-
Finger Force Precision for Computer Pointing
Ted Selker Joseph D. Rutledge
IBM TJ.Watson Research Center, Yorktown N.Y. 10598
ABSTRACTThis paper addresses motor control constraints
whichaffect analog pointing devices wsed with computer in-terfaces,
we have investigated th« accuracy and pre-cision with which
subjects can apply force to a smallisometric joystick with a
fingertip or finger andthumb, for osc and two dimensional force
speciUca*tion.
We find the icfonnation contest of A force applicationto be in
the range of 4 to 6 bits per dimension withvisual feedback, and
without time constraint, Theforce application task in two
dimensions is no moredifficult than in one dimension and gives
twice theinformation content. Us? of an opposing thumb andfinger
g?v«$ no improvement over * smg!e finger.Both inaccuracy «fil<
imprecision are concentratedalong the direction of the specified
force ^subjectstend to be both more accurate and more precise in
thedirection of force than in its magnitude.
Implications for aoatejj pub bag device} we discussed.
KSYWORDSMouse, Joystick* Dexterity. Force Precision
SWTRODUCTIONRecent work [5, 7} baa demonstrated that
pointingperformance can be significantly improved when hu-man motor
perceptual limitations ftre taken into ac-count.
This reopens questions of strategies for physical con-trol posed
at the time of radio knob desiga, etc,[2-4, 8] What ere the
perceptual motor constraintsof physical control design? now should
these con-straints affect the relationship between the
physicalcoatro! end a machine's response to its movement?By
understanding the relationships between these in-formation channels
we can improve design interfaces,
Motivated by work in designing a finger pressurecontrolled
pointer, we wanted to understand the con-trol channels available to
vinous analog motor taeke.We report here a stud,y of the
fmear.presstire channel.We have considered visual, auditory,
tactile and. pro*prioccptfve feedback. Some tactile feftd back is
inevi-tably present {barring anesthesia). Preliminaryexperiments
indicated that auditory feedback alone umuch less effective tbao
visual, even in the onc-di-mensional ease, wkik tactile feedback
aiono appearsto allow less than 3 bits in each dimension. Thisstudy
focuses on finger pressure with tactile and visualfeedback, without
tune constraint. The subject may
take several seconds to apply the specified force, andthe force
Is then measured as the mean of instantane-ous forces »mp!«d over a
2.4 second integrating pe-riod. Even under these conditions
accuracy ar»dprecision we surprisingly low.
The force range which we nave investigated Is thatappropriate
for one or two fingprs qn a small sccsor(a 3 mm by 8 mm cylinder),
0 - 225
While there are many studies of complex tasks suchas pointing
and tracking which use the finger or handforce channel, there seem
to be few if any which ad-dress the accuracy and precision
available in theehaiind itwlf. We have found only [6], which
studieswhole ann moveovsnt* *t much greater forces. Hisstudies
found sybiecis could apply a arm force towithin JOpercciit ol a
attempted target force,
METHOD
ApparatusSubjects were scaled at a standard office desk, m
achair adjusted to comfortable height by the subject.On the desk
were a CRT display »nd a J0!»kcy IBMPS/2 keyboard, plooed about 10
em bade from theedge of the desk. La the center of the keyboard,
be-tween the G and H keys, was an isometric joystick(the same
sensor used in the Pointing Stick, [5])topped by a dished 3*5 mm
diameter linger rest, 4nun above the level of the ksy caps Figure
I. Thejoystick top moves an undetsctable .13 mm at max-imum
force.
1. Fincertip grip oT an bometnc
An !BM 8514 b VGA graphics mode .4 mm square pixels on a 480 by
640 screen
3 0 0 ' 3 9 W d mtJw w a r 35 -B 15
-
For the fingertip grip condition the subject placed afinger tip
on the finger rest. For pen grip experimems,adjoining key caps wen;
rcraovec exposing an 8 mmlong 3 mm dierneter "per.', groped b*tw»*n
thumb«nd forefinger. Figure 2
tmt of an isometric joystick
Experiments were also run using a ainsJc finger orthumb pressure
on the side of the joystick to under-stand the value of the
opposing dint b the pen gripcondition. This is called the me,
gup.
Strain gauge signals from the sensor were processedto produce
signals on an IBM PS/2 pointing deviceinterface such that the
resulting cursor position re-presented the horizontal force Ibeiag
applied to thewtaa*, within the limits of the display screen.
Strain
Eige sensing and signal orocesstna was (performeda separate IBM
PC{XT with & Scientific Systemsbmastei data acquisition board
which communi-cated to the PS/2 through its mouse port.
A progrwis running on a IBM PS/2 model 80 pre-sented stimuli,
provided feedback, and recorded data.
Sub|«ct*Over several month periods one subject
performedexperiment* to calibrate and develop data
collectiontechniques.
We report here on results from four subjects, hiredthrough an
agency as office temporaries. They allnormally work in secretarial
ana clerical jobs, fre-quently using word processor, and bid sE
jjit famili-arity with puce, but ao prior experience with
Othercomputer pointing devices. All were women, between25 and *5CV.
who reported tywnE speeds between50 to 80 words per minute. Two
played or hadplayed a musical instrument; no other
lugh-maaua}-dSKterity hobbies were reported. Subjects
participatedin these experiments aa part of & two-day
sequenceof experiment* on pointing behavior, using thePointing
Stick in its normal mode and a mouse inaddition to the present
apparatus.
Proetaur*The cxpcrimtptaj paradigm is as follows; Subject
ini-tiates oacb. trial with • keypress. A target tone itpresented u
a position on the screen. The subject
attempts to apply the specified force, by bringing thecursor to
the target and holding it there.
An initial movement towards the target is ended whenthe
subject's precision limit is inevitably reached andthe cursor moves
away from the target. During thefollowing "hold" phase the cursor
is held as stably »spossible for wmc 2,4 seconds- The meen
appliedforce during the Tiolo" phase will be called the
trial'seffective force and its standard deviation is taker, asthe
trial's imprecision or dither. The miss vector is thedifference
between the target vector and the effectiveforce vector. The miss
angle is the direction of thetnlss vector.
Figure 3. DlspU) Showing subtlitjr andof force 4pplic»tion JB a
trial.
Feedback information u added to the target displayat the end of
e4ch trial, Th« computer displays thetrack of the cursor's
movement, A cross on the trackmarks the beginning of the hold phase
and an ellipseis centered at the mean hit position with shape
re-presenting the variation in applied force during theho54
phase.
Separating tubjeett target selection* ffotn target holdphke is
exemplified in Figure 3- hi this figure it isnoticeable that after
ft subject attempts to select atarget, stability at holding the
requested force can becharacterized as a cloud of dither.
After 10 such trials the computer displayed numbersindicating
feUtive accuracy and relative dither andstandard deviations for the
group of trials. A menuselection is selected to ruo ten groups of
these.
The subjects performed ihe« the trial type doing twodimensional
(circle) targets^ verttc*! one dimensional,end horizontal on*
dimensional targets, AU of theseconditions were investigated for
the Sneer tip as wellas pen crip conditions. A nominal 300 irsais
werecoliectec for each two dimensional condition and anominal 100
were collected for each one dimensionalcondition. For two subjects
an additional ipinica!100 trials were collected tor the vide grip
condition.
Target forces were pseudo-random, with uniformdistribution over
the rang* corresponding to the dis-
E00'3DUd DS HlbW HDyy3S3.il US I WOad 3S:9 IS, 91 D3Q
-
play screen., excluding an 1 cm »*rgb•no « disk of diameter 3
ceo to tins wrt'.«r.
An experuxtenter femaued in th* room to observerod to
demonstrate protocols. Written iasiruclionsexplained each phase of
the experiment, and directedthe sequence of phases; the sequence
was varied insom* cases to maintain subject motivation. Toe
ex-perimental protocol w*s otherwise Administered and«*«Ks recorded
by the computer.
RESULTSTarget force, mean applied force, dither, and timewere
recorded for each trial. Initial sets of trials werediscarded , as
were a lew trials invalidated by ints*-ruptlona or equipment
problems. Table I ave* over.all average! of mis* Mid dither for the
three targettypes, and for the two grips. Figures 4-12 look atthe
data in more detail,, examining dependencies ontarget force,
Figure 4 plot* error «.gs«n$t target fore* for the twopip
conditions. The figure shows that whik absoluteaccuracy of force
decreases with requested force, rela-tive accuracy increases with
force. The large range iaaccuracy inoicated by the quite tall cloud
shows thatrange in performance is wide- Note that the JWD
pipcondition adds * alight advantage over the finger
tipcondition.
Figure 5. Tb« relationship between force win!
Figure 5 plots dither against force for the two gripcondition*.
Notke the blank area at the bottom ofthe asreec Thi« represeDtt a
Emit tc steadiness. Thewidth of this band sliows that no one was
abb to holdforce steady to within 2 grams. The average instabil-ity
was 8 grams varying in individual from 5 to 12grams. Like the
error, other also increased somewhatwith forge.
Figure €. Dbecti&n oT miss i-erttis tarjd direction
show* th« relatloiuHip between the targetdirection and the mkz
angle. Note that th« miss tnjletendb ic be ISO degrees from lb«
target angls showmethat subjects tend to undershoot the target
forct. Ishows that swbjeeU wer? more accurate In the dirtc*lion
tban ic magatumte of force applicatiQa. Tcm dataKOuid be «cd to
help dtaiga velowty traasfcr fuac-tioaa Another OM ol this KUg^«t,s
that menut shouldbe made *d«tp* in the dlrectton of most likely
ap-proach. Edge menus, which are effectively infinitelydeep, are
instances of this.
DS HlbW HDdb3S3y UI3 1 WOdd ES'S 16 D3Q
-
Figure 7.
«•»!•
StaMHy in direction of target » wonc thusin ailhoBoitt}
direction _
A striking result cmer$es by looking at the sbe.pe ofth« data
aloud of subject hold data.. Figure 7. Theellipse that represents
the mean of the cither doudhas approximately a 2 to 1 axis ratio in
the directionof the target. We have collected data for pressure
to-wards targets above below to the right and to the left.Our data
collection procedure for angular forces docsnot distinguish this at
other angies.
i -
»t-
f Hi t* M» t.M *.«*»•»»»•
. • ..V.-A.&-** J v " " it*'
Figure 8. Demonstration that pen grip » not fccttothan pushing
with a side grip.
Experiments comparing the side grip to the pen griptest the role
of opposing grip for pressure accuracyand stability Figure 8. Note
that in the data the sidegrip is not d&iagujsbed from the pen
grip. Thissuggests that the slight advantage of the pen
gripcondition over the tip grip k due to improvement iacontrol by
pushing on the side of the post rather thanto any additional
stability provided by another finger.
Samples Error Dstber
Tip 1567
HIS
7.77.8
Pen
-
deviations to error and dither for the 2 di-mensional tip case
gives a value of (7.7 + 7.8) + (8.84 6.01™ 3Ujparr» « representing
aj> upper range forerror. This gives a precision of J to 12
percent, al-lowing 8 to 50 choices in each dimension Figure 9oo
page 4. Taking the log base two of these numbersrave* an
information coa**fii ranging iicfs 1 to 4.9bits for each trial.
Averaged over & lone bold time(2.4 seconds) subject accuracy
came to a precision of3 to 15 grams, giving 3.9 to 6.2 bits, for
each dimen-sion, Taking a long careful attempt, then, only al-lowed
o'or best subject to be ai>Jc to accurately selectOne of 70
force value* in each dkctjwson. Compiledtnto two dimeaaons this
gives a total informationcontent of 7.S to 12.2 bits.
The general leve! of information content reported bySchmidt
et.al. for arm force [6] (Standard deviationabout 10 percent of
applied force) is consistent with
Tat I dimmtiooal projection of &e 2 duneasianaJhorizontal
data results from setting aJQ horizontalcomponents to 0 fa the data
from 1 dimensional tri-ali. Note that it agrees closely vrith the
data for theone dimensional targets, Daia from trials with
verti-cal targets and the corresponding projectioo of the
2dimensional trial data are quite ttniulw, with the samedose
correspondence. The differences it error be*tween 1 and 2
dimensional trials is completely ac-counted for by the dimension of
the trail. Totsubjects appear to be able to add a second
dtaensteato their task without any interference giving twice
theinformation content without additional effort.
learning Curve
Figure 1i, Dejnonrtrttkss of th« experience lubjeclihad relative
to asymptotic improve-ffleat with use of pointing device.
The experiments described in this paper were includedin a two
day exercise designed to iratn subjects to us-ing isometric
pointing devices for movse-like selectionActivities, Aa with the
mcuee £tj w» find that a jub-j«dt awpraves at. selection tasks over
ft few thousandpointing selections. Figure Figure Hshows twopoints
at which Subjects performed these force ex*penmenu as part of a
longer set of mas designed togive them experience with isometric
pointing.
By noting the points along the performance curve atwhich dale
was gathered Tor force experimems, iheimpact of pointing experience
on these results can beestn»ftte$. Figure Figure 12 thaw* that this
data i$coBsisteat with all other data describing accuracy cffifcger
force.
l .Wt.U
i
Figure 12. Fwce accuracy data taken at tStffwwttimes on t user's
joystick pobtbglearning curve.
CONCLUSION
Limits of Digit Fore* Centre)Even under the optimal condition of
long integrationtbne, accuracy is quite Uwited. If we take the
workingrange for this kind of finger pressure as 0 - 225grn,average
ihort*terra preciaon, for oar fubjects, ss inthe range of 9 to 25
jrams or 4 to 10 percent of fullrange in each dknenaon,
eontspoDding to an infor-mation content qf 3.3 to 4,6 bits for each
trial. Overtime (2.4 jeeonds) this b sjiegrAted to a {trccitioc of3
to IS grams, saying 3-9 to 6.2 bits, for each dimen-sion, for a
total iniotinalioD content of 7.8 to 12.2bits. This compares with
18.22 bits represented bythe selection of a single pixel on a VGA
(640x480)screen. A force-to-positipn joystick is dearlv not
ad-equate as a pointing device. By mapping force intovelocity
ia«t«*d of pocition^ efficient time ieiepra'.icaa&d afbhra-riiy
precise pomtmg may be achieved,provided that the proper mapptng is
used. The im-precision of finger pressure impiies that speed can
beclosely controlled onJy when the mapping h&s a pla-teau,,
where the desired speed is maintained over arange of pressure*
larger than the 4 to 10 ptrcent un-certainty. In the Pointing stick
transfer function suchplateaus are found at zero speed for a
stopped cursor,t siow speed for accurate pixel and character
posi-tioning, and at maximum eye tracking speed for fasttccurate
rjaovements [5] - Figure id traosfe' un-known-
Over the range of target fare** t**t*
-
Subjects applying a force in two dimensions mademuch the same
errors along each axis «« when theywere concentrating on that a*i$
alone. One mightexpect that it would be easier to apply and hold
a$p*c&ed force to the right, for example, if one needcot contmt
in the up or down direction., This it notthe case.
Rote of Grip in Fere* ControlIt might have been expected that
the opposing thumband forefinger hold would greatly improve both
ac-curacy s,ud ti*ftdifi«*s of force ttpplicahpa. Data fromthis
Study shows that the pen grip improved per.fonnance very little
(Figure 4 on page 3, Figure 9 onpage 4). This improvement, however,
does not comefrom the opposing thumb as predicted; a anger placedOB
the aide, of the joystick in the side grip conditionslightly
outperform* the pen grip Figure 8 cm page4shows that the observed
force accuracy advantageof the pen grip is due to the finger or
thumb positionon the side of the sensor rather that) on the top,
andoot to the opposing digits grip. Two unsteady fingersare just as
unsteady as one unsteady finger,
Error and Otther mrm Align** in iho TargetApplied force leads to
be aligned to the diieetidu oftarget force but undershot. For force
to velocitytransfer function* this is an advantage; the Boise
alongthe intended direction of movement changes the cur-sor speed
but not its diroctbn.
A specific style of osenu that this tuggests u a m«oualong the
edge of a screen (which acts as if infinitelydeep).
The result* of these experiments help describe behav-ioral motor
issues which contribute to the perform-ance ucp-rovecaenii achieved
by tb* Pointing Sticktransfer function,
This study shows that individual performance differby a large
factor; could ibis be takes into account mpersonalized or adaptive
transfer functions?
The noise in a persons movement U greater Ln the axisdirectly
blersecting the target; could this analysis beutilized to to
augment user control?
1. S. K- Card, W. K. English, and B J. Burr.Evaluation of Mouse,
Ratc»Coctro!ied Iso-metric Joystick, Step Keys,,
Ergoiomia,21(8):6Gl-6i3, 1978.
2. P. M, Fitts The Information Capacity of thei Human Motor
System In Controlling T.
3. P. M. Fjtts and I. R Peterson. InformationCapacity of
Discrete Motor Responses,Journal of Experimental Psychology,67(2);
103- 1 1 2, 1*64,
Leroy jenkms *fid Mittfia B. Con-ner, Some Design Factors in
Making Set-tings on s linear Seaje, Applied Psychology*,33:7-25,
1949.
Joseph D. Ruticdgc. and Ted Sdkcr, Forceto Motion Functions For
Pointing. Interact'90 Prottedingx, North Holland Amsterdam,August
1990.
R.A. Schmidt, N.H. Zelasnik, B- Hawkins,J.S. Frank, and J,T.
Quints, Jr. Motor Out-put Variability, A Theory for the Accuracyof
Rapt. Psychological Review, 86:415-451,1979,
Andrew Sears and Ben Shneiderman, HighPrecision ToucbscreeEts;
Design Stratecit;Acd. Department of Computer Sdcnce Uni-vwshy Of
Maryland., CAR-TR-4SO- 1988,
R, S, Woodwortij. Accuracy of VoluntaryMovement. The
Prychologic&l Review, Seriesof Monograph St^piemenlj, 3( 3}:1
114, July
100•39yd DS Hiaw Hoyb3S3y wai SS:S 16 , 91 D3Q