See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3257723

Symbolic Derivation of Dynamic Equations of Motion for Robot Manipulators

Using Piogram Symbolic Method

Article in IEEE Journal on Robotics and Automation · January 1989

DOI: 10.1109/56.9298 · Source: IEEE Xplore

CITATIONS READS

26 958

3 authors, including:

Cheng-I Weng
National Cheng Kung University
153 PUBLICATIONS 3,235 CITATIONS

SEE PROFILE

All content following this page was uploaded by Cheng-I Weng on 15 March 2016.

The user has requested enhancement of the downloaded file.

IEEE JOURNAL OF ROBOTICS AND AUTOMATION, VOL. 4, NO. 6 , DECEMBER 1988 599

Symbolic Derivation of Dynamic Equations of

Motion for Robot Manipulators Using
Piogram Symbolic Method
PI-YING CHENG, CHENG-I WENG, AND CHAO-KUANG CHEN

Abstruet-In this paper, an algorithm for both manual and automatic INTRODUCTION
derivation of dynamic models of robotic manipulators using Piogram
symbolic method is presented. A program is also developed based on the
Newton-Euler formalism by Piogram symbolic representation method,
which is applicable to the manipulators of any degree of freedom. Two
T HE SYMBOLIC derivations have been extensively used
in research to generate and evaluate control algorithms for
multi-link robots. Generally, a symbolic tool is more powerful
examples are given to illustrate how to use this program for dynamic than a numerical one. From the design viewpoint, symbolic
equation generation. It is shown the present method has the great
advantage of simplifying the manipulation process, and thus a large
equations of motion can be easily analyzed in order to check,
memory space and much computing time are saved as implemented in a for instance, the influence of design parameters, such as
computer program. masses and lengths, on the model. Furthermore, in the usual
considered cases, a simulation run based on a symbolic (exact)
NOMENCLATURE model is faster in obtaining the result than one based on a
numerical model. A lot of algorithms for generating the
4 x 4 D-H transformation matrix from equations of motion of robots have been presented. Many of
coordinate system a to b about k , ( = k b ) , them, e.g., ARM, MACSYMA, etc., [8], [9], try to generate
j a (=j b ) , ia( = i b ) , respectively. equations of motion in symbolic form automatically. Never-
4 x 4 D-H transformation matrix from theless, most of the programs reveal an elevated memory and/
coordinate system k to k - 1 . or are time-consuming. The reason is that successive computa-
Total external vector force exerted on link tions of the coordinate transformations and of the position,
N. velocity, and acceleration of the last link with respect to the
Angular momentum vector of link N with manipulator base or any local coordinate system is required.
respect to its center of mass. Thus the cumbersome matrix multiplication induces the
Inertia matrix of link N about its center of tedious operations of sine and cosine functions for the
mass. transformations. Besides, the algebra expressions obtained by
Appropriate mass coefficient due to actua- symbolic manipulation need to be simplified to reduce the
tor. arithemetic operation terms. Therefore, most of those pro-
Unit vectors along the x , y , and z axes of grams require a large computer memory space and much
coordinate system a, respectively. computing time. Furthermore, in formulating the dynamic
Total external vector moment exerted on equations, the number of chains of multi-link mechanisms
link N . representing the velocity must grow linearly with the number
Column matrix whose entries are the i, j , of rotations, and the number of chains representing the
and k components of robor rc, coordinated acceleration must grow quadratically [12], [13], [17]. This is
in system c. unavoidable since the complexity of the problem and the
Column matrix whose four entries are the i, number of terms in the analytic expressions have a similar
j , and k components of rab or rc and 1 , growth. As for the algorithm presented in this study, it has the
coordinated in system c. advantage of eliminating the cumbersome matrix multiplica-
4 x 4 D-H homogeneous transformation tion. Moreover, in computation of velocity and acceleration,
matrix from the coordinate system k to the using matrix operations, many terms are computed but are
base frame. eventually dropped. This unnecessary computation is avoided
x, y , and z component of ra in the coordinate by using Piogram symbolic operation rules [ 121, [131.
system a, respectively. Formulation of manipulator dynamics in relation to the
Euler angle of rotation about j , , i,, and k, , computational efficiency and control analysis has been an
respectively. active research topic. Newton-Euler [3], [7] and Lagrange
Abbreviations of sin (ei), cos (6;). [l], [ 5 ] , [14] formalisms are the two well-known formulation
methods. The former has been regarded to be computationally
Manuscript received September 29, 1987; revised February 15, 1988. efficient and the latter to be insightful in representing
The authors are with the Department of Mechanical Engineering, National
Cheng Kung University, Tainan, Taiwan, ROC. manipulator dynamics. If the vectodmatrix equations were
IEEE Log Number 8821744. expanded symbolically to scalar form, the resultant equations

0882-4967/88/12OO-0599$01.OO 0 1988 IEEE

600 IEEE JOURNAL OF ROBOTICS AND AUTOMATION, VOL. 4, NO. 6, DECEMBER I988

Fig. 2. Configuration of a general-type robot.

mations rotating about X k , Y k , z k axes, denoted as R t -

P f A f - respectively. It is necessary to transform vectors
from one frame to another in cases where the alignment
between frames cannot readily be achieved by a simple
rotation about an axis of one of the frames. In such cases it is
possible to choose convenient intermidiate frames which can
be readily aligned by simple rotations. Considering the robot
system, as shown in Fig. 2 , the kinematics can be described
'k-I using homogeneous coordinate transformations. Based on
Denavit-Hartenberg method [4], [ 141, a coordinate frame is
assigned to each link, and the transformation from the
coordinate frame of link k - 1 to that of link k is expressed by
a series of simple translation and rotation transformations as
follows:
1) Rotate about the Zk-]-axis an angle e k to line up the
direction of the Zk- ,-axis with the direction of the zk-axis.
2 ) Translate along the Zk- ]-axis a distance dkand then along
the new xk-axis a distance akto bring the origin of frame k -
1 to the origin of frame k.
Fig. 1. Symbolic representation of all three D-H matrices. (a) R f - (b) 3) Rotate about the new xk-axis an angle (Yk to line up the
pi-,. direction Xk - I -axis with the direction xk-axis.
Four parameters d k , d k , ak, and (Yk are the link-geometric
parameters associated with link k. The variables are e k for a
from the Newton-Euler and Lagrange formalisms would be revolute joint and d k for a prismatic joint. The transformation
equivalent. The present paper adopts the Newton-Euler between the two adjacent coordinate frames of links k - 1 and
formalism in deriving the dynamic equations. Two typical k can be expressed by Piogram symbolic representation Bft-
examples are also presented as illustration. as shown in Fig. 3. The four inputs are three components of
r y and 1, and outputs are three components of rft:isb and 1.
KINEMATICS
OF LINKBASED
ON PIOCRAM The dots on the lines in symbolic representations are the
Manipulator kinematics can be described by using homoge- summation points. To obtain the value of the output rib, one
neous coordinate transformations. Based on the Denavit- has to propogate all of the inputs as well as all of the vertical
Hartenberg (D-H) method [4], a coordinate frame is assigned entries to each output through all possible paths using Piogram
to each link, and the transformation from the coordinate frame rules as
of link k - 1 to that of link k is expressed by a series of simple
translation and rotation transformations. The 4 x 4 matrix of
Denavit and Hartenberg is a symbolic notation which permits
the complete description of the kinematic properties of all where
lower pair mechanisms by means of equations. In this paper,
we propose Piogram symbolic representations of the 4 x 4
matrices, as shown in Fig. 1, with a translation [xk- Y k - means transformation from the coordinate frame of an
z k - 1 1 and followed by three Euler angle coordinate transfor- arbitrary link j - 1 to that of link i. The transformation Th,
CHENG el al.: EQUATIONS OF MOTION FOR ROBOT MANIPULATORS 60 1

yib BTy yk-l

k-1.b

kb k-1.b

-
k 'k-1

xk-l,b
k-1
-4k lxkb

STOP?

EXIT

1 1 I 1
(b)
Fig. 3. (a) Piogram symbolic representation: Bit-,. (b) Inversion of
symbolic representation: Bit-,.

abbreviated as Ti, describes the position and orientation of the

coordinate frame i with respect to the base coordinate frame. forward
dynamic (store
Note that the Piogram symbolic representation of Fig. 3(a) is
invertible; thus the inversion of B t - l is shown in Fig. 3(b).
Both of the Piogram operation patterns can be generated from
block a to e, as shown in Fig. 4. Using the Piogram symbolic
representation method, we do not have to operate a matrix
multiplication such as (1) and (2).For general six-link robots,
I I-
the displacement equation
OII 1'1'11'1'
T6 'r:b= 'r:b (3) monitor

can be expressed in Piogram symbolic form, as shown in Fig.

5. Since the link offsets such as ak or dkwill not influence the
orientation vectors, we set that
Fig. 4. Flowchart of PIODYM-I.
dk = 0
ak= (4)
for implementing the orientation vector transformation; and 'rGN and 'r:N. The differentiations of B t - , and Bf-I are
Fig. 5 is simplified as in Fig. 6. In general, a vector specified defined as Figs. 7 and 8. Thus we can completely derive each
by a column matrix has its time derivative determined by chain which is corresponding to each term of (5) symbolically
differentiating each element of the matrix with respect to time and numerically, and then sum them up to obtain the
only if it is referred to the frame used for its measurement. expressions of vector 'r:N.
Thus To obtain the acceleration of point PN with respect to point
'fON= T:(BlB2B3
0 1 2
... ',.EN 0, coordinated in system k, one can differentiate the
expression for the velocity given in (5) once. This, however,
+BAB: * BE-, 'rEN can also be accomplished directly by using the symbolic
+... + BAB: - BE- 'rEN representation. To show how this can be done, we differentiate
(5) once with respect to time and yield the following
+BAB: - * BE-l ' r E N ) (5) expression:
where
!YON= T:(B;B:B: BE-1 'rEN
T i = ( T,k)- I =(Tt)T. (6)
It is obvious that N u , the number of chains representing
+ BAB: . * BE-l ' r g N
velocity, is given by + . . . +BABT 'YEN
N u = n +1 (7) +2B,!,B: * BE-l 'rEN
where n is the number of rotations between the two vectors +.-*+2BAB: BE-l ' r g N ) . (8)
602 IEEE JOURNAL OF ROBOTICS AND AUTOMATION, VOL. 4, NO. 6, DECEMBER 1988

11 I 1
Fig. 5. Position form of symbolic equation.

xLblp
1+ I1
Fig. 6. Orientation form of symbolic equation.

ak
;k-1,
k-1 b

kb .k-l,b
k' 'k-1

kb .k-l,b
'
k --"k-l

l t - 4 0 1 + - - l o
(a) (b)
Fig. 7. Symbolic representation of the first derivative of E:- I (a) for a revolute joint, (b) for a prismatic joint.

k-1.b .kb
Xk-l Xk

k-l,b
'k-1

k-1.b .kb k-1.b- . kb

'k-1 k 'k-1 k

I C 4 0 I + - - 4 0
(a) (b)
Fig. 8. Symbolic representation of the first derivative of (a) for a revolute joint, (b) for a prismatic joint.

It shows that N u , the number of terms in the acceleration follows:

expression, is given by

if link N is rotational
Nu= (n + l)(n+ 2)/2 (9)

where n is defined the same as in (7). To find a symbolic if link N is translational

representation of (8), we first have to define, symbolically, the
second derivative of B t - l as shown in Fig. 9, in which
Piogram operation rules also apply. The components of the
acceleration are finally obtained when the corresponding
outputs of the Nu chains are summed up after multiplying the
appropriate terms by two as indicated in (8).
c=i CjN-l ,

We also define angular velocity and angular acceleration as if link N is translational

CHENG et a l . : EQUATIONS OF MOTION FOR ROBOT MANIPULATORS 603

4 0 l + - 0
(a) cp,
Fig. 9. Symbolic representation of the second derivative of B : _ , (a) for a revolute joint, (b) for a prismatic joint.

XO

Fig. 10. Forces and moments exerted on and from link i .

where OF(GF) is the angular velocity (acceleration) of link N Since the components of IcN are constantly relative to a
with respect to the base coordinate referred to as coordinate k , corotational basis, thus
and 2 N - l = [0, 0, 1, 1IT. Using the Piogram symbolic
representation method, the unnecessary computations of the (f$F)N+ 1 = H E = I c N G E . (16)
kinematic equations as mentioned above are avoided, since For the Newton-Euler equations we need the resultant force
each term, which eventually reaches a zero value, can easily FN and the momentum M c N of the force about the center of
and directly be eliminated before computation, and it is also mass of each link. As shown in Fig. 10, let f N be the force
clearly seen from the symbolic representation that the path applied to link N by link N - 1, and let -f N + I be the force
which involves these terms quickly reaches a zero end. applied to link N by link N + 1, then
EQUATIONS
OF MOTION
FOR MANIPULATORS
FN=fN+ (-fN+ 1) +m N g o = f N - f N + I +m N B k g . (17)
D’Alembert’s principle is applied to each link of the
manipulator. Assume the viscous damping terms are omitted. Also let MN be the moment applied to link N by link N + 1
Thus the Newton-Euler equations are expressed as and - MN+I be the moment applied to link N by link N + 1 ,
then
FN=mN ‘?kN (12)
MCN=MN-M N + I - ( S N - T N ) X ~ N +~ N X ~ IN. + (18)
McN = HkN. (13)
By projecting M N onto the direction of zN- axis, we obtain
For a rigid body, the angular momentum vector is expressed as
the input torque TN on the Nth link. There is, however,
HE= I N W E . (14) another important contribution to kinematic energy, that of the
actuators at the joints. Hence, we adopt the terms represented
The rate of change of the angular momentum of link N relative by an actuator inertia (or equivalent mass), referred to joint
to the base frame referred to coordinate k is angular acceleration (or linear acceleration) in the actuated
HzN=B;(BAB: * - - BE-,HE+BAB: BE-lHE torque (or force), as shown in (19) and (20).

+ * * + BAB: * * BE- ,I?;). (15) TN=MNBE-~zN-~+I~N~N. (19)

604 IEEE JOURNAL OF ROBOTICS AND AUTOMATION, VOL. 4, NO. 6, DECEMBER 1988

7 yo
The inertia matrix and mass center position vectors are

yo o o 01

'rE=[O, 0 , 0 , 1IT, N = l , 2. (22)

The gravity field vector is

Fig. 1 1 . Configuration of a double pendulum.

where the symbol g inside the brackets represents the

TABLE I gravitational constant. Considering the rigid body motion and
LINK PARAMETERS OF DOUBLE PENDULUM
the kinematic model shown in Fig. 1 1 , we can easily have the
condition
Link No. @i di ai ai Variable

Since the inertia matrices are null matrices, therefore, only the
linear accelerations ' F Y I and ' f y need to be derived for each
link's dynamics. From (8), the symbolic forms of forward
Similarly, the input force f N is dynamic equations can be derived by using the Piogram
symbolic representations method as follows:

For the evaluation of Newton-Euler equations, all derivations

must be with respect to the base frame, but it is most
convenient to carry out the required computations working
with components relative to a frame which is fixed in the link. Each term of (25) and (26) can be easily operated in the
In this paper, we propose a Piogram symbolic representa- Piogram symbolic representation method as shown in Figs.
tion method to manipulate the successive coordinate transfor- 12-15. The patterns shown in Figs. 12-15 for clear explana-
mations. The advantage is that the unnecessary operation paths tion are the only paths that need to be manipulated for
are avoided, and then the tedious simplification processes of generating ' rt2 and ' rY1. Adding the outputs on the right sides
expressions in equations are eliminated. Furthermore, the of each terms that are shown in the above figures, we obtain
Piogram symbolic representations and their derivatives have
been coded into an efficient computer program PIODYM-I. e
a2ef + a24 f + 2a2e2e,+ al f C, - al el S, 1
Thus the dynamic equations of motion for any degree of
freedom of robotic manipulators may be derived automatically
-02
r2
-
- - a2e1 + a2e, - al e f S, - al8, C,
0
and symbolically. The flowchart showing the automatic
derivation of dynamic equations for an arbitrary robot
0 J
configuration is shown in Fig. 4. The derivation of dynamic and
equations of motion for a double pendulum and the Stanford
arm are presented as illustrative examples. The CPU times for
the generation of the forward solutions and complete dynamic
robot models for double pendulum, three-degrees-of-freedom
Stanford arm, and the PUMA arm using PIODYM-I are listed
in Table IV in comparison with Leu and Hemati [9] and Next, considering Newton-Euler's equation for the center
Murray and Neuman [8]. of mass of the ith link, assume that no forces or moments are
applied at the far end of link two, i.e., f : = Mi = 0, thus
ILLUSTRATIVE
EXAMPLE
1 from (17), we have
We first consider a simple example to study the dynamics of
a double pendulum with coordinate frames shown in Fig. 1 1 . ff = m2 ' r p - m2Big (29)
The parameters of the coordinate frame transformation are f l = m 2 '"01
rl + B2f2-m
1 2 1qg. (30)
listed in Table I. The double pendulum consists of two
massless links with a weight attached to the end of each link. Substituting (27) and (28) into (29) and (30) and making the
An actuator is located at each joint to drive the double operation for the terms m2B:g, Bf f f , and mlByg using the
pendulum. Piogram symbolic operation rules, similar to the symbolic
CHENG et al.: EQUATIONS OF MOTION FOR ROBOT MANIPULATORS 605

-a2 -a

a i 2 + a ii’c -a ii s
2 1 1 1 2 1 1 2

-a 8 -a 62s -a i; c
2 1 1 1 2 1 1 2

O h + o

I C I O

Fig. 12. B:BAB: ’ r ; & ’

Fig. 16. Configuration of the Stanford arm.

TABLE I1
0- 1-a-B L 2
LINK PARAMETERS OF MODIFIED STANFORD ARM

Link No. 0, d, a, ff, Variable

1 0, 0 0 - 90 0,
2 02 0 90 02
3 0 d3 0 0 d3
1+- I O

Fig. 13. BtBAB: ‘ r :

[
and

mla14:+ m2(a& + a2Q+ 2a24241 + g C d - mlgCl

2a2 8, e, - rnla181+ m2(- a201- a2e2)+ mlgSl - m2gSl2
ft= -m2a14:~ -2m2a1el C2
o b -
0
0
(32)
O h 4 0
From (19), we thus obtain two equations of motion as
follows:
l i - -0
r1= -el( - m2a:- 2m2ala2C2- mla: - m2a:- I o l )
Fig. 14. 2B;BAB: ’ r ; .
+ $2 (m24 + m2a1a2 cd

-alw
0 c L
alii;

and
(33)

r2= - 8,( - m2ag- m2aIazC, )

O h 4 0 - e2(m2ai + IO2)
+g(m2a2S12)
4 0 +4:(m2ala2~2). (34)
Fig. 15. ByBA ‘ r l .
It is worthwhile to mention that the above results of manual
derivation can also be obtained by using PIODYM-I.
operations of Figs. 12-15, thus we have
ILLUSTRATIVE 2
EXAMPLE
m2(a24:+ a24i+ 2a2B2d1 + g c 1 2+ a,B:c2- al8, s,)

1
The second example is a manipulator with the same
m2(-a2BI- a 2 8 2 - g ~ 1 2 - a l B ~ ~ 2 - a 1 8 1 ~ 2 )geometry as that of the Stanford arm (as shown in Fig. 16) but
fi= 0 without the wrist and gripper, called the modified Stanford
0 arm. The link-geometric parameters of this three-degree-of-
(31) freedom manipulator are listed in Table 11. Presented in this
606 IEEE JOURNAL OF ROBOTICS AND AUTOMATION, VOL. 4, NO. 6, DECEMBER 1988

TABLE 111
LINK PARAMETERS OF 3-DOF PUMA ARM

Link No. 8, dl a, a, Variable

example is a method for deriving manipulator dynamic

equations using PIODYM-I. The deriving procedures are quite
similar to those of Example 1 that can be shown as Fig. 4.
Because of the symmetry of link geometry, the cross
products of inertia of each link are zero. The matrices of
moment of inertia IcNare

rINxy o o o i
(35)
L o 0 0 o j
The position vectors of center of mass coordinated in each
frame are as follows:

Equations (40)-(51) are prepared as an index of operation

processes that can be scanned by the program itself, term by
term, for deriving the forward equations and the rate change of
The g vector is angular momentum in symbolic form. Unlike in the matrix
operation process, it is not necessary here to implement the
cumbersome matrix multiplication processes and then simplify
them. Using the method presented in this paper, we can easily
Considering the derivation of the equations of motion, the eliminate the unnecessary operation path by means of Piogram
efficient technique for computing the joint input forces and symbolic operation rules. As illustration, all the necessary
torques we adopt is to have each link's dynamics referred to its symbolic representation's patterns of each term in (40) that are
own link coordinates as follows. First, when the parameters of implemented in the cycle block c to f of Fig. 4 are shown in
robot links have been determined and inputted from block a of Figs. 17-22. The components of the acceleration vector are
Fig. 4, the relations and their derivatives between two adjacent finally obtained when the corresponding outputs of the six
frames expressed in the pattern form of the Piogram symbolic chains are summed up as indicated in (40). Then we obtain the
representation such as in Figs. 3, 7, 8, and 9 can be generated expressions for 'f:' in symbolic form from block g of Fig. 4
automatically by PIODYM-I from block b of Fig. 4. The automatically. Thus
initial conditions are assumed as W O = &0 = 'to= ' f O= 0.
Letf4 = M4 = 0, and assume the links are rigid bodies, hence
the derivation of the position vectors which are the locations of
mass center with respect to its link frame are zero. The
forward dynamic equations and the rate change of angular
momentum for link-one, -two, and -three system can be
assembled by means of block b of Fig. 4, and listed,
respectively, as follows. (52)
i) For link-three system: Hence it is clear that only the necessary operation paths are
left to be implemented and, therefore, the simplest forms of
(40)-(5 1) are thus derived automatically and without any
simplification processes. Furthermore, the preceding proc-
esses can very conveniently be implemented in a computer
system, and are economic for the computer memory. The
following step is to derive the backward equations from (17)-
(20). By using block h to i of Fig. 4 the actuated torque or
CHENG et ul.: EQUATIONS OF MOTION FOR ROBOT MANIPULATORS 607

O F - 2b3B2

-(z 3 +d 3 )
-2i3 +Lo
\ o+- 4 0

01-
U
o c -lo
Fig. 22. 2B:Bi(B:E: ' r p ) .

force can be obtained automatically. So we have

TI = 6182(2m32i+ 4m3d3z3+ 2m3d - 2Z3,

o c 4 0
+ U3,- 2Z2, + 2Z2,,)
* C2S2+8:(d2m3z3+m3d2d3)S2-m3d3d2S2

o+ 4 0
+ 81 [m3(z:+2z3d3+ d : ) . (S2)2
+m2(~2+dd~+md:
d3
+(z2u+13u) * (s2)2+(12u+13zz)
z3 c L d,
- (c2)2+4yy+471+m1z:1
+ el [2m3( 2 3 + d3) - (&I2]
a3
1+ --lo
+e2c2[-m3dZ(z3+d3)1
Fig. 19. E:(& ' r y ) .
- 2m3d2d3e2C2 (53)
72= )2[- +&I2
- m 3 ( z 3 + d ~ ) g S 2 + ( e ~m3(z3
+ 4 , - 13,, + 12, - 1 2 u l s2c 2
- 8 (1113 d2 d3+ m3d2z3) C2
+ 8 2 [m3( ~ +3 d3 l2+ 13yy + IzyyI
o c + o +282d3[m3(z3+d3)1+Iu282 (54)
ji = m3[( - (z3+ d3 )e:S2 - d2el) S2
- ( ~ +3 d3)ei + d3 + gC2] + Z u 3 d 3 . (55)
The equations of motion of the PUMA arm (as shown in
Table III) have been also derived successfully. But because of
space limitation, only the formulating efficiency is presented
in Table IV for comparison.

DISCUSSION
The automatic equation derivation process is highly desir-
Ob- 4 0 able while manual symbolic expansion of manipulator matrix
equations is tedious, time-consuming, and error-prone. The
derivation of dynamic equations involves a large number of
symbolic operations. For an n-degree-of-freedom manipula-
608 IEEE JOURNAL OF ROBOTICS AND AUTOMATION, VOL. 4, NO. 6 , DECEMBER 1988

tor, it requires n(n + 1)/2 differential operations for the d T ; / TABLE IV

d q j matrices and an addition of n(n + l ) ( n + 2)/6 COMPARISON OF EFFICIENCYBETWEEN PIODYM-I, MACSYMA, AND
ARM
differential operations for the d2Tj/dqjdqk.Furthermore, it
requires O( n3) and O( n4) trace operations for obtaining the Subject PIODYM-I* MACSYMA** ARM**
inertial and the velocity-related terms, respectively. This
imposes a great demand on the computer memory space during Computer pc/AT (micro) DEC 2060 VAXl1/780
the equation derivation process. system
Deriving the dynamic equations symbolically, insights on Memory space 64K 512 K 512 K
the dynamics of a manipulator can be generated in two ways: required
one is examining the terms of the dynamic equations and the Formulation A 7 19 11
other is using the dynamic equations to simulate individual (sec.) B 30 69 -

force components. Another merit of expanding the vector/ C 45 - 65.93

matrix equations of motion is that, if most manipulator links Simplification A 0 45 -
(SE.) B 0 137 -
are symmetric in geometry, the resultant equations are more C 0 - -
computationally efficient than even the efficient recursive
Newton-Euler formulation in vectodmatrix form which is in a A Double pendulum; B 3-DOF Stanford arm; C 3-DOF Puma
numerical approach. It has been shown that for the Stanford arm.
arm with constant payload, the time required for the force Undefined values.
“-”
* Formulation and simplification processes are tracked by a real clock.
computation is based on an order of magnitude less than that ** Formulation and simplification processes are tracked in CPU time.
based on the recursive Newton-Euler formulation [7], [9].
This computer saving is particularly available for manipulator tion of symbolic models mentioned in the report, ARM
motion control when the controller requires an update of the (Algebraic Robot Modeler) is another computer program for
dynamic forces in real time. generating symbolically the forward solution and the complete
This paper presents a new technique for deriving the scalar Lagrangian dynamic robot model [8]. In Table IV, the
form of manipulator dynamic equations by symbolically efficiency of the presented method is also compared with
expanding the Newton-Euler equations using the Piogram ARM; it is clear that PIODYM-I is more efficient than ARM.
symbolic representation method. The derivation processes, Moreover, it is noted that the computer system applied in this
implemented by PIODYM-I automatically, are shown in two paper is a 16-bit microcomputer only. And the present method
illustrative examples. In deriving the dynamic equations, the becomes a very easy and powerful tool for deriving the
computer memory space and efficiency are two important dynamic equations of motion of multivarying-link mechanisms
factors that have been considered in this paper. In the present such as robot manipulators.
method, other than the multiplication of each of the two
SUMMARY AND CONCLUSION
matrices explicitly, the unnecessary operation paths which
quickly reach a zero value are eliminated. Especially, in the This paper presents a new technique for deriving the scalar
computation of velocity and acceleration using matrices, many form of manipulator dynamic equations by symbolically
terms are computed which eventually are dropped (since at the expanding the Newton-Euler matrix equations using
end of the chain of matrix multiplications they are multiplied PIODYM-I. The algebraic dynamic robot modeling program
by zero). Moreover, the other important contribution of this PIODYM-I has been implemented to enable the control
paper is the direct construction of the expression of functions, engineer to formulate and gain physical insight into Lagran-
for example, sin (0, + 02), other than the simplification of the gian dynamic robot models for the systematic design and
expression of functions such as evaluation of robotic feedback control system. Illustrative
examples 1 and 2 are presented to show both manually and
sin (0,) cos (02)+cos (0,) sin (02)-+sin(e, +e,). automatically derived processes. As compared with the results
of Leu and Hemati [9] and Murray and Neuman [SI, it is
PIODYM-I will simplify the operation path of each Piogram
obvious that the technique presented in this paper is more
symbolic chain avoiding the tedious simplification after
computationally efficient than either the MACSYMA or the
formulations, the way MACSYMA does. Hence it is obvious
ARM code. Furthermore, the tedious simplification processes
that the method presented in this paper can save much more
after derivations can also be eliminated. In addition, it needs
memory space. As a comparison with the program code based
much less computation time and less computer memory space.
on MACSYMA and ARM [9], [8], Table IV shows that
PIODYM-I needs much less computation time and less REFERENCE
computer memory space than the program code based on A. K. Bejczy, “Robot arm dynamics and control,” Tech. Memo 33-
MACSYMA and ARM. That is, for the cases of modified 66-9, Jet Propulsion Lab., Pasadena, CA, 1974.
Stanford arm (or double pendulum), the program code based B. C. Mcinnis and C. K. F. Liu, “Kinematics and dynamics in
robotics: A tutorial based upon classical concepts of vectorial mechan-
on MACSYMA takes 69 (19) CPU seconds to formulate the ics,” IEEE J. Robotics Automat., vol. RA-2, no. 4, Dec. 1986.
equations and 137 (45) CPU seconds for the simplification. C. S. G. Lee, “Robot arm kinematics, dynamics and control,” IEEE
While PIODYM-I needs only 30 (7) seconds which are tracked Computer, pp. 62-80, Dec. 1982.
J. Denavit and R. S. Hartenberg, “A kinematic notation for lower-pair
by a real clock to formulate the equations and it is unnecessary mechanisms based on matrices,” ASME J. Appl. Mech., vol. 22, pp.
to do simplification. Among the existing methods for forma- 215-221, June 1955.
 ~

CHENG el al.: EQUATIONS OF MOTION FOR ROBOT MANIPULATORS 609

J . M. Hollerhach, “A recursive formulation of Largrangion manipula- Cheng-1 Weng received the B.S. degree from
tor dynamics,” IEEE Trans. Syst., Man, Cybern., vol. SMC-IO, no. National Cheng Kung University (NCKU), Tainan,
11, pp. 730-736, 1980. Taiwan, ROC, in 1966, and the M.S. and Ph.D.
J. R. Birk and R. S. Kelly, “An overview of the basic research needed degrees from the University of Rochester, Roches-
to advance the state of knowledge in robotics,” IEEE Trans. Syst., ter, NY, in 1969 and 1973, respectively, both in
Man, Cybern., vol. SMC-11, no. 8, pp. 574-579, 1981. mechanical engineering.
J. Y. S. Luh, M. W. Walker, and R. P. C. Paul, “On-line From 1973 to 1977, he was an Associate Profes-
computational scheme for mechanical manipulators,” Trans. ASME sor of Mechanical Engineering of National Cheng
J. Dyn. Syst., Meas., Contr., vol. 102, pp. 69-76, June 1980. Kung University. From 1977 to 1980, he taught at
J. I. Murray and C. P. Neuman, “ARM: An algebraic robot dynamic the National Chaio Tung University as a Professor
modeling program,” IEEE Publ. CH 2008-1/84/oooO/0103. and Department Head of the Department of Me-
M. C. Leu and N. Hemati, “Automated symbolic derivation of chanical Engineering. In 1980, he returned to NCKU as a Professor. Form
dynamic equations of motion for robotic manipulators,” Trans. ASME 1980 to 1986 he was also the Department Head of the Department of
J . Dyn. Syst., Meas., Contr., vol. 108, pp. 172-179, Sept. 1986. Mechanical Engineering, and has been the Director of Tjing Ling Manufactur-
P. E. Nikravesh, R. A. Wehage, and 0. K. Kwon, “Euler parameters ing Center at the same University since 1986. His current research interests
in computational kinematics and dynamics. Part 1,” ASME J . Mech., include elastodynamics, hydrodynamic stability, metal forming, and machin-
Transmissions, Automat. Des., vol. 107, pp. 358-365, Sept. 1985. ing.
-, “Euler parameters in computational kinematics and dynamics.
Part 2,” ASME J . Mech., Transmissions, Automat. Des., vol. 107,
pp. 366-369, Sept. 1985.
R. L. Pio, “Euler transformations,” IEEE Trans. Automat. Contr.,
vol. AC-11, pp. 707-715, Oct. 1966.
-, “Symbolic representation of coordinate transformation,” IEEE
Trans. Aerosp. Navigat. Electron., Vol. ANE-11, pp. 128-134, June
1964.
R. P. Paul, “Manipulator Cartesian path control,” IEEE Trans. Syst.,
Man, Cybern., vol. SMC-9, no. 11, pp. 702-711, 1979.
-, Robot Manipulators: Mathematics, Programming and Con-
trol. Cambridge, MA: MIT Press, 1981.
W. 0 . Schiehlen and E. J. Kreuzer, “Symbolic computerized
derivation of equations of motion,” in Proc. IUTAM Symp. (Munich,
F.R. Germany, Aug. 29-Sept. 3, 1977).
Y. Bar-Itzhack, “Symbolic representation of translatory motion in
multivarying-link mechanisms,” J. Guidance, vol. 6, no. 3, pp. 193- Chao-Kuang Chen received the B.S. degree from
199, May-June 1983. the National Cheng Kung University, Tainan, Tai-
wan, ROC, in 1957, the M.S. degree from Georgia
Institute of Technology, Atlanta, in 1970, and the
Pi-Ying Cheng was born in Hsin-Tzu, Taiwan, Ph.D. degree from the University of Liverpool,
ROC, on January 17, 1958. He received the B.S. Liverpool, UK, in 1986, all in mechanical engineer-
degree in 1980, the M.S. degree in 1984, and the ing.
Ph.D. degree in mechanical engineering in 1988, From 1960 to 1966 he was a Mechanical Engi-
all from the National Cheng Kung University, neer in Chinese Petroleum Corporatoon. From 1966
Tainan, Taiwan, ROC. to 1971, and from 1971 to 1975, he was a Lecturer
He is now an Associate Professor in the Depart- and an Associate Professor, respectively, of Na-
ment of Mechanical Engineering, National Chiao tional Cheng Kung University in the Department of Mechanical Engineering.
Tung University, Hsin Chu, Taiwan, ROC. His He has been a Professor at the same Department since 1975. His current
current research interests include robotics, control research interests include mechatronics, nonlinear and variable structure
mechanisms, design, and automation. control theory, heat and mass transfer, and metal machining.

View publication stats