proprioceptive
Touch

Actuator
Force
Golgi
tendon
organ

Actuator
Position

Muscle
Spindle
Primary,

Secondary
Afferents

Table
2-‐1:
Human
Cutaneous
and
Proprioceptive
Touch

Neural
transducers
in
the
skin
comprise
cutaneous
touch
while
those
that
reside
in
the
musculature

comprise
proprioceptive
touch.
Cutaneous
touch
receptors
are
responsive
to
a
wide
range
of

mechanical
stimuli
including,
force,
deformation,
skin
stretch,
vibration,
temperature
and
pain
(not

listed).
Proprioceptive
touch
receptors
are
sensitive
to
the
orientation
and
position
of
the
skeletal

system
and
the
forces
exerted
on
it
by
the
musculature
of
the
body.

The
relationship
between
touch
and
movement
is
symbiotic.
Touch
is
also

required
to
enable
dexterous
movements
of
the
hand
(perception
for
action).
For

9

instance,
the
ability
to
detect
slip
becomes
a
critical
function
in
maintaining
stable

grasp
of
an
object
(Johansson
&
Westling
1987;
Johansson
&
Flanagan
2009).
This

coupling
can
be
appreciated
by
the
exceptionally
poor
performance
of
subjects
in

simple
tasks
such
as
maintaining
grasp
of
an
object
when
cutaneous
tactile
feedback

is
blocked
by
anesthesia
(Johansson
&
Westling
1984),
or
buttoning
a
shirt
when

tactile
feedback
is
not
available
due
to
severe
peripheral
neuropathy
(Rothwell
et
al.

1982).

The
skin
and
its
sensory
transducers
are
highly
evolved
and
specialized
in

structure.
Studying
their
designs
and
behavior
can
inform
advanced
tactile
sensing

technologies.
The
glabrous
skin
found
on
the
palmar
surface
of
the
human
hand

possesses
characteristics
and
receptors
(Figure
2-‐1)
that
differ
from
the
hairy
skin

found
elsewhere
on
the
human
body.
For
instance,
glabrous
skin
contains
a
higher

density
of
mechanoreceptors;
this
is
especially
true
for
fingertips
(Vallbo
&

Johansson
1978;
Johansson
&
Vallbo
1979a).
Glabrous
skin
is
also
thicker
and
more

compliant
than
hairy
skin,
allowing
the
fingers
to
conform
to
the
wide
range
of

objects
they
may
encounter.
An
additional
feature
unique
to
glabrous
skin
is
the

presence
of
papillary
ridges,
or
fingerprints,
which
have
been
hypothesized
to
assist

with
the
transduction
of
vibrations
(Scheibert
et
al.
2009;
Fishel
et
al.
2009).

Physiology
of
Cutaneous
Tactile
Sensing

In
human
glabrous
skin
there
are
four
types
of
mechanoreceptors:
Merkel
disc

receptors,
Ruffini
endings,
Meissner
corpuscles,
and
Pacinian
corpuscles
(Vallbo
&

10

Johansson
1984;
L.
A.
Jones
&
Lederman
2006).
Neurophysiologists
employing

single-‐unit
microneurography
techniques
have
similarly
identified
four
populations

of
afferents
sensitive
to
different
forms
of
mechanical
stimuli.
They
have
been

classified
in
two
dimensions
by
their
adaption
rate
to
mechanical
stimuli
and
their

receptive
field
size
(Knibestöl
&
Vallbo
1970;
Johansson
1978;
Johansson
&
Vallbo

1983).
The
first
class
of
afferents,
identified
as
producing
strong
responses
to

dynamic
loading
with
little
output
to
sustained
loading,
have
been
described
as
fast-‐
adapting
(FA)
while
those
that
produced
sustained
responses
to
constant

mechanical
force
but
little
response
to
high-‐frequency
stimulation
were
described

as
slowly-‐adapting
(SA).
The
second
classification
is
based
on
receptive
field
size.

Afferents
with
smaller
and
more
precise
receptive
fields
(described
as
roughly

10mm
2
)
were
labeled
as
type-‐I
receptors,
while
larger
receptive
fields
(described
as

roughly
100mm
2

or
larger)
were
classified
as
type-‐II
receptors
(Johansson
&
Vallbo

1983).
All
four
possible
combinations
(FA-‐I,
FA-‐II,
SA-‐I,
SA-‐II)
have
been
observed
in

the
human
hand.
Johansson
et
al.
(1982)
characterized
the
sensitivity
of
these

afferents
as
a
function
of
frequency
and
amplitude
to
further
illustrate
their

differences.

While
the
Pacinian
corpuscle
has
been
identified
as
the
receptor
structure
for

the
FA-‐II
afferents
(and
thusly
sometimes
referred
to
in
literature
as
PC
afferents),

there
has
been
no
conclusive
proof
determining
a
relationship
between
the

remaining
afferents
and
nerve
endings
to
date
(L.
A.
Jones
&
Lederman
2006).
There

is,
however,
a
general
consensus
among
scholars
correlating
the
SA-‐I,
SA-‐II
and
FA-‐I

11

afferents
with
the
Merkel
disk
receptors,
Ruffini
endings,
and
Meissner
corpuscles

respectively
(Johnson
2001;
L.
A.
Jones
&
Lederman
2006).
It
is
worth
noting
that

Ruffini
endings
and
SA-‐II
afferents
have
never
been
observed
in
primates,
which
are

frequently
used
as
an
animal
model
for
touch
(L.
A.
Jones
&
Lederman
2006).

The
proposed
pairings
of
mechanoreceptors
and
afferents
can
be
appreciated
by

the
physical
location
and
structure
of
the
mechanoreceptors
(Figure
2-‐1).
Merkel

disk
receptors
and
Meissner
corpuscles,
both
assumed
endings
of
type-‐I
afferents

characterized
by
their
small
receptive
fields,
are
located
in
the
dermal
tissue
of
the

skin
near
the
dermal/epidermal
tissue
border.
In
contrast,
the
Pacinian
corpuscles

and
Ruffini
endings,
the
first
known
to
be,
and
the
latter
assumed
to
be
type-‐II

afferents
characterized
by
their
large
receptive
fields,
are
located
deeper
in
the

subcutaneous
layers
of
the
skin.
These
relationships
seem
appropriate,
given
the

spatial
dispersion
of
stresses
and
strains
as
they
transmit
through
elastic
media

such
as
skin.
This
has
been
validated
by
finite
element
models
of
the
human

fingertip
(Maeno
et
al.
1998).

The
physical
structure
of
the
nerve
endings
also
suggests
relationships
between

fast
and
slowly
adapting
nerve
endings.
The
encapsulation
of
Meissner
and
Pacinian

corpuscles
in
layers
of
thin
lamellae
and
interlamellar
fluid
function
act
as
a
high-‐
pass
filter,
similar
to
a
spring-‐dashpot
system
(Loewenstein
&
Skalak
1966).
This

permits
transmission
of
high-‐frequency
stimuli,
giving
rise
to
their
fast-‐adapting

nature,
while
avoiding
saturation
from
large
but
slowly
changing
forces.
In
contrast,

Merkel
disk
endings
are
unencapsulated
and
would
be
expected
to
produce
linear

12

output
in
response
to
mechanical
strain.
Ruffini
endings
are
spindle
shaped

receptors
and
are
believed
to
respond
to
skin
stretch
in
preferred
orientation,
much

like
those
found
in
the
proprioceptive
system.
They
are
typically
found
in
large

densities
around
the
fingernail
(Johansson
&
Vallbo
1979b)
where
stresses
from
the

fingertips
are
focused
in
response
to
skin
stretch.
A
thorough
review
of
the

functions
of
all
four
mechanoreceptors
is
provided
by
Johnson
et
al
(2000)
as
well

as
Jones
&
Lederman
(2006)
in
their
book
Human
Hand
Function.

Figure
2-‐1:
Mechanoreceptors
Found
in
Human
Skin

Mechanoreceptors
and
their
various
locations
in
human
skin
are
displayed
in
the
center
panel.
The

Meissner
corpuscles
and
Pacinian
corpuscles
(left
panel)
respond
to
dynamic
(vibration)
responses

due
to
their
laminar
structure.
The
Merkel
discs
and
Ruffini
endings
(right
panel)
respond
to
static

loads
(shear
and
normal
forces).
The
location
to
the
surface
of
the
skin
indicate
receptive
field
size,

with
the
Meissner
corpuscles
and
Merkel
discs
having
the
smallest
receptive
fields
and
the
Pacinian

corpuscle
and
Ruffini
endings
having
the
largest
receptive
fields.
Adapted
from:
(Vallbo
&
Johansson

1984).

13

Object
Identification
by
Touch

Touch
is
necessarily
an
interactive
sense,
unique
from
the
senses
of
vision
and

hearing.
While
we
are
able
to
observe
the
sights
and
sounds
of
our
environment

without
any
physical
interaction,
the
tactual
properties
of
an
object
can
only
be

sensed
be
physical
contact.
Experimental
psychologists
have
identified
six
general

types
of
exploratory
movements
that
humans
make
when
tactually
exploring

objects
to
determine
their
properties
(Lederman
&
Klatzky
1987)
(Figure
2-‐2).

Figure
2-‐2:
Exploratory
Movements
to
Determine
Properties
by
Touch

Six
different
exploratory
movements
humans
make
to
extract
object
properties.
Source:
(L.
A.
Jones
&

Lederman
2006)

The
control
of
these
exploratory
movements
as
well
as
the
information
required

to
extract
these
properties
are
highly
dependent
on
the
sense
of
touch.
We
propose

14

mechanisms
for
how
both
proprioceptive
and
cutaneous
touch
might
contribute
to

these
strategies
in
Table
2-‐2.

Exploration
Control
Varible
Sensory
Information

Static
Contact
Fingertip
Position
Thermal

Pressure
Fingertip
Position
and
Force
Local
Deformation,
Force

Lateral
Motion
Fingertip
Velocity
and
Force
Vibrations

Contour
Following
Fingertip
Position
Contact

Enclosure
Hand
Joint
Torques
Hand
Joint
Positions

Unsupported
Holding
Arm
Joint
Positions
Arm
Joint
Torques

Table
2-‐2:
Exploratory
Movement
Control
and
Property
Estimation

The
sense
of
touch
is
essential
both
to
produce
exploratory
movements
and
to
extract
tactual

properties.
Theories
of
how
to
use
such
information
for
both
control
exploratory
movements
and

property
extraction
in
these
tasks
is
proposed
for
both
proprioceptive
and
cutaneous
tactile
feedback.

Cutaneous
tactile
feedback
is
presented
in
italics.

Physiology
of
Vibrotactile
Receptors

The
perception
of
microvibrations
is
critical
for
tool
usage,
slip-‐detection
for
grip

control,
and
the
discrimination
of
textures.
In
order
to
produce
artificial
tactile

sensors
that
can
replicate
these
tasks,
it
is
necessary
to
understand
the
sensitivity,

bandwidth,
and
performance
criteria
of
human
vibration
receptors.

In
the
biological
hand,
Pacinian
corpuscles
with
frequency
responses
of
60-‐
700Hz
(Mountcastle
et
al.
1972)
are
capable
of
measuring
vibrations
associated

with
slip
and
other
transient
events.
Studies
have
demonstrated
sensitivity
to

vibrations
less
than
1µm
in
amplitude
around
their
center
frequency
of
200Hz

15

(Westling
&
Johansson
1987).
Others
have
reported
sensitivity
as
low
as
10nm

(Brisben
et
al.
1999).

Individual
neurons
cannot
transmit
action
potentials
at
the
higher
frequencies
of

the
Pacinian
corpuscle
sensitivity.
Instead,
the
receptors
have
been
demonstrated
to

fire
in
a
phase-‐locked
sub-‐harmonic
pattern
similar
to
that
found
in
auditory

neurons
(Freeman
&
Johnson
1982).
The
receptive
field
of
the
Pacinian
corpuscle
is

the
largest
of
all
mechanoreceptors
and
capable
of
detecting
stimuli
even

centimeters
away
(Johansson
&
Vallbo
1979a).
The
large
receptive
field
permits

multiple
receptors
to
be
activated
by
a
single
stimulus.
Therefore,
it
may
be
possible

for
high
frequency
information
to
be
conveyed
by
a
population
of
receptors

(Freeman
&
Johnson
1982).
While
it
has
generally
been
thought
that
information

was
only
contained
in
the
firing
rate
of
mechanoreceptors,
more
recent
studies
have

suggested
that
the
relative
timing
of
neural
spikes
from
populations
of

mechanoreceptors
can
also
convey
information
about
the
stimulus
shape
and

direction
of
motion
(Johansson
&
Birznieks
2004).

Meissner
corpuscles
have
a
higher
spatial
resolution
than
Pacinian
corpuscles

and
a
much
lower
and
narrower
bandwidth
of
5-‐40Hz
(Johansson
et
al.
1982).

However,
they
appear
in
larger
densities,
particularly
in
the
fingertips
(Johansson
&

Vallbo
1979b).
They
have
been
identified
with
the
sense
of
“flutter”
(Mountcastle
et

al.
1990).
A
comparison
of
the
frequency
responses
of
Pacinian
and
Meissner

corpuscles
is
given
in
Figure
2-‐3.

16

Figure
2-‐3:
Frequency
Response
of
Meissner
and
Pacinian
Corpuscles

The
sensitivity
threshold
to
sinusoidal
displacements
at
various
frequencies
is
plotted
for
the

Meissner
and
Pacinian
corpuscles.
Red
curve
represents
human
detection
threshold
and
black
curves

represent
the
sensitivities
of
the
individual
receptors.
Source:
(Kandel
et
al.
2000),

Biological
Mechanisms
of
Reflexive
Grip
Adjustment

Roland
Johansson
(a
co-‐inventor
of
the
BioTac
discussed
in
this
dissertation)

and
colleagues
have
contributed
a
large
portion
of
the
literature
on
biological

mechanisms
of
grip
control.
Their
early
work
showed
that
human
subjects
were

capable
of
automatically
adjusting
their
grip
on
an
object
such
that
it
was
always

slightly
above
the
required
force
needed
to
prevent
the
object
from
slipping

(Westling
&
Johansson
1984).
This
was
robust
to
variations
in
surface
texture
or

friction;
subjects
automatically
increased
grip
force
when
holding
slippery
objects
of

17

the
same
weight
(Johansson
&
Westling
1984).
The
response
times
of
these

adjustments
were
roughly
60-‐80ms,
indicating
that
this
mechanism
was
part
of
a

spinal
reflex.
The
inability
to
perform
this
task
when
the
skin
was
anaesthetized
in

these
experiments
indicated
that
this
desirable
behavior
was
dependent
on
tactile

feedback
(Johansson
et
al.
1992).
Subsequent
research
by
the
same
group

demonstrated
that
the
FA-‐I
afferents
increased
their
firing
rates
in
response
to
local

slips
before
these
automatic
grip
adjustments
occurred
(Westling
&
Johansson

1987;
Macefield
et
al.
1996).
Other
research
has
shown
the
involvement
of
both
FA-‐I

and
FA-‐II
in
detecting
slip
(Srinivasan
et
al.
1990).

The
fast-‐adapting
tactile
signals
in
reflexive
pathways
were
found
to
be
quite

salient.
Johansson
&
Westling
(1988)
explored
perturbations
caused
by
dropping

weights
onto
a
gripper
held
between
the
fingertips.
Their
findings
indicated
similar

reflexive
adjustments
in
response
to
the
impulse.
This
occurred
when
the

experimenter
initiated
the
impulse
or
the
subject
initiated
the
impulse,
even
though

in
the
latter
case
subjects
prepared
for
the
grip
by
increasing
force
in
preparation.

The
presence
of
vibratory
signals
regardless
of
grip
stability
created
the
illusion
of

slip
and
provoked
a
degree
of
reflexive
response,
albeit
much
smaller.
The

autonomous
nature
of
this
pathway
was
also
demonstrated
when
subjects
were

instructed
to
let
an
object
slip
slowly
from
grasp;
despite
conscious
desire
for
slip,

small
reflexive
grip
increases
were
still
detected
(Johansson
&
Westling
1987).

Further
research
by
Cole
and
Abbs
(1988)
supported
the
contribution
of
tactile

afferents
in
this
mechanism
by
observing
that
reflexive
adjustments
were
reduced

18

significantly
when
perturbations
were
applied
to
the
fingers
with
a
yoke
at
a

proximal
joint
compared
to
an
object
gripped
between
the
fingers.
When
grasping

an
object
with
two
different
surfaces,
control
of
grip
force
was
found
to
be

independent
in
each
digit,
suggesting
independent
reflexive
pathways
for
each
digit

(Edin
et
al.
1992).

Further
investigation
into
the
contributions
of
cutaneous
afferents
in
grip

control
indicated
that
the
FA-‐I
and
FA-‐II
afferents
contributed
mostly
to
fast
loading

rates,
with
FA-‐II
responding
mostly
to
step
responses
(Macefield
et
al.
1996).
In
the

same
study,
lower
loading
rates
SA-‐I
and
SA-‐II
were
found
to
have
the
strongest

correlation
to
grip
force,
suggesting
that
once
frictional
properties
are
understood,

slip
signals
are
not
necessary
to
predict
adjustments
to
grip.
Other
studies
have

indicated
that
predictive
adjustments
to
grip
based
on
cutaneous
afferents
occur

also
during
self-‐applied
perturbations
when
oscillating
an
object
at
low
frequencies

and
are
decreased
by
cutaneous
nerve
block
(Augurelle
et
al.
2003).

In
summary,
cutaneous
tactile
sensing
plays
a
critical
role
in
control
of
grasp.
FA-‐
I
and
FA-‐II
afferents
provide
information
on
slip
when
grasp
is
critical,
initiating

rapid
reflexive
grip
adjustment
as
well
as
providing
information
to
the
user
about

the
frictional
properties
of
the
grasped
object.
Once
these
frictional
properties
are

understood,
SA-‐I
and
SA-‐II
afferents
that
sense
normal
and
shear
forces
applied
to

the
skin
play
a
critical
role
in
predicting
the
appropriate
amount
of
grip
force.

19

Biological
Mechanisms
of
Texture
Discrimination

David
Katz
provided
one
of
the
earliest
studies
on
the
physiology
of
perceiving

texture
in
his
book,
The
World
of
Touch
(1925)
2
.
In
his
studies,
he
observed
that

while
coarse
textures
could
be
discriminated
based
on
their
static
contours
by

simply
pressing
down
on
an
object,
fine-‐textures
instead
required
sliding
motion
in

order
to
generate
vibrations
for
their
discrimination.
This
work
has
become
what
is

known
as
the
duplex
theory
of
texture.
Katz
proposed
the
judgment
of
coarse

textures
could
be
resolved
spatially
based
on
point-‐to-‐point
discrimination,
while

the
judgment
of
fine
textures
required
lateral
motion
to
produce
vibrations
that

could
also
be
sensed
in
the
finger.
He
observed
that
while
velocity
was
required
to

generate
these
vibrations,
the
classification
of
texture
was
relatively
velocity

independent
over
a
wide
range
of
speeds.

The
major
objection
to
Katz’s
theory
comes
from
Susan
Lederman
who
did
a

series
of
experiments
with
rather
large-‐scale
textures.
In
early
work,
Lederman

(1974)
had
subjects
explore
gratings
with
spatial
periods
ranging
from
0.5-‐1.25mm.

Based
on
these
studies,
she
concluded
that
perceived
roughness
was
independent
of

sliding
velocity
and
instead
increased
with
contact
force.
However,
the
data
were

collected
only
at
two
experimenter-‐controlled
forces,
1
oz.
and
16
oz.,
and
one
force

based
on
the
subjects'
own
preferences.
When
the
subjects
were
given
the
option
to

control
their
force,
perceived
roughness
agreed
more
with
that
of
the
light
1
oz.

2

The
original
manuscript
is
titled
Der
Aufbau
der
Tastwelt,
which
translates
to
"The
World
of

Touch."

20

experimenter-‐controlled
force,
suggesting
that
this
increase
in
perceived
roughness

might
have
been
due
to
the
potentially
awkward,
uncomfortable,
or
even
painful

forced
exploratory
movements.
No
statistical
difference
was
found
between
active

and
passive
touch
in
the
perception
of
roughness
with
these
surfaces,
further

suggesting
that
velocity
(or
at
least
cortical
encoding
of
velocity)
does
not
play
a

critical
role
in
the
perceived
roughness
of
large
spacing
textures
(Lederman
1981).

Further
research
into
the
mechanisms
of
roughness
perception
with
textures
of
this

size
led
the
investigators
to
conclude
that
vibration
plays
no
role
in
the

discrimination
of
textures
(Lederman
et
al.
1982).
While
this
body
of
work
from

Lederman
et
al.
originally
sought
to
debunk
Katz’s
theory
on
texture
discrimination,

regarding
the
importance
of
velocity
in
sensing
textures,
surfaces
with
spatial

features
less
than
perceptual
thresholds
of
static
contact
were
never
used.
Indeed,

Katz
clearly
proposed
that
large
scale
textures
could
be
discriminated
spatially

without
movement,
while
fine
textures
were
those
that
required
sliding
movements

for
their
discrimination.
Studies
on
two-‐point
spatial
resolution
of
the
human

fingertip
touching
similar
size
gratings
(down
to
0.5mm)
have
shown
that
humans

possess
this
spatial
discrimination
(Johnson
&
Phillips
1981)
for
the
entire
range
of

textures
explored
by
Lederman
et
al.

To
better
understand
the
physiological
mechanisms
behind
texture

discrimination,
Darian-‐Smith
et
al.
(1980)
studied
responses
in
the
SA,
RA
and
PC
3

3

Because
primates
do
not
have
SA-‐II
afferents
as
observed
in
humans,
the
SA-‐I
afferents
are

typically
referred
to
as
simply
SA,
while
the
FA-‐II
are
referred
to
as
PC
for
the
known
correspondence

21

cutaneous
afferents
in
the
monkey
when
sliding
a
macro-‐texture
pattern
of
raised

dots
ranging
from
0.55-‐2.25mm
at
velocities
ranging
from
4-‐15
cm/s.
Responses
in

SA
were
shown
to
decrease
in
coherency
with
the
features
as
velocities
were

increased
(due
to
the
poor
dynamic
response
of
SA
receptors)
or
the
spacing
was

decreased
(due
to
the
spatial
resolution
limits).
By
contrast,
the
PC
pathways

showed
the
opposite
trend;
further
supporting
that
there
was
a
transition
between

two
exploratory
strategies.
Similar
results
to
these
studies
using
raised
dots
were

observed
with
gratings
of
the
same
dimensions
(Darian-‐Smith
&
Oke
1980).

Lamb
demonstrated
the
capabilities
in
humans
to
discriminate
small
differences

in
textured
bump
spacings
on
the
order
of
2mm
+/-‐
1-‐8%
with
a
75%
confidence

interval
of
successfully
selecting
whether
the
second
surface
presented
was
finer
or

coarser
from
only
a
2%
change
in
spacing
distance
(Lamb
1983b).
This
was
further

investigated
in
peripheral
neural
recordings
in
the
monkey
using
both
1
and
2mm

spacings
(Lamb
1983a).
Results
indicated
that
RA
afferents
provided
the
strongest

temporal
coding,
enabling
discrimination
of
the
larger
spacings.
In
these
single

afferent
studies
it
was
noted
that
such
sensitivity,
as
demonstrated
in
human

subjects,
would
require
a
large
number
of
afferents
for
proper
denoising.
At
smaller

spacings
of
1mm,
the
PC
afferents
provided
the
most
information,
suggesting
that
as

spacing
decreased,
the
mechanism
changed
to
higher-‐frequency
dynamic
coding.

to
the
Pacinian
corpuscle,
FA-‐I
is
simply
referred
to
as
RA
for
rapidly
adapting.
The
RA
and
PC
labels

are
also
sometimes
used
to
describe
FA-‐I
and
FA-‐II
in
biological
studies.

22

LaMotte
(1991)
explored
the
detection
thresholds
for
discriminating
small

raised
dots
and
bars
with
microtexture
spacings
of
100µm.
Results
indicated
that

feature
heights
greater
than
0.2µm
could
be
readily
detected
above
chance
when

compared
to
a
smooth
surface.
Sensitivity
was
highest
when
the
motion
was

perpendicular
to
the
fingerprints.
In
this
same
study,
nerve
recordings
indicated

that
PC
fibers
were
the
most
responsive
to
these
small
textures.
Action
potentials

were
recorded
in
response
to
features
as
small
as
0.06µm.
Further
investigations
by

Miyaoka
et
al.
(1999)
proposed
that
microtexture
amplitude
was
a
key

discriminator
for
texture
roughness.
In
their
studies
discrimination
was
explored

for
various
grits
of
fine
sand
paper
ranging
in
particle
size
of
1-‐40µm
as
well
as

single
ridges
ranging
in
height
of
6.3-‐25µm.
Their
findings
indicated
a
Weber

fraction
of
0.28
for
discriminating
various
textures
and
0.16
for
discriminating

single
ridges.
They
proposed
the
improved
sensitivity
of
discriminating
ridge

heights
was
due
to
the
coherence
of
sensory
receptor
firing
when
sliding
over
the

ridge.

Studies
in
microtexture
discrimination
continued
with
Hollins
&
Risner
(2000),

who
demonstrated
proof
of
Katz’s
duplex
theory.
When
subjects
were
asked
to

classify
objects
on
the
dimension
of
roughness-‐smoothness,
they
were
able
to
show

that
without
motion
subjects
were
incapable
of
identifying
differences
between
grits

of
fine
sand
paper
with
particles
smaller
than
100µm;
with
motion
they
were
able
to

classify
properly
(Hollins
&
Risner
2000).
The
study
further
went
on
to
show
that

roughness
of
abrasive
papers
with
larger
grits
around
200-‐500µm
could
be

23

classified
properly
without
motion.
Later
studies
by
the
same
group
recorded
skin

vibrations
with
a
Hall-‐effect
sensor
and
demonstrated
testure-‐specific
spectral

differences
that
were
within
the
bandwidth
of
the
Pacinian
receptors
(Bensmaïa
&

Hollins
2003;
Bensmaïa
&
Hollins
2005;
Yau
et
al.
2009).

Hollins
et
al.
(2000)
found
that
vibrating
a
surface
at
150Hz
made
a
surface
feel

rougher
in
comparison
to
the
control
of
no
vibration.
This
was
found
to
be
true
for
a

wide
range
of
frequencies
up
to
400Hz,
with
all
being
equally
effective
in
producing

the
illusion.
The
authors
concluded
that
the
Pacinian
channels
provide
an
intensity

code
that
is
used
to
discriminate
texture.
This
finding
demonstrating
increased

roughness
in
the
presence
of
vibration
is
particularly
significant
because
vibrating
a

surface
tends
to
reduce
the
actual
friction;
this
is
the
basic
operating
principle
of
a

haptic
display
device
known
as
the
T-‐Pad
4

by
Winfield
et
al.
(2007).
More
detailed

investigations
to
discriminate
the
contributions
of
the
FA-‐I
and
FA-‐II
pathways
were

done
using
vibrotactile
adaptation
at
either
10
or
100Hz
to
target
the
two
specific

pathways.
The
group
found
that
adaptation
at
100Hz
only
inhibited
the
ability
to

discriminate
fine
textures,
but
not
coarse
ones,
suggesting
the
important
role
of
the

Pacinian
corpuscles
in
classifying
microtextures
(Hollins
et
al.
2001).

Smith
et
al.
(2002a)
proposed
a
different
hypothesis
related
to
the
assessment
of

roughness.
They
found
that
the
rate
of
change
(vibrations
less
than
5Hz)
in

tangential
forces
had
a
strong
correlation
with
perceived
roughness
when
analyzing

4

The
T-‐Pad
operates
at
frequencies
beyond
the
sensitivity
of
Pacinian
corpuscles

24

exploratory
forces
applied
by
subjects
exploring
textured
ridges
ranging
from
1-‐
8.5mm
spacings.
This
was
found
to
have
a
stronger
correlation
with
perceptually

rougher
surfaces
than
kinetic
friction.

Recognizing
that
texture
was
more
than
simply
an
array
of
features,
Hollins
et
al.

(1993)
investigated
the
quality
characteristics
of
surfaces.
In
a
rather
simple,
yet

very
useful,
experiment
subjects
were
presented
17
different
textures
and,
without

any
guidance,
were
asked
to
group
them
in
anywhere
from
three
to
seven
groups

based
on
similarity.
Subjects
were
then
asked
to
rate
each
surface
on
scales
for

roughness/smoothness,
hardness/softness,
slippery/sticky,
flat/bumpy,
warm/cool.

Based
on
these
groupings
of
co-‐occurrence,
the
authors
used
multidimensional

scaling
analysis
to
find
a
reasonable
convergence
into
a
three
dimensional
space.
Of

the
most
salient
descriptors
that
made
these
surfaces
different,

roughness/smoothness
and
hardness/softness
proved
to
be
the
most
orthogonal,

with
slippery/sticky
being
the
third
most
independent
descriptor.
However,
it

should
be
noted
that
others
attribute
hardness/softness
to
an
object’s
compliance,

which
is
independent
of
texture
(Lederman
&
Klatzky
1987).

The
low
number
of
surfaces
in
Hollins
et
al.
(1993)
has
been
criticized
by

Bergnam
Tiest
and
Kappers
(2006)
noting
that
a
convergence
to
three-‐dimensional

space
is
rather
meaningless
with
only
17
textures,
as
16
profoundly
different

surfaces
would
be
required
to
occupy
the
corners
of
a
higher
four-‐dimensional

space.
To
expand
this
analysis,
a
total
of
125
surfaces
were
studied
using
similar

multidimensional
analysis
techniques.
Findings
from
this
expanded
database
have

25

suggested
that
there
are
at
least
four
perceptual
dimensions
of
surfaces,
although

not
all
could
be
correlated
specifically
with
named
properties.
It
is
proposed
that

some
of
this
dimensionality
could
be
attributed
to
non-‐textural
haptic
properties

such
as
compliance
and
thermal
properties
that
might
be
sensed
without
the
need

for
sliding
movements.
Other
dimensionality
in
vision
and
audition
have
also
been

proposed
for
surface
texture
(Klatzky
&
Lederman
2010).

The
literature
on
texture
discrimination
supports
Katz’s
duplex
theory
that
large

features
can
be
discriminated
by
tactile
pathways
capable
of
resolving
them

spatially,
while
features
that
are
too
small
to
resolve
require
the
finger
to
slide
over

a
surface
and
are
dependent
on
the
vibrations
thus
produced
and
sensed.
The

detailed
body
of
work
from
Hollins’
group
indicates
that
the
Pacinian
corpuscles
are

highly
involved
in
this
pathway
and
evidence
has
been
presented
that
information
is

conveyed
largely
in
the
form
of
an
intensity
code.
With
specific
regards
to
texture

and
sliding
movements,
the
dimensions
of
sticky/slippery,
rough/smooth
and

coarse/fine
seem
to
be
the
most
salient
descriptions
of
properties
that
make

textures
distinct.

Artificial
Touch

Artificial
Tactile
Sensing
of
Microvibrations

Mechanical
analogs
for
proprioceptive
touch
are
readily
available
and
widely

used
in
a
variety
of
devices
such
as
strain
gages,
position
sensors
and
the
like.

26

However,
finger-‐like,
compliant
and
deformable
sensors
capable
of
detecting
stimuli

similar
to
biological
mechanisms
of
cutaneous
touch
have
been
notably
absent
from

the
commercial
market.
This
is
not
due
to
lack
of
interest
or
effort.
This
section

outlines
the
progress
in
developing
such
sensors
with
particular
focus
on
the

sensing
of
microvibrations
and
dynamic
touch.

Many
approaches
to
transduce
dynamic
tactile
stimuli
into
measurable
electrical

signals
have
been
explored
over
the
years,
based
on
mechanisms
from
virtually

every
physical
phenomenon
imaginable
such
as
piezoelectrics
(Dario
et
al.
1984),

accelerometers
(Howe
&
Cutkosky
1989;
Romano
et
al.
2011),
optics
(Dubey
&

Crowder
2006;
Kamata
et
al.
2005),
electrical
capacitance
(Ko
et
al.
2006;
Schmidt
et

al.
2006),
acoustics
(Edwards
et
al.
2008)
and
other,
even
more
exotic
technologies.

Comprehensive
reviews
of
artificial
tactile
sensing
developed
over
the
last
three

decades
are
provided
(Nicholls
&
Lee
1989;
Lee
&
Nicholls
1999;
Howe
1994;

Dahiya
et
al.
2010).
Only
some
of
these
technologies
have
the
requisite
temporal

bandwidth
to
be
classified
as
dynamic
tactile
sensors.
While
performance
for
many

of
these
sensors
was
generally
good
enough
for
a
particular
experiment,
many

achieve
the
requisite
sensitivity
at
the
expense
of
fragility,
creating
many
challenges

for
producing
robust
sensors.
This
is
particularly
a
problem
if
the
fingertip
must

also
incorporate
a
deformable
skin,
an
important
contributor
to
stability
of
grip.

The
sensitivity
and
frequency
response
of
most
tactile
sensing
technologies
is

often
inadequate
for
sensing
the
vibrations
associated
with
texture
discrimination.

Resistive
array-‐based
tactile
sensing
technologies
are
one
of
the
most
common

27

forms
of
tactile
sensors
explored.
One
of
the
earliest
designs
was
developed
by
Hillis

et
al.
(1982)
who
used
a
conductive
silicone
rubber
separated
from
a
conductive

PCB
board
with
a
256
element
grid-‐like
spacer.
As
mechanical
forces
were
applied

to
the
rubber
it
would
come
in
contact
with
the
PCB
causing
the
resistance
to

decrease
for
that
element.
Other
designs
have
been
implemented
based
on
similar

principles
using
orthogonal
conductive
film
(Speeter
1990)
or
cylindrical

conductive
carbon
fibers
(Larcombe
1981).
These
strategies
typically
result
in
a

high
degree
of
noise
from
intermittent
electrical
connections
in
the
deformable

media
and
don’t
possess
the
requisite
sensitivity
or
bandwidth.

Similar
strategies
with
capacitive
sensing
arrays
have
been
used
with

deformable
materials
that
change
the
distance
and
sensed
capacitance
in
response

to
mechanical
forces(Fearing
1987;
Fearing
1990).
A
novel
implementation
of
this,

introduced
by
Schmidt
et
al.
(2006),
coupled
hair-‐like
extrusions
to
capacitive
plates

and
demonstrated
exquisite
sensitivity
and
dynamic
response
but
were
not
very

robust.
Sensors
based
on
arrays
of
small
elements
suitable
for
spatial
resolution
of

touch
have
a
limit
to
their
sensitivity
because
they
capture
less
incident
energy
and

often
have
high
electrical
impedances,
resulting
in
poor
signal-‐to-‐noise
ratios.

Larger
elements
become
increasingly
susceptible
to
electrical
interference.

Vranish
developed
a
magnetoinductive
skin
by
sensing
changes
in
applied

magnetic
field
as
conductive
wires
were
moved
mechanically
(1986).
Results
were

quite
impressive
with
a
theoretical
sensitivity
of
down
to
3
Pa
or
25
µm
of
deflection.

28

However
this
design
required
rather
intricate
electrical
connections
and
required

extensive
shielding
from
noise
sources
to
work
properly.

The
use
of
accelerometers
has
provided
promise
for
dynamic
tactile
sensing.

Howe
and
Cutkosky
(1989)
have
demonstrated
this
principle
using
a
MEMS

accelerometer
embedded
near
the
surface
of
a
rubber
skin
overlying
a
foam

substrate.
The
performance
of
the
sensor
indicated
sensitivity
to
vibrations
that

arise
under
slip
and
texture.
However,
the
placement
of
the
sensor
near
the
surface

of
the
skin
makes
the
transducer
vulnerable
to
damage.
More
recently,
measuring

vibrations
using
an
accelerometer
mounted
on
a
robotic
gripper
has
enabled
the

gripper
to
produce
reflexive
adjustments
to
grip
in
the
presence
of
slip-‐related

vibrations
(Romano
et
al.
2011).
This
approach,
while
robust,
did
not
possess
the

requisite
sensitivity
similar
to
human
touch.

Dario
et
al.
(1984)
was
the
first
to
utilize
thin
film
polyvinylidene
difluoride

(PVF2
or
more
commonly
PVDF
in
other
literature).
The
material
is
well
known
for

its
piezoelectric
and
pyroelectric
properties
that
can
be
obtained
by
applying
a
high

electric
field.
The
use
of
piezoelectrics
has
received
widespread
attention
in

dynamic
tactile
sensing.
Examples
of
this
technology
are
provided
in
literature

(Patterson
&
Nevill
1986;
Dario
&
Buttazzo
1987;
Gao
et
al.
1990;
Howe
&
Cutkosky

1993;
Son
et
al.
1994;
Yamada
&
Cutkosky
1994;
Dario
et
al.
1994;
Kolesar
&
Dyson

1995;
Dargahi
et
al.
2000;
Krishna
&
Rajanna
2004;
Hosoda
et
al.
2006;
Mirbagheri

et
al.
2007;
Cotton
et
al.
2007;
Chuang
et
al.
2008).
While
the
PVDF
film
commonly

used
in
these
sensors
possesses
suitable
bandwidth
and
sensitivity
for
dynamic

29

tactile
sensing,
the
film
needs
to
reside
close
to
the
region
of
contact
where
it
can

become
easily
damaged
making
it
unsuitable
for
applications
beyond
controlled

laboratory
experiments.

Tactile
sensors
based
on
ultrasound
have
been
used
to
measure
deformation

height
(Hutchings
et
al.
1994)
or
even
phased
arrays
to
localize
contact
location

(Shinoda
&
Ando
1994).
The
high
frequency
of
ultrasound
permits
these
to
be
used

as
dynamic
slip
sensors
with
sensitivity
to
deflections
down
to
7µm.
However,
the

complex
and
high-‐power
electronics
required
for
ultrasound
make
it
undesirable
for

these
applications.

Other
more
exotic
technologies
have
been
explored
as
well.
Shinoda
et
al.
(1993)

have
designed
a
sensor
that
uses
a
3-‐dimensional
structure
of
air
tubes
at
different

depths.
By
sensing
the
relative
pressure
in
these
tubes
they
were
able
to
extrapolate

the
force
applied
to
the
sensor
as
well
as
detect
spectral
content
in
the
presence
of

slip.
Later
work
from
Ando
and
Shinoda
(1995)
developed
a
method
of
measuring

ultrasonic
emission
from
touch
to
localize
and
detect
slip.

Many
of
these
technologies
have
been
used
to
create
high-‐density
arrays
of

sensors
for
detecting
forces
and
are
unsuitable
or
lack
the
sensitivity
for
detecting

the
small
vibrations
that
humans
are
capable
of
sensing.
In
almost
all
cases
the

requisite
sensitivity
for
dynamic
vibration
sensing
requires
the
transduction

elements
to
be
mounted
near
the
surface
of
the
skin,
where
they
are
prone
to

damage.

30

Dynamic
Tactile
Sensing
Technology
for
Texture
Discrimination

Biological
studies
indicate
that
two
different
types
of
tactile
receptors
are

involved
in
the
sensation
of
texture.
Macro-‐scale
features
can
be
encoded
by
high-‐
resolution
transducers
capable
of
imaging
at
this
scale.
Several
artificial

technologies
have
used
high-‐resolution
taxel
arrays
to
achieve
this
performance,
e.g.

(Maheshwari
&
Saraf
2006).
In
contrast,
fine-‐textures
can
be
encoded
by
the

patterns
of
vibrations
that
they
produce
in
the
skin
during
sliding
exploratory

movements,
similar
to
biological
mechanisms
of
fine-‐texture
discrimination.
Tactile

sensors
utilizing
this
approach
have
received
much
less
attention
in
the
artificial

domain.
This
review
focuses
on
those
sensors
that
seek
to
discriminate
textures

based
on
their
dynamic
properties
while
sliding.

Howe
and
Cutkosky
(1989)
investigated
the
dynamic
contributions
of
texture

exploration
using
a
sensor
with
an
accelerometer
embedded
in
an
elastic
substrate.

Their
findings
indicated
that
while
the
vibration
amplitude
increased
at
heavier

contact
forces,
the
frequency
of
these
vibrations
were
primarily
at
700Hz,

representing
the
natural
frequency
of
their
particular
sensors.
Later
studies
from

the
same
group,
this
time
using
PVDF
strips
embedded
into
a
rubber
skin
that
was

overlaid
on
a
foam
core
to
measure
the
strain
rate
of
the
skin,
demonstrated
that

features
as
small
as
6.5µm
could
be
resolved
as
the
device
slid
over
a
surface
(Howe

1991;
Howe
&
Cutkosky
1993).

Tada
and
Hosoda
et
al.
(2003;
2004)
also
investigated
vibrations
from
texture

with
embedded
PVDF
strips
molded
in
rubber;
however,
the
group
took
a
rather

31

interesting
approach
in
their
sensor
design.
Instead
of
creating
a
careful
array
of

elements,
as
many
sensors
as
possible
were
randomly
embedded
in
their
silicone

substrate.
This
sensor
was
then
slid
over
wood
and
paper.
Given
the
random
nature

of
sensor
placement,
many
transducers
produced
relatively
useless
output
for

discrimination,
but
two
channels
produced
some
degree
of
unique
clustering
when

plotting
their
respective
variances
in
two
dimensions.
Further
development
by
this

group
expanded
this
analysis
to
an
additional
sensory
dimension
using
an

embedded
strain
gage
(Hosoda
et
al.
2006).

Using
a
tri-‐axial
MEMS
force
sensor,
de
Boissieu
et
al.
(2009)
analyzed
the

spectrum
as
the
sensor
was
slid
over
various
types
of
paper
and
found
unique

spectral
components
for
some
of
the
samples.
The
group
was
able
to
demonstrate
a

neural
network
classifier
with
70%
accuracy
in
properly
classifying
ten
different

textures.
Two
different
methods
were
explored
as
inputs
to
the
classifier:
the
first

utilized
differences
in
the
Fourier
coefficients
while
the
second
used
more
direct

signal
features
such
as
the
mean
values,
variance
and
kurtosis
of
the
time
signals
as

well
as
the
mean
spectral
power
and
slope
in
the
bandwidth
of
0-‐250Hz
as
inputs.

Kurtosis,
mean
spectral
power
and
average
spectral
slope
seemed
to
generate
the

most
information
for
the
model.
A
similar
approach
was
taken
by
Giguere
and

Dudek
(2011)
using
accelerometers
on
a
rigid
tactile
probe
to
classify
surfaces
over

which
a
wheeled
robot
was
driving,
based
on
their
means,
variance
and
higher-‐
order
moments
in
a
neural
network,
achieving
90-‐94%
classification
accuracy
over

10
surfaces.

32

Jamali
and
Sammut
(2011)
analyzed
Fourier
components
of
eight
different

textures
using
a
novel
method
of
majority
voting.
In
this
study,
a
set
of
three

different
sliding
motions
was
performed
in
each
test
and
then
the
most
likely

texture
was
selected
for
that
test
as
a
“vote”.
Convergence
required
at
least
three

consecutive
“votes”
for
a
texture
from
the
initial
three
tests.
If
not
obtained

additional
tests
were
executed
until
one
texture
candidate
possessed
greater
than

80%
of
all
votes
or
the
exploration
exceeded
10
tests.
With
this
approach
they
were

able
to
demonstrate
classification
of
eight
different
textures
with
a
performance
of

95%.

Expanding
the
repertoire
of
exploratory
movement
types
and
sensory

information
has
been
shown
to
help
the
discrimination
of
textures
(Sinapov
&

Stoytchev
2010;
Sinapov
et
al.
2011).
In
this
study,
classification
accuracy
improved

from
65%
using
the
single
best
movement
to
80%
for
using
five
different

movements
to
classify
20
different
textures.

The
Role
of
Fingerprints
in
Texture
Discrimination

In
a
high-‐profile
paper
published
in
Science,
Scheibert
et
al.
(2009)
explored
the

role
of
fingerprints
in
generating
vibrations.
Using
a
pressure
sensor
with
a
near-‐
biomimetic
fingerprint
spacing
of
0.22mm,
they
slid
textured
surfaces
past
the

device
at
very
low
speeds
of
0.02cm/s
and
found
that
the
measured
~1Hz

"vibrations"
fit
the
equation
of:

33

(2.1)

Based
on
this
finding,
they
proposed
that
at
more
biomimetic
sliding
velocities

(roughly
200-‐300
times
faster),
they
would
expect
to
see
vibrations
of
200-‐300Hz,

which
corresponds
to
the
peak
sensitivity
of
the
Pacinian
corpuscle.
They
concluded

that
this
was
the
mechanism
of
fingerprints.
A
recent
review
(Dahiya
&
Gori
2010)

of
this
paper
brought
up
the
issue
of
using
such
a
low
exploratory
velocity
and

recommended
that
the
same
data
be
collected
at
more
biomimetic
speeds
to

validate
the
findings.

An
earlier
study
by
Mukaibo
et
al.
(2005)
reported
similar
results.
The
design
of

their
sensor
included
fingerprint-‐like
ridges
spaced
at
1.5mm
(~5
times
larger
than

biological
fingerprints).
Vibrations
were
recorded
in
embedded
strain
gages
near

the
surface
of
the
sensor.
They
reported
the
same
relationship
among
velocity,

wavelength
and
frequency
as
reported
by
Schiebert’s
group,
but
for
the
spatial

wavelength
of
the
explored
surface
rather
than
the
fingerprints.
Surfaces
were

explored
at
a
more
biomimetic
exploratory
speed
(2cm/s)
than
that
used
by

Schiebert's
group,
which
was
two
orders
of
magnitude
slower.
They
report
that

while
scanning
finest
of
their
textures
(spacing
of
0.2mm),
the
data
did
not
fit
this

relationship
as
well,
but
improved
somewhat
when
sliding
at
a
slower
speed
of

1cm/s.
In
another
study
Oddo
et
al.
(2009)
investigated
the
frequency
dependency

of
coarser
texture
patterns
of
2.6-‐4.1mm
explored
at
scanning
velocities
of
1.5-‐4.8

cm/s.
They,
too,
observed
a
correlation
between
scanning
velocity
and
frequency

f =
v
λ

34

(Oddo
et
al.
2009).
While
not
discussed
in
their
paper,
their
data
showed
that
as

sliding
velocity
increased,
so
did
the
error
in
this
relationship.

The
experimental
methods
of
all
Scheibert
et
al.
(2009),
Mukaibo
et
al.
(2005).,

and
Oddo
et
al.
(2009)
did
not
accurately
reflect
biological
dimensions
for
the

fingerprints
and/or
scanning
velocity
for
microtexture
discrimination
(Scheibert’s

group:
scanning
100x
too
slow,
Mukaibo’s
group:
fingerprints
5x
too
large,
Oddo’s

group
textures
about
10x
too
large).
The
simple
relationship
among
frequency,

scanning
velocity
and
spatial
period
described
in
equation
(2.1)
was
found
but

tended
to
fall
apart
as
the
parameters
became
more
biologically
realistic.
Reducing
a

regularly
spaced
texture
into
the
microtexture
range
or
increasing
the
sliding

velocity
into
biological
ranges
appears
to
introduce
some
substantial
non-‐linearity,

indicating
that
dynamics
play
a
significant
role.
Despite
reports
from
these
artificial

studies,
neither
the
encoding
of
microtextures
nor
the
role
of
fingerprints
seems
to

be
well-‐understood
in
human
texture
discrimination.

35

Chapter
3:
Preliminary
Findings

Jeremy
A.
Fishel,
Veronica
J.
Santos
and
Gerald
E.
Loeb

©
2008
IEEE.
Reprinted
with
permission
from:

Fishel,
J.A.,
Santos,
V.J.,
Loeb,
G.E.
A
Robust
Micro-‐Vibration
Sensor
for

Biomimetic
Fingertips.
In
Proceedings
of
the
IEEE
2nd
Biennial
IEEE/RAS-‐
EMBS
International
Conference
on
Biomedical
Robotics
and
Biomechatronics

2008,
Scottsdale,
AZ,
USA,
19-‐22
October,
2008;
pp.
659-‐663.

Preface

This
chapter
presents
the
preliminary
findings
in
the
detection
of

microvibrations
with
a
liquid
based
sensor.
The
promising
results
of
this
study
led

to
the
development
of
the
vibration
sensory
modality
of
the
BioTac.
Many

differences
exist
between
the
prototype
used
in
this
study
and
the
BioTac
as

discussed
in
Chapter
4.
However,
several
important
findings
are
uncovered.
In

particular,
the
observation
of
poor
correlation
between
spectral
frequency
and

sliding
velocity
was
first
proposed
here.
Three
possible
applications
were
presented

for
this
sensor:
slip
detection,
texture
perception
and
conscious
feedback.
The

application
of
texture
discrimination
ultimately
evolved
into
this
thesis.

36

Contributions
of
the
Authors

Jeremy
A.
Fishel
designed
and
developed
experimental
prototype,
conducted
the

experiments,
analyzed
the
data
and
generated
the
figures.
Dr.
Gerald
E.
Loeb

provided
guidance
on
the
development
of
experiments
and
data
analysis.
Both

Jeremy
A.
Fishel
and
Dr.
Gerald
E.
Loeb
wrote
the
manuscript.
Veronica
Santos

assisted
with
developing
the
experimental
protocol
and
edited
the
manuscript.

Abstract

Controlling
grip
force
in
a
prosthetic
or
robotic
hand
requires
detailed
sensory

feedback
information
about
microslips
between
the
artificial
fingertips
and
the

object.
In
the
biological
hand,
this
is
accomplished
with
neural
transducers
capable

of
measuring
microvibrations
in
the
skin
due
to
sliding
friction.
For
prosthetic

tactile
sensors,
emulating
these
biological
transducers
is
a
difficult
challenge
due
to

the
fragility
associated
with
highly
sensitive
devices.
Incorporating
a
pressure

sensor
into
a
fluid-‐filled
fingertip
provides
a
novel
solution
to
this
problem
by

effectively
creating
a
device
similar
to
a
hydrophone,
capable
of
recording

vibrations
from
lateral
movements.
The
fluid
conducts
these
acoustic
signals
well

and
with
little
attenuation,
permitting
the
pressure
sensing
elements
to
be
located

in
a
protected
region
inside
the
core
of
the
sensor
and
removing
them
from
harm’s

way.
Preliminary
studies
demonstrate
that
high
frequency
vibrations
(50-‐400Hz)

can
be
readily
detected
when
such
a
fingertip
slides
across
a
ridged
surface.

37

Introduction

Fine
control
of
grip
force
for
the
human
hand
is
made
possible
largely
by
the

wealth
of
tactile
sensory
information
delivered
to
the
central
nervous
system

(Johansson
&
Westling
1987).
When
making
a
precision
pinch
between
two
fingers,

the
muscles
in
the
human
hand
deliver
just
enough
grip
force
so
that
an
object
does

not
slip
out
of
grasp
(Johansson
&
Westling
1984).
This
desirable
behavior
requires

finely
tuned
sensory
neurons
capable
of
detecting
microslips
between
the
skin
and

the
object
when
the
ratio
of
gripping
to
lateral
forces
at
the
fingertip
approaches
a

critical
threshold
(Westling
&
Johansson
1984).
In
the
biological
hand,
Pacinian

corpuscles
with
frequency
responses
of
60-‐500Hz
(Mountcastle
et
al.
1972)
are

capable
of
measuring
these
vibrations
associated
with
slip
that
can
be
as
small
as
a

micrometer
around
their
center
frequency
of
200Hz
(Johansson
et
al.
1982).
These

neural
signals
delivered
by
these
sensory
cells
also
contain
a
rich
array
of

information
for
detecting
fine
texture
during
exploratory
movements
(Johnson
&

Hsiao
1992).

An
advanced
mechatronic
hand
or
robotic
manipulator
will
need
artificial

sensors
to
replicate
the
sensing
function
of
the
human
finger.
This
has
previously

been
divided
into
two
categories
of
events
to
be
sensed:
those
associated
with
static

forces
(normal
and
tangential
to
the
surface)
and
those
associated
with
the

dynamics
of
sliding
across
a
surface.
For
a
comprehensive
review
of
existing
robotic

tactile
sensing
technologies
see
(Nicholls
&
Lee
1989;
Lee
&
Nicholls
1999;
Howe

1994).
This
paper
investigates
robust
sensing
strategies
for
dynamic
tactile
sensing.

38

Pruski
and
Mutel
(1984)
suggested
that
is
might
be
possible
to
extract
slip

information
from
noise
in
a
resistive
sensor
of
normal
force
but
did
not
present
data

on
sensitivity
or
frequency
response.
Howe
and
Cutkosky
(1989)
took
a
more
direct

approach
to
sensing
vibration
in
compliant
skin
by
attaching
an
accelerometer
but

this
involved
mounting
a
fragile
MEMS
device
where
it
would
be
prone
to
damage.

Similarly,
Dario
et
al.
(1984)
mounted
a
piezoelectric
polymer
(polyvinylidene

fluoride)
near
the
skin
surface,
where
it
would
be
difficult
to
protect
such
a
high

impedance
device
from
ambient
moisture.
Other
variations
using
these
piezoelectric

materials
have
been
investigated
as
dynamic
sensors
(Son
et
al.
1994;
Yamada
&

Cutkosky
1994;
Dario
et
al.
1994),
however
using
fragile
elements
near
contact

surfaces
reduces
the
robustness
of
these
sensors.

While
these
sensors
work
fine
for
specific
applications
in
controlled
laboratory

environments,
they
all
share
the
common
trait
of
requiring
fragile
sensing

mechanisms
to
reside
near
the
contact
surface
of
the
sensor.
Biological
hands
are

afforded
with
the
ability
to
regenerate
damaged
skin
and
receptors,
but
an

alternative
approach
for
engineered
systems
is
to
keep
delicate
sensing
devices
a

safe
distance
from
possible
damage,
while
still
retaining
as
much
sensitivity
as

possible.
In
this
paper
we
propose
a
novel
class
of
tactile
sensors
capable
of

measuring
the
microvibrations
associated
with
slip
at
a
remote
location
that
is

protected
from
potential
damage
from
the
harmful
environments
our
hands

encounter
on
a
daily
basis.
By
taking
advantage
of
the
excellent
propagation
of

39

sound
waves
through
an
incompressible
fluid,
we
have
overcome
the
supposition

that
transduction
must
be
done
near
the
surface
of
a
tactile
sensor
(Howe
1994).

Methods

Designing
a
Robust
Micro-‐Vibration
Sensor

Figure
3-‐1:
Tactile
Sensor
Conceptual
Design

Prototype
for
micro-‐vibration
sensor
is
comprised
of
a
rigid
core
surrounded
by
a
deformable
fluid

contained
within
an
elastomeric
skin.
A
pressure
sensor
is
housed
away
from
the
contact
area

protecting
it
from
damage.
This
is
a
critical
design
feature,
as
many
alternative
approaches
require

delicate
sensors
to
be
placed
on
or
near
the
skin’s
surface.
As
the
sensor
slides
across
a
textured

surface
microvibrations
generated
from
sliding
friction
are
transduced
into
fluid
pressure
vibrations.

Our
design
challenge
was
to
develop
an
easily
repaired,
highly
durable,
yet

sensitive
and
precise
tactile
sensing
device
that
contains
a
replaceable
compliant

skin
with
no
sensing
elements
or
electronic
connections.
Instead
all
sensing

mechanisms
would
remain
on
or
in
a
rigid
core
to
which
the
skin
can
be
attached.
In

40

previous
work,
we
have
developed
a
biomimetic
tactile
array
that
measures
the

distribution
of
applied
forces
by
detecting
deformation
of
a
fluid-‐filled
space
that

functions
mechanically
like
the
finger
pulp
(Figure
3-‐1)
(Wettels,
Santos,
et
al.

2008a).
Sensing
electrodes
distributed
over
the
surface
of
the
core
detect
these

deformations
by
changes
in
the
AC
impedance
with
respect
to
reference
electrodes.

However,
the
fine
microvibrations
due
to
sliding
contact
do
not
produce
enough

fluid
deformation
to
detect
micro-‐slips
with
a
similar
impedance-‐based
approach.

This
paper
describes
the
incorporation
of
an
off-‐the-‐shelf
pressure
transducer
to

detect
vibrations
in
the
fluid
associated
with
texture
and
slip.

The
incompressible,
low-‐viscosity
fluid
is
an
efficient
conductor
of
acoustic

frequency
vibrations
(Figure
3-‐2).
Given
the
long
wavelengths
of
the
relevant

frequencies
(λ
=
3m
@
500Hz),
it
was
possible
to
locate
the
sensitive
MEMS

transducer
and
its
supporting
electronics
away
from
the
skin
and
within
the

protective
core
of
the
sensor.

41

Figure
3-‐2:
Sensing
of
Sliding
Microvibrations
vs.
Control

Sensing
of
sliding
microvibrations
(right)
compared
with
no-‐slip
control
(left).
In
the
no-‐slip
control

case,
the
sensor
was
manually
pressed
down
on
a
rigid
surface
(tabletop)
and
released
after
some

time
showing
an
increase
in
fluid
pressure
during
this
task.
In
the
sliding
case,
the
sensor
was

pressed
down
on
the
same
surface
and
slid
across
while
maintaining
a
constant
downward
force.

High
frequency
spectral
energy
was
readily
observed
in
this
task
and
has
been
attributed
to

microvibrations
generated
from
sliding.

Fabrication,
Inflation,
and
Leak
Detection

For
this
study,
a
simplified
prototype
was
constructed
with
the
pressure
sensor

system
and
without
the
impedance-‐
sensing
system.
First,
a
negative
mold
of
the

rigid
core
was
machined
out
of
jeweler’s
wax.
The
mold
was
prepared
by
routing

silicone
tubing
from
the
palmar
aspect
of
the
fingertip
to
the
dorsal
aspect
where
it

was
attached
to
a
3-‐
way
stopcock
that
protruded
from
the
cavity.
The
mold
was

then
filled
with
dental
acrylic
(Hygenic,
Perm
reline/repair
resin).
Once
the
acrylic

core
had
hardened,
the
silicone
tubing
was
removed,
leaving
a
patent
fluid
pathway.

42

A
pressure
sensor
(Honeywell,
#40PC015G1A)
and
a
10mL
syringe
were
attached
to

the
remaining
ports
of
the
3-‐way
stopcock.
The
elastomeric
skin
(Smooth-‐On

#PMC744,
45
durometer)
was
dip-‐coated
on
the
finger
surface
and
allowed
to
cure.

To
achieve
the
adhesion
of
the
skin
to
the
nail
bed,
mold
release
(polyvinyl
alcohol)

was
applied
to
all
surfaces
of
the
skin
except
this
area
where
sufficient
bonding

occurred.
Once
cured,
the
finger
was
carefully
filled
with
distilled
water
and
air

bubbles
were
removed
by
systematically
applying
pressure
and
suction
with
the

syringe.
The
fluid
was
returned
to
atmospheric
pressure
once
all
air
bubbles
had

been
purged.

To
characterize
the
relationship
between
fluid
pressure
and
volume,
hydrostatic

pressure
was
measured
as
distilled
water
was
injected
in
0.2mL
increments
up
to

1.0mL
for
5
trials.
To
ensure
that
no
leaks
were
present
in
the
system
the
finger
was

filled
with
0.6mL
of
fluid
and
allowed
to
rest
for
three
hours.
It
was
concluded
that

the
sensor
was
adequately
sealed
after
verifying
that
there
was
no
detectable

change
in
pressure
during
this
time.

Data
Acquisition
and
Signal
Processing

A
bandwidth
of
1000Hz
was
concluded
to
be
more
than
adequate
based
on
the

observed
signal
response
of
the
sensor
during
a
variety
of
tasks
using
an

oscilloscope
(unpublished).
Additionally,
this
was
well
beyond
the
biological

vibration
sensitivity
of
the
human
finger
(Mountcastle
et
al.
1972),
suggesting
that

frequencies
beyond
these
limits
do
not
provide
useful
biomimetic
tactile

43

information.
To
eliminate
potential
aliasing
from
unexpected
interference,
data
was

collected
using
a
first-‐order,
analog
low-‐pass
filter
with
center
frequency
at
1000Hz

and
a
digital
sampling
rate
of
2500Hz
(Nyquist
frequency:
1200Hz).
Normal
and

tangential
forces
applied
to
the
finger
were
also
recorded
from
a
6-‐axis
force
plate

(Advanced
Mechanical
Technology,
Inc.,
Model
HE6X6-‐16).
Data
were
digitized

(National
Instruments,
USB-‐6218
with
LabVIEW
8.0)
and
analyzed
offline

(MathWorks,
MATLAB
and
Signal
Processing
Toolbox).

Response
to
Sliding
Motion
over
a
Controlled
Surface

To
determine
the
signal
response
to
sliding
motion,
the
sensor
was
stroked

manually
over
a
ridged
surface
(3mm
regular
grating)
with
approximately
constant

downward
force
and
constant
tangential
velocity
(Figure
3-‐3).
The
pressure

response
was
analyzed
with
a
Short
Time
Fourier
Transform
(STFT)
and
presented

as
a
spectrogram
vs.
time.
A
time
window
of
80ms
was
selected
to
mimic
biological

grip
reflex
response
times
(Johansson
&
Westling
1987).
To
simulate
a
real-‐time

processing
environment,
this
window
was
shifted
such
that
the
spectrogram

presented
at
a
given
time
was
generated
from
the
previous
80ms
of
data.

The
narrow
time
window
selected
for
the
STFT
resulted
in
a
trade-‐off,
sacrificing

frequency
resolution,
an
inherit
problem
with
using
discrete-‐time
frequency

transforms.
As
a
result,
high-‐intensity,
low-‐frequency
spectral
content
leaked
into

many
bands
of
the
STFT,
overshadowing
any
useful
spectral
information
associated

44

with
slip.
To
combat
this,
a
first-‐order,
5Hz,
high-‐pass
filter
was
used
to
mitigate
the

STFT
spectral
leakage.

The
moment
of
slip
is
denoted
by
the
transition
in
the
tangential
force
from
a

ramp
to
a
plateau.
The
ramp
results
from
the
isovelocity
deformation
of
the
skin

while
it
is
locked
in
place
by
static
friction.
After
slip
occurs,
the
tangential
force

depends
on
the
constant
normal
force
and
dynamic
friction.
Velocity
was
estimated

by
observing
the
length
of
time
to
traverse
the
9
cm
long
test
surface.

Figure
3-‐3:
Procedure
for
Manually-‐Controlled
Sliding
Experiments

Sensor
is
slid
over
a
surface
while
downward
force
and
sliding
velocity
are
controlled
manually.

Normal
and
tangential
forces
are
recorded
from
a
force
plate
simultaneously
with
the
pressure
signal.

Multiple
trials
are
taken
with
variations
in
downward
force
and
velocity.

45

Results

Pressure
Sensor
Response
to
Inflation

The
relationship
between
hydrostatic
pressure
and
fluid
volume
was
consistent

over
the
five
trials
(Figure
3-‐4).
This
repeatability
over
multiple
trials
suggested
that

the
fluid
pathways
were
adequately
sealed.

Figure
3-‐4:
Pressure
vs.
Inflation
Volume

Pressure
vs.
Volume
curve
as
fluid
is
added
to
the
sensor.
Experiment
was
repeated
five
times
with

average
values
being
displayed,
error
bars
represent
the
standard
deviation
over
these
trials.

46

Spectral
Features
of
Contact
and
Slip

The
spectral
density
plot
in
Figure
5
shows
the
general
features
associated
with

all
test
movements.
There
is
an
initial
pressure
spike
at
first
contact
with
a
broad

spectrum
followed
by
a
relatively
silent
period
as
tangential
force
develops
during

the
stroking
motion.
The
moment
of
slip
is
denoted
by
the
vertical
dashed
line,

determined
as
above.
Vibrations
from
the
skin’s
surface
had
increased
high

frequency
spectral
energy
as
the
finger
transitioned
into
a
sliding
state.
There
is
a

moment
of
incipient
slip
at
this
transition
which
produces
the
weak
high
frequency

spectrum
just
after
the
dashed
line
(delayed
by
the
80ms
sampling
window).
During

the
steady
sliding
phase,
there
are
gross
rhythmic
fluctuations
in
the
force
and

pressure
signals
associated
with
the
ridges
of
the
textured
surface
and
a
rich
set
of

high
frequency
harmonics.

47

Figure
3-‐5:
Demonstration
of
Sliding
Vibration
Sensation

As
the
finger
makes
contact
with
a
surface,
tangential
force
begins
loading
until
it
reaches
the
plateau

at
which
it
begins
to
slip.
The
red
dashed
line
indicates
the
time
of
slippage.
High-‐frequency
spectral

power
from
30-‐200Hz
appears
to
be
a
reliable
indicator
of
sliding
while
loading
and
unloading

regions
are
indicated
by
increases
in
all
frequency
bands.

Effects
of
Force
and
Velocity

Surprisingly,
the
spectrograms
appeared
to
be
similar
in
distribution,
albeit

increasing
in
general
intensity,
for
higher
forces
and
velocities
(Figures
3-‐6
&
3-‐7).

48

Spectrograms
resulting
from
similar
sliding
movements
over
differently
textured

surfaces
appeared
to
have
distinctive
patterns,
but
this
has
not
been
analyzed

systematically.

Figure
3-‐6:
Response
to
Variations
in
Force

Increasing
downward
force
while
sliding
in
subsequent
trials.
Downward
force
was
determined
by

plateau
height
of
normal
force
for
these
four
trials
while
attempting
to
maintain
a
consistent
sliding

velocity.
While
increasing
downward
force
increases
the
general
spectral
intensity
no
obvious
shifts

in
spectral
content
were
observed.

49

Figure
3-‐7:
Response
to
Variations
in
Velocity

Increasing
sliding
velocity
in
subsequent
trials.
Velocity
(red
arrows)
was
estimated
by
the
amount
of

time
(black
arrows)
taken
to
traverse
a
surface
of
constant
length.
Throughout
these
trials

downward
force
was
kept
constant.
Surprisingly,
changes
in
velocity
did
not
produce
noticeable

shifts
in
spectral
content
indicating
that
this
frequency
band
is
a
strong
indicator
of
slip
regardless
of

sliding
velocity.

50

Discussion

Relationship
Between
Fluid
Pressure
and
Volume

Having
a
defined
fluid
pressure
to
volume
relationship
for
a
sensor
of
this
type

allows
for
a
quick
and
accurate
measurement
of
fluid
volume.
It
is
envisioned
that

the
DC
pressure
component
of
the
signal
can
be
used
to
detect
leakage
and

determine
if
the
skin
is
in
need
of
replacement.
This
illustrates
one
advantage
of

using
a
piezoresistive
pressure
sensor
as
opposed
to
a
piezoelectric
pressure
sensor,

which
cannot
detect
resting
pressures.

Dynamic
Response
of
the
Sensor

As
predicted
from
sound
conduction
in
a
fluid,
the
remote
pressure
sensor
is

highly
sensitive
to
microvibrations
of
the
skin.
This
suggests
an
opportunity
to

optimize
the
shape
and
mechanical
properties
of
this
skin
so
that
it
generates
more

distinctive
patterns
of
such
microvibrations
for
the
events
and
objects
to
be

discriminated.
In
the
current
prototype,
the
outer
surface
of
the
skin
is
smooth,

whereas
biological
fingertips
have
regular
patterns
of
ridges
with
0.3-‐0.5mm

spacing.
The
mechanical
interactions
between
these
ridges
and
variously
textured

objects
presumably
give
rise
to
temporospatial
patterns
of
microvibration
that
are

sensed
by
the
high
frequency
mechanoreceptors
of
the
glabrous
skin.
It
is
not

difficult
to
create
such
textures
by
molding
elastomers
against
surfaces
machined

into
a
negative
of
the
desired
texture
(Wettels,
Smith,
et
al.
2008b).
The
generation

51

and
propagation
of
such
microvibrations
across
a
prosthetic
skin
and
through
the

fluid
to
our
pressure
transducer
is
likely
to
depend
on
the
size
and
spacing
of
these

molded
surface
features
and
the
thickness
and
viscoelasticity
of
the
polymer
from

which
the
skin
is
formed.

While
these
results
provide
promising
preliminary
findings,
additional

experiments
are
currently
underway
to
characterize
the
ability
of
this
design
to

detect
incipient
slip.
The
microvibrations
associated
with
the
release
of
individual

skin
ridges
are
likely
to
be
smaller
and
less
coherent
than
those
associated
with

continuous
sliding
over
a
textured
surface.

Applications

Slip
Detection
for
Grip
Control

When
slid
over
a
textured
surface,
our
biomimetic
sensor
provides
robust,
high-‐
frequency,
spectral
output.
This
sensory
information
should
be
useful
for
detecting

lateral
slips
in
a
timely
manner
such
that
object-‐stabilizing
grip
adjustments
can
be

generated
similar
to
spinal
reflexes
in
the
biological
hand
(Westling
&
Johansson

1984).
With
this
approach,
grip
force
can
be
increased
at
the
onset
of
slip,
making

possible
advanced
prosthetics
that
behave
more
like
the
natural
hand
and
are
thus

easier
to
control
via
slower
visual
feedback.

It
should
be
noted
that
signal
processing
parameters,
such
as
the
duration
of
the

STFT
window
(and
thus
increased
response
time),
will
affect
the
spectral
analysis

results.
Reducing
this
response
time
window
will
reduce
the
resolution
in
the

52

frequency
domain,
making
it
difficult
to
distinguish
high-‐frequency
spectral
content

from
low-‐
frequency
spectral
content,
but
making
it
faster
to
detect
the
onset
of
the

weak
vibrations
associated
with
incipient
slip.

Autonomous
Texture
Discrimination

We
observed
(unpublished
results)
that
skin-‐surface
interactions
give
rise
to

distinctive
“spectral
signatures”
for
this
sensor.
When
sliding
our
prototype
over

different
surfaces
and
then
replaying
the
signals
(which
are
in
the
audible
band)

through
a
speaker
it
was
possible
to
distinguish
different
surfaces
by
their
tone,

much
like
the
human
finger
can
distinguish
textures
by
touch.
Further

experimentation
into
how
this
can
be
optimized
is
currently
underway.

For
the
purposes
of
texture
discrimination
in
robotic
applications,
the
problem

becomes
more
difficult
without
the
spectral
analyzing
capabilities
of
the
human

brain.
However,
it
is
proposed
that
a
robotic
manipulator
with
the
goal
of
texture

discrimination
could
make
slow
exploratory
movements
similar
to
biological

exploratory
movements.
These
slower
movements
do
not
require
the
same
rapid

response
time
as
grip
reflexes,
therefore
a
wide-‐windowed
STFT
can
be
used
for

high
spectral
resolution
which
is
likely
to
play
a
key
role
in
artificial
texture

recognition.

Conscious
Haptic
Feedback

It
should
be
possible
to
drive
a
mechanical
vibrator
with
the
waveforms
detected

by
our
pressure
transducer.
The
vibrator
could
be
located
on
the
intact
skin
on
the

53

residual
limb
of
an
amputee
or
incorporated
into
the
manipulandum
of
a

teleoperated
robot.
The
signal
processing
would
then
be
performed
by
the
central

nervous
system.

Conclusions

This
initial
report
suggests
that
hydrophonic
sensing
can
provide
a
rich
source
of

information
about
tactile
events
that
are
important
for
the
identification
and

manipulation
of
everyday
objects.
Additional
experiments
are
underway
to

investigate
the
effects
of
the
design
and
mechanical
properties
of
the
elastomeric

skin
on
the
generation
of
pressure
waves
and
their
transduction
by
the
remote

sensor.
The
general
design
of
the
finger
should
provide
a
relatively
inexpensive,

robust
and
easily
repaired
appliance
for
use
in
a
wide
range
of
environments.

54

Chapter
4:
A
Robust
Tactile
Sensor

Jeremy
A.
Fishel,
Chia-‐Hsien
Lin,
Raymond
Peck
and
Gerald
E.
Loeb

Preface

Early
BioTac
prototypes
as
discussed
in
Chapter
3
were
fabricated
using
very

simplistic
methods
that
limited
their
reproducibility
and
function.
Fluid
pathways

would
be
placed
into
an
open-‐faced
mold
that
was
then
filled
with
dental
acrylic
to

create
the
bone-‐like
core
of
the
sensor.
The
resulting
part
had
sharp
edges
on
the

ventral
surface
that
need
to
be
removed
with
a
file.
Skins
were
then
fashioned
by

dip-‐coating
the
final
core
in
silicone,
which
resulted
in
an
uneven
and
inconsistent

skin
thickness
between
parts.
Sealing
the
skin
to
the
core
and
inflating
with
fluid

was
also
a
difficult
challenge
that
worsened
over
the
life
of
the
prototype.
As
a
result

many
sensors
built
using
these
methods
did
not
last
longer
than
a
few
days
of

experimentation
before
needing
to
be
rebuilt
from
scratch.
To
remedy
this,

improvements
were
necessary
to
improve
the
reliability,
repeatability
and
function

of
the
device.

The
refinement
of
the
device
took
many
years
and
design
iterations,
the
result
of

which
was
the
BioTac,
a
tactile
sensor
now
commercially
distributed
by
SynTouch,

LLC.
The
core
of
the
redesigned
sensor
had
a
biomimetic
shape
and
integrated
all

55

electronics,
sensory
transducers
and
digital-‐to-‐analog
converters
in
a
small
package

similar
in
size
to
the
two
distal
phalanxes
of
the
human
fingertip.
Sensory
function

was
designed
to
mimic
all
three
sensory
modalities
found
in
human
cutaneous
skin

(force,
vibration,
temperature)
in
a
single
package.
The
skins
were
fabricated

separately
in
a
three-‐part
mold.
In
the
result
of
skin
wear
or
damage
the
skin
of
the

BioTac
could
be
easily
replaced
with
a
new
one.
Fluidic
seals
were
integrated
into

the
skin
to
allow
for
inflation
and
sealing
of
the
skin
to
the
core
without
additional

reinforcement.
A
fingerprint-‐like
ridge
pattern
was
added
to
the
outer
surface
of
the

skin;
this
was
found
to
greatly
improve
the
sensitivity
to
vibrations.

Contributions
of
the
Authors

Jeremy
A.
Fishel
led
the
entire
redesign
process
of
the
BioTac
and
is
responsible

for
the
shape
and
engineering
analysis.
Jeremy
designed
and
fabricated
all
of
the

molds
and
developed
the
injection
process
for
the
core
and
skin.
He
developed
the

fluidic
seals
and
assisted
with
the
design
of
the
flexible
circuit
to
fit
into
the
mold.

Gary
Lin
is
responsible
for
the
design
and
programming
of
the
electronics
in
the

flexible
circuit.
Raymond
Peck
provided
guidance
and
advisement
over
the
initial

fabrication
of
BioTac
prototypes
and
was
the
lead
effort
on
perfecting
the
final

fabrication
process.
Gerald
E.
Loeb
provided
guidance
and
advisement
over
the

entire
project.

56

Abstract

An
artificial
tactile
sensor
that
mimics
the
human
fingertip
in
both
multimodal

sensory
function
and
compliance
has
been
developed.
This
device,
known
as
the

BioTac,
can
detect
localized
deformations,
forces,
temperature
gradients,
and

microvibrations
similar
to
the
human
finger.
The
design
consists
of
a
rigid
bone-‐like

core,
which
houses
all
electronic
components,
and
an
elastomeric
skin
covering
that

is
textured
with
fingerprint-‐like
ridges.
The
space
between
the
skin
and
the
core
is

inflated
with
an
incompressible
liquid,
resulting
in
compliance
that
is
remarkably

similar
to
a
human
fingertip.
Microvibrations
that
arise
as
the
skin
slides
over
a

surface
are
sensed
by
a
pressure
transducer
in
the
core
that
acts
like
a
hydrophone,

similar
in
sensitivity
and
receptive
field
size
to
Pacinian
corpuscles.
Liquid
is
an

excellent
acoustic
conductor;
allowing
the
sensory
transducer
to
be
safely
housed

within
the
rigid
core
where
it
is
protected
from
damage.
The
addition
of
a

fingerprint-‐like
ridge
texture
resulted
in
a
profound
increase
in
these
vibrations.

The
Need
for
Robust
Tactile
Sensors

In
the
biological
hand,
Pacinian
corpuscles
with
frequency
responses
of
60-‐
700Hz
(Mountcastle
et
al.
1972)
are
capable
of
measuring
vibrations
associated

with
slip
that
can
be
less
than
a
micrometer
around
their
center
frequency
of
200-‐
250Hz
(Westling
&
Johansson
1987;
Brisben
et
al.
1999).
This
plays
an
integral
role

in
slip
detection
for
the
control
of
grip
(Johansson
&
Westling
1984;
Westling
&

Johansson
1984;
Johansson
&
Westling
1987;
Westling
&
Johansson
1987;
Macefield

57

et
al.
1996;
Srinivasan
et
al.
1990),
detection
of
vibrations
through
objects
held
in

the
hand
(Brisben
et
al.
1999),
as
well
as
the
perception
of
fine
textures
(Lamb

1983b;
Hollins
&
Risner
2000;
Hollins
et
al.
2001;
Bensmaïa
&
Hollins
2005).

Artificial
systems
seeking
to
obtain
these
capabilities
would
benefit
from

transducers
that
have
similar
sensory
performance,
structure
and
compliance
as
the

human
finger.
This
presents
a
difficult
challenge
in
developing
a
robust
tactile

sensor
with
the
compliance
and
vibration
sensitivity
of
the
human
fingertip:
the

sensor
must
be
capable
of
detecting
vibrations
smaller
than
a
micrometer,
yet

robust
enough
to
deform
a
few
millimeters
without
damage.

Many
other
groups
have
explored
dynamic
tactile
sensing.
Pruski
and
Mutel

(1984)
suggested
that
it
might
be
possible
to
extract
slip
information
from
noise
in
a

resistive
sensor
of
normal
force
but
did
not
present
data
on
sensitivity
or
frequency

response.
Howe
and
Cutkosky
(1989)
took
a
more
direct
approach
to
sensing

vibration
in
compliant
skin
by
attaching
an
accelerometer,
but
this
involved

mounting
a
fragile
MEMS
device
where
it
would
be
prone
to
damage.
Similarly,

Dario
et
al.
(1984)
mounted
a
piezoelectric
polymer
(PVDF)
near
the
skin
surface,

where
it
would
be
difficult
to
protect
such
a
high
impedance
device
from
ambient

moisture
and
damage.
Other
variations
using
these
piezoelectric
materials
have

been
investigated
as
dynamic
sensors
(Gao
et
al.
1990;
Howe
&
Cutkosky
1993;

Kolesar
&
Dyson
1995;
Dargahi
et
al.
2000;
Krishna
&
Rajanna
2004;
Hosoda
et
al.

2006;
Mirbagheri
et
al.
2007;
Cotton
et
al.
2007;
Chuang
et
al.
2008).
Various

alternatives
have
also
been
explored
using
acoustic
resonance
(Shinoda
et
al.
1997),

58

high-‐frequency
capacitive
sensors
(Ko
et
al.
2006;
Schmidt
et
al.
2006),
magnetics

(Hasegawa
et
al.
2006),
or
optics
(Dubey
&
Crowder
2006;
Kamata
et
al.
2005).

While
these
sensors
generally
perform
well
in
controlled
laboratory

environments,
they
all
share
the
common
trait
of
requiring
fragile
sensing

mechanisms
to
reside
near
the
contact
surface
of
the
sensor.
Furthermore,
those

systems
that
embody
a
compliant
sensor
design
are
even
more
susceptible
to

damage.
This
is
due
to
the
high
stress
concentrations
that
arise
as
two
dissimilar

medias
deform
together.
These
stress
concentrations
can
lead
to
delaminations
that

inhibit
transduction
of
forces
and
vibrations,
or
failure
of
the
tiny
wires
commonly

used
in
these
sensors.
A
novel
solution
to
improve
robustness
in
a
tactile
sensor

consists
of
coupling
these
mechanical
vibrations
with
a
fluid
(Fishel
et
al.
2008).

The
BioTac
(Figure
4-‐1)
was
designed
to
meet
this
need
for
robustness
and

sensitivity.
It
consists
of
a
rigid
core
that
contains
all
sensory
transducers
for
the

detection
of
forces,
vibrations
and
temperatures,
mimicking
the
full
cutaneous

sensory
suite
of
the
human
fingertip.
An
elastomeric
skin
covers
the
core.
The
space

between
the
skin
and
the
core
is
inflated
with
an
incompressible
liquid
to
give
it

compliance
similar
to
human
fingerpads.
No
sensory
transducers
are
contained
in

the
skin
making
the
design
robust
to
grit,
moisture
or
other
damage
that
typically

plague
other
tactile
sensors.
The
use
of
a
fluid
provides
substantial
advantages
over

solid
materials,
which
are
vulnerable
to
wear,
damage,
and
stress
concentrations
as

they
deform.

59

Figure
4-‐1:
Conceptual
Diagram
of
the
BioTac

The
BioTac
consists
of
a
rigid
core
housing
all
sensory
electronics
and
data
acquisition.
The
core
is

surrounding
by
an
elastic
skin
and
the
space
between
the
skin
and
the
core
is
inflated
to
give
it

compliance
similar
to
the
human
fingertip.

The
BioTac
consists
of
three
complimentary
sensory
modalities
(force,
vibration

and
temperature)
that
have
been
integrated
into
a
single
package:

• As forces are applied to the skin, the skin and fluid deform. Changes in
impedance as the fluid deforms are detected by an array of electrodes on
the surface of the core (Wettels, Santos, et al. 2008a; Wettels & Loeb
2011).
• As objects slide across the surface of the BioTac, they generate vibrations
that are detected by a pressure transducer inside the core (Fishel et al.
2008; Fishel et al. 2012).

60

• As objects of different thermal conductivity come into contact with the
core, the heat that flows from the heated BioTac into the object produces
thermal gradients that are detected as a change in temperature of the
thermistor in the BioTac tip (Lin et al. 2009).
The
BioTac
Design

Integrated
Electronics

The
complete
signal
processing
and
analog
to
digital
circuitry
for
the
sensory

modalities
of
the
BioTac
is
embodied
in
a
flexible
circuit
that
is
molded
into
the
core.

A
conceptual
schematic
of
the
electronics
is
provided
in
Figure
4-‐3.
The
BioTac

contains
three
classes
of
sensors:
impedance
sensing
electrodes
to
measure
force

and
deformation,
static
and
dynamic
fluid
pressure
to
measure
light
forces
and

vibrations,
and
temperature
and
thermal
flux
sensing.

61

Figure
4-‐2:
Conceptual
Schematic
of
the
BioTac
Electronics

Both
pressure
and
temperature
circuits
contain
a
first
stage
DC
gain
and
a
second
stage
high-‐pass
AC

gain.
These
four
signals
are
connected
directly
to
the
microcontroller.
Electrodes
are
multiplexed.

Referring
to
Figure
4-‐3,
the
19
platinum
sensing-‐electrodes
are
connected
to
the

four
excitation-‐electrodes
(four
similar
electrodes
distributed
around
the
fingertip)

62

through
the
conductive
fluid
of
the
BioTac.
As
the
skin
and
fluid
of
the
BioTac

deform,
the
measured
impedance
between
these
electrodes
also
changes.
When

sampling
an
electrode
it
is
first
selected
by
the
multiplexor
(Analog
Devices,

#ADG732)
prior
to
delivering
a
stimulation
pulse
(driven
by
an
AC-‐coupled,
3.3V

4kHz
clock).
The
output
of
the
multiplexor
is
connected
to
Rload
(10kΩ)
and
the

internal
analog-‐to-‐digital
converter
in
the
PIC
microcontroller
(Microchip

Technology
Inc.,
#dsPIC33FJ128GP802),
which
measures
the
peak
voltage
produced

by
the
current
passing
through
the
fluid
path
from
the
excitation-‐electrodes
to
the

enabled
sensing-‐electrode.
The
resulting
voltage
is
digitized
with
12bit
resolution

(En:
0-‐4095).
The
impedance
of
the
fluid
path
and
electrodes
can
be
calculated
with

the
following
formula:

(4.1)

Fluid
pressure
is
measured
with
an
off-‐the-‐shelf
MEMS
fluidic
pressure
sensor

(Honeywell
24PCSMT15).
Contacting
the
skin
of
the
BioTac
or
sliding
the
sensor

over
textured
surfaces
produces
changes
in
fluid
pressure.
The
transducer
contains

a
piezo-‐resistive
bridge
that
responds
to
pressure
with
a
sensitivity
of
2.18mV/kPa

(15mV/psi)
at
the
10V
supply
voltage
(Vcc).
The
output
of
the
transducer
is
biased

positive
to
ensure
the
operational
amplifier
(AD8624,
Analog
Devices)
doesn't

saturate
over
the
range
of
pressures
the
transducer
may
encounter.
This
signal
is

amplified
with
a
gain
of
10
to
produce
the
fluid
pressure
signal
(PDC)
and
furthered

impedance
n
=
3.3V
V
n
−1
"
#
$
%
&
'
10kΩ=
4095bits
E
n
−1
"
#
$
%
&
'
10kΩ

63

amplified
with
a
gain
of
99.1
with
a
high-‐pass
filter
(10Hz)
to
produce
the
vibration

signal
(PAC).
AC
Pressure
is
biased
positive
to
half
of
the
sampling
reference
voltage

to
ensure
the
entire
signal
is
captured,
without
saturation,
in
the
0-‐3.3V
analog-‐to-‐
digital
converters
of
the
PIC.
Both
signals
are
digitized
with
12bit
resolution.
The

pressure
of
these
two
signals
can
be
calculated
from
the
following
equations:

(4.2)

vibration= P
AC
−offset
( )
0.37
Pa
bit

(4.3)

Temperature
is
measured
with
a
thermistor
(Vishay,
NTCS0805E3474FXT)
in
a

voltage
divider
configuration
with
reference
to
a
30kΩ
resistor
and
a
10V
supply.

The
internal
electronics
of
the
BioTac
contains
four
heating-‐resistors
to
raise
the

temperature
of
the
BioTac
10°C
above
ambient.
Contacting
objects
at
room

temperature
produces
thermal
gradients
that
can
be
detected
by
this
circuitry.
The

output
of
the
voltage
divider
circuit
is
buffered
to
provide
the
absolute
temperature

signal
(TDC)
and
high-‐pass
filtered
(5Hz)
with
a
further
gain
of
98
to
produce
the

thermal
flux
signal
(TAC).
The
signals
are
digitized
with
12bit
resolution.
The

temperature
of
these
two
signals
can
be
calculated
with
the
following
equations:

(4.4)

fluid pressure= P
DC
−offset
( )
0.0365
kPa
bit
temperature=
4025
ln
155183−46555
T
DC
4095bits
T
DC
4095bits
"
#
$
$
$
%
&
'
'
'
°C+273.15°C

64

(4.5)

Upon
receiving
a
sampling
command
from
the
host
that
specifies
the
channel
to

sample
the
PIC
microcontroller
will
sample
that
channel
and
return
the

measurement.
Digital
communication
follows
serial
peripheral
interface
(SPI)
bus

protocol,
permitting
for
multiple
BioTacs
sharing
these
lines
to
be
sampled
at
the

same
time.
The
recommended
sampling
sequence
to
take
advantage
of
the
available

bandwidth
of
the
sensory
channels
(Table
4-‐1)
is
to
interleave
the
sampling
of
the

high-‐bandwidth
PAC
with
the
other
channels
(E1-‐E19,
PDC,
TAC,
TDC)
at
a
rate
of
4.4kHz

or
greater.

Sensory

Modality

Symbol
Range
Resolution
Frequency

Response

Impedance
En
0
-‐
3.3V
3.2
mV
0
-‐
100
Hz

Fluid
Pressure
PDC
0
-‐
100
kPa
36.5
Pa
0
-‐
1040
Hz

Microvibration
PAC
+/-‐0.76
kPa
0.37
Pa
10
-‐
1040
Hz

Temperature
TDC
0
-‐
75
C
0.1
C
0
–
22.6
Hz

Thermal
Flux
TAC
0
-‐
1
C/s
0.001
C/s
0.45
–
22.6
Hz

Table
4-‐1:
BioTac
Sensory
Modality
Details

Table
4-‐1
specifies
details
of
the
range,
resolution
and
frequency
response
of
the
sensory
modalities

available
in
the
BioTac.

Size
and
Shape

To
achieve
the
sensory
performance
and
function
of
the
human
fingertip
the

design
of
the
BioTac
mimics
the
human
fingertip
in
many
aspects,
including
its
size

dynamictemperature=
41.07
ln
155183−46555
T
DC
4095bits
T
DC
4095bits
"
#
$
$
$
%
&
'
'
'
°C+273.15°C

65

and
shape.
Careful
measurements
were
taken
of
a
human
fingertip
both
unloaded

and
when
compressed
to
drive
the
dimensions
of
the
BioTac.
The
inflated
shape,

compliance
and
deformed
shape
were
analyzed
using
finite
element
modeling

software
(COMSOL)
to
verify
the
similarity.
Other
research
investigating
the

compliance
of
fluid-‐filled
membranes
have
also
proposed
the
similarity
between
the

human
fingertip
and
a
fluid-‐filled
design
(Srinivasan
1989).

Fluidic
Seals

One
of
the
leading
design
requirements
of
the
BioTAC
was
to
have
a
skin
that

was
simple
to
replace
and
inflate
with
fluid
in
the
event
of
damage
or
wear.
This

presented
challenges
in
sealing
the
skin
at:
the
point
of
insertion
onto
the
core,
the

location
of
the
screws
holding
the
nail,
and
at
the
point
of
inflation.
Seals
for
each
of

these
points
were
designed
and
integrated
directly
into
the
geometry
of
the
skin
to

avoid
leaks
(Figure
4-‐2).

66

Figure
4-‐3:
Fluidic
Seals
of
the
BioTac

The
nail
screw
seal,
collar
seal
and
inflation
seal
comprise
the
three
fluid
seals
of
the
BioTac.
Black

arrows
indicate
the
direction
of
motion
that
creates
the
seal
and
red
arrows
indicate
resulting
sealing

force.
Left:
When
installing
the
nail,
the
screws
compress
the
excess
silicone,
creating
a
compression

seal
between
the
nail
and
the
skin
that
prevents
fluid
leaks.
Bottom:
The
collar
seal
was
designed
to

stretch
over
the
core
during
insertion
while
still
providing
residual
compression
to
prevent
fluid

leaks
when
resting
in
a
groove
around
the
neck
of
the
BioTac.
The
initial
diameter
of
the
seal
was

designed
to
ensure
it
did
not
exceed
80%
of
the
rated
maximum
elongation
of
the
silicone
when

stretching
over
the
core
during
installation.
Analytical
calculations
of
the
resulting
compression

between
the
skin
and
the
groove
were
found
to
be
sufficient
to
prevent
fluid
leaks
under
these

conditions.
Right:
A
self-‐sealing
inflation
port
was
integrated
into
the
skin
and
core
to
allow
inflation

using
a
hypodermic
needle,
similar
to
a
multiple
use
drug
vial.
To
achieve
the
seal
a
conical
extrusion

of
the
inner
surface
mates
with
a
slightly
smaller
conical
hole.
When
attached
to
the
core
with
screws,

the
nail
compresses
the
skin
into
the
hole
and
provides
a
guide
hole
through
which
the
needle
can
be

inserted.
After
the
syringe
is
removed
the
compression
forces
on
the
skin
reseal
holes
created
by
the

nail.

67

Fabricating
a
BioTac

To
create
the
skin
and
core
of
the
BioTac,
a
3-‐part
mold
for
both
parts
were

designed
using
commercially
available
computer
aided
design
(CAD)
packages

(SolidWorks,
Dassault
Systems).
Toolpaths,
for
machining
these
pieces
on
a

computer
numerical
controlled
(CNC)
mill,
were
generated
using
commercially

available
computer-‐aided
machining
(CAM)
packages
(Mill,
Level
3,
MasterCAM).

With
the
exception
of
the
mandrel
(as
discussed
below),
all
parts
were
machined
in-‐
house
on
a
3-‐axis
CNC
milling
machine
(DynaMite
2800,
Dyna
Metronics).
The
nails

for
the
BioTac
were
designed
using
CAD
and
rapid
prototyped
from
a
nylon-‐like

material
(QuickParts,
Atlanta,
GA).

The
Core

Molds
for
the
core
(Figure
4-‐4)
include
a
bottom
half
and
a
top
section
that
can

be
split
into
two
parts
to
permit
for
the
exit
of
the
flexible
circuit
connector.
Molds

were
machined
out
of
aluminum
and
treated
with
a
specialized
Teflon
anodizing

process
(K&L
Anodizing,
Burbank,
CA)
to
ease
part
release
and
improve
the
life
of

the
mold.
A
gasket
was
designed
to
seal
the
top
halves
of
the
mold
(Figure
4-‐4).

68

Figure
4-‐4:
Molds
to
Produce
the
Biotac
Core

Upper:
flexible
circuit
electronics
are
placed
in
the
bottom
half
of
the
mold.
Lower:
the
mold

assembled
with
the
top-‐back
mold
piece
and
gasket
to
seal
the
flexible
circuit.
The
electrodes
are

positioned
with
a
custom
SLS
part
to
hold
the
electrodes
to
the
surface
of
the
core
before
molding.

To
fabricate
the
core
of
the
BioTac
the
cleaned
molds
are
first
coated
with

polyvinyl
acetate
(PVA),
which
serves
as
a
mold
release.
The
flexible
electronic

circuit
is
prepared
with
all
appropriate
tubes
and
positioners
and
placed
inside
the

mold
of
the
core.
Two
tubes
for
the
epoxy
injection
and
ejection
port
are
connected

69

to
the
mold
before
it
is
sealed
with
bolts.
Using
a
vacuum-‐injection
process
the
core

is
infused
with
a
transparent,
low-‐viscosity,
high-‐strength
and
impact-‐resistant

epoxy
(Emerson
&
Cuming,
Stycast
1264).
After
mixing
two
parts
of
the
epoxy,
it
is

de-‐aired
in
a
vacuum
chamber
to
remove
bubbles
and
dissolved
air,
which

otherwise
would
escape
during
the
injection
process
to
create
voids
in
the
final
part.

When
completely
de-‐aired
the
epoxy
is
vented
to
atmospheric
pressure
and

connected
to
the
injection
port
of
the
mold.
The
molds
are
then
placed
under

vacuum
(25mmHg)
to
remove
air
from
the
cavity.
A
valve
on
the
injection
port
is

carefully
opened
to
slowly
pull
the
epoxy
into
the
mold.
Upon
completion
of
the

injection
process
the
bubble-‐free
epoxy
can
be
observed
leaving
the
exit
tube.
The

injection
process
is
stopped
by
venting
the
vacuum
chamber
to
atmosphere
and

eliminating
the
differential
pressure.
The
mold
is
removed
from
the
vacuum

chamber
and
allowed
to
cure
for
72
hours
before
removing
the
part.
The
remove

part
is
subjected
to
an
additional
24-‐hour
post-‐curing
phase
at
60°C
to
optimize
the

material
properties
of
the
epoxy.

The
Skin

The
top
and
bottom
halves
of
the
skin
molds
were
machined
out
of
acrylic

(Figure
4-‐5).
This
allows
for
visibility
during
the
injection
process
of
the
silicone

skins
that
do
not
adhere
to
the
acrylic.
A
custom
mandrel
for
creating
the
textured

inner-‐surface
of
the
skin
(which
improves
the
range
of
impedance
sensing
(Wettels,

70

Smith,
et
al.
2008b))
was
machined
from
aluminum
using
a
five-‐axis
milling
machine

(5
th

Axis
Machining,
San
Diego,
CA).

Figure
4-‐5:
Molds
to
Produce
the
BioTac
Skin

Upper:
Top
and
bottom
halves
of
the
skin
mold.
Posts
are
present
on
the
top
half
of
the
mold
to

produce
the
negative
feature
of
holes
for
installing
the
nail
onto
the
core.
Bottom:
The
assembled

skin
mold
with
mandrel
inside.

The
skin
is
fabricated
using
a
similar
injection
process
as
the
core,
but
instead

molded
with
a
silicone
(Silastic
S,
Dow
Corning).
After
injection,
the
skins
are
cured

71

in
an
oven
at
60°C
for
4
hours
before
the
parts
are
removed.
They
are
then
post-‐
cured
for
an
additional
24
hours
in
the
oven
to
ensure
complete
curing.
Flashing
is

removed
with
a
scalpel
and
the
skins
are
inspected
to
ensure
they
are
free
from

defects.

Assembly

Figure
4-‐6:
Skin
Installation
Procedure

Step
1:
Lubricate
the
core
of
the
BioTac
with
BioTac
fluid.
Step
2:
Insert
skin
onto
the
core
of
the

BioTac,
taking
care
to
ensure
there
is
an
adequate
seal
around
the
neck.
Step
3:
Install
the
nail
onto

the
core
of
the
BioTac.
Step
4:
Inflate
the
BioTac
with
approximately
0.5cc
of
BioTac
fluid.
Step
5:

remove
air
bubbles
by
pinching
the
skin
around
the
neck.
Step
6:
Verify
the
correct
inflation
height
of

the
BioTac
as
15.1mm,
add
or
remove
fluid
if
necessary.

1
4 3 5 6
2

72

The
skin
can
be
assembled
to
the
BioTac
(Figure
4-‐6)
by
first
lubricating
the
core

with
the
BioTac
fluid
(82%
Polyethylene
Glycol
and
18%
distilled
water
that
is

mixed
with
NaBr
to
produce
a
1M
solution),
then
inserting
the
skin
onto
the
core.

The
skin
is
held
in
place
by
the
nail
with
two
small
nylon
screws.
The
BioTac
can

then
be
inflated
with
fluid
by
inserting
a
syringe
with
stopper
into
the
inflation
port.

Trapped
air
is
removed
by
squeezing
the
collar
seal
until
all
air
bubbles
are
release.

Once
properly
inflated
the
distance
from
the
top
of
the
nail

to
the
ventral
portion
of

the
skin
should
be
15.1mm
(Figure
4-‐6).
The
assembled
an
inflated
BioTac
is
shown

in
Figure
4-‐7.

Figure
4-‐7:
Assembled
BioTac

Photograph
of
an
assembled
BioTac.

73

Fingerprints

The
addition
of
fingerprints
was
hypothesized
to
play
a
critical
role
in
the

transduction
of
vibrations
as
the
BioTac
slides
over
textured
surfaces.
This
is

proposed
to
result
in
a
coherent
pattern
of
stick-‐slip
behavior
that
amplifies

individual
vibrations
by
abrupt
release
of
the
fingerprints
when
their
loading
from

tangential
force
reaches
the
limits
of
static
friction
(Fishel
et
al.
2009).

To
test
this,
two
prototype
skins
were
created,
one
with
a
smooth
skin
and
the

other
with
a
textured
pattern
that
resembled
dermal
ridges
(Figure
4-‐8).
The
ridges

were
fabricated
by
machining
a
negative
pattern
in
the
mold
with
a
pattern
from
a

1/64”
ball
end
mill.
The
resulting
features
were
human-‐sized
fingerprint-‐like

patterns
with
a
width
and
spacing
of
0.4mm
and
a
height
of
0.2mm.

74

Figure
4-‐8:
Fingerprint
Pattern
of
the
BioTac
Skin

Fingerprint
pattern
of
the
BioTac
skin
under
magnification.
Pattern
spacing
is
0.4mm.

Vibration
data
were
collected
to
compare
a
smooth
skin
and
one
with
the

fingerprint-‐like
pattern
when
sliding
over
60-‐grit
sandpaper.
Contact
force
and

velocity
were
manually
controlled
and
were
estimated
by
the
experimenter
to
be

approximately
0.5N
and
1cm/s.
When
compared
to
a
smooth
skin,
the
addition
of
a

fingerprint-‐like
texture
pattern
to
the
external
surface
of
the
elastomeric
skin

resulted
in
a
profound
increase
in
the
amplitude
and
complexity
of
the
vibration

spectra
(Figure
4-‐9).

This
was
found
to
be
true
for
a
range
of
materials
(silk,
suede,

sandpaper)
over
different
exploratory
conditions
(Figure
4-‐10).

75

Figure
4-‐9:
Sliding
Vibrations
from
Smooth
and
Fingerprinted
Skin

Acoustic
spectra
recorded
from
AC
pressure
signal
contrasting
smooth
skin
to
skins
with
human-‐
sized
fingerprints
when
slid
over
a
textured
surface
(sandpaper,
60
grit),
at
experimenter
controlled

speeds
and
forces
(estimated
to
be
~1cm/s
and
~0.5N).
Results
indicate
substantial
increase
in

signal
amplitude
and
complexity
with
the
addition
of
fingerprints.

76

Figure
4-‐10:
STFTs
of
Smooth
and
Fingerprint
Skin
over
Different
Materials

Short-‐Time
Fourier
Transforms
(STFT)
for
smooth
and
fingerprinted
skin
when
sliding
over
a

surface
at
different
forces
(speed:
approximately
1cm/s).
The
x-‐ordinate
is
time
in
seconds
and
the
y-‐
ordinate
is
frequency
in
Hz.
The
color
scale
represents
frequency
domain
spectral
components.
Silk,

suede
and
sandpaper
textures
for
both
smooth
and
fingerprinted
skins
are
compared.
Three
different

force
levels
are
presented
(Low:
~0.5N,
Med:
~1N,
High:
~5N).

Discussion

Using
the
above
processes,
we
are
able
to
create
a
robust
fluid-‐based
tactile

sensor
that
contains
all
of
the
sensory
modalities
of
human
cutaneous
touch.
The

process
improvement
over
earlier
prototypes
permits
for
a
more
consistent

performance
between
sensors
and
permits
for
the
simple
replacement
of
skins
if

they
become
damaged.
Furthermore,
fingerprint-‐like
patterns
on
the
skin
were

shown
to
enhance
sensed
vibrations
and
have
been
integrated
into
the
design.

77

Chapter
5:
Sensing
Tactile
Microvibrations

Jeremy
A.
Fishel
and
Gerald
E.
Loeb

©
2012
IEEE.
Reprinted
with
permission
from:

Fishel,
J.A.,
Loeb
G.E.,
Sensing
Tactile
Microvibrations
with
the
BioTac
–

Comparison
with
Human
Sensitivity.
In
Proceedings
of
the
IEEE
4th
Biennial

IEEE/RAS-‐EMBS
International
Conference
on
Biomedical
Robotics
and

Biomechatronics,
2012.

Preface

This
chapter
presents
the
characterization
and
performance
of
the
vibration

sensing
modality
of
the
BioTac.
Results
are
presented
that
compare
the
sensitivity
of

humans
and
the
BioTac
to
applied
vibrations
and
small
impact
forces.
The
results
of

this
study
demonstrate
the
BioTac
as
a
promising
research
tool
that
achieves

sensitivity
to
vibrations
that
exceeds
even
human
performance.

Contributions
of
the
Authors

Jeremy
A.
Fishel
designed
and
developed
the
experimental
protocol
conducted
in

this
study,
carried
out
the
experiments,
analyzed
the
data,
and
generated
the
figures.

Gerald
E.
Loeb
provided
guidance
in
the
study.
Both
authors
wrote
the
manuscript.

78

Acknowledgements

The
authors
would
like
to
thank
Gary
Lin
for
the
layout
and
design
of
the
flexible

circuit
and
Raymond
Peck
for
the
fabrication
of
the
BioTac
sensors
used
in
these

tests.

Abstract

The
human
fingertip
is
exquisitely
sensitive
to
vibrations
that
are
essential
to

detect
slip
and
discriminate
textures.
Achieving
similar
functions
with
prosthetic

and
robotic
hands
will
require
tactile
sensors
with
similar
sensitivity.
Many

technologies
have
been
developed
to
sense
such
vibrations,
yet
none
have
achieved

the
requisite
sensitivity
in
a
package
that
is
robust
enough
to
meet
practical

applications.
The
BioTac®,
developed
by
the
authors,
uses
an
incompressible
liquid

as
an
acoustic
conductor
to
convey
vibrations
from
the
skin
to
a
wide
bandwidth

pressure
transducer
located
deep
in
the
rigid
core
of
the
mechatronic
finger,
where

it
is
protected
from
damage.
Signal
conditioning
electronics
were
designed
to

achieve
sensitivity
down
to
the
theoretical
noise
floor
of
the
transducer,
making
the

device
very
sensitive
to
the
smallest
of
vibrations,
even
sound.
We
demonstrate
here

that
this
device
exceeds
human
performance
in
detecting
sustained
vibrations

(capable
of
sensing
vibrations
as
small
as
a
few
nanometers
at
~330Hz)
as
well
as

very
small
transient
events
that
arise
when
small
particles
are
dropped
on
the
finger.

This
overcomes
the
supposition
that
such
sensitivity
requires
fragile
sensory

elements
to
reside
near
the
vulnerable
contact
surfaces.

79

Introduction

Humans
have
exquisite
sensitivity
to
mechanical
vibrations
in
the
skin.
Pacinian

corpuscles
with
frequency
responses
of
60-‐700Hz
(Mountcastle
et
al.
1972)
are

capable
of
measuring
vibrations
associated
with
slip
that
can
be
less
than
a

micrometer
around
their
center
frequency
of
200
Hz
(Westling
&
Johansson
1987;

Brisben
et
al.
1999).
The
sensitivity
to
such
vibrations
plays
an
integral
role
in
many

important
tasks
such
as
slip
detection
for
the
control
of
grip
(Johansson
&
Westling

1987;
Macefield
et
al.
1996;
Srinivasan
et
al.
1990)
and
the
perception
of
fine

textures(Lamb
1983b;
Hollins
et
al.
2001;
Bensmaïa
&
Hollins
2005).
It
has
been

demonstrated
that
eliminating
tactile
sensitivity
with
nerve
block
renders
the

human
hand
almost
useless
in
even
the
simplest
of
tasks
(Johansson
&
Westling

1984).
Artificial
systems
seeking
to
obtain
these
capabilities
will
require
tactile

sensors
with
performance
similar
to
that
found
in
the
human
finger;
otherwise
they

are
likely
to
be
severely
limited,
similarly
to
the
anesthetized
human
hand.

Development
of
artificial
tactile
sensors
is
not
new
and
there
are
many
reviews

that
cover
their
evolution
over
the
years
(Nicholls
&
Lee
1989;
Howe
1994;
Lee
&

Nicholls
1999;
Dahiya
et
al.
2010).
Most
of
the
attention
has
focused
specifically
on

the
sensing
of
high-‐resolution
normal
forces,
which
only
represents
a
small
subset

of
human
touch.
Human
fingertips
are
also
sensitive
to
shear
forces
and
skin
stretch,

temperature
and
thermal
fluxes,
as
well
as
vibration.
Such
multimodal
sensitivity
is

necessary
to
achieve
the
dexterity
and
performance
of
the
human
hand
(L.
A.
Jones

&
Lederman
2006).

80

Developing
a
tactile
sensor
with
similar
form
and
function
as
the
human
finger

presents
a
unique
design
challenge,
particularly
for
the
sensing
of
vibrations.
The

sensor
must
be
sensitive
enough
to
detect
vibrations
smaller
than
a
micrometer
yet

robust
enough
to
deform
a
few
millimeters
without
damage.
Many
groups
have

explored
dynamic
tactile
sensing
with
a
variety
of
sensory
technologies
such
as

accelerometers
(Howe
&
Cutkosky
1989),
piezoelectric
polymers
(Dario
et
al.
1984;

Son
et
al.
1994;
Yamada
&
Cutkosky
1994;
Dario
et
al.
1994),
magneto-‐inductance

(Vranish
1986),
and
ultrasonic
technologies
(Hutchings
et
al.
1994;
Ando
&
Shinoda

1995).
While
these
sensors
work
fine
for
specific
applications
in
controlled

laboratory
environments,
they
all
share
the
common
trait
of
requiring
fragile

sensing
mechanisms
to
reside
near
the
contact
surface
of
the
sensor.
Some

developers
have
even
concluded
that
this
was
a
requirement
for
such
sensory

capabilities
(Howe
1994).
Both
mechatronic
and
human
fingers
are
frequently

exposed
to
conditions
where
they
can
be
damaged,
but
biological
appendages

possess
the
ability
to
regenerate
damaged
skin
and
tactile
receptors
therein.
An

alternative
approach
for
engineered
systems
is
to
keep
delicate
sensing
devices
a

safe
distance
from
possible
damage,
while
still
retaining
as
much
sensitivity
as

possible.

The
BioTac
(Figure
5-‐1)
was
designed
to
meet
this
need
for
robustness
and

sensitivity.
It
consists
of
a
rigid
core
that
contains
all
sensory
transducers,
covered

by
an
elastomeric
skin.
The
space
between
the
skin
and
the
core
is
inflated
with
an

incompressible
and
conductive
fluid
to
give
it
a
compliance
that
mimics
human

81

fingerpads.
No
sensory
transducers
or
wires
are
contained
in
the
skin
making
the

design
robust
to
grit,
moisture
and
physical
damage
to
sensory
mechanisms
that

typically
plague
other
tactile
sensors.
If
the
skin
of
the
BioTac
becomes
damaged
or

worn,
it
can
be
easily
replaced.

Figure
5-‐1:
The
BioTac
Conceptual
Diagram
and
Picture

A)
Cross-‐sectional
schematic
of
the
BioTac.
B)
Photograph
of
an
assembled
BioTac.
Fingerprint-‐like

ridges
can
be
seen
on
the
ventral
surface.

82

The
BioTac
consists
of
three
complimentary
sensory
modalities
(force,
vibration

and
temperature)
that
have
been
integrated
into
a
single
package.
Contact
forces

distort
the
elastic
skin
and
underlying
conductive
fluid,
changing
impedances
of

electrodes
distributed
over
the
surface
of
the
rigid
core
(Wettels
&
Loeb
2011;

Wettels,
Smith,
et
al.
2008b;
Wettels,
Santos,
et
al.
2008a).
Vibrations
in
the
skin

propagate
through
the
fluid
and
are
detected
as
AC
signals
by
the
hydro-‐acoustic

pressure
sensor
(Fishel
et
al.
2008).
Temperature
and
heat
flow
are
transduced
by
a

thermistor
near
the
surface
of
the
rigid
core
(Lin
et
al.
2009).
This
paper
describes

the
design
and
performance
of
the
vibration
sensing
modality
of
the
BioTac.

Methods

Physics
of
Fluidic
Vibration
Sensing

Preliminary
studies
demonstrated
the
feasibility
of
sensing
slip-‐related

microvibrations
in
a
liquid-‐filled
fingertip
by
measuring
the
dynamic
pressure
of
the

liquid
(Fishel
et
al.
2008).
The
incompressibility
of
the
liquid
and
long
wavelengths

at
the
frequencies
of
interest
(λ
=
3m
at
500Hz
in
water)
allow
for
the
sensing

elements
to
be
moved
away
from
the
contact
region
where
it
would
be
susceptible

to
damage.
This
permits
for
a
class
of
highly
robust
and
simple
tactile
vibration

sensitivity
that
has
been
integrated
into
the
multimodal
sensory
suite
of
the
BioTac.

The
tube
that
connects
the
liquid
under
the
skin
to
the
pressure
transducer
(i.e.

hydrophone)
in
the
core
also
has
a
resonant
frequency
(f),
which
can
be
calculated

83

from
the
media’s
speed
of
sound
(v)
and
the
length
of
the
tube
(L)
by
the
acoustic

resonance
formula
of
a
closed
tube:

(5.1)

For
a
2cm
tube
filled
with
saltwater
(v
=
1497m/s),
this
resonant
frequency
is

18.7kHz.
The
desired
bandwidth
to
mimic
biological
performance
is
1kHz
so
the

resonance
of
a
2cm
tube
is
substantially
high
enough
to
be
ignored.
Another
factor

to
consider
is
the
inertial
damping
of
the
fluid.
Increasing
the
length
of
the
tube,

viscosity
or
compressibility
of
the
fluid
increases
the
attenuation
of
pressure
signals,

particularly
at
higher
frequencies.
Using
a
low
viscosity
liquid
such
as
water
and

keeping
the
fluid
tube
as
short
as
possible
can
mitigate
these
effects.

Fabrication
of
the
BioTac

The
BioTac
core
includes
a
flexible
circuit
that
contains
all
of
the
sensory

electronics,
signal
conditioning
circuits
and
microcontroller
(Lin
et
al.
2009).
The

flexible
circuit
is
loaded
into
a
three-‐part
mold
and
filled
with
an
epoxy
(Stycast

1264,
Emerson
&
Cuming)
to
produce
the
core
of
the
BioTac.
Skins
are
molded
from

a
silicone
(Silastic
S,
Dow
Corning)
in
a
three-‐part
mold
that
has
features
to
mold

artificial
ridges
similar
to
fingerprints.
The
fingernail
is
machined
from
acrylic.
Once

the
parts
are
assembled
(Figure
5-‐1B),
the
BioTac
is
filled
with
a
solution
that

minimizes
changes
in
volume
due
to
diffusion
through
the
skin
(82%
Polyethylene

Glycol
and
18%
distilled
water
that
is
mixed
with
NaBr
to
produce
a
1M
solution).

f =v 4L

84

Care
is
taken
to
ensure
there
are
no
bubbles
inside
the
fingertip
after
filling
with

fluid.
The
salinity
of
this
mixture
provides
the
conductivity
required
for
the

impedance
sensing
modality
of
the
BioTac
(Wettels,
Santos,
et
al.
2008a).
This

solution
was
indistinguishable
from
pure
water
with
regard
to
the
vibration
sensing

modality
discussed
in
this
paper.

Electronics
Design
and
Noise
Analysis

The
BioTac
contains
all
necessary
electronics
for
signal
conditioning,
analog
to

digital
conversion
and
serial
transmission
of
data
from
all
sensors.
The
signal

conditioning
electronics
were
redesigned
to
detect
signals
as
close
as
possible
the

theoretical
noise
floor
of
the
piezoresistive
transducer.

Thermal
Noise

Electronic
noise
arises
in
all
conductors
regardless
of
applied
voltage
because

the
charge
carriers
inside
a
conductor
vibrate
stochastically
due
to
thermal
energy.

The
root-‐mean-‐square
(RMS)
amplitude
of
the
noise
can
be
calculated
from
the

Boltzmann
constant
(kb),
the
temperature
(T),
the
resistive
load
(R),
and
the

bandwidth
(Δf)
from
the
following
equation:

(5.2)

Thermal
noise
was
minimized
by
low-‐pass
analog
filtering
with
a
1040Hz
cutoff

frequency
and
selecting
a
pressure
transducer
(Honeywell
24PC15SMT)
with
low

V
thermal
= 4k
b
T RΔf

85

output
resistance.
An
alternative
sensor
(Honeywell
26PC15SMT)
offered

temperature
compensation
to
reduce
low-‐frequency
drift
with
changing

temperature
but
had
a
much
larger
output
resistance
and
would
have
increased
this

noise
source.
The
selected
transducer
operates
as
a
piezoresistive
bridge
with
an

output
resistance
of
5kΩ
(Figure
5-‐2A).
The
BioTac
is
heated
to
about
35C
for

thermal
flux
measurements
(Lin
et
al.
2009).
The
thermal
noise
is
thus
0.29µVrms.

Figure
5-‐2:
Signal
Conditioning
for
Pressure
Transducer

A)
Electrical
schematic
of
the
piezoresistive
strain
bridge
used
in
the
pressure
transducer.
The

transducer
includes
two
fixed
resistors
and
two
variable
resistors
that
change
in
response
to
applied

pressure.
B)
Conceptual
diagram
of
amplifier
circuit
used
to
create
DC
pressure
and
AC
pressure

signals.

The
output
of
the
transducer
in
response
to
pressure
is
specified
as

0.218mV/V·kPa
(1.5mV/V·psi)
and
increases
as
the
supply
voltage
to
the
bridge

increases.
Given
that
the
thermal
noise
stays
constant,
a
very
simple
trick
to

improve
the
signal-‐to-‐noise
ratio
is
to
increase
the
supply
voltage.
While
most

electronics
included
in
the
BioTac
are
powered
at
3.3V,
we
added
a
10V
regulated

signal
for
the
transducer.
The
output
of
the
transducer
with
a
10V
supply
is

86

2.18mV/kPa
(15mV/psi).
The
thermal
noise
of
0.29µV
would
then
be
the
equivalent

of
0.133Pa
(0.0193
×10-‐3psi).

Signal
Conditioning
and
Amplifier
Noise

The
transducer
output
is
amplified
with
a
gain
of
10
and
a
low-‐pass
anti-‐aliasing

filter
at
1040Hz
to
produce
a
DC
pressure
signal
(Figure
5-‐2B).
A
second
stage
with

a
band-‐pass
filter
of
10-‐1040Hz
and
an
additional
gain
of
99.1
produces
the
high-‐
resolution
vibration
signal
(AC
pressure).
Both
the
DC
and
AC
pressures
are

sampled
by
a
microcontroller
inside
the
BioTac
with
12-‐bit
resolution
for
the
range

0-‐3.3V.
This
accommodates
the
full
range
of
transducer
output
(103.4kPa
=
15psi)

with
a
resolution
of
36.5Pa
(5.3x10-‐3psi)
at
the
DC
stage
and
+/-‐758Pa
(0.110psi)

range
with
a
0.37Pa
(0.054x10-‐3psi)
resolution
at
the
AC
stage.

First
stage
amplifier
noise
tends
to
be
a
significant
source
of
noise
so
a
low-‐noise

amplifier
was
selected
(AD8624,
Analog
Devices).
For
1024Hz
bandwidth,
it

generates
0.39µV
of
voltage
noise
referred
to
input.
The
device
also
generates
4.8pA

of
current
noise,
but
with
a
5kΩ
load
this
source
is
comparatively
negligible

(~0.025µV).
The
second
stage
amplifier
has
similar
input
noise,
but
the
gain
of
10

on
the
first
stage
amplifier
makes
the
noise
contribution
of
the
second
stage

negligible
when
referred
to
input.

Capacitively
Coupled
Extrinsic
Noise

Any
extrinsic
AC
signal
can
contribute
noise
by
capacitive
coupling
into
the

conductors
comprising
the
measurement
circuitry.
One
common
source
is
the
60Hz

87

electrical
power
lines.
In
our
system
the
primary
source
of
this
noise
is
the
high-‐
speed
digital
communication
lines
operating
at
up
to
10MHz.
To
reduce
stray

capacitance,
care
was
taken
to
redesign
the
circuitry
to
shorten
leads
of
low
voltage

un-‐amplified
signals
and
move
them
away
from
digital
communication
lines.
To

further
reduce
interference,
the
microcontroller
was
configured
to
sample
the

analog
inputs
at
times
when
the
digital
communication
between
the
BioTac
and
the

host
was
silent.

Total
Noise
Calculation
and
Validation

Computed
noise
is
summarized
in
Table
5-‐I.
The
total
RMS
noise
of
0.49µV
was

computed
as
the
root
sum
of
squares
of
thermal
and
input
noise.
This
does
not

include
extrinsic
sources
of
noise
such
as
stray
capacitance
or
power
supply
noise.

Based
on
the
sensitivity
of
the
transducer
this
amount
of
noise
would
be
equivalent

to
0.224Pa
(0.033
×10-‐3psi)
of
pressure
or
0.6
of
the
least-‐significant-‐bit
(lsb)
value

of
digitized
AC
pressure
(the
lsb
value
is
hereafter
referred
to
as
“bits”).

Noise
Source
RMS
Noise
Notes

Thermal
Noise
0.29µV
Inherent
in
transducer

Amplifier
Input
Noise
0.39µV
Noise
of
first
stage
amplifier

Stray
Capacitance
-‐
Not
considered

Total
Noise
0.49µV

Table
5-‐1:
Summary
of
Theoretical
Noise
Sources

Total
theoretical
RMS
noise
is
found
to
be
0.49µV
when
considering
the
thermal
noise
of
the
sensor

and
amplifier
input
noise.
Stray
capacitance
was
not
considered,
but
was
minimized
by
using
short

leads.

88

Data
Acquisition

The
BioTac
uses
SPI
protocol
for
digital
communication.
A
high-‐speed
SPI/USB

device
(Cheetah,
Total
Phase)
was
used
to
command
the
BioTac
to
sample
the
AC

and
DC
pressure
signals
at
2.2kS/s
(kilosamples
per
second)
each
for
a
total
of

4.4kS/s.
Software
(LabVIEW,
National
Instruments)
was
developed
to
communicate

with
the
SPI
controller
and
queue
up
batches
of
asynchronous
sampling
commands

in
100ms
blocks
of
data
that
were
transferred
to
the
host
through
USB
as
they

became
available
at
the
controller.
Data
were
saved
to
text
files
and
analyzed
offline

in
MATLAB
(MathWorks).

Signal-‐to-‐Noise
Estimation

Background
RMS
noise
and
signal
power
were
calculated
as
the
standard

deviation
and
variance
of
the
AC
pressure
signal
when
the
BioTac
was
at
rest
and

without
any
vibration
disturbances.
Signal-‐to-‐noise
ratio
was
computed
for
common

tasks
with
the
BioTac:
1)
sliding
over
a
textured
piece
of
foam,
2)
making
light
taps,

and
3)
clicking
on
a
computer
mouse.
The
signals
are
presented
as
time
plots.

Comparison
with
Human
Performance

Psychophysical
and
physiological
studies
have
explored
the
sensitivity
of
the

human
fingertip
to
vibrations
(Mountcastle
et
al.
1972;
Johansson
et
al.
1982),
but

they
are
not
necessarily
comparable
to
stimuli
applicable
to
the
BioTac.
We

measured
the
thresholds
for
detection
of
applied
vibrations
and
small
impacts
in

89

five
normal
human
volunteers
who
gave
their
informed
consent
to
participate
in

this
study.
We
compared
these
to
the
detectability
of
the
corresponding
AC
pressure

signals
as
defined
below.

Frequency
Sensitivity

Many
different
devices
exist
to
apply
vibrations
to
human
skin
with
technologies

such
as
eccentric
motors
or
inductive
coils
(Yao
&
Hayward
2010),
but
their
output

amplitudes
are
often
sensitive
to
mechanical
load.
We
used
a
piezoelectric
actuator

(AE0203D16F,
NEC/TOKIN)
that
produces
0.113µm/V
of
applied
voltage
with
a

17.4µm
displacement
for
maximal
applied
drive
of
150V.
The
device
had
a
high

blocking
force
of
200N,
making
it
stiff
enough
to
be
insensitive
to
the
range
of
loads

presented
by
the
BioTac
or
human
finger.
Sinusoidal
signals
were
generated
by
a

function
generator
and
amplified
by
a
commercial
driver
for
piezoelectric
devices

(Piezomaster
VP7206-‐24L105,
Viking
Industrial
Products).
The
input
to
the

actuator
was
monitored
on
an
oscilloscope
to
confirm
the
frequency
and
amplitude.

To
determine
frequency
sensitivity
of
the
biological
finger,
subjects
were
asked

to
place
their
finger
on
the
device
with
any
force
they
chose
and
amplitude
was

varied
over
a
set
number
of
frequencies
between
10
and
900Hz.
At
each
frequency

test,
vibration
amplitudes
were
increased
from
below
the
perceptual
limits
until

subjects
first
reported
the
vibration
sensation.
All
trials
were
repeated
3
times
at

each
frequency
and
the
mean
value
was
determined.
Thresholds
determined
in
this

manner
were
consistent
among
the
repeat
trials
and
similar
among
subjects,
so

90

more
systematic
methods
such
as
Bekesy
tracking
were
not
employed.
The
average

amplitudes
at
threshold
of
detectability
and
the
performance
of
the
most
sensitive

subjects
are
reported
here.

To
test
frequency
sensitivity
of
the
BioTac,
a
1µm
drive
signal
was
used
at

frequencies
between
10
and
900Hz.
The
AC
pressure
signals
were
analyzed
in
the

frequency
domain
with
a
1s
window.
The
frequency
of
maximum
power
was
found

to
coincide
with
the
drive
stimulus
frequency
in
all
tests.
Total
signal
power
was

calculated
from
the
total
energy
in
a
2Hz
band
centered
on
this
peak
frequency.
The

RMS
value
of
AC
pressure
(square
root
of
total
power)
was
found
to
vary

proportionally
with
applied
vibration
amplitude
at
all
frequencies,
confirming
the

linearity
of
vibration
amplitude
to
sensed
vibrations
in
the
system.
This
linearity

made
it
possible
to
determine
the
theoretical
sensitivity
at
different
frequencies

based
on
the
power
estimated
from
a
1µm
vibration.
The
background
noise
power

with
no
vibration
applied
in
the
largest
2Hz
band
was
found
to
be
0.2bits
2
.
The

lowest
detectable
vibration
amplitude
was
calculated
as
the
point
when
the
signal

power
was
double
the
this
background
signal
power
using
the
following
formula:

(5.3)

This
equation
was
tested
at
multiple
frequencies,
including
the
peak
sensitivity,

and
was
found
to
be
accurate
in
all
cases.

d
min
=1µm×
0.4bits
2
P
1µm

91

Impact
Sensitivity

To
determine
the
sensitivity
to
impact,
small
spheres
of
various
sizes
and

densities
were
dropped
on
either
the
BioTac
or
the
human
fingertip
from
a
height
of

7
cm
through
a
guide
hole
drilled
in
a
block
of
acrylic.
Both
the
mass
of
the
ball
and

impact
energy
(estimated
from
potential
energy
before
the
drop)
were
calculated.

Six
different
sized
spheres
were
used
as
outlined
in
Table
5-‐2.
It
was
observed
that

sensitivity
scaled
with
impact
energy
rather
than
mass
as
smaller
masses
were

readily
detected
from
larger

Sphere
Mass
(mg)
Energy
(µJ)

0.25
mm
Solder
Ball
0.072
0.049

0.45
mm
Solder
Ball
0.422
0.289

1
mm
Aluminum
Bearing
1.461
1.002

1
mm
Steel
Bearing
4.089
2.805

3/64
in
Steel
Bearing
6.902
4.734

5/64
in
Aluminum
Bearing
11.415
7.830

Table
5-‐2:
Spheres
Used
for
Impact
Tests

The
mass
and
energy
from
dropping
from
a
7cm
height
is
for
the
6
spheres
used
in
these
studies.

For
assessment
of
human
performance,
subjects
were
asked
to
close
their
eyes

while
holding
their
finger
under
the
guide
hole
and
report
when
they
felt
any

impacts.
Five
or
more
trials
were
completed
for
all
spheres
except
for
some
of
the

larger
spheres
when
it
was
quite
obvious
that
the
subject
had
no
difficulty

identifying
the
impacts.
Results
are
presented
as
percentage
of
correct

identifications
averaged
across
all
subjects.

92

To
assess
the
sensitivity
of
the
BioTac
we
used
an
algorithm
to
count
the
number

of
points
that
deviated
past
four
standard
deviations
of
noise.
A
confirmed
detection

of
impact
was
classified
as
any
event
that
produced
more
than
3
points
outside
of

this
range
within
a
100ms
window.
At
a
sampling
rate
of
2200S/s
the
likelihood
of

false
positives
in
the
100ms
window
was
0.00027%
(once
every
10
hours)
and
was

never
observed
when
the
BioTac
was
at
rest.

Results

Vibration
Signals
During
Common
Tasks

The
BioTac
was
found
to
be
highly
sensitive
to
vibrations
and
transient
signals

during
common
tasks
(Figure
5-‐3).
Background
noise
for
vibration
signal
had
an

RMS
value
of
approximately
1.4bits.
Sliding
over
a
foam
texture
therefore
had
a

signal
to
noise
ratio
of
greater
than
1000.
Signals
such
as
the
click
and
release
of
a

mouse
button
were
readily
observable.
The
BioTac
was
even
found
to
demonstrate

such
sensitivity
that
when
resting
on
a
table
it
could
easily
detect
vibrations
from
a

person
walking
by
as
well
as
loud
acoustic
emissions
such
as
shouting.

93

Figure
5-‐3:
Vibration
Signals
During
Common
Tasks

Signals
from
common
tasks
demonstrate
superior
signal
to
noise
response
(background
noise
is

approximately
1.4bits).

Theoretical
vs.
Actual
Noise

Actual
RMS
noise
of
the
AC
pressure
signal
was
verified
to
be
1.4+/-‐0.1bits,

which
is
slightly
more
than
double
the
theoretical
noise
of
0.6bits
based
on
the

conservative
analysis.
Much
of
this
noise
is
likely
due
to
quantization
noise
of
the

analog-‐to-‐digital
conversion
itself.
This
1.4bits
of
noise
corresponds
to
0.52Pa

(0.078
×10-‐3psi)
in
terms
of
transducer
noise.

94

Comparison
with
Human
Performance

Frequency
Sensitivity

Plots
comparing
amplitudes
of
human
sensitivity
to
BioTac
sensitivity
as
a

function
of
frequency
are
shown
in
Figure
5-‐4.
The
BioTac
had
a
rather
complex

transfer
function
that
varied
slightly
with
inflation
volume
and
the
location
of
the

vibrational
stimuli.
For
all
frequencies
and
conditions,
however,
the
sensitivity
of

the
BioTac
was
better
than
the
human
subjects.

Figure
5-‐4:
Frequency
Sensitivity
of
the
BioTac
and
Human
Subjects

Green
trace
=
average
subject
performance;
purple
trace
=
best
subject
performance;
red
trace
=

BioTac
sensitivity.

95

Impact
Sensitivity

Table
5-‐3
compares
human
to
BioTac
sensitivity
for
correctly
identifying
the

impact
of
various
spheres.
The
BioTac
was
able
to
readily
detect
contact
of
all

spheres
except
the
smallest
(0.25mm
Solder
Ball).
Humans
had
a
great
deal
of

variation
in
performance
with
some
subjects
unable
to
detect
reliably
even
the
1mm

Aluminum
Bearing
(1.461mg).
In
most
cases
subjects
were
occasionally
able
to

detect
the
0.45
Solder
Ball
(0.422
mg)
and
only
one
subject
was
able
to
detect
these

every
time;
the
BioTac
had
a
100%
classification
rate
for
the
same
object.
Signals

from
the
BioTac
(Figure
5-‐5)
resemble
decaying
sinusoids
with
a
resonant

frequency
around
330Hz
and
a
time
constant
of
20ms.

Sphere
Energy
(µJ)

Human

Classification

BioTac

Classification

0.25mm
Solder
Ball
0.049
0%
0%

0.45mm
Solder
Ball
0.289
47.5%
100%

1mm
Aluminum
Bearing
1.002
74.86%
100%

1mm
Steel
Bearing
2.805
100%
100%

3/64in
Steel
Bearing
4.734
100%
100%

5/64
Aluminum
Bearing
7.830
100%
100%

Table
5-‐3:
Summary
of
Performance
for
Impact
Tests

The
summary
of
identification
of
impact
events
for
human
subjects
and
the
BioTac
is
presented

averaged
across
all
trials.

96

Figure
5-‐5:
Sample
Signals
Measured
by
the
BioTac
During
Impact
Tests

Lower
plot
is
zoomed
in
on
the
y-‐axis
to
show
background
noise
for
impact
sensitivity
on
the

0.45mm
solder
ball,
which
was
the
lowest
detectable
stimuli
tested.

Discussion

The
vibration
transmission
and
transduction
system
of
the
BioTac
was
found
to

be
operating
near
the
theoretical
sensitivity
of
the
transducer
and
signal

conditioning
system.
Its
sensitivity
exceeded
that
of
human
fingertips
for
the
full

range
of
sinusoidal
frequencies
for
which
humans
have
tactile
receptors
and
for

impacts
of
tiny,
low-‐mass
objects.
This
contradicts
previous
claims
(Howe
1994)

that
such
sensitivity
to
vibrations
would
require
transducers
near
the
contact

surface,
where
they
would
be
vulnerable
to
damage
during
normal
use.

97

Compared
to
the
strain-‐gage
pressure
transducer
in
the
BioTac,
piezoelectric

transducers
are
known
to
have
far
superior
signal-‐to-‐noise
performance
and
would

likely
produce
even
better
sensitivity
when
coupled
with
transmission
of
vibrations

from
skin
through
an
incompressible
fluid.
They
have
two
major
drawbacks
in
this

application,
however:
1)
they
are
not
suitable
for
determining
resting
fluid
pressure,

and
2)
they
generally
require
more
complex
signal
processing
electronics.

The
effects
of
silicone
skin
thickness
and
hardness
on
vibration
sensitivity
of
the

BioTac
were
examined
anecdotally
in
preliminary
research
and
were
found
to
be

relatively
minor.

Future
Work

The
BioTac
was
designed
to
be
biomimetic,
a
strategy
that
assumes
that
sensing

capabilities
at
least
similar
to
those
of
humans
would
be
necessary
for
mechatronic

systems
to
achieve
human-‐like
haptic
performance.
Preliminary
studies
have

demonstrated
that
the
BioTac
does
actually
generate
rich
patterns
of
vibrations
as
it

contacts
and
slides
over
surfaces,
but
these
patterns
depend
on
the
force
and

velocity
of
the
exploratory
movements
that
are
applied
to
those
surfaces
(Fishel
et

al.
2008).
Thus,
much
work
is
still
required
to
develop
algorithms
to
control
such

movements
and
to
interpret
the
robust
but
complex
sensory
signals
that
the
BioTac

generates.
Because
these
signals
are
similar
in
modalities,
sensitivity
and
dynamic

range
to
those
from
human
fingertips,
it
should
be
possible
to
use
biomimetic
design

to
develop
such
algorithms.
Given
such
algorithms,
it
should
be
possible
for
robots

98

to
achieve
humanlike
capabilities
to
discriminate
and
identify
textures,
detect
slip,

estimate
coefficient
of
friction
and
adjust
stability
of
grip.

99

Chapter
6:
Bayesian
Exploration

Jeremy
A.
Fishel
and
Gerald
E.
Loeb

©
2012
Jeremy
A.
Fishel
and
Gerald
E.
Loeb.
Reprinted
from:

Fishel,
J.A.,
Loeb,
G.E.
Bayesian
exploration
for
intelligent
identification
of

textures.
Frontiers
in
Neurorobotics.
2012.

Preface

This
study
presents
a
novel
exploratory
algorithm
that
makes
use
of
vibrotactile

information
from
the
BioTac
called
Bayesian
exploration.
The
algorithm
is
one
of
the

first
applications
developed
to
take
advantage
of
the
sensory
information
in
the

BioTac
and
was
used
to
test
theories
of
how
humans
use
tactile
information
when

exploring
objects
(Loeb
et
al.
2011).

Contributions
of
the
Authors

Jeremy
A.
Fishel
designed
and
developed
the
experimental
protocol
and

developed
the
mechanism
of
Bayesian
exploration
used
in
this
study.
Gerald
E.
Loeb

provided
the
initial
vision
for
Bayesian
exploration
and
helped
with
the

interpretation
of
results
and
guidance
of
experiments.
Gerald
E.
Loeb
and
Jeremy
A.

Fishel
co-‐authored
the
manuscript.

100

Acknowledgements

The
authors
would
like
to
thank
Raymond
Peck
for
fabrication
of
the
BioTac

sensors
and
apparatus
used
in
these
tests
and
Michelle
Willie
and
Lee
Ann
Lumilan

for
preparation
of
the
texture
samples.
The
authors
would
also
like
to
thank
the

entire
team
at
the
Medical
Device
Development
Facility
at
the
University
of

Southern
California
and
at
SynTouch
LLC,
particularly
Gary
Lin,
Matthew
Borzage,

Nicholas
Wettels,
David
Groves
and
Tomonori
Yamamoto
for
their
advice
and

assistance
on
this
project.
This
research
was
supported
by
the
Department
of

Defense
Advanced
Research
Projects
Agency
contract
D11PC20121.

Abstract:

In
order
to
endow
robots
with
human-‐like
abilities
to
characterize
and
identify

objects,
they
must
be
provided
with
tactile
sensors
and
intelligent
algorithms
to

select,
control
and
interpret
data
from
useful
exploratory
movements.
Humans

make
informed
decisions
on
the
sequence
of
exploratory
movements
that
would

yield
the
most
information
for
the
task,
depending
on
what
the
object
may
be
and

prior
knowledge
of
what
to
expect
from
possible
exploratory
movements.
This

study
is
focused
on
texture
discrimination,
a
subset
of
a
much
larger
group
of

exploratory
movements
and
percepts
that
humans
use
to
discriminate,
characterize,

and
identify
objects.
Using
a
testbed
equipped
with
a
biologically
inspired
tactile

sensor
(the
BioTac),
we
produced
sliding
movements
similar
to
those
that
humans

make
when
exploring
textures.
Measurement
of
tactile
vibrations
and
reaction

101

forces
when
exploring
textures
were
used
to
extract
measures
of
textural
properties

inspired
from
psychophysical
literature
(traction,
roughness,
and
fineness).

Different
combinations
of
normal
force
and
velocity
were
identified
to
be
useful
for

each
of
these
three
properties.
A
total
of
117
textures
were
explored
with
these

three
movements
to
create
a
database
of
prior
experience
to
use
for
identifying
these

same
textures
in
future
encounters.
Performance
of
99.6%
in
correctly

discriminating
pairs
of
similar
textures
was
found
to
exceed
human
capabilities.

Absolute
classification
from
the
entire
set
of
117
textures
generally
required
a
small

number
of
well-‐chosen
exploratory
movements
(median=5)
and
yielded
a
95.4%

success
rate.
The
method
of
Bayesian
exploration
developed
and
tested
in
this
paper

may
generalize
well
to
other
cognitive
problems.

Introduction

"The
tactual
properties
of
our
surroundings
do
not
chatter
at
us
like
their
colors;

they
remain
mute
until
we
make
them
speak...
Eye
movements
do
not
create
color
the

way
finger
movements
create
touch."
–
David
Katz,
The
World
of
Touch
(1925)

Touch,
by
necessity,
is
an
interactive
sense,
unique
from
the
senses
of
vision
and

hearing.
While
we
are
able
to
observe
the
sights
and
sounds
of
our
environment

without
any
physical
interaction,
the
tactual
properties
of
an
object
can
only
be

sensed
by
physical
contact.
When
interacting
with
an
object,
humans
not
only
need

to
interpret
the
tactile
information
they
sense,
they
also
need
to
decide
which
types

of
movements
to
make
in
order
to
produce
these
tactual
percepts.
Artificial
systems

102

will
require
similar
strategies
(Loeb
et
al.
2011).
Experimental
psychologists
have

identified
six
general
types
of
exploratory
movements
that
humans
make
when

tactually
exploring
objects
to
determine
their
properties:
enclosure
to
determine

global
shape
and
volume,
hefting
to
determine
weight,
pressure
to
determine

hardness,
static
contact
to
determine
thermal
properties,
contour
following
to

determine
exact
shape,
and
lateral
sliding
movements
to
determine
surface
texture

(Lederman
&
Klatzky
1987).
Performing
all
of
these
movements
and
their
many

variants
when
identifying
objects
by
touch
may
not
be
practical
or
useful.
Instead,

prior
knowledge
can
be
used
to
intelligently
guide
the
selection
of
which

exploratory
movements
to
make.
In
this
work
we
present
a
novel
methodology
for

selecting
these
optimal
exploratory
movements
called
Bayesian
exploration.
The

process
works
by
using
prior
experience
to
determine
which
of
the
many
possible

exploratory
movements
is
expected
to
produce
the
greatest
distinction
between
the

most
plausible
candidate
objects.
To
simplify
the
analysis,
we
reduced
the
scope
of

the
discrimination
task
to
only
texture
discrimination,
a
modality
for
which
human

strategies
and
capabilities
have
been
well-‐described
in
the
literature.

Early
work
from
David
Katz
provided
some
of
the
first
insights
into
the

psychophysics
of
texture
discrimination.
In
his
studies
he
observed
that
while

coarse
textures
could
be
discriminated
based
on
their
static
contours
by
simply

pressing
down
on
an
object,
fine
textures
instead
required
sliding
motion
in
order
to

generate
vibrations
for
their
discrimination
(Katz
1925).
More
recent
studies
have

supported
that
lateral
movements
and
vibrations
do,
in
fact,
play
a
critical
role
in

103

the
perception
of
fine
textures
(Hollins
&
Risner
2000),
for
which
discrimination
is

impaired
after
vibrotactile
adaptation
(Hollins
et
al.
2001).
The
human
hand
has
a

wealth
of
sensory
receptors
responding
to
mechanical
and
thermal
stimuli
(L.
A.

Jones
&
Lederman
2006).
Pacinian
corpuscles
with
frequency
responses
of
60-‐
700Hz
(Mountcastle
et
al.
1972)
are
capable
of
sensing
vibrations
associated
with

slip
and
texture
that
can
be
less
than
a
micrometer
in
amplitude
at
their

characteristic
frequency
of
200Hz
(Johansson
et
al.
1982;
Brisben
et
al.
1999),

suggesting
that
highly
sensitive
transducers
will
be
required
if
this
capability
is
to

be
reproduced
artificially.

A
great
body
of
research
has
focused
on
the
physical
stimuli
and
perceptual

properties
that
permit
the
identification
and
discrimination
of
textures.
The

perception
of
roughness
has
been
one
of
the
most
studied
of
these
properties.
Early

psychophysical
experiments
attributed
this
to
the
friction
coefficient
between
the

skin
and
object
for
fine
textures
(Katz
1925).
Other
studies
using
coarser
textures

(spatial
periods
greater
than
0.5
mm)
have
proposed
that
spatial
period
and
contact

force,
and
not
friction,
are
correlated
with
perceived
roughness
(Lederman
et
al.

1982).
Smith
et
al.
(2002a)
contested
these
findings,
providing
additional
support

for
the
role
of
friction
and
shear
force
in
the
perception
of
these
coarse
textures.

Further
studies
involving
fine-‐textures
have
indicated
that
the
power
of
vibrations

as
sensed
by
the
Pacinian
corpuscles
could
play
an
integral
role
in
the
perception
of

roughness
(Bensmaïa
&
Hollins
2005).
Early
studies
into
the
perceptive

dimensionality
of
surfaces
have
suggested
that
sticky/slippery,
hard/soft,
and

104

rough/smooth
represent
three
independent
dimensions
of
a
surface
(Hollins
et
al.

1993).
However,
these
studies
only
used
17
surfaces
and
the
relevance
of
these

results
have
been
contested
by
Bergmann
Tiest
and
Kappers
(2006)
who
have

explored
dimensionality
with
a
total
of
125
surfaces.
Findings
from
this
expanded

database
have
suggested
that
there
are
at
least
four
perceptual
dimensions
of

surfaces
(and
likely
more),
although
not
all
could
be
correlated
specifically
with

named
properties.
We
propose
that
some
of
this
dimensionality
could
be
attributed

to
non-‐textural
properties
such
as
compliance
and
thermal
properties
that
might
be

obtained
without
the
need
for
sliding
movements.
With
specific
regards
to
texture

and
sliding
movements,
the
dimensions
of
sticky/slippery,
rough/smooth,
and

coarse/fine
seem
to
be
the
most
salient
descriptions
of
properties
that
make

textures
distinct,
based
on
both
the
descriptive
words
that
people
use
to
describe

textures
and
their
utility
as
demonstrated
by
the
experimental
literature.

In
Katz’s
original
work
(1925)
he
proposed
a
duplex
theory
for
texture

perception,
hypothesizing
that
coarse
textures
can
be
discriminated
spatially
while

fine
textures
are
discriminated
dynamically
through
sensed
vibrations.
This
was

prescient,
as
the
structure
and
function
of
cutaneous
mechanoreceptors
was
then

unknown.
We
now
know
that
vibrations
and
static
pressure
are
sensed
by
separate

populations
of
cutaneous
mechanoreceptors
(Knibestöl
&
Vallbo
1970;
L.
A.
Jones
&

Lederman
2006).
Artificial
tactile
sensors
have
developed
along
similar
lines,

offering
either
high
spatial
resolution
or
high
temporal
bandwidth.
Reviews
of
the

various
tactile
sensing
technologies
over
the
last
30
years
can
be
found
in
(Nicholls

105

&
Lee
1989;
Howe
1994;
Lee
&
Nicholls
1999;
Dahiya
et
al.
2010).
For
dynamic

tactile
sensing
and
texture
discrimination,
a
number
of
technologies
have
been

introduced
implementing
accelerometers
(Howe
&
Cutkosky
1989;
Howe
&

Cutkosky
1993),
piezoelectric
film
(Tada
et
al.
2003)
microphones
(Edwards
et
al.

2008),
and
tri-‐axial
MEMS
force
sensors
(de
Boissieu
et
al.
2009).
More
recently,

many
tactile
sensors
have
introduced
fingerprint-‐like
ridges
(Mukaibo
et
al.
2005;

Oddo
et
al.
2009;
Scheibert
et
al.
2009),
which
have
been
proposed
to
enhance
the

correlation
of
spatial
frequency
of
explored
textures
with
temporal
patterns
in
the

sensed
vibrations.

The
use
of
tactile
sensors
for
artificial
texture
discrimination
has
received
a

great
deal
of
attention
in
recent
years.
Tada
et
al.
(2004)
were
able
to
demonstrate

that
signal
variance
of
two
different
polyvinylidene
difluoride
(PVDF)
films

embedded
at
different
depths
in
a
complaint
sensor
could
be
used
to
distinguish

among
five
different
textures.
Further
development
by
this
group
expanded
this

analysis
to
an
additional
sensory
dimension
using
an
embedded
strain
gage
(Hosoda

et
al.
2006).
Mukaibo
et
al.
(2005)
developed
a
tactile
sensor
with
fingerprints
and

embedded
strain
gauges
to
discriminate
surfaces
based
on
roughness
and
friction.
A

force
sensor
with
an
elastic
covering
developed
by
de
Boissieu
et
al.
(2009)
was

used
with
sliding
movements
to
identify
ten
different
kinds
of
paper
using
two

different
analytical
approaches:
the
first
utilized
differences
in
Fourier
coefficients

in
the
recorded
vibrations
while
the
second
used
more
direct
signal
features
such
as

the
mean,
variance
and
kurtosis
of
the
signals
as
well
as
spectral
properties
in
a

106

neural
network
classifier.
A
similar
approach
was
taken
by
Giguere
and
Dudek

(2011)
using
accelerometers
on
a
rigid
tactile
probe
to
classify
driving
surfaces

based
on
their
means,
variance
and
higher-‐order
moments
in
a
neural
network.

Oddo
et
al.
(2011)
used
a
robotic
finger
producing
a
stereotyped
sliding
movement

to
discriminate
between
three
fine
textured
gratings
ranging
from
400-‐480um

based
on
their
spectral
properties.
Jamali
and
Sammut
(2011)
analyzed
Fourier

components
of
vibrations
measured
from
eight
different
textures
using
a
novel

method
of
majority
voting
to
classify
these
textures
with
a
high
accuracy
of
95%.

This
study
took
advantage
of
multiple
exploratory
movements,
starting
with
three

movements
initially
and
adding
movements
until
80%
or
more
of
these
movements

indicated
a
particular
texture.
With
exception
to
the
last
study
that
explored
three

different
sliding
velocities,
all
of
these
cases
considered
only
a
single
exploratory

movement.
It
has
been
demonstrated
that
multiple
exploratory
movements
as
well

as
multiple
features
can
boost
the
performance
of
such
a
classifier
for
texture

discrimination
(Sinapov
&
Stoytchev
2010).
In
a
subsequent
study
by
the
same

group
20
textures
were
explored
with
a
fingernail-‐like
tactile
sensor
measuring

accelerations
at
five
different
exploratory
movements.
The
frequency
components

from
these
different
movements
were
used
to
obtain
a
classification
performance

increase
from
65%
using
the
best
single
exploratory
movement
to
80%
for
all
five

exploratory
movements
(Sinapov
et
al.
2011).

In
all
of
these
cases,
the
discrimination
of
textures
with
these
artificial
systems
is

for
the
most
part
a
“passive”
exploratory
procedure.
Force
and
speed
are
preset
to

107

some
standard
values
and
a
fixed
movement
(or
sequence
of
movements)
is

executed.
Furthermore,
the
selection
of
signal
processing
measures
is
seemingly

arbitrary
in
many
of
these
studies.
Various
methods
are
attempted
and
those
that

appear
to
yield
the
best
performance
are
used
as
classifier
input.
Instead,
we

propose
that
findings
from
psychophysical
literature
can
guide
the
development
of

more
intuitive
and
useful
signal
measures.

As
reviewed
above,
multiple
signals,
multiple
exploratory
movements,
as
well
as

repeated
trials
can
boost
classification
performance.
However,
as
the
complexity
of

discrimination
tasks
increase
beyond
simply
texture
discrimination
and
more

movements
and
signals
are
added
to
the
repertoire,
the
feasibility
of
doing

everything
(especially
for
multiple
trials)
can
become
highly
impractical
if
not

completely
infeasible.
Instead,
the
task
of
discrimination
can
take
advantage
of
a

hypothesis-‐testing
approach
that
we
have
proposed
humans
likely
use
when

exploring
objects
(Loeb
et
al.
2011).
Each
successive
exploratory
movement
can
be

used
to
reduce
the
set
of
possible
candidates;
this
information
can
be
used
to

determine
the
optimal
next
exploratory
movement
that
would
yield
the
best

discrimination
among
these
most
likely
candidates
rather
than
the
entire
population.

Such
an
approach
would
have
advantages
in
reducing
the
number
of
exploratory

movements
needed
to
classify
a
texture
or
in
the
general
case,
any
object.
Here
we

introduce
a
novel
algorithm
where
the
selection
of
the
exploratory
movement

becomes
a
critical
process
of
the
identification
task.
Using
a
database
of
prior

108

experience,
optimal
exploratory
movements
are
selected
and
executed
to
aid
in
the

discrimination
task.

Material
and
Methods

An
overview
of
Bayesian
exploration
is
presented
first
in
the
context
of
a
texture

discrimination
task,
followed
by
a
description
of
the
BioTac
multimodal
tactile

sensor
and
the
experimental
apparatus
used
to
control
sliding
movements
when

exploring
textures.
The
descriptive
words
humans
use
when
discriminating
textures

are
used
to
define
quantifiable
properties
of
textures,
followed
by
a
method
for

determining
the
most
useful
exploratory
movements
to
estimate
those
properties.

The
final
three
sections
describe
the
classifier
training
over
this
refined
set
of

movements,
the
methods
employed
for
comparing
performance
of
this
classifier
to

human
performance
in
discriminating
pairs
of
similar
textures
and
methods
for

absolute
texture
discrimination
from
a
broad
set
of
117
textures.

Classification
Theory
and
Strategy

Classification
is
a
topic
of
wide
interest
in
artificial
intelligence
and
is
a
subset
of

the
larger
fields
of
pattern
recognition
and
machine
learning.
The
goal
of
a

classification
task
is
to
identify
which
class
or
classes
best
explains
a
set
of

observations.
Many
tools
exist
involving
both
supervised
and
unsupervised
training

(Jain
et
al.
2000).
In
the
majority
of
classification
problems,
inputs
are
given
and
the

109

class
with
the
maximal
posterior
likelihood
determines
the
classification
5
.
This

introduces
a
fundamental
deficiency
in
the
typical
approach
to
classification

problems:
the
decision
must
be
made
with
the
currently
available
information.
To

compensate
for
this
deficiency,
it
is
common
to
collect
as
much
information
as

possible
before
the
classification
is
made.
The
time
and
effort
required
to
produce

each
exploratory
movement
to
collect
tactile
data
suggests
that
this
would
be
highly

inefficient.
Decisions
are
first
required
to
determine
which
exploratory
movements

to
make
before
any
tactile
information
can
be
obtained.
The
selection
of
these

movements
would
benefit
greatly
from
iterative
decision-‐making,
in
which
the

observations
of
previous
movements
are
used
to
identify
the
most
likely
candidates

to
select
the
next
movement
that
is
most
likely
to
disambiguate
them.
Here
we

introduce
a
novel
method
of
texture
discrimination
implementing
these
strategies.

This
method
of
Bayesian
exploration
should
be
generalizable
to
any
identification

task
requiring
such
intelligence.

Bayesian
Inference
For
Discrimination
of
Textures

Bayesian
inference
is
a
widely
implemented
statistical
classification
method

used
to
estimate
the
likely
causes
of
an
observation
after
it
has
occurred.

Considering
a
set
of
textures
(T)
and
the
observable
measurements
that
they

generate
(X)
when
performing
an
exploratory
movement
(M),
we
can
estimate
the

likelihood
that
a
given
texture
had
caused
these
observations
with
Bayes’
rule:

5

Other
methods
may
use
cost
functions
to
reduce
the
occurrence
of
Type-‐I
or
Type-‐II
errors
for

particular
classes
where
such
errors
are
detrimental.

110

(6.1)

Where
Ti
belongs
to
a
set
of
textures
T,
X
is
a
set
of
observable
properties
(which

are
introduced
in
the
later
sections),
Mm
is
a
particular
exploratory
movement
that

gives
rise
to
these
sensed
properties,
and
P(Ti)
represents
the
prior
probability
of

texture
Ti.
P(X,Mm)
is
the
probability
of
observation
X
occurring
given
all
known

causes
from
the
set
T
at
exploratory
movement
Mm
and
can
be
found
by
the
law
of

total
probability:

(6.2)

Substituting
(6.2)
into
(6.1)
yields
a
common
formulation
of
Bayes’
rule:

(6.3)

The
probability
of
a
measurement
occurring
given
a
known
texture
and

exploratory
movement
can
be
estimated
from
its
probability
density
function.
In
the

absence
of
other
evidence,
the
central
limit
theorem
suggests
that
these
values

should
fall
within
a
normally
distributed
probability
density
function
that
can
be

defined
according
to
a
mean
(µ)
and
standard
deviation
(σ):

P T
i
|X,M
m
( )
=
P X |T
i
,M
m
( )
P T
i
( )
P X,M
m
( )
P X,M
m
( )
= P X |T
j
,M
m
( )
P T
j
( )
j
∑
P T
i
|X,M
m
( )
=
P X |T
i
,M
m
( )
P T
i
( )
P X |T
j
,M
m
( )
P T
j
( )
j
∑

111

(6.4)

It
is
important
to
note
that
the
probability
density
function
is
not
a
true

probability,
but
rather
a
density,
and
can
take
on
a
value
greater
than
unity.

However,
this
measure
is
proportional
to
the
actual
probability;
the
unknown

scaling
can
be
ignored
as
it
is
cancelled
out
by
the
denominator
of
(6.3),
which
has

the
same
scaling
factor.
The
formulas
(6.3)
and
(6.4)
can
be
used
to
update
the

posterior
probability
of
a
texture
given
observation
X
from
a
normally
distributed

set
of
expected
observations.
In
practice,
as
evidenced
by
human
performance,

multiple
observations
and
exploratory
movements
need
to
be
made
to
refine
this
to

an
acceptable
level
of
confidence
before
determining
the
most
likely
texture.

Adaptive
Selection
of
Optimal
Exploratory
Movements

We
have
proposed
that
humans
use
a
careful
selection
of
exploratory

movements
to
test
hypotheses
when
exploring
objects
by
touch
(Loeb
et
al.
2011).

Consider
a
simple
example
of
identifying
a
brick
by
touch.
Absent
prior
information

about
the
object,
a
reasonable
first
exploratory
movement
might
be
an
enclosure

movement,
yielding
information
about
the
object’s
size
and
indicating
that
it
is
a

large
rectangular
prism.
Based
on
this
information,
the
examiner
may
then
conclude

that
it
is
either
a
brick
or
a
block
of
wood.
Useful
subsequent
movements
to
extract

the
most
information
between
these
two
objects
would
probably
focus
on
its
mass,

such
as
pushing
or
hefting
the
object.

P X |T
i
,M
m
( )
∝p X |T
i
,M
m
( )
=
1
2πσ
i,m
2
e
−
x−µ
i,m ( )
2
2σ
i,m
2

112

The
process
of
determining
which
exploratory
movement
is
optimal
requires
a

prediction
of
the
perceived
benefit
based
on
prior
experiences.
A
similar

methodology
has
been
presented
by
Rebguns
et
al.
(2011),
where
movements
and

sensing
actions
are
selected
to
reduce
Shannon
entropy.
In
that
study
exploratory

movements
cease
when
there
is
no
perceived
reduction
of
this
entropy,
a
feature
the

authors
refer
to
as
“burying
its
head
in
the
sand”
to
avoid
getting
additional

information
that
might
increase
uncertainty.
While
the
performance
of
this
study

was
quite
impressive,
the
concept
of
additional
information
being
undesirable
is

peculiar.
By
contrast,
our
approach
is
not
to
infer
the
reduction
of
entropy;
instead

we
simply
select
the
movement
that
would
best
discriminate
between
likely
objects.

The
decision
to
make
this
next
movement
or
not
depends
on
whether
the

information
and
a
higher
level
of
confidence
is
worth
the
time
and
energy
required

to
make
the
exploratory
movement.

To
estimate
which
movement
would
best
discriminate
among
likely
objects,
we

can
use
prior
experience
to
infer
the
expected
similarity
between
signals
from
pairs

of
objects
at
each
of
these
movements.
Movements
that
produce
the
greatest

difference
in
measured
signals
from
different
objects
would
be
optimal
for
the

discrimination
task,
while
the
movements
that
produce
similar
signals
would
not
be

useful.
One
suitable
measure
of
this
degree
of
confusion
is
the
amount
of
overlap

between
two
probability
density
functions.
An
estimation
of
this
is
provided
by
the

Bhattacharyya
coefficient,
defined
as:

113

(6.5)

The
Bhattacharyya
coefficient
varies
between
0
and
1
depending
on
the

overlapping
region
of
the
two
probability
density
functions.
For
a
given
movement,

observation,
and
pair
of
textures,
a
low
value
would
indicate
no
confusion
(so
this

would
be
a
very
useful
movement
to
make
in
order
to
disambiguate
these
objects),

while
a
high
value
would
indicate
an
undesirable
movement
because
substantial

ambiguity
would
remain.
For
all
possible
pairs
of
textures
(i
and
j)
we
can
define
an

expected
confusion
probability
matrix
for
each
possible
exploratory
movement
(m)

as:

(6.6)

For
normally
distributed
populations
this
reduces
to:

(6.7)

We
can
estimate
the
expected
uncertainty
for
a
particular
texture
and
movement

(ui,m)
that
would
remain
after
making
an
exploratory
movement
from
this
confusion

probability
matrix:

BC= p
1
x
( )
p
2
x
( )
∫
dx
C
ij,m
= p x |T
i
,M
m
( )
p x |T
j
,M
m
( ) ∫
dx
C
ij,m
=
2σ
i,m
σ
j,m
σ
i,m
2
+σ
j,m
2
e
−
µ
i,m
−µ
j,m ( )
2
4 σ
i,m
2
+σ
j,m
2
( )

114

(6.8)

Equation
(6.8)
measures
the
degree
of
confusion
between
a
specific
texture
and

all
other
likely
textures,
weighted
by
their
priors,
divided
by
the
total
amount
of

weighted
confusion
including
between
that
texture
and
itself.
If
no
other
textures

produce
overlapping
probability
distribution
curves
with
this
texture,
the
value

then
becomes
zero,
as
there
would
be
no
expected
uncertainty
for
this
texture
and

movement
combination.

The
total
expected
uncertainty
for
all
textures
for
a
given
exploratory
movement

(Um)
can
be
estimated
as:

(6.9)

Substituting
(8)
into
(9)
yields:

(6.10)

which,
given
that
the
coefficient
Cij,m
is
equal
to
1
when
i
is
equal
to
j,
can
be

shown
to
reduce
to:

u
i,m
=
C
ij,m
P T
j
( )
j,j≠i
∑
C
ij,m
P T
j
( )
j
∑
U
m
= u
i,m
P T
i
( )
i
∑
U
m
=
C
ij,m
P T
j
( )
j,j≠i
∑
C
ij,m
P T
j
( )
j
∑
P T
i
( )
#
$
%
%
%
&
'
(
(
( i
∑

115

(6.11)

The
value
from
(6.11)
can
be
used
to
determine
which
movement
would

produce
the
lowest
expected
uncertainty.
We
define
the
perceived
benefit
of
making

an
exploratory
movement
as:

(6.12)

Which
depends
on
parameter
α,
which
we
define
as
inversely
proportional
to
the

number
of
times
an
exploratory
movement
has
been
made
previously
(n)
for
the

current
discrimination
task:

(6.13)

To
promote
diversity
in
exploratory
movements
and
collect
a
richer
database
of

information,
we
need
to
reduce
the
benefit
of
repeated
movements
that
did
not
yield

satisfactory
discrimination
performance
in
prior
explorations.
Because
the

uncertainty
is
a
value
that
ranges
from
0
to
1,
a
larger
value
of
n
reduces
the
benefit

of
a
repeated
movement.
By
calculating
this
benefit
for
all
possible
exploratory

movements,
the
movement
that
produces
the
maximal
benefit
can
be
identified.
The

iterative
selection
and
execution
of
these
optimal
exploratory
movements
when

investigating
an
object
is
the
process
that
we
call
Bayesian
exploration.

U
m
=1−
P T
i
( ) ( )
2
C
ij,m
P T
j
( )
j
∑
#
$
%
%
%
&
'
(
(
( i
∑
B
m
=1−U
m
α
α=
1
n

116

Biomimetic
Tactile
Sensor

The
BioTac®
(SynTouch,
Los
Angeles,
CA)
(Figure
6-‐1)
was
designed
to
provide

both
robustness
and
sensitivity
for
multimodal
tactile
sensing.
It
consists
of
a
rigid

core
that
contains
all
sensory
transducers,
covered
by
an
elastomeric
skin.
The

space
between
the
skin
and
the
core
is
inflated
with
an
incompressible
liquid
to
give

it
a
compliance
that
mimics
human
fingerpads.
No
transducers
or
electrical

components
are
contained
in
the
skin,
making
the
design
robust
to
grit,
moisture
or

other
damage
that
typically
plagues
tactile
sensors.
The
BioTac
consists
of
three

complimentary
sensory
modalities
(force,
vibration
and
temperature)
that
have

been
integrated
into
a
single
package.
Contact
forces
distort
the
elastic
skin
and

underlying
conductive
liquid,
changing
impedances
of
electrodes
distributed
over

the
surface
of
the
rigid
core
(Wettels,
Santos,
et
al.
2008a;
Wettels
&
Loeb
2011).

Vibrations
in
the
skin
propagate
through
the
fluid
and
are
detected
by
the
pressure

sensor
(Fishel
et
al.
2008).
These
vibrations
can
be
amplified
and
filtered
to
obtain
a

dynamic
(AC)
pressure
signal
with
even
greater
sensitivity
than
the
human
fingertip

(Fishel
et
al.
2012).
Temperature
and
heat
flow
are
transduced
by
a
thermistor
near

the
surface
of
the
rigid
core
(Lin
et
al.
2009).

117

Figure
6-‐1:
The
BioTac
Conceptual
Schematic
and
Picture
with
Fingerprints

(A)
Cross-‐sectional
schematic
of
the
BioTac,
the
multimodal
tactile
sensor
used
for
these
studies.

Vibrations
of
the
skin
are
induced
when
sliding
over
textured
surfaces
and
propagate
efficiently

through
the
liquid-‐filled
sensor
where
they
can
be
sensed
by
the
pressure
sensor.
(B)
Photograph
of

an
assembled
BioTac
and
fingerprint-‐like
ridges
(inset).
These
fingerprint-‐like
ridges
that
have
a

biomimetic
size
(0.4mm
spacing)
and
have
been
observed
to
greatly
enhance
the
vibrations
that
are

detected
with
the
BioTac
(Fishel
et
al.
2009).

The
BioTac
exhibits
high
sensitivity
to
induced
vibrations
when
sliding
over

textured
surfaces
(Fishel
et
al.
2008).
More
recent
quantitative
tests
with
controlled

small
impacts
and
applied
vibrations
demonstrated
higher
sensitivity
than
human

118

fingertips
(Fishel
et
al.
2012).
In
this
study
it
was
demonstrated
that
the
BioTac
is

capable
of
detecting
small
vibrations
only
a
few
nanometers
in
amplitude
around
its

peak
frequency
sensitivity
of
330Hz,
nearly
two
orders
of
magnitude
better
than

human
subjects.
To
achieve
this
sensitivity,
the
BioTac
takes
advantage
of
carefully

designed
signal
processing
electronics
that
allow
a
sensitivity
near
the
theoretical

noise
floor
of
the
pressure
sensor.
First
the
output
from
the
piezoresistive
pressure

transducer
(24PC15SMT,
Honeywell)
is
amplified
by
a
gain
of
10
with
a
low-‐pass

anti-‐aliasing
filter
(1040Hz)
obtain
a
measurement
of
fluid
(DC)
pressure

(sensitivity:
21.8mV/kPa).
This
is
then
passed
through
a
band-‐pass
filter
(10-‐
1040Hz)
and
amplified
with
an
additional
gain
of
99.1
to
obtain
a
sensitivity
of

2.16mV/Pa
for
dynamic
(AC)
pressure.
The
background
noise
at
this
stage
was

found
to
be
only
1.2mV
(0.52Pa
of
dynamic
pressure).
Dynamic
(AC)
pressure
as

well
as
static
(DC)
pressure
were
sampled
at
2200Hz
and
digitized
with
a
resolution

of
12
bits
in
the
range
of
0-‐3.3V
(AC
Pressure
is
biased
to
1.65V)
through
onboard

electronics
inside
the
BioTac.
Sampling
and
data
transmission
are
controlled

through
a
serial
peripheral
interface
(SPI)
protocol
provided
with
the
BioTac.

The
compliance,
shape
and
material
properties
of
the
liquid-‐inflated
elastomeric

skin
(Silastic
S,
Dow
Corning)
give
rise
to
a
natural
resonant
frequency
around
200-‐
350Hz,
which
happens
to
be
similar
to
the
peak
sensitivity
of
the
Pacinian

corpuscles.
The
surface
of
the
BioTac
has
a
fingerprint-‐like
pattern
(cylindrical

shaped
ridges
with
a
height
of
0.2mm
and
spacing
of
0.4mm)
that
has
been

observed
to
enhance
the
amplitude
of
these
vibrations
in
the
BioTac
(Fishel
et
al.

119

2009).
Given
its
similarities
to
the
mechanical
properties
and
sensitivity
of
the

human
fingertip,
the
BioTac
provides
an
opportunity
to
test
theories
of
human

texture
discrimination
(Loeb
et
al.
2011)
and
to
explore
if
they
can
be
used
by

artificial
systems
seeking
to
achieve
similar
performance
in
tactile
object

identification.

Experimental
Apparatus

We
hypothesize
that
humans
utilize
a
variety
of
lateral
sliding
movements
when

exploring
textures.
The
magnitude
of
contact
force
and
velocity
of
the
sliding

movement
are
the
two
most
obvious
parameters
that
define
these
movements.

Compelling
artificial
texture
percepts
can
be
recreated
based
on
only
these
two

parameters
of
an
exploratory
movement
(Romano
&
Kuchenbecker
2011).
The

apparatus
developed
for
these
experiments
is
capable
of
precision
control
of
contact

force
and
sliding
velocity
while
collecting
sensory
data
from
the
BioTac
as
it

explores
a
texture.
The
apparatus
makes
use
of
a
stepper
motor
to
set
contact
force

and
a
precision
linear
stage
to
control
the
sliding
velocity
relative
to
the
textured

surface
(Figure
6-‐2).

120

Figure
6-‐2:
Texture
Exploration
Apparatus

A
stepper
motor
(left)
is
attached
to
a
lever
(blue)
that
can
raise
or
lower
the
BioTac
on
textures

(red).
Adjusting
the
vertical
position
of
the
stepper
motor
provides
control
of
contact
force.
To

produce
lateral
motion,
a
special
vibration-‐free
linear
stage
is
used
to
slide
textures
past
the
BioTac.

Textures
are
adhered
to
flat,
square
magnets
that
can
be
mounted
and
dismounted
rapidly
on
a
steel

plate
attached
to
the
linear
stage.

Force
Control
with
Stepper
Motor

Normal
force
of
the
BioTac
onto
the
explored
texture
is
adjusted
with
a
stepper

motor
(L4118
and
SMCI33,
Nanotec)
that
positions
a
lever
with
a
BioTac
on
the
end

(Figure
6-‐2).
Observations
indicated
that
the
change
in
fluid
pressure
of
the
BioTac

was
linearly
correlated
with
contact
force
(slope
11.5mN/bit,
R
2
=0.995)
at
forces

less
than
2N
(Figure
6-‐3).
This
was
verified
by
pressing
down
on
a
force
plate

(Nano17,
ATI)
positioned
to
be
at
the
same
height
as
the
textures.
At
forces
greater

121

than
2N
the
skin
of
the
BioTac
comes
into
contact
with
the
core
and
the
relationship

between
contact
force
and
fluid
pressure
is
no
longer
linear.
This
relationship
in
the

linear
range
was
used
to
control
the
stepper
motor
in
order
to
achieve
the
desired

contact
force
prior
to
an
exploratory
movement.
The
BioTac
was
lowered
slowly

onto
a
texture
(0.5mm/s)
while
monitoring
the
actual
DC
pressure.
When
this
value

reached
the
target
change
in
DC
pressure,
the
stepper
motor
was
stopped.
The

sliding
movements
and
associated
shear
forces
tended
to
produce
modest
changes

in
the
DC
pressure,
but
no
adjustments
were
made
to
the
stepper
motor
position

while
sliding
and
collecting
vibration
data
to
avoid
introducing
spurious
vibrations.

Figure
6-‐3:
Relationship
Between
Normal
Force
and
DC
Pressure

A
single
trial
is
shown
in
both
loading
and
unloading
(blue)
as
normal
force
increases
and
decreases

on
the
tip
of
the
BioTac.
The
best
fitting
line
is
shown
in
green
and
a
correlation
value
of
R
2

=
0.995
is

observed.

122

Velocity
Control
with
Linear
Stage

Sliding
velocity
of
the
textures
under
the
BioTac
was
controlled
with
a
precision,

low-‐vibration
linear-‐stage
(ANT130,
Aerotech).
The
high-‐quality
cross-‐roller

bearings
of
the
motor
produced
extremely
smooth
sliding
motions
and
no

mechanical
vibrations
could
be
detected
even
with
human
touch
while
the
stage
was

moving.
A
motion
controller
(Soloist,
Aerotech)
controlled
sliding
velocity
and

distance
based
on
preset
commands.
Motor
current
and
sliding
velocity
were

sampled
by
the
motion
controller,
which
could
be
queried
in
LabVIEW
using
built-‐in

software
libraries
provided
by
the
manufacturer.

Textures

A
total
of
117
textures
were
used
in
these
experiments
(Table
6-‐1).
These
were

selected
from
a
large
library
of
everyday
materials
found
in
art
supply,
fabric
and

hardware
stores.
Using
a
variety
of
commonly
occurring
textures
provides
a
more

realistic
database
of
surfaces
than
have
been
previously
used
in
other
studies
of

psychophysical
and
artificial
texture
discrimination,
which
tend
to
use
surfaces

made
from
the
same
material
varying
along
a
single
parameter
such
as
spatial

period.
These
textures
were
cut
into
75mm
x
75mm
squares
and
attached
to
square

magnets
of
the
same
size
with
adhesive
backing.
The
magnetically
backed
textures

could
be
rapidly
mounted
and
dismounted
to
a
steel
plate
attached
to
the
linear

stage.

123

Table
6-‐1:
List
of
117
Textures
Used
in
Study

Textures
can
be
grouped
into
the
following
categories:
1-‐9:
paper-‐like
materials;
10-‐23:
art
supplies

and
miscellaneous
materials;
24-‐28:
types
of
glass;
29-‐31:
types
of
foam;
32-‐35:
tiles
and
laminates;

36-‐38:
types
of
wood;
39-‐46:
engineering
materials;
47-‐54:
types
of
rubber;
55-‐66:
types
of
vinyl;
67-‐
72:
leathers
and
suedes;
73-‐83:
cottons
and
silks;
84-‐99:
other
fabrics
and
textiles;
100-‐115:
coarse

weaves;
116-‐117:
furs.

Software

The
sampling
of
the
BioTac,
control
of
stepper
motor
and
linear
stage
were
done

using
LabVIEW
(National
Instruments).
Sampling
of
the
BioTacs
was
achieved
using

a
USB/SPI
adapter
(Cheetah
SPI,
Total
Phase)
and
software
libraries
developed
by

124

and
available
from
SynTouch
(Los
Angeles,
CA).
Both
DC
and
AC
pressure
were

sampled
at
2200Hz
each.
Data
was
sampled
continuously
and
transmitted
back
to

the
computer
in
batches
every
100ms.
The
digital
controls
for
the
stepper
motor

were
also
updated
every
100ms
through
a
DAQ
card
(NI
USB-‐6218,
National

Instruments).
Control
of
the
linear
stage
was
maintained
continuously
by
the

motion
controller;
the
motor
current
and
stage
position
and
velocity
were
queried

every
100ms
through
software
libraries
developed
by
Aerotech.
For
each
texture,

the
exploratory
process
was
automated
to
produce
multiple
trials
at
each

exploratory
movement
before
proceeding
to
the
next
texture.
Data
were
analyzed

offline
in
MATLAB.

Analytical
Measures
of
Descriptive
Texture
Properties

A
system
that
uses
orthogonal
measurements
as
inputs
is
ideal
for
a
machine

classifier
problem.
Several
machine
learning
algorithms
exist
to
reduce
unnecessary

dimensionality
of
inputs,
such
as
principle
component
analysis
and
other

multidimensional
scaling
techniques
(Jain
et
al.
2000).
Moderate
success

discriminating
a
small
numbers
of
textures
has
been
achieved
using
various

statistical
measures
and
signal
processing
approaches
as
input
to
these
classifiers.

We
hypothesized
that
reasonably
orthogonal
measures
could
be
obtained
by

studying
the
language
people
use
to
describe
textures.
The
human
brain
is
a
very

effective
classifier
and
language
has
evolved
as
a
tool
to
describe
the
percepts

associated
with
texture
discrimination.
For
this
study,
we
selected
simple
and

125

intuitive
measures
of
descriptive
properties
frequently
used
in
psychophysical

literature
exploring
texture
discrimination,
bypassing
many
of
the
artificial
and

convoluted
statistical
techniques
commonly
used
in
classifier
methods.
Three

distinct
properties
have
been
identified
in
literature:
traction
(sticky/slippery),

roughness
(rough/smooth)
and
fineness
(coarse/fine).

Traction
of
Texture

Descriptive
words
such
as
slippery
and
sticky
6

are
commonly
used
to
describe

the
resistance
to
movement
when
sliding
over
a
texture.
This
dimension
has
been

suggested
to
be
relatively
orthogonal
to
the
perceptual
dimension
of
roughness

(Hollins
et
al.
1993).
To
measure
this
percept,
we
chose
to
use
traction
or
resistance

to
motion,
although
other
literature
has
reported
that
this
force
is
correlated
with

the
perception
of
roughness
(Smith,
Chapman,
Deslandes,
Langlais,
et
al.
2002a).
In

physics,
the
kinetic
coefficient
of
friction
between
two
objects
is
typically
used
to

quantify
this
property.
It
is
important
to
note
that
when
measuring
traction,
we
are

measuring
the
force
required
to
slide
the
BioTac
skin
(Silastic
S,
Dow
Corning)
over

the
texture;
the
measured
traction
is
a
property
of
these
two
surface
interactions.

Amontons’
first
and
second
laws,
as
well
as
Coulomb’s
law
of
friction,
state
that
the

coefficient
of
friction
multiplied
by
the
normal
force
is
equal
to
the
maximal
shear

force
that
two
objects
will
exert
on
one
another
when
sliding.
While
this
is
an

idealization
that
does
not
address
some
of
the
more
complex
properties
of
friction

6

The
descriptive
word
sticky
is
also
used
to
describe
adhesive
properties.
However,
even
in
this

context
it
still
refers
to
resistance
to
movement,
albeit
away
from
the
surface.

126

(i.e.
viscosity),
we
propose
that
the
measurement
of
average
frictional
force

between
the
skin
of
the
BioTac
and
explored
texture
while
sliding
can
provide
a

useful
measure
in
discriminating
textures.

In
our
experimental
testbed,
the
linear
stage
that
produces
sliding
movements

can
be
queried
for
instantaneous
motor
current.
This
was
found
to
vary
linearly

with
shear
force
by
placing
the
stage
at
a
45-‐degree
angle
and
attaching
various

weights
to
it.
In
dynamic
sliding
it
was
observed
that
initially
a
high
amount
of

current
was
required
to
accelerate
the
stage
from
rest
but
only
a
low
amount
of

current
was
required
to
overcome
the
dynamic
friction
within
the
stage
in
the

unloaded
state
once
the
stage
had
reached
its
target
velocity.
When
the
BioTac
was

pushed
against
a
texture
sample
on
the
stage,
the
motor
current
required
to
sustain

constant
velocity
sliding
increased
linearly
with
the
applied
normal
force.
Given
the

linearity
of
this
response,
the
average
motor
current
from
the
stage
while
sliding

was
used
to
estimate
the
traction
between
the
BioTac
and
textures
being
explored.

This
was
simpler
and
more
accurate
than
using
the
force-‐sensing
modality
of
the

BioTac
itself,
which
can
extract
tangential
force
from
the
distributed
change
in

impedance
of
its
impedance
sensing
electrodes
(Wettels
&
Loeb
2011).

(6.14)

Examples
of
this
signal
and
calculations
are
provided
in
Figure
6-‐4,
frame
B.

Traction∝MotorCurrent

127

Roughness
of
Texture

When
sliding
over
surfaces
with
different
roughness
properties
with
the
BioTac,

we
observed
that
the
amplitude
of
vibration
as
measured
by
the
dynamic
pressure

sensor
in
the
BioTac
(PAC)
was
correlated
with
the
perceived
roughness
of
the

texture,
similar
to
the
observations
of
(Bensmaïa
&
Hollins
2005).
In
our
own

findings,
smooth
surfaces
were
found
to
produce
virtually
no
vibrations,
while

rougher
surfaces
produced
vibrations
of
much
greater
amplitudes.
To
quantify
this,

we
computed
the
logarithm
of
signal
power
after
subtracting
the
background
noise

with
the
equation:

(6.15)

(6.16)

AC
pressure
was
filtered
with
a
20-‐700Hz
digital
band-‐pass
filter
(66
th

order
FIR

filter)
to
simulate
the
frequency
response
of
the
Pacinian
corpuscles
that
are

thought
to
mediate
texture
perception
and
eliminate
low
frequency
oscillations

from
contributing
to
this
estimate.
An
example
of
the
filtered
signal
is
shown
in

Figure
6-‐4,
frame
B.
It
was
also
observed
that
even
when
locked
in
a
fixed
position,

the
stepper
motor
produced
some
background
noise
that
was
not
consistent
from

trial
to
trial.
By
measuring
this
background
noise
power
prior
to
sliding
and

subtracting
it
from
the
power
while
sliding
we
were
able
to
obtain
more
consistent

Power=
1
N
filt P
AC
n
( ) ( ) ( )
2
n=1
N
∑
Roughness∝log Power−background noise
( )

128

measurements
of
the
signal
power
contributed
from
sliding
alone.
The
logarithm
of

this
signal
was
found
to
better
reflect
a
more
evenly
distributed
set
of
roughness

properties
with
similar
variances
(as
presented
later
in
Figure
6-‐5).
A
similar

justification
for
using
the
logarithm
of
surface
amplitude
in
psychophysical

discrimination
of
textures
has
been
proposed
by
Bergnam
Tiest
and
Kappers

(Bergmann
Tiest
&
Kappers
2006).

Fineness
of
Texture

The
role
of
spatial
periodicity
in
the
perception
of
textures
has
been
explored
in

a
number
of
studies.
It
has
been
observed
that
the
coarser
textures
produce
lower-‐
frequency
vibrations
when
sliding
over
an
object,
while
finer
textures
produce

higher-‐frequency
vibrations,
suggesting
the
simple
relationship:

(6.17)

with
λ
equal
to
the
spatial
wavelength
of
the
texture
or
fingerprints
and
v
equal
to

the
velocity
of
lateral
motion.
This
is
the
operating
principle
of
many
algorithms
for

texture
discrimination
in
artificial
sensors
(Mukaibo
et
al.
2005;
Scheibert
et
al.

2009;
Oddo
et
al.
2009;
Oddo
et
al.
2011).
Our
own
findings
have
indicated
that
this

relationship
breaks
down
for
finer
textures
and
higher
velocities
(as
presented
later

in
Figure
6-‐6).
Although
not
directly
reported,
this
can
be
observed
in
the
results

from
Mukaibo
et
al.
for
higher
spatial
frequencies
and
from
Oddo
et
al.
for
higher

velocities,
in
which
the
estimation
of
spatial
wavelength
becomes
much
more
prone

f = v λ

129

to
errors
when
approaching
these
limits.
The
experiments
of
Scheibert
et
al.
were

conducted
at
very
low
exploratory
speeds
(0.02
cm/s),
two
to
three
orders
of

magnitude
slower
than
common
exploratory
movements
employed
by
humans

(Dahiya
&
Gori
2010),
which
is
clearly
not
fast
enough
to
observe
this
dynamic

behavior.
Furthermore,
smoother
surfaces
have
been
demonstrated
to
be
free
of

spectral
harmonics,
instead
generating
signals
that
represent
1/f
noise
in
the

frequency
domain
(Wiertlewski
et
al.
2011).
Measured
frequencies
do
not
always

relate
linearly
to
the
spatial
wavelength
of
the
texture,
particularly
for
fine
or

smooth
textures,
but
the
estimation
of
this
frequency
can
still
yield
useful

information
about
the
relative
fineness
or
coarseness
of
the
texture.

We
propose
a
measure
of
spectral
centroid
to
determine
the
weighted
frequency

power
of
the
vibrations
recorded
by
the
BioTac.
The
dynamic
pressure
is

transformed
into
the
frequency-‐domain
using
the
single-‐sided
fast
Fourier

transform
and
the
spectral
centroid
is
calculated
using
the
weighted
average
of
the

frequency
power
from
the
following
equation:

(6.18)

To
smooth
the
frequency
domain
response,
a
sliding
window
was
used
to
collect

the
Fourier
transform
at
different
sections
of
the
signal,
which
were
averaged
and

used
for
the
spectral
centroid
estimation.
Examples
of
this
measurement
are
shown

in
Figure
6-‐4.
A
logarithmic
scale
of
this
measure
is
used
as
an
input
to
the
texture

SC=
fft P
AC
( )
2
× f
( )
∑
fft P
AC
( )
2
∑

130

discrimination
model.
Note
that
this
method
does
not
attempt
to
compensate
for
the

rather
complex
frequency
response
of
the
BioTac
itself
(Fishel
et
al.
2012).

(6.19)

Fineness∝log SC
( )

131

Figure
6-‐4:
Typical
Signals
that
Occur
During
an
Exploratory
Movement

132

Figure
6-‐4
caption:
In
(A)
the
change
in
DC
pressure
(top)
and
sliding
velocity
(bottom)
are
shown

over
the
course
of
the
trial.
The
loading
of
DC
pressure
by
the
stepper
motor
occurs
between
t
=
-‐1.5

and
-‐1
seconds,
in
this
example
it
is
equal
to
roughly
17
bits
or
0.2N.
Once
the
desired
contact
force
is

reached,
the
position
of
the
stepper
motor
is
held
for
about
0.5s
before
the
linear
stage
is
actuated
to

the
controlled
sliding
velocity
(6.31
cm/s
in
this
example).
The
measurement
region
of
signals

indicated
as
vertical
black
lines
occurs
shortly
after
the
sliding
and
stops
before
the
sliding
is

completed.
In
(B)
the
time
axis
is
zoomed
with
respect
to
(A)
and
motor
current
and
filtered
PAC

signals
are
displayed.
In
the
top
trace,
the
motor
current
of
the
linear
stage
before
the
measurement

of
signals
indicated
by
the
vertical
black
lines
is
initially
high
due
to
the
acceleration
of
the
linear

stage.
The
horizontal
dashed
line
represents
the
average
motor
current
over
the
measurement
region

and
is
used
to
estimate
the
traction
between
the
texture
and
the
BioTac
while
sliding.
In
the
lower

trace
filtered
PAC
signals
are
presented.
The
root
mean
squared
(RMS)
power
is
indicated
by
the

dashed
lines
as
upper
and
lower
bounds;
the
logarithm
of
the
actual
power
is
used
as
the
roughness

signal.
In
(C)
the
fast
Fourier
Transform
is
presented
of
the
unfiltered
PAC
signal.
The
spectral

centroid
is
calculated
as
the
weighted
average
of
spectral
power
components
and
is
presented
as
the

vertical
dashed
line.
This
measurement
is
used
to
estimate
the
fineness
of
the
texture.

Normality
of
Signals

The
classifier
discussed
in
the
previous
section
makes
the
assumption
that

signals
arise
from
a
normally
distributed
population.
Due
to
the
large
number
of

textures
explored,
only
a
small
number
of
trials
could
be
collected
for
each
of
the

exploratory
movements.
Therefore,
a
thorough
analysis
of
the
exact
probability

density
function
for
these
signals
could
not
be
conducted.
We
make
the
assumption

that
signals
are
normally
distributed
for
the
purposes
of
this
study,
however
as

more
samples
are
collected
a
clearer
understanding
of
the
true
probability
density

function
would
serve
to
improve
the
performance
of
this
classifier.
This
will
be
even

more
important
when
classifying
surfaces
that
have
some
heterogeneity
in
their

textures.

133

The
Curse
of
Dimensionality

Additional
input
dimensions
will
typically
improve
performance
of
a
classifier
if

they
are
well
defined.
In
practice,
however,
it
has
been
observed
that
additional

dimensions
will
actually
degrade
performance
of
a
classifier
for
a
constant
sample

size,
a
property
that
has
become
known
as
the
curse
of
dimensionality
(Jain
et
al.

2000).
To
overcome
this,
an
exponential
increase
in
training
data
is
required
for

each
new
dimension.
To
avoid
the
need
for
such
a
vast
amount
of
training
data,
our

model
considers
only
univariate
distributions
of
a
single
property
at
each

exploratory
movement.
Indeed,
we
have
found
that
our
classifier
performance
was

severely
degraded
when
considering
all
three
properties
at
once
for
each

exploratory
movement
using
multidimensional
probability
density
functions.
The

result
was
an
algorithm
that
quickly
converged
in
one
or
two
exploratory

movements
but
frequently
to
the
wrong
texture.
As
additional
training
is
obtained,

such
a
multidimensional
approach
would
be
optimal,
but
this
is
infeasible
for
the

large
number
of
textures
used
in
this
study.
To
avoid
this
shortcoming
we
only

considered
a
single
signal
during
an
exploration
movement.
While
this
may
not

appear
to
take
full
advantage
of
all
available
information,
it
is
actually
preferable
for

a
classifier
with
such
limited
experience,
allowing
it
to
focus
solely
on
the
property

that
it
determines
to
be
most
relevant
for
a
given
exploration.

134

Selection
of
Set
of
Exploratory
Movements

While
there
exist
infinite
combinations
of
contact
force
and
sliding
velocities
that

can
be
used
when
exploring
textures,
experimental
studies
have
demonstrated
that

individual
subjects
are
quite
consistent
in
reproducing
exploratory
movements
in

these
tasks,
although
there
is
a
high
degree
of
variability
among
subjects
(Smith,

Gosselin
&
Houde
2002b).
This
suggests
that
certain
combinations
of
exploratory

movement
tend
to
be
more
efficient
and
that
an
individual
person
discovers
and

uses
such
combinations
consistently.
The
internal
representation
of
the
objects
in

the
external
world
could
then
be
based
on
predictable
sensations
obtained
when

well-‐learned
and,
hence,
dependable
exploratory
movements
are
made.

To
identify
these
useful
movements,
ten
textures
were
chosen
for
the
pilot
study

based
on
their
perceived
dissimilarity
in
the
multidimensional
space
of
identified

texture
properties
(low/high
traction,
rough/smooth,
coarse/fine)
(Table
6-‐2).

Given
the
diversity
of
this
sample,
this
was
believed
to
represent
most
of
the

perceptual
range
of
the
complete
set
of
117
textures.

135

Table
6-‐2:
Perceived
Properties
of
10
Textures
used
in
Pilot
Study

Estimations
of
traction,
roughness
and
fineness
when
explored
by
the
human
finger
for
the
ten

textures
used
in
the
pilot
study.
These
samples
were
selected
to
represent
the
range
of
material

properties
to
be
expected
over
the
larger
population
of
textures.

A
total
of
36
exploratory
movements
were
chosen
based
on
all
combinations
of

six
speeds
and
six
forces.
The
ranges
of
these
parameters
were
chosen
to
mimic
the

ranges
humans
typically
use
when
exploring
textures
(1-‐10
cm/s,
0.2-‐2
N).
Force
(F)

and
velocity
(v)
at
these
6
steps
were
calculated
as
a
geometric
series:

v
i
=10
i−1
5
×1
cm
s

(6.20)

(6.21)

Using
such
a
scale
permits
for
comparisons
to
be
made
between
exploratory

movements
that
had
equivalent
frictional
sliding
power.
For
a
given
texture

F
i
=10
i−1
5
×0.2 N

136

interacting
with
the
BioTac
skin,
sliding
power
is
proportional
to
the
tangential

force
times
sliding
velocity.
Given
the
assumptions
of
Amoltons’
first
law
of
friction,

the
tangential
force
would
be
proportional
to
the
normal
force
for
a
given
pair
of

surfaces
with
a
constant
coefficient
of
friction.
With
this
set
of
parameters,
pairs
of

force
and
velocity
with
equal
sliding
power
could
be
found
from
the
following

equation:

P
a
=F
i
×v
a−i
×µ=10
a−2
5
×µ×20mW
(6.22)

Six
repetitions
of
the
36
movements
were
collected
for
each
of
the
ten
textures.

At
each
trial
the
starting
location
on
the
texture
was
randomized
to
ensure
collected

signals
were
properties
of
the
texture
itself
and
not
necessarily
an
isolated
feature

on
a
given
portion
of
the
texture.
Exploratory
movements
were
automated
by

software
and
the
data
were
saved
to
file
for
post
processing.
Data
were
collected
for

each
of
the
exploratory
movements
for
a
particular
texture
before
moving
on
to
the

next.
The
degree
of
uncertainty
for
each
movement
was
analyzed
for
each
signal

property
independently
rather
than
as
a
multivariate
system.
This
method
was

chosen
to
avoid
the
curse
of
dimensionality,
as
discussed
in
the
previous
section.

It
was
observed
that
high
frictional
sliding
power
exploratory
movements

(combinations
of
high
force
and
velocity)
led
to
an
increase
in
skin
wear
and

removal
of
fingerprints,
resulting
in
substantial
changes
in
the
vibration
signals

recorded
by
the
BioTac;
these
pairs
of
exploratory
movement
parameters
were

avoided.
Of
the
remaining
options,
we
selected
three
combinations,
each
of
which

137

provided
the
lowest
uncertainty
for
one
of
the
three
properties
(see
Figure
6-‐7).

These
three
most
useful
movements
were
used
to
explore
the
entire
set
of
117

texture
samples.

Classifier
Training
and
Data
Collection

Five
trials
were
completed
at
each
of
the
three
selected
exploratory
movements

for
the
entire
set
of
117
textures.
All
trials
were
completed
on
a
single
texture
before

moving
to
the
next.
During
these
trials,
the
skin
was
checked
regularly
to
identify
if

the
fingerprints
were
still
intact.
It
was
observed
that
fingerprint
wear
had
a

detrimental
effect
on
the
repeatability
of
data,
particularly
in
the
measurement
of

roughness
from
vibration
power.
To
compensate
for
this,
the
skin
of
the
BioTac
was

replaced
if
there
were
any
visible
signs
of
wear.
With
this
approach
we
were
able
to

avoid
any
signal
drift
resulting
from
wear,
which
was
verified
by
comparing
signals

before
and
after
the
skin
replacement.
The
skin
of
the
BioTac
was
replaced
two

times
under
these
conditions.
Data
collection
for
all
117
textures
took
roughly
20

hours
and
spanned
4
days.

The
data
from
these
trials
served
to
build
a
prior
experience
database
that
could

be
used
to
identify
presented
textures
and
to
compute
expected
benefit
of
a
given

exploratory
movement.
During
the
course
of
these
tests,
three
textures
were

damaged
during
the
higher
force
exploratory
movements.
Data
for
these
textures

were
not
used
in
the
sample.
Outputs
of
the
various
properties
at
their
optimal

movements
are
presented
(Figure
6-‐8)
along
with
the
confusion
probability

138

matrices
for
all
combinations
of
signals
and
movements
(Figure
6-‐9).
Signal

correlation
is
presented
to
show
the
independence
of
each
dimension
(Table
6-‐3).

Texture
Discrimination
and
Comparison
with
Human
Performance

Analyzing
the
resulting
confusion
probability
matrices
of
the
large
texture

dataset
yielded
surprising
findings
in
the
confusion
between
textures.
Many
pairs
of

textures
that
were
perceived
as
difficult
to
discriminate
by
touch
were
readily

distinguishable
based
on
at
least
one
dimension
of
the
three
calculated
texture

properties,
while
some
pairs
of
textures
that
appeared
simple
to
discriminate
by

human
touch
were
determined
to
be
more
challenging
to
the
artificial
system
based

on
the
observed
confusion
matrices
(Figure
6-‐9).
Eight
pairs
of
textures
(16
textures

total)
were
selected
for
a
study
of
discriminability,
including
pairs
that
were

perceived
to
be
similar
to
human
observers
but
not
the
artificial
system,
the
reverse,

or
similar
to
both.
Care
was
taken
to
select
texture
pairs
that
did
not
have
other

properties
that
were
readily
discriminable
by
other
non-‐textural
mechanisms
such

as
compliance
or
thermal
properties,
for
which
human
subjects
would
have
an

obvious
advantage
(the
BioTac
does
provide
signals
that
can
be
used
to
estimate

both
properties
(Lin
et
al.
2009)
but
these
were
not
used
in
this
study).

Five
human
subjects
consented
to
participate
in
a
study
to
explore
biological

abilities
to
discriminate
between
similar
textures.
Prior
to
these
experiments,

subjects
were
informed
that
they
would
be
presented
with
one
of
the
eight
pairs
of

textures
at
a
time,
which
they
could
see
and
explore
by
sliding
their
fingers
over

139

them
for
as
long
as
they
desired
in
order
to
feel
comfortable
discriminating
between

the
two
textures
in
the
pair.
They
were
informed
that
after
they
were
finished

exploring,
they
would
begin
the
testing
phase
and
would
not
be
allowed
to
explore

both
textures
again.
No
additional
guidance
was
provided
on
which
properties
or

exploratory
movements
would
be
optimal
for
performing
the
discrimination
task.

They
were
also
informed
that
when
ready,
they
would
have
their
vision
occluded

and
be
presented
with
a
random
selection
of
one
of
the
two
textures
for
four
trials.

Subjects
were
aware
that
the
selection
of
the
presented
texture
was
predetermined

from
a
random
number
generator
(i.e.
it
would
not
always
be
each
texture
twice
for

the
four
trials,
by
chance
it
could
even
be
the
same
texture
four
times).
The

experimenter
suggested
that
the
subjects
could
call
these
textures
A
or
B,
however

all
subjects
preferred
to
refer
to
the
textures
based
on
their
visual
properties
(i.e.,

“the
blue
one”).
For
the
testing
phase,
which
started
immediately
after
the

exploratory
phase,
a
small
platform
was
placed
in
front
of
the
subject
where

textures
were
to
be
placed.
The
platform
was
short
and
unobtrusive,
allowing

subjects
to
assume
the
same
posture
used
in
the
exploratory
phase.
The

experimenter
placed
the
randomized
texture
on
this
platform
and
the
subject
held

his
finger
over
the
texture
until
an
auditory
command
was
given
to
start
exploration.

After
making
exploratory
movements
(which
ranged
from
2
or
3
to
dozens
of

movements
depending
on
the
difficulty
the
subject
was
experiencing),
subjects

notified
the
experimenter
which
of
the
two
textures
they
thought
they
were

touching.
This
was
repeated
for
four
trials
for
each
of
the
eight
pairs
of
textures.

140

While
subjects
were
eager
to
know
their
performance,
this
was
not
disclosed
to

them
until
the
completion
of
the
experiment.
Average
performance
across
all
5

subjects
for
the
8
texture
pairs
in
terms
of
percentage
of
correct
classifications

(chance
=
50%)
is
presented
in
Table
6-‐4.

Comparison
in
performance
of
humans
with
the
artificial
system
in
this

discrimination
task
requires
two
separate
populations
of
data,
one
to
represent
the

information
obtained
in
the
exploratory
phase
and
the
other
to
represent
novel

information
being
encountered
in
the
identification
phase.
This
is
commonly

referred
to
in
machine
classifier
problems
as
a
training
set
and
a
validation
set.
The

training
set
consists
of
data
to
be
used
as
the
previous
experience
that
the
Bayesian

exploration
algorithm
refers
to
when
encountering
an
unknown
texture
in
order
to

determine
optimal
exploratory
movements
and
to
compute
posterior
probabilities

after
these
movements.
The
original
set
of
data
obtained
in
the
previous
section
was

used
to
create
this
training
set.
A
second
set
of
novel
data
was
collected
and
used
as

a
validation
set
for
the
same
textures
that
were
used
in
the
human
texture

discrimination
studies.
Similar
to
the
training
set,
five
trials
for
each
of
the
three

exploratory
movements
were
collected
for
these
16
textures.

The
computational
speed
of
computer
processors
made
it
attractive
to
analyze

the
performance
of
this
artificial
discrimination
task
offline
in
a
virtual
texture

exploration.
When
performing
a
virtual
exploratory
movement,
a
randomly
selected

trial
from
the
unseen
validation
set
of
the
texture
being
explored
was
given
to
the

classifier.
Due
to
the
high
degree
of
randomness
of
these
simulations,
a
total
of
1000

141

simulations
as
described
below
for
each
of
the
8
pairs
of
textures
were
conducted
to

establish
a
more
accurate
measure
of
performance.

During
a
texture
discrimination
task
for
the
artificial
system,
a
pair
of
textures

was
selected
and
their
prior
probabilities
were
set
equally
to
50%.
The
probability

for
all
other
textures
in
the
database
was
set
to
zero,
effectively
eliminating
them

from
the
classifier’s
decision
process.
One
of
the
two
textures
was
selected
at

random
as
the
unknown
texture
to
be
identified
by
the
system.
The
Bayesian

exploration
algorithm
used
data
in
its
previous
experience
(from
the
training
set)
to

decide
which
exploratory
movement
and
signal
would
discriminate
optimally

between
them.
The
signal
from
this
movement
in
the
validation
set
was
delivered
to

the
classifier
and
the
posterior
probabilities
of
the
two
textures
were
updated
using

Bayesian
inference.
The
process
of
performing
optimal
combinations
of
exploratory

movements
and
properties
to
measure
through
Bayesian
exploration
was
repeated

until
one
of
the
two
textures
converged
to
a
probability
of
greater
than
99.9%
7
.
Data

from
these
discrimination
tasks
was
not
added
to
the
database.
Results
comparing

the
performance
of
human
subjects
and
the
Bayesian
exploration
are
presented
as

the
percentage
of
correct
identifications
in
Table
6-‐4.

7

The
99.9%
convergence
criterion
for
texture
pairs
was
higher
than
the
99%
used
in
the

absolute
classification
task
as
discussed
in
the
following
section.
It
was
found
that
when

discriminating
between
a
smaller
number
of
textures
(i.e.
2
as
used
in
this
experiment)
the
algorithm

would
quickly
converge
in
only
one
or
two
movements
and
frequently
to
a
wrong
decision
if
the

required
probability
threshold
was
not
set
to
a
high
enough
value.
By
increasing
the
required

probability
to
a
higher
level,
additional
exploratory
movements
would
be
required,
resulting
in

better
overall
classification
performance.
At
this
level
most
solutions
converged
to
the
correct
values

with
a
median
of
three
exploratory
movements
with
satisfactory
results.

142

Absolute
Texture
Identification

The
new
validation
data
from
the
16
textures
obtained
in
the
previous
section

were
also
used
against
the
entire
set
of
117
textures
to
evaluate
the
performance
of

absolute
texture
identification.
The
classifier
was
not
aware
of
the
16
textures
it
was

being
presented
and
initially
set
the
probabilities
for
all
textures
to
the
same
value

(1
divided
by
117).
The
same
process
of
Bayesian
decision
making
to
determine

optimal
pairs
of
exploratory
movements
and
signals
for
virtual
exploratory

movements
was
followed
as
discussed
in
the
previous
section.
The
performance
of

this
Bayesian
exploration
approach
was
compared
with
two
alternative
exploratory

strategies.
In
the
first,
the
most
useful
movements
for
each
of
these
signals
as

determined
previously
were
cycled;
in
the
second,
exploratory
movement
and
signal

combinations
were
randomly
selected.
A
maximum
of
10
exploratory
movements

were
allowed
and
the
classifier
was
run
until
any
texture
converged
to
greater
than

99%
probability
or
until
the
10
exploratory
movements
were
conducted.
If
not

converged,
the
texture
with
the
greatest
probability
after
the
10
exploratory

movements
was
determined
to
be
the
most
likely
candidate.
A
total
of
8000
Monte

Carlo
simulations
over
the
16
textures
were
conducted
and
performance
is

presented
as
percentage
of
correct
classifications
(Table
6-‐5).
Examples
of
the

evolving
probabilities
of
possible
textures
and
the
selected
exploratory
movements

for
some
of
these
trials
are
shown
in
Figure
6-‐10.

143

Results

Analysis
of
Descriptive
Texture
Properties

In
the
pilot
study
of
10
textures,
descriptive
properties
were
found
to
reflect

expected
values.
For
instance,
graphite
produced
the
lowest
measure
for
traction

between
the
surface
and
the
skin
of
the
BioTac
while
polished
aluminum
and
rubber

had
the
highest.
Foam,
satin,
rayon
and
burlap
produced
the
highest
measures
of

roughness,
while
polished
aluminum
produced
the
lowest
measure
of
roughness.
In

calculating
the
spectral
centroid,
the
finer
textures
such
as
graphite
and
satin

produced
higher
values
while
the
coarser
textures
such
as
burlap
and
velcro

produced
lower
values.
Featureless
textures
such
as
polished
aluminum
and
rubber

tended
to
produce
low
spectral
centroids
as
well
due
to
their
1/f
noise
as
discussed

in
(Wiertlewski
et
al.
2011).
The
exploratory
movements
that
produced
the
most

discriminability
within
the
10-‐texture
dataset
as
calculated
by
the
minimal

uncertainty
are
presented
in
Figure
6-‐5.

144

Figure
6-‐5:
Measures
of
Texture
Properties
at
their
Optimal
Movements

Texture
IDs:
T2
=
Linen
Paper,
T10
=
Velcro
Hooks,
T30
=
Foam,
T43
=
Graphite,
T45
=
Milled

Aluminum,
T46
=
Polished
Aluminum,
T47
=
Rubber,
T84
=
Satin,
T87
=
Rayon,
T100
=
Burlap.
Top

frame:
Traction
as
measured
from
motor
current
to
overcome
sliding
friction
between
the
skin
of
the

BioTac
and
the
texture.
Middle
frame:
roughness
as
measured
from
vibration
power
as
recorded
by

the
PAC
signal
from
the
BioTac.
Bottom
frame:
spectral
centroid
as
measured
by
weighted
spectral

power
of
PAC
signal
from
the
BioTac.
Six
trials
are
shown
for
each
movement
to
demonstrate

clustering
of
each
signal.

145

A
notable
finding
of
these
trials
was
that
the
spectral
centroid
did
not
scale
with

sliding
velocity
for
all
textures
(Figure
6-‐6).
Such
scaling
was
observed
only
for

certain
coarse
textures
that
were
also
rough
(Velcro
hooks,
foam,
rayon,
burlap).

This
constitutes
additional
evidence
that
fingerprints
do
not
simply
convert
sliding

velocity
and
spatial
frequency
into
temporal
signals
as
concluded
by
(Scheibert
et
al.

2009).

Figure
6-‐6:
Spectral
Centroid
as
a
Function
of
Sliding
Velocity

Spectral
centroid
as
a
function
of
sliding
velocity
at
the
lightest
contact
force
(0.2N).
Results
indicate

that
only
the
spectral
centroid
of
coarser
textures
(T10
=
Velcro
Hooks,
T30
=
Foam,
T87
=
Rayon,

and
T100
=
Burlap)
consistently
increased
as
a
function
of
velocity.
Finer
textures
produced

idiosyncratic
functions
of
velocity,
while
the
spectral
centroid
of
graphite
(T43)
decreased
as
a

function
of
velocity.

146

Identifying
the
Most
Useful
Exploratory
Movements

The
most
useful
movement
for
each
property
was
selected
from
the
set
of
36

movements
(Figure
6-‐7).
It
was
observed
that
combinations
of
high-‐power

exploratory
movements
(high
force
and
high
velocity)
resulted
in
a
high
rate
of

fingerprint
wear
on
the
BioTac’s
skin.
These
high-‐wear
movements
(grey
boxes
with

white
text
in
Figure
6-‐7)
were
eliminated,
although
some
appeared
to
be
useful
for

discrimination
(e.g.
6.31cm/s
and
1.26N
for
discriminating
roughness).
As
a
general

trend,
it
was
observed
that
the
ability
to
discriminate
traction
improved
at
lower

velocities
and
higher
forces,
while
the
ability
to
discriminate
roughness
improved
at

higher
velocities
and
lower
forces.
These
findings
supported
our
own
intuition
on

the
exploratory
movements
humans
make
to
extract
these
properties.
We
propose

that
when
humans
make
a
movement
to
determine
surface
traction,
they
tend
to
use

a
high
amount
of
force
while
slowly
moving
their
finger.
Presumably
this
is
to

maximize
the
amount
of
shear
force
sensed
in
the
finger;
it
is
easier
to
control
these

forces
by
moving
slowly.
Similarly
in
our
experimental
testbed,
sliding
at
slower

velocities
produced
more
stable
sliding
forces
as
measured
by
the
motor
current;

this
resulted
in
more
consistent
measurements
between
multiple
trials.
To

discriminate
roughness,
we
have
observed
that
humans
tend
to
use
very
light
forces

while
sliding
over
the
surfaces
of
textures
to
feel
their
vibrations.
Our
observations

with
the
BioTac
have
demonstrated
a
possible
utility
of
this
strategy.
In
general,

faster
velocities
tend
to
produce
higher
amplitude
vibration
signals,
while
greater

forces
tend
to
dampen
vibrations
sensed
by
the
BioTac.
The
findings
of
these
two

147

exploratory
movements
as
the
most
useful
for
these
properties
in
our
artificial

system
provide
additional
support
to
these
hypotheses
about
biological
exploratory

movements
for
perception
of
traction
and
roughness.

While
a
certain
movement
may
be
classified
as
generally
“useful”
for
a
given

property,
other
movements
may
actually
be
more
useful
for
discriminating
a
given

pair
of
materials
along
this
property.
Because
all
properties
are
collected
with
each

movement,
the
classification
algorithm
takes
advantage
of
all
nine
combinations
of

available
movements
and
material
properties
during
its
decision
process
of

determining
the
optimal
movement
and
signal
to
sense.

148

Figure
6-‐7:
Selection
of
Optimal
Exploratory
Movements

149

Figure
6-‐7
caption:
Selection
of
optimal
exploratory
movements
for
pilot
study
of
10
textures.
Tables

present
the
uncertainty
calculated
for
each
measurement
property
for
combinations
of
contact
force

and
sliding
velocity.
Gray
boxes
with
white
numbers
in
the
lower-‐right
half
plane
represent

exploratory
movements
that
were
excluded
due
to
the
high
wear
rate
high
force
and
velocity

combinations
had
on
the
skin.
Values
in
the
upper
left
half
plane
are
coded
from
blue
to
white
to

represent
decreasing
uncertainty
with
lower
values
being
ideal
for
discrimination
of
the
10
textures.

From
this
the
three
of
the
most
useful
movements
were
selected
as
1.26N
and
1cm/s
for

discrimination
based
on
traction,
0.2N
and
6.31cm/s
for
discrimination
based
on
roughness,
and

0.5N
and
2.5cm/s
for
discrimination
based
on
fineness.
Movements
at
0.2N
and
2.5cm/s
were
not

selected
for
the
roughness
or
fineness
measures
because
they
appeared
to
be
outliers
and
did
not
fit

the
general
trend
of
performance
from
neighboring
movements.

Training
Dataset

The
measured
properties
of
the
117
textures
were
individually
tightly
clustered

for
repeated
measures
but
spanned
most
of
the
three
dimensional
property
space,

an
ideal
situation
for
an
efficient
classifier.
The
complete
set
of
measured
properties

for
each
signal
at
the
most
useful
movement
for
that
property
is
shown
in
Figure
6-‐8.

In
many
cases,
textures
that
had
similar
values
for
one
property
tended
to
be

dissimilar
along
other
dimensions,
suggesting
the
utility
of
well-‐chosen
next

exploratory
movements.

150

Figure
6-‐8:
Summary
of
All
Texture
Properties
for
117
Textures

Summary
of
all
texture
properties
at
their
most
useful
movements
for
entire
set
of
117
textures.

Similar
types
of
materials
are
grouped
by
color
as
shown
in
the
top
panel.
Five
trials
are
shown
for

each
property
and
texture
to
demonstrate
the
clustering
of
the
measurements.
In
many
cases
the

clustering
is
so
tight
that
all
five
trials
appear
as
a
single
marker.

A
graphical
representation
of
confusion
probability
matrices
was
generated
for

each
combination
of
movements
and
properties
(Figure
6-‐9).
The
geometric
mean

(calculated
by
multiplying
confusion
probability
matrices
and
taking
the
n
th

root
of

151

the
result)
for
all
of
these
movements
and
properties
provides
additional
insight

into
which
pairs
of
textures
have
the
most
confusion
across
all
exploratory

movements
and
properties.
Results
indicate
that
most
textures
are
readily

distinguishable
with
a
few
exceptions.
When
the
confusion
probability
matrices
are

combined
for
only
the
optimal
movements,
there
are
considerably
more
off-‐diagonal

dark
spots
representing
textures
that
are
likely
to
be
confused.
The
identification

algorithm
chooses
the
combination
of
movement
and
signal
type
that
will
be
most

likely
to
discriminate
among
the
most
probable
alternatives
at
any
given
point
in
the

identification
process.

152

Figure
6-‐9:
Confusion
Probability
Matrices

Confusion
probability
matrices
for
each
combination
of
exploratory
movement
and
texture
property

and
geometric
mean
of
all
signals
and
only
the
most
useful
combinations
of
signals
and
movements.

Spots
are
scaled
from
white
to
black
to
represent
low
to
high
confusion.
Axes
in
each
inset
represent

the
texture
indexes.
The
solid
black
line
on
each
diagonal
indicates
the
value
of
unity,
as
each
texture

has
a
100%
chance
of
being
confused
with
itself
at
all
movements.
Blue
outlines
indicate
the
optimal

movement
for
each
signal.
The
geometric
mean
of
all
confusion
probability
matrices
have
very
few

off-‐diagonal
dark
spots,
whereas
geometric
mean
of
only
the
most
useful
exploratory
movements
for

each
signal
have
substantially
more
dark
spots
indicating
potential
confusion
between
similar

textures
and
the
value
of
multiple
movements
for
each
signal.
By
taking
advantage
of
all

combinations
of
exploratory
movements
and
properties,
rather
than
just
the
properties
at
their

“optimal”
movements,
we
see
an
improvement
of
60%
in
the
geometric
average
of
the
confusion

probability
matrices
and
overall
uncertainty.

153

Correlation
between
calculated
texture
properties
were
analyzed
for
all
117

textures
for
each
movement
and
the
average
values
for
each
movement
are

provided
(Table
6-‐3).

Table
6-‐3:
Correlation
Matrices
for
Texture
Properties

Correlation
matrix
between
each
texture
property
at
each
of
the
three
movements.

An
interesting
finding
was
a
strong
negative
correlation
between
roughness
and

traction.
This
can
be
observed
in
materials
such
as
rubber
and
glass,
which
generally

have
smooth
surfaces,
yet
produce
a
high
amount
of
friction
against
the
silicone
skin

of
the
BioTac.
This
contradicts
(Smith,
Chapman,
Deslandes,
Langlais,
et
al.
2002a),

154

who
reported
a
strong
positive
correlation
between
friction
and
roughness
for

human
fingertips,
but
did
not
include
such
a
diverse
set
of
materials.

Texture
Discrimination
and
Comparison
with
Human
Performance

Results
of
the
performance
for
discriminating
between
two
textures
for
both

humans
and
our
artificial
algorithm
are
provided
in
Table
6-‐4.
The
performance
of

the
classifier
exceeded
human
performance
for
all
pairs.
The
average
performance

of
human
subjects
across
all
of
the
texture
pairs
was
found
to
be
81.3%
while
the

average
performance
of
our
classifier
was
found
to
be
99.6%.
This
result
was
quite

unexpected,
as
human
capabilities
have
previously
been
thought
of
as
the
“gold

standard.”
Our
results
in
this
study
demonstrate
that
our
artificial
exploratory

algorithm
can
surpass
this
capability
even
when
the
methods
were
designed
to

mimic
the
strategies
that
humans
employ
(but
see
Discussion).

155

Table
6-‐4:
Comparison
of
AB
Discrimination
to
Human
Subjects

Comparison
of
AB
discrimination
of
similar
texture
pairs
between
human
subjects
and
the
Bayesian

exploration
classifier.
In
all
cases
Bayesian
exploration
outperformed
human
subjects
with
many

pairs
of
textures
yielding
100%
classification
over
the
1000
simulations
for
each
pair.

Absolute
Texture
Classification

The
texture
classification
algorithm
was
validated
by
using
it
to
identify
the
best

match
from
the
117
textures
in
the
database
by
selecting
the
most
efficient

sequence
of
exploratory
movements
from
a
novel
set
of
data.
Figure
6-‐10
shows
a

few
examples
of
these
simulations,
which
exhibit
a
wide
range
of
sequences
of

exploratory
movements
and
properties,
depending
on
the
actual
texture
being

classified
and
those
in
the
data
set
with
which
it
might
be
most
easily
confused.
In

all
cases
the
first
movement
is
0.2N
and
6.31cm/s
to
determine
texture
roughness.

156

This
is
due
to
all
textures
starting
with
equal
probability
(1
divided
by
117).
In
this

scenario,
data
from
the
training
set
has
indicated
that
this
first
movement
will

produce
the
largest
benefit.
After
information
from
this
first
movement
is
collected,

each
simulation
went
through
a
set
of
exploratory
movements
that
was
optimal
for

discriminating
among
the
most
likely
candidates
for
the
particular
simulation.
This

was
found
to
be
unique
for
each
texture
and
even
different
between
simulations
of

the
same
texture
due
to
the
random
presentation
of
various
trials
from
the

validation
dataset.

157

Figure
6-‐10:
Evolution
of
Probabilities
Through
Bayesian
Exploration

158

Figure
6-‐10
caption:
Evolution
of
estimated
probabilities
as
virtual
exploratory
movements
are
made

to
identify
textures
from
the
entire
training
set
of
117
textures.
In
each
of
these
plots
the
steps
along

the
x-‐axis
represent
discrete
exploratory
movements,
and
the
y-‐axis
represents
the
estimated

probabilities
of
likely
texture
candidates.
The
movement
and
signal
taken
at
each
step
are
indicated

below
the
tick
marks
(Movements
(M):
1
=
1.26N,
1cm/s;
2
=
0.5N,
2.5cm/s;
3
=
0.2N,
6.31cm/s.

Signals
(S):
1
=
Traction,
2
=
Roughness,
3
=
Fineness).
The
color-‐coded
key
for
probability
traces

shows
the
numbers
of
the
textures
being
classified
in
the
validation
trial.
Dashed
line
represents
the

99%
confidence
required
to
end
the
simulation
before
all
10
movements
are
made.
In
(A)
texture
54

(Silicone)
was
rapidly
identified,
as
was
the
case
for
many
of
the
simulations.
In
(B)
texture
58

(Textured
Vinyl
#2)
was
eventually
identified
after
a
few
initially
more
probable
candidates
were

ruled
out.
In
(C)
texture
3
(smooth
cardstock)
is
shown
being
misidentified
as
balsa
wood
(T38)
and

in
(D)
correctly
identified,
although
with
only
60%
confidence
at
the
end
of
the
simulation.
In
both

cases
no
texture
reached
a
confidence
of
above
99%
to
stop
the
simulation
so
it
ran
for
the
complete

10
trials.

In
this
study
we
compared
the
Bayesian
exploration
algorithm
with
alternative

algorithms
such
as
cycling
through
the
most
useful
movements
for
each
signal
and

randomly
selecting
combinations
of
exploratory
movements
and
signals
to
measure.

A
summary
of
performance
for
the
8000
Monte
Carlo
simulations
of
the
16
textures

tested
is
provided
(Table
6-‐5).
Our
Bayesian
exploration
algorithm
was
found
to
be

superior
in
both
classification
accuracy
and
number
of
exploratory
movements

required
to
converge
to
99%
confidence.
Furthermore
the
Bayesian
exploration

strategy
was
more
likely
to
converge
on
the
correct
texture
before
reaching
the

maximum
of
10
movements.
Of
the
16
textures
explored
for
global
classification

among
the
set
of
117
textures
in
these
simulations,
Bayesian
exploration

outperformed
uninformed
cycling
and
random
selection
for
all
but
one
of
these

textures.

159

Table
6-‐5:
Summary
of
Performance
for
Bayesian
Exploration

Summary
of
performance
for
absolute
classification
task
for
uninformed
cycling,
random
selection,

and
Bayesian
Exploration.
A
total
of
8000
Monte
Carlo
simulations
for
16
textures
from
unique

validation
data
were
compared
against
the
training
data
from
all
117
textures
to
determine
which
of

the
117
textures
best
fit
the
observed
data
when
performing
virtual
explorations.
Results
of
Bayesian

exploration
are
compared
to
uninformed
cycling
through
exploratory
movements
between
the
three

signals
at
their
most
useful
movements
and
random
selection
of
exploratory
movements
from
all

combinations
of
movements
and
signals.
The
percentage
of
correct
identifications
are
shown
for
each.

The
algorithm
that
produced
the
best
performance
for
each
texture
is
highlighted
in
blue.
*For
the

case
of
uninformed
cycling
the
median
number
of
movements
to
convergence
could
not
be
obtained

as
the
simulation
was
stopped
at
10
movements
before
half
of
the
simulations
could
converge.

160

Discussion

Summary
of
Findings

Previous
investigations
into
the
psychophysics
of
textures
and
their

classification
from
sensory
data
have
looked
at
a
fairly
narrow
range
of
coarse

textures
and
exploratory
movements.
This
has
resulted
in
simplistic
classifiers

based
on
one
or
two
dimensions
of
the
sensory
information
that
tend
to
break
down

when
extrapolated
beyond
their
original
data.
Such
circumscription
makes
study

design
more
tractable
but
it
may
foreclose
opportunities
inherent
in
considering
the

larger
problem.
The
large
set
of
textures
and
large
range
of
movements
explored
in

this
study
forced
us
to
develop
systematic
and
scalable
methods
for
dealing
with
a

problem
whose
scale
is
more
similar
to
that
faced
by
the
human
nervous
system.

These
methods
extend
conventional
Bayesian
decision
making
to
encompass

optimal
strategies
for
acquiring
the
data
for
such
optimal
decision-‐making,

suggesting
the
term
Bayesian
exploration.

For
this
study
we
have
found
the
method
of
Bayesian
exploration
to
be
far

superior
to
other
methods
previously
used
for
discriminating
textures:
de
Boissieu

et
al.
(2009)
were
able
to
demonstrate
discrimination
among
10
different
textures

with
62%
performance
classification;
Giguere
and
Dudek
(2011)
were
able
to
obtain

a
classification
performance
of
89.9-‐94.6%
with
10
surfaces;
Oddo
et
al.
(2011)

demonstrated
a
classification
performance
of
97.6%
classification
across
three
fine

gratings;
Jamali
and
Sammut
(2011)
demonstrated
95%
classification
across
8

161

textures;
Sinapov
et
al.
(2011)
demonstrated
a
classification
performance
of
95%

across
20
different
textures
when
using
multiple
exploratory
movements.
Bayesian

exploration
yielded
a
performance
of
99.7%
when
choosing
between
two
difficult

textures,
surpassing
even
human
capabilities,
and
95.4%
when
choosing
from
a

database
of
117
textures.
Normally
classification
accuracy
would
be
expected
to

decline
as
the
number
of
possible
textures
increased.

A
guiding
principle
throughout
this
study
was
biomimicry.
We
used
a
tactile

sensor
that
shares
many
mechanical
features
with
the
human
fingertip
and
we
slid
it

over
textured
surfaces
with
exploratory
movements
similar
to
those
that
humans

make
when
exploring
textures.
The
exploratory
movements
and
properties
to

measure
were
inspired
by
observations
of
human
behavior
and
the
descriptive

language
that
humans
use
to
describe
textures.
The
Bayesian
exploration
algorithm

intelligently
selects
the
optimal
exploratory
movements
to
make
based
on
current

knowledge
of
what
the
texture
may
be
and
prior
experience
of
which
movement

would
best
discriminate
the
likely
possibilities,
a
method
also
inspired
by
theories

of
biological
behavior
(Loeb
et
al.
2011).
Gratifyingly,
we
were
able
to
obtain

performance
in
discriminating
textures
that
surpassed
human
capabilities
for
both

accuracy
and
speed
of
classification.

The
use
of
descriptive
properties
inspired
by
human
language
was
probably

important.
The
human
brain
is
an
outstanding
classifier,
so
naturally
one
would

expect
it
understands
what
makes
textures
different.
Therefore,
the
language
that

humans
use
to
describe
textures
are
inherently
low-‐hanging
fruit
for
inspiring

162

analytical
measures
of
texture
properties.
In
this
study
we
implemented
relatively

simple
algorithms
for
estimating
these
signal
properties
(motor
current
to
estimate

sliding
force,
vibration
power
to
estimate
roughness,
vibration
frequency
to

estimate
coarseness).
The
performance
of
our
classifier,
even
using
this
simplified

set
of
inputs,
far
exceeded
our
expectations
for
such
a
large
database
of
textures.

This
approach
of
language-‐guided
signals
may
be
useful
for
other
artificial

discrimination
tasks.

There
are,
of
course,
substantial
differences
between
our
machine
and
a
human

hand
or
even
other
robotic
systems.
On
the
plus
side,
the
human
fingertip
has
a

much
richer
set
of
sensors
than
the
BioTac,
which
has
a
similar
dynamic
range
but

lacks
the
spatial
resolution
of
the
dynamic
Meissner’s
receptors
in
the
individual

fingerprint
ridges
(L.
A.
Jones
&
Lederman
2006).
On
the
minus
side,
the
movements

of
the
human
hand
are
subject
to
considerably
more
motor
noise
than
our
electrical

motors
(K.
E.
Jones
et
al.
2002;
L.
A.
Jones
&
Lederman
2006).
Apparently
these
two

differences
tended
to
cancel
each
other
in
terms
of
overall
performance.
This
may

be
a
general
property
of
the
Bayesian
strategy
for
selection
of
exploratory

movements
and
interpretation
of
the
resulting
sensory
data.
Noise
affecting
either

the
movements
or
the
sensory
transduction
is
represented
automatically
in
the

database
and
biases
the
process
away
from
choices
that
provide
less
useful

information
for
any
reason.
Extending
this
algorithm
to
a
complete
robotic
system

working
in
unstructured
environments
is
expected
to
degrade
the
quality
of

measured
signals,
which
was
enhanced
by
the
careful
design
of
a
custom-‐built

163

experimental
apparatus.
In
particular,
the
actuators
in
humanoid
robots
are
likely
to

be
considerably
noisier
than
our
apparatus,
introducing
both
variability
into
the

exploratory
movements
and
noise
into
the
sensor
signals.
Additional
training
to

better
understand
the
characteristics
of
noise
and
variability
is
one
way
to

compensate
for
this.
We
expect
the
Bayesian
exploration
method
to
be
robust
to
this

and
evolve
to
make
the
most
of
available
information.

Considerations
for
Improving
the
Classifier

The
results
presented
in
this
study
implement
what
is
known
in
machine

learning
as
supervised
learning.
A
set
of
textures
and
their
properties
to
measure

were
given
to
the
classifier
and
it
was
told
that
the
textures
were
unique
and

therefore
belonged
to
separate
classes.
In
the
real
world,
the
existence
of
discrete

entities
must
be
inferred
in
the
first
place
from
the
clustering
of
data
points
that

may
arise
from
the
existence
of
multiple
discrete
entities,
continuous
gradations
of

material
properties,
or
simply
noise
in
the
measurement
system.
Any
novel
sensory

experience
might
be
taken
to
be
a
distorted
sampling
of
a
previously
known
entity

or
a
first
example
of
a
new
entity.
Such
situations
can
be
accommodated
by

extending
the
classification
algorithm
to
continuously
refine
its
experience
for

known
textures
as
well
as
to
identify
when
new
textures
are
encountered
and
a
new

entity
needs
to
be
added
to
the
database.
One
method
to
do
this
would
be
to

calculate
the
Bhattacharya
coefficient
between
the
object
currently
being
explored

and
the
existing
database
of
objects.
If
the
newly
observed
data
are
not
similar

164

enough
to
known
textures,
a
new
class
could
be
created.
In
addition
to
the

distributions
of
the
tactile
data
themselves,
the
classifier
may
be
able
to
use
other

information
such
as
the
visual
appearance
(used
by
our
subjects
when
first

comparing
the
two
similar
textures
in
the
discrimination
task)
or
the
probability

that
an
entity
could
have
changed
or
been
replaced
from
one
exploration
to
the
next.

In
our
system,
the
internal
representation
of
a
texture’s
properties
consists
only

of
a
mean
value
and
a
standard
deviation
for
each
property
and
each
movement
that

can
be
made.
After
successfully
identifying
a
texture,
the
system
would
benefit
from

adding
these
new
data
to
its
library
so
future
encounters
with
the
same
texture
will

be
identified
more
efficiently.
Furthermore,
as
more
explorations
are
made,
the
true

probability
density
function
could
be
identified,
which
may
in
fact
deviate
from
the

initially
assumed
normal
distributions.
This
would
serve
to
improve
classifier

performance.
Adding
these
results
to
the
system
would
also
increase
the
amount
of

training
data
it
has
available,
eventually
enabling
multivariate
analysis
as
opposed

to
the
univariate
methods
used
in
this
study
to
avoid
the
curse
of
dimensionality

(Jain
et
al.
2000).
Updating
the
mean
and
standard
deviation
with
the
new
data
can

accomplish
this,
but
it
is
not
trivial.
If
all
observations
are
assumed
to
be
equally

valid
regardless
of
when
they
occurred,
then
updating
requires
knowledge
also
of

the
previous
number
of
experiences
with
that
entity.
If
a
new
entity
is
created,
it
is

possible,
even
likely,
that
a
substantial
number
of
the
previous
observations
have

been
misclassified.
Creating
two
new
means
and
standard
deviations
from
one

previously
learned
distribution
may
not
be
feasible,
in
which
case
the
algorithm
will

165

need
to
“forget”
much
of
the
old
data
and
explore
the
two
new
entities
intensively
to

create
new
internal
representations.

Collecting
data
sufficient
for
multivariate
analysis
was
impractical
for
the
large

number
of
textures
employed
in
these
experiments,
but
something
like
it
may
be

feasible
over
the
life
of
an
organism
or
robotic
system
learning
progressively
about

its
world.
We
propose
that
a
strategy
of
initially
focusing
only
on
salient
properties

for
novice
systems
with
little
experience
is
preferable.
As
more
experience
is

obtained,
however,
such
systems
could
benefit
from
the
efficiency
of
multivariate

analysis.

Considerations
for
Identifying
Objects
by
all
Available
Sensory
Modalities

The
strategies
used
in
this
study
could
be
generally
applied
to
a
more
diverse

class
of
problems
involving
object
identification.
As
discussed
in
the
introduction,

texture
discrimination
is
only
a
small
subset
of
tools
that
humans
employ
when

discriminating
objects
by
touch,
others
include:
compliance,
thermal
properties,

shape,
volume
and
weight.
The
development
of
biologically
inspired
exploratory

movements
and
signal
measures
for
these
properties
would
enhance
the
capabilities

and
performance
of
the
system.
Furthermore,
these
must
be
integrated
with
other

exteroceptive
modalities
such
as
vision,
sound
and
smell.
Iterative
decisions
must
be

made
about
other
exploratory
movements
of
the
fingers,
the
eyes
(e.g.
saccadic
gaze

shifts)
and
other
attentive
mechanisms.
Anthropomorphic
robots
provide
both
the

need
and
the
ability
to
implement
biomimetic
strategies
for
coping
with
such
high

166

dimensional
data.
In
doing
so,
they
may
provide
insights
into
those
strategies
that

are
difficult
to
obtain
from
studying
biological
systems
alone.

167

Chapter
7:
Conclusions

Jeremy
A.
Fishel

This
dissertation
presents
the
development
of
a
robust
tactile
microvibration

sensor
and
its
application
to
texture
discrimination.
The
simple
and
elegant

technology
draws
inspiration
from
the
human
fingertip
in
size,
shape,
compliance

and
sensitivity
to
tactile
vibrations.
The
performance
in
sensitivity
with
respect
to

the
human
fingertip
surpassed
even
our
own
expectations,
but
its
utility
depends
on

an
application
that
makes
effective
use
of
such
sensitivity.
Continuing
along
the
path

of
“bio-‐inspiration,”
theories
were
developed
and
tested
for
how
humans
might
use

similar
tactile
information
for
a
texture
discrimination
task.
The
result
of
this
was

Bayesian
exploration,
which
appears
to
be
a
promising
general
algorithm
for

exploratory
decision-‐making
by
autonomous
robots.

What
is
Necessary
and
Sufficient
for
Sensing
Tactile
Microvibrations?

This
dissertation
is
divided
into
two
parts:
the
development
of
a
robust
tactile

sensor
capable
of
sensing
microvibrations
and
the
application
of
this
sensory

modality
to
texture
discrimination.
Perhaps
the
most
difficult
challenge
was

accepting
when
the
sensor
was
good
enough
and
that
it
was
time
to
move
on
to
the

application
phase.
Many
engineering
tools
are
available
to
understand
and
to

168

optimize
the
physical
mechanisms
of
systems.
The
BioTac
has
not
been

systematically
optimized;
there
is
still
much
that
might
be
done
to
enhance

sensitivity
by
changing
its
design
parameters,
including
skin
thickness,
skin
material

composition,
the
geometry
of
the
fingerprints,
the
composition
of
the
fluid,
etc.
It

became
quite
clear
at
the
transitional
phase
of
these
studies
that
optimizing
the

BioTac
for
vibration
sensing
could
become
a
dissertation
of
its
own.
Instead

thresholds
for
good
enough
needed
to
be
determined.

Nature,
specifically
human
capability,
was
referred
to
as
the
gold
standard
of

vibration
sensing.
Instead
of
arbitrarily
optimizing
the
sensitivity
of
the
BioTac
or

accepting
the
device
as-‐is,
we
sought
to
achieve
a
sensitivity
that
approximated

human
performance;
then
we
stopped.
In
many
regards
human
capabilities
provide

an
invaluable
reference
to
determine
these
requirements.
The
observation
that

humans
are
capable
of
doing
many
tactile
tasks
(such
as
discriminating
texture)

with
the
tactile
feedback
they
are
afforded
in
their
fingertips
presents
an
existence

proof
that
such
tools
are
sufficient
for
the
task.
That
is
not
to
say
that
these
tasks

could
not
be
done
even
better
with
enhanced
transducers
or
alternative
strategies.

Rather,
it
demonstrates
that
these
sensory
capabilities
(if
they
are
properly

understood)
embody
a
list
of
minimal
requirements
for
one
such
approach.
One

might
consider
human
tactile
capabilities
to
be
optimized
for
the
combination
of

tasks
that
humans
need
to
do
or,
perhaps
more
properly,
that
humans
have
wisely

matched
their
tasks
to
their
capabilities.
Mimicking
this
would
provide
an
optimal

starting
point
for
advanced
tactile
sensors
to
enable
robotics
with
human-‐like

169

dexterity
and
perception.
Furthermore,
mimicking
the
structures
of
the
human

fingertip,
such
as
the
fingerprints
in
Chapter
4,
can
reveal
useful
mechanisms
that

enhance
biological
function
and
can
be
applied
to
engineered
devices.

In
order
to
identify
whether
humanlike
sensitivity
to
vibration
was
actually

sufficient
to
achieve
humanlike
performance,
we
picked
an
application
for
which

vibration
information
appeared
to
be
necessary.
Discrimination
and
identification
of

fine
textures
was
found
to
be
quite
good,
but
there
remain
many
other
tasks
such
as

grip
control
and
tool
usage
for
which
vibration
appears
to
be
necessary.
Whether

the
BioTac
is
as
effective
in
these
tasks
remains
to
be
determined
and
will
be
the

focus
of
future
research.

Bayesian
Exploration

The
process
of
Bayesian
exploration
as
detailed
in
Chapter
6
can
be
extended
by

learning
new
exploratory
movements
and
signal
processing
strategies
to
extract

additional
object
properties.
Obvious
candidates
include
palpation
to
measure

compliance
or
static
contact
to
measure
thermal
properties
of
objects.
As
both
the

number
and
complexity
of
these
exploratory
movements
increase,
exhaustive

exploration
of
an
object
to
elucidate
all
of
its
properties
rapidly
becomes
impractical.

However,
Bayesian
exploration
provides
a
solution
to
this
by
selecting
the
optimal

movements
to
disambiguate
between
the
then
most
likely
candidates
of
objects;

those
movements
not
expected
to
contribute
to
this
disambiguation
are
ignored.

170

While
this
thesis
demonstrates
the
benefit
of
such
an
approach
for
texture

discrimination,
we
are
at
the
beginning
of
understanding
its
broader
capabilities.

Of
particular
interest
is
explaining
how
these
systems
might
theoretically
evolve

in
the
first
place.
In
Chapter
7
we
presented
the
goal
of
identifying
a
texture
and

proposed
that
properties
such
as
roughness,
traction
and
fineness
could
be
used
to

enable
discrimination.
While
we
were
able
to
obtain
success
by
using
human

language
as
an
inspiration
for
signal
property
measurement,
the
question
still

remains:
how
would
these
properties
evolve
as
useful
to
the
task
of
texture

discrimination
and
deserving
of
unique
language
in
the
first
place?

In
order
to
develop
a
hypothesis
that
answers
this
question,
the
utility
of

identifying
an
object
is
first
explored.
Identifying
an
object
simply
for
the
purpose
of

classification
in-‐and-‐of
itself
has
little
value.
Instead
consideration
of
why
an
object

needs
to
be
identified,
or
what
the
intended
action
after
identification
is,
can
provide

insight.
It
is
proposed
that
the
answers
can
be
simplified
into
two
categories:

identifying
a
specific
object
that
can
perform
a
needed
function
(such
as
your
car

keys
needed
to
start
your
car),
or
identifying
an
object
that
would
be
suitable
to

perform
a
task
(such
as
any
rigid
and
heavy
object
to
hit
a
small
nail
into
a
soft

material
when
a
hammer
cannot
be
found).
In
the
first
case
of
identifying
an
object,

distinguishing
features
need
to
be
recognized
(i.e.
what
makes
my
keys
different

from
your
keys).
In
the
second
case
a
set
of
properties
or
features
to
perform
the

task
are
identified
(i.e.
something
rigid
and
heavy
to
hit
a
nail,
preferably
with
a
long

handle
and
heavy
mass
at
the
end,
similar
to
but
not
necessarily
a
hammer).
In
both

171

cases,
the
relevant
properties
to
detect
are
known
from
previous
experience
with

the
task.
It
is
proposed
that
those
properties
that
were
useful
in
performing
the
task

in
the
past
will
be
sought
out
when
the
need
for
performing
that
task
in
the
future
is

presented
again.

Properties
that
are
frequently
desired
over
a
wide
range
of
tasks
(or
useful
for
a

particularly
important
task)
require
tests
to
identify
them.
For
haptic
perception,

these
tests
come
in
the
form
of
exploratory
movements
to
produce
sensory

perception
and
the
signal
processing
methods
to
interpret
this
data.
Similar
to
the

process
of
Bayesian
exploration,
the
benefit
of
these
tests
depend
on
how
well
they

produce
outputs
to
discriminate
between
objects
that
possess
a
certain
property

and
those
that
do
not.
Furthermore,
experience
plays
an
essential
role
in
the
ability

to
perform
these
tests.
This
is
a
fundamental
feature
in
all
classification
problems:

better
training
(experience)
yields
better
performance.
The
implications
of
this
are

intriguing
when
put
into
the
context
of
Bayesian
exploration,
which
actively
seeks

out
the
most
useful
movements
for
discrimination.
The
execution
of
these
most

useful
movements
more
often
than
others
results
in
additional
experience
that
the

system
can
rely
on
in
the
future,
enhancing
the
training
and
providing
further

reinforcement
for
the
benefit
of
these
movements
in
future
explorations.

The
summation
of
these
thoughts
leads
to
an
interesting
theory:
why
not
start

out
a
novice
system
by
trying
everything?
If
every
exploratory
movement
physically

possible
and
every
signal
processing
strategy
conceivable
was
attempted,
such
a

system
that
used
strategies
inspired
by
Bayesian
exploration
would
adaptively

172

converge
on
those
movements
and
signal
processing
strategies
that
offered
utility
in

performing
desired
tasks
and
ignoring
the
rest.
The
gained
experience
of
the
useful

tests
would
serve
to
reinforce
these
strategies
and
ultimately
a
smaller
set
of
useful

tests
would
evolve
on
their
own
depending
on
the
set
of
tasks
the
system
frequently

performs.
Such
an
approach
offers
promise
for
evolving
haptic
systems
rather
than

simply
programming
them.

Nature
to
Inspire
Robotics,
Robotics
to
Understand
Nature

The
accomplishments
of
this
thesis
contribute
to
the
advancement
of
practical

applications
of
robotic
tactile
sensing.
The
methods
and
strategies
used
were
deeply

inspired
by
biology
and
psychophysical
studies
of
human
performance.
The

biomimicry
approach
has
proven
fruitful
in
a
great
number
of
applications
to
inspire

advanced
artificial
systems,
now
including
haptically
enabled
robots.

Humans
possess
a
unique
combination
of
manual
dexterity,
a
dense
supply
of

cutaneous
neural
transducers,
and
highly
developed
brains
to
process
and
use
this

information.
Only
a
few
other
species
have
anywhere
near
this
sensitivity
and

dexterity
(e.g.
raccoon,
monkey),
but
they
are
difficult
to
train
to
perform
such
tasks

in
a
reproducible
and
observable
manner.
Instead,
we
have
proposed
the
use
of
a

robotic
model
for
testing
theories
of
biological
function
(Loeb
et
al.
2011).
The

ability
of
a
biomimetically
designed
robot
to
match
(or
even
exceed)
human

performance
in
a
difficult
perceptual
task
supports,
but
does
not
prove,
the

173

hypothesis
that
humans
use
similar
mechanisms
and
algorithms
for
sensing,

perception
and
exploratory
movements.

While
robotics
can
learn
a
lot
from
nature,
we
ask
the
question
what
can
we

learn
about
nature
using
robotics?
The
development
of
this
sensor
and
general

algorithms
for
its
use
provide
a
basis
for
future
research
that
can
answer
this

question.

174

References

Ando,
S.
&
Shinoda,
H.,
1995.
Ultrasonic
emission
tactile
sensing.
IEEE
Control

Systems,
15,
pp.61–69.

Augurelle,
A.S.
et
al.,
2003.
Importance
of
cutaneous
feedback
in
maintaining
a

secure
grip
during
manipulation
of
hand-‐held
objects.
Journal
of

Neurophysiology,
89(2),
pp.665–671.

Bensmaïa,
S.J.
&
Hollins,
M.,
2005.
Pacinian
representations
of
fine
surface
texture.

Perception
and
Psychophysics,
67(5),
pp.842–854.

Bensmaïa,
S.J.
&
Hollins,
M.,
2003.
The
vibrations
of
texture.
Somatosensory
and

Motor
Research,
20(1),
pp.33–43.

Bergmann
Tiest,
W.M.
&
Kappers,
A.M.L.,
2006.
Analysis
of
haptic
perception
of

materials
by
multidimensional
scaling
and
physical
measurements
of

roughness
and
compressibility.
Acta
Psychologica,
121,
pp.1–20.

Brisben,
A.J.,
Hsiao,
S.S.
&
Johnson,
K.O.,
1999.
Detection
of
vibration
transmitted

through
an
object
grasped
in
the
hand.
Journal
of
Neurophysiology,
81(4),

pp.1548–1558.

Chuang,
C.H.,
Dong,
W.B.
&
Lo,
W.B.,
2008.
Flexible
piezoelectric
tactile
sensor
with

structural
electrodes
array
for
shape
recognition
system.
In
IEEE

International
Conference
on
Sensing
Technology.
pp.
504–507.

Cole,
K.J.
&
Abbs,
J.H.,
1988.
Grip
force
adjustments
evoked
by
load
force

perturbations
of
a
grasped
object.
Journal
of
Neurophysiology,
60(4),

pp.1513–1522.

Cotton,
D.P.J.
et
al.,
2007.
A
novel
thick-‐film
piezoelectric
slip
sensor
for
a
prosthetic

hand.
IEEE
Sensors
Journal,
7(5),
pp.752–761.

Dahiya,
R.S.
&
Gori,
M.,
2010.
Probing
with
and
into
fingerprints.
Journal
of

Neurophysiology,
104,
pp.1–3.

Dahiya,
R.S.
et
al.,
2010.
Tactile
sensing—from
humans
to
humanoids.
IEEE

Transactions
on
Robotics,
26(1),
pp.1–20.

175

Dargahi,
J.,
Parameswaran,
M.
&
Payandeh,
S.,
2000.
A
micromachined
piezoelectric

tactile
sensor
for
an
endoscopic
grasper-‐theory,
fabrication
and
experiments.

Journal
of
Microelectromechanical
Systems,
9(3),
pp.329–335.

Darian-‐Smith,
I.
&
Oke,
L.E.,
1980.
Peripheral
neural
representation
of
the
spatial

frequency
of
a
grating
moving
across
the
monkey's
finger
pad.
Journal
of

Physiology,
309,
pp.117–133.

Darian-‐Smith,
I.,
Davidson,
I.
&
Johnson,
K.O.,
1980.
Peripheral
neural
representation

of
spatial
dimensions
of
a
textured
surface
moving
across
the
monkey's

finger
pad.
Journal
of
Physiology,
309,
pp.135–146.

Dario,
P.
&
Buttazzo,
G.,
1987.
An
anthropomorphic
robot
finger
for
investigating

artificial
tactile
perception.
International
Journal
of
Robotics
Research,
6(3),

pp.25–48.

Dario,
P.
et
al.,
1994.
An
investigation
on
a
robot
system
for
disassembly
automation.

In
IEEE
International
Conference
on
Robotics
and
Automation.
pp.
3515–
3521.

Dario,
P.
et
al.,
1984.
Ferroelectric
polymer
tactile
sensors
with
anthropomorphic

features.
In
IEEE
International
Conference
on
Robotics
and
Automation.
pp.

332–340.

de
Boissieu,
F.
et
al.,
2009.
Tactile
texture
recognition
with
a
3-‐axial
force
MEMS

integrated
artificial
finger.
In
Robotics:
Science
and
Systems.
pp.
49–56.

Dubey,
V.N.
&
Crowder,
R.M.,
2006.
A
dynamic
tactile
sensor
on
photoelastic
effect.

Sensors
and
Actuators,
128(2),
pp.217–224.

Edin,
B.B.,
Westling,
G.
&
Johansson,
R.S.,
1992.
Independent
control
of
human

finger-‐tip
forces
at
individual
digits
during
precision
lifting.
Journal
of

Physiology,
450,
pp.547–564.

Edwards,
J.
et
al.,
2008.
Extracting
textural
features
from
tactile
sensors.

Bioinspiration
&
Biomimetics,
3,
pp.1–12.

Fearing,
R.S.,
1987.
Some
experiments
with
tactile
sensing
during
grasping.
In
IEEE

International
Conference
on
Robotics
and
Automation.
pp.
1637–1643.

Fearing,
R.S.,
1990.
Tactile
sensing
mechanisms.
International
Journal
of
Robotics

Research,
9(3),
pp.3–23.

176

Fishel,
J.A.
et
al.,
2008.
A
robust
micro-‐vibration
sensor
for
biomimetic
fingertips.
In

IEEE/RAS-‐EMBS
International
Conference
on
Biomedical
Robotics
and

Biomechatronics.
pp.
659–663.

Fishel,
J.A.,
Loeb,
G.E.
&
(null),
2012.
Sensing
tactile
microvibrations
with
the
BioTac

–
comparison
with
human
sensitivity.
In
IEEE/RAS-‐EMBS
International

Conference
on
Biomedical
Robotics
and
Biomechatronics.

Fishel,
J.A.,
Loeb,
G.E.
&
(null),
2009.
The
role
of
fingerprints
in
vibrotactile

discrimination,
Whitepaper
for
DoD
Physics
of
Biology.

Freeman,
A.W.
&
Johnson,
K.O.,
1982.
Cutaneous
mechanoreceptors
in
macaque

monkey:
temporal
discharge
patterns
evoked
by
vibration,
and
a
receptor

model.
Journal
of
Physiology,
323,
pp.21–41.

Gao,
G.,
Wang,
Z.
&
Gao,
R.,
1990.
A
PVDF
film
sensor
for
material
identification.

Sensors
and
Actuators,
23,
pp.886–889.

Giguere,
P.
&
Dudek,
G.,
2011.
A
simple
tactile
probe
for
surface
identification
by

mobile
robots.
IEEE
Transactions
on
Robotics,
27(3),
pp.534–544.

Harmon,
L.D.,
1982.
Automated
tactile
sensing.
International
Journal
of
Robotics

Research,
1(2),
pp.3–32.

Hasegawa,
Y.
et
al.,
2006.
An
active
tactile
sensor
for
detecting
mechanical

characteristics
of
contacted
objects.
Journal
of
Micromechanics
and

Microengineering,
16(8),
pp.1625–1632.

Hillis,
W.D.,
1982.
A
high-‐resolution
imaging
touch
sensor.
International
Journal
of

Robotics
Research,
1(2),
pp.33–44.

Hollins,
M.
&
Risner,
S.R.,
2000.
Evidence
for
the
duplex
theory
of
tactile
texture

perception.
Perception
and
Psychophysics,
62(4),
pp.695–705.

Hollins,
M.
et
al.,
1993.
Perceptual
dimensions
of
tactile
surface
texture:
a

multidimensional
scaling
analysis.
Perception
and
Psychophysics,
54(6),

pp.697–705.

Hollins,
M.,
Bensmaïa,
S.J.
&
Washburn,
S.,
2001.
Vibrotactile
adaptation
impairs

discrimination
of
fine,
but
not
coarse,
textures.
Somatosensory
and
Motor

Research,
18(4),
pp.253–262.

Hollins,
M.,
Fox,
A.
&
Bishop,
C.,
2000.
Imposed
vibration
influences
perceived
tactile

smoothness.
Perception,
29(12),
pp.1455–1465.

177

Hosoda,
K.,
Tada,
Y.
&
Asada,
M.,
2006.
Anthropomorphic
robotic
soft
fingertip
with

randomly
distributed
receptors.
Robotics
and
Autonomous
Systems,
54(2),

pp.104–109.

Howe,
R.D.,
1991.
A
tactile
stress
rate
sensor
for
perception
of
fine
surface
features.

In
IEEE
International
Conference
on
Transducers.
pp.
864–867.

Howe,
R.D.,
1994.
Tactile
sensing
and
control
of
robotic
manipulation.
Advanced

Robotics,
8(3),
pp.245–261.

Howe,
R.D.
&
Cutkosky,
M.R.,
1993.
Dynamic
tactile
sensing:
perception
of
fine

surface
features
with
stress
rate
sensing.
IEEE
Transactions
on
Robotics
and

Automation,
9(2),
pp.140–151.

Howe,
R.D.
&
Cutkosky,
M.R.,
1989.
Sensing
skin
acceleration
for
slip
and
texture

perception.
In
IEEE
International
Conference
on
Robotics
and
Automation.

pp.
145–150.

Hutchings,
B.L.,
Grahn,
A.R.
&
Petersen,
R.J.,
1994.
Multiple-‐layer
cross-‐field

ultrasonic
tactile
sensor.
In
IEEE
International
Conference
on
Robotics
and

Automation.
pp.
2522–2528.

Jain,
A.K.,
Duin,
R.P.W.
&
Mao,
J.,
2000.
Statistical
pattern
recognition:
a
review.
IEEE

Transactions
on
Pattern
Analysis
and
Machine
Intelligence,
22(1),
pp.4–37.

Jamali,
N.
&
Sammut,
C.,
2011.
Majority
voting:
material
classification
by
tactile

sensing
using
surface
texture.
IEEE
Transactions
on
Robotics,
27(3),
pp.508–
521.

Johansson,
R.S.,
1978.
Tactile
sensibility
in
the
human
hand:
receptive
field

characteristics
of
mechanoreceptive
units
in
the
glabrous
skin
area.
Journal

of
Physiology,
281,
pp.101–125.

Johansson,
R.S.
&
Birznieks,
I.,
2004.
First
spikes
in
ensembles
of
human
tactile

afferents
code
complex
spatial
fingertip
events.
Nature
Neuroscience,
7(2),

pp.170–177.

Johansson,
R.S.
&
Flanagan,
J.R.,
2009.
Coding
and
use
of
tactile
signals
from
the

fingertips
in
object
manipulation
tasks.
Nature
Reviews:
Neuroscience,
10,

pp.345–359.

Johansson,
R.S.
&
Vallbo,
Å.B.,
1979a.
Detection
of
tactile
stimuli;
thresholds
of

afferent
units
related
to
psychophysical
thresholds
in
the
human
hand.

Journal
of
Physiology,
297,
pp.405–422.

178

Johansson,
R.S.
&
Vallbo,
Å.B.,
1979b.
Tactile
sensibility
in
the
human
hand:
relative

and
absolute
densities
of
four
types
of
mechanoreceptive
units
in
glabrous

skin.
Journal
of
Physiology,
286(1),
p.283.

Johansson,
R.S.
&
Vallbo,
Å.B.,
1983.
Tactile
sensory
coding
in
the
glabrous
skin
of

the
human
hand.
Trends
in
Neurosciences,
6,
pp.27–32.

Johansson,
R.S.
&
Westling,
G.,
1988.
Programmed
and
triggered
actions
to
rapid

load
changes
during
precision
grip.
Experimental
Brain
Research,
71(1),

pp.72–86.

Johansson,
R.S.
&
Westling,
G.,
1984.
Roles
of
glabrous
skin
receptors
and

sensorimotor
memory
in
automatic
control
of
precision
grip
when
lifting

rrougher
or
more
slippery
objects.
Experimental
Brain
Research,
56,
pp.550–
564.

Johansson,
R.S.
&
Westling,
G.,
1987.
Signals
in
tactile
afferents
from
the
fingers

eliciting
adaptive
motor
responses
during
precision
grip.
Experimental
Brain

Research,
66,
pp.141–154.

Johansson,
R.S.,
Häger,
C.
&
Bäckström,
L.,
1992.
Somatosensory
control
of
precision

grip
during
unpredictable
pulling
loads.
III.
Impairments
during
digital

anesthesia.
Experimental
Brain
Research,
89(1),
pp.204–213.

Johansson,
R.S.,
Landström,
U.
&
Lundström,
R.,
1982.
Responses
of

mechanoreceptive
afferent
units
in
the
glabrous
skin
of
the
human
hand
to

sinusoidal
skin
displacements.
Brain
Research,
244,
pp.17–25.

Johnson,
K.O.,
2001.
The
roles
and
functions
of
cutaneous
mechanoreceptors.

Current
Opinion
in
Neurobiology,
11(4),
pp.455–461.

Johnson,
K.O.
&
Hsiao,
S.S.,
1992.
Neural
mechanisms
of
tactual
form
and
texture

discrimination.
Annual
Review
of
Neuroscience,
15,
pp.227–250.

Johnson,
K.O.
&
Phillips,
J.R.,
1981.
Tactile
spatial
resolution.
i.
two-‐point

discrimination,
gap
detection,
grating
resolution,
and
letter
recognition.

Journal
of
Neurophysiology,
46(6),
pp.1177–1192.

Johnson,
K.O.,
Yoshioka,
T.
&
Vega-‐Bermudez,
F.,
2000.
Tactile
functions
of

mechanoreceptive
afferents
innervating
the
hand.
Journal
of
Clinical

Neurophysiology,
17(6),
p.539.

Jones,
K.E.,
Hamilton,
A.F.
&
Wolpert,
D.M.,
2002.
Sources
of
signal-‐dependent
noise

during
isometric
force
production.
Journal
of
Neurophysiology,
88(3),

pp.1533–1544.

179

Jones,
L.A.
&
Lederman,
S.J.,
2006.
Human
Hand
Function,
New
York,
NY:
Oxford

University
Press,
USA.

Kamata,
M.
et
al.,
2005.
Optical
vibration
sensor
fabricated
by
femtosecond
laser

micromachining.
Applied
Physics
Letters,
87(051106),
pp.1–3.

Kandel,
E.R.,
Schwartz,
J.H.
&
Jessell,
T.M.,
2000.
Principles
of
neural
science,
McGraw-‐
Hill/Appleton
&
Lange.

Katz,
D.,
1925.
The
World
of
Touch,
Hillsdale,
NJ:
Lawrence
Erlbaum.

Klatzky,
R.L.
&
Lederman,
S.J.,
2010.
Multisensory
texture
perception.
In

Multisensory
Object
Perception
in
the
Primate
Brain.

Springer
Verlag,
pp.

211–230.

Knibestöl,
M.
&
Vallbo,
Å.B.,
1970.
Single
unit
analysis
of
mechanoreceptor
activity

from
the
human
glabrous
skin.
Acta
Physiologica
Scandinavica,
80(2),

pp.178–195.

Ko,
C.T.,
Tseng,
S.H.
&
Lu,
M.S.C.,
2006.
A
CMOS
micromachined
capacitive
tactile

sensor
with
high-‐frequency
output.
Journal
of
Microelectromechanical

Systems,
15(6),
pp.1708–1714.

Kolesar,
E.S.,
Jr
&
Dyson,
C.S.,
1995.
Object
imaging
with
a
piezoelectric
robotic

tactile
sensor.
Journal
of
Microelectromechanical
Systems,
4(2),
pp.87–96.

Krishna,
G.M.
&
Rajanna,
K.,
2004.
Tactile
sensor
based
on
piezoelectric
resonance.

IEEE
Sensors
Journal,
4(5),
pp.691–697.

Lamb,
G.D.,
1983a.
Tactile
discrimination
of
Textured
surfaces:
peripheral
neural

coding
in
the
monkey.
Journal
of
Physiology,
338,
pp.567–587.

Lamb,
G.D.,
1983b.
Tactile
discrimination
of
textured
surfaces:
psychophysical

performance
measurements
in
humans.
Journal
of
Physiology,
338,
pp.551–
565.

LaMotte,
R.H.
&
Srinivasan,
M.A.,
1991.
Surface
microgeometry:
tactile
perception

and
neural
encoding.
In
Wenner-‐Gren
International
Symposium
Series.
pp.

49–58.

Larcombe,
M.H.E.,
1981.
Why
carbon
fibres
can
give
robots
a
strong
sense
of
grip.

Sensor
Review,
1(2),
pp.58–59.

180

Lederman,
S.J.,
1974.
Tactile
roughness
of
grooved
surfaces:
the
touching
process

and
effects
of
macro-‐and
microsurface
structure.
Perception
and

Psychophysics,
16(2),
pp.385–395.

Lederman,
S.J.,
1981.
The
perception
of
surface
roughness
by
active
and
passive

touch.
Bulletin
of
the
Psychonomic
Society,
18(5),
pp.253–255.

Lederman,
S.J.
&
Klatzky,
R.L.,
1987.
Hand
movements:
a
window
into
haptic
object

recognition.
Cognitive
Psychology,
19,
pp.342–368.

Lederman,
S.J.,
Loomis,
J.M.
&
Williams,
D.A.,
1982.
The
role
of
vibration
in
the

tactual
perception
of
roughness.
Perception
and
Psychophysics,
32(2),

pp.109–116.

Lee,
M.H.
&
Nicholls,
H.R.,
1999.
Tactile
sensing
for
mechatronics
-‐
a
state
of
the
art

survey.
Mechatronics,
9,
pp.1–31.

Lin,
C.H.
et
al.,
2009.
Signal
processing
and
fabrication
of
a
biomimetic
tactile
sensor

array
with
thermal,
force
and
microvibration
modalities.
In
IEEE

International
Conference
on
Robotics
and
Biomimetics.
pp.
129–134.

Loeb,
G.E.
et
al.,
2011.
Understanding
haptics
by
evolving
mechatronic
systems.

Progress
in
Brain
Research,
192,
pp.129–144.

Loewenstein,
W.R.
&
Skalak,
R.,
1966.
Mechanical
transmission
in
a
Pacinian

corpuscle:
an
analysis
and
a
theory.
Journal
of
Physiology,
182(2),
pp.346–
378.

Macefield,
V.G.,
Häger-‐Ross,
C.
&
Johansson,
R.S.,
1996.
Control
of
grip
force
during

restraint
of
an
object
held
between
finger
and
thumb:
responses
of
cutaneous

afferents
from
the
digits.
Experimental
Brain
Research,
108,
pp.155–171.

Maeno,
T.,
Kobayashi,
K.
&
Yamazaki,
N.,
1998.
Relationship
between
the
structure
of

human
finger
tissue
and
the
location
of
tactile
receptors.
Bulletin
of
JSME

International
Journal,
41(1),
pp.94–100.

Maheshwari,
V.
&
Saraf,
R.F.,
2006.
High-‐resolution
thin-‐film
device
to
sense
texture

by
touch.
Science,
312(5779),
pp.1501–1504.

Mirbagheri,
A.
et
al.,
2007.
Design,
fabrication,
and
testing
of
a
membrane

piezoelectric
tactile
sensor
with
four
sensing
elements.
American
Journal
of

Applied
Sciences,
4(9),
pp.645–652.

181

Miyaoka,
T.,
Mano,
T.
&
Ohka,
M.,
1999.
Mechanisms
of
fine-‐surface-‐texture

discrimination
in
human
tactile
sensation.
Journal
of
Acoustical
Society
of

America,
105(4),
pp.2485–2492.

Mountcastle,
V.B.,
LaMotte,
R.H.
&
Carli,
G.,
1972.
Detection
thresholds
for
stimuli
in

humans
and
monkeys:
comparison
with
threshold
events
in

mechanoreceptive
afferent
nerve
fibers
innervating
the
monkey
hand.

Journal
of
Neurophysiology,
35(1),
pp.122–136.

Mountcastle,
V.B.,
Steinmetz,
M.A.
&
Romo,
R.,
1990.
Frequency
discrimination
in
the

sense
of
flutter:
psychophysical
measurements
correlated
with
postcentral

events
in
behaving
monkeys.
Journal
of
Neuroscience,
10(9),
pp.3032–3044.

Mukaibo,
Y.
et
al.,
2005.
Development
of
a
texture
sensor
emulating
the
tissue

structure
and
perceptual
mechanism
of
human
fingers.
In
IEEE
International

Conference
on
Robotics
and
Automation.
IEEE,
pp.
2565–2570.

Nicholls,
H.R.
&
Lee,
M.H.,
1989.
A
survey
of
robot
tactile
sensing
technology.

International
Journal
of
Robotics
Research,
8(3),
pp.3–30.

Oddo,
C.M.
et
al.,
2009.
Artificial
roughness
encoding
with
a
bio-‐inspired
MEMS-‐
based
tactile
sensor
array.
Sensors,
9,
pp.3161–3183.

Oddo,
C.M.
et
al.,
2011.
Roughness
encoding
for
discrimination
of
surfaces
in

artificial
active-‐touch.
IEEE
Transactions
on
Robotics,
27(3),
pp.522–533.

Patterson,
R.W.
&
Nevill,
G.E.,
Jr.,
1986.
The
induced
vibration
touch
sensor–a
new

dynamic
touch
sensing
concept.
Robotica,
4,
pp.27–31.

Pruski,
A.
&
Mutel,
B.,
1984.
Direct
contact
sensors
based
on
carbon
fibre.
In
IEEE

International
Conference
on
Robotics
and
Factories
of
the
Future.
IEEE

International
Conference
on
Robotics
and
Factories
for
the
Future,
pp.
409–
415.

Rebguns,
A.,
Ford,
D.G.
&
Fasel,
I.,
2011.
InfoMax
control
for
acoustic
exploration
of

objects
by
a
mobile
robot.
In
AAAI
Conference
on
Artificial
Intelligence.
pp.

22–28.

Romano,
J.M.
&
Kuchenbecker,
K.J.,
2011.
Creating
realistic
virtual
textures
from

contact
acceleration
data.
IEEE
Transactions
on
Haptics.

Romano,
J.M.
et
al.,
2011.
Human-‐inspired
robotic
grasp
control
with
tactile
sensing.

IEEE
Transactions
on
Robotics,
27(6),
pp.1067–1079.

182

Rothwell,
J.C.
et
al.,
1982.
Manual
motor
performance
in
a
deafferented
man.
Brain,

105,
pp.515–542.

Scheibert,
J.
et
al.,
2009.
The
role
of
fingerprints
in
the
coding
of
tactile
information

probed
with
a
biomimetic
sensor.
Science,
323(5920),
pp.1503–1506.

Schmidt,
P.A.,
Maël,
E.
&
Würtz,
R.P.,
2006.
A
sensor
for
dynamic
tactile
information

with
applications
in
human-‐robot
interaction
and
object
exploration.

Robotics
and
Autonomous
Systems,
54(12),
pp.1005–1014.

Shinoda,
H.
&
Ando,
S.,
1994.
Ultrasonic
emission
tactile
sensor
for
contact

localization
and
characterization.
In
IEEE
International
Conference
on

Robotics
and
Automation.
pp.
2536–2543.

Shinoda,
H.,
Matsumoto,
K.
&
Ando,
S.,
1997.
Acoustic
resonant
tensor
cell
for
tactile

sensing.
In
IEEE
International
Conference
on
Robotics
and
Automation.
pp.

3087–3092.

Shinoda,
H.,
Uehara,
M.
&
Ando,
S.,
1993.
A
tactile
sensor
using
three-‐dimensional

structure.
In
IEEE
International
Conference
on
Robotics
and
Automation.

IEEE,
pp.
435–441.

Sinapov,
J.
&
Stoytchev,
A.,
2010.
The
boosting
effect
of
exploratory
behaviors.
In

AAAI
Conference
on
Artificial
Intelligence.
pp.
1613–1618.

Sinapov,
J.
et
al.,
2011.
Vibrotactile
recognition
and
categorization
of
surfaces
by
a

humanoid
robot.
IEEE
Transactions
on
Robotics,
27(3),
pp.488–497.

Smith,
A.M.,
Chapman,
C.E.,
Deslandes,
M.,
Langlais,
J.,
et
al.,
2002a.
Role
of
friction

and
tangential
force
variation
in
the
subjective
scaling
of
tactile
roughness.

Experimental
Brain
Research,
114,
pp.211–223.

Smith,
A.M.,
Gosselin,
G.
&
Houde,
B.,
2002b.
Deployment
of
fingertip
forces
in
tactile

exploration.
Experimental
Brain
Research,
147(2),
pp.209–218.

Son,
J.S.,
Monteverde,
E.A.
&
Howe,
R.D.,
1994.
A
tactile
sensor
for
localizing

transient
events
in
manipulation.
In
IEEE
International
Conference
on

Robotics
and
Automation.
pp.
471–476.

Speeter,
T.H.,
1990.
A
tactile
sensing
system
for
robotic
manipulation.
International

Journal
of
Robotics
Research,
9(6),
pp.25–36.

Srinivasan,
M.A.,
1989.
Surface
deflection
of
primate
fingertip
under
line
load.

Journal
of
Biomechanics,
22(4),
pp.343–349.

183

Srinivasan,
M.A.,
Whitehouse,
J.M.
&
LaMotte,
R.H.,
1990.
Tactile
detection
of
slip:

surface
microgeometry
and
peripheral
neural
codes.
Journal
of

Neurophysiology,
63(6),
pp.1323–1332.

Tada,
Y.
et
al.,
2003.
Sensing
the
texture
of
surfaces
by
anthropomorphic
soft

fingertips
with
multi-‐modal
sensors.
In
IEEE
International
Conference
on

Intelligent
Robots
and
Systems.
IEEE,
pp.
31–35.

Tada,
Y.,
Hosoda,
K.
&
Asada,
M.,
2004.
Sensing
ability
of
anthropomorphic
fingertip

with
multi-‐modal
sensors.
In
IEEE
International
Conference
on
Intelligent

Autonomous
Systems.
Citeseer,
pp.
1005–1012.

Vallbo,
Å.B.
&
Johansson,
R.S.,
1984.
Properties
of
cutaneous
mechanoreceptors
in

the
human
hand
related
to
touch
sensation.
Human
neurobiology,
3(1),
pp.3–
14.

Vallbo,
Å.B.
&
Johansson,
R.S.,
1978.
The
tactile
sensory
innervation
of
the
glabrous

skin
of
the
human
hand.
In
G.
Gordon,
ed.
Active
Touch,
the
Mechanism
of

Recognition
of
Objects
by
Manipulation.
Oxford:
Pergamon
Press
Ltd.,
pp.
29–
54.

Vranish,
J.M.,
1986.
Magnetoinductive
skin
for
robots.
In
IEEE
International

Conference
on
Robotics
and
Automation.
pp.
1292–1318.

Westling,
G.
&
Johansson,
R.S.,
1984.
Factors
influencing
the
force
control
during

precision
grip.
Experimental
Brain
Research,
53,
pp.77–284.

Westling,
G.
&
Johansson,
R.S.,
1987.
Responses
in
glabrous
skin
mechanoreceptors

during
precision
grip
in
humans.
Experimental
Brain
Research,
66,
pp.128–
140.

Wettels,
N.
&
Loeb,
G.E.,
2011.
Haptic
feature
extraction
from
a
biomimetic
tactile

sensor:
force,
contact
location
and
curvature.
In
IEEE
International

Conference
on
Robotics
and
Biomimetics.

Wettels,
N.,
Santos,
V.J.,
et
al.,
2008a.
Biomimetic
tactile
sensor
array.
Advanced

Robotics,
22(7),
pp.829–849.

Wettels,
N.,
Smith,
L.M.,
et
al.,
2008b.
Deformable
skin
design
to
enhance
response
of

a
biomimetic
tactile
sensor.
In
IEEE/RAS-‐EMBS
International
Conference
on

Biomedical
Robotics
and
Biomechatronics.
pp.
132–137.

Wiertlewski,
M.,
Hudin,
C.
&
Hayward,
V.,
2011.
On
the
1/f
noise
and
non-‐integer

harmonic
decay
of
the
interaction
of
a
finger
sliding
on
flat
and
sinusoidal

surfaces.
In
IEEE
World
Haptics
Conference.
IEEE,
pp.
25–30.

184

Winfield,
L.
et
al.,
2007.
T-‐pad:
tactile
pattern
display
through
variable
friction

reduction.
In
EuroHaptics/World
Haptics
Joint
Conference
and
Symposium

on
Haptic
Interfaces
for
Virtual
Environment
and
Teleoperator
Systems.
Ieee,

pp.
421–426.

Yamada,
Y.
&
Cutkosky,
M.R.,
1994.
Tactile
sensor
with
3-‐axis
force
and
vibration

sensing
functions
and
its
application
to
detect
rotational
slip.
In
IEEE

International
Conference
on
Robotics
and
Automation.
pp.
3550–3557.

Yao,
H.Y.
&
Hayward,
V.,
2010.
Design
and
analysis
of
a
recoil-‐type
vibrotactile

transducer.
Journal
of
Acoustical
Society
of
America,
128(2),
pp.619–627.

Yau,
J.M.,
Hollins,
M.
&
Bensmaïa,
S.J.,
2009.
Textural
timbre:
the
perception
of

surface
microtexture
depends
in
part
on
multimodal
spectral
cues.

Communicative
&
Integrative
Biology,
2(4),
pp.1–5.

Asset Metadata

Creator Fishel, Jeremy A. (author)

Core Title Design and use of a biomimetic tactile microvibration sensor with human-like sensitivity and its application in texture discrimination using Bayesian exploration

Contributor Electronically uploaded by the author (provenance)

School Andrew and Erna Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Biomedical Engineering

Publication Date 06/20/2012

Defense Date 06/04/2012

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag artificial texture discrimination,Bayesian exploration,Bayesian inference,biomimetic,BioTac,classification,fluid-filled sensor,human performance,impact,machine learning,microvibration,oai:digitallibrary.usc.edu:usctheses,OAI-PMH Harvest,object identification,Pacinian corpuscle,robot,tactile sensing,tactile sensor,texture discrimination,Touch,vibration

Language English

Advisor Valero-Cuevas, Francisco J. (committee chair), Loeb, Gerald E. (committee member), Schaal, Stefan (committee member), Yen, Jesse T. (committee member)

Creator Email jeremyfishel@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c3-47124

Unique identifier UC11290122

Identifier usctheses-c3-47124 (legacy record id)

Legacy Identifier etd-FishelJere-894.pdf

Dmrecord 47124

Document Type Dissertation

Rights Fishel, Jeremy A.

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Abstract (if available)

Abstract The cutaneous sensing of microvibrations in human fingertips plays a central role in the detection of slip-related and dynamic information critical for tool usage, reflexive grip control, and the perception of microtextures. This is made possible by the Pacinian corpuscle, a small sensory receptor in subcutaneous tissues that is capable of detecting vibrations up to 1000Hz in frequency. These receptors are sensitive to vibrations less than a micrometer in amplitude at frequencies around their maximal sensitivity of 250Hz. Artificial systems seeking to provide human-like dexterity and perception would benefit from similar sensory capabilities. ❧ In this dissertation, a novel tactile sensor capable of robustly sensing vibrations with a bandwidth and sensitivity that exceeds human performance is presented. The device, known as the BioTac, has a biomimetic structure that consists of a rigid bone-like core covered with an elastomeric skin. The space between the skin and core is filled with an incompressible low-viscosity liquid that is in contact with a pressure transducer. Vibrations that originate on the surface of the skin are transmitted as sound waves through the liquid and are readily sensed by the transducer. The incompressible liquid conducts these acoustic signals with little attenuation, permitting the transducer to be located in a protected region inside the core of the device, where it is less likely to get damaged. The addition of a biologically inspired fingerprint-like pattern on the surface of the skin was found to enhance vibrations sensed by the BioTac. The BioTac exceeded human capabilities in sensitivity thresholds to applied sinusoidal vibrations and impacts from small spheres. ❧ Biologically inspired strategies to use this tactile information for a texture discrimination task were developed. A specialized robot was built to make sliding movements similar to those humans make when exploring textures. The BioTac was slid over a total of 117 different textured surfaces collected from art supply, fabric and hardware stores. Signal processing methods were developed to extract properties modeled after the descriptive language that humans use when describing textures (rough/smooth, sticky/slippery, coarse/fine). Different sliding exploratory movements (defined by a combination of contact force and sliding velocity) were found to be optimal for discriminating each of these properties. All 117 textures were tested repeatedly with these three exploratory movements to collect a database of prior experience similar to human memory. A novel process of intelligently selecting exploratory movements was developed to guide the discrimination task when presented with an unknown texture. When exploring a texture, the Bayesian exploration algorithm selects the optimal movement to make and the property to measure based on previous experience to disambiguate the most-probable candidates. The combination of biomimetic hardware and software achieved performance that matched (and even surpassed) human capabilities in discriminating and identifying textures.