Upload
others
View
25
Download
0
Embed Size (px)
Citation preview
L1 Adaptive Output-Feedback Controller forNon-Strictly-Positive-Real Reference Systems:
Missile Longitudinal Autopilot Design
Chengyu Cao∗
University of Connecticut, Storrs, Connecticut 06269
and
Naira Hovakimyan†
University of Illinois at Urbana–Champaign, Urbana, Illinois 61801
DOI: 10.2514/1.40877
This paper presents an extension of the L1 adaptive output-feedback controller to systems of unknown relative
degree in the presence of time-varying uncertainties without restricting the rate of their variation. As compared with
earlier results in this direction, a new piecewise continuous adaptive law is introduced, along with the low-pass-
filtered control signal that allows for achieving arbitrarily close tracking of the input and the output signals of the
reference system, the transfer function of which is not required to be strictly positive real. Stability of this reference
system is proved using a small-gain-type argument. The performance bounds between the closed-loop reference
system and the closed-loop L1 adaptive system can be rendered arbitrarily small by reducing the step size of
integration. Missile longitudinal autopilot design is used as an example to illustrate the theoretical results.
I. Introduction
M ODERN fighter aircraft and munitions need to operate inhighly nonlinear and uncertain flight regimes, for which
accurate aerodynamic modeling is not possible or is overlyexpensive. The control design challenge for these vehicles is toensure safe operation in the presence of such uncertainties, withoutsacrificing the maneuverability of the vehicle. Both classical andmodern control design methods have been extensively investigatedfor performance improvement of such vehicles in the presenceof uncertainties. Their performance is well known to depend uponaccurate knowledge of the system dynamics and is limited bythe presence of unmodeled high-frequency effects. The same isparticularly true for adaptive methods that lose their robustnessin the presence of fast adaptation [1–4]. The recently developed L1
adaptive controller [5–7] opened new opportunities for addressingthe control challenge for highly uncertain vehicles in the presenceof component failures, aerodynamic uncertainties, and disturban-ces. It is important at this point to emphasize the significance ofoutput-feedback architectures, which have the potential to relax thematching assumption, typically appearing in state-feedback adaptivecontrol architectures. One of the fundamental limitations of adaptiveoutput-feedback approaches is the requirement for the reference-system dynamics to have a strictly-positive-real (SPR) transferfunction between the input and the regulated output [8]. When theSPR condition does not hold, then the obtained results lead toultimate boundedness, which are hard to quantify rigorously and/orto reduce in desirable way [9,10]. On the other hand, the SPRrequirement for the reference system restricts the class of reference
behaviors that can be achieved by adaptation for the given unknownplant [6,11].
This paper extends the results of [6] to an output-feedbackframework by considering reference systems that do not verify theSPR condition for their input–output transfer function. The keydifference from the results in [6] is the new piecewise continuousadaptive law. The adaptive control is defined as the output of a low-pass filter, resulting in a continuous signal despite the discontinuityof the adaptive law. Similar to [6], theL1-norms of both input/outputerror signals between the closed-loop adaptive system and thereference system can be rendered arbitrarily small by reducing thestep size of integration.
We note that adaptive algorithms achieving arbitrarily improvedtransient performance for a system’s output were reported in [12–26]. As compared with those results, [5,6] presented the opportunityto also regulate the performance bound for a system’s input signal byrendering it arbitrarily close to the corresponding signal of a boundedlinear time-invariant reference system. Unlike conventional adaptivecontrollers, the L1 adaptive controllers adapt fast, leading to desiredtransient and asymptotic tracking with a guaranteed, bounded awayfrom zero, time-delay margin [27]. The results from [5,6] have beenintensively applied in flight tests [11,28–31] and various mid- tohigh-fidelity simulation environments [32–35]. Insights into theperformance of L1 adaptive controller can be obtained from theanalysis of a simple linear system in [36], in which sensitivity andcosensitivity transfer functions are analyzed for disturbance rejec-tion and noise tolerance in the presence of a large adaptation rate.Further, the results in [37] provide systematic design guidelines forselection of the underlying filter to achieve the desired performancebound while retaining a guaranteed time-delay margin.
This paper presents the adaptive output-feedback counterpartof the results in [5,6], without enforcing the SPR condition on theinput/output transfer function of the desired reference system,which typically appears in conventional adaptive output-feedbackschemes. The paper also uses the longitudinal dynamics of themissile from [38] to demonstrate the usefulness of the methodology.Different applications of this methodology can be found in [39,40].
The paper is organized as follows. Section II gives the problemformulation. In Sec. III, the closed-loop reference system is intro-duced. In Sec. IV, some preliminary results are developed towardthe definition of the L1 adaptive controller. In Sec. V, the novel L1
adaptive control architecture is presented. Stability and uniformperformance bounds are presented in Sec. VI. In Sec. VII, simulation
Presented as Paper 7288 at the AIAA Guidance, Navigation, and ControlConference, Honolulu, HI, 18–21 August 2008; received 8 September 2008;revision received 24 November 2008; accepted for publication 22 December2008. Copyright © 2009 by Chengyu Cao and Naira Hovakimyan. Publishedby the American Institute of Aeronautics and Astronautics, Inc., withpermission. Copies of this paper may be made for personal or internal use, oncondition that the copier pay the $10.00 per-copy fee to the CopyrightClearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923; includethe code 0731-5090/09 $10.00 in correspondence with the CCC.
∗Assistant Professor, Department of Mechanical Engineering; [email protected]. Member AIAA.
†Professor, Department of Mechanical Science and Engineering;[email protected]. Associate Fellow AIAA.
JOURNAL OF GUIDANCE, CONTROL, AND DYNAMICS
Vol. 32, No. 3, May–June 2009
717
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77
results are presented, and Sec. VIII concludes the paper. The small-gain theorem and some basic definitions from linear systems theoryused throughout the paper are given in the Appendix. Unlessotherwise mentioned, k � k will be used for the 2-norm of the vector.
II. Problem Formulation
Consider the following single-input/single-output (SISO) system:
y�s� � A�s��u�s� � d�s��; y�0� � 0 (1)
where u�t� 2 R is the input; y�t� 2 R is the system output; A�s� is astrictly proper unknown transfer function of unknown relative degreenr, for which only a known lower bound 1< dr � nr is available;d�s� is the Laplace transform of the time-varying uncertainties anddisturbances d�t� � f�t; y�t�� and f is an unknown map, subject tothe following assumptions.
Assumption 1. There exist constants L > 0 and L0 > 0 such that
jf�t; y1� � f�t; y2�j � Ljy1 � y2j; jf�t; y�j � Ljyj � L0
hold uniformly in t � 0, where the numbers L and L0 can bearbitrarily large.
Assumption 2. There exist constants L1 > 0, L2 > 0, and L3 > 0such that
j _d�t�j � L1j _y�t�j � L2jy�t�j � L3
for all t � 0, where the numbers L1, L2, and L3 can be arbitrarilylarge.
Let r�t� be a given bounded continuous reference input signal. Thecontrol objective is to design an adaptive output-feedback controlleru�t� such that the system output y�t� tracks the reference input r�t�following a desired reference model: that is,
y�s� M�s�r�s�
whereM�s� is a minimum-phase stable transfer function of relativedegree dr. We rewrite the system in Eq. (1) as
y�s� �M�s��u�s� � ��s��; y�0� � 0 (2)
��s� � ��A�s� �M�s��u�s� � A�s�d�s��=M�s� (3)
Let �Am; bm; cm� be the minimal realization of M�s�; that is, it iscontrollable and observable andAm is Hurwitz. The system in Eq. (2)can be rewritten as
_x�t� � Amx�t� � bm�u�t� � ��t��y�t� � c>mx�t�; x�0� � x0 � 0 (4)
III. Closed-Loop Reference System
Consider the following closed-loop reference system:
yref�s� �M�s��uref�s� � �ref�s�� (5)
�ref�s� ��A�s� �M�s��uref�s� � A�s�dref�s�
M�s� (6)
uref�s� � C�s��r�s� � �ref�s�� (7)
where dref�t� � f�t; yref�t��, and C�s� is a strictly proper system ofrelative order dr, with its dc gain C�0� � 1. Further, the selection ofC�s� andM�s� must ensure that
H�s� � A�s�M�s�=�C�s�A�s� � �1 � C�s��M�s�� (8)
is stable and
kG�s�kL1L < 1 (9)
where G�s� �H�s��1 � C�s��. This in turn restricts the class ofsystems A�s� in Eq. (1), for which the proposed approach in thispaper can achieve stabilization. However, as discussed later inRemark 3, the class of such systems is not empty. Letting
A�s� � An�s�Ad�s�
; C�s� � Cn�s�Cd�s�
; M�s� �Mn�s�Md�s�
(10)
where the numerators and the denominators are all polynomials of s.It follows from Eq. (8) that
H�s� � Cd�s�Mn�s�An�s�Hd�s�
(11)
where
Hd�s� � Cn�s�An�s�Md�s� �Mn�s�Ad�s��Cd�s� � Cn�s�� (12)
A strictly proper C�s� implies that the order of Cd�s� � Cn�s� andCd�s� is the same. Because the order of Ad�s� is higher thanthat of An�s�, the transfer function H�s� is strictly proper.
Lemma 1. If C�s� and M�s� verify the condition in Eq. (9), theclosed-loop reference system in Eqs. (5–7) is stable.
Proof. It follows from Eqs. (6) and (7) that
uref�s� �C�s�M�s�r�s� � C�s�A�s�dref�s�C�s�A�s� � �1 � C�s��M�s� (13)
From Eqs. (5) and (6), one can derive
yref�s� �H�s��C�s�r�s� � �1 � C�s��dref�s�� (14)
Because H�s� is strictly proper and stable,
G�s� �H�s��1 � C�s��
is also strictly proper and stable; therefore,
kyrefkL1 � kH�s�C�s�kL1krkL1 � kG�s�kL1
�LkyrefkL1 � L0�
Using the condition in Eq. (9), one can write
kyrefkL1 � �; ��kH�s�C�s�kL1
krkL1 � kG�s�kL1L0
1 � kG�s�kL1L
<1
(15)
Hence, kyrefkL1 is bounded, and the proof is complete. □
IV. Preliminaries for the Main Result
Let
H0�s� � A�s�=�C�s�A�s� � �1 � C�s��M�s�� (16)
H1�s����A�s��M�s��C�s��=�C�s�A�s���1�C�s��M�s�� (17)
H2�s� �H�s�C�s�=M�s� (18)
H3�s� � ��M�s�C�s��=��C�s�A�s� � �1 � C�s��M�s��� (19)
Using the expressions from Eqs. (10) and (12),
H0�s� � Cd�s�An�s�Md�s�=Hd�s�
and H1�s� can be rewritten as
H1�s� � �Cn�s�An�s�Md�s� � Cn�s�Ad�s�Mn�s��=Hd�s� (20)
Because the degree of Cd�s� � Cn�s� is larger than Cn�s� by dr,the degree of
Mn�s�Ad�s��Cd�s� � Cn�s��
is larger than
718 CAO AND HOVAKIMYAN
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77
Cn�s�Ad�s�Mn�s�
by dr. Because the degree of Ad�s� is larger than An�s� by � drand the degree ofMn�s� is larger thanMd�s� by dr, the degree of
Mn�s�Ad�s��Cd�s� � Cn�s��
is larger than that of
Cn�s�An�s�Md�s�
Therefore, H1�s� is strictly proper with relative degree dr. Wenote from Eqs. (11) and (20) that H1�s� has the same denominatoras H�s�, and it therefore follows from Eq. (9) that H1�s� is stable.Using similar arguments, it can be verified that H0�s� is proper andstable. Similarly, H2�s� is strictly proper and stable. Let
�� kH1�s�kL1krkL1 � kH0�s�kL1
�L�� L0�
� ��
�kH1�s�=M�s�kL1
� LkH0�s�kL1
kH2�s�kL1
1 � kG�s�kL1L
�(21)
where �� > 0 is an arbitrary constant. Because both H1�s� andM�s� are stable and strictly proper with relative degree dr andM�s�is minimum-phase, H1�s�=M�s� is stable and proper. Hence,kH1�s�=M�s�kL1
exists. Therefore, � is bounded. Further, becauseAm is Hurwitz, there existsP� P> > 0, which satisfies the algebraicLyapunov equation:
A>mP� PAm ��Q; Q > 0
From the properties of P, it follows that there exits nonsingular����Pp
such that
P� �����Pp�>
����Pp
Given the vector c>m�����Pp��1, let D be a �n � 1� n matrix that
contains the null space of c>m�����Pp��1:
D�c>m�����Pp��1�> � 0 (22)
and further let
�� c>mD
����Pp
� �
Lemma 2. For any
�� yz
� �2 Rn
where y 2 R and z 2 Rn�1, there exist p1 > 0 and positive definiteP2 2 R�n�1��n�1� such that
�>���1�>P��1�� p1y2 � z>P2z
Proof. Using P� �����Pp�>
����Pp
, one can write
�>���1�>P��1�� �>�����Pp
��1�>�����Pp
��1��
We note that
������Pp��1 � c>m�
����Pp��1
D
� �
Let
q1 � �c>m�����Pp��1��c>m�
����Pp��1�>; Q2 �DD>
From Eq. (22), we have
�������Pp��1����
����Pp��1�> � q1 0
0 Q2
� �
Nonsingularity of � and����Pp
implies that
�������Pp��1����
����Pp��1�>
is nonsingular, and therefore Q2 is also nonsingular. Hence,� ����Pp
��1�>� ����
Pp
��1�����
����Pp��1��
������Pp��1�>
��1����
����Pp��1��>� ����
Pp
��1��
q�11 0
0 Q�12
" #
Denoting p1 � q�11 and P2 �Q�12 completes the proof. □
Let T be any positive constant, which can be associated with thesampling rate of the available CPU, and let 11 2 Rn be the basisvector with the first element equal to 1 and all other elements equal to0. Let ��T� 2 Rn�1 be a vector, which consists of 2 to n elements of1>1 exp��Am��1T�, and let
��T� �ZT
0
j1>1 exp��Am��1�T � ����bmjd� (23)
Further, let
&�T� � k��T�k�������������������
�
max�P2�
r� ��T��
�� max���>P��1��
2�k��>Pbmkmin���>Q��1�
�2
(24)
Letting
1 >1 exp��Am��1t� � �1�t�>2 �t��
where 1�t� 2 R and 2�t� 2 Rn�1 contain the first and 2 to nelements of the row vector 1>1 exp��Am��1t�, respectively, weintroduce the following functions:
�1�T� � maxt2�0;T�j1�t�j; �2�T� � max
t2�0;T�k2�t�k (25)
Further, let ��T� be the n n matrix:
��T� �ZT
0
exp��Am��1�T � ����d� (26)
�3�T� � maxt2�0;T�
3�t�; �4�T� � maxt2�0;T�
4�t� (27)
where
3�t� �Zt
0
j1>1 exp��Am��1�t
� ������1�T� exp��Am��1T�11jd�;
4�t� �Zt
0
j1>1 exp��Am��1�t � ����bmjd�
Finally, let
�0�T� � �1�T�&�T� � �2�T��������������������
�
max�P2�
r� �3�T�&�T� � �4�T��
(28)
Lemma 3. The following limiting relationship is true:
limT!0
�0�T� � 0
Proof. Note that because �1�T�, �3�T��, and � are bounded, it issufficient to prove that
limT!0
&�T� � 0 (29)
limT!0
�2�T� � 0 (30)
limT!0
�4�T� � 0 (31)
CAO AND HOVAKIMYAN 719
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77
Because
limT!0
1>1 exp��Am��1T� � 1>1
then
limT!0
��T� � 0n�1
which implies
limT!0k��T�k � 0
Further, it follows from the definition of ��T� in Eq. (23) that
limT!0
��T� � 0
Because � and � are bounded,
limT!0
&�T� � 0
which proves Eq. (29). Because 2�t� is continuous, it followsfrom Eq. (25) that
limT!0
�2�T� � limt!0k2�t�k
Because
limt!0
1>1 exp��Am��1t� � 1>1
we have
limt!0k2�t�k � 0
which proves Eq. (30). Similarly,
limT!0
�4�T� � limt!0k4�t�k � 0
which proves Eq. (31). The boundedness of �, �3�T�, and� implies
limT!0
��1�T�&�T� � �2�T�
��������������������
max�P2�
r� �3�T�&�T�
� �4�T���� 0
which completes the proof. □
V. L1 Adaptive Output-Feedback Controller
We consider the following state predictor (or passive identifier):
_x�t� � Amx�t� � bmu�t� � ��t�y�t� � c>mx�t�; x�0� � x0 � 0 (32)
where ��t� 2 Rn is the vector of adaptive parameters. Note thatalthough ��t� 2 R in Eq. (4) (i.e., the unknown disturbance) ismatched, the uncertainty estimation in Eq. (32), ��t� 2 Rn, isunmatched. This is the key step of the solution and the subsequentanalysis.
Letting ~y�t� � y�t� � y�t�, the update law for ��t� is given by
��t� � ��iT�; t 2 �iT; �i� 1�T���iT� � ���1�T���iT�; i� 0; 1; 2; . . .
(33)
where ��T� is defined in Eq. (26) and
��iT� � exp��Am��1T�11 ~y�iT�; i� 0; 1; 2; 3; . . . (34)
The control signal is defined as follows:
u�s� � C�s�r�s� � C�s�c>m�sI � Am��1bm
c>m�sI � Am��1��s� (35)
whereC�s�wasfirst introduced in Eq. (7). TheL1 adaptive controllerconsists of Eqs. (32), (33), and (35), subject to the condition inEq. (9).
We will now proceed with the computation of error bounds. Let~x�t� � x�t� � x�t�. The error dynamics between Eqs. (4) and (32) are
_~x�t� � Am ~x�t� � ��t� � bm��t�~y�t� � c>m ~x�t�; ~x�0� � 0 (36)
Lemma 4. Let e�t� � y�t� � yref�t�. Then
ketkL1 �kH2�s�kL1
1 � kG�s�kL1Lk ~ytkL1 (37)
Proof. Let
~��s� � C�s�c>m�sI� Am��1bm
c>m�sI� Am��1��s� � C�s���s� (38)
It follows from Eq. (35) that
u�s� � C�s�r�s� � C�s���s� � ~��s� (39)
and the system in Eq. (2) consequently takes the form
y�s� �M�s��C�s�r�s� � �1 � C�s����s� � ~��s�� (40)
Substituting u�s� from Eq. (39) into Eq. (3) gives
��s� � ��A�s� �M�s���C�s�r�s�� C�s���s� � ~��s�� � A�s�d�s��=M�s�
and hence,
��s� � �A�s� �M�s���C�s�r�s� � ~��s�� � A�s�d�s�M�s� � C�s��A�s� �M�s�� (41)
Using the definitions of H0�s� and H1�s� in Eqs. (16) and (17), wecan write
��s� �H1�s�r�s� �H1�s�C�s� ~��s� �H0�s�d�s� (42)
Substitution into Eq. (40) leads to
y�s� �M�s��C�s� �H1�s��1 � C�s����r�s� � ~��s�
C�s�
��H0�s�M�s��1 � C�s��d�s�
Recalling the definition of H�s� from Eq. (8), one can verify that
M�s��C�s� �H1�s��1 � C�s��� �H�s�C�s�
H�s� �H0�s�M�s�
which implies
y�s� �H�s��C�s�r�s� � ~��s�� �H�s��1 � C�s��d�s�
Using the expression for yref�s� fromEq. (14) and letting de�s� be theLaplace transform of
de�t� � f�t; y�t�� � f�t; yref�t��
one can derive
e�s� �H�s���1 � C�s��de�s� � ~��s��
Lemma 5 in the Appendix and Assumption 1 give the followingupper bound:
ketkL1 � LkH�s��1 � C�s��kL1ketkL1 � kr1tkL1 (43)
where r1�t� is the signal, with its Laplace transform being
720 CAO AND HOVAKIMYAN
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77
r1�s� �H�s� ~��s�
Using the expression for ~��s� from Eq. (38), along with the ex-pression for y�s� from Eq. (2), and taking into consideration that
y�s� �M�s�u�s� � c>m�sI� Am��1��s�
we have
~y�s� � c>m�sI� Am��1��s� �M�s���s�
�M�s�C�s�
C�s�M�s� c
>m�sI � Am��1��s� �
M�s�C�s� C�s���s�
�M�s�C�s� ~��s� (44)
This implies that r1�s� can be rewritten as
r1�s� �C�s�H�s�M�s�
M�s�C�s� ~��s� �H2�s� ~y�s�
and hence
kr1tkL1 � kH2�s�kL1k ~ytkL1
Substituting this back into Eq. (43) completes the proof. □
VI. Analysis of L1 Adaptive Controller
In this section, we analyze the stability and the performance oftheL1 adaptive controller. Using the definitions fromEq. (10),H3�s�in Eq. (19) can be rewritten as
H3�s� ��Cn�s�Ad�s�Mn�s�
Hd�s�(45)
where Hd�s� is defined in Eq. (12). Because
deg�Cd�s� � Cn�s�� � degCn�s� � dr
it can be checked straightforwardly thatH3�s� is strictly proper. Wenote from Eqs. (11) and (45) that H3�s� has the same denominatoras H�s�, and therefore it follows from Eq. (9) that H3�s� is stable.Because H3�s� is strictly proper and stable with relative degree dr,H3�s�=M�s� is stable and proper and, therefore its L1 norm is finite.Consider the state transformation
~���~x
It follows from Eq. (36) that
_~��t� ��Am��1 ~��t� ����t� ��bm��t�; ~��0� � 0 (46)
~y�t� � ~�1�t� (47)
where ~�1�t� is the first element of ~��t�.Theorem 1. Given the system in Eq. (1) and the L1 adaptive con-
troller in Eqs. (32), (33), and (35), subject to Eq. (9), if we choose Tto ensure
�0�T�< �� (48)
where �� is an arbitrary positive constant introduced in Eq. (21), then
k ~ykL1 < �� (49)
ky � yrefkL1 � �1; ku � urefkL1 � �2 (50)
with
�1 � kH2�s�kL1��=�1 � kG�s�kL1
L��2 � LkH2�s�kL1
�1 � kH3�s�=M�s�kL1��
Proof. First, we prove the bound in Eq. (49) by a contradictionargument. Because ~y�0� � 0 and ~y�t� is continuous, then assumingthe opposite implies that there exists t0 such that
k ~y�t�k< ��; 8 0 � t < t0 (51)
k ~y�t0�k � �� (52)
which leads to
k ~yt0 kL1 � �� (53)
Because y�t� � yref�t� � e�t�, the upper bound in Eq. (15) can beused to arrive at
kyt0 kL1 � kyreft0 kL1 � ket0 kL1 � �� kC�s�H�s�=M�s�kL1
��=�1 � kG�s�kL1L� (54)
It follows from Eqs. (42) and (44) that
��s� �H1�s�r�s� �H1�s� ~y�s�=M�s� �H0�s�d�s�
and hence Eq. (53) implies that
k�t0 kL1 � kH1�s�kL1krkL1 � kH1�s�=M�s�kL1
��
� kH0�s�kL1�Lkyt0 kL1 � L0�
which, along with Eq. (54), leads to
k�t0 kL1 � � (55)
It follows from Eq. (46) that
~��iT � t� � exp��Am��1t� ~��iT�
�ZiT�t
iT
exp��Am��1�iT � t� ������iT�d�
�ZiT�t
iT
exp��Am��1�iT � t � ����bm����d�
� exp��Am��1t� ~��iT� �Zt
0
exp��Am��1�t � ������iT�d�
�Zt
0
exp��Am��1�t � ����bm��iT � ��d� (56)
Because
~��iT� � ~y�iT�0
� �� 0
~z�iT�
� �
it follows from Eq. (56) that
~��iT � t� � �iT � t� � ��iT � t� (57)
where
�iT � t� � exp��Am��1t�~y�iT�0
� ��
Zt
0
exp��Am��1�t � ������iT�d�(58)
��iT � t� � exp��Am��1t�0
~z�iT�
" #
�Zt
0
exp��Am��1�t � ����bm��iT � ��d� (59)
CAO AND HOVAKIMYAN 721
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77
In what follows, we prove that for all iT � t0, one has
j ~y�iT�j � &�T� (60)
~z >�iT�P2 ~z�iT� � � (61)
where &�T� and � are defined in Eq. (24). Because ~��0� � 0, it isstraightforward that j ~y�0�j � &�T� and ~z>�0�P2 ~z�0� � �. For any�j� 1�T � t0, we will prove that if
j ~y�jT�j � &�T� (62)
~z >�jT�P2 ~z�jT� � � (63)
then Eqs. (62) and (63) hold for j� 1 too. Hence, Eqs. (60) and (61)hold for all iT � t0.
Assume that Eqs. (62) and (63) hold for j and, in addition,
�j� 1�T � t0 (64)
It follows from Eq. (57) that
~���j� 1�T� � ��j� 1�T� � ���j� 1�T� (65)
where
��j� 1�T� � exp��Am��1T�~y�jT�0
" #
�ZT
0
exp��Am��1�T � ������jT�d� (66)
���j� 1�T� � exp��Am��1T�0
~z�jT�
" #
�ZT
0
exp��Am��1�T � ����bm��jT � ��d� (67)
Substituting the adaptive law from Eq. (33) in Eq. (66), we have
��j� 1�T� � 0 (68)
It follows from Eq. (59) that ��t� is the solution of the followingdynamics:
_��t� ��Am��1��t� ��bm��t� (69)
��jT� � 0
~z�jT�
� �; t 2 �jT; �j� 1�T� (70)
Consider the following function
V�t� � �>�t���>P��1��t�
over t 2 �jT; �j� 1�T�. Because � is nonsingular and P is positivedefinite, ��>P��1 is positive definite, and hence V�t� is a posi-tive definite function. It follows from Lemma 2 and the relationshipin Eq. (70) that
V���jT�� � ~z>�jT�P2 ~z�jT�
which further, along with the upper bound in Eq. (63), leads to thefollowing:
V���jT�� � � (71)
It follows from Eq. (69) that over t 2 �jT; �j� 1�T�,
_V�t���>�t���>P��1�Am��1��t���>�t���>A>m�>��>P>��1��t��2�>�t���>P��1�bm��t����>�t���>Q��1��t��2�>�t���>Pbm��t�
Using the upper bound from Eq. (55), one can derive over t 2�jT; �j� 1�T�
_V�t� � �min���>Q��1�k��t�k2 � 2k��t�kk��>Pbmk� (72)
Note that for all t 2 �jT; �j� 1�T�, if
V�t� � � (73)
we have
k��t�k �����������������������������������
�
max���>P��1�
r� 2�k��>Pbmkmin���>Q��1�
and the upper bound in Eq. (72) yields
_V�t� � 0 (74)
It follows from Eqs. (71), (73), and (74) that
V�t� � �; 8 t 2 �jT; �j� 1�T�
and therefore
V��j� 1�T� � �>��j� 1�T����>P��1����j� 1�T� � � (75)
Because
~���j� 1�T� � ��j� 1�T� � ���j� 1�T� (76)
the equality in Eq. (68) and the upper bound in Eq. (75) lead to thefollowing inequality:
~� >��j� 1�T����>P��1� ~���j� 1�T� � �
Using the result of Lemma 2, one can derive that
~z>��j� 1�T�P2 ~z��j� 1�T�
� ~�>��j� 1�T����>P��1� ~���j� 1�T� � �
which implies that the upper bound in Eq. (63) holds for j� 1.It follows from Eqs. (47), (68), and (76), that
~y��j� 1�T� � 1>1 ���j� 1�T�
and the definition of ���j� 1�T� in Eq. (67) leads to the followingexpression:
~y��j� 1�T� � 1>1 exp��Am��1T�0
~z�jT�
� �
� 1>1
ZT
0
exp��Am��1�T � ����bm��jT � ��d� (77)
The upper bounds in Eqs. (55) and (63) allow for the following upperbound:
j ~y��j� 1�T�j � k��T�kk~z�jT�k
�ZT
0
j1>1 exp��Am��1�T � ����bmjj��jT � ��jd�
� k��T�k�������������������
�
max�P2�
r� ��T��� &�T� (78)
where ��T� and ��T� are defined in Eq. (23), and &�T� is defined inEq. (24). This confirms the upper bound in Eq. (62) for j� 1. Hence,Eqs. (60) and (61) hold for all iT � t0.
For all iT � t � t0, where 0 � t � T, using the expression fromEq. (56), we can write that
722 CAO AND HOVAKIMYAN
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77
~y�iT � t� � 1>1 exp��Am��1t� ~��iT�
� 1>1
Zt
0
exp��Am��1�t� ������iT�d�
� 1>1
Zt
0
exp��Am��1�t � ����bm��iT � ��d�
The upper bound in Eq. (55) and definitions of 1�t�, 2�t�, 3�t�, and4�t� allow for the following upper bound:
j ~y�iT � t�j � j1�t�jj ~y�iT�j � k2�t�kk~z�iT�k� 3�t�j ~y�iT�j � 4�t��
Taking into consideration Eqs. (62) and (63) and recalling thedefinitions of �1�T�, �2�T�, �3�T�, and �4�T� in Eqs. (25–27) for all0 � t � T and for any nonnegative integer i subject to iT � t � t0,we have
j ~y�iT � t�j � �1�T�&�T� � �2�T��������������������
�
max�P2�
r� �3�T�&�T� � �4�T��
Because the right-hand side coincides with the definition of �0�T� inEq. (28), then for all t 2 �0; t0�, we have
j ~y�t�j � �0�T�
which, along with the assumption on T introduced in Eq. (48), yields
k ~yt0 kL1 < ��
This clearly contradicts the statement in Eq. (53). Therefore,
k ~ykL1 < ��
which proves Eq. (49). Further, it follows from Lemma 4 that
ketkL1 �kH2�s�kL1
1 � kG�s�kL1L��
which holds uniformly for all t � 0 and therefore leads to the firstupper bound in Eq. (50).
It follows from Eqs. (39) and (41) that
u�s� �M�s�C�s�r�s� �M�s� ~��s� � C�s�A�s�d�s�C�s�A�s� � �1 � C�s��M�s�
To prove the second bound in Eq. (50), we use the expression ofuref�s� from Eq. (13) to derive
u�s� � uref�s� � �H2�s�de�s� �H3�s�C�s� ~��s�
� �H2�s�de�s� �H3�s�M�s�
M�s�C�s� ~��s� (79)
It follows from Eqs. (44) and (79) that
ku � urefkL1 � LkH2�s�kL1ky � yrefkL1
� kH3�s�=M�s�kL1k ~ykL1
which, along with Eqs. (49) and (50), leads to the second bound inEq. (50). The proof is complete. □
Thus, the tracking error between y�t� and yref�t�, as well be-tween u�t� and uref�t�, is uniformly bounded by a constant propor-tional to T. This implies that during the transient one can achievearbitrary improvement of tracking performance by uniformlyreducing T.
Remark 1. Note that the parameter T is the fixed time step in thedefinition of the adaptive law. The adaptive parameters in ��t� 2 Rn
take constant values during �iT; �i� 1�T� for every i� 0; 1; . . ..Reducing T imposes hardware (CPU) requirements, and Theorem 1further implies that the performance limitations are consistent with
the hardware limitations. This in turn is consistent with the earlierresults in [5,6], in which improvement of the transient performancewas achieved by increasing the adaptation rate in the continuous-time adaptive laws.
Remark 2. We note that the following ideal control signal
uideal�t� � r�t� � �ref�t�
is the one that leads to desired system response
yideal�s� �M�s�r�s�
by canceling the uncertainties exactly. Thus, the reference system inEqs. (5–7) has a different response from the ideal one. It only cancelsthe uncertainties within the bandwidth ofC�s�, which can be selectedto be compatible with the control channel specifications. This isexactly what one can hope to achieve with any feedback in thepresence of uncertainties.
Remark 3. We note that stability of H�s� is equivalent tostabilization of A�s� by
C�s�M�s��1 � C�s�� (80)
Indeed, consider the closed-loop system, composed of the systemA�s� and negative feedback of Eq. (80). The closed-loop transferfunction is
A�s���
1� A�s� C�s�M�s��1 � C�s��
�(81)
Incorporating Eq. (10), one can verify that the denominator of thesystem in Eq. (81) is exactly Hd�s�. Hence, stability of H�s� isequivalent to the stability of the closed-loop system in Eq. (81). Thisimplies that the class of systems A�s� that can be stabilized by theL1
adaptive output-feedback controller Eqs. (32), (33), and (35) is notempty.
Remark 4. Finally, we note that although the feedback in Eq. (80)may stabilize the system in Eq. (1) for some classes of unknownnonlinearities, it will not ensure uniform transient performance in thepresence of unknown A�s�. On the contrary, the L1 adaptivecontroller ensures uniform transient performance for both of thesystem’s signals, independent of the unknown nonlinearity andindependent of A�s�.
VII. Simulations
A. Numerical Example
Consider the system in Eq. (1) with
A�s� � �s� 1�=�s3 � s2 � 2s� 8�
We note that A�s� has poles in the right-half complex plane. Weconsider the L1 adaptive output-feedback controller defined viaEqs. (32), (33), and (35), where
M�s� � 1
s2 � 1:4s� 1
C�s� � 100
s2 � 14s� 100; T � 10�4
First, we consider the step response for d�t� � 0. The simulationresults of L1 adaptive controller are shown in Figs. 1a and 1b. Next,we consider
d�t� � f�t; y�t�� � sin�0:1t�y2�t� � sin�0:4t�
and apply the same controller without retuning. The control signaland the system response are plotted in Figs. 2a and 2b. Further, weconsider a time-varying reference input r�t� � 0:5 sin�0:3t� and notethat without any retuning of the controller, the system response andthe control signal behave as expected (Figs. 3a and 3b).
CAO AND HOVAKIMYAN 723
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77
0 5 10 15 200
0.2
0.4
0.6
0.8
1
Time t0 5 10 15 20
−2
0
2
4
6
8
time t
a) y(t) (solid) and r(t) (dashed) b) Time history of u(t)
Fig. 1 Performance for r�t� � 1 and d�t� � 0.
0 5 10 15 20−2
0
2
4
6
8
time t0 5 10 15 20
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
Time t
a) y(t) (solid) and r(t) (dashed) b) Time history of u(t)
Fig. 2 Performance for r�t� � 1 and d�t� � sin�0:1t�y2�t� � sin�0:4t�.
0 10 20 30 40−0.5
0
0.5
Time t0 10 20 30 40
−5
0
5
time t
a) y(t) (solid) and r(t) (dashed) b) Time history of u(t)Fig. 3 Performance for r�t� � 0:5 sin�0:3t� and d�t� � sin�0:1t�y2�t� � sin�0:4t�.
0 0.5 1 1.5−4
−3
−2
−1
0
1
2
Time, [s]
Con
trol
sig
nal,
[rad
]
K=10%K=25%K=50%K=100%
0 0.5 1 1.50
0.2
0.4
0.6
0.8
1
1.2
1.4
Time, [s]
Ang
le o
f atta
ck, [
deg]
K=10%K=25%K=50%K=100%
a) System response b) Control historyFig. 4 Simulation results for missile longitudinal autopilot design: Scaled response for scaled (unmatched) uncertainties.
724 CAO AND HOVAKIMYAN
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77
B. Missile Longitudinal Autopilot Design
We consider the missile dynamics from [38], which are given bythe following state-space representation:
_x�t� �
ZaV
1 ZdV
0
Ma 0 Md 0
0 0 0 1
0 0 �!2a �2�a!a
266664
377775x�t� ��Ax�t�
�
0
0
0
!2a
266664
377775�u�t� � k>mx�t��
y�t� � � 1 0 0 0 �x�t�
where x�t� � �Az q �a _�a �> is the system state, in which Az(ft=s2) is the vertical acceleration, q (rad=s) is the pitch rate, �a (rad)
is the fin deflection, _�a (rad=s) is the fin deflection rate, km is thevector of matched (in the sense of state-feedback) parametricuncertainties, and�A contains unmatched uncertainties. For detailedanalysis of the missile dynamics, the reader is referred to [38].
In simulations, the following numerical values have been used forthe missile dynamics [38]:
ZaV��1:3046
�1
s
�;
ZdV��0:2142
�1
s
�
Ma � 47:7109
�1
s2
�; Md ��104:8346
�1
s2
�
�A�
�0:4996K 0 �0:4996K 0
0:4996K 0 �0:4996K 0
0 0 0 0
0 0 0 0
266664
377775
km � 0:3
0
0
�1�2 �a
!a
2666664
3777775
Here, the matrix �A is selected to enforce the worst direction forthe change of aerodynamic parameters, withK being the uncertaintyscaling factor for unmatched uncertainties. In this example, K � 0corresponds to nonperturbed aerodynamics, and the value K � 1corresponds to uncertainties that would turn the system into anoscillator in the presence of an linear quadratic regulator state-feedback controller only.
For simulation of the L1 adaptive output-feedback controller, thefollowingM�s� and C�s� have been selected, along with a samplingrate T � 0:0001 s:
M�s� � 1
1=!s2 � 2�!s� 1; C�s� � 1
1=!cs2 � 2�c!cs� 1
where !� 8 rad=s, �� 0:9, !c � 120 rad=s, and �c � 1:85.Figures 4a and 4b show the scaled response for scaled uncertainties,dependent upon the changes inK, achieved via scaled control efforts.
VIII. Conclusions
We presented the L1 adaptive output-feedback controller forreference systems that do not verify the SPR condition for theirinput–output transfer function. The new piecewise constant adaptivelaw, along with the low-pass-filtered control signal, ensures uniform
performance bounds for the system’s input/output signals simul-taneously. The performance bounds can be systematically improvedby reducing the integration time step.
Appendix: Facts from Linear Systems Theory
Definition 1 [41]. For a signal ��t�, where t � 0 and � 2 Rn, itsL1and truncated L1 norms are
k�kL1 � maxi�1;...;n
�sup��0j�i���j
�
k�tkL1 � maxi�1;...;n
�sup0���tj�i���j
�
where �i is the ith component of �.Definition 2 [41]. The L1 gain of a stable proper SISO system is
defined:
jjH�s�jjL1�Z 10
jh�t�jdt
where h�t� is the impulse response of H�s�.Lemma5 [41]. For a stable propermulti-input/multi-output system
H�s� with input r�t� 2 Rm and output x�t� 2 Rn, we have
kxtkL1 � kH�s�kL1krtkL1 8 t � 0
Acknowledgments
This research is supported by the U.S. Air Force Office ofScientific Research under contract FA9550-08-1-0135 and byNASA under contracts NNX08AB97A and NNX08AC81A. Theauthors are thankful toKevinWise for sharing hismissile code and toEvgeny Kharisov for integrating L1 adaptive output-feedbackcontroller into that code.
References
[1] Khalil, H. K., and Esfandiari, F., “Semiglobal Stabilization of a Classof Nonlinear Systems Using Output Feedback,” IEEE Transactions on
Automatic Control, Vol. 38, No. 9, 1993, pp. 1412–1415.doi:10.1109/9.237658
[2] Krstic, M., Kanellakopoulos, I., and Kokotovic, P., Nonlinear and
Adaptive Control Design, Wiley, New York, 1995.[3] Marino, R., and Tomei, P., “Global Adaptive Output Feedback Control
of Nonlinear Systems, Part 1: Linear Parameterization,” IEEE
Transactions on Automatic Control, Vol. 38, No. 1, 1993, pp. 17–32.doi:10.1109/9.186309
[4] Marino, R., and Tomei, P., “Global Adaptive Output Feedback Controlof Nonlinear Systems, Part 2: Nonlinear Parameterization,” IEEE
Transactions on Automatic Control, Vol. 38, No. 1, 1993, pp. 33–48.doi:10.1109/9.186310
[5] Cao, C., and Hovakimyan, N., “Design and Analysis of a Novel L1
Adaptive Control Architecture with Guaranteed Transient Perform-ance,” IEEE Transactions on Automatic Control, Vol. 53, No. 2,Mar. 2008, pp. 586–591.doi:10.1109/TAC.2007.914282
[6] Cao, C., and Hovakimyan, N., “L1 Adaptive Output FeedbackController for Systems of Unknown Dimension,” IEEE Transactions
on Automatic Control, Vol. 53, No. 3, Apr. 2008, pp. 815–821.doi:10.1109/TAC.2008.919550
[7] Cao, C., and Hovakimyan, N., “L1 Adaptive Controller for Systemswith Unknown Time-Varying Parameters and Disturbances in thePresence of Nonzero Trajectory Initialization Error,” International
Journal of Control, Vol. 81, No. 7, 2008, pp. 1147–1161.doi:10.1080/00207170701670939
[8] Calise, A. J., Hovakimyan, N., and Idan, M., “Adaptive OutputFeedback Control of Nonlinear Systems Using Neural Networks,”Automatica, Vol. 37, No. 8, 2001, pp. 1201–1211.doi:10.1016/S0005-1098(01)00070-X
[9] Hovakimyan, N., Nardi, F., and Calise, A., “A Novel Error ObserverBased Adaptive Output Feedback Approach for Control of UncertainSystems,” IEEE Transactions on Automatic Control, Vol. 47, No. 8,
CAO AND HOVAKIMYAN 725
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77
2002, pp. 1310–1314.doi:10.1109/TAC.2002.800766
[10] Hovakimyan, N., Nardi, F., Calise, A., and Kim, N., “Adaptive OutputFeedback Control of Uncertain Systems Using Single Hidden LayerNeural Networks”, IEEE Transactions on Neural Networks, Vol. 13,No. 6, 2002, pp. 1420–1431.doi:10.1109/TNN.2002.804289
[11] Cao, C., Hovakimyan, N., Kaminer, I., Patel, V., andDobrokhodov, V.,“Stabilization of Cascaded Systems via L1 Adaptive Controller withApplication to a UAV Path Following Problem and Flight TestResults,” American Control Conference, Inst. of Electrical andElectronics Engineers, Piscataway, NJ, 2007, pp. 1787–1792.
[12] Tsakalis, K. S., and Ioannou, P. A., “Adaptive Control of LinearTime-Varying Plants,” Automatica, Vol. 23, No. 4, 1987, pp. 459–468.doi:10.1016/0005-1098(87)90075-6
[13] Middleton, R. H., and Goodwin, G. C., “Adaptive Control of Time-Varying Linear Systems,” IEEE Transactions on Automatic Control,Vol. 33, No. 2, 1988, pp. 150–155.doi:10.1109/9.382
[14] Bartolini, G., Ferrara, A., and Stotsky, A. A., “Robustness andPerformance of an Indirect Adaptive Control Scheme in the Presence ofBounded Disturbances,” IEEE Transactions on Automatic Control,Vol. 44, No. 4, Apr. 1999, pp. 789–793.doi:10.1109/9.754819
[15] Sun, J., “A Modified Model Reference Adaptive Control Scheme forImproved Transient Performance,” IEEE Transactions on Automatic
Control, Vol. 38, No. 8, July 1993, pp. 1255–1259.doi:10.1109/9.233162
[16] Miller, D. E., and Davison, E. J., “Adaptive ControlWhich Provides anArbitrarily Good Transient and Steady-State Response,” IEEE
Transactions on Automatic Control, Vol. 36, No. 1, Jan. 1991,pp. 68–81.doi:10.1109/9.62269
[17] Costa, R., “Improving Transient Behavior of Model-ReferenceAdaptive Control.” Proceedings of the American Control Conference,Inst. of Electrical and Electronics Engineers, Piscataway, NJ, 1999,pp. 576–580.
[18] Ydstie, B. E., “Transient Performance and Robustness of DirectAdaptive Control,” IEEE Transactions on Automatic Control, Vol. 37,No. 8, Aug. 1992, pp. 1091–1105.doi:10.1109/9.151091
[19] Krstic, M., Kokotovic, P. V., and Kanellakopoulos, I., “TransientPerformance Improvement with a New Class of Adaptive Controllers,”Systems and Control Letters, Vol. 21, No. 6, 1993, pp. 451–461.doi:10.1016/0167-6911(93)90050-G
[20] Ortega, R., “Morse’s New Adaptive Controller: ParameterConvergence and Transient Performance,” IEEE Transactions on
Automatic Control, Vol. 38, No. 8, Aug. 1993, pp. 1191–1202.doi:10.1109/9.233152
[21] Zang, Z., and Bitmead, R., “Transient Bounds for Adaptive ControlSystems,” Proceedings of the 30th IEEE Conference on Decision and
Control, Inst. of Electrical and Electronics Engineers, Piscataway, NJ,Dec. 1990, pp. 2724–2729
[22] Zang, Y., and Ioannou, P. A., “Adaptive Control of Linear Time-Varying Systems,” Proceedings of the 35th IEEE Conference on
Decision and Control, Inst. of Electrical and Electronics Engineers,Piscataway, NJ, 1996, pp. 837–842.
[23] Datta, A., and Ioannou, P., “Performance Analysis and Improvement inModel Reference Adaptive Control,” IEEE Transactions on Automatic
Control, Vol. 39, No. 12, pp. 2370–2387.doi:10.1109/9.362856, Dec. 1994.
[24] Narendra, K. S., and Balakrishnan, J., “Improving Transient Responseof Adaptive Control Systems Using Multiple Models and Switching,”IEEE Transactions on Automatic Control, Vol. 39, No. 9, Sept. 1994,pp. 1861–1866.doi:10.1109/9.317113
[25] Marino, R., and Tomei, P., “An Adaptive Output Feedback Control fora Class of Nonlinear Systems with Time-Varying Parameters,” IEEE
Transactions on Automatic Control, Vol. 44, No. 11, 1999, pp. 2190–2194.doi:10.1109/9.802943
[26] Arteaga, A. M., and Tang, Y., “Adaptive Control of Robots with anImproved Transient Performance,” IEEE Transactions on Automatic
Control, Vol. 47, No. 7, July 2002, pp. 1198–1202.doi:10.1109/TAC.2002.800672
[27] Cao, C., and Hovakimyan, N., “Stability Margins of L1 AdaptiveController: Part 2,” American Control Conference, Inst. of Electricaland Electronics Engineers, Piscataway, NJ, 2007, pp. 3931–3936.
[28] Beard, R.W., Knoebel, N., Cao, C., Hovakimyan, N., andMatthews, J.,“An L1 Adaptive Pitch Controller for Miniature Air Vehicles,” AIAAGuidance, Navigation and Control Conference, AIAA Paper 2006-6777, 2006.
[29] Kaminer, I., Yakimenko, O., Dobrokhodov, V., Pascoal, A.,Hovakimyan, N., Cao, C., Young, A., and Patel, V. V., “CoordinatedPath Following for Time-Critical Missions of Multiple UAVs via L1
Adaptive Output Feedback Controllers,” AIAA Guidance, Navigationand Control Conference, Hilton Head Island, SC, AIAA Paper 2007-6409, Aug. 2007.
[30] Dobrokhodov, V., Kitsiois, I., Kaminer, I., Jones, K., Xargay, E.,Hovakimyan, N., Cao, C., Lizarraga, M., and Gregory, I., “FlightValidation of Metrics Driven L1 Adaptive Control,”. AIAA Guidance,Navigation, and Control Conference, Honolulu, HI, AIAA Paper 2008-6987, 2008.
[31] Aguiar, P., Pascoal, A., Kaminer, I., Dobrokhodov, V., Xargay, E.,Hovakimyan, N., Cao, C., and Ghabchello, R., “Time-CoordinatedPath Following ofMultiple UAVs over Time-Varying Networks UsingL1 Adaptation,”AIAAGuidance, Navigation and Control Conference,Honolulu, HI, AIAA Paper 2008-7131, 2008.
[32] Wise, K., Lavretsky, E., Hovakimyan, N., Cao, C., and Wang, J.,“Verifiable Adaptive Control: UCAV and Aerial Refueling,”AIAAGuidance, Navigation, and Control Conference and Exhibit, AIAAPaper 2008-6658, 2008.
[33] Lei, Y., Cao, C., Cliff, E., Hovakimyan, N., Kurdila, A., Bolender, M,and Doman, D., “Design of an L1 Adaptive Controller for an Air-Breathing Hypersonic Vehicle Model with Unmodelled Dynamics,”AIAA Guidance, Navigation and Control Conference, Hilton HeadIsland, SC, AIAA Paper 2007-6527, 2007.
[34] Gregory, I., Patel, V., Cao, C., and Hovakimyan, N., “L1
Adaptive Control Laws for Flexible Wind Tunnel Model of High-Aspect Ratio Flying Wing,”AIAA Guidance, Navigation, and ControlConference, Hilton Head Island, SC, AIAA Paper 2007-6525,2007.
[35] Cotting, M. C., Cao, C., Hovakimyan, N., Kraus, R., and Durham, W.,“Simulator Testing of Longitudinal Flying Qualities with L1 AdaptiveController,” AIAA Guidance, Navigation and Control Conference,Honolulu, HI, AIAA Paper 2008-6551, 2008.
[36] Cao, C., Patel, V. V., Reddy, K., Hovakimyan, N., and Lavretsky, E.,“Are the Phase and Time-Delay Margins Always Adversely AffectedbyHigh-Gain?,”AIAAGuidance, Navigation andControl Conference,AIAA Paper 2006-6347, 2006.
[37] Li, D., Hovakimyan, N., Cao, C., and Wise, K., “Filter Design forFeedback-Loop Trade-Off of L1 Adaptive Controller. A Linear MatrixInequality Approach,” AIAA Guidance, Navigation and ControlConference, AIAA Paper 2008-6280, 2008.
[38] Wise, K., “Robust Stability Analysis of Adaptive Missile Autopilots,”AIAA Guidance, Navigation, and Control Conference and Exhibit,AIAA Paper 2008-6999, 2008.
[39] Wang, J., Cao, C., Hovakimyan, N., Hindman, R., and Ridgely, B. “L1
Adaptive Controller for a Missile Longitudinal Autopilot Design,”.AIAA Guidance, Navigation, and Control Conference and Exhibit,AIAA Paper 2008-6282, 2008.
[40] Kharisov, E., Gregory, I., Cao, C., and Hovakimyan, N., “L1 AdaptiveControl Law for Flexible Space Launch Vehicle and Proposed Plan forFlight Test Validation,” AIAA Guidance, Navigation, and ControlConference and Exhibit, AIAA Paper 2008-7128, 2008.
[41] Khalil, H. K., Nonlinear Systems, Prentice–Hall, Englewood Cliffs,NJ, 2002.
726 CAO AND HOVAKIMYAN
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77
This article has been cited by:
1. Ahmed Awad, Haoping Wang. 2016. Roll-pitch-yaw autopilot design for nonlinear time-varying missile using partial stateobserver based global fast terminal sliding mode control. Chinese Journal of Aeronautics 29:5, 1302-1312. [CrossRef]
2. Fabian Hellmundt, Jens Dodenhöft, Florian HolzapfelL1 Adaptive Control with Eigenstructure Assignment for PolePlacement considering Actuator Dynamics and Delays . [Citation] [PDF] [PDF Plus]
3. Hamidreza Jafarnejadsani, Hanmin Lee, Naira HovakimyanConvex Multi-Objective Filter Optimization for OutputFeedback L1-Adaptive Controller . [Citation] [PDF] [PDF Plus]
4. Dan Xu, James Ferris Whidborne, Alastair Cooke. 2016. Fault tolerant control of a quadrotor using ℒ 1 adaptive control.International Journal of Intelligent Unmanned Systems 4:1, 43-66. [CrossRef]
5. Zhen Liu, Xiangmin Tan, Ruyi Yuan, Guoliang Fan, Jianqiang Yi. 2016. Immersion and Invariance-Based Output FeedbackControl of Air-Breathing Hypersonic Vehicles. IEEE Transactions on Automation Science and Engineering 13:1, 394-402.[CrossRef]
6. A. Awad, H. P. Wang. 2016. Integrated Pitch-Yaw Acceleration Autopilot Design for Varying-Velocity Man PortableMissile. International Journal of Modeling and Optimization 6:1, 11-17. [CrossRef]
7. Evgeny Kharisov, Carolyn L. Beck, Marc Bloom. 2015. Design of <mml:math altimg="si0002.gif"overflow="scroll" xmlns:xocs="http://www.elsevier.com/xml/xocs/dtd" xmlns:xs="http://www.w3.org/2001/XMLSchema" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns="http://www.elsevier.com/xml/ja/dtd" xmlns:ja="http://www.elsevier.com/xml/ja/dtd" xmlns:mml="http://www.w3.org/1998/Math/MathML"xmlns:tb="http://www.elsevier.com/xml/common/table/dtd" xmlns:sb="http://www.elsevier.com/xml/common/struct-bib/dtd" xmlns:ce="http://www.elsevier.com/xml/common/dtd" xmlns:xlink="http://www.w3.org/1999/xlink"xmlns:cals="http://www.elsevier.com/xml/common/cals/dtd" xmlns:sa="http://www.elsevier.com/xml/common/struct-aff/dtd"><mml:msub><mml:mrow><mml:mi mathvariant="script">L</mml:mi></mml:mrow><mml:mrow><mml:mn>1</mml:mn></mml:mrow></mml:msub></mml:math> adaptive controllers forhuman patient anesthesia. Control Engineering Practice 44, 65-77. [CrossRef]
8. Lei Pan, Jie Luo, Chengyu Cao, Jiong Shen. 2015. <mml:math altimg="si4.gif"overflow="scroll" xmlns:xocs="http://www.elsevier.com/xml/xocs/dtd" xmlns:xs="http://www.w3.org/2001/XMLSchema" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns="http://www.elsevier.com/xml/ja/dtd" xmlns:ja="http://www.elsevier.com/xml/ja/dtd" xmlns:mml="http://www.w3.org/1998/Math/MathML"xmlns:tb="http://www.elsevier.com/xml/common/table/dtd" xmlns:sb="http://www.elsevier.com/xml/common/struct-bib/dtd" xmlns:ce="http://www.elsevier.com/xml/common/dtd" xmlns:xlink="http://www.w3.org/1999/xlink"xmlns:cals="http://www.elsevier.com/xml/common/cals/dtd" xmlns:sa="http://www.elsevier.com/xml/common/struct-aff/dtd"><mml:mrow><mml:msub><mml:mrow><mml:mi mathvariant="script">L</mml:mi></mml:mrow><mml:mrow><mml:mn>1</mml:mn></mml:mrow></mml:msub></mml:mrow></mml:math> adaptivecontrol for improving load-following capability of nonlinear boiler–turbine units in the presence of unknown uncertainties.Simulation Modelling Practice and Theory 57, 26-44. [CrossRef]
9. Jie Luo, Chengyu Cao. 2015. L 1 adaptive control with sliding-mode based adaptive law. Control Theory and Technology13:3, 221-229. [CrossRef]
10. Quan Nguyen, Koushil SreenathL<inf>1</inf> adaptive control for bipedal robots with control Lyapunov function basedquadratic programs 862-867. [CrossRef]
11. Jie Luo, Chengyu Cao. 2015. Flocking for multi-agent systems with unknown nonlinear time-varying uncertainties undera fixed undirected graph. International Journal of Control 1-12. [CrossRef]
12. Nikolaos D. Tantaroudas, Andrea Da Ronch, Kenneth J. Badcock, Yinan Wang, Rafael PalaciosModel Order Reductionfor Control Design of Flexible Free-Flying Aircraft . [Citation] [PDF] [PDF Plus]
13. Dongkyoung Chwa. 2015. Fuzzy adaptive disturbance observer-based robust adaptive control for skid-to-turn missiles.IEEE Transactions on Aerospace and Electronic Systems 51:1, 468-478. [CrossRef]
14. Venkata S. Akkinapalli, Guillermo P. Falconi, Florian HolzapfelAttitude control of a multicopter using L<inf>1</inf>augmented quaternion based backstepping 170-178. [CrossRef]
15. Giovanni Mattei, Salvatore Monaco. 2014. Nonlinear Autopilot Design for an Asymmetric Missile Using RobustBackstepping Control. Journal of Guidance, Control, and Dynamics 37:5, 1462-1476. [Abstract] [Full Text] [PDF] [PDFPlus]
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77
16. Evgeny Kharisov, Naira Hovakimyan, Karl J. Åström. 2014. Comparison of architectures and robustness of model referenceadaptive controllers and L1 adaptive controllers. International Journal of Adaptive Control and Signal Processing 28:7-8,633-663. [CrossRef]
17. Ruyi Ren, Zaojian Zou, Xuegang WangL<inf>1</inf> adaptive control used in path following of surface ships 8047-8053.[CrossRef]
18. Nikolaos D. Tantaroudas, Andrea Da Ronch, Guanqun Gai, Kenneth J. BadcockAn Adaptive Aeroelastic Control Approachusing Non Linear Reduced Order Models . [Citation] [PDF] [PDF Plus]
19. Hanmin Lee, Venanzio Cichella, Naira HovakimyanL<inf>1</inf> adaptive output feedback augmentation of ModelReference Control 697-702. [CrossRef]
20. Shao-Ming He, De-Fu Lin. 2014. Missile two-loop acceleration autopilot design based on ℒℒ<sub>1</sub> adaptiveoutput feedback control. International Journal of Aeronautical and Space Sciences 15:1, 74-81. [CrossRef]
21. Hui Sun, Naira Hovakimyan, Tamer BasarReference tracking of uncertain nonlinear Multi-Input Multi-Output quantizedsystems using L<inf>1</inf> adaptive control 7546-7551. [CrossRef]
22. Elisa Capello, Fulvia Quagliotti, Roberto Tempo. 2013. Randomized Approaches for Control of QuadRotor UAVs. Journalof Intelligent & Robotic Systems . [CrossRef]
23. Mario Cassaro, Manuela Battipede, Piergiovanni Marzocca, Enrico Cestino, Aman Behal. 2013. ℒ 1 Adaptive FlutterSuppression Control Strategy for Highly Flexible Structure. SAE International Journal of Aerospace 6:2, 693-702.[CrossRef]
24. Justin Vanness, Evgeny Kharisov, Naira HovakimyanL<inf>1</inf> adaptive output-feedback controller for linear time-varying reference systems 4183-4188. [CrossRef]
25. Elisa Capello, Fulvia Quagliotti, Roberto TempoRandomized approaches and adaptive control for quadrotor UAVs 461-470.[CrossRef]
26. John Cooper, Chengyu CaoL<inf>1</inf> adaptive controller with additional uncertainty bias estimation 607-611.[CrossRef]
27. John Cooper, Jiaxing Che, Chengyu Cao. 2013. The use of learning in fast adaptation algorithms. International Journalof Adaptive Control and Signal Processing n/a-n/a. [CrossRef]
28. Costin Ene, Adrian-Mihail Stoica, James F Whidborne. 2013. Application of L1 Adaptive Controller to LongitudinalDynamics of a High Manoeuvrability Aircraft. IFAC Proceedings Volumes 46:19, 447-452. [CrossRef]
29. Lukas R.S. Theisen, Roberto Galeazzi, Mogens Blanke. 2013. Unmanned Water Craft Identification and Adaptive Controlin Low-Speed and Reversing Regions. IFAC Proceedings Volumes 46:33, 150-155. [CrossRef]
30. Elisa Capello, Giorgio Guglieri, Fulvia Quagliotti, Daniele Sartori. 2013. Design and Validation of an ${\mathcal{L}}_{1}$Adaptive Controller for Mini-UAV Autopilot. Journal of Intelligent & Robotic Systems 69:1-4, 109-118. [CrossRef]
31. Zhiyuan Li, Naira HovakimyanL<inf>1</inf> adaptive controller for MIMO systems with unmatched uncertainties usingmodified piecewise constant adaptation law 7303-7308. [CrossRef]
32. Jiaxing Che, Chengyu Cao&ℒ8466;<inf>1</inf> adaptive control of system with unmatched disturbance by usingeigenvalue assignment method 4823-4828. [CrossRef]
33. Ronald A. Hess. 2012. Frequency Domain-Based Pseudosliding Mode Flight Control Design. Journal of Aircraft 49:6,2077-2088. [Citation] [PDF] [PDF Plus]
34. Keum W. Lee, Sahjendra N. Singh. 2012. adaptive control of flexible spacecraft despite disturbances. Acta Astronautica80, 24-35. [CrossRef]
35. Jiaxing Che, Irene Gregory, Chengyu CaoIntegrated Flight/Structural Mode Control for Very Flexible Aircraft Using L1Adaptive Output Feedback Controller . [Citation] [PDF] [PDF Plus]
36. Ali Elahidoost, John Cooper, Chengyu Cao, Khanh PhamSatellite Orbit Stabilization Using L1 Adaptive Control .[Citation] [PDF] [PDF Plus]
37. Ali Elahidoost, Chengyu CaoControl and navigation of a three wheeled unmanned ground vehicle by L<inf>1</inf>adaptive control architecture 13-18. [CrossRef]
38. J. Luo, X. Zou, C. Cao. 2012. Eigenvalue assignment for linear time-varying systems with disturbances. IET ControlTheory & Applications 6:3, 365. [CrossRef]
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77
39. Evgeny Kharisov, Carolyn L. Beck, Marc Bloom. 2012. Control of Patient Response to Anesthesia using ℒ 1 AdaptiveMethods. IFAC Proceedings Volumes 45:18, 391-396. [CrossRef]
40. Jie Luo, Chengyu Cao&ℒ8466;<inf>1</inf> adaptive output feedback controller for a class of nonlinear systems 5425-5430.[CrossRef]
41. Kwang Ki Kevin Kim, Evgeny Kharisov, Naira HovakimyanFilter Design for &ℒ8466;<inf>1</inf> Adaptive Output-Feedback Controller 5653-5658. [CrossRef]
42. Vladimir Dobrokhodov, Isaac Kaminer, Ioannis Kitsios, Enric Xargay, Chengyu Cao, Irene M. Gregory, Naira Hovakimyan,Lena Valavani. 2011. Experimental Validation of L1 Adaptive Control: The Rohrs Counterexample in Flight. Journal ofGuidance, Control, and Dynamics 34:5, 1311-1328. [Citation] [PDF] [PDF Plus]
43. Keum Lee, Sahjendra SinghOutput Feedback Variable Structure Adaptive Control Approach to Missile Autopilot Design .[Citation] [PDF] [PDF Plus]
44. Markus Geiser, Enric Xargay, Naira Hovakimyan, Thomas Bierling, Florian HolzapfelL1 Adaptive Augmented DynamicInversion Controller for a High Agility UAV . [Citation] [PDF] [PDF Plus]
45. Evgeny Kharisov, Kwang Ki Kim, Naira Hovakimyan, Xiaofeng WangLimiting Behavior of L1 Adaptive Controllers .[Citation] [PDF] [PDF Plus]
46. Evgeny Kharisov, Naira Hovakimyan&ℒx2112;<inf>1</inf> adaptive output feedback controller for minimum phasesystems 1182-1187. [CrossRef]
47. Dapeng Li, Naira Hovakimyan, Tryphon GeorgiouRobustness of &ℒ8466;<inf>1</inf> adaptive controllers in the gapmetric in the presence of nonzero initialization 2723-2728. [CrossRef]
48. Robert Fuentes, Jesse Hoagg, Blake Anderton, Anthony D'Amato, Dennis BernsteinInvestigation of CumulativeRetrospective Cost Adaptive Control for Missile Application . [Citation] [PDF] [PDF Plus]
49. Evgeny Kharisov, Naira Hovakimyan, Karl ÅströmComparison of Several Adaptive Controllers According to TheirRobustness Metrics . [Citation] [PDF] [PDF Plus]
50. Keum W. Lee, Sahjendra N. Singh. 2010. Noncertainty-Equivalent Adaptive Missile Control via Immersion and Invariance.Journal of Guidance, Control, and Dynamics 33:3, 655-665. [Citation] [PDF] [PDF Plus]
51. Isaac Kaminer, Antonio Pascoal, Enric Xargay, Naira Hovakimyan, Chengyu Cao, Vladimir Dobrokhodov. 2010. PathFollowing for Small Unmanned Aerial Vehicles Using L1 Adaptive Augmentation of Commercial Autopilots. Journal ofGuidance, Control, and Dynamics 33:2, 550-564. [Citation] [PDF] [PDF Plus]
52. Keum Lee, Sahjendra SinghNoncertainty-Equivalent Adaptive Missile Control via Immersion and Invariance . [Citation][PDF] [PDF Plus]
53. Irene Gregory, Chengyu Cao, Enric Xargay, Naira Hovakimyan, Xiaotian ZouL1 Adaptive Control Design for NASAAirSTAR Flight Test Vehicle . [Citation] [PDF] [PDF Plus]
Dow
nloa
ded
by B
EIH
AN
G U
NIV
ER
SIT
Y o
n D
ecem
ber
20, 2
016
| http
://ar
c.ai
aa.o
rg |
DO
I: 1
0.25
14/1
.408
77