Tsuneo Yoshikawa

Automation Research Laboratory

Manipulability of
Kyoto University
Uji, Kyoto 611, Japan
Robotic Mechanisms

Abstract interested in can be describedby m variables fi, j =

1,
2, ... , m (m ; n) and that the relation between Oi
This paper discusses the manipulating ability of robotic and ri is given by
mechanisms in positioning and orienting end-effectors and
proposes a measure of manipulability. Some properties of this
measure are obtained, the best postures of various types of
manipulators are given, and a four-degree-of-freedom finger
is considered from the viewpoint of the measure. The pos-
where B =
[0,, 82, ... , 8n]T is the joint vector,
tures somewhat resemble those of human arms and fingers.
r [r, , r2 , ... , rm ] T is the manipulation vector, and
=

the superscript T denotes the transpose. The manipu-

lation velocity r is related to the joint velocity 6 by
1. Introduction

Determination of the mechanism and size of a robot

manipulator at the design stage and determination of
where r E Rm (m-dimensional Euclidean space),
the posture of the manipulator in the workspace ~ d0/dt E ~&dquo;, and J(0) E R’nxn (the set of all m X n
=

real matrices). The matrix J(O) is called the Jacobian.

(Gupta and Roth 1982) for performing a given task at
the operation stage have been done largely on the We assume that the following condition is satisfied:
basis of experience and intuition. One of various mea-
sures used for these determinations seems to be the
ease of changing arbitrarily the position and orienta-
tion of the end-effector at the tip of the manipulator. It Failing to satisfy this condition usually means that the
will be beneficial for design and control of robots and selection of manipulation variables is redundant and
for task planning to have a quantitative measure of that the number of these variables m can be reduced.
manipulating ability of robot arms in positioning and When condition (3) is satisfied, we say that the degree
orienting the end-effectors. The concept &dquo;manipula- of redundancy of this manipulator is (n - m). More
bility measure&dquo; has been proposed in a previous paper detailed discussion on the degree of redundancy and a
(Yoshikawa 1983) as one such measure. related concept of redundant space is given in Hana-
In this paper, some properties of this measure and fusa, Yoshikawa, and Nakamura ( 1981 ).
its utilization for determining the best postures of If, for some 0*,
various types of manipulators and of an articulated
robot finger are discussed.

2. then we say that the manipulator is in a singular state.

Manipulability Measure This state 0* is not desirable because the manipulation
vector r cannot move in a certain direction, meaning
We consider a manipulator with n degrees of freedom
whose joint variables are denoted by Oi, i = 1, that the manipulability is seriously deteriorated.
To analyze this problem, Yoshikawa (1983) pro-
2, ... n. We assume that a class of tasks we are
,

posed the following quantitative measure of manipula-

bility.
Definition: A scalar value w given by

3
is called the manipulability measure at state 0 with Therefore, w is equal to the volume of the manipula-
respect to manipulation vector r. bility ellipsoid except for the constant coefhcient d.
The following three facts have been established in 3. When m =
n, that is, when we consider nonre-
Yoshikawa ( 1983). dundant manipulators, the measure w reduces to
1. Let the singular value decomposition (KJema and
Laub ( 1980) of J be

This type of measure has been used by Paul and Ste-

venson( 1983) for analysis of robot wrists.
where U E Rmxm and V E Rnxn are orthogonal matri- Now we establish several new properties of the mea-
ces and sure.
When m =
n, the measure has the following physical
interpretation as well as that of number 2 above. The
subset of the realizable velocity r such that

is a parallelepiped in Rm, and its volume is 2mw. In

other words, the measure w is proportional to the
volume of the parallelpiped. This result can easily be
obtained from a property of determinants.
Next we discuss the relation between the measure w
and the maximum velocities of joints. So far, we have
implicitly assumed that the maximum velocities of all
Then the measure w can be expressed as the product joints are the same. When this assumption does not
of the singular values a&dquo; (j2,
...am : , hold, the velocities of joints should be normalized.
After fixing a set of units for distance, angle, and time
(for example, m, rad, s), we denote the maximum
(angular) velocity of joint i by 8 iO . We also select the
2. We can show that the subset Sv of the realizable desirable maximum (angular) velocity of each manip-
velocity /’ in the space Rm using joint velocity 6
such ulation variable ’jO taking into consideration the class
that I 0 ~~2 = aT 1 + 62 2 + + 02 n 1 is an ellipsoid of tasks the manipulator is supposed to perform.
with principal axes (JI&dquo;¡’ Q2u2, ... , CT.U,,, where Then, letting
ui E Rm is the ith column vector of U, that is,
[u,u2 ’ ’ ’ urn] U. This ellipsoid can be called the
=

manipulability ellipsoid and could be a good means for

the analysis, design, and control of robot manipulators.
The volume of this ellipsoid is given by
we obtain

where the constant d is given by

where

4
 Fig.1. Two-joint link mech-
anism.

Since lê¡1 ~ 1 and 101 ~ 1, we can define the manipula-

bility measure w using the normalized Jacobian i(o).
Defining the measure w for 1(8) T(j’ 1 as W 8, and the
measure for J(8) T~ J(0) Tgias w, we have
=

and, especially when n =

m, we have

Hence the transformation ( 18) has only the effect of

multiplying the scalar value fl fi (1/rjo), The relative
shape of w as a function of 0 is independent of the Therefore, the manipulator takes its best posture when
transformation T,. Furthermore, when n m, the rel-
= 02 ± 90 °, for any given values of £ I’ f 2, and 01. If
=

the lengths I , and £2 can be specified under the condi-

ative shape of w is independent of both Tr and Two.
tion of constant total length, that is, l + £2 = con-
stant, the manipulability measure attains its maximum
3. Best Postures of Various Robotic when £1 e2 for any given 8, and 82.
=

Mechanisms from the Viewpoint of When the human arm is regarded as a two-joint link
mechanism by neglecting the degree of freedom of
Manipulability sideward direction at the shoulder and the degree of
3.1. TWO-JOINT LINK MECHANISM freedom of the wrist, it approximately satisfies the
relation t Q2. Moreover, when we handle some ob-
=

In this section, we calculate the manipulability mea- ject with our hands, the elbow angle is usually in the
sure for various robotic mechanisms and determine neighborhood of 90 ° . Hence it could be said that peo-
the best postures and the best points in the workspace ple are unconsciously taking the best arm postures
of these mechanisms from the viewpoint of manipula- from the viewpoint of manipulability.
bility. These will be called the optimal postures and
the optimal working positions.
Consider the two-joint link mechanism shown in 3.2. SCARA-TYPE ROBOT MANIPULATORS
Fig. 1, which is the simplest case of multijoint manip-
ulators. When the hand position [x, y] T is taken as the Consider the SCARA-type manipulator with four
manipulation vector r, the Jacobian matrix is given by degrees of freedom shown in Fig. 2. Let r [x, y, z,
=

a] T, where [x, y, z] T is the hand position and a is the

rotational angle of the hand about Z-axis. The Jaco-
bian matrix for this case is given by

where c, =
cos01 c12 cos (01 + 0~), sl sin B1,
= =

s~2
=
sin (01 + B2). Hence the manipulability measure
w is

5
Fig. 2. SCARA-type robot. Fig. 3. PUMA-type robot.

Hence

Therefore, as in the case of two-joint link mechanism

of Fig. 1, the best posture is given by 9z ± 90 ° , for
=

any given values of e I’ e 2, 01, and 83, and 84, Also

under the constraint of e + ~2 constant, the ma-
=

nipulability measure attains its maximum when

e1 = 2z , 02 = ± 90 ° . Notice that there are many com-
mercial SCARA-type manipulators satisfying e 1 = e 2’
and the manipulability measure is

3.3. PUMA-TYPE ROBOT MANIPULATORS

Most of the commercially available PUMA-type robot The best posture for given ~2 and ~3 is obtained as
manipulators have five or six degrees of freedom. follows. First, 01 is not related to w and can take any
Many of them have links with some displacements in value. Second, from aw~ae2 0 we have
=

the direction of joint axes. However, we consider only

the main three joints shown in Fig. 3, neglecting the
degrees of freedom placed at the wrist and neglecting
the displacements in the direction of joint axes. The
joint vector is 0 [01, 82, B3] T. The manipulation
=
This that the tip of the arm should be on the X,
means
vector is taken to be r =
[x, y, z] r’. Then the Jacobian Y-plane, that is, at the same height as the second joint.
matrix is This can further be interpreted as maximizing the

6
Fig. 4. Best arm posture far Fig. 5. Finger witJ2 four
PUMA-type robot. degrees of freedom.

contribution of the angular velocity of the first joint to

the manipulability measure.
Substituting Eq. (27) into Eq. (26) yields

The value of 03 that maximizes w is

polar coordinate manipulators, the best posture is

attained on the boundary of their workspace. Although
Figure 4 shows the best postures for the cases e3 ye 2’
=
this is inconvenient, it is true for all robotic mecha-
y =
0. 5, 1, 2 (only those satisfying 0 ° = 82 = 90 are nisms that have a prismatic joint whose axis is in the
shown in the figure). If the manipulator is regarded as radial direction of a rotational joint. This is also intui-
a two-joint mechanism consisting Of 02 and B3, the tively understandable. For these manipulators, consid-
optimal angle for e3 is 90°, from the discussion in Sec- erations other than the manipulability measure will be
tion 3. l. In the present case, however, the optimal 03 necessary to determine the best working position.
is smaller than 90 because the contribution of 61 to w
can be made larger by placing the tip of the arm far-
ther from the first joint axis. 3.5. FOUR-JOINT ROBOTIC FINGER

Various robotic hands with multi-articulated fingers

3.4. ORTHOGONAL, CYLINDRICAL, AND POLAR have been developed (Okada 1980; Salisbury and Craig
COORDINATE MANIPULATORS 1982; Hanafusa, Yoshikawa, and Nakamura 1983) to
realize the dexterity and flexibility of human hands in
Only the main three degrees of freedom of manipula- handling and assembling jobs. In this section the four-
tors and the hand position are considered in this sec- joint finger shown in Fig. 5 is considered from the
tion, as in the previous section. viewpoint of the manipulability measure.
The manipulability measure w of orthogonal coordi- The Jacobian matrix relating 0 = [~;, ~, ~3, ~4] ~ to
nate manipulators is 1 everywhere in the workspace. r =
[x, y, Z] T is
The best posture of cylindrical coordinate manipula-
tors is attained when the arm is fully stretched out.
Similarly, the best posture of polar coordinate manip-
ulators is attained when the arm is fully stretched out
in the horizontal direction. Hence, for cylindrical and

7
Fig. 6. Maximum value of w Fig. 7. Best finger posture.
as a function off,,.

where

The manipulability measure is calculated as

where

The maximum value of w(82, 93, 0,) for a given

distance t between joint 2 and the tip of the finger in
Fig. 5 is shown in Fig. 6 for the case t 2 = e 3 = 0.4, Salisbury and Craig (1982) have used the condition
f4 0.3. The corresponding finger posture is shown in
=
number c(lT(8» of the transpose the Jacobian ma-
of
Fig. 7. Notice that these postures are independent of trix J(B) as a measure of workspace quality. This is a
the angle 82. Figure 8 shows the maximum value of w measure of the accuracy with which forces can be
as a function of the finger tip position in the X,Z’- exerted. Hence, the manipulability measure w dis-
plane (only the lower half-portion is shown since the cussed in this paper is quite different from the condi-
value in the upper half is symmetric with respect to tion number. For example, Fig. 9 compares the best
the X-axis). The best finger posture is also shown in postures of a simple two-joint link mechanism for
Fig. 8 by a broken line. manipulability measure and for condition number.
Notice that the finger postures given in Fig. 7 are The figure clearly shows the difference between the two
quite similar to those taken by human fingers during measures. Note, however, that the condition number
the manipulation of small objects. Hence, we can can also be interpreted as a measure of directional

expect that these postures would be useful in deter- uniformity of the manipulability illipsoid. Further
mining the grasping posture of a robotic hand with study is necessary on this aspect of the condition
several four-degree-of-freedom fingers. number.

8
Fig. 8. Maximum value of w Fig. 9. Comparison of the
as a function of the finger tip best postures of two-link
position. mechanism. (a) c (JT ); (b) w.

Paul, R. P. and Stevenson, C. N. 1983. Kinematics of robot

wrists. Int. J. Robotics Res. 1(2):31-38.
Salisbury, J. K., and Craig, J. T. 1982. Articulated hands:
Force control and kinematic issues. Int. J. Robotics Res.
1(1):4-17.
4. Conclusion Yoshikawa, T. 1983 (Aug. 28-Sept. 2, Bretton Woods,
N.H.). Analysis and control of robot manipulators with
Properties of the manipulability measure, which was redundancy. Preprints 1st Int. Symp. Robotics Res.
proposed in a previous paper (Yoshikawa 1983) as a
measure of manipulating ability of robotic arms in
positioning and orienting end-effectors, have been
studied. Utilization of the manipulability measure for
determining the best postures of various types of ma-
nipulators and an articulated robot finger has been
discussed. The best postures obtained have some re-
semblance to those taken by human arms and fingers.
These postures will be useful for planning the working
position of robots for various tasks.

REFERENCES

Gupta, K. C., and Roth, B. 1982. Design considerations for

manipulator workspace. Trans. ASME, J. Mech. Design
104(4):704-711.
Hanafusa, H., Kobayashi, H., and Terasaki, N. 1983. Fine
control of the object with articulated multi-finger robot
hands. Proc. 1983 Int. Conf. Adv. Robotics. Tokyo: Robo-
tics Society of Japan, pp. 245-252.
Hanafusa, H., Yoshikawa, T., and Nakamura, Y. 1981.
Analysis and control of articulated robot arms with redun-
dancy. Prep. 8th IFAC World Congress, vol. 14. Kyoto:
IFAC, pp. 78-83.
Klema, V. C., and Laub, A. T. 1980. The singular value
decomposition: Its computation and some applications.
IEEE Trans. Automatic Contr. AC-25(2):164-176.
Okada, T. 1980. Analysis of finger tip motion for precise
object-handling. Trans. SICE 16(4):597-602. In Japanese.