Elan Microelectronics Corporation v. Apple, Inc.

Filing 293

Declaration of Jennifer Liu in Support of Plaintiff Elan Microelectronics Corporation's Reply to Apple, Inc.'s Opposition to Elan's Motion for Partial Summary Judgment of Infringement filed byElan Microelectronics Corporation. (Attachments: # 1 Exhibit A, # 2 Exhibit B, # 3 Exhibit C, # 4 Exhibit D, # 5 Exhibit E, # 6 Exhibit F, # 7 Exhibit G, # 8 Exhibit H, # 9 Exhibit N, # 10 Exhibit O)(Liu, Jennifer) (Filed on 6/16/2011)

EXHIBIT H

The nternationtf

Journal

of

tnt

of

robot

SE

Cur

ic reerch
Shel

UNIVERSITV
SN DIEGO
Received

VoL

1982-

Cur

7\

f\L
/\/

OF

on
no

CALIFORNIA

LI8RRIES
Q627-9Q

Cprinq

Editor

The

Michael

Brady Oxford

Paul University of Pennsylvania

Richard

Number

Volume
Editorial

ISSN 0278-3649

Research
review

June

April

December

and

August October

bimonthly

is

published February

London

Offices

changes

and mailing

Tactile

Sensing

Journals

to

55

$140.00

$17.00

Single

of

copies

missing

issues

issues

free

claims

617

Department

tocopy

Adaptive

Chung and Gary

the

clients

is

for

Robotic

directly

MA

Salem
of

the

fee

$5.00

that

have

with

license

been

CCC

1990

Sadegh and Roberto Horowitz

fee

code

Reporting
For

Dataflow
Steve

pro

74

Multiprocessor

Gef/in

System

for

Robot

Arm

Control

93

and Borko Furht

those

granted

BOOK REVIEW

is

per copy

Street
for users

Advanced
Service

Research

in

VLSL

Proceedings

of the 1989 Decennial

is

Caltech
organiza

Conference

104

photocopy
system

of

Institute

separate

payment has been

Controllers

with the

27 Congress

The

Transactional

0278-3649/90
tions

CCC

to

Class of Adaptive

pho

to

copy-

Service

of $5.00

01970

Analysis of

Manipulators

CCC

Center

Transactional Reporting
the

the

by

registered

Copyright Clearance

paid

63

Leininger

or personal

granted

for users

and Robustness

Stability

issue

253-2889

or personal use of

internal

that

Hybrid Manipulator Control

for

Permission

for internal

articles

use or

vided

to an Assembly Task

49

Task-Level

Nader

owner

Applied

must be made immediately

Copyright Information

Lght

Dexterity

Sturges

Jack

air

for

$28.00

of the next published

receipt

of Machine

Subscribers

current

To be honored

Circulation

Street

for individuals

postage

$25.00

24

and Canada add

States

for surface

Paul

Depart

Hayward

for institutions

the

with Sliding Contacts

Manipulation

and

Quantification

United

outside

Dexterous

Trinkle

02

Rates $70.00

Subscription

for

Planning

correspondence

list

Press

MA

Cambridge

specific

Mechanisms

Fearing

address

Subscriptions

be addressed

ment The MIT

upon

MIT

and

England

Business

should

the

by

Cambridge Massachusetts

Press

mail

1990

Contents

Journal of Robotics

International

and

June

Assistant

Batten

Jennet

The

Journal of

Robotics Research

Editor

Founding

International

University

of

arranged

The Massachusetts

Technology

Postmaster

Send

55

Research

Cambridge
paid

at

Advertising

55

changes

to

Street

and

Journals

Street

Second

class

at additional

Advertising

Press

Hayward

617

Hayward
02142

MA

Boston

The MIT

address

Journal of Robotics

International

The

post

postage
offices

Manager
Department

Cambridge

MA

02142

253-2866

APE L0007542

Fearing
of Electrical

Department

Tactile Sensing

and Computer

Engineering

Mechanisms

Science
of California

University

California 94720

Berkeley

Abstract

from

orientation

Cameron

in
This
tactile

sensor

and

has been

depth

made between

been

dimensional
and

sensor

with

implemented

Jbr

are

sensitivity

contact

and

has

simple

one

The

1985
has an

contact

have

sensors

been

element

tactile

of

for

dextrous

trol for

dextrous

mine contact

with

Some

point

it

nice to

to

sensors to

finger

the finger

at

1984

It

is

necessary to

contact

on the

finger

for

know

accurate

This article addresses tactile finger
transduction

mechanism

tact

and magnitude using

location

niques

paper

Fearing

987b

iffTnverse

force

of the

linear

of

to

con
tech

based

on

curvature

and

hemispherical
ing elements

and

International

Vol
1990

No

June

Journal of Robotics

Research

1990

Massachusetts

Institute

of Technology

12

for

total

133-element

and
The

is

is

rubber
in

is

finger

Figure

on

the

cy
the

underneath

12

sens

capacitive

on cylinder
connected

externally

to

now

work

adequate

cylindrical
in

Fearing
for local
finger

the

in

sensor density has been

Other

256-element

roll

tactile

tactels

of

device described

main goal

shown

for

better

subset

The

and

gauges

but an

manipulation

are

elements

from previously reported

determination

1985
1988

as

elements

Only an

aliasing

for

molded

hand

tactile

gives

Strain

of force

shapes

in

and

electronics

reduce

tact

tip

point

Fearing 1986

under the tip

creased
to

The

12

portion

contacts

describing surfaces

fingers

are

the object

if

sensor that

single

Round

the Stanford/JPL

and

single

multiple

quantities

for

hands

about

ing objects

interface

recover surface

better

fool

sensors are not very useful

dextrous

lindrical

discusses

filtering

is

Object

other as well as the object

at

measuring

objects

measure

using

type

addition

and can

mea

difficult

is

It

contact

each

force

There

the

filtering

1988

with

In

touch

tip for

application

and determination

for

surface

explore

single

one

to

simultaneous

the object

sensor array was packaged

design issues

Fearing and Binford

nor
con

Fearing

the location

good

flat

to

parameters

example

are

array sensor

tips

surface

common

only the resultant

deter

to

and type of

curvature
see for

are quite

con

fingers

with

with the fingers

have

determine

where the

impor

force

shown the need

of the most useful

plane

line

is

multifingered

angle and magnitude of force

are

mals location of contact
tact

open-loop

hand have

types and location

Fearing 1986
recover

with

manipulation

Experiments

for

advantage

However

sensors

must remain grasped

hands

al

et

measurement

equivalent

while being manipulated

time consuming

fingers

be

sensor can

over the whole finger

ment

tant

Hillis

Siegel

single

measurements

tactile

surement at each

from the

from

shape

In one sense

and

force

Introduction

information

1987 and

transducers

1984 Chun and Wise

motion makes

contact

structure

and Dario

Bicchi

of deflection

location

gives

gauge

Preliminary

total

move

Local

strain

in

contact

obtaining

array sensor approach

1985

discussed

are also

sensor that

using

uses arrays

Boie

two-

to

finger-tip
force

Brock and Chiu 1985
that

measurements

and Speeter1987

approaches

are

resultant

1981

contact

line

formation

and

of strain

set

1988

et

different

determining

analyzed

normal deflection

only

of

of skin thickness

effects

location

infor

localization

comparison

sensor

model The

basic

of appropriate

analysis

accurate

the cylindrical

on

force

An

allow

Two

cylindrical

planar linear elas

spatial aliasing

methods for determining
tangential

to

of

for determining

contact

stress-strain

depth

using

analysis

used

and reduced

contacts

design

analysis

signal

mation about the applied
sensor

themes

mechanical

models

ticity

main

has three

paper

static

and inversion methods

Transduction

987a
con

tips are

Allen and Bajczy

fiber-optic

device Begej

the use of this finger

as

an

ex

APE
Fearing

LJOO7543

Tactile

Fig

sensing

Fig

finger

hand

Stanford/JPL

for

sensor

Capacitive

block

Drive

diagram

100

Ki-Ix

ultiplexor

Multiplexer

Output

Detector

device

perimental

The

search

and
It

in

is

not intended

not robust

probably

however

is

sensory and manipulation

for
is

finger

rather

as

enough

final

re

prototype

and reasonably

sensitive

use

for industrial

durable

laboratory use

output

improve

Hz

to

been

speed

were

tip

good

so that

many

designing

and low

sensitivity

designs could

be

quickly

and the success

with

Siegel

arrays

Bole and Miller

from capacitors
oped here

of other workers

al 1985

Boie

1984

devel

the sensor analysis

the most part

mechanism

transduction

evaluated

us to build tactile arrays

led

However

for

is

1985

et

described

as

and ease of fabrica

These considerations
capacitor

the tactile finger

aliasing

1985

and Hollerbach

by Fearing
tion

in

used

of the

independent

by having the
was not

intersection

As

in

tance

in

but

it

was
This

since

only

forces

the

has

not

was an issue thus the

remotely

shielded

cables

driving

from the sensor

distance

at

built

of about

into

30

the finger

of installing

cost

The

are

large

is

tip

it

enough
inter

thus lost in
of this

capacitance

Two

cable capacitance

significant

cm

electronics

of the signal

fraction

required to reduce

at

improved

significantly

the sensor design was stable

one

coupling

the array and one for sensing the array

output Improvements

used

the rate

constraints

could

be

in

sensitivity

and reduced

accomplished

embedded

in

by

wir

custom

the core of the finger

Design

Figure
of rows

the method
at

for static

capacitance

sensor allows

for

but

be increased

electronics

significant

the shielding

2.2

As shown

hand

that

felt

hybrid circuit

Electrical

done

the considerable

nally

ing

2.1

could

While performance would be

for

Hz

15

to

potential

The

improved noise performance

are mounted

the base of the
criteria

at

Simplicity of construction

Criteria

ground

to

cross-talk

limitation

electronics

primary

scanned

for

has been

analysis

The

and reduce

array was originally
reduced

Construction/Design

row and column Unused

selected

shielding

low scanning

Finger

for

voltage

and columns are switched

rows

junction

capacitors

are

formed

at

and columns of conductive

used
is

by

Siegel

measured

1986
by

the

the

strips

capaci

amplitude

Measurement

the

of the

Using an amplifier with input impedance
voltage

output

of the tactile sensor

is

RL

given

the

by

APE L0007544
The International

Journal

of Robotics

Research

in

Capacitor

Fig

elastic

model

strip

of width

mm and

2.5

0.5
tric

constant

From

00

surface

mm The

of

separation

pF

about

is

the dielec

for

of about
in

the arguments given

1985

bach

mm and

5.1

capacitance

normal

the

an important

beneath

strain

measuring the subsurface

the surface

is

To approximate
we use

measure

to

quantity

and Holler-

Fearing

strain

Ad

where Ad

is

capacitor

CSRLo

CSRLW

CL2

of the top

is

small and Ad

d0

d0

the bottom

since

VI

displacement

long

very

has no effect

the percent

WRLCL
C5

jVdI

C50

VSVCL__CL_1

_M

d0

d0

V5

where

Vd

the drive

is

tances between

for

and columns

CL

and columns

rows

thus the cable capacitance

dominate

The
will

significantly

CL C5

larger

and

shows

Figure

eq

large
is

simple

model

capacitor

plate

the sensor capacitance

plates

that

can

be assumed
are

the fractional

cell

at

the

same

twisting
for thin

well-behaved

each

and given

RL

and CL parameters

independent

of these

to
is

The

fractional
force

deflection

depends

elasticity for that

depth Percent

strain
to

directly

for

if

cell
will

all

be

par
cells are

used

throughout

the normal

con

order only

first

to

deflection
to

approximation

and compared

paper

for

measured with the same wiring and

deflection

parameters

first-order

valid

capacitance
is

V5 and

that

amplifier and the same

ticular

as

undeflected

measurements

cell

it

particular

stant

to the

without

compression

as

this

strain

is

C5K00
where K0
ty

Since

on the modulus of

The

for

to

valid

sumes capacitor
formula

made

capacitance

frequency

the nominal

function

with

than the sensor capacitance

sufficiently
in

approximation

capacitances

from the sensor

the amplifier and the amplifier input

is

This ratio of voltage

change
be

C0

where

cable
are

should

With the cable capacitance

d0Ad

CL

ca

sensor

of stray capaci

effects

rows

the load capacitance

into

the unselected

pressure

to

unselected

the

is

and amplifier input capacitance

capacitance

lumped

The

one junction

at

pacitance

and C5

voltage

deflec

by

obtained

is

signal

is

plate

of the

plate

Ignoring tilting of

of the plates

parallel translation

plates

tion

the

of free

separation

is

the dielectric

space
and

is

2.3 Mechanical Design

d0Ad

constant

the plate

Ad is

the plate

actual

capacitance

fields

because

pared

to the separation

may

the plate

is

The

d0

is

the permittivi

nominal

as

not

result

very

plate

The

displacement

be larger
area

is

area

of fringing

large

com

sensor has conductors

The

finger

Curved

size

surfaces

and shape
are

the fingers Fearing

can occur

anywhere

palmar surface
finger

is

Fearing

1986

chosen

are

needed

on the finger not

The

good grasping

objects

In these operations

thus complete

necessary

for

for rolling

finger

just

tip

made

of

contact

on the

sensor coverage
is

about

of the
cylin

APE L007545

Fig

Finger

construc

tip

dielectric

Outer
Electrodes

Inner
Electrode

JSA

of molded

Structure

Fig

tion

layer

STRIP

ON CORE

RUBBER

drical

mm long

25

section

with

hemisphere
with

coverage

For improved

design
versions

use 3.3

free

should

design parameter

be

twice

But

response

overall

finger

mm width
as

compliant

copper

rings

DPR

In the

is

inversely

sen

ap

ring

for

measured

is

incompressible

to

very

sensory

mm

covering

as

Poissons

on

10

force
is

the

rigid

coverage
strips

by

which

transmis

molded

modulus which

Nm2 The rubber

ratio

0.5 was veri

reduced
obtain

tric

section

spatial

The

im
without

sensitivity

for

as

case

sensor

of the wider

discussion

of

soft

the

and

the dielectric

to

struc

foam

the skin layer

give

Fig

of the

with

solid

layer

had only

of the molded

Siegel

1986

This

some

it

mm thickness

0.7

some

Thus

test finger

the sensitivity

1985

to

spaces

foam

of about

al

et

hollow

as

dielectric

in

as

in

cells

of some

strength and homogeneity

one-fourth

tabs

molding

during

was changed from

includes

Siegel

rubber

from the same rubber

sensitivity

using

because

the sensitivity

decreasing

increased

about

entire

improved

See

foam absorbed

structure

rubber

be

move

to

issues

molded

same

softer

In the limiting

wide as the

to

increasing aliasing

places

initially

The

foundation the

elastic

as

is

be

can

for

rubber This rub

separate sample

response

pulse response

The

Nm2 modulus

deforming

on an

plate

skin depth

layer

was

the harder rubber

for

without

layer

ture
disc

i0

at

dielectric

rubber

helps to widen the sensor response

also

the dielectric

of

thickness

For

low tensile

2.5

design

from the

Hardman Inc
its

be about

experimentally

0.5

the finger

4280-LV

chosen

give

layer

Figure

mm and

3.3

separated

of about

in

rings

outer

whole
of

the surrounding

of the tendency

aliasing

diameter pro

in

dielectric

foam

sensitivity

than

measured

impulse

is

was chosen

newest

at
to

open-cell rubber

is

fled

strain

shown

is

mm

17.7

spaced

are

sion and protection

ber was

structure

the seven

are

dielectric

molded

that

compromise between

core

integrity

the eighth

the tip The

with

strain

The

needed

tip

lDelrin

electrodes

connected

is

subsurface

equal to the sensor spacing

lightweight

at

alias

ratio

softer

cause

for negligible

and aliasing the sensor depth

vides structural

2.5

From

the sensor depth

signal-to-noise

also

is

For increased

much
used

the sensor depth

is

1985

depth As

to

proximately

driven

along

of the continuous

good

sensitivity

proportional

The

and 180

the length

the sensor spacing

ing from sampling

sitivity

and

initial

the latest

performance

spatial

mm spacing

and Hollerbach

Fearing

good

the

around the circumference

spacing

The

for

UPPER
RING

to

along the length

spacing

spacing around the circumference

450

diameter

of complete

goal

array on the cylinder led

mm center-to-center

3.8

mm in

and 25.4

end The

the

at

dielec

describe

increase compress

to

ibility

Unwrapped

the linear

around the circumference

mm which

would cause

separation
for

450

lot

of the elements

spacing

of aliasing

is

about

Since

AP
The International

Journal

of Robotics

Research

the

L0007546

Percent

Fig

load

applied

vs

deflection

to

one

celL

50-gm load

LLneorlty

Fearing

in

sured strain

C20

contact

pressure

area can

mate method

15

for

to

for

is

almost

reduce

the force

in

40

20

80

100

80

gnoms

toad

is

3.2-mm

does

orders of magnitude

is

copper

not stretch

beam

like

the sensor

but can

This

much

beam-like
better

might he obtained
rings

the length

around

localization

tion Fearing

the

tactcl

straight

load

0.5-N
rubber
relation

is

The

resolution

along

important

circumference

for

than

manipula

1986

of the

in

true

of

spond

sample

is

important

tion

cell

AID
The
for

for

of each
in

cells

diameter

element

tactile

the finger

units

gives

seven

7.8-mm2

voltage

pressure

sensitivity

meaning

because

figure

of about

The

0.5

real

gm

is

of

of 0.06

the response

for

cell

1t change

pressure distribution
sensitivity

output

of signal

of this

6%

sensor

6%

table

relation

medium

but

capacitor

the parallel

can
estimate

an

this

wasnt

would be sensed
linear The

truly

function

vernier and found to
in

for

force-deflection

make

to

the

for

very large even

force-strain

as

than

less

the strain

of load was
have

similar

the sensor appears
of

plate

capacitor

raw

sensitivity

if

deflec

to

re

eq

area probe

is

6% deflection

0.06

A/D

gm

mm2

unit

function

also

of merit

AID

units

Calibration

is

for

More

important

The
little

of the
force

than

measurement

capabilities

is

how

of the sensor that is

performance

of 2000

gm mm2 has

figure

100

l/

3.1

deflec

50-gram load on

voltage

bits

pressure sensitivity
sensed

is

the strain that

of rubber

with

counterbal

is

The

tactel

array give

hemisphere

with an undeflected

tactile

evaluating

50-gram weight on

for

3.2-mm

is

one

at

between

Evaluation

parameter

the sensitivity

least sensitive

as

the

in

to

friction

instrument

placed

nonlinear

medium were

the sensor and

relation The

An

look-up

strain

needed here This

tion

This

expected

used

measured

Performance

and thus

is

by

of the pivot point

through

origin

of 10%

strain

material

easily be

more

from the

This
trials

halfway

There

side

1-N load

to

response

line

than

conductor

improved
is

aliasing

degree

greater

of the finger which

The

caused

probe

it

of the rings gives

and reduces

compliant
for

the rubber

approximately

property

to

with

on the outside

are

bend

overlap

around the circumference

than

stiffer

only

cell

output

of seven

with weights

diameter

on the opposite

weight

tactel

application

and the weight

the pivot point
ance

beam

balance
the

force

of one

deflections

be

may

that

The

device

simply

end and

The

of the large

small variations

larger

approxi

in section

from the average

was generated

plot

of

from the capacitor

eq

using

in spite

An

response

as determined

output

linear

applied

described

is

to

the load from the strain

determining
sizes

mea

the

frequencies

peak strain

larger

increasing weight

sensor voltage
-o

force

shows the monotonic

Figure
10

give

contact

varying

of the spatial

The same

As

probe

1985

and Hollerbach

function

is

the applied

-I-

with this particular

specified

described

Sensor

25

probe

with

50-gm load

at

the cylinder
elements
for

while the voltage
are

recorded
location

for

row

each

cell

tip

is

increments

equal time to each

deflection

of

hemispherical

0.6-mm

the pressure

vary from element

ments and with the sensor geometry
ter

the spatial

to

ele

3-mm diame
with

applied

along the length

outputs

The

is

of

from seven

The

probe

applied

maximum

the gain parameter

is

percent

stored

APE L0007547
Fearing

Fig

Responses along

cylinder

Fig

moving

length for

variation

Sensitivity

among

els

tact

probe

.4

Cain

Tocteis

of

7xSi

Gytnder

on

40

30
in

o20

18

10

010

0.2

0.4

0.8
bit

0.8

sensitivity grams

.._

c4

$8

01

7.8-mm2

area probe

Five cells were very sensitive

with

threshold of only 0.2 grams

ments

are

10

25

mm

position

finger

is

which

20

iS

in

each

and used

table

There
by

is

some

variations

to

normalize

variation

in

around the core during
ferent

dielectric

Figure

calibration

which

ing elements

Note the

There

elastic

set

is

as

tactel

the right side

elements
tactel

at

the

the probe

tip

for

within

rubber

5%

about

Figure

tactile

finger The

change

in

mean
the

A/D

probe

first

of the calibrated

elements

on

sensitivity

converter

at

sensitiv

enough

hand

is

Superposition

row

the

tip

to

of

deflection
at the base

month

but then

stabilize

to

many

that

theory
response

can

is

be

all

portion

of the

1-unit

0.4 grams with

the

the

linear

used

to

analyze

and obeys the

space

probes applied
the individual

invariant

independently
strain

strain

valid

assumption

profile

space

profiles

invariant

from

techniques

shows interpolated

joint
for

as

powerful

linear

the linear

Figure
as

gain value

cylindrical

the sensor behaves

tion

The

molding

after

If

there are

which can be seen

threshold for
is

3.2

between sens

of the sensitivities
the

up
more than

caused

giving dif

moves from

cures further

histogram

is

the

hand

400 grams

to

wrapped

interpolation

as residual

for the

gains change

the isoprene

when

is

responses along

important

of Figure

may

have

exerts

at different

good overlap

is

it

be

fabrication

the base left to right in the figure
at

as

and thickness

shows the

cells
after

dielectric

finger

stiffness

finger

has

usable.The

use with the Stanford/JPL

our configuration

We

is still

the sensor output

gain which can

the rubber

in

for

designed

gram which

0.8

with this tactile sensor for this

ity
in

and the worst element

functional

threshold of about

the array ele

All

it If

of superposi

assumption
profiles

least

two

sum

and jointly The
corresponds

with forces

less

hold

will
for

well

Thus superposition seems
at

system

the sensor

principle

strain

system

linear

to

be

to

than

of

the

100

grams or so
Sensor

two ways

hysteresis

The

interferes

first is

that

with superposition

the sensor will not be

in

time

APEL000754S
The International

Journal

of Robotics

Research

Fig

Superposition

loads

two

of

10 Sensor

Fig

on finger

for repeat

Superposition

25

--probe

Loads

of

output

jitter

ed force

Probe

Repeoted

probe

18al.ement

oprobe

20

F0.SN

AppLicotior

12

41

16

probe

botb

15

-.--

014

probes

12

10

C-

010
4-ci

eLements

51

31

s-

-5

toctot

0-

position

20

40

60

80

100

scrnpLe

The

invariant
tions

second

relation Because

will

of the offset
for

the

if

that

byfp
some
2fp

described

2fp

is

zero the sensor

not

are

if

input

strain

initial

not

condi

linear

have

errors

the sensor output
load

then

f2p

are

the finger friction

offset

on only

sources can

Noise

With
be

Limitations

mum

from the

detectable

eq

Vo

tion

The

step

about

size
0.3

force

AJJ

applications
is

of the signal
weight

and

is

applied
jitter

in

Figure

11 The

under the probe
This error

balance

beam

is
is

when
tactel

tactels

The

location

of

deviation

to

1-gram

errors The

by quantization
force

experiments

these

error
all

tactel

ac

calibration

Finger

errors

Behavior

application

be

removed

12

response

it

force

makes

be

may

more

how

the temporal

see

strain

long

it

of creep

and

the

useful

indicate

force

removed What

how much
but

elastic

remains

is

when

takes to decay
strain which

delayed

elastic

is

re

1985 The creep strain causes prob
can make the sensor output appear
as

Johnson

much

exhibit

loop to

force

flow of material

lems because

to

of the sensor to

arrows in the figure

sensor shows the effects

sponse

compliant

polymers undergoing

expected

indicate

not just

is
is

be

hysteresis

The

and dots

continued

too

can

steps of Figure

of creep

important

Soft

time response

the traditional

in

high sensitivity

hysteresis

The

force

50-gm

if

for

pay

and creep

The

100

histogram

equivalent

so

they will affect

simultaneously

deformations

large

the force

directly
for

may

sensor

applied

standard

caused

since

limited by these

to

price

than

50-

the same

at
for

measurement

be neglected

Viscoelastic

The

set

of 0.5 grams

higher

and two adjacent

shown

3.4

effects
is

of the 50-gram weight

variation

several

to

shown

cells

deviation of the sensitivity

variation

repeatedly

the probe

jitter

quantiza

quantization

10 shows the output

beneath

larger

0.14 grams

sensor output

gram weight
Figure

step

the standard

pivot

made Most

voltage

implying

sensitive

is

on

and varying

sen

0.05% corresponds

of

most

for the

output

f/

the deflection

larger

deflection

Figure

The

the

grams

implies that

the

smaller the voltage

the

in

the balance

of loca

friction

surface

to

mini

The

converter

of the capacitor

deflection

separation

plate

in

AID

depends on the nominal capacitor

sor

is

curacy

no load the dominant sensor noise was found

quantization

combination

of

contact

as

single

measurements
3.3

in

velocities

impact
rely

result

probably

wobble of the probe varying

tion

is

it

is

being applied

The

hard to determine

delayed

elastic

whether contact

APE L0007549
Fearing

11 Probe

Fig

noise histo

Fig

gram

12

Viscoelastic

re

of sensor

sponse

around

13 Response

Fig

for

circumference

moving

probe

Force

on

F0.SN

tI

40
eement

41

mean

std

IS.9t

dev

1.0

gin

j30
20

etement
10

31

meon

4.7t

std

dev

0.4

10

groin

12

iS

14

10

18

defLection

has been broken

If

mined one
ual

method

setting

strain

contact

is

is

it

0.1

residual

the

delayed

sor

output

3.5

error

or

for

an

13

good

shows

normally

reduced

as

is

three

this

5%

sensor

10% of

to

is left

of the peak
or so

rate

25% sen

represent

will

to

hard to deter

be

measurements

from

are

Circumference

cells

with
30

sufficient

spacing to

and Binford

limit

remnant

around the circumference

450

1989

overlap

length
to

3.6

is

Tip Behavior

The

tip

around the circumfer

response

180

probe

spacing as

intervals

along the finger

this

may
It

deflection

at

For

to decay

25-Hz scan

more uniform along the length
Fearing

after

and the response

of about

response

phantom

ence There
as

the

upper

scanned

be

or greater

earlier finger

applied

the resid

point

practical

Beyond

Response Around

Figure

set

deflection

elastic

mine whether
contact

deflection

deflection

deflection

fractional

35

30

be deter
with

dealing

the response

for

fractional

peak

with

can

elasticity

takes about

the

new zero

the sensor should

fast

for

25

20

see

broken

Delayed

how

removal can

contact

possible

15

time

The

make

but

newest

it

is

the responses

and circumference

consider

around the cylinder circumference

in

not

finger

the response

more detail

portion

extensively

has been analyzed

of the finger

than

the

cylindrical

portion

The

less

core

ex

tends slightly into the tip area to improve sensitivity

and

to

prevent

hemisphere

dead spot

at

and the cylinder

the junction

From the

top

between

the

of the

APE L0007550
I9

The international

Journal

of Robotics

Research

14 Responses

Fig

normal

probe

ing

mov

for
to

15

Fig

plane

tip

geometry for

Finger

assumption

stress

Pressure

The

thick

normal
tact

to

for

response

the surface

has been

analysis

The

used
is

for

space

invariant

the tip because

not constant

above

sphere underneath

done because

the thickness

hemi

This

to

is

and

the

medium

eter

of the cylinder in

with position

hemi

tip This was not
of forming

thickness

Fearing
instead
length

The

will

Model

for

Of

Half Plane

course

linear

two-dimensional
used

in

Phillips

Hollerbach
strain

on the
finger
in

the

and

stress-strain

and Johnson

1985

will

fractional

surface

of

coordinate
direction

der Consider
with the applied

1981

be used

capacitor
half

slice

here to predict

plane Figure

15

reasonable

between

Goodier

defines
is

plane strain
finger
closer

so that

materials rubber

makes

first-order

are not

dielectric

sensor nonisotropic
is

only valid

up

to

30%

approximation For

isotropic

and

the

theory

elasticity

the stresses

as

here

of three

that

and

diam

show

medium

strains

are

for

but

this

linear

the relations

Timoshenko and

1951

the

for stresses

deflection

sensor at the top

vc5

the
defined

oJ

va ar

of the cylin

of elastic material in the x-z plane

force

tion and the stresses

to

and Fearing

deflection

axes Capacitor
for

similar

analysis

used

the

they assumed infinite

but

which ours

three-dimensional

to

the responses of Figure

definitely

small strains
is

be

inclusion

and copper
Elastic

way
1985 used

the plane stress response

to

sumption

corresponds

crude

and Hollerbach

However

slab

of the finger

portion

of the slab

of plane stress

agreement

nondevelopable

surface

Linear

the cylindrical

to

corresponds

The

the plane stress assumption

be

sensors on

on

plates

impulse

cannot
of the

of the added complexity
capacitor

con
14

be

yet

the

varies

it

mount

to

tip

strain

assumption

the hemispherical

strips for

copper

the

applied
point

Figure

beneath

the sensors

would be reasonable

It

for
in

mm

10

force

done on the

of analyzing

responses of the sensing elements
sphere

the

would be used

as

of the difficulty

cause

about

is

the tip with

given for two elements

is

grasping

Not much

base the rubber

its

to

hemisphere

on top constant
on the

face

in

the

of the slab

direc

va

equal

c4
APEL0007551

Fearing

11

Fig

16 Plane

and

experimental

stress

model

data

lv

1.2

2Pz1v

vx

0.8

the relation

of plane stress to

the following

tuting

1980

from Gladwell

substitution

useful
0.6

shows
substi

By

plane strain

eq

into

0.4

11

10

0.2

where

is

lence can

be

mm

position

0.5

shown For example

corresponds

4.1 Impulse
The

strains

The

each

in

stresses

rameter

is

are

direction

Poissons

our finger The

is

for

Nm2 for

0.6-mm
one

crude

cell

cr
where

and

mation

a1

per unit thickness

z2 For

x2

r2

az

rr4

the force

is

Nm
at

2Pzx2

Thus

for

2Pz3

of the slab

in

the plane stress

line

fit

mm and

for

most of the

retical

with no free
fitting

az2

curve

Poissons

vx2

Eirr4

tactile

solid

an

strain

ible

infinitely

half space

approximation corresponding
long line of force

on an

elastic

to

all

3.8

ay

va

in

results

parameters
to

It

is

equal

The

compressibility

mm thick
estimate

mm depth

layer acts

which

The
is

predicted

0.5

gives

for

the

layer
is

capacitor

postulated

Fig
by
0.4

best

data

to

rubber

top

theo

in

not an arbitrary

but this thickness

it

The

mm and

3.8
is

the experimental

ratio

is

of

depth

compared

are

pa

consistent

the array

the figure

in

was

eq

ratio

are

an incompress

material like rubber but the dielectric

depth
3.2

elements

curve

space for expansion

For the plane

normalization

0.4 These parameters

plane stress model with

the

sin

amplitude

the plane stress model of

and experimental

ure 16 The

2P

to

an

is

of the previous

assumption

after

from

which corre

done by adjusting the depth and Poissons
The best fitting parameters were
rameters

approxi

load and indentation

stress

Accordingly

3.8

var

cos

the plane

least-squares

from normal

angle

to

section

the finger

to

load at the surface

at

length

the output

while recording

line

with

load

along the finger

approximation

with

slab

line

approximates

normally

increments

elastic

1985

which

was applied

0.5

load normal to the sur

Fearing and Hollerbach

are

33

Response Along Length

edge

Youngs

is

10

2.5

knife

sponds

face

plane

0.33

and

given by

approximately

stresses

with

strain

with

stress

plane

by

material like rubber

an incompressible

modulus which

are

The pa
cy and
which is equal to 0.5 for
ratio

given

to

ratio the equiva

Poissons

the equivalent

that

finger

layer

over

above the core

coincidental
plate

is

at

out

to

is

the

about

the softer

to spread the strain response

has

some

dielectric

more giv

APE L0007552
12

The International

Journal

of Robotics

Research

ing an effective

greater

The

square

of

mean

root

scale which

full

of 0.5

made

that

tional

jitter

tion

in

is

model

plane stress
flection

with

for

It

of the

region

is

this

sonable the contact
diameter

finger

rigid

tion
in

very

The

much

was done

shown

the overall

rubber

Linear Elastic Model

an

stress

plane

pacitor plates
to

are

very

the sensor depth

rate

Because

model would be an

tion in two dimensions
fects

of the

finite

The
the

the impulse
given
should

plate

response
depth

be

Figure

each

strip

layer

and V24

u2

to

YW1XW2YW3t
has three displacement

The

vj

Youngs

12

and Poissons

modulus

com

shear modulus

is

by

ratio

M2lv

13

the
The

deflections

for

the two dimensional

from

case

1980

Gladwell

strips

14

2pu4l vyIiJ-x911zyar3

The

2uw4l

might

vw3xwi zw34

15

and

V3

are

plane harmonic functions

16

the capacitor

and

important
is

strip

and rubber

be

in

is

strains

are

obtained

8w

respect

rigid

founda
where

medium on

question
whether

strain

will

rigidly

shear strain The

obtained

from the

2vc2p

consider

attached

are

the

stresses

strains

by

ow\

Iou

for

17

plane

via

vvc
2va2ji vvE

for

there

the capacitor

We

is

j-

from the displacements

ca

models the ef

for

The

more accu

with

model
to

the

that

other with

of the elastic

hard or soft foundation

the dielectric

Vw

move indepen

of the core

and separation

shows the plane

and

the

effect

stiffness

This section

An

modulus

vector

displacement

ponents

next sec

in the

to

that

extent
to

elastic

thickness

11

for Strip

of infinite
close

potential

stiffness

model assumes

medium

elastic

shear

the

where Wi

The

as

very well to the rubber

dently of the copper Thus the copper
not increase

to

by the copper

which would allow the rubber

skin

in

the impulse

affect

do not have much

not be bonded

may

copper

is
4u

rxIy9z2

de

Com

the capacitors

to

caused

inhomogeneities

the sensor apparently

given

of the

the small
theory

small compared

be

as will

is

scalar

the sensor matches

why

backing

ji and

potential

model may not be unrea
is

vector

with

related

slab

length

vector

Papko

the

Using

Gladwell 1980 with no

2pd2lvyVr4
where

data

of the sensor would

rigid core of the finger does not

response

is

model The

by the

experiment

speculate

The

methods

potential

force the displacement

body

sensor

to

interesting

the model so well

the experimental

elasticity

the nonlinearity

potential

elastic

can be solved

plates

by sensor varia

of violating

spite

elastostatic

using

and posi

friction

of the capacitor

deflections

vich-Neuber

by

cell

The

function of

assumption

the ideal behavior

to

of linear

help for larger loads
linear

close

in

assumption

pensation

in

explained

statistically

sensor performs

by

than

the

constant

above

caused

rather

variance

is little

not

is

1.3%

just
that

with

output

scale The

full

beam

the

sensitivity
is

error

fitting

applied

2.0% of

variance

this

There

that

is

losing

good considering

tactel

repeatedly

beam

balance

RMS

quite

deviation of

standard
force

is

without

depth

18
19

to

20

the core

Fearing

APEL07553

where

is

the

For an

Fourier transform
other potentials

are

W3

W2

is

1980

representation Gladwell

the

F_s_h Bse
Dsesz

22

and

frequency

is

2ve

Ds

where

indentation

frictionless

the strip

the surface

at

adhesion

for perfect

cosine transform

can

are

To

24

wxzh0

PsHs

determine

tact

solution

tion directly

the deflec

necessary to

is

beneath

The

The

applied

normal

stress at

the center

previous

the surface

contact

px

is

the contact

four equations

The

gain for the capacitor

and

normal

to

with d0 the nominal

the four unknowns

the system of equations

deflection

the surface

can

be

solved

The

FFsHs

27

after

kernel

Hs

matching boundary conditions

is

the

is

In the
gain

Finally

plate

taken

is

Crsz

KB

szB

De5t

28

to

capacitor

medium
The

zero even

so this case

when

boundary layer Although
The terms

in

the above

are

expanded

as

the actual

more
_Ke2w

2w

KC2

2w

29

signal

to

In the second
3.8

mm depth

work
case

3.8

plate

shows gain

plate

pre

Figure

is

mm

width

and the

of the top
is

fixed

at

the

function

as

gain does

gets close

the gain

in

mm thick

d0 The

value

capacitance

is

as

pressure in the

of the depth

also

the top

shown

is

bottom

separation

plate

36

of the probe

function

plotted as

given by

is

separation

plate

surface

parabola

of the capacitor

foundation

obtained

szA

35

w0zd
hd

where the

as

case the

first

of nominal

Hs

depth

gain for plane stress

two cases

for

given by
plane

2iwx

The

viously
17

is

certain

at

26

pressure

for

zds

is

must equal the

Az

With

to

FsHs

25

con

for frictionless

p4
where

wx

of pressure
simplifies

equation

so

pressure

at

deflection

know

it

25

the foundation here the adhesive

at

found

34

cos 2xsxds

sensor gain the percent

depth

2jiwO
the

gives

func

to

Jo

bottom of the layer

ux

1980

used

07

certain

Gladwell

be

real

deflection

23
the

and

even

2iwx

at

and

of the layer

are

assuming

rx
and

Hs Fs

Because

the inverse

determine
for

33

the thickness

sh with

tions

boundary conditions

2ve 32

CsDl2vAB

the inverse

Fourier transform

The

31

AscxBs

21

Fr2

spatial

30

2w

in

given

4v
where

ea

2w

e_t0c4c

po

appropriate

are

w1x

Bs

the

gives

1980

Gladweil

choose

to

ap

stress

v/

strip of thickness

elastic

functions

tential

vt

substitution

stresses

corresponding

For the plane

shear stress

the

proximation

to

slightly

not go

the

reduced

gets larger so there

is

with
the capacitor

and the

plate

bottom

separation

d0

plate
is

is

fixed

at
at

APE L0007554
14

The International

Journal

of Robotics

Research

17 Gain

Fig

vs

and

separation

Fig

capacitor
thick

strip

half

18 Impulse response
plane and strip

for

ness

Coin

For

E2.5x105

Modet

Strip

20

F20

N/m2

ImpuLse

N/rn

--hotf

vs

gain

s.depthC

g15

thickness

For

Responses

ond

Strip

Hoif

PLone

ptcne

strip

cadet

00.8
.4-

008
ID

mm

3.3
3.8

.4-

mm

mm

3.5

ZJ////11//d/.//fJJ//J//t

z77z

j//////77/////////////

-15

0.5

mm The
beneath

layer

now

gain

has

the

soft

the increased

adequate

is

It

above

does

apparently

foundation

10%

about

predicts

3.8

mm

the capacitor

approach

the infinite

Because

gain changes

thickness

plane strip
half

in
is

agrees

quite similar
in

more

closely
is

layer

to

The

soft

backing

no

strip

it

loss

model does not

The

than

should

be
of

along the length

analysis

for

layer

it

can

be

to

much

rigid

important

for

and

prob

The

simpler

The
for

pressure from
sis

for

is

the length

is

useful

of the

cylindrical

much

than

better

good position

get

of peak deflection

is

also

on the

estimating load magnitudes
describes

to

of the coarse

to

to

important

recovery

section

recovering

an interpolation

method

location and magnitude of the center
discrete

restricted

measurements

contact

type

is

in

better

Fearing

of

analy

and Bin-

1989

Assume

concluded
plate

spacing

it

signal Because

of the finger localization

finger This

data

measurements

strain

continuous

information

ford

of sensitivity

yield

the

here
it

Reconstruction

spacing of sensors along

the tactel

response

elastic

recover

section

the skin above

multiple

discrete

portion

half plane

the bottom capacitor

fabrication

From

simpler

for

response

the impulse

can be attached

significant

makes

for

much

of the previous

the plane strip analysis

not required

dation

experimental

Strain Signal

Fig

with increasing

with experimental

softer

lem has not been considered

with

plane

the half plane assumption

value As the

use the

and width

shape

cause this effect

From

to

little

18 the impulse

Figure

plane

dielectric

that

half plane

Nm1

200

the model

thicker

so

half plane analysis

stress

As shown

may

to

re

impulse

stress

the finger

half plane of the previous

seems reasonable

it

Ft

and

gets

section

The

that

for

adequate

enough

similar

assumption

the

the

for

calibration

with the experimental

should

model

have

is

sponse

but the calculated

response

obtain

to

deflection

used in the finger

as

well

agrees

layer beneath

the

increase

substantially

not necessary to

compliant

mm

0.5

plane

the capacitor

so

seen from the plot

be

not

the impulse

of the

15

err

tion

p051

sensitivity

same parameters

which

As can

thickness

model

This

increases

capacitor

backing

the sensor gain
capacitor

plotted as the thickness

is

10

-5

mm

thi ok ness

ond

depth

-10

discrete

sensors at spacing

then

seven

sensors of one row along the cylinder give

is

foundation
rigid

The

foun

hx

plane

closed form expression

nb

rect

37

for

APE L0007555
Fearing

is

Fig

Hs
As

19 Frequency response
and aliased response

maximum

The

assume

I.8

that

the low-pass
is

energy

The

but

continuous

1/b03

0.4

is

the normal

eq

strain

impulse

we

filtering

Ers
version

filtered

low-pass

recovered

Ers

and the

for

length

is

38

vx2

xEr4

defined

now

have

42

2bs

rect

has both

As

.tAs

As

and

the desired

2bs

rect

43

39

The

relative

from the

38

neglecting

spectrum

higher

by

leads

error

Hs

sinc

is

caused

by aliasing

an

Assuming

ideal

obtained

signal

low-pass

error

E0

filter

the

by

given

40

6xbs

is

44

of energy in the reconstruction

the energy in the original

to

1978

Pratt

to

error

ratio

ER

E0
domain eq

aliased

AsH4sHsQ

ifIxI

In the frequency

the

is

order terms given

by

rectxl ifx

rII2b

Hs2ds

IHsI2ds

ijim
with

spatial

frequency

Fourier transform
is

which

is

frequency

in

of

periodic in frequency

widened

Because

of

Ess

of sampling

and the frequency

of having

response

and

cycles/mm

by the convolution
result

ex

finite

components

from

response

2z

function

Ess

by low-

by

hx
rect

aliasing

be

principle

limited by aliasing

is

Es

where

The

in

Ignoring the finite finger

0.5

The

defined

The

in practice

filtering of the periodic spectrum

of

the aliased

hx

may improve

slightly

cyctes/mm

fncquency

where

we

along the

occurs

reduced

can

signal

with ideal low-pass

0.2

is

If

pressure distributions

but this process

0.1

Hs

Figure

mm

3.8

sensor width

finite

significant

by low-pass

tent

of deflection

properties of

pass contact

a.1

the

1/3md

ford

response

integration

plate

capacitor

cC

occurs at

response

shows the frequency

19

with

only seven

hx eq

the plane stress assumption

the sinc

is

39

given

at

Ers

response

is

6irbs term

samples The
depth

with

This

cluded

are

by

the effects

e211

we

41

for localization

lated

function

eq

45

The

will

effects

errors

caused

Hs Not

low-pass

and amplitude

Fs

is

an upper bound

of aliasing

on

in the

sufficiently

and

3.8to

20%

interpo

bandlimited

on the expected

localization

in

filtering

0.20 which implies up

If

be

of

the sensor depth

for

get

error

2ird

reconstruction

of nonideal

numerically

mm spacing

3xdsI

includes

expression

by the cutoff of the high frequencies
Evaluated

Hs2irdIsllv

45

the

error

and amplitude

APE L0007556
16

The International

Journal

of Robotics

Research

20

Fig

Reconstruction

Fig 21 Interpolated

error

caused by aliasing
3.8

lion

mm

nterpototion
30

Error

continuous

ci

terpoo

in

25

ted

somptes
20

tocotizotion

15

1-

error

offiptitude

ci
-t-

dejiec

surface

10

ci

toctel

postion
in

force

estimation

now

will

be shown

row on the finger The

along

windowed

using

sinc

the seven

for

strain

sensors
estimate

response

angle of force

strain

be

principle

when

there

1985

no

is

and the applied pressure

hx0

function

nh

__________

46

This

to

is

find

with

window Oppenheim

the Hanning

1975 which

is

makes

the

ringing

the total

amplitude
point force
to

the

20

error
in

filter

of samples The

finite

extent

in

flection

function

N-IN-I

and

when compared

Here
are

response

tion

forces

accurately

for

Small

location

can

certain

finger

errors

example
lead to

configurations

location to

force

Localization

accurate

be

but can

tool

processing

the tactels

all

zj

where

The

is

11

the

de

interpolated

is

srnirxisrnnyj

48

wx zwyj

are

in tactel

21

shows

with

spacing units where

plot

of the deflection

soft object

large

useful

apply

contours

contours

distances

by dividing by the sensor spacing

normalized

tool

The

of height

than
is

surface

another

plot

conditions

plot

onto
for

of equal percent
for

where

finger

unwrapped
the surface

for

plane
analyz
deflec

equal deflection

are

1986

Fearing

large

maximum

single

by interpolation

strain

of element

ing the sensor output

the contact

ap

the surface

fl

more

know

at

array formed from the simul
of

the cylinder and the tip are

to

an even

real-time

10

Localization

important

force

to

contact

is

in

fxy

window

Figure

It

from the

used mainly for sensor evalua

be an

deflection

and reduces

the interpolated response

strain

47

forjnIN2l

the plane stress model

continuous

11

response

percent

shows the peak displacement

for

41

is

relatively

is

than as

rather

taneous

number

low-pass

Figure

tion

Let

is

wn4

and

and Schafer

is

and

loca

for

more general

tangential

the peak

scheme

simple

time consuming

force

sensors as was proposed

and Hollerbach

Fearing

done by solving

and magnitude of

at three

equations

proach

is

interpolation

could

tion

of

errors

in

xy

line

x_

APE L0007557
Fearing

17

22

Fig

Fig 23 Measured

contours

Deflection

on

probe

location

array

Contours

Strain

interpolated

for

IS

iN

degrees

III

Probe

interpoLated

Location

6.5---ideaL

-I-

6-

position

esensed

Cs

Experiment

position

--5.5-

-0

1CA

ID35

-0
around

tactel

The

of constant

direction

on the

height

4.5

3.5

circumference

surface

given

is

The

5.5

toctot

position

probe

maximum

conditions for

of

are

surface

by

4tan1f

.feee0
Ox

50

The

contours

found

are

from

iteratively

slopes

are

Oy

search

Using Newton-Raphson
where the

zero we

Ax\
\AJ

Thomas 1968

where Ar

is

the step size The

locationfx

Ax

and

to

locations

is

for

contact

sor

spacing

The

the deflection

with

finger

low-pass

surface

is

profile decreases

spacing

45

has

out at about

tion on the cylinder
rings

This

to

45

is

behaved

zero

at just

90
may

Fig

by 3.8-mm

sen

one

as

spacing

loca

by the cop

caused

of the

as

find

the

maximum deflection

an unknown pressure

iterative

and

its

techniques

location

spaced

are

are

needed

Fig

steps

location and
of the

portion

50-gm load
will

strips

is

be

the

The

higher
is

probe

if

moved

along the strip the tactile ele

mm along

the

row

the figures are given in tactile element

The

abscissas

tactel spac

ing units

The

ripple

on amplitude

caused

for

error

between the copper

3.8

three

of probe

effects

The

strip

by

aliasing

plitude

as well

rubber

To

in

is

from the max

two or

along the central

boundary

250-jim increments

ments

unit

Note how the cyl

by the incompressibility

probe

above

starting

recovery

deflection

avoid

cylinder to

at

the

tactel

0y2

in typically

and 24 show

centered

ensures that

45

32f

to 0.01 tactel

observed

is

peak percent

from the contact

be

tactel

the finger

The

Ox

OxOy

32f

Convergence

ures 23

not sufficient

around the circumference

inder bulges

per

to

new

tracking algo

characteristic

well

the

adjust

on the contour

rod at

around the circumference
strain

to

find

new

the

at

error

of this contour

where the

cylinder axis

used

ensure staying

ure 22 shows the output
rithm

height

Ox2

OyOx

51

\ArsinJ

to

have

a2ff

02

arcos4\

52

and
Ox

are

sensed

ing error

These

of

results

0.5

and peak location

The position of the probe
12% of

within

mm along

imply that

ideal

the length

aliasing

is

less

is

mainly

and

for

am

position

of the finger
severe than

APE L0007558
18

..The.Jmernational

JQzirjial

of Robotics Research

24 Measured

Fig

25

Fig

peak

around

amplitude

Poa

Interpolated

Localization

errors

circumference

Strain

30

4.5
ideaL

position

sensed

25

position

circumFerence

around

Q20
ci

15

a2

10

.5

1.5

3.5

probe

is

that

the

error

aliasing

above

directly

polation

tactel

when

zero

is

because

the

is

the nonuniform

10

probe

line

An

60

100

60

120

degrees

positior

HstHOdE

the

sinc

the sample points

passes through

source of errors

tional

plane stress model with

by the simple

load Note

20

6.5

toictet

position

probe

predicted

5.5

inter

addi

Then

normal force

for

of

positioning

the sensor rings from manufacture
Localization
of the finger
ing the probe
is

tested

in

the same

25

10 Fig

manner

The

increments

at

larger about

mm error

This

need good

of

line

contact

localization

along

on the

by apply

localization

determine

Fearing l987a The

the circumference

was too large so improved

built

with

180

we

and

and around

the length

the circumference

were

HOF Fsdx

the

cylinder

finger

around

error

fingers

where

Force

is
is

the total force

recovered

the interpolated
different
It

is

useful

sensor

to

This

estimate

know
section

this

the total normal
uses

simple

force For contacts

three times the sensor depth
proximated

Fs

will

as

tend

constant
to

be

56

HOF1Psdx

spacing

Total

Estimating

55

error

2.3-

about

is

To

along the circumference

orientation

L. FHsFsdx

performance around the circumference

was

force

aproximation

longer

Hs can

because

bandlimited

on the

than
be

to

two or

crudely

the surface

finger

ap

pressure

The

contact

integrals

disagreement
Aliasing can
ing

987b

method
simple

fractional

the normal direction

lengths

contribute

errors

do not show

filter

in

Fear

the integrals

more accurate

but

contact

large

approximation

description of

fast for large

three

for

SO-gram load

deflection

to

and

shows

26

profile

with

in spite of the crude

has

Figure

deflection

each

of fractional

using an inverse

and

in

by interpolation

the method

here

is

areas

APE L0007559
Fearing

19

Fig 26 Contacts

and

total

load

contact

Strain

Interpol.oted

car

Bar

normal

Loads

Smith and

there

stress

40

aS

run

fc2dx

A1
21

30

mnff

rnmJcdx

dx

123
132
115

zOpx

crx
where

Now

tan

/3

sensor at constant

fx

1-

for

the

px

we

fx

where

px

is

of an

lution

Thus

components

Force Angle

would be

useful

to

get

result

normal pressure information

as well

as

sensor

From eq

with
be

the plane

both

stress

contribute

and

for

contacts

where the contact

1985
method

is

function

an even

pressure

is

contacts

with

type

here for the

proposed

edges

vertices

it

The

the

convolution

angle of force

planes

normal

stress

and the

function

odd function

as

tangential

can

be

stress

The

measured and interpolated
and odd

us even

gives

fx
fx

fx

corresponds

to

eq

an

57
vx2

cos

skew index

sin

which
Note

gives

is

defined

Similarly

are

and odd impulse

even

responses

The

assumption
at

the surface

can
at

be

made

any point

that
is

the tangential

proportional

to

best

be

63

maxJ

index

In general

function

plicated

might

heven

stress

as

maxf

an indication

the skew

that

indentation
there

62

into

2F
vx2

fx

strain

Ejj

2F

function

deflection

by

parts

skew

Cy

61

for

and odd components

even

60

ddx0

pxx0h0x0ddx0

and cyl

spheres

seen by breaking

an

stress

in

x0hx0

symmetric

an even

to

corresponds

in

xc

fx

inders
The

results

tangential

and

have

known
priori
case when the contact

for example

and the convo

function

normal and

by the

of two even

can

Fearing and Hollerbach
is

by convolution

caused

respectively

./

stresses

strain Methods

the

strain

load

line

pressure

tactile

normal

surface

tangential

recovering

cylindrical

from

for

for

assumption

the normal

to

been proposed
line

the expression

normal and

seen that

information

force

tangential

the

at

even

58

hoxo ddx0

px
It

have

and an odd function

even

odd function

Recovering

strain

by an

the strain at

functions results in an even

Length

normal

xoEhcvenXo

pressure on the surface

aLong

is

Jx fdx
Jx0

toctel

that

z0Jipx

caused

depth

px

distribution

020

cAx

and

1953

Liu

of the relative
wili

the

be

force

skew index

will

as

com

be

of the pressure distribution

used

angle

zero for pure normal

and

means of tracking

force

angle

be simply evaluated

for

line

changes
the

The

skew index can

APE L0007560
20

The International

Journal

of Robotics

Research

27 Even

Fig

and odd

Fig 28 Measured

strain

odd percent

model

components from

300

and

even

deflection

for

contact

Knife

40-

Experiment

Edge

at30

CF

degrees

oeven

port

odd

0.9
30

port

--interpoLated

0.6
--3
-3-

020
0.4
4-

0.2

-D

-0.2

mm

position

toctot

indentor
tion
force

0.5

angles

The

this

the plane

for

Unfortunately

with

is

stress

low-fidelity

maximum

even

signal

strain

assumpfor

Knife

Edge

at

Experiment

t30

degrees

40

small

occurs at

Oandisgivenby

30

2Pcosa

max

64
020

The

maximum

odd

strain

occurs at
010

_____________

65
and the

maximum

odd

strain

ratio

0.5

2F0.2441

maxec
The

for

of odd to even

strain

is

given

by
toctet

sin

66

is

strain

even

max
max

Figure

knife

and Figure
edge

driven

from normal with
recovered

by

along the edge

0.2441

67

tan

67
analytic

result

28 gives

experimental

into

the surface

100-gm force The

interpolation

for

line force

at
strain

from the seven

of the cylinder

Since

the

at

results

response
tactels

maximum

esis

offer
is

0.1

for

however

and

aliasing

300

the

maximum

and odd functions

with this point as the origin The
of about

zcvee

27 shows the

30

for

E0

to

corresponds

sponse the even

the skew

10%

also

improved

signal

well

agrees

can

be

problem
force

Strain

result

with

obscured
errors

gauge

angle recovery

re

strain

determined

experimental

errors Amplitude

calibration
are

ratio

are

eq

by hyster
caused

methods

by

may

Brock and Chiu

1985
For

more general even

the line load the force

Fearing

than

pressure distribution

angle can

also

be

obtained

by

APEL907561

inverse

using
the

Taking Fourier transforms of

filtering

and odd impulse

even

we

responses

allows partial

68

neous

real

is

the normal

function the

frequency

of surface

component

and

stress

is

which

for

sponse

The

an

is

strain

Fs

frequency

component

of surface

inclusions

not an isotropic

medium

from the

theoretical

existence

Fs

of the inverse

by

filter

Es

71

thanks

ration

be

in

Rise and

sensor design and

ment and
comments

for

to

This work

tried

of inverse

limitations

and Binford

Ess

Tm

filtering

1989

72

by

Stanford

at

DARPA

provided

K-0002

and F336l5-82K-5

and

nition

Conclusion

elastic

Simple
tactile

it

for the

is

for

many

0.2

Perhaps
to

results

tactile

finger

is

forces

is

ade

and
Intrinsic

is

to

Robotics

Boie

for the

within
to

im

Stanford

improvements required
and the delayed

elastic

are

re

IEEE

sen

tactile

optical

Automat

Robot

31

Boie

eds

Roth

Research
1984

Proc

Atlanta

artificial

of the 4th

Capacitive

and

1985

Miller

Chiu

ASME

Winter Annual
Daniel

robot

hand

Meeting
and

mt Symp

Bolles

on

Press
readout

and

on Robotics

Conformable

4526043
1985
Miami

Califor

In

impedance

IEEE mt Conf
370378
pp

and

hands

Mass MJT

image sensor

an articulated

Cameron

for

Cambridge

sensor U.S Patent
Brock

August Santa Cruz

sensing

tion of

the

recog

Conf on

1137

1131

finger-shaped

1987

tactile

Automation

sufficient

Object

mt Joint

9th

applications

Dario

to

servo system were available

aliasing

robotic

nia

tactile

of applied

are already

for

and

Planar

Los Angeles

touch

pp

Intelligence

1988

and

Using interpola

reliability greatly

the biggest

reduce

tasks

is

with

of the sensor

and pressure estimation

initial

if

cylindrical

localization

manipulation

JPL hand

of this sensor

and

using vision

Biechi

interpretation

feature

1985

Bajczy

45472 484

design the

to

low-level

sensitivity

manipulation

tactel

These

data

in

tip The

techniques

within

need

tactile

encapsulated

hemispherical
quate

models were used both

sensor and to develop

strategies

prove

108

References

sors

20%

MDA9O3-86-

has not been

Artificial

tion

with

University

contracts

experimentally

Begej

that

was invaluable

are discussed in

This method

Mien

10

whose collabo

Gorali

fabrication

was performed while the author

student

graduate

funding

Fearing

are

discrepancies

large

guidance insight and encourage
Shekhar for helpful
Armstrong and

and

The

it

response

to

Binford

to

was

Hts

Presently these effects

70

flHs

Re

ftPs

the

for

the rubber

in

Acknowledgments

Ps

the

would be

account

to

of copper

but do not seem to cause

Many

solved assuming

now

hard

very

more homoge

conductors

stress

real the total pressure and angle can

is

is

which

strips

re

by

given

Ess
Since

function the

imaginary
the tangential
is

69

It

by imperfect

and copper

compliant

stiff

ignored

hx

analyze

of the

effects

for

response

to

caused

interface

at their

slip

sensor using

easier

may be

of the creep

between the rubber

adhesion

have

hx
which

Some

sponse

tactile

Environment

using contact

percep

sensors

pp 8996

Durrant-Whyte

1988

APE L0007562
22

The International

Journal

of Robotics

Research

Touch

Philadelphia

and

Elect

3271
1984

Fearing

1986
for

strategy
Robotics

San

Engineering

and
for

tactile

force

Implementing

on

sensor

during

Some exper

perception

and

and

shape

Philadelphia

determining

1985

partment of

Active

Electrical

1986

Siegel

gence

Contact sensors

TR-900

hands

den

in

SPIE Conf

on

sensing

Engineering

and

S.M

some

chanics

thesis

Computer

De
Science

Speeter
ulation

1953

normal loads on an
contact

McCammon

Biggcrs

Fl 1987

Ph.D

Thomas
Reading

ity

Intelli

based

stress

1985
sensor

tactile

and Computer
Stresses

due

Vision

to the

solid with

elastic

problems

Me

Applied

.1

tan

applica

ASME75l57l66

Trans

Analysis

thesis

1968

Mass

Timoshenko
and

robotic

Artificial

Hollerbach

Robots

Liu

neering Case Western
touch

and

capacitive

Intelligent

and

tion to

the Classical

for dextrous

Mass MIT

Cambridge

Garabieta

me

Rijn Netherlands

New York

Laboratory

434054

solid

1225

Processing

Sons

and

MIT
Jacobsen

John Wiley

of skin pre

bars edges and

4661204
Image

Signal

spatial

model

to

responses

Digital

Tactile

mechanics

Neurophysiol
1978

gential

Nordhoffl

1981

.1

Smith

pp 765771
Basic

Problems

aan

Using

curvature

Res

Robot

Contact

Alphen

mechanoreceptor

Siegel

Electrical

andAutomation

mm

dicting

Digital

Prentice-Hall

1981

continuum

III

En

Cambridge

1975

N.J

Cliffs

and Johnson

Phillips

13

Press

Schafer

Cambridge Mass.
1988

for

sensing

Sijthoff

of

Department

Hollerbach

of Elasticity

and

for

systems

Syst 81

Control

Mechanics

Contact

Englewood

Processing

sensing

tactile

IEEE

University

gland Cambridge

gratings

grasping

sensing

Robotics

Theory

1985

Pratt

University

1980

Gladwell

Hillis

Con

American

North Carolina

Binford

tactile

Fearing

grasping

hands

IEEE mt Conf
pp 96102

thesis

Stanford

IEEE mt Conf on
chanics

Francisco

Tactile

Ph.D

and

cylindrical

and

iEEE liv
andAutomation pp 16371643

1987b

Fearing

on

of

Design

manipulators

Johnson

resolution

sensing

Robotics

interpretation

Simplified

robot

Raleigh

tactile

Fearing

Trans

re-orientation

987a

Conf on

IEEE

silicon

1988

Oppenheim

Diego

dextrous

object

iments with

cell

capacitive

andAutomation

Fearing

performance

pp 3238

Conference

Fearing

1067

high

196120

San

with

manipulation
trol

on

based

imager

dextrous

1985

Wise

Devices

Phillips

pp 1062

and Automation

Robotics

Chun
tactiie

motion IEEE mt Conf on

and

New York

and

control

Department
Reserve

Calculus

of

of robotic

manip

Biomedical Engi

University

and Analytic

Geometry

Addison-Wesley

and

Goodier

1951

Theory

of Elastic

McGraw-Hill

APE L0007563
Fearing

23