WA6

- 11:15

Proceedings of the 271h Conference

on Decisionand Control
Austin, Texas December 1988

IMPLEMENTATION OF A LAG-LEAD COMPENSATOR FOR ROBOTS

Yilong Chen
Mathematics Department
General Motors Research Laboratories
Warren, Michigan 48090

ABSTRACT
Because of their simplicity, PD (or PID) controllers are widely
used with various robot control strategies. For dynamic control
of robots, this algorithm can be shown to lead to unsatisfactory
trade-offs between static accuracy, system stability and insensitivity to disturbances. These trade-offs become even more serious
as sampling rate decreases. By deriving a more realistic discrete
time system model and using frequency response analysis, a Laglead compensator is designed and implemented. Both theoretical
analysis and real tests are given for the comparisons of the PD
(or PID) controller and the Lag-lead compensator. Replacing PD
(or PID) controllers by Lag-lead compensators results in a small
increase of off-line tuning effort and on-line computational load,
but the improvements in robot performance are significant.

and insensibility to disturbances on the frequency domain. The

implementation and real tests verify the encouraging results on
the time domain. The Lag-lead compensator reduces the tracking
errors considerably. In Section 2, we briefly derive the discrete
time system model which includes a computational time delay
and a zero order holder for a robot system. The conventional
PD controller and its limitations are analyzed in the frequency
domain in detail. In Section 3, we designed and analyzed the
Lag-lead compensator by the frequency-response approach. Also
we accomplished the realization of the Lag-lead compensator and
derived the algorithm for the proposed feedback control law. The
system and test results for the implementation are presented and
discussed in Section 4. The conclusion is given in Section 5.
2. DISCRETE TIME MODEL AND ITS

1. INTRODUCTION
Because of their simplicity, PD (or PID) controllers are widely
used with various robot arm control methods, like the approximate
linearized method [Paul 1972, Pieper 1968, Whitney 19691; the
computed torque method [Markiewicz 1973, Luh, Walker, Paul
19801; the hierarchical control method [Albus et al. 1981, C. S.
G. Lee et al. 19821; the feedforward compensation method [Bejczy 1974; Khalil 19781; nonlinear feedback control method [Freund 1982, T. Tarn, A. Bejczy and Y. Chen 19841; the adaptive
control method [Dubowsky and DesForges 1979, Goor 1982,19841
and etc. Our analysis and tests of a Unimation PUMA 560 have
shown that the use of a PD feedback, even with a relatively faster
sampling rate (5ms) than those sampling rates usually used for
robot dynamic control, leads to serious trade-offs between static
accuracy, system stability and damping of high frequency disturbances. Adding an integral of error to the PD to make a PID
feedback often makes the overall system less stable, although it
improves the static accuracy.

The purpose of this paper is to propose the design of a new

compensator for the robot feedback control which reduces the
trade-offs caused by the conventional PD (or PID) controllers.
Our simulations and real tests have shown that the sampling
time is critical to the system performance of robots. So it is necess a r y to use discrete time model instead of continuous time model
for the design of compensators. The tool that we propose for
designing the feedback loop is the frequency response approach
based on the z-transform of the discrete time system model cascaded with delay and hold devices.

FREQUENCY-RESPONSE
In a sampled data system, the choice of sampling rate is often
governed by conflicting criteria. In practice, the sampling rate for
dynamic control of robots often becomes critical. It is not high
enough for a high performance robot arm to be considered as a
continuous time system. In such cases, we need to deal with a
discrete time system model, which we use for feedback control
design. First let us derive the discrete time model of a robot, including a one period computing time delay and a zero order holder
(ZOH) for the example shown in Figure 1 with a feedforward and
feedback loop. To be focus on the comparison of a Lag-lead compensator to a PD controller, a simple linear model is used in this
paper. A more sophisticated discrete nonlinear system model is
preferred if it is available [Neuman and Tourassis, 19851. In that
case, we can use a feedback control method, or a computed torque
method, or a feedforward compensation method to have a desired
linear and decoupled (or approximately linear and decoupled) system for which the following analysis of the frequency-response of
a discrete time system is still valuable.
In Figure 1 each a x i s of a robot arm is described by a 3rd
order motor model [Goor, 19841. For the jth axis, we have
LjJjx,

K$
where
= (applied) armature voltage,
= motor position,
= resistive component of armature circuit ,
L j , = inductive component of armature circuit,
KZ = voltage constant of motor (back emf),
Bj = viscous dampimg coefficient,
Jj
= motor inertia,
K+ = torque constant of the motor.
uj

xj
Rj

In this paper, using the frequency-response approach we have

designed and realized a so-called Lag-lead compensator for a robot
arm. The analytical comparison between a PD (or PID) controller
and a Lag-lead compensator shows that the Lag-lead compensator
reduces the trade-offs between the static accuracy, system stability

88CH2531-2/88/0000-0174$1.OO

1988 IEEE

174

for ease of notation, we drop the subscript j, since we will deal

with the axis individually. We may define

T 2 ( a- P + 7)
4(a+P+7)
az
T
--.
4 b [ l - 2e-01T/2cos(wo?')

d2=

calT]

bT 2bal
*[---+
az

az

where X(s) and U(s) are the Laplace transformations of x(t) and
u(t) respectively, and

a1 = ( L B R J ) / L J ,
az = ( R B K , K T ) / L J ,
b = KT/LJ.

(3)

We can convert the Laplace transfer function (2) to a ztransfer function by defining
oo

X ( Z )= Z ( z ( t ) ) =

z(nT)z-"

(4)

n=O

(where T is the sampling period) and derive the transfer function

of the discrete time system for a robot arm.
In consideration of a zero order holder and a one period computing time delay (see Figure 2) the transfer function of G block

The Bode plot of G(w) with sampling time T = 0.005s is

shown in Figure 3. To understand better the limitations of PD
(or PID)controllers, it is useful to know how a continuous time
system is modified by sampling, computing delay and a zero order
(5)
holder. These can be seen from comparing the transfer functions
Gm(s) (Eq. 2) of the continuous time case and G(w) (Eq. 10) of
the discrete time case (and their Bode plots, Figure 3 and Figure
4 ) . More important, the comparison shows the creation of righthand plane zeros of G(w) at 2/T and the creation of two fast zeros
when compared to Gm(s). These zeros can be attributed to the
sampling-and-hold operations and thus depend on the sampling
rate. They become slower and thus more important to the transient response of the system as the sampling rate is decreased.
More significantly, in addition to increasing the gain at high frequency, the zero at 2/T is in the right hand side of the complex
(7)
plane, so that it causes an extra phase lag.
After observing these important features, now it is interesting
to see what happens when we use a conventional PD controller for
a robot arm system. Let us see an example. In our real test shown
in Figure 1, after carefully tuning, the PD controller was chosen
as H ( s ) = g1 g2s, where g1 = 2 4 , s ~= 0.225. From Figure 5,
we may see that the continuous time system model with the PD
controller has a crossover frequency w, = 180 rad/sec and static
With this model we now are ready to design the feedback loop. error coefficient K = 100. Note that this system model has very
Frequency design methods based on Bode plot are useful and con- large positive gain margin (=+ 00 db) and phase margin (=+goo).
venient for designing compensators for continuous time systems Also, for the continuous time system model, the gain at high fredescribed by transfer functions. However, frequency curves for quency attenuates quickly. When g1,ga were increased, according
discrete-time systems are more difficult to analyze since the pulse- to the continuous time model, the system would remain stable.
transfer functions are not rational function in s, but in z=exp(sT).
However, this conclusion conflicts with the real test results. To
In order to circumvent this difficulty and to use Bode's design understand this fact we need to analyze the more realistic discrete
techniques, we may use the w-transformation which is defined as time system model G(w). The Bode plot of G(w) with the digital
w=
implementation of a PD controller is shown in Figure 6. Comparing Figure 6 with Figure 5, we can see that the crossover frequency
Under the w-transformation, we substitute z =
into
w, and static error coefficient K remain about the same. However,
(5). Then we have
the gain margin (=+ldb) and phase margin (=+15O) from Figure
6 become critical. It is clear that the system could be unstable
G(w) G(z(w))
when we increase the gain or when the system parameters vary
from those of the model. Figure 6 shows that the PD controller
leads to serious trade-offs between the static accuracy, stability,
and high frequency noise rejection. For example, in order to have
larger positive gain and phase margins for system stability, we
must decrease the gain K, which means we must sacrifice static
accuracy. On the other hand, to have better static accuracy and
therefore to increase the gain K, the high-frequency noise response

,w.

175

of the compensated system with the digital PD controller will be

even worse than that we see from Figure 6. Normally, high frequency noise does not significantly influence the function of analog
control devices because of their natural low-pass behavior. However, in the case of digital signal processing, noise is sampled and
transmitted and it may be amplified. These trade-offs become
more serious as sampling rate decreases. We would like to point
out that the system behavior is well predicted by the discrete time
system model (i.e., Figure 6) and is consistent with the real robot
tests performed on a PUMA 560 arm at the Advanced Engineering
Staff of General Motors.
Detailed study shows that even for a relatively fast sampling
rate (T=O.O05s), the real discrete dynamic system of a robot behaves quite differently from the continuous time dynamic system
model. Further, conventional PD controllers cannot meet the requirement for high performance because of unacceptable tradeoffs between static accuracy, system stability, insensitivity to high
frequency noise. Therefore, we must use a discrete time system
model and design a more sophisticated feedback compensator.

3. LAG-LEAD COMPENSATOR AND ITS

REALIZATION ALGORITHM
In order to have a high value of the error coefficient, and thus
high static accuracy, and yet satisfactory stability, insensitivity of
model structure inaccuracies and noise rejection, we find it necessary to reshape the system frequency-response curve. The gain in
the low frequency region should be large enough to achieve high
static accuracy. Near the crossover frequency the slope of the logmagnitude curve in the Bode diagram should be -2Odb/decade
and thus to have a less than 180 phase lag. This slope should
extend over a reasonably wide frequency band to assure proper
phase margin and gain margin. For the high frequency region,
the gain should be attenuated as rapidly as possible in order to
minimize the effects of noise and model structure inaccuracies.
Based on these requirements, we propose to design a Lag-lead
compensator, which has more flexibility than PD (or PID) controllers to reshape the frequency- response curve for our system.
The suggested Lag-lead compensator in w-plane is:

Lag-lead compensator on the frequency domain can be found in

[Chen, 19861
By substituting the w-transformation w = (2/T)(z-l)/(2+1)
into (16), we have the following compensator in terms of z, which
can be realized in a digital computer:
A
[(2 + LT) + (LT - 2)]
() = H ( z ) =
H ( w ( z ) )= /32K.

[(LTt 2PI)

E()

[(dTt 2)
. [(PZdT t 2)

which can be viewed as a special case of (16) when is very large.

The phase lead portion &(w) of this compensator in (Eq.16)
alters the frequency- response curve by adding phase lead angle
and increasing the phase margin at the crossover frequency. It
basically improves the system stability and noise rejection. The
phase lag portion Hz(w) of the Lag-lead network provides attenuation near and above the crossover frequency and thereby allows
an increase of gain at the low-frequency range to improve the
static accuracy. The Lag-lead network possesses two poles and
two zeros. Therefore, such compensation increases the order of
the system by two. The detailed design and improvements by the

+ (LT - 2/31)]

+ (/32dT - 2)]

The predescribed Lag-lead compensator is a dynamic system.

Different realizations may be obtained by transforming the statespace coordinates. The choice of a suitable realization is very
important for numerical conditioning. In particular, the companion forms make the system sensitive to changes in the coefficients,
especially with a high sampling rate and with multiple eigenvalues
close to the unit circle[Astrom, et. al., 19841, To avoid this numerical difficulty, we may represent the compensator as a combination of first-(for distinct real poles) and second-order (for complex
poles) systems. In our case, the Lag-lead compensator has simple
real poles and zeros and thus may be decomposed as follows:
From (18) we have

[ ( d T + 2)
[(PadT t 2)

+ (dT - 2 ) ]

+ (PzdT - 2)]
(19)

- The state diagram for this cascade form of the Lag-lead compensator from equation (19) is shown in Figure 9.
We let LLI = (LT - 2)/(LT t 2), LL2 = (dT - 2)/(dT
2), ddl = (LT - 2/3I)/(LT 2/31), ddz = (P2dT - 2)//3zdT
2), kkl = (LT 2)/(LT 2/31), and kkz = (dT 2)/(/3zdT -t2).
Then we have state equations with states z1,zz and ez, input e
and output Au as follows:

where K, L, d,
and Pz are constant parameters to be selected. The number K is the gain of the compensator and L and
d are known as the corner frequencies. The Bode diagram of the
Lag-lead compensator can be seen in Figure 7.
Note that the suggested compensator in (16) can also have
the following form in w-plane:

+ (dT - 2)]

+
+

1) = -ddlzl(n) t ( L L I - ddl)kkl e(n)

ez(n) = z1(n) k h e(n)
zz(n 1) = -ddzzz(n) t (LLz - ddz)kkz . q ( n )
q(72

(20)

A 4 n ) = PzK(za(n)+ kk2 . ez(n))

Notice, the input position error e of this Lag-lead compensator

can be obtained by the desired position and measuring the real
joint position from the optical encoder. The computed output Au
is the corrective voltage (or current) which we add to and adjust
the input voltage (or current) of the robot motor. (Eq. 20) can
be easily realized in a digital computer[Chen 19861.
4. IMPLEMENTATION RESULTS
Although the overall compensated system with the Lag-lead
compensator looks good on the frequency domain (See Figure 8
and compare to Figure 6), we stillneed to verify further the system
performance on the time domain. Simulation results show that a
Lag-lead compensator based on a frequency- response approach
indeed leads to higher accuracy while ensuring the stability of the
system and insensitivity to disturbances. The overall system is

176

more stable and robust. The detailed simulation results can be

found in [Chen, 19861.
Recently we conduct some real implementation tests. The results are also encouraging. To test OUI new control algorithm, we
use a PUMA 560 robot arm but bypass the VAL controller. A
MicroVax I1 is used as a target computer to control the arm. A
finished VAXELN Pascal system runs on the target computer by
itself, without VAX/VMS or any other operating system present.
VAXELN is a software product for the development of dedicated,
realtime systems for VAX processors[Digital Equipment Corporation, 19851. It's a faster way to design and implement time-critical
applications on micros. Both multitasking and multiprogramming
are supported by the VAXELN Kernel and VAXELN Pascal programming language. In the process of developing and building the
VAXELN system for controlling the arm, we also use the VAXELN
Debugger, thru Ethernet from a VAX 780 computer, to debug the
programs in a developed, executing VAXELN system. The remote
debugger can display the states of all VAXELN processes and jobs
in the local-area network and can dynamically change the user's
"view" from one process of node to another. Finished VAXELN
systems can be loaded down line from VAX 780 into the target
computer Macrovax 11.
In the standard Unimation system, the VAL software interprets the operating instructions for a robot arm, and the controller
transmits these instructions from the computer memory to the
arm. From incremental encoders and potentiometers in the robot
arm, the controller/computer receives data about arm position.
This provides a closed loop control of arm motions. The VAL controller (LSI-11 computer) is interfaced to the joint interface board
through a parallel 1/0 board. The joint interface board then relays joint related information received from the VAL controller to
the individual joint microprocessor boards[Unimation, 19821. For
our implementation we severed the connection between the joint
interface board and the VAL controller by removing the parallel
1/0 board connection from the VAL controller board. We then
connected the parallel 1/0 connection to a MicroVax I1 Q - bus.
Thus the MicroVax I1 is in direct control of the robot arm. we can
now communicate with the joint processors by sending the joint
microprocessors commands to perform various functions.
Using MicrovaxII, we receive point data and build a queue.
And the motion planning, inverse kinematics and joint interpolation are then accomplished. By modifying the VAXELN software,
the sampling rates of different executing processes can be varied.
We now have two ways to follow a predescribed path:
(1)Input the desired position command from the MicrovaxII
to the joint microprocessors and use the PUMA PID servo control
strategy to get the required currents for the power amplifiers of
six joints.
(2)Implemente our own control algorithm, the Lag-lead compensator, and compute the required current in MicrovaxII and
then send to the joint power amplifiers bypassing the PUMA PID
servo control.
We can now compare the conventional PID controller and our
Lag-lead compensator by following the same path with the same
speed. The only difference between these two tests is the servo
control sampling rate. Besides the LSI 11 main CPU, Unimation
uses six microprocessors for the servo control of six joints. So the
sampling rate, as high as 0.875 ms are used for PUMA PID servo
control. In our case, however, in addition of doing queuing, path
planning, joint interpolating, timing, error checkingand so on, the
CPU of MicrovaxII has to do the servo control for all six joints also.
Due to the large computational load, we can only use no faster
than 5 ms sampling rate. Even so, the performance of the Lag-lead
compensator is better than the PID controller. As a example, we
let the end effector of a PUMA 560 follow a path which is a part

of a circle with diameter of about 400 mm. The speed is about

70 "/sec.
The running time is 5 seconds. The load is 4 lb. Our
system is robust and the parameters of the Lag-lead compensator
were fairly easy to be tuned. While the parameters of the PID
controller were carefully set by Unimation service people and they
are believed to be well tuned. The comparisons of tracking errors
for all six joints are shown in Figures 10 - 15. The improvements
by the Lag-lead compensator are obvious. They can be also seen
in the follwing table. Different paths were used for tracking. But
similar results were obtained.

CONTROLLER

0.009 0.013

0.008 0.017

0.009

0.007

0.0013 0.003

0.004 0.003

0.003 0.0035

LAG-LEAD
CCMPENSATOR

IMPROVING
FACTOR

Table : Maximum Tracking Errors (unit: rad.) By Using PID and Lag-lead

Although the design and tuning for a Lag-lead compensator

need a little more effort than for a PID controller, the on-line
computational load increased by using the Lag-lead compensator
is very small comparing to other parts of the system. Moreover,
we developed a method which can automatically tune parameters
of robots and therefore save time and labour[Chen, 19871, the
increased tuning effort is not a big problem.
5. CONCLUSION
In many cases the sampling rate is critical for dynamic control of robots. Thus the widely used PD (or PID) contro!ers
have serious trade-offs between the static accuracy, system stability and insensibility to disturbances. When used with the discrete
time system models, the classic frequency response analysis and
Lag-lead compensators show their important roll in improving system performance. Replacing PD (or PID) controllers by Lag-lead
compensators results in a small increase of off-line tuning effort
and on-line computational load, but the improvements in robot
performance are significant.

REFERENCES
(1) J. S. Albus, A. J. Barbera, R. N. Nagel, 1981, "Theory
and practice of Hierarchial Control," 23rd IEEE Computer Society
International Conference, September, 1981.
(2) K. J. Astrom, B. Wittenmark, 1984, Computer-Controlled
Systems, Theory and Design, Prentice-Hall, Inc., Englewood
Cliffs, New Jersey.
(3) A. K. Bejczy, T. J. Tarn, Y. L. Chen, 1985. "Computer
Control of Robot Arms", Proc. of 1st. IEEE International Conference on Robotics and Automation, St. Louis, Missouri, March
1985.
(4) Y. L. Chen, 1984, "Nonlinear Feedback and Computer
Control of Robot Arms." D. Sc. Dissertation, Department of System Sciences and Mathematics, Washington University, St. Louis,
Missouri, December 1984.
(5) Y.L. Chen, 1986, "A Lag-lead Compensator for Robot
Feedback Control", General Motors Research Publication GMR5480, August 1986.
(6) Y.L. Chen, 1987, "A Formulation Of Automatic Parameter Tuning For Robot Arms", Proc. 23rd IEEE Conference on

177

Decision and Control, Los Angeles, California, December 1987.

(7) Y.L. Chen, 1987, "Auto-tuning by Learning for Robot
Feedback Control", General Motors Research Publication GMR5766, April 1987.
(8) S. Dubowsky, D. T. DesForges, 1979, "The Application of
Model-Referenced Adaptive Control to Robotic Manipulators,"
J. Dynamic Systems Measurement and Control, VlOl (1979), pp.
193-200.
(9) G. F. Franklin, J. D. Powell, 1980, "Digital Control of
Dynamic Systems." Mass. Addison - Wesley.
(10) E. Freund, 1982, "Fast Nonlinear Control with Arbitrary
Pole-Placement for Industrial robots and Manipu8lators," The International Journal of Robotics Research, Vol. 1, No. 1.
(11) R. M. Goor, 1982, "Continuous Time Adaptive Feedforward Control: Stability and Simulatins," General Motors Research Publicatin GMR-4105.
(12) R. M. GOOI, 1984. "A New Approach to Minumum Time
Robot Control," General Motors Research Publication GMR4869.
(13) H. Herman and P. C. Camana, 1976. "Nonlinear Feedback in Simple Locomotion System," IEEE Transactions on Automatic Control, Vol. AC-19, pp 355-860.
(14) Rolf Isermann, 1981, Digital Control Systems, SpringerVerlag Berlin Heidelberg New York.
(15) M. E. Kahn, B. Roth, 1971, "The Near-Minimum-Time
Control of Open-Loop Articulated Kinematic Chains," ASME J.
Dynamic Systems, Measurement and Control, September, 1971,
pp. 164172.
(16) A. J. Koivo, T. H. Guo, 1981, "Control fo Robotic Manipulator with a Adaptive Controller," Proc. 20th IEEE Conference
on Decision and Control, San Diego, CA, Dec. 16-18, 1981.
(17) B. C. Kuo, 1970,Dismete - Datu Control Systems,
Prentice-Hall, Englewood Cliffs, New Jersey.
(18) C. S. G. Lee, T. N. Mudge, J. L. Turney, 1982, "Hierarchial Control Structure Using Special Purpose Processors for hte
Control of Robot Arms", Proc. of the 1982 Pattern Recognition
and Image Processing Conference, pp. 634-640.
(19) Y. S. Luh, M. Walker, R. Paul, 1980, "Resolved Acceleration Control for Mechanical Manipulators," IEEE Trans. Auto
Control, Vol. AC-25, No. 3, June 1980, pp. 468-474.
(20) B. R. Markievicz, 1973, "Analysis of the Computed
Torque Drive Method and Comparision with Conventional Position Servo for a Computer - Controlled Manipulator,'' Jet Propulsion Lab. Tech Memo 33-601.
(21) C. P. Newman, V. D. Tourassis, 1985, "Discrete Dynamic
Robot Model." IEEE Trans. on Systems, Man, and Cybernetics,
Vol. SMC-15, No. 2, March/April 1985.
(22) Katsuhiko Ogata, 1970, Modern Control Engineering,
Prentice-Hall, Englewood Cliffs, New Jersey.
(23) R. P. Paul, 1972, "Modelling, Trajectory Calculation and
Solving of a Computer - Controlled Arm," Stanford A.I. Lab,
Stanford University, AIM 177.
(24) D. L. Pieper, 1968, "The Kinematics of Manipulators Under Computer Control." Ph.D. Dissertation, Stanford University.
Memo 72, Stanford, CA. Stanford University Artificial Intelligence
Laboratory.
(25) T. J. Tarn, A. K. Bejczy, A. Isidori and Y. L. Chen, 1984,
"Nonlinear Feedback in Robot Arm Control," Proc. 23rd IEEE
Conference on Decision and Control," Las Vegas, Nevada, Dec.
12-14, 1984.
(26) Unimation Robotics, 1982, Unimate PUMA Robot 550/
560 series Volume I - Equipment Manual 398HlA, May 1982.

(27) Digital Equipment Corporation, 1985, "VAXELN Reference Manual", Maynard, Massachusetts.
(28) D. E. Whitney, 1969, "Resolved Motion Rate Control
of Manipulators and Human Prosthesis. "IEEE Transaction on
Man-Machine System, Vol. MMS-10, pp. 47-53.

a"(*)'--

-_-_ _ -

----

--A

Compensatw

Figure 1. A Control System With

Feedforward and Feedbadr h o p

One Period Delay Due to the Computational Lag

Figure 2.

Z M

100
0

"

I
-100

-IW

Figure 3.

Bode Diagram of Uncompensated Discrete System C ( W )

Figure 4.

Bode Diagram of Uncompensated Continuous System C(s)

- m(2)

.W(ll
Figure 9.

State Diagrams f o r the Cascade Ddcomposition of Lag-lead

Compensator

ir

loo
wos/stc

(0

m C w t w v IN

Figure 5 .

(WO

Lag-lead

GO:

Bode Diagram of C ( s ) w i t h PD Controller

100

300

200

400

500

600

700

800

900

1000

900

1000

Figure 10. Tracking errors of first joint

n
0

IM

I-

Lag-lead

a
a
0

B O
0

100

200

300

500

400

600

700

800

Figure 11. Tracking errors of second joint

~UG/SCC

mcgyC*cr "4

.005

Bode D i . g r a m of C(U) with Digital PD controller

Figure 6.

55

50

-01

~ ~ " " ' ' " " ' ' " " " ' " ' ' " ' ' " ' ' " ' ' ' " ' . ' . J

100

200

300

400

500

600

700

800

900

I000

800

900

1000

800

900

1000

800

900

1000

Figure 12. Tracking errors of third joint

I5

Figure 7 .

100

nun/scc
Bode Diagram of Lag-lead Compensator
rnLodiNn

300

200

400

500

600

700

Figure 13. Tracking errors of fourth joint

#ti

100

200

300

400

500

600

700

Figure 14. Tracking errors of fifth joint

/'
I

Figure 8.

IO
r"cr

IN

loo
aros/scc

1010

Bode Diagram of Compensated System by Lag-lead Compensator

179

100

200

Lag-lead

300

400

500

600

700

Figure 15. Tracking errors of sixth joint