Proceedings of the 17th World Congress

The International Federation of Automatic Control

Seoul, Korea, July 6-11, 2008

Robust Output Feedback Controller Scheme for a Class of Uncertain Nonlinear

Systems

J. Kuvulmaz* E. Zergeroglu**

*Yildiz Technical University,Istanbul,34349

TURKEY (Tel:90-212-2597070; e-mail :janset@ yildiz.edu.tr).
**Gebze Institute of Tecnology,Kocaeli, 41400 TURKEY (e-mail:
[email protected])

Abstract: In this study, we present a new continuous output feedback type controller mechanism for the
tracking problem of a class of uncertain nonlinear systems. The proposed strategy requires the
uncertainties of the dynamical system to be first order differentiable and achieves semi-global asymptotic
tracking when only the system outputs are measurable. The Controller design is based on a Lyapunov-type
stability argument. Simulation studies on a two link planar robotic system are presented to illustrate the
feasibility of the proposed strategy.

of PI type controllers for many practical systems, a nonlinear

1. INTRODUCTION
Proportional-Integral (NPI) type controller was presented for
The tracking control problem for uncertain nonlinear systems a class of uncertain dynamical systems (Ortega et.al., 2002).
has been extensively studied for decades. This extensive Though the proposed controller had some discontinuities,
research interest is not only due to the theoretically owing to its simple structure has fewer parameters to tune.
challenging nature of the problem but also due to practical Moreover, the effects of the controller gains on the system
needs, as the mathematical model for nearly all dynamical are predictable owing to the similarities of the controller
systems in control theory contain some uncertainty and structure to linear PI type counterpart. Therefore the
feedback alone is not enough for preferential performance. implementation is quite easy. In 2004, Xian-et.al. presented a
Researchers have proposed many different types of continuous tracking controller strategy for second order
controllers depending on the nature of the uncertainty. To differentiable uncertain dynamics systems. The proposed
give a few examples, when the parametric uncertainties are controller strategy achieves asymptotic tracking and was
constant or slowly time-varying and the function containing backed up by a novel Lyapunov based analysis. However, our
the overall uncertainties can be linearly parametrizable, due experience with the proposed method have shown that the
to its continuous nature, adaptive control (Kristic et al., 1995; control input signal, similar to variable-structure controller,
Sastry and Bodson, 1989) would be the preferred choice. has high order frequency components, which in practical
Unfortunately in adaptive controllers each uncertain implementations might trigger chattering like phenomenon.
parameter has to be adapted separately, making the tuning
The development of controllers that only require output
process (due to the parameter update gains) moderately
measurements (i.e., output feedback (OFB)) has received
tedious. On the other hand when the uncertainties of the
considerable interest in literature due to the advantages of
system are bounded by some known norm-based function,
eliminating many sensors (e.g., reduced system complexity,
the theory of robust control (Qu, 1998) can be applied. From
cost, and noise). Global1 solutions to the OFB link position
the implementation point of view, unlike adaptive controllers,
setpoint control problem have been presented by several
the robust controllers have fewer gains to deal with.
researchers. For example, model-based global regulating
However, in most cases the convergence of the tracking error
OFB controllers were proposed in (Berguis and Nijmeijer,
into an ultimate bound can be assured with the robust
1993; Burkov, 1993; Kelly, 1993). With the intent of
controllers and over shrinking this ultimate bound, for better
overcoming the requirement of exact model knowledge, an
performance, might cause undesirable system responses (like
OFB regulator was designed (Ortega et al., 1995); however,
chattering). When the uncertainties of the dynamics are
periodic, learning controllers (Arimoto et al., 1984; Messner
et al., 1991; Dixon et al., 2002) can be used, the down part
is, there are only limited number of systems with periodic 1
Global position tracking means that the controller must
dynamics (same is also true for the desired dynamics). drive the link position error to zero for any finite, initial
Recently researches have proposed alternative methods to position and velocity tracking errors, with no conditions on
overcome the aforementioned disadvantages of adaptive and the size of the initial tracking errors.
robust controllers. Motivated by the satisfactory performance

978-1-1234-7890-2/08/$20.00 © 2008 IFAC 14125 10.3182/20080706-5-KR-1001.3165

17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

the stability result was semi-global2 asymptotic. On the other differentiable nonlinear, uncertain function and u (t ) is the
hand, a limitation that exists in almost all of the proposed control input.
OFB link position tracking controllers is the semi-global
nature of the stability results. To give a few examples, a Property 1: The inertia matrix can be upper and lower
model-based observer was used to construct a semi-global bounded by the following inequalities (Lewis et al., 1993)
exponential link position tracking controller (Lim et al.,
1996). Variable structure OFB controllers were designed to m1 ≤ m ( x) ≤ m 2 (2)
compensate for parametric uncertainty (Canudas de Wit and where m1 and m2 are positive constants.
Fixot, 1991; Canudas de Wit and Slotine 1991). Filter-based
robust control schemes which produce semi-global, Our control objective is to ensure that the state signal x(t )
uniformly ultimately bounded (UUB) were designed in would track the given smooth reference trajectory x d (t ) . To
(Berghuis and Nijmeijer, 1994; Qu et al., 1995; Yuan and
Stepanenko, 1991). Then adaptive OFB controllers were quantify this objective we defined the tracking error signal
presented in (Burg et al.,1994; Burg et al., 1996; Kaneko and e(t ) in the following form
Horowitz, 1997; Zergeroglu et al., 1999) that yield semi-
e = xd − x (3)
global asymptotic link position tracking in the presence of
parametric uncertainty. In 1996, Loria developed a model- In our analysis we will utilize the common assumption that
based controller which produces global uniform asymptotic the reference trajectory signal x d (t ) and its first three time
tracking but the proposed method is only valid for a one derivatives are always bounded (i.e.
degree-of-freedom (DOF) system. Then, a global OFB x d (t ), x& d (t ), &x&d (t ), &x&&d (t ) ∈ L∞ ).
adaptive controller was designed for n-DOF robot
manipulator (Zhang et al.,2000) Finally, a robust OFB Velocity variables can not be measured because of that the
tracking controller was proposed with a global, uniformly related variables are obtained by filtering technique of
ultimately bounded link position tracking (Dixon et al.,2004). position error. While position tracking error is the input, the
In this work, we extend our previous result (Kuvulmaz and velocity tracking error is the output of the system. The
Zergeroglu, 2007) to output feedback case. When compared proposed filter dynamics is as follows
to (Ortega et al., 2002) our approach does not possess any r f = p − (k1 + 1) e (4)
singularities and when compared to (Xian et al., 2004) can
compensate the uncertainties for a larger class of nonlinear p& = − p + e − e f − ( k1 + 1) r f (5)
systems. However, our approach requires a high gain e& f = − e f + r f (6)
condition on the feedforward compensation gain which might
be considered as a theoretical weakness. where the filter output r f (t ) ∈ℜ will be used for the link
velocity variable. p(t ), e f (t ) ∈ ℜ are the auxiliary variables
The remaining of the paper is organized as follows: the
model under consideration and the control problem are stated used to establish the velocity variable. k1 is the positive
in Section 2. Section 3 contains the error system control gain which is defined as
development, while controller design with the stability
analysis to ensure asymptotic tracking and boundedness of 1
k1 = (k n + 1) (7)
the closed loop system are given in section 4. Simulations m1
performed on a two-link, planar robotic mechanism are where k n is the positive nonlinear damping gain. In order to
presented in Section 5 and lastly some concluding remarks
with possible future research are presented in Section 6. form the open loop position error system, dynamics of the
filter output would be obtained. Taking the time derivative of
2. PROBLEM STATEMENT (4) and inserting for p& from (5) we get the dynamics of the
filter output
For the ease of presentation3, we consider a second order
single-input single-output dynamical system having the r& f = − (k1 + 1)e& − p + e − e f − (k1 + 1) r f (8)
following form Inserting the value of p from (4) in equation (8),
m( x) &x& + f ( x, x& , θ ) = u (t ) (1)
r&f = − (k1 + 1)η + e − e f − r f (9)
where x is the output and the notations x& , &x& are used to
identify the first and second order derivatives of the output where the filtered error signal η (t ) ∈ ℜ is defined as
respectively. m( x) 〉 0 and f ( x, x& , θ ) is first order η = e& + e + rf (10)
If (10) is rearranged, we get the error dynamics as
e& = − e − r f + η (11)
2
In a semi-global stability result, a control gain often has to
In addition, we define an auxiliary term named integral effect
be adjusted according to the “size” of the initial conditions.
3 injection term ξ as
Extension to multi-input multi output and higher order
versions are also possible with a considerably small effort.

14126
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

t Based on the subsequent stability analysis we propose the

ξ (t ) = e (t ) + ∫ (e(τ ) + r f (τ )) dτ + e(0) (12) following nonlinear PI control law
0
u (t ) = − (k1 + 1) r f + (k i + 1) e + β tanh (e + e f )
It is obvious that the equation (12) can also be expressed as
t
(22)
∫
+ k i (e f + r f ) (τ ) dτ
t
ξ (t ) = ∫ η (τ ) dτ (13) 0
0
where k1 , k i and β are positive control gains. Substituting
(22) into (18) the closed loop dynamics for the filtered
tracking error is obtained as
3. ERROR SYSTEM DEVELOPMENT
1 ~
Taking the time derivative of (10), multiplying both sides of m η& = − m& η + N + N d − k mη + ( k + 1) r
2 1 1 f
the resultant equation by m ( x) and inserting for t (23)
&x& (t ), r&f , e&( f ) from (1), (9) and (11) respectively, we have − k i ∫ η (τ ) dτ − β tanh (e + e ) − e
0 f
m η& = − k1 mη + N − m (2 r f + e f ) − u (14) 4. ANALYSIS
1 Before going into the stability analysis, we state the
After adding and subtracting m& η term to the both sides of
2 following Lemma which will be useful by helping us prove
the (14), we get the Lyapunov candidate function is lower bounded.
m η& = − k1 mη + N − m (2 r f + e f ) − u Lemma1: Consider the auxiliary function β I (t ) defined as
1 1 (15)
+ m& η − m& η t
2 2 β I := − ∫ wI + δ b (24)
where the auxiliary signal is defined as 0

N = m &x&d + f ( x, x&,θ ) (16) with

At this point we define the desired version of the auxiliary wI :=η ( N d − β tanh(e + e f )) (25)
signal N , N d such that where the auxiliary constant term δ b , explicitly given in the
Nd := N (17) following form
x& = x& d , x = x d
δ b := β (ln cosh(e(0) + e f (0)) +1)
Note that due to assumption that the reference trajectory term (26)
x d is third order differentiable, the newly defined “desired” − (e(0) + e f (0)) N d (0).
version of the auxiliary term can be proven to be at least first When the constant scalar control gain β of (26) is selected to
order differentiable (i. e. N d (t ), N& d (t ) ∈ L∞ ) . Adding and satisfy
subtracting N d to the right hand side of (17), we obtain
β 〉 k Nd [ ∞
+ N& d
∞
] (27)
~ 1
m η& = − k1 mη + N + N d − u − m& η (18) with the high but bounded design gain k defined as
2
~ ⎧⎪ ⎫⎪
where the function N is defined as 1
k = max ⎨(1 + ε ), ⎬ (28)
⎪⎩ tanh( w)
~ 1 w≠ 0 ⎪⎭
N := N − N d − m (2 rf + e f ) + m& η (19)
2 where ε 〉 0 . Then β I (t ) will always be lower bounded by
Remark 1: Since the auxiliary function N defined in (16) is
~ zero (i. e. β I ≥ 0 ) or equivalently
continuously differentiable, we can show that N can be upper
t
bounded in the following manner:
~ ∫ w (τ ) dτ ≤ δ
I b
(29)
N ≤ ρ ( j) j (20) 0

Proof: For presentation easiness, a simple transformation is

where ⋅ denotes the standard Euclidean norm, j (t ) is the given in the following form by using (6) and (10)
vector function as η = w+
& w (30)
[
j := e e f rf η ] (21) w=e + ef (31)

and ρ (⋅) is a positive defined non-decreasing bounding To prove the Lemma, we take only the part inside the integral in
(24) and substitute (30) to obtain
function.

14127
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

t t
and s (t ) is the vector function defined as
∫ w (τ ) dτ = ∫ [ w (τ ) ( N
I d (τ ) − β tanh( w (τ )))] dτ
0
t
0

d w (τ ) (32)
[
s = jT z ξ ]
T
(39)
+ ∫
0
dτ
N d (τ ) dτ
Proof: To prove the Theorem we define a non negative
t
function of the form
d w (τ )
− β ∫
0
dτ
tanh ( w (τ ))dτ . 1 1
V ( t ) = mη 2 + Z + k i ξ 2 (40)
2 2
evaluating the third term and integrating the second term on
the right hand side by parts, we have with Z (t ) is defined to have the form
t t
⎡ dN d (τ ) ⎤ Z := β P + β I (41)
∫ wI (τ ) dτ =
0
∫ w (τ ) ⎢⎣ N
0
d (τ ) −
dτ
− β tanh( w(τ ) ⎥ dτ
⎦ where β P is selected as
+ [ w (t ) N d (t ) − β (ln (cosh ( w(t ))) + 1 ] (33)
1
1 1
+ [ β (ln (cosh ( w ( 0 ))) + 1) − w ( 0 ) N d ( 0) ] , β P = e 2 + e 2f + r f2 (42)
22 2
where the β term has been added and subtracted from the
and β I term is the lower bounded function defined in (24).
right hand side. Now we can upper bound (33) in the
following way We can upper and lower bound (40) by using λ1 and λ 2 terms
in the following way
t t
⎡ dN (τ ) ⎤
∫0 I
w (τ ) dτ ≤ ∫0 w(τ ) ⎢⎢⎣ Nd (τ ) + d (dτ ) − β tanh( w(t)) ⎥⎥⎦ dτ 2
λ1 s ≤V ≤ λ2 s
2
(43)

[
+ w (t ) Nd (t ) − β ( ln(cosh ( w (t ))) +1) ] (34) Taking the time derivative of (40) substituting (7) and (23)
we obtain
+ [β ( ln(cosh ( w (0))) +1) − w (0) Nd (0)]
~
V& = − e2f − rf2 − e2 −η 2 − kn η 2 +η N (44)
Where the w (t ) tanh( w (t )) ≥ 0 has been utilized. Notice that And from (20), we can state the following upper bound for
in (34), the integral term (the first line) on the right hand side the time derivative of function V given in (40)
of the inequality will exactly be zero when w(t ) = 0 and will
have negative values for all other values of w(t ) when the
2
[
V& ≤ − j + η ρ ( j ) j − kn η 2 ] (45)
2
controller gain β is selected to satisfy (27). Similarly the ρ 2 ( j) j
Adding and subtracting term to the right hand
same selection of β will also ensure the negative semi 4 kn
definiteness of second term on the right hand side of (34) (as side of (45) yields
k (ln (cosh ( w)) + 1) − w ≥ 0 for k ≥ (1 + ε ) ). And from the 2
2 ρ 2 ( j) j
definition of δ b given in (26), it is straight forward to show V& ≤ − j +
4 kn
that (29) holds. (46)
⎡ ρ 2 ( j) j ⎤
2

We are now ready to present the following Theorem − ⎢kn η 2 − η ρ ( j ) j + ⎥

⎣⎢
4 kn ⎥
⎦
Theorem 1: The control law of (22) ensures that all the In the above equation the terms in the brackets is square of
signals in the closed loop system of (23) will remain bounded
⎛ ⎞
and the semi-global asymptotic convergence and stability of ⎜ k η − ρ ( j ) j ⎟ and due to the negative sign on its front,
the error signal e(t ) is guaranteed in the sense that ⎜ n
2 k n ⎟⎠
⎝
lim e(t ) = 0 (35) is always negative, this enables us to further upper bounded
t →∞
(46) to have the following form
provided that the control gain β is adjusted according to (27)
⎡ ρ 2 ( j) ⎤ 2
and damping gain k n is selected to satisfy V& ≤ − ⎢1 − ⎥ j (47)
⎣ 4 kn ⎦
λ2 When the damping gain k n is selected sufficiently large we
kn 〉 ρ 2 ( s ( 0) ) (36)
λ1 can obtain
where λ1 and λ 2 are defined as ρ2( j )
2
V& ≤ −ψ j , kn 〉 (48)
1 4
λ1 = min {m1 , 1, k i , 0} (37)
2 for some ψ 〉 0 . Equation (48) can also be defined as
1
λ 2 = max {m 2 , 1, k i , δ b } (38)
2

14128
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

ρ2( s ) ⎡ xd 1 (t ) ⎤ ⎡0.7 sin (t ) (1 − exp (−0.3 t 3 ))⎤

V& ≤ −ψ j
2
, kn 〉 (49) ⎢ ⎥=⎢ ⎥ rad (54)
⎣ xd 2 (t )⎦ ⎢⎣1.2 sin (t ) (1 − exp (−0.3t )) ⎥⎦
3
4
With the help of (43) we can rearrange (49) in the following For both joints the initial position values are selected to be
way 0.5236 rad (30 degrees). After the tuning process a fair
performance for the controller was achieved when the
2 1 2 ⎛⎜ V (t ) ⎞⎟ controller gains were selected as follows
V& ≤ −ψ j , kn 〉 ρ (50)
4 ⎜⎝ λ1 ⎟⎠
k n = 40 , ρ = 5 , k1 = 500 k i = 200 , β = 100 (55)
Notice that, V (0) is the maximum value of V (t ) . So we can The simulation results are shown in Figures 1-2. Figure 1
adjust k n according to V (0) such that presents the link position tracking errors while Figure 2
presents the control torques applied to each link motor during
2 1 2 V (0) the simulations. As can be observed from the simulation
V& ≤ −ψ j , kn 〉 ρ ( ) (51)
4 λ1 results the control gain, β , is not high compared to the other
controller gains. During the simulation studies we have also
and finally when the damping gain is selected as follows we
observed that cranking up β does not affect on the controller
can upper bound the time derivative of V (t ) according to
performance much. Therefore, it is our belief that the high
initial conditions
gain condition given in (27) is only a theoretical drawback.
2 1 2 λ2
V& ≤ −ψ j , kn 〉 ρ ( s (0) ) (52) 6. CONCLUSION
4 λ1
In this paper4, we have presented a new output feedback
From the structure of (40) and (52) it is clear that
controller strategy for the tracking control of a class of
( )
V ≥ 0 , V& ≤ 0 V ∈ L∞ and due to the structure of V all the uncertain nonlinear systems. Despite the parametric
signals contained in V , are also bounded, that is uncertainties in the system dynamics, both constant and/or
e(t ), e f (t ), r f (t ), η (t ) ∈ L∞ . From (11) and (21) we can time-varying, the proposed controller guarantees semi-global
asymptotic tracking, and only requires the parametric
conclude that e& (t ), j (t ) ∈ L∞ . Thus the control input signal of
uncertainties to be first order differentiable. Due to the
(22) is bounded. It follows from (23) that η&∈ L∞ that is all continuous nature of the controller the proposed method can
signals in the closed loop error system are bounded. Finally, also be used in backstepping type controller designs.
from the structure of (52) we can conclude that j , therefore Moreover since the controller proposed can be formulated as
e∈ L 2 . With the above information and direct application of a N-PI type controller, implementation and gain tuning are
straightforward compared to other uncertainty compensating
Barbalat’s Lemma (Sadegh and Horowitz, 1990) we can controllers in the literature. However the proposed controller
wrap up that the tracking error term, e(t ) will approach to requires a theoretical high gain condition on the
zero as timpe approaches to infinity, as proposed in (35). compensation gain, β , given in (27). Future studies will
concentrate on reducing this high gain condition.
5. SIMULATION RESULTS
To illustrate the performance of the proposed control scheme,
we have performed simulations on a 2-link, revolute, direct-
drive robot manipulator with the following dynamics (DDM
Operations Manual, 1992)
⎡τ 1 ⎤ ⎡ p1 + 2 p 3 cos( x 2 ) p 2 + p 3 cos( x 2 )⎤ ⎡ &x&1 ⎤
⎢ ⎥=⎢ ⎥ ⎢ && ⎥
⎣τ 2 ⎦ ⎣ p 2 + p 3 cos( x 2 ) p2 ⎦ ⎣x2 ⎦
⎡− p 3 sin( x 2 ) x& 2 − p 3 sin( x 2 )( x&1 + x& 2 )⎤ ⎡ y& 1 ⎤
+⎢ ⎥⎢ & ⎥ (53)
⎣ p 3 sin( x 2 ) x&1 0 ⎦⎣ y2 ⎦
⎡ f d1 0 ⎤ ⎡ x&1 ⎤
+⎢ ⎥⎢ ⎥
⎣0 f d 2 ⎦ ⎣ x& 2 ⎦
Fig. 1. Position Tracking Errors
where x , x& , &x& are link position, velocity and acceleration
vectors respectively. The unknown but constant parameters
representing the robot parameters are taken as p1=3.473
[kg.m2], p2 = 0.193 [kg.m2], p3 = 0.242 [kg.m2], fd1 = 5.3
[Nm.s], fd2 = 1.1 [Nm.s] during the simulation studies. The 4
robot’s reference trajectory is selected as This work was mainly supported by the grands provided
from The Scientific Technological Research Council of
Turkey-TÜBİTAK Project No: 104E061

14129
17th IFAC World Congress (IFAC'08)
Seoul, Korea, July 6-11, 2008

Kelly, R. (1993). “A Simple Set-Point Robot Controller by

Using Only Position Measurements,” Proc. IFAC World
Congress, Sydney, Australia (July, 1993) pp. 173-173.
Kristic, M., I. Kanellakopoulos and P. Kokotovic (1995).
Nonlinear and Adaptive Control Design, New York:
Wiley.
Kuvulmaz, J. and E. Zergeroglu (2007). “A New High-Gain
Continuous Controller Scheme for a Class of Uncertain
Nonlinear Systems” Accepted to appear in Proc. IEEE
Conf. Decision and Control, New Orleans, LA (Dec.
2007 ).
Lewis, F., C. Abdallah and D. Dawson (1993 ). Control of
Fig. 2. Control Torque Inputs Robot Manipulators (New York: MacMillan Publishing
Co.
Lim,S. Y., D. M. Dawson and K. Anderson (1996). “Re-
Examining the Nicosia-Tomei Robot Observer-Controller
from a Backstepping Perspective,” IEEE Trans on
REFERENCES Control Systems Technology 4, N.3, 304-310.
Loria, A. (1996). “Global Tracking Control of One Degree of
Arimoto, S., S. Kawamura, and F. Miyazaki (1984). Freedom Euler-Lagrange Systems Without Velocity
“Bettering Operations of Robots by Learing,” J. Robot Measurements,” European J. of Control 2, No. 2 144-151.
Syst., vol. 1, no. 2, pp. 123–140. Messner, W., R. Horowitz, W. W. Kao and M. Boals (1991).
Berguis, H. and H. Nijmeijer (1993). “A Passivity Approach “A new Adaptive Learning Rule,” IEEE Trans. on
to Controller-Observer esign for Robots,” IEEE Trans. Automatic Control, cilt: 36, s: 188–197.
Robotics and Automation 9 No. 6, 740-754. Ortega, R., A. Loria and R. Kelly (1995). “A Semiglobally
Berghuis, H. and H. Nijmeijer (1994). “ Robust Control of Stable Output Feedback PI2D Regulator for Robot
Robots via Linear Estimated State Feedback,” IEEE Manipulators,” IEEE Trans. Automatic Control 40, No 8,
Trans. Automatic Control 39 No. 10 2159-2162. 1432-1436.
Burg, T., D. Dawson and P. Vedagarpha (1994). “A Ortega,R., A. Astolfi and N. E. Barbanov (2002). “Nonlinear
Redesigned DCAL Controller Without Velocity PI Control of Uncertain Systems: An Alternative to
Measurements: Theory and Demonstration,” Proc. IEEE Parameter Adaptation,” Systems and Control Letters, cilt:
Conf. Decision and Control, Lake Bueno Vista, FL (Dec. 47, s: 259–278.
1994 ) pp 824-828. Qu, Z., D. Dawson, J. Dorsey and J. Duffie (1995). “ Robust
Burg, T., D. Dawson, J. Hu and M. de Queiroz (1996). “An Estimation and Control of Robotic Manipulators,”
Adaptive Partial State Feedback Controller for RLED Robotica 13 Part 3, 223-231.
Robot Manipulators,” IEEE Trans. On Automatic Control Qu, Z. (1998).Robust Control of Nonlinear Uncertain
41, No. 7, 1024-1031. Systems, New York: Wiley.
Burkov, I. (1993). “Asymptotic Stabilization of Nonlinear Sadegh, N. and R. Horowitz (1990). “Stability and
Lagrangian Systems without Measuring Velocities,” Robustness Analysis of a Class of Adaptive Controller for
Proc. of the Int. Symp. Active Control in Mech. Engg., Robotic Manipulators”, International Journal of Robotic
Lyon, France. Research, Vol. 9, No. 9 s:74–92.
Canudas de Wit, C. and N. Fixot (1991a). “Robot Control Sastry, S. and M. Bodson (1989). Adaptive Control:
via Robust Esimated State Feedback,” IEEE Trans. Stability, Convergence, and Robustness, Upper Saddle
Automatic Control 36 No. 12 1497- 1501. River, NJ: Prentice-Hall.
Canudas de Wit, C. and J. Slotine (1991b). “Sliding Xian, B., D. M. Dawson, M. S. de Queiroz and J. Chen
Observers for Robot Manipulators,” Automatica 27, No. 5 (2004). “A Continuous Asymptotic Tracking Control
859-864. Strategy for Uncertain Nonlinear Systems,”IEEE Trans.
Direct Drive Manipulator Research and Development on Automatic Control, cilt: 49, No: 7,s:1206–1211.
Package Operations Manual. (1992) Yuan, J. and Y. Stepanenko (1991). “Robust Control of
Dixon, W. E., E. Zergeroğlu, D. M. Dawson and B. T. Costic Robotic Manipulators without Velocity Measurements,”
(2002). “Repetitive Learning Control: A Lyapunov-Based Int. J. Robust and Nonlinear Control 1, 203-213.
Approach,” IEEE Trans. on Systems, Man, and Zergeroglu, E., W. E. Dixon, D. Haste and D. M. Dawson
Cybernetics, Kısım B: Cybernetics, cilt: 32, No: 4, s: (1999). “A Composite Adaptive Output Feedback
538–545. Tracking Controller for Robotic Manipulators,” Robotica
Dixon, W. E., E. Zergeroglu and D. M. Dawson (2004). 17, Part 6 591-600.
“Global Robust Output Feedback Tracking Control of Zhang, F., D. M. Dawson, M. S. de Queiroz and W. E. Dixon
Robot Manipulators,” Robotica 22, 351-357. (2000). “Global Adaptive Output Feedback Tracking Control of
Kaneko, K. and R.Horowitz (1997). “Repetitive and Robot Manipulators,” IEEE Trans. Automatic Control 45, No. 6,
Adaptive Control of Robot Manipulators via Velocity 1203-1208.
Estimation,” IEEE Trans. Robotics and Automation 13
No.2, 204-217.

14130