4 DIFFERENTIALKINEMATICS

Purpose:
Thepurposeofthischapteristointroduceyoutorobotmotion.Differentialformsof
the homogeneous transformation can be used to examine the pose velocities of
frames.Thismethodwillbecomparedtoaconventionaldynamicsvectorapproach.
TheJacobianisusedtomapmotionbetweenjointandCartesianspace,anessential
operationwhencurvilinearrobotmotionisrequiredinapplicationssuchaswelding
orassembly.

4.1 DifferentialDisplacements
Often it is necessary to determine the effect of small displacements on kinematic
transformations. This can be useful in trajectory planning and path
interpolation/extrapolationorininteractingwithsensorsystemssuchasvisionortactile.
NowifwehaveaframeTrelatingasetofaxes(primedaxes)toglobalorbaseaxes,then
asmalldifferentialdisplacementoftheseaxes(dp,d=dk)relativetothebaseaxes
resultsinanewframe T'= T + dT = H(ddp) T. Althoughfiniterotationscan'tbe
consideredvectors,differential(infinitesimal)rotationscan,thusd=dk.Thus,the
formofdTcanbeexpressedas
dT=(H(ddp)I)T=T

(4.1)

Equation (4.1) expresses the differential displacement in base coordinates.

Displacements made relative to the primed axes ( dp, dexpressed in primed
axes)changetheformofdTto
dT=T(H(ddp)I)=TT

(4.2)

Consideringtranslationandrotationseparately,anddefiningorT(dependingupon
theaxesofreference)equaltoH(ddp)I,wedeterminethat
H (d)=

R(d)0
0T1

where
1 kzdky d
R(d)= kzd1kx d
ky dkx d1

41

(4.3)

Equation(4.3)comesfrom(2.1)usingsindd,andcosd1.
Forpuredifferentialtranslation,
H(dp)=
where

Idp
0T1

(4.4)

dpT= dpx dpy dpz

NowsinceH(d,dp)=H(dp)H(ditiseasilyshownthat,
(d,dp)=H(dp)H(d)I

0
k d
d dp z
- k y d

- k z d
0
k x d
0

k y d
- k x d
0
0

dp x
dp y

dp z
0

(4.5)

Given(4.5),d anddp,dT canbedeterminedineither(4.1)or(4.2)andthenatural

extensiontovelocitiesofpointswithintheseframesappropriatelyfollows.
Inthecaseofpurerotation,(4.5)reducestoadifferentialscrewrotationabouttheunit
vector k sincethedifferentialtranslationinthe4th columnreducestozero.Takingthe
derivativeof(4.5)wegeneratethescrewangularvelocitymatrixofequation(4.7)in
Tsai'stext:

0
z
- y
0

- z
0
x
0

y
- x
0
0

0
0

0
0

(4.6)

4.2 RelatingDifferentialTransformationstoDifferentCoordinate
Frames
GivendTrelativetothebaseaxes,whatisthedifferentialtransformationinx'y'z'such
thatT=TT?Inotherwords,givendThowareandTrelated?Therelationship
isdefinedbyequating(4.1)and(4.2)sothat

T=TT
whichcanbesolvedfor
=TTT1

(4.7)
42

or
T=T1T

(4.8)

Todeterminethecomponentsof(4.7)and(4.8),let
ax bx cx px
ay by cy py
T= azbzczpz
0001
anddefinethedifferentialrotationsandtranslationsrelativetobaseaxesby
=x i+y j+zk
d=dx i+dy j+dzk
where
x =d x dx =dpx
y =d y dy =dpy
z=d zdz=dpz
Thus,
0zy dx
z0x dy
= y x 0dz
0000

(4.9)

whichcomesfrom(4.5)bynotingthat x=kx d, y=ky d,and z=kz d (the

directioncosinesresolvedintoitsi,j,andkcomponents).
PremultiplyingTbyitcanbeshownthatTassumestheform
xa x xb x xc x xp +d x
xa y xb y xc y xp +d y
T =
xa z xb z xc z xp +d z
0
0 0
0

(4.10)

PremultiplyingbyT1 toobtain(4.8),
a (xa)a (xb)a (xc)a (xp )+d
b (xa)b (xb)b (xc)b (xp)+d
T =T1 T = c (xa)c (xb)c (xc)c (xp)+d
0
0 0
0
43

(4.11)

Equation(4.11)canbereducedusing

totheform,

a(bxc)=b(axc)=b(cxa)
a(bxa)=0
axb=c
bxc=a
cxa=b
0cb(pxa)+dxa
c0a(pxb)+dxb
T =
ba0(pxc)+dxc
0000

(4.12)

Butintermsofdifferentialdisplacementsrelativetotheprimedaxes
0z' y' dx'
z' 0x' dy'
T =
y' x' 0dz'
0000

(4.13)

Equating(4.12)and(4.13),
x'=a=Ta
y'=b=Tb
z'=c=Tc

(4.14a)
(4.14b)
(4.14c)

dx'=(pxa)+da
dy'=(pxb)+db
dz'=(pxc)+dc

(4.15a)
(4.15b)
(4.15c)

Note that we can easily determine the elements of given the elements of T by
applyingtherelationship=TTT1whichcanbegatheredintotheformof(4.7)as,

=(T1)1TT1

(4.16)

Thuswecanapply(4.14)and(4.15)bysimplyusingTandtheinverseofT.

4.3 Velocities
Vectorwcanberesolvedintocomponentspinthebaseframebytheequationp=Tw.
Nowunderasmallrotation/displacement,frameTcanbeexpressedbythenewframe

44

T'=T+dT=T+T

(4.17)

AvectorwinframeT,onceperturbed,canbelocatedgloballybyp'suchthat
p'=(T+T)w

(4.18)

Thedeltamove,p'p,becomes
p'p=(T+T)wTw=Tw
z

Z
T

y
x

z'

T'
p'

y'
x'

X
Figure41Velocityofapointinamovingframe
Dividingbytandtakingthelimitast0(takederivative),wedeterminethevelocity
inthebaseframe:
Tw
v

(4.19)

where

0 k z k y p x

k z
0 k x p y
- k y k x
0 p z
0
0
0
0

(4.20)

whichcanbereducedtothesimplerform

0

z

- z
0
x
0

y
- x
0
0

45

vx
vy

vz
0

(4.21)

expressedinangularandcurvilinearvelocitycomponents.
4.3.1

Example

ApointPislocatedbyvectorinframeCasshowninFigure42.Atthisinstantthe

P(3,5,0)relative
toframeC
FrameC

O
(2,3,0)

2rad/s
X

Figure42Example
frameCoriginisbeingtranslatedbythevelocityvt=2I+2Jm/srelativetothebase
frame.Theframeisalsobeingrotatedrelativetothebaseframebytheangularvelocity
=2rad/s K where IJK aretheunitaxisvectorsforthebaseframe.Determinethe
instantaneousvelocityofpointPusingboththeconventionalvectordynamicsapproach
andalsousingthedifferentialmethodsinthischapter.
Conventional:
v=vo+x
wherevo=vt+xr. At this instant note that frame C is aligned with the base frame
so that the unit axes vectors are parallel.
It can be shown that xr = -6I + 4J so that
vo== 2I+2J6I+4J = 4I+6Jm/s
Itcanbeshownthatx = -10 I + 6 J,sothat
v= 4I+6J-10 I + 6 J = 14I+12Jm/s

46

Differentialtransformations:
Apply

Tw
v

wherew=and

. 0 2
2 0
0 0
0 0

2
2
0
0

0
0
0
0

1
0
0
0

0
1
0
0

0
0
1
0

2
3
0
1

to give

14
m/s
v 12
0
0

Note that the same values are obtained!

4.4 JacobiansforSerialRobots
The Jacobian is a mapping of differential changes in one space to another space.
Obviously,thisisusefulinroboticsbecauseweareinterestedintherelationshipbetween
Cartesianvelocitiesandtherobotjointrates.
Ifwehaveasetoffunctions
yi=fi(xj)

i=1,...m;j=1,...n

(4.22)

thenthedifferentialsofyicanbewrittenas

y i
j1

f i
x j
x j

i=1,...m

(4.23)

orinmatrixform,
where

y=Jx

(4.24)
f i

x j

J=Jacobian=

(4.25)

In the field of robotics, we determine the Jacobian which relates the TCF velocity
(positionandorientation)tothejointspeeds:

V Jq

(4.26)
47

 .
q1

V
J (q) .
v
.
q
m

(4.27)

where

vistheTCForiginvelocity
istheorientationangularvelocity(equivalentEulerangles,etc.)
q isthejointvelocities(jointrotationalortranslationspeeds)

Waldroninthepaper"AStudyoftheJacobianMatrixofSerialManipulators",June,
1985, ASME Journal of Mechanisms, Transmissions, and Automation in Design,
determinessimplerecursiveformsoftheJacobian,givenpointsonthejointaxes(ri)and
jointdirectionvectors(ei)forthejointaxes.ThevelocityofpointP,apointonsome
endeffector,canbedeterminedbythefollowingequation,referenceFigure43:
P

rn
en
r1

ri

r i+1

e i+1

ei
e1

Jointn

Jointi+1
Jointi

Joint1

Figure43Waldron'snotation
n
v e i x ri

(4.28)

i 1

Equation(4.28)canalsobereducedtotheform
n

v i x i

(4.29)

i1

48

whereiistheabsoluteangularvelocityoflinkiandiisthepositionvectorbetween
jointaxesi+1andi:

i
i je j

(4.30)

j 1

The Jacobian is determined by

collecting the velocities for each joint
intotheform:
J=

ei
x
e i ri

(4.31)

andthevelocityrelationshipbecomes:
"R""P"

e
V i
v e i x ri

Figure44Singularityforserialrobot

0
q
e i

(4.32)

GiventheTCFvelocity,wedeterminethejointvelocitiesbytheinverseJacobian:
q=J1 V

(4.33)

Waldron,et.al.,illustrateseveralclosedformsolutionsfortheJacobian,but,ingeneral,
thereisnogeneralclosedformsolutionforrobotswitharbitrarystructures.NotethatJ1
onlyexistsforasixaxisrobot.Robotswithlessthan6jointsmapthejointspaceintoa
subspaceofCartesianspace.Forexample,acylindricalrobotwilltypicallymaptheone
or two rotational joints into one angular velocity component (usually about the base
frameZaxis).
Sixaxisrobotsoftenwilldisplayconfigurationswherejointslineup.Werefertothisas
jointredundancy.InthesecasesJ1isundefinedbecausetherankofJislessthan6and
the J determinant vanishes. There are also singular configurations where det(J) = 0
becausemotioninCartesianspacecauseslargejointratesseeFigure44.

4.5 SerialRobotSingularities
ConsideraserialmechanismanditsJacobian J thatmapsthejointmotiontotheend
effector(tool)motion:

v=J q

(4.34)

Vector v isamx1vectoroftoolvelocitycomponents inCartesianspace(sometimes

isanx1vectorofthemechanismjointratesforn
referredtoastaskspace),while q

49

joints.Normallym=6.Thespaceofthetoolmotionissometimesreferredtoasthe
operationalspace.
Forconvenienceweassumethattherobotstoolorigincoincideswiththeoriginofthe
terminaljoint.Inthecasewherethemechanismhasa3jointterminalsequencearranged
asasphericaljoint,wechoosetheconcurrentpointofthelast3joints.Thischoicewill
simplifytheinvestigationofmotionsingularities.
Assuming full (R6) Cartesian mobility, the Cartesian velocity components in v
correspond to 3 orthogonal angular components and 3 orthogonal linear components
(showninrowform):
vT=[xyzvxvyvz]

(4.35)

TheJacobianmay beresolvedinto therobotsbaseframeorintoanyselected joint

frame.Ifarobothasaterminalsphericaljointsequence,asisthecasewithmanyrobots,
anddependingontherobottype,itmaybeusefultoresolvetheJacobianintojoint4to
decoupletheorientationandlinearcomponents.
Mechanismsthathave < 6jointsshouldhavetheirzero(ordependent)rowsremoved
fromtheJacobian,alsoremovingthecorrespondingCartesianvelocitycomponentsfrom
v (leavingm=6orlesscomponents)ThesecomponentscorrespondtoCartesiantask
motion components that cannot be generated by any mechanism joint motion. For
example,amechanismthathasonlyonerevoluteaxiswillonlybeabletogenerateoneof
theCartesianangularvelocitycomponentsin(4.35).
IfCartesianmotionisspecified,thenrobotcontrolmustdeterminetheappropriatejoint
ratesfrom
=J1v
q

(4.36)

This requires that J be square and invertible. We can determine invertibility by the
determinant of J, namely |J|. The Jacobians rank depends on the mechanisms
kinematicsconfiguration,numberofjoints,andjointtypes.Motionatasingularitywill
reducetherank,resultingin|J|=0.Motionnearasingularitywillcause|J|toapproach
zero and generally lead to higher joint velocities near singularities. At or near a
singularity the robotstoolwilleither losesomedirectional Cartesianmobility (even
thoughrobotjointscanmove),orrequiresomejointstomoveathighspeedstoattainthe
mobility.

410

The robot is degenerate in

configurations that cause it
to lose mobility in certain
directions. These directions
arereferredtoasdegenerate
directions that cannot be
instantaneously achieved
with any joint motion.
Effectively,robotmobilityis
limitedatornearthecurrent
configuration.
4.5.1

a)Boundarysingularity

SingularityTypes

b)Singularitydue
tojoints4and6
alignment

Three singularities are

c)Sphericaljointsorigin
showninFigure45.Fig.4
losespositionmobility
bylyingonfirstjoint
5a,theworkspace boundary
axis.
singularity, occurs when
joint 3 is zero (or 180
degrees) and links 2 and 3
are radially extended (or Figure45Typicalsingularities
folded). The line along
extended(orfolded)links2
and3iscalledadegeneratedirectionbecauseitisnotpossibletomovetherobotstool
pointinthisdirectionwithoutverylargejoint2and3motions.
InFig.45bjoints4and6ofthesphericaljointsequencearecolinearwhenjoint5is
zero.Thus,joints4and6areredundantwithrespecttotheiralignedrotationalaxes.This
singularity is called a redundancy singularity.Others call ita selfmotion singularity
becausejoints4and6canberotatedinacorrelatedfashionsothatthetooldoesnot
rotateaboutthedirectioncolinearwithjointaxis4andjointaxis6.Infactthetoolwill
notmoveorrotateinCartesianspaceifjoints4and6arerotatedoppositely.Insuch
caseswesetv=0andreduce(4.34)totheform:
=0
Jq

(4.37)

Equation(4.37)expressesconfigurationswhere J hasa nullspace.Thetrivialsolution

=0existswhenJisnotsingular.WhenJissingularandthus|J|=0,
representedby q
isnotzeroand(4)issatisfied.Thisdefinesmathematicallya
therearesolutionswhere q
selfmotionsingularity.Mechanismswithselfmotionsingularitieswillexhibit(4.37)at
selectedconfigurations.Somemethodsformovingthroughselfmotionsingularitieswill
reconfigurethemechanismusingtherelation(4.37)sincethetoolisnotmoved,andthen
escapethesingularityusingthisnewconfiguration.Thismeansthatthemechanismis
slowed,thenstopped,atthesingularity,reconfigured,andthenanewpathplannedto
escapethesingularity.Thismaynotbefeasibleinmanytasks.
NoteinFigure46thatifthetaskmotionrequiresthatthetoolberotatedaboutascrew
axisinstantaneouslyperpendiculartotheplanecontainingjointaxes4and5,without
411

changingthepositionofthetoolpoint,
it is not possible without an infinite
joint5rate.Thisperpendicularaxisis
thedegeneratedirection.

z5

z5

z4,z6

InFig.45cthesphericaljointorigin
Degeneratedirection
(concurrentintersectionpointoflast3
joints)liesontheverticaljointaxisof
joint1.Thus,jointonemotioncannot
change the spherical joint origin
position and the robot loses design
mobility.Infact,thereisnomotionof
the first 3 joints that can instantly
move the tool point perpendicular to
theplaneoflinks2and3whenlinks2
Fig.46Degeneratedirection
and 3 are not colinear. This is a
degeneratedirectionanotherexample
ofaselfmotionsingularitywhichisremovedifthetoolpointdoesnotlieonthefirst
jointaxis.Itispossibletomovethesphericaljointsinsuchawayastocanceloutthe
joint1rotationofthetool,keepingthetoolpointfixedinspace.Thus,thenullspace
equation(4.37)holdsforthissingularity.
Morethanonesingularitycanoccuratthesametime. Forthis robot,iftherobotis
extendedradiallyandverticallyupandjoint5isalsozero,thentherankoftheJacobian,
normally6,willbereducedto3.
Theselfmotionsingularitiesjustdescribedmayhaveaninfinitenumberofnonfeasible
directions,sincethereareaninfinitenumberofuniquetaskspacedirectionsthatgivea
nonzeroprojectionontothedegeneratedirection.Forexample,considertheFig.45b
singularityshowninFigure46.Thedegeneratedirectionisperpendiculartotheplaneof
thez4/z6andz5jointaxes.Therobotcannotinstantlyrotateaboutthisaxisandkeepthe
robottoolpositionfixed.Infact,anytaskspacerotationalaxisthatdoesnotlieinthe
planeofz4/z6andz5isnotfeasibleuntilaxis5isfirstrotatedseethedashedscrewaxis
inFigure46.
4.5.2

MotionPlanningNearorThroughSingularities

Itwouldseemthattherearetwocasesthatmustbeconsideredwhenmovingnearor
throughsingularities,whenthetaskcannotavoidthesingularities:
Case1movingthroughasingularity
Case2movingnearasingularity,butnotthroughasingularity
Case1shouldbeavoided,butthismaynotalwaysbepossible,dependingonthetask
requirementsandthemechanismtype.Mechanismslikerobotshavedifferenttypesof
singularities and related degenerate directions. Some singular configurations can be
traversed with proper motion planning, while others cannot without an additional
sequenceofinternaljointmotionsthatmaynotmaintainthedesiredtoolmotion.For
412

exampleiftherobotofFig.45ismovedinjointspacetoaconfigurationwherejoint5is
zero,thenitisnotpossibletorotatethetoolaroundthedegeneratedirectioninFigure46
withoutinfinitejoint5rates.
NowconsiderthattherobotstoolinFigure45a(boundarysingularity)istobemoved
radiallyinwardataspecifiedvelocity.Thisisnotfeasiblewithoutimpossiblylargejoint
2and3rates,yetpossibleifthetoolvelocitycanberelaxedlocallybyhavingthejoint2
and3ratesslowlyincrease.Thetoolvelocitycanbeplannedtoslowlyincreaseuntilthe
desiredvaluecanbeachievedastherobotmovesawayfromthesingularconfiguration.
Reference[1]describesthistypeoftrajectorymotionnearthistypeofsingularity,but
doesnotofferanymethodsnewtotheDMAClookaheadmethods.
Ingeneral,Case2motionnearsingularitieswillcauseexcessivejointrates,requiringthe
Cartesianmotionbeslowedaccordingly.Cartesianslowingmaynotalwaysbefeasible,
dependingonthetask(e.g.,welding,painting). Motionat ornearasingularity must
consider the degenerate directions and the type of singularity. Motion along/about a
degeneratedirectionisnotpossiblewithoutoneormorejointsexperiencingexcessive
rates.Infactanymotionthathasaprojectioncomponentonthedegeneratedirectionmay
notbefeasiblebecauseoflargejointrates.InthesesituationstheCartesianmotionmust
eitherbeslowedsignificantlytomovethroughoraboutthesingularityortheCartesian
motionbeconvertedtojointmotion.Inapplicationsthatdemandclosepathfollowing,
whereonlyslightpathdeviationsareallowed,thereisnoalternative.
One method often referenced is the damped least squares method [2], [3] which
damps/slows the task motion in proximity to singular configurations. Reference [4]
describesarelatedmethod,callednaturalmotion,whichalsoslowsmotioninproximity
tosingularconfigurationsbutwithsomelossofpathaccuracy.Researchersarequickto
notethat theresultingpathmay notapproximate truepaths welland,inthecaseof
references [2] and [3], the endeffector path may not reach or pass through the
singularity, if required. Although these methods may be viable for tasks like part
assemblythatdonotrequiretruepathfollowing,theyfailwhenpathfollowingmustbe
accurate.
ObservationDMACispresentlydesignedwithlookaheadmethodstodeterminejointvaluesalong
the path, for identifying cases where large joint changes occur over small path displacements,
signalingsingularityornearnesstosingularity.Whensuchcasesoccur,theCartesianpathmotion
can be slowed to bring the joint rates (including joint accelerations) to within limits or,
alternatively,themotioncanbetransitionedtojoint spacewhennearsingularities.Shiftingthe
motiontojointspacewillrequireafinerspacingoftrajectorypointsalongthepathtomaintainpath
accuracy.Aquestiontobeansweredishowmuchthejointspacetrajectorydeviatesfromthetrue
pathwhenusingajointspacetransitionscheme.

Nearnessofsingularitycanbemeasuredbythevalueof|J|asitapproaches zeroto
withinsometolerance.Incaseswherethemechanismexhibitsmorethanonesingularity
theJacobiandeterminantcanbeexpandedintotheform[4.38]:
|J|=s(q)=s1(q)s2(q)sk(q)(k=numberofsingularities),

413

(4.38)

where one or more of the singularity functions si(q) will vanish at a singularity
configuration. Given (4.38) it is not clear how to determine nearness to singularity
configuration,sinceanumberoffunctionssi(q)areinvolved.Onecouldassignnearness
bythecondition:

if|si(q)|<iforanyi=1,2,,kconfigurationnearsingularity(s)

(4.39)

bysettingappropriatevaluesforthetolerances i.Unfortunately,whatdefinesproper
tolerancesisnotclear.
Alternatively,wecouldusetheJacobiandeterminantinthefollowingcheck:
if|J|<,motionnearsingularity(s)

(4.40)

wherewemustagaindetermineanappropriate.Theadvantageof(4.39)isthatitwill
identifythesingularitytype.
Reference[6]isanexamplebyoneofanumberofresearchersthatconsidershowto
movearoundsingularities.Theapproachistoremovethedegeneratedirection(s)from
the Jacobian by removing the appropriate row(s) at or near the singularity. Motion
planningisthenconductedinthespaceorthogonaltothedegeneratedirection(s),since
thereducedJacobianonlyallowsthistypeofmotion.Thisalsoconstrainsthepossible
pathdirectionswhichmaynotbefeasibleiftheactualpathhasmotioncomponentsalong
thedegeneratedirection.Inthesecasesaconversiontojointmotionmaybenecessaryto
makethetransition.WewillnametheconversiontojointmotiontheJointTransitional
Method(JTM)andconsideritinaseparatesection.
AnotherapproachoftenusedtoidentifynearnesstosingularconfigurationsisSingular
ValueDecomposition(SVD)onJ,whichwillgenerateandsortindescendingorderthe
diagonal terms in the diagonal matrix that get smaller as the mechanism motion
approaches singularities. If the mechanism is far removed from any singularity
configurationsnodiagonaltermswillbesmall.
Thedecompositionformis
J=UVT,

(4.41)

whereUandVTarebasismatricesforCartesianspaceandjointspace,respectively,with
their columns serving as orthonormal basis vectors for their two respective spaces.
Reference[4]notesthat,atsingularconfigurations,thelastcolumnsofU,withnumber
equal to the rank deficiency of the Jacobian, represent basis vectors for the singular
subspace.Anyvectorinthissubspaceisdegenerate,andanypathwhoseprojectiononto
thissubspaceisnonzerowilllikewisebedegenerate.Anypathsthatlieinthesubspace
orthogonaltothesingularsubspace(expressedbythesubspaceofthefirstcolumnsofU)
willnotbedegenerateandwillnotexperiencelargejointrates.Thus,SVDmethodscan
establishfeasiblepathdirectionsnearsingularpointsthatdonotcauselargejointvalues.

414

If you have the freedom to deviate the path, then these methods may be useful. A
limitationisthatcalculatingtheSVDofJcanbecomputationallyexpensive.
4.5.3

Conclusions

Researchers have examined motion around singularities using a number of methods,

many not mentioned here. But a fundamental premise of these methods is that path
deviations are allowed around singularities (in some cases a sharp reduction in path
velocitiesmaybealsoallowed).Aquestionnotansweredinthesepapersishowmuch
deviationisallowedandhowtoselectpathswhenseveralpathchoicesareavailable.
Unfortunately,manyrealtasksdonotpermitthesedeviations.
Afewofthepapershavementionedconversiontojointmotionasasolution,butwithout
furtherdetail.Thissimplertransitionalapproachisprobablythepreferredapproachused
bymanyvendors.

4.6 DifferentialMotionSummaryforSerialRobots
Differentialformscanbeappliedtogeneratemotionofframesorofpointsinframes
whentheframe experiences translational and/orangularmotion.This method canbe
simplertoapplythantheconventionalvectorequations.
The Jacobian is important in robotics because the robot is controlled in joint space,
whereastherobottoolisappliedinphysicalCartesianspace.Unfortunately,motionin
CartesianspacemustbemappedtomotioninjointspacethroughaninverseJacobian
relation.Thisrelationmightbedifficulttoobtain.
TheequationsfortheJacobianareeasytodetermineforarobot,giventhecurrentrobot
configuration,asoutlinedbyWaldronet.alinthepaper"AStudyoftheJacobianMatrix
ofSerialManipulators",June,1985,ASMEJournalofMechanisms,Transmissions,and
AutomationinDesign.

4.7 JacobianforParallelRobots
Letusdefinetwovectors: x isavectorthatlocatesthemovingplatform,while q isa
vectorofthenactuatorjointvalues.Ingeneralthenumberofactuatedjointsisequalto
thedegreesoffreedomoftherobot.
The kinematic constraints imposed bythelimbs will lead toaseries ofnconstraint
equations:
f(x,q)=0

(4.42)

Wecandifferentiate(4.42)withrespecttotimetogeneratearelationshipbetweenjoint
ratesandmovingtablevelocities:

Jx x =Jq q

(4.43)

415

whereJx=f/xandJq=f/q.
Thus, the Jacobian equation can be
writtenintheform

q =J x

(4.44)

whereJ=Jq1Jx.
Notethatequation (4.44)is alreadyin
thedesiredinverseform,giventhatwe
candetermineJq1andJx.
4.7.1

SingularityConditions

Figure47
Aparallelmanipulatorissingularwhen
either Jq or Jq or both are singular. If
det(Jq)=0,werefertothesingularityasan inversekinematicssingularity.Thisisthe
kindofsingularitywheremotionoftheactuationjointscausesnomotionoftheplatform
alongcertaindirections.Forthepickerrobotthisoccurswhenthelimblinksalllieina
plane(2i=3i=0).ThiscaseissimilartothecaseshowninFigure44foraserialrobot.
Ifdet(Jx)=0,werefertothesingularityasa direct kinematicssingularity.Thiscase
occursforthepickerrobotwhenallupperarmlinkagesareintheplaneofthemoving
platformseeFigure47.Notethattheactuatorscannotresistanyforceappliedtothe
movingplatforminthezdirection.

4.8 JacobianforthePickerRobot
Referringbacktoequation(3.30)inthenotes,wecanwritetheloopequation
OP+PCi=OAi+AiBi+BiC

(4.45)

to close the vector from the base frame to the center of the moving platform. We
differentiate(4.45)toobtaintheequation
vp=1ixai+2ixbi

(4.46)

forthevelocityofthemovingplatformatpointPandwhereai=AiBiandbi=BiCi.
Thelasttermin(4.46)isafunctionofthepassivejoints.Wecaneliminatethemfromthe
equationbydotmultiplying(4.46)bybitoget
bivp=1i(aixbi)

(4.47)

416

Expressingthesevectorsintheithlimbframe,wecanget
.

jixvpx+jiyvpy+jizvpz=as2is3i 1

(4.48)

where
jix=c(1i+2i)s3icic3isi

(4.49a)

jiy=c(1i+2i)s3isi+c3ici

(4.49b)

jiz=s(1i+2i)s3i

(4.49c)

Wecanassemble(4.48)and(4.49)intoequationsforeachlimbbythematrixequation:

Jxvp=Jq q

(4.50)

where

j1x j1y j1z

Jx= j2x j2y j2z

j3x j31y j3z

(4.51)

0
0
s21s31

0
s22s32
0
Jq=a

0
0
s23s33

(4.52)

Question:Whatobservationcanyoumakefromtheformsof(4.50)(4.52)?

4.9 JacobianSummary
TheJacobianinroboticsisparticularlyimportantbecauseitrepresentstheratemapping
between joint and Cartesian space. Unfortunately, for serial robots the desire is to
determinethejointratestogetdesiredmotioninCartesianspaceasthetoolismoved
alongsomelineinspace.ThisusuallyrequirestheinverseoftheJacobian,whichmaybe
difficultforsomemechanisms.
TheJacobianforparallelrobotsmaybeeasiertofind,asisthecaseforthePickerrobot.

4.10 References
1. Lloyd,J.E.,andHayward,V.,SingularityRobustTrajectoryGeneration,IntlJ.
ofRoboticsResearch,Vol.20,No.1,pp3856,2001.

417

2. Nakamura, Y., and Hanafusa, H. 1986. Inverse kinematics solutions with

singularityrobustnessforrobotmanipulatorcontrol,Trans.ASME,J.Dynamic
Systems,MeasurementandControl108:163171.
3. C.W. Wampler 11, Manipulator inverse kinematics solutions
based on vector formulations and damped least-squares
methods, IEEE Trans. on Systems, Man and Cyb., Vol. SMC-16,
No. 1, pp. 93-101,1986.
4. D. N. Nenchev, Y.Tsumaki, and M. Uchiyama, SingularityConsistent
Parameterization of RobotMotionand Control, Intl J.ofRobotics Research,
Vol.19,No.2,pp159182,2000.
5. KS.Chang,andO.Khatib,ManipulatorControlatKinematicsSingularities:A
DynamicallyConsistentStrategy,Proc.IEEE/RSJInt.ConferenceonIntelligent
RobotsandSystems,Pittsburgh,August1995,vol.3,pp.8488.
6. D. Oetomo, M. H. Ang, and S. Y. Lim, Singularity handling on PUMA in
operational space formulation, in Lecture Notes in Control and Information
Sciences, Eds. Daniela Rus and Sanjiv Singh, Eds., vol. 271, pp. 491501.
SpringerVerlag,2001.

418