New Tactile Sensing by Robotic Fingertip with Soft Skin
Kouji Murakami and Tsutomu Hasegawa
Graduate School of Information Science and Electrical Engineering
Kyushu University
6-10-1 Hakozaki, Higashi-ku, Fukuoka 812-8581 JAPAN
{mkouji, hasegawa}@irvs.is.kyushu-u.ac.jp
Abstract— We propose a new method of tactile sensing
for a fingertip of robotic hand. Using a simple fingertip
with soft surface cover and a 6-axes force/torque sensor,
developed is a new function of measuring direction of edge
when the fingertip contacts with an edge of the object.
The developed function is useful not only to plan strategies
for stable grasp and dexterous manipulation but also to
recognize a geometrical shape of an object. Two necessary
components, a 6-axes force/torque sensor and the soft skin,
are commercially available. They will neither restrict design
of a robotic fingertip nor badly affect essential functions of
stable grasping and manipulation.
Index Terms— tactile sensing, soft fingertip
I. I NTRODUCTION
To achieve dexterous manipulation by a multi-fingered
robotic hand, recognition of geometric shape and pose
of an object is indispensable. There have been many
research works on vision sensing and tactile sensing for
the dexterous multi-fingered manipulation. However, there
still remain many problems. The vision sensing can not
avoid such situation that a grasped object is occluded by
the grasping fingers themselves. Lighting condition can not
be constant through the dynamic manipulation process. The
background scene of the object may be visually compli-
cated. All these may lead to failure of the recognition.
Although the tactile sensing is free from these problems,
it suffers drawbacks of different nature: a high resolution
tactile sensor with discreetly distributed sensing elements
is still under development in research laboratories. Shaping
the sensor to fit to a complicated curved surface of a finger,
wiring of all elements, signal processing, and information
processing are still open to be solved.
This paper proposes a new method of sensing direction
of an edge of an object through the contact with a soft
fingertip equipped with 6-axes force/torque sensor (Fig.1).
Measurement of edge direction of an object enables recog-
nizing geometric shape and pose of the object. The tactile
sensing method proposed in this paper will lead to a new
manipulation skill utilizing edges of an object and to a
tactile exploration by active motion of fingertip. Specific
features of this method are summarized as followings.
1) Required hardware elements are fingertip covered
with soft rubber skin and a 6-axes miniature
force/torque sensor which is commercially available.
2) Edge direction is measured while grasping an object
stably and even statically without obstructing manip-
ulation and grasping.
0-7803-8692-2/04/$20.00 ©2004 IEEE.
3) Edge direction is measured based solely on fingertip
sensor readings without relying on a geometric model
of the object.
4) There is no constraint on the shape of fingertip. The
soft cover of the tip enables adjusting the friction
coefficient to augment capability of the finger [1].
Fig. 1. Edge direction of object.
Several works have been reported so far on the tactile
recognition of geometric shape and pose of an object.
Fearing[2] measured pose of a stick using tactile image.
The pose is measured as direction of major axis of ellipse
appeared in the tactile image when the object is pushed
against the sensor. Shimojo et al. [3] transformed tactile im-
age into video signal and obtained the contour using vision
processing. These methods assumes that the global tactile
image of the object is obtained from widely distributed
tactile sensing elements. However, actual implementation
of tactile sensing elements on whole surface of the multi-
fingered robotic hand is few due to the cost and difficulties
of wiring. Gifu Hand[4] has been exceptionally covered
with a film sheet of distributed tactile sensor as an attempt.
The sensor was not completely fitted with the finger shape.
Thus tactile sensing with distributed sensor elements can
not yet be used with the multi-fingered robotic hand.
Feasible surface area for tactile sensing is generally lim-
ited on a multi-jointed robotic finger as compared with the
size of an object to be manipulated. Since the tactile sensor
itself provides local information, global image of the object
is not obtained by a single touch. To avoid this problem,
tactile sensor is often combined with actuator to achieve
the so-called active sensing[5]. A method of recognizing
geometric shape of a polyhedron has been proposed using
position and surface normal of the multiple contact points
measured by active point-to-point contact with fingertip[6].
However, this method relied on the assumption that the
fingertip will not contact the polyhedron on its edge.
The tactile sensing method proposed in this paper is free
824
from disadvantages of the methods described above. Not
only point but also edge is measured. Geometric informa-
tion thus obtained will be effectively used to estimate pose
of the grasped object with the model matching. In addition,
it will be used to recognize geometric shape of the object.
II. C ONTACT OF SOFT FINGERTIP WITH A PLANE AND
AN EDGE
Physical constraint of the soft fingertip contacting with
an object differs depending on the geometric property of the
contacting part of the object: a planar surface and a sharp
edge. This difference of constraint causes difference of
transmissible friction torque to the soft fingertip. We have
already developed a method of discriminating fingertip
contact with a planar surface from that with a sharp edge
[7]. As an extension to our previous work mentioned above,
we propose a new method of measuring direction of an
edge with respect to the fingertip coordinate system when
the fingertip is contacting with the edge.
We have empirically studied the characteristics of a soft
fingertip contacting with a sharp edge of a solid object.
As a result, we have observed that the transmitted torque
at the contact of the soft fingertip differs depending on
the direction of the edge with respect to the fingertip
coordinate system. Although this phenomenon may be
resulted from micro interaction among the components of
the soft materials and the hard faces of the edge, it is quite
difficult to interpret it based on various physical properties
of the material such as elasticity, surface friction, hysterisis
of deformation. On the other hand, concept of the soft
contact (Fig.2) has been widely used in many previous
works to theoretically analyze the contact between robotic
finger and objects. Therefore, we also uses the soft contact
model to analyze the contact between the soft fingertip and
an edge of an object and to formalize a new method of
measuring direction of the edge.
[contact with plane] [cotnact with edge]
Fig. 2. Soft contact model.
Within the soft contact model, only the rotational friction
torque (spin torque) around the surface normal of the object
can be transmitted at the contact in addition to three dimen-
sional force vector. Therefore, detectable transmitted torque
component by the fingertip is only around the surface
normal when the fingertip contacts with a planar surface.
When the fingertip contacts with an edge of the object, the
situation is different : the peak of the edge is buried slightly
into the soft material and the soft material contacts with two
side faces of the edge due to the self deformation. Therefore
825
two different components of friction torque corresponding
to two surface normals can be transmitted and detected.
Each spin torque is perpendicular to the corresponding side
faces of the edge. This spin torque is decomposed into two
components : the one which is parallel to surface normal
of the fingertip at the contact, and another one which is
perpendicular to both the surface normal of the fingertip
at the contact and the edge itself (Fig.3). The latter is
obtained from the spin torque measured at the fingertip
link by subtracting the former. In this paper we call the
latter the tangential component of the spin torque. Then the
direction of the edge is finally obtained as vector product
of the latter and the surface normal of the fingertip.
Fig. 3. Geometric relationship at contact.
We now formalize how to measure the direction of an
edge of an object being grasped by the fingertip cov-
ered with soft skin and equipped with 6-axes force/torque
sensor. We assume that geometric shape information of
the fingertip is known and that position of contact on
the fingertip and a surface normal are measurable. Let
f , m, r, q be force, moment, contact point, spin
torque, respectively. Force f and moment m exerted by
the fingertip are measured by 6-axes force/torque sensor.
The position r of contact on the fingertip is measured by
implementing Bicchi’s method[8]. The spin torque q is
obtained from the balance equation of force and moment
as:
m = q + r×f (1)
By removing the spin torque component parallel to the
surface normal at r from q, we define h as Eq.(3). Vector h
indicates a tangential component of the spin torque, which
is perpendicular to both the surface normal and the edge.
Then we can detect the edge direction as outer product of
vector n and h.

S(r)
n = (2)

S(r) 
h ≡ q − (q · n ) · n (3)

is the gradient operator, S(r) is the function describing
the surface of the fingertip.
III. I MPLEMENTATION AND F INGERTIP
We have developed a robotic fingertip with soft skin[1].
Fig.4 shows the inner structure and its appearance. The
fingertip is composed of an inner shell, hard nail and
soft skin. The inner shell and the hard nail are made of
aluminum. The inner shell is a cylinder with a hemisphere
on its top. Its radius is 11 [mm]. The soft skin is made
of silicone rubber covering the inner shell. The skin has
constant thickness. The thickness of the skin is 5 [mm]. The
inner shell is fixed to 6-axes force/torque sensor which is
then connected to the upper link of the finger. The exerted
force/torque to the fingertip is measured by the force/torque
sensor.
Fig. 4. Soft fingertip with nail.
When the fingertip contacts with a planar surface, the po-
sition of contact on the fingertip is obtained from measured
force/torque using a 6-axes sensor together with geometric
shape information of the fingertip. We are using a method
proposed by Bicchi[8] for this position measurement. How-
ever, this method assumes the spin torque only around the
surface normal at the contact point. This is not true in
the edge contact. Consequently the method suffers larger
error in contact position measurement when applied to
edge contact with larger deformation by grasping force. To
eliminate such error, we measure the position of the contact
point at the beginning of the grasping process where the
grasping force is not yet large. Assuming that the position
of the contact point on the fingertip surface does not move,
it is used for measuring direction of the edge.
It should be noted that, the spin torque at the contact
point on the fingertip has to be exerted sufficiently to be
detected by the force/torque sensor in order to measure the
direction of the edge. In our experience, the spin torque
at the contact is usually sufficient in the ordinary grasping
operation even without mechanical interaction between the
grasped object and its environment. This is due to the
gravity and joint torque of the finger for stable grasping.
Only in exceptional grasping configurations, the method
suffers from larger error without sufficient spin torque at
the contact. However, it can be solved by introducing a par-
ticular motion strategy of the finger to generate sufficient
spin torque since such situation is at least detectable by
observing magnitude of the vector h defined in the previous
section.
IV. E XPERIMENT
We made experiments in practical condition to measure
the direction of an edge being grasped by two fingers.
The robotic hand composed of 2 fingers each with 4 joints
has been mounted on the 6DOF manipulator (Fig.5). The
developed soft fingertips are mounted on each finger. A
geometrical shape of the grasped object is unknown to the
system. The sequence of the experiment is as followings.
1) Two fingers are set open to grasp. Then an object is
put into the open space between the two fingertips
826
and is kept still.
2) Each finger moves to close the hand from initial
position aiming at the position of another fingertip
respectively.
3) When the fingertip contacts the object, a position of
contact on the fingertip is measured.
4) In order to keep stable grasp, each fingertip exerts
constant inner grasping force toward another contact
point respectively.
Fig. 5. Two-fingered hand mounted on 6DOF manipulator.
Direction of the edge of an object is measured at the end
of this motion. It should be noted that the proposed tactile
sensing is implemented in such a simple and ordinary
grasping strategy. The object to be grasped is a cube with
a cone attached together. The robot grasped the object in
three different pose as shown in Figs.6 through 8. The
left fingertip in Fig.5 contacts the edge with three different
directions, -45, 0, 45 [degree] respectively with respect
to the fingertip coordinates system as shown in Fig.9.
The result of tactile measurement at the left fingertip
is shown in Figs.10 through 14. Horizontal axis of each
chart shows the elapsed time. The fingers begin grasping
motion at the time 0. The servo cycle of the finger control
is 20ms. Sensing is made in the same cycle time. When the
magnitude of the tangential component of the spin torque is
enough, the direction of edge of the object is successfully
measured with good accuracy.
Two charts in Figs.10 through 12 show output of the
force sensor: force and moment respectively. The mag-
nitude of the tangential component of spin torque at the
contact is shown in Fig.13. The measured direction of
the edge is shown in Fig.14. The three different pose is
measured with good accuracy.
We have already investigated influence of the sharpness
of the edge on accuracy of measurement (Fig.15). The
result indicates that the sharpness of edge of the object
doesn’t influence the accuracy of measurement.
V. C ONCLUSION
A new method of tactile sensing is proposed to measure
the direction of an edge of an object being grasped by a
multi-fingered robotic hand. Required hardware elements
are fingertips covered with soft rubber skin and a 6-
axes miniature force/torque sensor which is commercially
available. In the experiments, edge direction has been
successfully measured based solely on fingertip sensor
readings in a wide range of mutual relationship of contact
 [Kgf] [Kgf-mm]
1.0 2.0
force x :
0.8 force y : 1.5
force z : moment x :
0.6 moment y :
1.0 moment z :
0.4
0.5
0.2

0 0

-0.2
-0.5
-0.4
[front view] [side view] 0 2.0 4.0 6.0 8.0 10.0 0 2.0 4.0 6.0 8.0 10.0
[s] [s]
Fig. 6. Grasp with the edge(-45 degree).
Fig. 10. Contact with the edge (-45 degree).

[Kgf] [Kgf-mm]
1.0 2.0
force x :
0.8 force y : 1.5
force z :
0.6
1.0 moment x :
0.4 moment y :
0.5 moment z :
0.2

0 0

-0.2
-0.5
-0.4
[front view] [side view] 0 2.0 4.0 6.0 8.0 10.0
[s]
0 2.0 4.0 6.0 8.0 10.0
[s]

Fig. 7. Grasp with the edge(0 degree).

Fig. 11. Contact with the edge (0 degree).

[Kgf] [Kgf-mm]
1.0 2.0
force x :
0.8 force y : 1.5
force z :
0.6
1.0
0.4
0.5
0.2

0 0 moment x :
moment y :
-0.2 moment z :
-0.5
-0.4

[front view] [side view] 0 2.0 4.0 6.0 8.0 10.0

[s]
0 2.0 4.0 6.0 8.0 10.0
[s]

Fig. 8. Grasp with the edge(45 degree). Fig. 12. Contact with the edge (45 degree).

[Kgf-mm]
1.0
45 deg :
0.8 0 deg :
-45 deg :

0.6

0.4

0.2

Fig. 9. Direction of the edge. 0

0 2.0 4.0 6.0 8.0 10.0
[s]

Fig. 13. Magnitude of the tangential component of the spin torque.

between the edge and the fingertip. The proposed method
is applicable to a variety of different edge angle. [degree]
90 45 deg :
R EFERENCES 0 deg :
-45 deg :
[1] K. Murakami and T. Hasegawa, “Novel Fingertip Equipped with Soft 45

Skin and Hard Nail” Proc.IEEE ICRA, pp. 708-713 2003

[2] R. S. Fearing, “Some Experiments with Tactile Sensing during 0
Grasping” Proc. IEEE ICRA, pp. 1637-1643 1987
[3] M. Shimojo, M. Ishikawa, and K. Kanaya, “A Flexible High Res-
-45
olution Tactile Imager with Video Signal Output” Proc.IEEE ICRA
pp. 384-391 1991
[4] H. Kawasaki, T. Komatsu, and K. Uchiyama, “Dextrous Anthropo- -90
0 2.0 4.0 6.0 8.0 10.0
[s]
morphic Robot Hand with Distributed Tactile Sensor: Gifu Hand II”
IEEE/ASME Trans. on Mechatronics, Vol.7, No.3, Sep 2002 Fig. 14. Measurement of the edge of the grasped object.
[5] M. Shimojo and M. Ishikawa, “An Active Touch Sensing Method
Using a Spatial Filtering Tactile Sensor” Proc.IEEE ICRA, pp. 948-
954 1993
[6] K. Nagata, T. Keino, T. Omata, “Acquisition of an Object Model
by Manipulation with a Multifingered Hand” Proc.IEEE/RSJ IROS
pp. 1045-1051 1996
[7] K. Murakami and T. Hasegawa, “A New Method of Tactile Sensing
Using Fingertip with Soft Skin” Proc. of IEEE/RSJ IROS, pp. 535-
pp.540 2003
[8] A.Bicchi: “Intrinsic Contact Sensing for Soft fingers” Proc. of IEEE Fig. 15. Edges with different sharpness.
ICRA, pp. 968-973 1990

827