The International Journal of

Robotics Research http://ijr.sagepub.com/

Friction, Stability and the Design of Robotic Fingers

Mark R. Cutkosky and Paul K. Wright
The International Journal of Robotics Research 1986 5: 20
DOI: 10.1177/027836498600500402

The online version of this article can be found at:

http://ijr.sagepub.com/content/5/4/20

Published by:

http://www.sagepublications.com

On behalf of:

Multimedia Archives

Additional services and information for The International Journal of Robotics Research can be found at:

Email Alerts: http://ijr.sagepub.com/cgi/alerts

Subscriptions: http://ijr.sagepub.com/subscriptions

Reprints: http://www.sagepub.com/journalsReprints.nav

Permissions: http://www.sagepub.com/journalsPermissions.nav

Citations: http://ijr.sagepub.com/content/5/4/20.refs.html

>> Version of Record - Dec 1, 1986

What is This?

Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014

Mark R. Cutkosky
Mechanical Engineering Department
Stanford University
Stanford, California 94305
Paul K. Wright
Mechanical Engineering Department and
The Robotics Institute
Carnegie-Mellon University
Pittsburgh, Pennsylvania 15213
Abstract
For practical reasons, compliant materials are often used on
the gripping surfaces of robotic hands. Such materials are not
well described by the Coulomb friction law or by simple
point-contact or line-contact kinematics. In this paper, a
shearing model is used to describe the contact friction.
Models ofpointed, curved, flat, soft, and soft-curved fingertips
are then developed and compared in terms of their contribu-
tion to the stiffness and stability of a simple grasp. There is a
spectrum of contact conditions defined by the fingertip radius
and contact area relative to the object size. This spectrum
provides insights for designing and controlling robotic fingers.
1. Introduction
The broad goals of this research are the design and
analysis of robotic hands for industrial assembly and
metal processing environments. Examples of grippers
and active wrists for robots in factory automation
have been described in earlier publications (Cutkosky
and Kurokawa 1983; Cutkosky, Jourdain, and Wright
1984; Wright and Cutkosky 1985). Some considera-
tions in designing such grippers and wrists are grip
strength, the number of degrees of freedom, &dquo;finger&dquo;
construction materials, finger actuation, and finger
control. Inevitably, the final design involves many
compromises. As grippers and wrists grow more com-
plex, it becomes increasingly important to understand
the interaction among different elements of the design.
This paper presents several models of fingertip/object
contact conditions and evaluates their influence on
the overall strength and controllability of the grip.
20
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
Friction, Stability and
the Design of Robotic
Fingers
2. Previous Investigations
One starting point for gripper analysis is to consider
the object to be grasped and to determine a stable
static grip. An analysis of the stability and strength of
a multifingered hand grasping a rigid object is a tracta-
ble problem for single-point loading of the fingers (see
Fig. 1 ). Hanafusa and Asada (1982) present a two-
dimensional analysis in which a potential function,
based on the shape of the object, is used to determine
stable positions for the placement of pointed, friction-
less fingers. Okada (1979; 1982) derives the equations <
of motion for a sphere and a rectangular box manipu-
lated by three fingers with hemispherical tips. Salis-
bury and Craig (1982) develop a more general 3-D
analysis in which a Jacobian relates forces and veloci-
ties of the fingers to an equivalent force and velocity
of the object. Salisbury’s and Craig’s analysis implicitly
assumes that the frictional forces do not exceed the
limits imposed by a Coulomb friction law. Mason
( 1982) acknowledges that friction forces are not con-
centrated at a point but are often distributed uniformly
over an area of contact. He then uses the Coulomb
friction law to determine how, for example, a block
may rotate when pushed along a flat surface. Such
analyses can be extended to include fingers with arbi-
trary numbers of joints and to incorporate simple
models of curved and deformable fingertips where
large coefficients of friction may be present (Cutkosky
1985).
From an analytical viewpoint, the single-point con-
tact and the Coulomb friction model are attractive
because the boundary conditions between the fingers
and object become unambiguous. Undue focus on the
point-contact grasping problem, however, will direct
attention away from the potentially more useful results
Fig. 1.A two-dimensional
object held by three fingers.
obtained with large contact areas and deformable grip-
ping surfaces. The advantages of using curved finger-
tips with soft materials and large coefficients of friction
are manifold. With a high coefficient of friction, the
hand can grip gently. A light grip helps to prevent
damage to the object and fingers and keeps the finger
sensors working over a low range of forces (with a
correspondingly high sensitivity). As discussed in Sec-
tion 6, there are also significant indirect advantages
that result from lower forces and higher degrees of
stability. Thus, there is a strong motivation to develop
friction models that account for soft materials and
kinematic models that account for curved and de-
formable surfaces in grasping.
In Section 3, the frictional properties of hard and
soft materials are compared. With soft materials, the
Coulomb friction law must be replaced with a shearing
model. Section 4 explores the effects of fingertip rolling
and deformation using this shearing model. The results
are not only quantitatively different from point con-
tact, but they also produce qualitative differences in
the overall behavior of the grasp. The differences are
further illustrated with some examples in which two
fingers grasp a rectangular block. In general, rolling
and deformation increase the stability of the grip-
especially for light gripping forces. It also becomes
clear that the various simplified models in Section 4
represent extreme cases, among which a spectrum of
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
realisticfingertip materials can be found. Finally, the
new fingertip models and the analysis suggest some
guidelines for designing and controlling robotic hands.
3. Modeling Friction
The following discussion compares the contact of two
hard materials, such as steel and brass, with the con-
tact of an elastomer on a hard surface. Real surfaces
are not flat and consist of micro-irregularities: one of
Bowden and Tabor’s metaphors (1956) is that the
mating of two surfaces is akin to taking Austria and
placing it upside down on Switzerland. When a normal
force, f&dquo;, pushes two metallic surfaces together, the
peaks are deformed (as in Fig. 2) to form junctions of
bonded material. The real area of contact, An is the
integrated area of all these junctions and is consider-
ably smaller than the apparent area of contact, Aa.
The maximum shear force that the contact can sustain
is proportional to the real area of contact, or
where k is the shear strength of the parent material. A
plasticity analysis (Bowden and Tabor 1950) shows
that A, is proportional to the normal load. As a result,
the maximum shear force is also directly proportional
to the normal load, and the ratio can be described by
a constant coefficient, p.
We next consider a rubber fingertip pressed against
ahard surface. The basic geometry in Fig. 2 still ap-
plies, but a plasticity analysis is not applicable since the
rubber asperities remain elastic for all but very large
strains. If the rubber asperities are approximated as
small hemispheres, a Hertzian analysis (Schallamach
1952) predicts that A, does not increase linearly with
the normal load, but asf2ll. In practice, A, may in-
crease slightly faster (more nearly as fn) because new
asperities are brought into contact with the steel when
the original ones deform. The rubber is sheared when
the surfaces slide past each other (dragging a pencil
eraser, even lightly, against any hard surface leaves a
film of rubber that is evidence of shearing) and, as
with metallic contacts, the coefficients of friction is a
function of the material shear strength and the real
area of contact. Since A, increases less rapidly than the
21
Fig. 2. Deformation and
friction of mating surfaces.
normal load, the coefficient of friction decreases as the
load increases. This phenomenon has been observed
experimentally by the authors and others (Schallamach
1952; Ludema 1971 ) for several rubbers, polyure-
thanes, and other elastomers. Since the coefficient of
friction is not a constant, it becomes more convenient
to speak of a limiting shear stress, ’[slip, over the con- contact, there is no rolling and consequently no move-
tact such that
where k is the shearing strength of the fingertip mate-
rial. This shearing model of friction is used in the
following discussion of rolling and deforming fingertips. basketball or a cardboard box, the point contact as-
4. Fingertip Contact Models
As seen in the last section, friction between the gripper
and the object depends on material properties and
surface deformation. The effects of the fingertip geom-
etry are now investigated for some simple cases shown
schematically in Fig. 3. The Coulomb friction law
may be used for hard pointed or curved fingertips, but
the shearing model is preferable for soft fingertips.
22
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
4.1. POINT CONTACT
point contact with friction, forces are transmitted
In a
fingertip and the object, but torques are
between the
not. Similarly, translation of the fingertip is coupled
with that of the object, but rotation is not. With point
ment of the contact area upon the object or the fingertip.
Point contact is the simplest case to analyze, but as
the following models reveal, it becomes inaccurate
when the fingertips deform or when the radius of cur-
vature of the fingertips is not small compared to the
size of the object. For example, when a person holds a
sumption may be accurate, but when a person holds a
racquetball or a matchbook it is not.
4.2. CURVED FINGER CONTACT
A hard, curved fingertip is similar to a pointed finger-
tip in that the contact area is small enough that forces
are transmitted, but torques are not. The main differ-
ence between hard, curved and pointed fingertips
arises from the fingertip rolling upon the surface of the
Fig. 3. Examples of fingertip Fig. 4. Rolling contact.
geometry.

After the fingertip rolls a small amount, the new con-

tact point will be at the location r&dquo; on the body and
the new tangent will have the direction
object. As the finger rolls, the location of the contact
point shifts. This, in turn, changes the way in which
the grasp forces act upon the object.
For the present discussion, we limit ourselves to the
two-dimensional case in Fig. 4. The fingertip and the The contact point on the fingertip will be at the loca-
object profiles are described parametrically as rf) and tion r’with respect to the finger coordinate system and
rb(S), where s is the arc length along either curve. The the direction of the tangent will be
conditions for pure rolling are:
1. There is a common tangent, u, at the points of
contact.
2. The instantaneous contact points on the fin-
For pure rolling, 6sr = (5sb, where t5s IA Ar for
=

gertip and the object must have the same

translational velocity. small motions. Thus, for a small rolling motion, the
3. The arc length.5s, traversed along rb(s) and
contact point translates ~rb upon the body and rotates

rf(s) must be equal when the fingertip rolls on through the angle between Ub and u’. At the same
the object. time, the fingertip must translate by drb - Art (the
distance between bp’ and fp’ ) and rotate through the
At the contact point, the tangent is the same for both angle between ufand u~. The translations and rota-
curves so that tions are functions of rf(s), rb(s), and,5s.
Expressing Ar and u’ as Taylor’s series expansions
in r and 6s and eliminating second-order terms, the
above vector quantities become (Cutkosky 1985):

23
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
 Fig. 5. Cross section of a Fig. 6. Cross section of a
large-radius hemispherical small-radius hemispherical
fingertip on a flat object fingertip on a flat object
surface. surface.

1. Translation of contact point with respect to The motion reduces to a rotation of the (a, b) coordi-
object: nate system about the contact point in the (x, y) sys-
tem. In other words, as the radius of curvature be-
comes smaller, the model reduces to the case of a

2. Rotation of contact point with respect to object:
3. Rotation of fingertip with respect to
object:
pointed finger rotating about its tip.
In Fig. 5, drb and ~~rf are significant. To a first ap-
proximation, the finger still rotates about the contact
point, but the point of application of the grasp force
has moved by Arb on the object surface.
,

4.2. 1.
Effects of Rolling Motion
The meaning of the above equations becomes apparent
in Figs. 5 and 6, which show fingers with curved tips
of constant radius rolling on a flat surface. In both
cases, uf = ub = ( 1 ) i ~- (0)j. Since the object is flat, the
second derivative of rb is zero. The finger undergoes
virtually the same motion in Figs. 5 and 6, but there is
a significant difference in
Arb and Arf between the two
cases, stemming from the difference in 5s. In Fig. 6,
there is no appreciable change from rb to rb . There is
also virtually no difference between r and rf, when
expressed with respect to the (a, b) coordinate frame.
4.2
Flat-tipped Finger
A flat-tipped finger can be seen as a limiting case in
which the radius of curvature becomes infinite so that
Arb and Arb become infinite and produce an infinite
displacement of the contact area for any rotation of the
finger with respect to the object. In practice, of course,
the contact point will tilt onto the edge of the flat fin-
gertip, at which point the radius of curvature becomes
zero.
4.3. VERY SOFT FINGER
The fourth fingertip example shown in Fig. 3 repre-
sents the extreme case of a compliant fingertip pressing

24
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
Fig. 7. Elastic fingertip in
contact with object surface
( perspecti ve).
against the object surface. In this model, the fingertip
conforms to the object surface and adheres slightly.
Such characteristics are found in many natural grip-
ping surfaces, including the fingertips of the human
hand. The coefficient of friction for such a fingertip
may be greater than one. As discussed in Section 1, a
shearing model of friction is more appropriate for
such materials.
The soft finger model is further specialized with the
assumption that no rolling occurs and that the compli-
ant medium at the fingertip is elastic. With these as-
sumptions, the fingertip becomes a less accurate model
of human fingertips. Human skin is viscoelastic, and
after being deformed will not immediately return to its
original position. Depending on the curvature of the
object being held and the degree of adhesion present,
the human fingertip will also roll slightly upon the
object, exhibiting a rolling resistance of the kind dis-
cussed in Section 4.4. Nonetheless, the elastic soft-
finger model is useful to demonstrate a limiting case
in which there is kinematic coupling between the fin-
gertip and the object in all six degrees of freedom.
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
Figure 7 shows the fingertip in contact with the sur-
face of an object. The fingertip material is assumed to
be much softer than the object and the finger sub-
strate, which are treated as rigid bodies. A finger coor-
dinate system (a, b, c) is located at the interface be-
tween the fingertip material and the hard finger
substrate. A second coordinate system, (l, m, rt), is
attached to the object.
The grasping forces at the object surface can be
expressed as integrals of the stresses over the contact
area:
The soft fingertip represents a compliant coupling
in which small motions of the finger with respect to the
object are possible in any direction. An elastic model
of the system permits the deflection/force relationships
to be expressed jointly as the stiffness of the contact.
The fingertip can be treated as a short elastic
member clamped between two rigid boundaries. Ob-
taining the exact stress field for such a problem is a
formidable task-even assuming that the material is
perfectly elastic and isotropic. Numerical results could
be obtained using a finite element analysis, but the
analysis would be time-consuming and would have to
be recomputed for different cross sections and mate-
rials. Approximate analytical solutions exist for related
problems, such as a rubber block bonded to two rigid
plates (Gent and Meinecke 1970), but the present
analysis can be simplified by observing that the stresses
at any given location within the material are of little
interest, provided that
estimates of the integral quantities can be computed
at the object surface,
the combined stress- field nowhere exceeds the
strength of the material, and
the normal stress, ann, never becomes sufficiently
tensile to cause the fingertip material to separate
from the object surface.
25
 Fig. 8. Elastic fingertip in
contact with object surface
(section).

The last requirement can be satisfied by assuming a

large grasping force normal to the object surface and/
or some adhesion between the fingertip and the object.
If one edge of the fingertip does start to separate from
the object when the finger rotates slightly, the finger is
starting to roll.
Since an accurate elastic solution is complicated
(and would in any event be an approximation to the
viscoelastic behavior of compliant polymers and skin-
like materials), a first-order model of the fingertip is
used. The behavior of the fingertip in shear, torsion,
compression, and bending is discussed below, and the
separate solutions are superposed to produce a stiff-
ness matrix for the contact.

4.3.1. Bending Stiffness and Resistance to Rolling

The bending model for the fingertip is similar to that
used in classical beam theory. A rotation about the
a-axis by the finger produces a rotation in the material
of 601t per unit thickness. The bending strain and
stress at a distance m above the centerline are
where Imm is the moment of inertia of the cross section
about the m-axis and V is the volume of the material.
Similarly, the bending stiffness for rotations about
the m-axis is found as
and

As mentioned earlier, the maximum bending mo-

where E is the modulus of elasticity. As in beam
theory, it is assumed that plane sections remain plane
and wi~ (o 0. It is also assumed that since the
= =
stresses rl,, (Jmm, and ull are zero at the outer surfaces,
they are approximately zero throughout. This assump-
tion is somewhat less supportable than in beam theory
since the elastic element cannot be considered slender.
It is not actually necessary that ~t,m, amm, and (J1l be
zero everywhere, but only that their resultant does not
greatly affect the bending rigidity of the element. The
bending rigidity may then be found by equating the
energy stored in rotating the finger with the energy
stored in deforming the material.
ment that the contact can sustain is limited by the
adhesion between the fingertip and the object surface.
The limitation is easily demonstrated for the example
of a square contact area of length w on each side. Re-
ferring to Fig. 8, a normal force of magnitude In pro-
duces a uniform contribution to the normal stress of
A bending moment of magnitude fol produces a contri-
bution to the normal stress that is maximum at the
edges of the contact. I

26 Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014

The combined normal stress will become tensile at
one edge when
Thus, unless the adhesion between the fingertip and
the object is able to resist tensile loads, the finger will
start to roll whenever the bending moment is more
than one-sixth the normal load times the length of the
side. For small contact areas, the fingertip is likely to
start rolling unless considerable adhesion is present.
4.3.2. Shear Stiffness and Resistance to Slipping
For a beam with an end load, the variation in the
moment over the length of the beam is balanced by a
distribution in shearing stress over the cross section of
the beam (Crandall, Dahl, and Lardner 1972). For the
elastic fingertip, however, it is assumed that the varia-
tion in the moment produced by a shear force in the
(a, b, c) system is negligible compared to the effect of
rotating the finger. Consequently, the bending mo-
ment is approximately constant over t and the shear
stress is assumed to be uniform over the cross section.
The shear stiffness is found by equating the energy
required to displace the finger in shear with the energy
stored internally in the material.
or
where G is the shear modulus of the material and A is
the cross section area, wh.*
Using the definition of ’slip from Eq. (2), the maxi-
mum shear force that the contact can sustain is just
A ’slip .
*
Thus, A corresponds to the apparent area, ~4,., of Section
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
4.3.3. Compressive S’tiffness
Displacement of the fingertip toward the object results
in a uniform compressive strain, Em!’ over the cross
section. The compressive stiffness is found in the same
way as the shear stiffness, with G replaced by E.
. 4.3.4. Torsional Stiffness and Resistance toSlipping
The torsional rigidity of the fingertip is most easily
found for the case in which the fingertip is a short
cylindrical element. The rigidity is
where ro is the radius of the cylinder and Ip is the polar 1B
I
moment of inertia (Crandall, Dahl, and Lardner 1972).
For noncircular cross sections, the expression becomes
more complicated due to warping of the cross sec-
tions, although for the present case the warping may
be negligible since t is small and since the material is
constrained by a rigid boundary at each end.
For a round bar, the shear stress in torsion is
Thus, if the fingertip were a cylindrical pad, slipping
would begin at the periphery when
where Tslip is given above for shear loading. Once slip-
ping has occurred at the periphery, the fingertip will
not return to exactly the same orientation when the
torque is removed. As the torque is increased, the
region of slipping spreads from the periphery toward
the center. The phenomenon resembles the yielding of
1. an elastic, perfectly plastic bar in torsion. At any stage,
27
the moment balance is given by Table 1. Soft Fingertip Deflections for 4.0 N Load and
100 mm2 or 400 mm2 Contact Area

Integrating the above equation and expressing ’cl,

and rsrrp in terms of r and the angle of rotation of the
finger, don gives

Thus, the torque required for complete slipping is ~/3

the torque required to initiate slipping at the periphery,
although this would theoretically only be reached for
an infinite rotation of the fingertip. For a square or

rectangular contact area, the qualitative behavior is the

same, with slipping initiating at the periphery and
spreading inwards. However, the expression for Jon be-
comes more complex due to the more involved ex-

pression for! 1m.

4.3.5. Effects of Deforming Fingertips

The comparative importance of the above quantities
can be determined for a fingertip of given proportions.
Table 1 shows the results for two fingertips. For the
first, w h = 10.0 mm and t = 5.0 mm; for the sec-
=
ond, w ~t 20.0 mm and t = 5.0 mm. The modulus
= =
of elasticity, E, is assumed to be 2.5 MPa and Pois-
son’s ratio is 1/2, so that G E/3. These are typical
=
values for soft rubber. A force of 4.0 N (a little less
than one lbf) is used to produce deflections for com-
parison.
For the smaller area, the rotational stiffness terms
are much lower that the translational terms, and the
fingertip is clearly less constrained with respect to
rotations than translations. However, the bending and
torsional stiffnesses increase as the square of the con-
tact area, while the shear and compressive stiffness
increase linearly with the contact area. Thus, for the
larger contact patch, the rotational and translational
stiffnesses become comparable. lf w and h were dou-
bled again, bending and torsional deflections would
become small in comparison to shear deflections. This
result matches what one would expect intuitively.
If the grasping force is varied proportionately with
the contact area, then, as the contact area becomes
small, the fingertip begins to behave like a point con-
tact in which significant rotations are possible but
translations are not. As the contact area becomes large,
rotations are negligible compared to shear deflections.
If the grasping force is held constant for different con-
tact areas, then the fingertip becomes much less com-
pliant as the area increases, and rotations become
negligible more quickly than translations.
For the forces given in Table 1, unless some adhe-
sion exists between the fingertip and the object, the
bending moment will cause the fingertip to roll for
both the 100 mm2 and the 400 mm2 area. From Eq.
(8), the largest bending moment that the contact could
sustain without tensile stresses occurring is 0.007 Nm
for the 100 mm2 case and 0.013 Nm for the 400 mm2
case. In torsion, depending on the shear strength of
the interface, the contact will probably slip for the
100 mm2 area but might not for the 400 MM2 area.
From Eqs. ( 13) and (15), if the shear strength is
roughly equal to 4.0 X 104 N/m2 in the first case (cor-
responding to a coefficient of friction of 1.0) and 1.5 X
104 N/m2 in the second (corresponding to a coefficient
of friction of 1.5), the maximum torques that can be
exerted are 0.02 Nm and 0.05 Nm, respectively. This

28
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
 Fig. 9. Pressure distributions
for elastic, soft, and very soft
fingertips.

supports the idea that a soft finger with a small con-

tact area can exert torques about an axis perpendicular
to the contact surface more readily than it can exert
torques in the plane of the surface. For a soft, curved
fingertip, as discussed below, the difference is more
pronounced.
Once the fingertip stiffness matrix has been com-
puted, it may be inverted and added to the finger
compliance to form a net compliance matrix for the
contact.

4.4. SOFT, CURVED FINGERTIP

The hard, curved fingertip and the very soft fingertip
represent extremes between which real, deformable
fingertips may be expected to lie. Human fingers and
rounded robot fingers with rubber surfaces exhibit
both rolling and substantial deformation. The analysis
of such fingertips becomes quite involved, combining
the rolling calculations in Section 4.2 with the defor-
mation calculations in Section 4.3. A complete model
is not attempted in the following discussion, but the would behave in the same manner as the hard, curved
properties of soft, rolling fingers are bracketed by the finger discussed earlier, the only difference being that
models developed in the last two sections. rf would vary due to flattening of the fingertip under
A number of insights can be gained by considering the load. If the degree of flattening could be predicted
the analyses applied to rolling rubber tires and metal as a function of fingertip loading, then the methods

cylinders or spheres. For a hard, elastic sphere rolling
on an elastic surface, the pressure distribution is de-
scribed by the Hertzian model, which predicts a hemi-
spherical pressure distribution (Timoshenko and
Goodier 1970). For the much larger deformations that
occur when a soft, curved finger presses against an
object, the distribution is expected to be only qualita-
tively similar. The pressure will be maximum at the
center of the contact, diminishing smoothly to zero at
the periphery. For progressively softer fingertips, the
pressure distribution becomes more uniform, espe-
cially toward the center of the contact area. In the lim-
iting case, the pressure is essentially uniform through-
out, as assumed in the very soft finger model
described in Section 4.3. The pressure distributions are
shown for elastic, soft, and very soft fingertips in Fig. 9.
For a perfectly elastic curved finger, it is impossible
to transmit moments in the plane of the contact since
the finger rolls easily upon the object. Thus, in the
absence of rolling resistance, the soft, curved finger
discussed in Section 4.2 could be used to predict the
motion of the finger and the contact point. Elastic
flattening formulas have been developed for cylinders
and spheres, but these are unlikely to give accurate
results for a soft fingertip.
In practice, there is a resistance to rolling. At low
speeds, the rolling resistance is due largely to hysteresis
losses and microslip at the contact area. Rolling resist-
ance is discussed at length in the literature on wheels
and tires (Ludema 1971; Halling 1975; Wong 1978).
For an elastic sphere or cylinder rolling upon a plane,
the deformation of the material results in a loss of
hysteresis that can be used to derive a &dquo;coefficient of
rolling resistance&dquo; (Halling 1975). Microslip results
from the elastic strain of the fingertip material as it is
pressed against the surface. If the fingertip is loaded
with a normal load, f,, the material ahead of the cen-
terline of the contact will spread forwards slightly and
the material behind the centerline will spread back-
wards slightly. This microslip produces rolling losses

Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014

29
 Fig. 10. Maximum shear
stress for moment about
finger axis.

and &dquo;creep,&dquo; so that soft, curved fingertips do not

rotate as freely with respect to the object as pointed or
hard, curved fingertips.
The static resistance to slipping of a soft, curved
fingertip will be similar to that of the very soft finger
discussed in Section 4.3, except that since the pressure
distribution is not uniform over the contact area, the
value of the stress at which slipping occurs also varies
over the contact. As in Section 4.3, the interface shear

strength ’slip may be expressed as a fraction of the

material shear strength, where the fraction, ~3, is a
function of factors including the normal pressure and
the surface roughness. Since the pressure is least at the
edges of the contact, slipping will initiate there. For
loads tangent to the contact, the shear stress may be
uniform inside the region where there is no sliding,
but will have an upper limit of ’slip outside the region.
For a moment about the axis normal to the contact,
the shear stress inside the sticking region will have the
same distribution as for the very soft finger, with the

magnitude given as in Eq. (12). In the slipping region,

the shear stress will again be equal to the upper limit
of A cross section of shear stress distribution is
shown in the lower part of Fig. 10. The distribution for
the very soft fingertip of Section 4.3 is shown in the
upper part for comparison. The maximum torque
about the axis of the finger is equal to the polar mo-
ment of the shear stress shown in Fig. 10:
where L¡m(r) is proportional to r and rli,(r) is a func-
tion of (Jnyr)·
Thus, unlike the hard, curved finger or the pointed
finger, the soft, curved finger is able to exert small
torques about its own axis.
5. Examples
The concepts of the last section can be demonstrated
with some short examples that explore the differences
in grip behavior produced by different fingertips. Fig-
ure 11 shows a rectangular block held by three sets of
fingers: pointed, curved, and soft. In all three cases,
the fingers are assumed to have three degrees of free-
dom, being restricted to motions within the plane of
the paper. For simplicity, it is assumed that the finger
joints correspond to translations, a and b, and a rota-
tion, a. The grasping force is [hI, 0, 0] for the left
finger and [&horbar;/~ 0, 0]’ for the right finger.
The sizes and orientations of the object and fingers
and the finger stiffnesses are identical in each case.
The finger stiffnesses are due partly to the structural
stiffness of the finger links and partly to the servo
stiffness at the joints. For the following examples, it is
assumed that the fingers are described by a 3 X 3 diag-
onal stiffness matrix Kq in (cz, b, c) coordinates.
In each case, a change [dfa, dfb, dir]1, in the grasping
force is calculated for small displacements of the ob-
ject. The force/deflection characteristics can then be
compared for each of the fingertips, which in turn
permits us to compare the stiffness and stability of
each grasp. The results show that the point-contact
model may produce results that are qualitatively, as

30
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
Fig. 11. Three examples of
holding a rectangle between
two fingers.
well as quantitatively, in error. For example, a grip
that may appear to be unstable with pointed fingertips
may be stable for both rounded and soft fingertips.
5.1. POINTED FINGERS
As the object is moved an arbitrary small amount
[dx, dy, dy]1, the motions of the contact points on the
object are given by [dm, , dn I, d,61]1 and
[dm2’ dn2, dP2F. However, only the translational mo-
tions (dm, dn), are transmitted to the fingers. The next
step is to determine the motions of each finger given
dm and dn. Motion in the dm direction can be accom-
modated either by a movement of the finger in the b
direction or by a rotation a. In practice, both will
occur, and the contribution from each will be balanced
to minimize the potential energy of the finger. The
result for the left finger is shown in Table A-1 in the
Appendix.
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
The change in the grasping force depends both on
the restoring forces at the finger joints ([dJ;&dquo; dfb, dk]t =
[Kq][da, db, da] t ) and on the relative motion between
the finger and the rectangle. The relative motion
changes the way in which the grasping force acts upon
the rectangle. The results for the left finger are sum-
marized in Table A-2.
The contribution from the right finger follows from
symmetry. When the two fingers are combined, the
change in the grasping force upon the object is given
by Table A-3.
Whenever any of the forces in Table A-3 become
positive, the grasp will be unstable for infinitesimal
displacements in the corresponding direction. Thus, if
the rotational stiffness of the fingers is small
(ka < farf), the change in the grasp force for a motion
in the y direction will be positive, amplifying the mo-
tion. This result matches one’s intuition that a rectan-
gle squeezed between two fingers will be unstable if
the fingers pivot freely, without springs.
Similarly, the last term in Table A-3 shows that the
rectangle will be unstable with respect to rotations
about the z-axis if
where
This result is less intuitively clear but it becomes ap-
parent if k,, is very large, in which case the fingers do
not rotate. For this case, C ~ kb/ka and the force/de-
flection relationship becomes
Thus, as the length of the rectangle and the stiffness of
the fingers increase, the grasp becomes more stable.
On the other hand, as the grasping force, fa, increases,
the infinitesimal stability decreases. One can verify
these results by holding a pencil lengthwise (horizon-
tally) between the index fingers of each hand. Applying
more force makes the grip less stable and increasing
31
the stiffness of one’s arms and fingers makes the grip
more stable. Finally, the grip is less stable with a short
pencil than a long one. Evidently, human fingertips
pressed against the ends of a pencil can be approxi-
mated by point contacts. However, repeating the above
experiment with a small box, such as a matchbox,
gives results that do not agree with the point-contact
model. In fact, a small object may be stable no matter
how hard one presses. For a small object, the curva-
ture and deformation of the fingers must be consid-
ered, as in the following examples.
5.2. CURVED FINGERS
Many of the results from the point-contact example
also apply for fingers with curved surfaces. The differ-
ence is that the contact is no longer fixed with respect
to the object because the finger rolls. Rolling changes
the way in which the grasp force acts upon the rectal-
gle. The changes in the forces from the left finger are
summarized in Table A-4, and the contribution from
both fingers is given in Table A-5. In this example,
since the center of curvature of the finger is also the
origin of the (a, b) system, the rolling translation of the tions may be set equal, so that ka kb.
contact point exactly cancels the rotational cross prod-
uct (dy - C rfdm} X rf. As Table A-5 shows, the
change in fy is exactly the same as it is for pointed
fingertips but the torque, df., is different.
As in the previous example, the expression for
torque about the z-axis simplifies for the limiting case
in which the rotational stiffness of the fingers is large
(ka » kb). The change in the torque about the Z-axis point contact case.
reduces to
w
In Table A-5, the expression for ~, reduces the change in the grasp force is given in Table A-7.
if rf = 2 ,
to - 1/2 C k,, W2 dy. The physical interpretation is that
the translation of the contact point due to rolling of the
finger with respect to the object exactly cancels the
effect of rotating the object. Thus, for large radii of
curvature
~ &h2orbar; t,
rf ~ the grasp is infinitesimally stable
with respect to rotations regardless of the stiffness of
the fingers. This explains why a small box grasped
32
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
between two human fingers (where the radii of curva-
ture of the fingers are large compared to the dimen-
sions of the box) is more stable than the pencil dis-
cussed in the previous example.
5.3. VERY SOFT FINGERS
For contacts with soft fingers, a fingertip stiffness ma-
trix is established from the rigidity terms in Section
4.3. The stiffness matrix can be inverted to form a
compliance matrix that may be combined with the
finger compliance matrix, Kq-1. Since the shear mod-
ulus, G, of rubberlike materials is about one-third the
compression modulus, E, the shear stiffness can be
written using Eqs. (9) and (10) as 1/3 kp. From Eqs. (6),
(7), and (11), the bending and torsional stiffnesses are
approximately B kp and T3 B kp, respectively, where B
is equal to one-twelfth the contact area.
The general expression for the change in the grasp
force with soft fingers is lengthy, but it is simplified
considerably for the limiting cases in which the contact
area is very small or very large. To further simplify the
algebra, the finger joint stiffnesses in the a and b direc-
=
For a small contact area, B - 0 and the bending
and torsional stiffnesses become negligible in compari-
son to the shear and compressive stiffnesses. For two
fingers, the final results are given in Table A-6. If it is
further assumed that kp » kb, as is usually the case, it
can be shown that the results for the change in grasp-
ing force become identical to those obtained in the
For the case when the contact area is large, the
bending and torsional stiffnesses become infinite. If it
is again assumed that kp » kb, the problem reduces to
that of a finger glued to the surface of the object, and
5.4. SUMMARY
In the first was held between two
example, a rectangle
hard, pointed fingertips. When the rectangle was ro-
tated slightly, the finger stiffnesses produced restoring
forces that tended to stabilize the grip, but a change in
geometry (resulting from rotation of the object with
respect to thefingers) made the grip less stable. The
overall stability of the grip was a function of the finger
stiffness, the length of the rectangle, and the magni-
tude of the initial grasping forces. Interestingly, the
grip became less stable as the gripping forces were in-
creased. Thus, while an increase in the gripping forces
may make the contacts more resistant to slipping, it
does not always make the grip more secure.
The second example used curved fingertips. With
curved fingers, the stability of the grasp increased over
the pointed-finger case due to rolling of the fingertips.
If the fingertip radii were larger than one-half the
length of the rectangle, the grip became stable with
respect to rotational displacements no matter how
small the finger stiffnesses were. The relationship be-
tween the fingertip radii and the length of the rectangle
raises an important point; the definitions of &dquo;large&dquo; or
&dquo;small&dquo; radii of curvature and contact areas depend
on the size of the object being handled. This is why
the point-contact model is reasonable when we hold a
basketball or a large box, but not a coin or a matchbox.
The third example involved soft fingers. The result-
ing expression for the force/deflection relationship was
lengthy but reduced to the point contact case when
the fingertip areas became small and reduced to the
flat-tipped case when the contact areas became large.
6. Conclusion
There are numerous advantages to using compliant
materials for the fingers of a robot gripper. This is
perhaps obvious based on the experience of the human
hand. Soft rubber gripping surfaces are commercially
available (Barry Wright Corp. 1983). The direct ad-
vantages include:
1.Compliant materials distribute contact forces,
avoiding damage to the gripped object.
2. Compliant materials have large coefficients of
friction, making it possible to use lower grip-
ping forces.
3. Compliant gripping surfaces have more kine-
matic coupling at each contact with an object
so that fewer contacts are required to com-
pletely restrain an object. For example, two soft
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
fingers can always restrain an object in space
but two hard
fingers cannot.
There are also significant indirect advantages. With
lower gripping forces, it is possible to use smaller ac-
tuators with faster response. In fact, the switch from
hard to soft fingertips may permit some of the external
actuators (that are located away from the hand and
drive the fingers through cables) to be replaced with
internal actuators (that drive the finger joints directly).
With lower gripping forces, it is also possible to mea-
sureforces and pressures in the hand with a finer reso-
lution and to achieve more delicate control of the
forces required for a manipulation task. The examples
in Section 5 show that in some cases a light gripping
force reduces the likelihood that the grip will become
unstable when subjected to task-induced disturbances.
With so many practical advantages to recommend
compliant fingertip materials, there is a strong motiva-
tion to account for them in an analysis of grasping.
However, the properties that make such materials
desirable for grippers also make them difficult to model:
l. With elastomers, the real area of contact gener-
ally increases less rapidly than the contact
pressure, with the result that the coefficient of
friction is highest for low gripping forces.
Thus, unlike hard materials for which the Cou-
lomb law of friction can be used to describe a
constant ratio between the gripping force and
maximum tangential force at the contact,
compliant materials are best described by a
shearing model for which there is a maximum
stress, T slip.
2. With soft gripping surfaces, fingertip rolling
and deformation may have a significant effect
on the kinematics of the grasp.
To evaluate the importance of rolling, deformation,
and adhesion, several fingertip models are considered
in Section 4. The idealized pointed, curved, flat, and
soft fingertips represent limiting cases of the behavior
of practical fingertips. The main features of the con-
tact models are summarized in Figure 12. A contin-
uum of contact behavior becomes apparent when we
look at large variations in the contact area and the
fingertip curvature. At extreme values of A and r, the
rolling and soft contacts reduce to simpler point or
33
Fig. 12. Summary of finger- Fig. 13. Relations among
tip and contact models. fingertip models.

contacts. The increased stability of curved

planar contact models: fingertips was due to rolling. The increased sta-
&dquo;

bility of soft fingertips was due to more kine-

rf ~ 0 --~- point contact with friction
rf --:> 00plantar contact with friction
A ~ 0 -~= point contact with friction
A --:> 00 = planar contact with friction
Figure 13 illustrates the overlapping regions in which
the different contact models apply.
In Section 5, the effects of different fingertips on the
force/deflection behavior of a grasp were compared for
pointed, curved, and soft fingertips with the following
results:
1. Finger stiffnesses tended to stabilize the grip,
but relative motion between the object and
fingers produced destabilizing forces.
2. Increasing the gripping force reduced the
chance of slipping but also made the grip less
stable with respect to disturbances.
3. With curved and soft fingertips the stability of
a two-fingered grip was greater than for point
matic coupling between the object and the
fingers.
4. The importance of rolling and deformation
depended on the relative sizes of the radius of
curvature and the contact area, compared to
the characteristic length of the object.
For control purposes, the small-motion, force/de-
flection behavior of a grip amounts to a linearized
description of the &dquo;plant,&dquo; giving a relationship be-
tween displacements of the object and the resulting
changes in force. The results show that point-contact
finger models and the Coulomb friction law are not
always adequate. The contributions of adhesion, roll-
ing, and deformation depend on the relative dimen-
sions of the fingertips and the object. For a single grip-
per, any of the models of Section 4 may apply,
depending on the size, shape, and hardness of the
grasped object. Thus, it is necessary to know where we
are in the space of Fig. 13 at any given time. For some
applications there will be enough a priori knowledge

34
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
 about the grasped object and the task to determine ann
=
normal component of stress
which model to use, but in less structured environ- T 1M
=
shear stress
ments (such as a robot picking up unfamiliar objects E =
modulus of elasticity
on the ocean floor), it becomes useful to have fingertip G =
shear modulus
sensors that can measure normal and shear forces and v =
Poisson’s ratio
can indicate the approximate size and location of the w =
contact width
area of contact. With this information it becomes h =
contact height
possible to estimate whether a finger is close to slipping A =
contact area
in translation, rotation, or both. From an approximate t =
contact thickness
pressure distribution the robot could also distinguish ro0
=
outer radius of contact
between soft and hard contacts. Recently, a number of rsJip
=
critical radius for slipping
sensors have been developed that can determine pres-
sure and/or shear distributions at a fingertip (Purbrick
Section 3
1981; Raibert and Tanner 1982; Hackwood et al. [dx, dy, dylt =
small planar translation and rotation
1983) and some of these have been demonstrated in of object
applications involving the recognition of small objects [dm, dn, d~3] =
small planar translation and rotation
(Grimson and Lozano-P6rez 1984). However, in less of contact
structured environments such sensors may prove even small planar translation and rotation
more useful for monitoring the contact conditions
[da, dh, da J1 =

of finger
between robotic hands and objects.
I£ , £ , £1 ’ ~ finger force (If., 0, 0] initially)
[fm,h,f.8]l = contact force ([0, 0, f&dquo;] initially)
LG, .1;&dquo;.1;]1 = object force ([fx, 0, 0] initially)
Appendix Kq (diagonal) finger stiffness matrix
=

ka finger stiffness in a direction

=

Nomenclature kb finger stiffness in b direction

=

Section 1 k~ rotational finger stiffness

=

In = grasping force kp compressive stiffness of fingertip

=

C =
constant defined in Section A.1 1
j~ = shear force at contact a

coefficients of friction B = a constant (equal to 1/12 the contact area)

j
Aa a
=
apparent area of contact w = width of rectangular block
A~ r
=
real area of contact rf = radius of curvature of fingertip
! slip
=
shear stress for slipping at soft finger
contact
.
k =
shear strength of material
A.1. FORCES AND MOTIONS POUR EXAMPLES IN
SECTION 3
Section 2
u =
unit tangent at curved contact In the following tables,
If
=
(vector) radius of fingertip surface
rb ‘
=
(vector) radius of object surface
s =
arc length along rolling path

(x, y, z) =
object coordinate system
(I, m, n) =
contact coordinate system .

(a, b, c) =
finger coordinate system
(fi,fm,fn) =
contact forces
{JBl ~ ~Bm 9 fBn) = contact torques
Enn
=
normal component of strain

35
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
36
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014
Table A-6. Total Change in Force
Contact Area (Soft Fingertips)
Table A-7. Total Change in Force
Contact Area (Soft Fingertips)
on Object for Small Hanafusa, H., and Asada, H. 1982. Stable prehension by a
robot hand with elastic fingers. In Robot motion planning
and control, ed. M. Brady. Cambridge: MIT Press, pp.
323-336.
Ludema, K. C. 1971. Friction of rubber. In Mechanics of
pneumatic tires, NBS Monograph 122, ed. S. K. Clark.
Washington, D.C.: U.S. Govt. Printing Office.
Mason, M. T. 1982. Manipulator grasping and pushing
operations. Ph.D. thesis, Massachusetts Institute of Tech-
nology, Department of Computer Science.
Okada, T. 1979. Object handling system for manual in-
dustry. IEEE Trans. Sys. Man, Cyber. SMC-9(2):79-89.
on Object for Large Okada, T. 1982. Computer control of multijointed finger
system for precise handling. IEEE Trans. Sys. Man,
Cyber. SMC-12(3):289-299.
Purbrick, J. A. 1981. A force transducer employing conduc-
tive silicone rubber. 1st Int. Conf. Robot Vision and Sen-
sory Controls. IFS Conferences, Ltd., pp. 73-80.
Raibert, M. H., and Tanner, J. E. 1982. A VLSI tactile array
sensor. Proc. 12th Int. Symp. Industr. Robots. Paris: Soci-

ety of Manufacturing Engineers, pp. 417-425.

Salisbury, J. K., and Craig, J. J. 1982. Articulated hands:
REFERENCES force control and kinematic issues. Int. J. Robotics Res.
1(1):4-17.
Barry Wright Corporation. 1983. Sensoflex gripper pad. Schallamach, A. 1952. The load dependence of rubber fric-
Bulletin GP-183. tion. The Proceedings of the Physical Society.
Bowden, F. P., and Tabor, D. 1950. The friction and lubrica- 65(393B):657-661.
tion of solids. London: Oxford University Press. Timoshenko, S. P., and Goodier, J. N. 1970. Theory of
Bowden, F. P., and Tabor, D. 1956. Friction and lubrication. elasticity. New York: McGraw-Hill.
New York: Wiley. Wong, J. Y. 1978. Theory of ground vehicles. New York:
Crandall, S. H., Dahl, N. C., and Lardner, T. J. 1972. Theory Wiley.
of elasticity. New York: McGraw-Hill. Wright, P. K., and Cutkosky, M. R. 1985. Design of grippers.
Cutkosky, M. R. 1985. Robotic grasping and fine manipula- In Handbook of industrial robotics, ed. S. Nof. New York:
tion. Hingham, Mass.: Kluwer Academic Publishers. Wiley.
Cutkosky, M., Kurokawa, E. 1983 (April). Grippers for an
unmanned forging cell. CMU-RI-TR-83-3. Pittsburgh:
Carnegie-Mellon University Robotics Institute.
Cutkosky, M. R., Jourdain, J. M., and Wright, P. K. 1984.
Testing and control of a compliant wrist. 14th Int. Symp.
Industr. Robots. Gothenburg, Sweden: Society of Manu-
facturing Engineers.
Gent, A. N., and Meinecke, E. A. 1970. Compression, bend-
ing, and shear of bonded rubber blocks. Polymer Engi-
neering and Science 10(1):48-53.
Grimson, W. E. L., and Lozano-Perez, T. 1984. Model-based
recognition and localization from sparse range or tactile
data. Int. J. Robotics Res. 3(3):3-35.
Hackwood, S., et al. 1983. A torque sensitive array for robo-
tics. Int. J. Robotics Res. 2(2):46-50.
Halling, J. 1975. Principles of tribology. London:
Macmillan.

37
Downloaded from ijr.sagepub.com at Virginia Tech on August 24, 2014