364 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 12, NO.

3, MAY 2004

Disturbance-Rejection High-Precision Motion

Control of a Stewart Platform
Y. X. Su, Member, IEEE, B. Y. Duan, C. H. Zheng, Y. F. Zhang, G. D. Chen, and J. W. Mi

Abstract—A simple robust autodisturbance rejection controller

(ADRC) in linkspace is proposed to realize high precision tracking
control of a general 6 degrees of freedom (DOF) Stewart platform
in this paper. In practice, the performance of the controlled system
is limited by how to select the high-quality differential signal in
the presence of disturbances and measurement noise. Moreover,
unmodeled nonlinear friction provides degradation on the motion
precision. So, a nonlinear tracking differentiator in the feedfor-
ward path and an extended states observer in the feedback path
are designed to obtain high quality differential signal and the real
action component of unknown disturbance signals including non-
linear friction without a precise mathematical model. The non- Fig. 1. Typical Stewart platform.
linear PD (proportional derivative) controller is used to synthesize
the control action to give a superior performance. Extensive simu-
Duan [4] developed a 6 DOF parallel manipulator serving
lations and experimental results are presented to verify the effec-
tiveness and ease of engineering implementation of the proposed as a fine-tuning platform for the feed tracking for the square
method. The developed ADRC controller is simple and directly in- kilometer array (SKA) and obtained the optimal kinematically
tuitive to the practitioners. structural parameters using genetic algorithms [5].
Index Terms—Nonlinear control, observer, parallel manipula- A robot control scheme stems from two frameworks. One is
tors, state estimation, tracking control, tracking filters. to design a control on the workspace and the other is on the
linkspace. The linkspace control is a conventional control, and
is a kind of tracking control to follow the desired link length
I. INTRODUCTION computed from the position command of the mobile plate by in-

S INCE Stewart developed a prototype of 6 degrees of

freedom (DOF) parallel manipulator in 1965, this parallel
manipulator has attracted tremendous attention from re-
verse kinematics. Normally, most controllers in applications are
based on the linkspace coordinates, which they consider only an
approximated manipulator model [2], [6]. On the contrary, the
searchers involving robot manipulators, robotic end-effectors, workspace control implies that the control is designed based on
and robotic devices for high-precision robotic tasks where the the dynamics expressed by the workspace coordinates. How-
requirements of accuracy and sturdiness are more essential ever, the control based on the workspace coordinates needs in-
than those of large workspace and maneuverability [1]–[5]. A formation of a 6 DOF sensor to measure the displacement or ve-
typical 6 DOF parallel manipulator-Stewart platform, as shown locity of the mobile plate. Otherwise, it needs direct kinematics
in Fig. 1, consists of six variable-length actuators connecting a commands which rely on the numerical method or observer de-
mobile plate to a base plate with spherical joint and universal sign to estimate the 6 DOF information. It is well known that
joint, respectively. Motion of the mobile plate with respect the direct kinematics of a parallel manipulator have always been
to the base plate is produced by shortening or extending the a difficult and challenging problem [2], [7], [8]. Hence, the
actuator lengths, and the proper coordination of the actuator workspace control for parallel manipulator is seldom used in
length trajectories enables the mobile plate to track the complex practice [2], [6].
trajectory with high accuracy. Considering a control aspect, a classical PID (proportional
Merlet [2] conducted a systematic study of an in-parallel- integral derivative) control has been employed in industry but
actuated robot involving kinematics, dynamics, workspace, it does not always guarantee a high performance for a parallel
singularity and control. Kim et al. [3] designed a redundantly manipulator. Thus, a sophisticated control scheme is required
actuated parallel mechanism for rapid machining. Su and for an enhanced control performance [6], [9], [10]. However,
the geometry, high nonlinearity, and uncertainty prevent a gen-
Manuscript received May 13, 2002; revised December 30, 2002. Manuscript uine control algorithm from being developed compared to a se-
received in final form May 15, 2003. Recommended by Associate Editor D. W. rial manipulator [6]. In practice, a conventional PID control ap-
Repperger. This work was supported in part by the National Natural Science pending some new modern control methods is still the main
Foundation under Grant 50075065, the Natural Science Foundation of Shaanxi
Province under Grant 2000C22, and Robot Laboratory in Shenyang Institute of scheme to enhance the trajectory tracking precision of a parallel
Automation, Chinese Academy of Sciences under Grant RL200104. manipulator, just as done in [9].
The authors are with the School of Electro-Mechanical Engineering, Moreover, unmodeled friction has been found to have
Xidian University, Xi’an 710071, China (e-mail: [email protected];
[email protected]). unwanted influence on performance of the controlled system
Digital Object Identifier 10.1109/TCST.2004.824315 [11]–[16]. Usually, a model-based approach is employed
1063-6536/04$20.00 © 2004 IEEE
SU et al.: DISTURBANCE-REJECTION HIGH-PRECISION MOTION CONTROL OF A STEWART PLATFORM 365

to reduce the effect from friction. Even in adaptive control

of servo systems, an initial friction model is also crucial to
ensure smooth control signals and rapid parameter convergence
[14]. However, under practical conditions, because of the
inherent nonlinearity of friction, precise friction modeling has
always been a difficult and challenging problem [11]–[16].
Other efforts have been made toward more advanced control
of precision motion systems. Otten et al. [15] proposed a
neural-network based feedforward assisted PID controller to
realize high precision motion control of a linear motor. Wang
et al. [16] developed a new well-trained neural-network-based
friction model to self-tune the PI controller with suitable
gain according to each operating condition for precise motion
control of a permanent-magnet synchronous ac motor. These
methods need an explicit system model for the control design,
and the performance achievable depends on the accuracy of the
model. Fig. 2. Prototype of 6 DOF Stewart platform.
The conventional PID controller is a widely used industrial
controller which provides a combination of proportional, inte- The inertial frame - - and moving frame - - are
gral, and derivative action on the control error to form the output fixed on the base plate and mobile plate, respectively, with their
of the controller [17]. It is known that a linear combination of origins at the centers of the two circulars, as shown in Fig. 1.
these components can achieve a compromised performance in The coordinates of the joints and on the base plate and
terms of system response speed and stability. A nonlinear com- mobile plate are defined as and , separately. According to
bination can provide additional degrees of freedom to achieve a coordinate transformation, the coordinates of with respect to
much improved system performance [18]–[21]. the inertial frame can be expressed as
In this paper, a robust auto disturbance rejection controller
(ADRC) is designed in linkspace to implement high precision (1)
tracking of a general 6 DOF Stewart platform. The ADRC uses a
nonlinear tracking differentiator (TD) to yield a high quality dif- in which, denotes the orientation matrix, which can be repre-
ferential signal in the presence of disturbances and measurement sented by the following three angles with respect to the
noise in the feedforward path, and an extended states observer -axis, -axis, and -axis of the inertial frame, separately
(ESO) to provide the system’s state and the real action com- [5]
ponent of the unknown disturbances including nonlinear fric-
tion in the feedback path. The nonlinear PD (N-PD) scheme is
used to synthesize the control action. The design is carried on
directly in discrete-time and the structure chosen for the realiza-
tion permits fast computation and avoids parameter sensitivity
problems. Second, feedforward TD is motivated by difficulties where the symbols and denote and .
posed by lower-order reference trajectories which are quantized According to geometrical relationship, the vector of the actu-
at the sensor resolution [22]. The ESO serves as a soft sensor of ator can be written as
general unknown disturbances arising in the control system.
(2)
The organization of this paper is as follows. The kinematic
model of a general 6 DOF Stewart platform is developed in
in which, denotes the position vector of the inertial origin to
Section II. The design of ADRC is presented in details in Sec-
the moving origin, is the length of the th actuator, and is
tion III, with extensive simulations to validate its effectiveness.
the unit vector along this actuator.
The experimental results are presented in Section IV. Finally,
So, if the pose of the mobile plate with respect to the base
some concluding remarks are summarized in Section V.
plate is given, the appropriate length of the actuator can be cal-
culated by the above formulas, and then the six extendable actu-
II. KINEMATIC MODEL ators can be driven through certain favorable control algorithms
to realize the trajectory tracking.
The prototype of a general 6 DOF Stewart platform is shown
in Fig. 2. It is consisted of an upward base plate, a mobile plate,
III. ADRC DESIGN
and six extendable actuators. The actuators are connected to the
base plate and mobile plate with a universal joint and a spherical In practice, the performance of the controlled system is lim-
joint, respectively. The actuator is driven by an ac servomotor, ited by how to select the differential signal out of the noncontin-
and the perfect coordination of the actuator length trajectory uous measured signal with stochastic noise. In most cases, the
enables the mobile plate to track the complex trajectory with differential signal (velocity) is usually obtained by the backward
high precision. difference of the given signal (position), which is very noisy and
366 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 12, NO. 3, MAY 2004

in which, denotes a sign function, and and can be

determined as follows:

(5)

with

Fig. 3. Schematic diagram of ADRC.

limits the overall performance [23]. In a PID controller, deriva- (6)

tive action provides a degree of predictive control capability
to yield faster response without an excessive overshoot/under-
shoot, but its practical merits are often questionable. In the pres- For any bounded integrable function , the system described
ence of measurement noise or rapidly changing disturbance sig- by (3) can be used as a high performance tracking differentiator
nals, it is unclear as to whether derivative action will give any of the reference signal , that is,
control improvement. In some cases, a first-order filter is used [26].
in conjunction with the differentiator. In other cases, derivative The performance of TD is shown in Fig. 4. The reference
action may just be switched off together [14]. input is , and it is perturbed by an additive white
In this section, a nonlinear tracking differentiator (TD) [24] in noise component with the maximum amplitude of 0.01. For the
the feedforward path is used to replace the differential compo- comparison, the differential signals obtained, respectively, by
nent of the conventional PID controller for more effective and the developed TD and the general backward difference method
robust performance in the presence of noise and other distur- are shown in this figure. The simulations were programmed in
bances. An extended states observer (ESO) [25] in the feedback Matlab and run on a PC Celeron-433. The simulation step is
path is developed to act as a soft sensor of general unknown determined as , and the initial value of and is
disturbance signals arising in the control system. So, a feedfor- and . It can be seen that the differential
ward control action may then be taken in response to the un- tracking performance of TD is much better compared to that of
known disturbance before they degrade the system performance. the conventional backward difference method.
The concepts behind the TD and ESO have been proposed in a
general empirical form [24]–[26]. Although a rigorous stability
analysis has not been available, the viability for practical appli- B. ESO Design
cations has been demonstrated via simulation and experiments
[14], [26]–[29]. In recent years, due to the great advances in nonlinear con-
The schematic diagram of ADRC is shown in Fig. 3. It is trol theory, the observer-based controller has become one of the
comprised of a TD in the feedforward path, and an ESO and a most commonly used schemes in industrial applications. The
N-PD synthesizer in the feedback path. The design is carried on ESO is proposed to estimate the system’s state and the external
directly in discrete-time and the structure chosen for the realiza- disturbance, which offers several attractive features [22]. In the
tion permits fast computation and avoids parameter sensitivity absence of large model errors, it allows independent tuning of
problems. disturbance rejection characteristics and command following
characteristics. Further, compared to integral action, ESO al-
A. TD Design lows more flexibility via the selection of the order, relative de-
gree, and bandwidth of low-pass filtering. Although the tech-
The discrete form of the reference signals and of the
nique of appending disturbance states to a traditional state es-
proposed TD in the feedforward path can be expressed as the
timator to construct an ESO is well known [22], using the idea
following [26]:
developed by Han [25] to design a ESO allows a simple and
intuitive tuning of the disturbance observer loop gains indepen-
dent on the state feedback gains.
(3)
Consider a discrete uncertain nonlinear system with unknown
external disturbance
where is the sampling step, and is a velocity factor that
determine the transition. The function denotes
(7)

(4) where is an unknown function, and

is the bounded unknown external disturbance. The
SU et al.: DISTURBANCE-REJECTION HIGH-PRECISION MOTION CONTROL OF A STEWART PLATFORM 367

Fig. 5. Performance of the third-order ESO.

, and are design parameters to be

determined. The function can be defined as
Fig. 4. Comparison of TD and general differentiator.
(9)
th-order ESO for this system only using the measure-
ment output of the system , can be described as shown in To validate the effectiveness of ESO, the simulation of a dis-
(8) crete disturbed nonlinear system is carried out, which can be
described as

(10)

.. where is a bounded unknown disturbance, which is assumed to

. be expressed as . For this second-order
control system, the third-order ESO can be expressed as shown
in (11)
(8)

where are the estimated system’s state, that is,

is the estimated real
action component of the bounded unknown disturbances, (11)
368 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 12, NO. 3, MAY 2004

Fig. 6. Hardware configuration of the Stewart platform.

The initial value of , and are determined as: TABLE I

, and . The simulations were programmed SYSTEM PARAMETERS
in Matlab with a fourth-order Runge–Kutta and run on a PC
Celeron-433. The output of ESO , and and the output
, and of the system are shown in Fig. 5. It can be seen that
the system’s state and the real action component of the unknown
disturbance can be estimated precisely by the designed third-
order ESO. The simulations have validated that the ESO has
great versatility and robustness for a class of nonlinear system.
There is no tracking degradation with the variation of the param-
eters by 50% [25]. For the third-order ESO, one feasible choice
of the parameters is determined as follows:
.

C. N-PD and Disturbance Compensation

TABLE II
Once obtained the high-quality differential signal and the es- PARAMETERS FOR ADRC
timated state, the essential elements in classical PD can be re-
placed by the following two components:

(12)

Therefore, the control action of this N-PD can be used to syn-

thesize the preliminary control action, which can be described
as [21]
(13)
where , and are design parameters to be de-
termined.
Taking the feedforward control action to cancel out the un-
known external disturbance before they degrade the system per-
formance, the total control action of ADRC can be expressed as
[27]

(14)
and represent the radii of the base plate and mobile plate, re-
where is a preset constant.
spectively. stands for the splaying angle between the adjacent
Empirically, the design parameters may be determined ac-
joints on the base plate and mobile plate, respectively, repre-
cording to the following guidelines: –
sents the angle between the center of the two pair joints, and
– , and can be adjusted using the many
is the initial vertical distance of the two plates while they are
PD tuning rules available for linear PD control.
parallel and symmetrical. This parallel manipulator was initial
designed to serve as a virtual-axis machining tool [30]. At the
IV. EXPERIMENTAL RESULTS present stage, it is temporarily used as the fine tuning platform
The prototype of the experiment is shown in Fig. 2. The struc- of the feed tracking system for the scaled model experiment of
tural parameters can be listed as follows. mm, the SKA [29]. The modified hardware configuration of this con-
mm, , and mm, in which trol system is shown in Fig. 6. PCL-726 is a six-channel D/A
SU et al.: DISTURBANCE-REJECTION HIGH-PRECISION MOTION CONTROL OF A STEWART PLATFORM 369

Fig. 7. Experimental result of a step command.

Fig. 8. Experimental result of a sinusoidal command.

output card, PCL-833 is a three-axis quadrature encoder card, ments of the coarse-fine cooperation control of the feed tracking
and HY-6220 is a timer to provide external interrupt signal and system for the scaled model experiment [4], [5], [29], at present,
sampling. The extendable actuator is comprised of an AC ser- the sampling period is determined as ms.
vomotor & drive unit, compliant timing belt and screwball. The The systematic parameters for this control system are listed
servomotor and driver used is Panasonic Minas A series (low in- in Table I.
ertial moment) MSM042A1F AC servomotor and the matching After several tests, the parameters of the ADRC for this ap-
driver. The ADRC algorithms were implemented in C on an plication are determined as shown in Table II.
Advantech industrial computer with a Pentium-233. Due to the In order to verify the effectiveness of the developed ADRC,
limitation of the serial-port communication in the future experi- the following experiments are carried out. Because of the high
370 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 12, NO. 3, MAY 2004

Fig. 9. Experimental result of a low slope command.

Fig. 10. Experimental result of a step command by 25% variation of the controller parameters.

Fig. 11. Length increment and tracking error of the first actuator.

robustness of ADRC and the same selected servomotor systems, error for mm is less than mm. The
the parameters for all the six extendable actuators are deter- large tracking error results from the sampling period of
mined as identical primarily, and then make the necessary small ms, which is large compared to the time constant of the used
tuning according to the respective experimental results. So, the servomotor. For this reason, the peak current has been lowered
first test examines the tracking performance of one of the ex- to 1 A in order to reduce the motor speed. The tracking error is
tendable actuators for a step command and a periodical sinu- expected to be lower with the reduction of the sampling period
soidal signal, as shown in Figs. 7 and 8, respectively. It can be or tracking a low velocity signal, just as shown in Fig. 8(b) for
seen that the error is significant only in the transition period with mm. The tracking error is less than mm.
a fast convergence and no overshoot. The steady-state tracking The result is quite favorable.
SU et al.: DISTURBANCE-REJECTION HIGH-PRECISION MOTION CONTROL OF A STEWART PLATFORM 371

Fig. 12. Length increment and tracking error of the second actuator.

Fig. 13. Length increment and tracking error of the third actuator.

Fig. 14. Length increment and tracking error of the fourth actuator.

To address the effectiveness of friction compensation, the variation of the above selected parameters. The experimental
third experiment of the tracking performance to a low slope result is shown in Fig. 10.
command is carried out. Friction, being the dominant, error-in- Compared Figs. 7 and 10, it can be seen that the tracking per-
ducing nonlinearity, will make the dominant contribution to the formance is no degradation with the variation of the selected
tracking with low velocity [18]. The experimental results are parameters by 25%. The developed ADRC has high robustness
shown in Fig. 9. It can be seen that the tracking error is only to the variation of the parameters, which makes it is very con-
mm, which is a superior performance for low velocity. venient to apply in practice.
To test the performance robustness of the developed ADRC, After the above experiments of one of the extendable actua-
the tracking performance of the same extendable actuator as the tors, the Stewart platform is reassembled together, and the mo-
first test for a step command is concerned, with 25% increased bile plate is desired to track an - planar circle with the radius
372 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 12, NO. 3, MAY 2004

Fig. 15. Length increment and tracking error of the fifth actuator.

Fig. 16. Length increment and tracking error of the sixth actuator.

It can be seen that, once entered into the steady tracking, the
tracking error of the actuator is less than mm, and the
tracking errors in coordinates are less than mm.
From the above experimental results, it may be concluded
that high tracking precision can be achieved. Furthermore, the
ADRC controller has high robustness with the variation in the
range of 25% of the selected parameters.

V. CONCLUSION
A high robust ADRC controller in linkspace is proposed to re-
alize the high precision trajectory tracking of a general 6 DOF
Stewart platform. The ADRC uses a nonlinear tracking differen-
tiator (TD) to yield a high-quality differential signal in the pres-
ence of disturbances and measurement noise in the feedforward
path, and an ESO in the feedback path to serve as a soft sensor
of general unknown disturbances arising in the control system,
Fig. 17. Trajectory of the X -Y planar circle. including nonlinear friction. The N-PD scheme is used to syn-
thesize the control action. This design is carried on directly in
of 12 cm at the linear velocity of 17 mm/s. The length increment discrete-time and the structure chosen for the realization per-
of the actuators and its tracking error is shown in Figs. 11–16, mits fast computation and avoids parameter sensitivity prob-
respectively. With the measurement of the center of the mobile lems. The experimental results show that high precision tracking
plate using Leica TC1101 total station [29], [31], the tracking for a general 6 DOF Stewart platform can be guaranteed. Fur-
of the mobile plate is shown in Fig. 17. The tracking errors of thermore, minor prior knowledge of the system is required; the
position in coordinates is shown in Fig. 18. ADRC controller can be applied to other servo system requiring
SU et al.: DISTURBANCE-REJECTION HIGH-PRECISION MOTION CONTROL OF A STEWART PLATFORM 373

Fig. 18. Position errors in X; Y coordinates.

high tracking precision. Above all, the structure of the controller [10] C. C. Nguyen, S. S. Antrazi, and Z.-L. Zhou, “Adaptive control of a
is simple and directly intuitive to the practitioners. stewart platform-based manipulator,” J. Robot. Syst., vol. 10, no. 5, pp.
657–687, 1993.
[11] B. Armstrong-Helouvry, P. Dipont, and C. C. de Wit, “A survey of
models, analysis, tools and compensation methods for the control of
ACKNOWLEDGMENT machines with friction,” Automatica, vol. 30, no. 6, pp. 1083–1138,
1994.
The authors would like to express their sincere gratitude to [12] J. T. Teeter, M.-Y. Chow, and J. J. Brickley Jr., “A novel fuzzy friction
Prof. J. Q. Han at Academy of Mathematics and System Sci- compensation approach to improve the performance of a DC motor con-
trol system,” IEEE Trans. Industr. Electron., vol. 43, pp. 113–120, 1996.
ences, Chinese Academy of Sciences, China, for his excellent [13] R. H. A. Hensen, M. J. G. van de Molengraft, and M. Steinbuch, “Fre-
guidance and detailed helps in the determination of the param- quency domain identification of dynamic friction model parameters,”
eters of the controller. They would also like to thank Prof. H. S. IEEE Trans. Contr. Syst. Technol., vol. 10, pp. 191–196, Mar. 2002.
[14] K. K. Tan, T. H. Lee, and H. X. Zhou, “Micro-position of linear-piezo-
Ding, Dr. J. Hao, and Ph.D. candidates C. G. Li, J. Q. Li, and electric motors based on a learning nonlinear PID controller,” Trans.
T. Fu at Beijing Institute of Technology, for their help in the Mechatron., vol. 6, no. 4, pp. 428–436, 2001.
experiment. Also, they would like to express their appreciation [15] G. Otten, T. J. A. de Vries, J. van Amerongen, A. M. Rankers, and E.
W. Gaal, “Linear motor motion control using a learning feedforward
to Prof. Y. Zheng, Prof. Z. G. Xia and Master candidate Y. B. controller,” Trans. Mechatron., vol. 2, pp. 179–187, 1997.
Luo for their help in the measurement of the experiments. They [16] G.-J. Wang, C.-T. Fong, and K. J. Chang, “Neural-network-based self-
are also grateful to the anonymous reviewers for the numerous tuning PI controller for precise motion control of PMAC motors,” IEEE
Trans. Industr. Electron., vol. 48, pp. 408–415, Feb. 2001.
comments and suggestions, which have lead to significant im- [17] K. J. Astrom and T. Hagglund, PID Controllers: Theory, Design, and
provements of this paper. Tuning, 2nd ed, NC: Instrument Society of America, 1995.
[18] B. Armstrong, D. Neevel, and T. Kusik, “New results in NPID control:
Tracking, integral control, friction compensation and experimental re-
REFERENCES sults,” IEEE Trans. Contr. Syst. Technol., vol. 9, pp. 399–406, Mar. 2001.
[19] W.-H. Chen, D. J. Balance, P. J. Gawthrop, J. J. Gribble, and J. O’Reilly,
[1] D. Stewart, “A platform with six degrees of freedom,” in Proc. Institu- “Nonlinear PID predictive controller,” in Inst. Elect. Eng. Proc.–Contr.
tion of Mechanical Engineers, vol. 180, 1965, pp. 371–386. Theory Applicat., vol. 146, 1999, pp. 603–611.
[2] J.-P. Merlet, Parallel Robots. Dordrecht, The Netherlands: Kluwer [20] Y. Xu, J. M. Hollerbach, and D. Ma, “A nonlinear PD controller for
Academic, 2000. force and contact transient control,” IEEE Contr. Syst. Mag., vol. 15, pp.
[3] J. Kim, F. C. Park, S. J. Ryu, J. Kim, J. C. Hwang, C. Park, and C. C. 15–21, 1995.
Iurascu, “Design and analysis of a redundantly actuated parallel mech- [21] J. Q. Han, “Nonlinear PID controller,” Acta Automatica Sinica, vol. 20,
anism for rapid machining,” IEEE Trans. Robot. Automat., vol. 17, pp. no. 4, pp. 487–490, 1994. In Chinese.
423–434, 2001. [22] C. J. Kempf and S. Kobayashi, “Disturbance observer and feedforward
[4] Y. X. Su and B. Y. Duan, “The mechanical design and kinematics accu- design for a high-speed direct-drive positioning table,” IEEE Trans.
racy analysis of a fine tuning stable platform for the large spherical radio Contr. Syst. Technol., vol. 7, pp. 513–526, Sept. 1999.
telescope,” Mechatronics, vol. 10, no. 7, pp. 819–834, 2000. [23] L. Xu and B. Yao, “Output feedback adaptive robust precision motion
[5] Y. X. Su, B. Y. Duan, and C. H. Zheng, “Genetic design of kinematically control of linear motors,” Automatica, vol. 37, no. 7, pp. 1029–1039,
optimal fine tuning Stewart platform for large spherical radio telescope,” 2001.
Mechatronics, vol. 11, no. 7, pp. 821–835, 2001. [24] J. Q. Han and W. Wang, “Nonlinear tracking differentiator,” J. Syst. Sci.
[6] D. H. Kim, J.-Y. Kang, and K.-I. Lee, “Robust tracking control design Math. Sci., vol. 14, no. 2, pp. 177–183, 1994. In Chinese.
for a 6 DOF parallel manipulator,” J. Robot. Syst., vol. 17, no. 10, pp. [25] J. Q. Han, “The “extended state observer” of a class of uncertain
527–547, 2000. system,” Contr. Decision, vol. 10, no. 1, pp. 85–88, 1995. In Chinese.
[7] R. Boudreau, S. Darenfed, and C. M. Gosselin, “On the computation [26] J. Q. Han and L. L. Yuan, “The discrete form of tracking-differentiator,”
of the direct kinematics of parallel manipulators using polynomial J. Syst. Sci. Math. Sci., vol. 19, no. 3, pp. 268–273, 1999. In Chinese.
networks,” IEEE Trans. Syst., Man, Cybern., pt. A, vol. 28, no. 2, pp. [27] J. Q. Han, “Auto-disturbance-rejection controller and its applications,”
213–220, 1998. Contr. Decision, vol. 13, no. 1, pp. 19–23, 1998. In Chinese.
[8] L. Baron and J. Angles, “The direct kinematics of parallel manipulators [28] Y. Huang and J. Q. Han, “Analysis and design for the second order non-
under joint-sensor redundancy,” IEEE Trans. Robot. Automat., vol. 16, linear continuous extended states observer,” Chinese Sci. Bull., vol. 45,
pp. 12–19, 2000. no. 21, pp. 1938–1944, 2000.
[9] D. Xiu, “Trajectory Control of a New Class of CNC Machine Tools,” [29] Y. X. Su, “Optimization, Analysis, and Control of the Stewart Platform
Ph.D. dissertation, Faculty of the College of Engineering, Florida At- for the SKA,” Ph.D. disertation, School of Electro-Mechanical Engi-
lantic University, 1998. neering, Xidian University, China, 2002. In Chinese.
374 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 12, NO. 3, MAY 2004

[30] H. S. Ding, D. H. Wang, and H. L. Liu, “Kinematic simulation modeling C. H. Zheng was born in Xi’an, China, in 1969. She
and its implementation in BKX-I variable-axes NC machine tool,” J. received the B.S. degree in electrical engineering
Beijing Inst. Technol., vol. 21, no. 4, pp. 434–439, 2001. In Chinese. from Xidian University, Xi’an, China, in 1991 and
[31] Guidelines to Leica TC1101 Total Station, 2001. the M.S. degree in automatic control and application
from Gansu University of Technology, Lanzhou,
China, in 1996. She is currently working toward the
Ph.D. degree in signal processing in the National
Key Laboratory of Radar Signal Processing at
Xidian University.
Y. X. Su (M’03) was born in Liaoning, China, in From June 1996 to January 2000, she was a
1969. He received the B.S. and M.S. degrees in Teaching and Researching Assistant in Xi’an
mechanical engineering from Gansu University of Petroleum Institute, Xi’an, China. From February 2000 to present, she has been
Technology, Lanzhou, China, in 1992, and 1995, a Lecturer in Xidian University. Her research interests include automation,
respectively, and the Ph.D. degree in mechatronics fuzzy signal processing, and support vector machine.
from Xidian University, Xi’an, China, in 2002.
From April 1996 to June 1998, he was a Teaching
and Researching Assistant in Xidian University.
Y. F. Zhang was born in Inner Mongolia, China. She received the B.S. degree in
From July 1998 to June 2000, he was a Lecturer in
mechanical engineering from Gansu University of Technology, Lanzhou, China,
Xidian University, where he has been an Associate
in 1998, and the M.S. degree in mechatronics from Xidian University, Xi’an,
Professor in the School of Electro-Mechanical En-
China, in 2003.
gineering, since July 2000. He has been a Research Fellow in City University
Her research interests include automation.
of Hong Kong, Hong Kong, since March 2003. His research interests include
mechatronics, motion control, robotics, and design optimization.

G. D. Chen was born in Hangzhou, China, in 1966.

He received the B.S. and M.S. degrees in mechanical
engineering from Xidian University, Xi’an, China, in
1987 and 1990, respectively.
B. Y. Duan was born in Hebei, China, in 1955. He Since 1990, he has been a faculty member of
received the B.Sc. and M.Sc. degrees from North- the University of Xidian, where he is currently an
west Telecommunication Engineering Institute, Associate Professor. His research interests include
Xi’an, China, in 1981 and 1984, respectively, and linear and nonlinear stochastic dynamical systems
the Ph.D. degree from Xidian University, Xi’an, and adaptive control.
China, in 1989. Mr. Chen is a recipient of the 1998 National Sci-
From 1991 to 1994, he was a Postdoctoral Fellow ence Progress Award.
at the University of Liverpool, Liverpool, U.K.
During that time, he also worked at Hokkaido
University, Hokkaido, Japan, as a Research Fellow
from 1993 to 1994. He has made a short academic J. W. Mi was born in Gansu, China. He received the
visit to France and Australia. Currently, he serves as Professor of Mechanical B.S. degree in mechanical engineering from Xidian
Engineering at Xidian University and is also the President of the university. His University, Xi’an, China, in 2001. He is currently
research has concentrated on structural optimization, CAD, and mechatronics. working toward the M.S. degree in mechatronics at
Prof. Duan is a Member of International Society of Computational Me- Xidian University.
chanics, Fellow of Chinese Institute of Electronics, Vice Chairman of Society His research interests include software engineering
of Electromechanical Engineering in China; Member of Directing Committee and adaptive control.
of Mechanical and Electronic Engineering in China; Member of Expert Group
of CAD Applied Engineering of Shaanxi province; Chairman of Society of
Electromechanical Engineering of Shaanxi province.