Universal Adaptive Control via Output Feedback for Nonlinear Systems with
Parametric and Measurement Uncertainty
Xu Zh ang, Wei Lin an d Chunjiang Qian
Abstract— The problem of global state regulation via output
feedback is studied for a class of nonlinear systems with
unknown parameters in the state and output equations. In
contrast to the existing results, both the value and sign of the
unknown parameter in the system output are not required to
be known a priori. Moreover, the controlled plant is assumed
to be nonlinearly dependent of the output and the unknown
parameters but linearly in the unmeasured states, with a
lower-triangular structure. Using the idea of K-filter, we fi rst
construct a nonlinear observer with a dynamic-gain for the
uncertain system. We then develop, by virtue of the universal
control philosophy with a Nussbaum function, a universal
adaptive control scheme that achieves global state regulation
and bounded ness of the closed-loop system.
I. INTRODUCTION
We consider the problem of global adaptive c ontrol by
output fe edback fo r the uncertain nonlinear system
˙z
1
= z
2
+ C
11
(y)z
1
+ φ
1
(t, z, u)
˙z
2
= z
3
+ C
21
(y)z
1
+ C
22
(y)z
2
+ φ
2
(t, z, u)
.
.
.
˙z
n
= u + C
n1
(y)z
1
+ · · · + C
nn
(y)z
n
+ φ
n
(t, z, u)
y = θz
1
, (1.1)
where z = [z
1
, · · · , z
n
]
T
∈ IR
n
, u ∈ IR and y ∈ IR are
the system state, input and measured output, respectively,
C
ij
: IR → IR, i = 1, · · · , n, j = 1, · · · , i, are known
smooth functio ns, and θ 6= 0 is an unknown parameter w ith
unknown sign, representing the measureme nt uncertainty
[5], [10], [14], [8]. The C
1
mappings φ
i
: IR × IR
n
× IR →
IR, i = 1 , · · · , n, are not required to be precisely known but
satisfy the following parametric condition.
Assumption 1.1: For i = 1, · · · , n, there are a C
1
func-
tion h(y) ≥ 1 and an unknown constant Θ ≥ 0, such that
|φ
i
(t, z, u)| ≤ Θh(y)|y|. (1.2)
The objective of this paper is to prove that under Assump-
tion 1 .1, there is an adaptive ou tput feedbac k controller
˙
ζ = ψ(ζ, y), ζ ∈ IR
2n+4
,
u = β(ζ, y), (1.3)
where ψ : IR
2n+4
× IR → IR
2n+4
and β : IR
2n+4
× IR → IR
are smooth functions with ψ(0, 0) = 0 and β(0, 0) = 0,
Xu Zhang is with the School of Automation, Beihang University,
China. Wei Lin is affiliated with Dongguan Univ. of Technology, and is
with the Dept. of Electrical Eng. and Computer Science, Case Western
Reserve University, Ohio. Chunjiang Qian is with the Dept. of Electrical
and Computer Eng., University of Texas at San Antonio, Texas. This
work was supported in part by NSFC under Grants 61533009, 61773033
and 61673038, China Postdoctoral Science Foundation, Research Projects
JCY20150403161923519 and KCYKYQD2017005. Corresponding author:
Prof. Wei Lin (email: linwei@case.edu).
asymptotically regulating the uncertain system (1.1) with
global boundedness, despite th e present of parametric un-
certainty in the state and output equations.
When θ = 1, many re sults have been reported in the
literature d evoting to the problem of global stabilization
of n onlinear systems including (1.1). For instance, g lobal
stabilization by output feedback was studied under the
condition that C
ij
(y) and h(y) are bounded by a known
constant [1], [20] (i.e., the linear growth condition), or
the perturbation φ
i
(t, z, u) satisfies a n output-dependent
growth condition with a triangular structu re [19], [12], [23].
Notably, the results in [19], [12], [23] can indeed be applied
to a larger class of nonlinear systems than the sy stem
(1.1) with (1.2). In the presence of parametric uncertainty,
global adaptive control of the uncertain system (1.1) and its
variations were also investigated, f or example, in [15], [16].
When θ is uncertain and represents the measurement
uncertainty [5], [10], [14], [8], global output feedback
stabilization has been studied fo r uncertain systems that are
composed of a controllable/observable linear system [21],
[22] and a perturbed nonlinearity [25], [24], [2]. However,
these results have been based o n the two hypotheses. One
is a linear or homogeneous growth condition imposed on
the perturbed n onlinear dynamics. The other one is that
both the sign of the parameter θ and its bounds are needed
to be known a priori. When the nonlinear perturbation is
present in (1.1) and the parameter θ is completely unknown
— neither its bound nor its sign, the question of how to
globally adaptively stabilize the uncertain system (1.1) by
output feedback is certainly interesting and deserves a study.
In this paper we address the p roblem and provide a
preliminar y solution. In particu lar, we show that asymp-
totic state regulation of the uncertain system (1.1) can
be achieve d by a universal-like adaptive output feedback
compen sator. As a matter of fact, motivated by the idea
of K-filter [11], [13], [1 7], we first design a nonlinear
observer tha t con sists of two K-filters with a dynam ic
gain to be updated dynam ically. Then, we use the adaptive
backstepping method to design an observer-based controller
with a Nussbaum gain [ 18], [6] and two adaptive gain
update laws for achieving global state regulation of the
uncertain system (1.1) with stability. The novelty of the
proposed adaptive control strategy lies in the development
of a universal-like output feedback compensator, which con-
sists of a K-filter based observer and a controller involving
with four dynamic gains, namely, a dynamic high-gain, a
Nussbaum gain and two adaptive gains. All of them are
updated online and play a crucial role in counteracting
the e ffect of the completely unknown parameter in the
system output, the output-d e pendent nonlinearity and the
parametric uncertainty in the perturbed dynamics.
Notation: A Nussbaum function N (k) = k
2
cos(k) will
be used through out this paper. Clearly it satisfies the pr op-
erties [18], [6]:
lim
k→∞
sup
1
k
Z
k
0
N(s)ds=+∞, lim
k→∞
inf
1
k
Z
k
0
N(s)ds=−∞ (1.4)
2018 IEEE Conference on Decision and Control (CDC)
Miami Beach, FL, USA, Dec. 17-19, 2018
978-1-5386-1394-8/18/$31.00 ©2018 IEEE 5530