You are on page 1of 14

68

IEEE TRANSACTIONS ON ROBOTICS, VOL. 29, NO. 1, FEBRUARY 2013

Development of Tool-Type Devices for a


Multifingered Haptic Interface Robot
Takahiro Endo, Member, IEEE, Satoshi Tanimura, and Haruhisa Kawasaki, Senior Member, IEEE

AbstractThis paper presents the design of tool-type devices for


a multifingered Haptic Interface Robot (HIRO), and summarizes
the experimental results. HIRO consists of a robot arm with a
five-fingered hand to which a variety of tools can be attached. The
system is able to present the force sensation of many tool-type
devices. In the medical field, manufacturing industry, and other
fields, there are tools of a variety of shapes with a range of uses,
and a haptic interface that can present the force sensation for
many tools will be important for virtual training systems. HIRO
has five fingers, and thus, we must clarify how many fingers need
to be connected to the tool-type device and which fingers should
be used for the connection. Solving these problems is important
with regard to presenting an operator the force feeling through the
tool-type device. To solve these problems, we propose an optimal
connection method from the mobility and singularity points of view,
and we have developed the tool-type devices for HIRO based on
the proposed method. We describe here several experiments that
were carried out to investigate the performance of the developed
devices.
Index TermsClosed-loop robots, haptic interfaces, virtual
reality (VR).

I. INTRODUCTION
T is possible to communicate with a virtual environment
via a haptic interface. An operator using the haptic interface can feel force sensations from the virtual environment and
can in turn provide force and position information to the virtual environment. Unlike the traditional interface using visual
and audio cues, the haptic interface is unique, as it provides a
bidirectional interaction between a human being and the virtual environment [1][3]. Therefore, the haptic interface is a
key input/output device for communication with highly realistic sensations and has the potential for use in many application
areas.
One of the application areas for the haptic interface is virtual
training systems in the medical field, manufacturing industry,
and other fields. For example, during surgical training, medical
doctors use various surgical tools, such as scissors, tweezers, and
surgical knives, and they must train with these tools to master
specific procedures or techniques. However, it is neither easy

Manuscript received December 28, 2011; revised June 8, 2012; accepted


August 3, 2012. Date of publication August 28, 2012; date of current version
February 1, 2013. This paper was recommended for publication by Associate
Editor T. Asfour and Editor W. K. Chung upon evaluation of the reviewers
comments. This work was supported by the Strategic Information and Communications R&D Promotion Programme of the Ministry of Internal Affairs and
Communications and by the Japan Society for the Promotion of Science under
Grant-in-Aid for Young Scientists (B) (23700143).
The authors are with the Gifu University, Gifu 501-1193, Japan (e-mail:
tendo@gifu-u.ac.jp; q3128020@edu.gifu-u.ac.jp; h_kawasa@gifu-u.ac.jp).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TRO.2012.2212831

nor safe for this training with surgical tools to occur in a real
environment, and thus, a training system that uses virtual reality
(VR) and a haptic interface technology has been researched
aggressively. With the construction of a virtual training system,
training can be safely carried out and a trainee can practice
in various situations that might be difficult to experience in the
real world. Furthermore, the results of some studies indicate that
such a system could increase the skill with which real surgery
can be performed [4] and contribute to the learning of real motor
skills [5], [6].
Based on the need for such systems, many researchers have
been developing tool-type haptic interfaces [7][13]. For example, Okamura et al. [7] developed a scissors-type haptic interface. It has two degree of freedom (DOF) of motion and force
feedback: one for cutting, namely, the single blade rotation, and
one for translational motion of the device. Their group also presented an analytical model to compute force applied to scissors
during cutting of a slab of material [8] and evaluated the cutting
model using the aforementioned scissors-type haptic interface.
Sato et al. [9] developed a brain retractor-type haptic interface
to train surgeons in brain surgery and investigated the soft tissue pushing operation using the haptic device for simulation
of brain tumor resection. In another study [10], a microscissortype haptic device was developed, which presented the cutting
resistance forces to the operator. Goksel et al. [11] developed a
needle-type haptic device and a probe-type haptic device, and
a haptic simulator for prostate brachytherapy with simulated
needle and probe interaction. The use of haptic technology in
medical simulators has attracted attention for many years, and
various commercialized products are already available, such as
LapSim [12], LAP mentor [13], and others. By using a tool-type
haptic interface in a virtual environment, an operator can carry
out virtual training, while feeling force sensations; however,
these tool-type haptic interfaces present the force sensation of
only the corresponding single type of tool. To present the force
sensations of a variety of tools, many tool-type haptic interfaces
are required, which requires multiple installation locations and
costs a great deal.
For this reason, we previously developed a haptic system
that presents the force sensations of a variety of tools [14].
This system consists of a multifingered haptic interface robot
named Haptic Interface Robot (HIRO) and numerous tool-type
devices, including a surgical knife, scissors, and syringe. HIRO
has five haptic fingers, and a variety of tool-type devices can
easily be attached to and removed from HIROs haptic fingers,
enabling the system to present the force sensations of many
tool-type devices. However, we must consider how many haptic
fingers need to be connected to the tool-type device and which

1552-3098/$31.00 2012 IEEE

ENDO et al.: DEVELOPMENT OF TOOL-TYPE DEVICES FOR A MULTIFINGERED HAPTIC INTERFACE ROBOT

69

TABLE I
SPECIFICATIONS OF HIRO III

Fig. 1.
holder.

Multifingered haptic interface robot. (a) HIRO III and (b) a finger
TABLE II
SPECIFICATIONS OF HIRO IIIS FORCE SENSOR

haptic fingers should be used for the connection. In our previous


research, we determined the connection by trial and error, and
we did not clarify the optimal connection between HIRO and the
tool-type device. Technically clarifying the optimal connection
is important with regard to presenting the operator the force
feeling through the tool-type device.
To solve these problems, we propose an optimal connection
method from mobility and singularity points of view, and based
on the proposed method, we have developed new tool-type devices. There are many types of tools in our real world, and they
can be divided into two main classes: tools with no joints and
tools with joints. For example, the knife has no joints, and thus,
its DOF is 6. On the other hand, the tweezers and the scissors
have one joint, allowing 1 DOF, and thus, the DOF is 7. We
developed a knife-type device as the tool with no joints and a
tweezers-type device as the tool with joints. Furthermore, we describe our experimental investigation of the tools performance.
The preliminary versions of this paper have been published
[15], [16]. This extended version contains a development of
new knife-type device, new experimental results, and the new
discussion about the applicability of the proposed connection
method to other tool-type devices.
This paper is organized as follows: In the next section, a
multifingered haptic interface robot, HIRO, and our previous
tool-type devices for HIRO are introduced. Section III presents
the optimal connection method between HIRO and the tooltype device, and a newly developed knife-type device and a
tweezers-type device are presented in Section IV and V, respectively. The experimental results that are described in Section VI
demonstrate the great potential of our system. Finally, Section
VII presents our conclusions.
II. FIVE-FINGERED HAPTIC INTERFACE ROBOT
AND PREVIOUS TOOL-TYPE DEVICES
A. Five-Fingered Haptic Interface Robot
We have developed a multifingered haptic interface robot,
named HIRO III [17], which is shown in Fig. 1. HIRO III can
present three-directional forces at an operators five fingertips.
The specifications of HIRO III are shown in Table I. HIRO III
can be briefly summarized as follows.

HIRO III consists of an arm and a haptic hand. The arm


consists of an upper arm, a lower arm, and a wrist. The arm has
3 DOF at the arm joint and 3 DOF at the wrist joint. The arm,
therefore, has six joints, allowing 6 DOF. The haptic hand is
constructed of five haptic fingers. Each haptic finger has three
joints, allowing 3 DOF. The first joint relative to the hand base
allows abduction/adduction, while the second and the third joint
allow flexion/extension. The total DOF of HIRO III is 21, and
its working space covers VR manipulation on the space of a
desktop. A three-axis force sensor is installed at the top of each
haptic finger. The force sensor was custom made for HIRO
III [17]. Its specifications are shown in Table II.
Note that the maximum displayable stiffness given in Table I
expresses the maximum spring coefficient of a virtual wall that
one haptic finger of HIRO III can present stably, with the virtual
wall made using a springdamper model. This value was measured by a contact experiment involving a virtual wall [18]. In
this experiment, the haptic finger was in a configuration that can
be really used in haptic display mode. Furthermore, the maximum output force of the haptic finger in Table I depends on the
configuration of the haptic finger. The maximum output force
of the haptic finger is 3.6 N in the worst configuration case. For
more details, see [17].
To manipulate HIRO III, an operator wears a finger holder,
a sample of which is shown in Fig. 1(b), on each of his/her
fingertips. The finger holder has a steel sphere, and the haptic
finger has a permanent magnet at its fingertips. By means of the
magnet force, the finger holder can be connected to HIRO III,
as shown in Fig. 1(a). Here, note that the sphere, when attached
to the permanent magnet at the force sensor tip, forms a passive
spherical joint. Its role is to adjust for differences between the
human and haptic finger orientations. Hereafter, this version of
the multifingered haptic interface robot is described as HIRO.

70

IEEE TRANSACTIONS ON ROBOTICS, VOL. 29, NO. 1, FEBRUARY 2013

Fig. 3.

Fig. 2. HIRO with surgical knife. (a) HIRO with surgical knife. (b) Tool
devices steel sphere. (c) Surgical knife and operators hand.

Haptic fingers and knife device.

not correspond to the DOF of the overall system. In this case, the
mobility index M [19] implies the DOF of the overall system,
as follows:
m

li
(1)
M = 6(N m 1) +
i=1

B. Previous Tool-Type Devices for HIRO


Our previously developed haptic interface presents the force
sensation of plural tools using HIRO. As tool-type devices, a
surgical knife, scissors, and syringe devices were developed. As
described previously, HIRO has a permanent magnet at each of
its fingertips. Thus, if we install the steel spheres in the tool
device, the device is easily attached to HIRO by the magnetic
power, and it is easy to exchange the plural tool-type devices.
As an example, Fig. 2 shows the surgical knife device, and
the steel spheres are installed as shown in Fig. 2(b), and this
setup presents the force sensation of the knife device to an
operator. Here, note that HIRO has the force sensors, motors,
and encoders required to create the force sensation, and thus,
the tool-type devices only require a structure similar to that of
the associated real tool.
III. CONNECTION BETWEEN HIRO
AND TOOL-TYPE DEVICES
HIRO has five haptic fingers. It is a big challenge to determine how many haptic fingers need to be connected to a
given tool-type device, and which fingers should be used for
the connections. These problems are bound up with presenting
the operator the force feeling through the tool-type device. To
solve these problems, we propose an optimal connection method
from mobility and singularity points of view. Here, note that the
arm is not included in the analysis because the force display of
HIRO is accomplished by haptic finger parts. For more details
concerning the control of HIRO, see Section VI.
A. Connection Analysis by Mobility
When HIRO is connected to the tool-type device via several
haptic fingers, the system acts as a parallel mechanism. (We
show an example of the connection between three haptic fingers
and a knife-type device in Fig. 3.) Having the features of a
parallel mechanism, the system is highly accurate, with high
output force, a high level of stiffness, and other advantageous
features that are important elements for a haptic interface. We
note that, in the parallel mechanism, the number of joints does

where N is the number of links, m is the number of joints,


and li is the DOF of the ith joint. Thus, using the mobility, we
determine the number of fingers necessary for HIRO to have
connected to the tool-type device. Here, we assume that the
system has no passive or idle DOF, which does not affect the
motion of the other links [19], [20]. Under this assumption, we
consider the number of the required fingers.
In particular, by using the mobility, we consider the following
two points: 1) the realization of DOF of the tool-type device
and 2) the control of all DOF of the overall system. For item 1),
to realize the tool-type devices DOF Fto ol , the mobility must
satisfy
M Fto ol .

(2)

For item 2), to control all DOF of the overall system, the
number of active joints D must satisfy
D M.

(3)

From these two points, we determine the required number of


haptic fingers that satisfies conditions (2) and (3). For a concentrated method of determination, see Sections IV and V.
B. Connection Analysis by Singularity
In the previous section, we determined the number of haptic
fingers for the connection, but we did not clarify which fingers should be used. Remembering that the parallel mechanism
has the drawback of singularity [21], we remember also that
singularity is the particular configuration where the system becomes uncontrollable [22] and that we must, therefore, avoid
singularity. By considering the avoidance of singularity, we can
determine which fingers should be used.
When the inputoutput relationship of the parallel mechanism
can be written as
Aq = Bv
where q R is the actuated joint velocity vector, v Rm
is the platform Cartesian coordinate velocity vector, and A
Rn n and B Rn m (n m) are appropriate matrixes, the
n

ENDO et al.: DEVELOPMENT OF TOOL-TYPE DEVICES FOR A MULTIFINGERED HAPTIC INTERFACE ROBOT

singularity of the parallel mechanism consists of three kinds of


singularities [23]. The first kind of singularity is defined when
A is singular, the second kind of singularity is defined when B
is singular, and the third kind of singularity is defined when A
and B are singular at the same time. To avoid the three kinds of
singularity, we introduce the following performance index:
PI = WA WB + Wq
WA = |det A| ,
Wq =

p


WB =

(4)


det(B T B)

i [e(q i a i ) + e(q i b i ) ]

71


= 12, and m
i=1 li = 18 (nine revolute joints and three spherical joints). The DOF of the knife-type device Fto ol is 6, and
condition (2) is satisfied. The number of active joints D is nine,
because three haptic fingers are used to hold the device, and
thus, condition (3) is satisfied. Although conditions (2) and (3)
are satisfied when the number of haptic fingers is four or more,
we used three haptic fingers in our device, which is the minimal
acceptable number, at the connection.

(5)
B. Kinematics of the Knife-Type Device
(6)

i=1

where , , and i are the weighting coefficients of


WA WB , Wq , and the ith actuated joint, respectively, p is the
number of the actuated joint, qi is the angle of the ith actuated
joint, is the parameter to adjust an exponential function, and
ai and bi are the lower and upper limits of the angle of the ith
actuated joint, respectively.
In (4), WA is well-known manipulability, and the manipulability becomes high and the configuration of the system is away
from the first kind of singularity when the value of WA becomes large. WB evaluates the second kind of singularity, and
if the value of WB becomes large, the system is away from the
second kind of singularity. We set the multiplication WA WB
in (4). Thus, if WA WB becomes large, then the configuration
of the system is away from the first and the second kinds of
singularity, and this corresponds to the third kind of singularity.
Here, note that the joint of the robot usually has a limit to its
movable range. If we consider the value of WA WB only, it is
possible that the configuration of the system is away from the
singularities but the joints go to the outside of the limit of the
movable range. We therefore add the penalty function Wq to the
performance index, which corresponds to the limits of the joint
angles. From these points of view, which haptic fingers should
be used is decided upon by maximizing PI. For a concentrated
method of determination, see Sections IV and V.
IV. DEVELOPMENT OF A KNIFE-TYPE DEVICE
This section describes our development of a knife-type device
based on the proposed connection method.
A. Connection Analysis by Mobility
First, we analyze the number of haptic fingers necessary to
have connected to the knife-type device. Haptic fingers and a
knife-type device are connected through the passive spherical
joint described in Section II. Three or more haptic fingers are
necessary to support the device. Here, note that when we use two
haptic fingers at the connection, the system has an idle DOF, i.e.,
the system allows spin movement around an axis through the
center of two spherical joints at the connection points between
HIRO and the tool-type device. Thus, to support the knife-type
device with the haptic fingers, three or more haptic fingers are
necessary.
When the device is connected to HIRO with three haptic fingers, the mobility index (1) is M = 6 because of N = 11, m

To evaluate the connection using a performance index (4), we


first consider the kinematics of the overall system, as illustrated
in Fig. 3. The system consists of three haptic fingers and a knifetype device, in which the ith haptic finger contacts the device at
point Ci . The coordinate systems are defined as follows: b is
the base coordinate system, to ol is the object coordinate system
fixed on the device, and F i is the ith fingertip coordinate
system fixed on the ith haptic fingertip. In addition, the following
notations are used: b pF i R3 is the position vector of F i with
respect to b , b RF i R33 is the orientation matrix of F i
with respect to b , b pto ol R3 is the position vector of to ol
with respect to b , and b Rto ol R33 is the orientation matrix
of to ol with respect to b .
Since the contact at Ci is fixed, the following constraint between the haptic fingertip position and the knife-type device
position is obtained:
b

pto ol + b Rto ol to ol pC i = b pF i + b RF i F i pC i

(7)

where to ol pC i R3 is the position vector of Ci with respect to


to ol , and F i pC i R3 is the position vector of Ci with respect
to F i . Differentiating (7) yields
b

p to ol [(b Rto ol to ol pC i )]b to ol


= b p F i [(b RF i F i pC i )]b F i

(8)

where b to ol R3 and b F i R3 are the angular velocities of


the knife-type device and the ith haptic fingertips, respectively,
(p) R33 is a skew-symmetric matrix expressing the crossproduct form of a vector p R3 , and the fact that to ol pC i and
Fi
pC i are constants was used in the derivation of (8).
Now, we define the following matrices:
ol
= [I 33 , (b Rto ol to ol pC i )] R36
D to
i

D F i = [I 33 , ( RF i
b

Fi

pC i )] R

36

(9)
(10)

where I 33 R33 is an identity matrix. Furthermore, the ith


haptic fingertip velocity and the joint angle velocity are related
by

b
p F i
(11)
= J F i q i
b
F i
where J F i R63 is a Jacobian, and q i R3 is the joint angle
vector of the ith haptic finger.
From (8)(11), we can obtain the following kinematics:
D to ol v to ol = J C F q

(12)

72

IEEE TRANSACTIONS ON ROBOTICS, VOL. 29, NO. 1, FEBRUARY 2013

TABLE III
PARAMETERS IN CONNECTION ANALYSIS OF THE KNIFE-TYPE DEVICE

TABLE IV
JOINT ANGLES FOR THE KNIFE-TYPE DEVICE

ol
ol
ol
where D to ol = col[D to
, D to
, D to
] R96 , v to ol =
1
2
3
b
b
6
col[ p to ol , to ol ] R is the velocity vector of the knife-type
device, J C F = diag[D F 1 J F 1 ,D F 2 J F 2 ,D F 3 J F 3 ] R99 ,
and q = col[q 1 , q 2 , q 3 ] R9 .
On the other hand, from the principle of virtual work, the
following relation is obtained:

F to ol = D Tto ol f C

(13)

where F to ol = col[f to ol , nto ol ] R6 is the total force/moment


vector applied to the knife device by the haptic fingers, f c =
col[f C 1 , f C 2 , f C 3 ] R6 , and each haptic finger applies a
force f C i at the contact point Ci .
C. Connection Analysis by Singularity
Based on the kinematics and the performance index, we can
determine which haptic fingers should be used for the connection. From Section IV-A, we know that three haptic fingers will
be used. For the combination of all three haptic fingers, the joint
angles that maximize the value of (4) are derived, and we set
the knife-type device to be connected to the haptic fingers that
maximize the performance index. In this case, the performance
index (4) can be rewritten as
PI = WA WB + Wq
WA = |det J C F | , WB =
Wq =

3 
3


(14)


det(D Tto ol D to ol )

(15)

j [e(q i j a i j ) + e(q i j b i j ) ]

(16)

i=1 j =1

where j is the weighting coefficient of the jth joint angle of the


haptic finger, qij is the jth joint angle of the ith haptic finger,
and aij and bij are the lower and upper limits of the jth joint
angle of the ith haptic finger, respectively.
In the derivation of the haptic fingers that maximize PI, the
conjugate gradient method and the parameters in Table III were
used. In addition, it is easy to see that the position and the
orientation of the object coordinate system to ol are not related
to the value of PI. For example, the variables of the function
WA and Wq are qij , and these functions are not related to
to ol . Furthermore, D to ol , in this case, becomes a grasp matrix
because of (9), and it is well known that WB is not related to
to ol [24]. Thus, in the derivation, to ol was fixed on the blade
edge of the knife device, and to ol was set on a centroid of a
triangle formed by three haptic fingertips.
The value of (14) reaches a maximum when the combination
of the thumb, index finger, and pinky finger is used. In this case,

Fig. 4. Developed knife device. (a) Knife-type device. (b) Knife-type device
that connects to HIRO.

WA = 2.49 1011 , WB = 2.01 102 , Wq = 1.47 1013 ,


and PI = 3.52 1013 . The joint angles in this case are shown
in Table IV.
For the development of the knife-type device, we considered
the following guidelines: The device is connected to the haptic
fingertip positions that maximize the value of (14), namely, the
knife device is connected to HIRO at the angles shown in Table IV, HIRO and the device are connected by passive spherical
joints, the device is set to the direction that most helps an operators grasp, and an actual surgical knife is used to maintain the
appearance of a real surgical knife. Based on these guidelines,
the knife device was developed. Fig. 4 shows the developed
knife device. The actual surgical knife is connected to HIRO
through the white connections known as the offset arm. The offset arm was made of acrylonitrile-butadiene-styrene resin and
was screwed to the actual surgical knife. In addition, three steel
spheres were screwed to the offset arm, and thus, the knife-type
device and three haptic fingers were connected by magnetic
power.
We investigated the effect of variation of the weights (, , ,
i ) on the optimal solution, and the results are shown in Fig. 5.
Fig. 5(a) shows the value of PI in the case that changes,
Fig. 5(b) shows the value of PI in the case that changes,
Fig. 5(c) shows the value of PI in the case that changes, and
Fig. 5(d) shows the value of PI in the case that i changes. (Here,
we set 1 , 2 , and 3 to the same value.) The horizontal axis is
the value of the corresponding weight, and the vertical axis is the
value of PI. The parameters that we used are shown in the caption

ENDO et al.: DEVELOPMENT OF TOOL-TYPE DEVICES FOR A MULTIFINGERED HAPTIC INTERFACE ROBOT

Fig. 5. Effect of variation of the weights (, , , i ) on the knife-type device.


(a) Effect of variation of . The following were used as parameters other than
: = 1.0 101 3 , 1 = 1.2, 2 = 1.0, 3 = 1.1, and = 10. (b) Effect of
variation of . The following were used as parameters other than : = 1.0,
1 = 1.2, 2 = 1.0, 3 = 1.1, and = 10. (c) Effect of variation of . The
following were used as parameters other than : = 1.0, = 1.0 101 3 , 1
= 1.2, 2 = 1.0, and 3 = 1.1. (d) Effect of variation of i . The following were
used as parameters other than i : = 1.0, = 1.0 101 3 , and = 10.

of the figure. As shown in Fig. 5(a), when becomes large, PI


also becomes large. However, in this case, the joint angles of the
haptic fingers approach their movable limit, and the joint angles
that maximize PI go outside of the movable range when >
3.1. In contrast, if becomes small, the manipulability of the
haptic fingers is reduced, because the influence of Wq becomes
large, and thus, the systems performance of the force display is
poor.
The effect of variation of , as shown in Fig. 5(b), was contrary to the case of , and PI becomes large when becomes
small. However, in this case, the joint angles of the haptic fingers
approach their movable limit, and the joint angles that maximize
PI go outside of the movable range when < 4.0 1014 . In
contrast, if becomes large, the manipulability of the haptic
fingers is reduced, because the influence of Wq becomes large.
As shown in Fig. 5(c), if approaches a small value, the joint
angles of the haptic fingers become small, and the joint angles
that maximize PI go outside of the movable range when < 3.0.
Finally, in Fig. 5(d), i has the same effect as . That is, if i
becomes small, then PI becomes large, but in this case, the joint
angles of the haptic fingers approach their movable limit, and the
joint angles that maximize PI go outside of the movable range
when i < 0.4. If i becomes large, the manipulability of the
haptic fingers is reduced, because the influence of Wq becomes
large, and the systems performance of the force display is poor.

Fig. 6.

Haptic fingers and tweezers-type device.

Fig. 7.

Tweezers-type device and the object coordinate system.

73

A. Connection Analysis by Mobility


Using the same method as described in Section IV-A, we can
determine the number of haptic fingers necessary to have connected to the tweezers-type device. Here, note that, unlike the
case of the knife-type device, a pair of tweezers has 7 DOF. Haptic fingers and a tweezers-type device are connected through the
passive spherical joint. Therefore, to support the tweezers with
the haptic fingers and not to generate an idle DOF, three or more
haptic fingers are necessary. When the device is connected to
HIRO with three haptic fingers,
the mobility index is M = 7 because of N = 12, m = 13, and m
i=1 li = 19 (ten revolute joints
and three spherical joints). At this time, the number of active
joints D is nine, and thus, condition (3) is satisfied. Furthermore,
the DOF of the tweezers is 7, and condition (2) is satisfied. Although conditions (2) and (3) are satisfied when there are three
or more haptic fingers, we used three haptic fingers, which is a
minimal acceptable number, at the connection.
B. Kinematics of the Tweezers-Type Device

V. DEVELOPMENT OF A TWEEZERS-TYPE DEVICE


In Section IV, we developed the knife-type device for HIRO.
The knife device has no joints, and thus, its DOF is 6. However,
some commonly used tools have joints. In this section, we focus
on tweezers as an example of a tool with joints. We describe our
development of a tweezers-type device.

Next, we consider the kinematics of the tweezers-type device,


as illustrated in Fig. 6. The system consists of three haptic
fingers and tweezers, in which the ith haptic finger contacts the
tweezers at point Ci . The coordinate systems are the same as in
Section IV-B.
We connected two haptic fingers to one side of the blade
(blade A), as shown in Fig. 7, and we connected one haptic finger
to other side of the blade (blade B). Based on this setup plan, we

74

IEEE TRANSACTIONS ON ROBOTICS, VOL. 29, NO. 1, FEBRUARY 2013

set the object coordinate system. The object coordinate system


to ol is fixed on blade A. The origin of to ol is the fulcrum
point of the tweezers, and its z-axis coincides with the openingand-closing axis. The y-axis is set in a tip-to-fulcrum direction
along blade A. The x-axis is set so that to ol becomes the
right-handed coordinate system. On the other hand, we define
the object coordinate system to ol2 on blade B. to ol2 is the
coordinate system that rotates to ol rad at the z-axis. Here,
is the opening-and-closing angle of the tweezers.
Since the contact at Ci (i = 2, 3) is fixed, we can obtain
the following velocity relation, as in the case of the knife-type
device:
b

p to ol [(b Rto ol to ol pC i )]b to ol


= b p F i [(b RF i F i pC i )]b F i , for i = 2, 3

(17)

Here, note that we cannot obtain the same relation for C1 because of . For the contact at C1 , we can obtain the following
constraint:
b

pto ol + b Rto ol to ol Rto ol2 to ol2 pC 1 = b pF 1 + b RF 1 F 1 pC 1


(18)
where to ol2 pC 1 R3 is the position vector of C1 with respect
to to ol2 . Differentiating (18) yields
b

p to ol [(b Rto ol2 to ol2 pC 1 )]b to ol


b Rto ol [(to ol Rto ol2 to ol2 pC 1 )]to ol to ol2
= b p F 1 [(b RF 1 F 1 pC 1 )]b F 1

(19)

where to ol to ol2 R3 is the angular velocity of to ol2 . Here,


T because to ol2 rotates around
note that to ol to ol2 = [0, 0, ]
the z-axis with respect to to ol . From this property, (19) can be
rewritten as follows:
b

p to ol [(b Rto ol2 to ol2 pC 1 )]b to ol


+ b Rto ol [([0, 0, 1]T )]to ol Rto ol2 to ol2 pC 1
= b p F 1 [(b RF 1 F 1 pC 1 )]b F 1

(20)

Now, we define the following matrices:


ol
D to
= [I 33 , [(b Rto ol2 to ol2 pC 1 )],
1
b

Rto ol [([0, 0, 1]T )]to ol Rto ol2 to ol2 pC 1 ]

R37
ol
D to
i

(21)

= [I 33 , [( Rto ol2
b

to ol2

pC i )], O 31 ] R
for i = 2, 3

37

(22)

where O 31 R is a zero matrix. As in the case of the knifetype device, we can obtain the relation (11) between the ith
haptic fingertip velocity and the joint angle velocity. Thus, from
(17), (20)(22), (10), and (11), we can obtain the following
kinematics:
3

D to ol v to ol = J C F q

(23)

where D to ol =
R , v to ol =
col[b p to ol , b to ol , ] R is the velocity vector of the tweezerstype device, and J C F and q are the same as in the case of the
knife-type device.
ol
ol
ol
col[D to
, D to
, D to
]
1
2
3
7

97

TABLE V
PARAMETERS IN CONNECTION ANALYSIS OF THE TWEEZERS-TYPE DEVICE

On the other hand, from the principle of virtual work, the


following relation is obtained:
F to ol = D Tto ol f C

(24)

where F to ol = col[f to ol , nto ol , n ] R7 , f to ol and nto ol are


the total force and moment vector applied to the tweezers-type
device by the haptic fingers, respectively, and n is the torque
at the opening-and-closing axis.
C. Connection Analysis by Singularity
Based on the performance index (14) and using an actual pair
of tweezers, we developed a tweezers-type device. From Section V-A, we know that three haptic fingers are used. By using
the conjugate gradient method for all three haptic fingers, we
derived the joint angles of haptic fingers, the position b pto ol ,
and the orientation b Rto ol that maximize (14). Here, note that,
unlike the knife-type device, D to ol does not result in the grasp
matrix because the opening-and-closing angle exists. Therefore, b pto ol and b Rto ol of the object coordinate system to ol are
related to the value of PI. Furthermore, note that, if we set no
condition for b pto ol , the derived b pto ol diverges. Thus, we must
set a condition for b pto ol . It is preferable that b pto ol be small,
because the required haptic fingertip force to hold the tweezerstype device f c increases if b pto ol becomes large. However, if
b
pto ol becomes small, there is a danger that the tweezers and the
haptic hand of HIRO will collide. As stated previously, we used
actual tweezers, and we found that the distance between the tip
and the opening-and-closing axis was 106.9 mm; therefore, we
established that b pto ol was located 106.9 mm from a centroid
of a triangle formed by three haptic fingertips in the direction
of outward normal, and we set b pto ol so that we could avoid a
collision between the tweezers and the haptic hand.
According to the instruction manual for tweezers, we operate
the tweezers by moving the handheld handle. It is important to
avoid a collision between the operators hand and HIROs hand
during the use of the tweezers, and thus, we made a condition
for b Rto ol so that the z-axis of to ol is parallel to the triangle
formed by three haptic fingers. Furthermore, the parameters in
Table V were used.
As a result, the value of (14) reaches a maximum when the
thumb and index fingers are connected to blade A and the pinky
finger is connected to blade B. Furthermore, WA = 2.47
1011 , WB = 1.69 103 , Wq = 1.28 1014 , and PI =
2.90 1014 . The joint angles in this case are shown in Table VI,

ENDO et al.: DEVELOPMENT OF TOOL-TYPE DEVICES FOR A MULTIFINGERED HAPTIC INTERFACE ROBOT

75

TABLE VI
JOINT ANGLES FOR THE TWEEZERS-TYPE DEVICE

Fig. 9. Effect of variation of the weights (, , , i ) on the tweezers-type


device. (a) Effect of variation of . The following were used as parameters other
than : = 1.0 101 4 , 1 = 1.2, 2 = 1.0, 3 = 1.1, and = 10. (b) Effect
of variation of . The following were used as parameters other than : = 1.0,
1 = 1.2, 2 = 1.0, 3 = 1.1, and = 10. (c) Effect of variation of . The
following were used as parameters other than : = 1.0, = 1.0 101 4 , 1
= 1.2, 2 = 1.0, and 3 = 1.1. (d) Effect of variation of i . The following were
used as parameters other than i : = 1.0, = 1.0 101 4 , and = 10.

Fig. 8. Developed tweezers-type device. (a) Tweezers-type device.


(b) Tweezers-type device that connects to HIRO.

and b Rto ol has the following values:

0.086 0.888
b
Rto ol = 0.286 0.457
0.954 0.057

0.453
0.842
0.293

For the development of the tweezers-type device, we considered


the following guidelines: The device is connected to the fingertip
positions that maximize the value of (14), namely, the tweezerstype device is connected to HIRO at the angles of the haptic
fingers shown in Table VI, HIRO and the device are connected
by the passive spherical joints, b Rto ol satisfies the aforementioned conditions, and the device uses actual tweezers. Based
on these guidelines, the tweezers-type device was developed.
Fig. 8 shows the developed tweezers-type device. We can see
that the actual pair of tweezers is connected to HIRO through
the offset arm as in the case of the knife-type device.
We investigated the effect of variation of the weights (, , ,
i ) on the optimal solution, like the case of the knife-type device.
The results are shown in Fig. 9. Fig. 9(a) shows the value of PI
in the case that changes, Fig. 9(b) shows the value of PI in
the case that changes, Fig. 9(c) shows the value of PI in the
case that changes, and Fig. 9(d) shows the value of PI in the
case that i changes (Here, we set 1 , 2 , and 3 to the same
value.). The parameters that we used are shown in the caption of
the figure. The tendency of the effect of variation of the weights
(, , , and i ) is the same as that in the case of the knife-type
device. As shown in Fig. 9(a), when becomes large, PI also
becomes large, and in this case, the joint angles of the haptic
fingers approach their movable limit, and the joint angles that

maximize PI go outside of the movable range when > 3.9. In


contrast, if becomes small, the manipulability of the haptic
fingers is reduced, because the influence of Wq becomes large,
and thus, the systems performance of the force display is poor.
The case of , as shown in Fig. 9(b), was contrary to the
case of , and PI also becomes large when becomes small.
In this case, the joint angles of the haptic fingers approach their
movable limit, and the joint angles that maximize PI go outside
of the movable range when < 2.5 1015 . On the contrary,
if becomes large, the manipulability of the haptic fingers is
reduced, because the influence of Wq becomes large. As shown
in Fig. 9(c), if approaches a small value, the joint angles of the
haptic fingers become small, and the joint angles that maximize
PI go outside of the movable range when < 2.5.
Finally, as shown in Fig. 9(d), i has the same effect as .
That is, if i becomes small, then the value of PI becomes large,
but in this case, the joint angles of the haptic fingers approach
their movable limit, and the joint angles that maximize PI go
outside of the movable range when i < 0.3. In contrast, if
i becomes large, the manipulability of the haptic fingers is
reduced, because the influence of Wq becomes large, and thus,
the systems performance of the force display is poor.
D. Applicability of the Proposed Connection Method
to Other Tool-Type Devices
As described previously, we developed the tweezers-type device based on the proposed connection method. The proposed
method can also be applied to the scissors-type device, which
like the tweezers is a tool with a joint. Since the proposed method
can also be applied to tools without joints, we considered the
applicability of the proposed connection method to other tools.
Here, we consider the applicability from the mobility point of
view.
First, we assumed a tool consisting of k revolute joints and k
+ 1 links. We show an example of such a tool in Fig. 10. The link

76

IEEE TRANSACTIONS ON ROBOTICS, VOL. 29, NO. 1, FEBRUARY 2013

VI. EXPERIMENTS

Fig. 10.

Haptic fingers and a tool consisting of three revolute joints.

To evaluate the developed devices, we carried out two experiments. One was to manipulate the knife-type device, and
the other was to manipulate the tweezers-type device. In each
experiment, the manipulability-optimized control was used as
the control law of HIRO [25]. This is a mixed control method
consisting of a haptic finger-force control and an arm position
control intended to maximize the control performance index
(27). The force control of the haptic finger is given by
t
F e (s)ds
F (t) = K 1 J TF F e (t) + K 2 J TF
0

is connected to the joint, and all joints are revolute joints, and
its rotational axes are all in the same direction. Let ljto ol be the

DOF of the jth joint of the tool. Then, we obtain kj=1 ljto ol = k.
Now, we set n haptic fingers to be connected to the tool. In this
case, the number of links of the haptic fingers is given as Nn =
3n, and the number of joints of the haptic fingers is given as
n
be the DOF of the jth joint of the
mn = 3n. In addition, let lij


n
ith haptic finger, and we obtain ni=1 3j =1 lij
= 3n.
The haptic fingers and the tool are connected through the passive spherical joint described in Section II, and the number of
n. Thus, the DOF
joints at n contact points is given as mcont = 
of the joint at the ith contact point licont is ni=1 licont = 3n.
By substituting these values into (1), we obtain M = k + 6 be1, m = k + m
+ mcont , and
cause of N 
= (k + 1) + N
n + 

n

m
k
n
3
n
to ol
n
cont
l
=
l
+
l
+
. From
i=1 i
j =1 j
i=1
j =1 ij
i=1 li
this, we found that condition (2) is always satisfied. By considering the relationship between D active joints and k tool
joints, which satisfies condition (3) (namely, D k + 6), we
obtain the following results: 1) When three haptic fingers are
used in the connection, D = 9, and thus, we can connect a tool
with up to three joints to HIRO; 2) when four haptic fingers are
used in the connection, D = 12, and thus, we can connect a tool
with up to six joints to HIRO; and 3) when five haptic fingers
are used in the connection, D = 15, and thus, we can connect
a tool with up to nine joints to HIRO. In addition, if we obtain
the concrete form of the tool and if we derive the kinematics
of the overall system, we can clarify which fingers should be
employed with the given tool by using the performance index
(14).
In the newly developed knife-type and tweezers-type devices,
the use of three haptic fingers is sufficient for the connection
between HIRO and the tool-type device. Thus, the remaining
two haptic fingers that have not connected to the tool-type deice
are fixed in the straight state (for example, see Fig. 4 as the case
of the knife-type device). However, for devices other than the
knife-type or the tweezers-type device, there is a possibility that
four or more haptic fingers are needed for connection between
HIRO and the tool-type device. Since HIRO has five fingers, it
can respond to such a situation, and we believe there is a big
merit and the potential to use a five-fingered hand to manipulate
tool-type devices.

J TF

F d K 3 q f (t)

(25)

where F = [ T1 , T2 , T3 ]T R9 is a joint torque vector of the


haptic finger in use, J F is a Jacobian, F = [F T1 , F T2 , F T3 ]T
R9 is a force vector whose subvector is the force vector
at the fingertip, F d = [F Td1 , F Td2 , F Td3 ]T R9 is the desired
force, F e = F d F , and q f = [q T1 , q T2 , q T3 ]T R9 is a
joint angle vector of the haptic finger. Furthermore, K i is the
positive feedback gain matrix. The control of the arm is given
by the following PD (proportional and derivative) control with
gravitational and external force compensators:
A (t) = K A 1 (q A d q A ) + K A 2 (q A d q A ) + g A (q A )

3

F di

i=1
T

(26)
+ JA

3


(pi phb ) F di
i=1

where q A R is the arm joint angle vector, q A d R6 is


the desired arm joint angle vector, which is to be determined,
A R6 is the arm joint torque, K A i is the positive feedback
gain matrix, g A (q A ) is the gravitational compensator term, J A
is a Jacobian, pi R3 is the ith fingertip position vector, and
phb R3 is the tip of the arm. Here, note that q A d is defined
to maximize the following control index (27) under a constraint
condition in which the five haptic fingertip positions are fixed
to the operator fingertip positions:
6

CPI = WA + Wq + QA

(27)

1
QA = (q A d q A )T (q A d q A )
2
where and are weighting coefficients, WA is a manipulability measure of the haptic finger (15), Wq is a penalty function
to keep the finger joint angles within the movement range (16),
QA is the penalty function to prevent a large change of the arm
angle, and > 0 is a weighting matrix. Here, a finger/arm that
reaches the limit of the movable range is switched to a position
control to keep the joint angle within the movable range, and the
rest joints of the fingers/arm are controlled by (25) and (26). After returning to within the movable range, the control is switched
back again to (25) and (26). Thus, the force display of HIRO
is accomplished by the haptic finger. Details of the control law
have been shown in [25]. For the experiment, the control PC

ENDO et al.: DEVELOPMENT OF TOOL-TYPE DEVICES FOR A MULTIFINGERED HAPTIC INTERFACE ROBOT

Fig. 11.

VR environment of the knife-type device.

Fig. 12.

Reaction forces of the virtual knife.

used a real-time OS (ART-Linux) to guarantee a 1-ms sampling


time.
A. Manipulation of the Knife-Type Device
To evaluate the developed knife-type device, we considered
manipulation of the device in a constraint space.
In the experiment, a virtual plane was made in the VR environment, as shown in Fig. 11. When the virtual knife, which
is in the VR environment, touches the virtual plane, the force
is presented to the operator through the knife-type device. The
desired presented force is calculated by the penetration depth
of the blade edge of the virtual knife. For this, we first set the
several contact points V Ci at the blade edge of the virtual knife,
as shown in Fig. 12. For each V Ci , the force f i is calculated
as f i = f ci + f fi , where f ci andf fi are the constraint and the
friction force, respectively. In the experiment, we set f ci and f fi
as the following: f ci = Kani + Dv ni
f fi =

i f ni  ti + di v ti (in the case of the static friction force)
i f ni  ti + i v ti (in the case of the dynamic friction force)
where the penetration depth vector of the V Ci into the virtual plane is decomposed to a normal directional vector ani
and a frictional directional vector ati , v ni and v ti are the normal and the frictional directional relative speeds between the
V Ci s velocity and virtual plane velocity, respectively, K is the
stiffness of the plane, and D is the damping coefficient of the
plane. Furthermore, i is the coefficient of static friction given
by i = ati  / ani  , di is the damping coefficient, i is the
coefficient of the dynamic frictional force, i is the damping
coefficient at the dynamic friction state, and ti is the unit vector
of the frictional force direction. (For technical details, see [26].)
Then, the force of the virtual knife, F to ol = col[f to ol , nto ol ], is
calculated, where f to ol = fi , nto ol = (b pV C i b pto ol )f i ,
and b pV C i is the position vector of V Ci with respect to b . Finally, using F to ol and (13), we calculate the desired haptic
fingertip forces, and the haptic fingers are controlled by (25) to
accomplish the desired haptic fingertip forces. Here, note that

77

Fig. 13. Time responses of f to o l in the knife-type devices. (a) Previously


developed knife-type device. (b) Newly developed knife-type device.

the aim of this experiment was to investigate the force display


of the device, and thus, we did not consider the cutting of the
virtual plane by the knife.
We compared the results gained with the newly developed
knife-type device, which is shown in Fig. 4, with the results
gained with the previously developed knife-type device, which
is shown in Fig. 2. In both experiments, an operator grasped
the knife-type device and moved the knife-type device toward a
virtual plane. When the virtual knife touched the virtual plane,
the operator felt the reaction force through the knife-type device.
The touching operation was repeated several times during each
experiment. Fig. 13 shows the experimental results. Fig. 13(a)
and (b) shows the responses of the z-axis force f to ol of the
previously and newly developed knife-type device, respectively.
In the figures, a dashed line is the desired value and the solid line
is the measured value. We only show the z-axis force because
the value of the z-axis force is larger than those of the x- and
y-axis forces. First, the operator moved the device toward the
virtual plane, and then, the operator touched the virtual plane
through the device. The average values of the absolute force
error of the previously and newly developed surgical knives
were 0.28 and 0.21 N, respectively. The difference was only
0.07 N, and the force response of the newly developed device
is only slightly better than the force response of the previously
developed device.
We also compared the values of PI of both devices, and we
confirmed that the PI of the previously developed knife-type
device was smaller than the PI of the newly developed one in all
intervals. We believe this was caused by the difference between
the values of WB in the two devices. Although there is no large
difference in the values of WA and Wq in the devices (because
HIRO is controlled so that the value of CPI in (27) becomes
large), the values of WB of the two devices are very different.
The values of WB in the two devices during the aforementioned
experiments are shown in Fig. 14. In this figure, the solid line
shows WB of the newly developed knife-type device, and the
dashed line shows WB of the previously developed knife-type

78

Fig. 14.

IEEE TRANSACTIONS ON ROBOTICS, VOL. 29, NO. 1, FEBRUARY 2013

Time responses of W B in the knife-type devices.

Fig. 15. Norms of fingertip force in the knife-type devices. (a) Previously
developed knife-type device. (b) Newly developed knife-type device.

device. To facilitate visualization, we used the logarithmic scale


in the vertical axis. We see that WB of the previously developed
knife-type device is about 100 times smaller than the WB of
the newly developed knife-type device. WB is related to the
second kind of singularity, and if the value of WB becomes large,
the system is away from the second kind of singularity. Here,
note that WB is also related to the well-known manipulatingforce ellipsoid, which denotes the force transmission from the
contact force f c to the total force/moment vector F to ol applied
on the knife-type device because of (13). The value of WB
indicates how F to ol can be more or less easily produced by
the haptic fingers, and thus, if the system is close to the second
kind of singularity, the resulted haptic fingertip force becomes
large. The norms of the haptic fingertip forces are shown in
Fig. 15. Fig. 15(a) and (b) show the norms of the three-axis
fingertip forces at the thumb, index, and pinky haptic fingers
in the experiments with the previously and newly developed
knife-type device, respectively. In fact, the fingertip forces in
the newly developed device are small values in all intervals and
the fingertip forces in the previously developed device were
large. This shows that the newly developed knife-type device
has good force transmission ability from f c to F to ol .
Note that the maximum value of the force in Fig. 13(b) is only
1 N, but the developed knife device can present greater force to
the operator. The maximum value of the presented force of the
developed knife device is over about 11 N. As an example, we
show the large force response of the developed knife device in
Fig. 16. As shown in the figure, the developed knife device can
present a high level of force. The presentation of about 4.5 N
of force is enough to cut an elastic object [27], [28], and it is
obvious that the developed device can fully display this level of
force.

Fig. 16.

Time responses of large f to o l in the knife-type device.

Fig. 17.

Displayable regions of the KD plane in the knife-type device.

Next, to investigate the displayable stiffness of the developed


knife-type device, we carried out a contact experiment involving a virtual wall [18]. In this experiment, a user grasps the
developed knife-type device. We made a virtual wall by using
a springdamper model, and the desired force at contact point
V Ci at the blade edge of the virtual knife was calculated by
f i = f ci = Kani + Dv ni . Four people in their twenties participated in the experiment. The experimental procedure was as
follows: 1) A damping coefficient D was set; 2) the participant
grasped the knife-type device; 3) the participant enlarged the
stiffness coefficient K from 0 N/m at intervals of 100 N/m; and
4) the participant touched the virtual wall through the knifetype device and moved the device on the surface of the virtual
wall in every case. If the participant felt vibration, the stiffness
coefficient K before one-step was the maximum displayable
stiffness coefficient at the damping coefficient, which was set
in step 1). Then, the participant returned to step 1) and set the
damping coefficient D to the next value. Here, the step size of
D was 10 Ns/m. The displayable stiffness levels for four participants are shown in Fig. 17. In the figure, the region formed
by the D-axis, the K-axis, and each participants curve is the region where the corresponding participant could feel the smooth
surface of the virtual wall with no vibrations. We observed no
large differences among the different subjects curves, and the
maximum displayable stiffness was about 20 kN/m.
B. Manipulation of the Tweezers-Type Device
To evaluate the developed tweezers-type device, we considered the manipulation of the device in a constraint space. Here,
note that we previously did not develop a tweezers-type device,
and thus, we did not consider the difference between a newly developed tweezers-type device and a previously developed one,
as in the case of the experiment with the knife-type device. In
the experiment, a virtual object was made in a VR environment,
as shown in Fig. 18. In the figure, the blue object is the virtual
object. When the virtual tweezers touched a virtual object, the
force was presented to an operator through the tweezers-type

ENDO et al.: DEVELOPMENT OF TOOL-TYPE DEVICES FOR A MULTIFINGERED HAPTIC INTERFACE ROBOT

Fig. 20.

Fig. 18.

VR environment of the tweezers-type device.

Fig. 19. Experimental results for the tweezers-type device. (a) Torque n .
(b) Angle . (c) Times responses of PI. (d) Norms of fingertip forces.

device. Furthermore, when the operator grasped the object with


the tweezers-type device, the torque at the opening-and-closing
axis was presented to the operator through the tweezers-type
device. The presented force and torque were calculated using
the penetration depth of the tips of the tweezers-type device
into the object in the same manner as the case of the knife-type
device. The obtained reaction force F to ol was translated to f c
using (24), and then, f c was applied by controlling HIRO. In
the experiment, the operator held the handle of the tweezerstype device, and grasped the object. Then, the operator lifted
the object and released it.
Fig. 19 shows the experimental results. Fig. 19(a) and (b)

79

Time responses of large torque n in the tweezers-type devices.

shows the responses of the torque n and the angle at the


opening-and-closing axis, respectively. In Fig. 19(a), a solid line
shows the response of n , and a dashed line shows the response
of the desired torque. The average value of the absolute torque
error during the experiment was 0.024 Nm. From the figures,
we see that the absolute value of the torque was large whenever
the operator grasped the object with tweezers, and the operator
felt the grasping sensation of the object through the tweezerstype device. Here, note that the initial value of angle , namely,
the angle in the case that the tweezers has no external forces as
shown in Fig. 8(a), is = 12 . On the other hand, Fig. 19(c)
shows the time response of the performance index, PI. From
the figure, although there were some variations, we see that the
developed tweezers-type device maintained a high PI value, and
we were able to confirm the good operability of the device in the
experiment. Fig. 19(d) shows the norms of the haptic fingertip
forces. From this figure, we can confirm that the haptic fingers
were able to display the required forces.
The maximum value of the torque in Fig. 19(a) is only
0.2 Nm, but the developed tweezers-type device can present
larger torque to the operator. The maximum value of the presented torque of the developed tweezers-type device is over
about 0.39 Nm. As an example, we show the torque response
of the developed device in Fig. 20. From Fig. 20, we can see that
the developed tweezers-type device can present larger torque.
For example, for a suturing task, which closes an incision in
a wound closure pad (i.e., a simulated skin pad), the tweezers
are used to grasp the tissue lips in passing the suture across the
incision. In this case, the presentation of about 3-N force at the
handle of the tweezers is sufficient to accomplish the task [29].
If this value is converted into the torque n , it is 0.32 Nm, and
we can see that the developed tweezers-type device can fully
display the required torque.
Finally, we measured the displayable stiffness of the developed tweezers-type device, like the case of the knife-type device. We made a virtual object by using a springdamper model.
When the operator grasps the virtual object with the tweezerstype device, the desired force at contact point V Ci at the virtual
tweezers was calculated by f i = f ci = Kani + Dv ni . Four people in their twenties participated in the experiment. The experimental procedure was as follows: 1) A damping coefficient D
was set; 2) the participant handled the tweezers-type device; 3)
the participant enlarged the stiffness coefficient K from 0 N/m
at intervals of 50 N/m; and 4) the participant grasped the virtual
object using the tweezers-type device and moved the device on
the surface of the virtual object in every case. Here, note that the
virtual object was fixed in the environment, and the object did

80

Fig. 21.

IEEE TRANSACTIONS ON ROBOTICS, VOL. 29, NO. 1, FEBRUARY 2013

Displayable regions of the KD plane in the tweezers-type device.

not move. If the participant felt the vibration, the stiffness coefficient K before one-step was the maximum displayable stiffness
coefficient at the damping coefficient, which was set in step 1).
Then, the participant returned to step 1) and set the damping
coefficient D to the next value. Here, the step size of D was
5 Ns/m. The displayable stiffness levels for four participants
are shown in Fig. 21. We observed no large differences among
the different subjects curves, and the maximum displayable
stiffness was about 6 kN/m.
VII. CONCLUSION
We have described a knife-type device and a tweezers-type
device that we developed for the multifingered haptic interface
robot HIRO. The knife-type device represents a tool device
with no joints, and the tweezers-type device represents a tool
with joints. To determine the optimal connection between HIRO
and the tool-type devices, we have proposed the optimal connection method from mobility and singularity points of view.
After we analyzed the kinematics of the knife-type device and
the tweezers-type device, we have developed the devices for
HIRO that satisfied the optimal connections and tested them experimentally. In the experiment with the knife-type device, we
have compared the newly developed knife-type device with a
previously developed one and found that the newly developed
knife-type device has good force transmission ability. In the experiment with the tweezers-type device, we have confirmed that
the operator feels the grasping sensation of the object through
the tweezers-type device, and device has good operability. These
results show the great potential of our tool-type haptic interface
system that is able to present the force sensation of many tooltype devices, as well as the validity of the proposed connection
method.
In this paper, we have developed a knife-type and a tweezerstype device. However, human beings work with many tools. For
example, in the medical field, there are many tools of different shapes and uses, such as surgical knives, scissors, syringes,
and many others; therefore, a haptic interface that can create
the force sensation of many different tools is important for virtual training systems. We plan to develop many other tool-type
devices for use in a virtual training system in future.
REFERENCES
[1] K. Hale and K. Stanney, Deriving haptic design guidelines from human physiological, psychophysical, and neurological foundations, IEEE
Comput. Graph. Appl., vol. 24, no. 2, pp. 3339, Mar./Apr. 2004.

[2] V. Hayward, O. Astley, M. Cruz-Hernandez, D. Grant, and G. Robles-DeLa-Torre, Haptic interfaces and devices, Sens. Rev., vol. 24, pp. 1629,
2004.
[3] B. Hannaford and A. Okamura, Haptics, in Handbook of Robotics,
B. Siciliano and O. Khatib, Eds. New York: Springer, 2008, pp. 719
739, ch. 30.
[4] N. Seymour, A. Gallagher, S. Roman, M. OBrien, V. Bansal, D. Andersen,
and R. Satava, Virtual reality training improves operating room performance: Results of a randomized, double-blinded study, Ann. Surg.,
vol. 236, pp. 458464, 2002.
[5] P. Youngblood, S. Srivastava, M. Curet, W. Heinrichs, P. Dev, and S. Wren,
Comparison of training on two laparoscopic simulators and assessment
of skills transfer to surgical performance, J. Am. Coll. Surgeons, vol. 200,
pp. 546551, 2005.
[6] D. Morris, T. Hong, F. Barbagli, T. Chang, and K. Salisbury, Haptic
feedback enhances force skill learning, in Proc. WorldHaptics, 2007,
pp. 2126.
[7] A. Okamura, R. Webster III, J. Nolin, K. Johnson, and H. Jafry, The
haptic scissors: Cutting virtual environments, in Proc. IEEE Int. Conf.
Robot. Autom., Sep. 2003, pp. 828833.
[8] M. Mahvash, L. Voo, D. Kim, K. Jeung, J. Wainer, and A. Okamura,
Modeling the forces of cutting with scissors, IEEE Trans. Biomed. Eng.,
vol. 55, no. 3, pp. 848856, Mar. 2008.
[9] D. Sato, R. Kobayashi, A. Kobayashi, S. Fujino, and M. Uchiyama, Soft
tissue pushing operation using a haptic interface for simulation of brain
tumor resection, J. Robot. Mechatronics, vol. 18, pp. 634642, 2006.
[10] S. Fujino, D. Sato, K. Abe, A. Konno, and M. Uchiyama, Displaying
feeling of cutting by a micro-scissors type haptic device, in Proc. IEEE
Int. Conf. Robot. Autom., May 2008, pp. 20672072.
[11] O. Goksel, K. Sapchunk, and S. Salcudean, Haptic simulator for prostate
brachytherapy with simulated needle and probe interaction, IEEE Trans.
Haptics, vol. 4, no. 3, pp. 188198, May/Jun. 2011.
[12] Surgical Science Ltd., LapSim. (2012). [Online]. Available: http://www.
surgical-science.com/productsmain/
[13] Simbionix, LAP mentor. (2012). [Online]. Available: http://simbionix.
com/simulators/lap-mentor
[14] H. Kawasaki, T. Mouri, S. Ikenohata, Y. Otsuka, and T. Endo, Multifingered haptic interface robot handling plural tool devices, in Proc.
WorldHaptics, 2007, pp. 397402.
[15] T. Endo, S. Tanimura, and H. Kawasaki, Development of a surgical knife
device for a multi-fingered haptic interface robot, in Proc. 18th Int. Feder.
Autom. Control World Congr., 2011, pp. 64606465.
[16] T. Endo, S. Tanimura, and H. Kawasaki, Development of a tweezers-type
device for a multi-fingered haptic interface robot, in Proc. IEEE/SICE
Int. Symp. Syst. Integr., Dec. 2011, pp. 10061011.
[17] T. Endo, H. Kawasaki, T. Mouri, Y. Ishigure, H. Shimomura,
M. Matsumura, and K. Koketsu, Five-fingered haptic interface robot:
HIRO III, IEEE Trans. Haptics, vol. 4, no. 1, pp. 458463, Jan./Feb.
2011.
[18] J. Colgate and J. Brown, Factors affecting the z-width of a haptic display,
in Proc. IEEE Int. Conf. Robot. Autom., May 1994, pp. 32053210.
[19] K. Hunt, Kinematic Geometry of Mechanisms. Oxford, U.K.: Oxford
Univ. Press, 1978.
[20] L-W. Tsai, Robot Analysis: The Mechanism of Serial and Parallel Manipulators. New York: Wiley, 1999.
[21] J. Merlet and C. Gosselin, Parallel mechanisms and robots, in Handbook
of Robotics, B. Siciliano and O. Khatib, Eds. New York: Springer, 2008,
pp. 269285, ch. 12.
[22] X. Liu, J. Jeong, and J. Kim, A three translational DOFs parallel cubemanipulator, Robotica, vol. 21, pp. 645653, 2003.
[23] C. Gosselin and J. Angles, Singularity analysis of closed loop kinematic
chains, IEEE Trans. Robot. Autom., vol. 6, no. 3, pp. 281290, Jun. 1990.
[24] Z. Li and S. Sastry, Task-oriented optimal grasping by multifingered
robot hands, IEEE J. Robot. Autom., vol. 4, no. 1, pp. 3244, Feb. 1988.
[25] H. Kawasaki and T. Mouri, Design and control of five-fingered haptic
interface opposite to human hand, IEEE Trans. Robot., vol. 23, no. 5,
pp. 909918, Oct. 2007.
[26] H. Kawasaki, Y. Ohtuka, M.O. Alhalabi, and T. Mouri, Haptic rendering
and perception of frictional moment, in Proc. EuroHaptics Conf., 2006,
pp. 201206.
[27] M. Mahvash and V. Hayward, Haptic rendering of cutting: A fracture
mechanics approach, Haptics-e, vol. 2, no. 3, 2001.
[28] T. Chanthasopeephan, J. P. Desai, and A. C. W. Lau, Measuring forces
in liver cutting for reality-based haptic display, in Proc. IEEE/RSJ Intl.
Conf. Intell. Robots Syst., Oct. 2003, pp. 30833088.

ENDO et al.: DEVELOPMENT OF TOOL-TYPE DEVICES FOR A MULTIFINGERED HAPTIC INTERFACE ROBOT

[29] A. Rafiq, F. Tamariz, C. Boanca, V. Lavrentyev, and R. C. Merrell, Objective assessment of training surgical skills using simulated tissue interface
with real-time feedback, J. Surg. Edu., vol. 65, no. 4, pp. 270274, 2008.

Takahiro Endo (M06) received the Dr. Eng. degree from the Tokyo Institute of Technology, Tokyo,
Japan, in 2006.
Since April 2006, he has been with the faculty
of Engineering, Gifu University, Gifu, Japan, where
he is currently an Assistant Professor. His research
interests include haptic interfaces, robotics, and the
control of infinite dimensional systems.

Satoshi Tanimura received the B.S. degree in human and information systems engineering from Gifu
University, Gifu, Japan, where he is currently working toward the M.S. degree in human and information
systems engineering.
His research interests focus on haptic interfaces in
virtual reality and robot control.

81

Haruhisa Kawasaki (M91SM10) received the


M.S. and Dr. degrees from Nagoya University,
Nagoya, Japan, in 1974 and 1986, respectively.
He is currently a Professor with the Faculty of Engineering, Gifu University, Gifu, Japan. From 1974 to
1990, he was a Research Engineer with NTT Laboratories. From 1990 to 1994, he was a Professor with the
Kanazawa Institute of Technology, Kanazawa, Japan.
From July 1998 to January 1999, he was a Visiting
Professor with the University of Surrey, Surrey, U.K.
His research interests are in the areas of robot control,
humanoid robot hands, haptic interfaces in virtual reality, and computer algebra
of robotics.
Dr. Kawasaki has contributed to the community as a member of many organizations, such as the Japan Society of Mechanical Engineers (JSME), the
Robotics Society of Japan (RSJ), the Society of Instrument and Control Engineers, and the Virtual Reality Society of Japan. He is a Fellow of JSME and RSJ.
He has received several awards, such as the Best Paper Award from the World
Automation Congress in 2004 and the Prizes for Science and Technology from
the Commendation for Science and Technology by the Minister of Education,
Culture, Sports, Science and Technology of Japan in 2006, as well as JSME
Funai Award in 2009. He was the National Organizing Committee Chair of
the Ninth International Federation of Automatic Control Symposium on Robot
Control in 2009.

You might also like