Beruflich Dokumente
Kultur Dokumente
Eugene Lavretsky∗
The Boeing Company, Huntington Beach, California 92647
DOI: 10.2514/1.G004328
In this paper, a multi-input/multi-output servomechanism primary flight controller is presented. Also discussed are
flight data related to the system performance and pilot assessments of handling qualities. The data were collected
during flight-test experiments (March 2018) on the Calspan Variable Stability In-Flight Simulator Learjet-25B
aircraft. Aircraft body angular rates and linear accelerations are used to enable decoupled command tracking in
pitch, roll, and yaw axes. Angle of attack and angle of sideslip measurements are not required for the system to
operate. The developed primary flight controller translates pilot controls into the corresponding pitch, roll, and yaw
Downloaded by MCMASTER UNIVERSITY on September 30, 2019 | http://arc.aiaa.org | DOI: 10.2514/1.G004328
commands for the aircraft to track. The system consists of a gain-scheduled robust linear baseline augmented with a
direct adaptive model reference output feedback. The baseline is designed via the Observer-Based Loop Transfer
Recovery method. The adaptive component incorporates the closed-loop reference model, whereby a state observer
from the baseline Observer-Based Loop Transfer Recovery system also acts as the reference model for the adaptive
augmentation. In March 2018, the Observer-Based Loop Transfer Recovery controller was successfully flight verified
on the Calspan Learjet-25B aircraft, at the U.S. Air Force Test Pilot School, Edwards Air Force Base, California.
Flight-test results are discussed, and a brief summary is given at the end.
robust control definitions are introduced in Sec. V. Signals with hats closed-loop system response in the presence of modeling errors,
represent estimated values of their respective components. These control deficiencies, and environmental uncertainties. These
quantities are computed by the pitch and the roll/yaw state observers, unknowns may cause significant changes in aircraft aerodynamics,
the generic definitions of which are also presented in this section. propulsion, and structures.
OBLTR adaptive output feedback augmentation is defined in Sec. VI. It is expected that in a symmetric aircraft configuration in level
Section VII gives an overview of OBLTR flight evaluations, related flight conditions coupling between the vehicle pitch and roll/yaw
test data, and pilot comments. The paper ends with a brief summary in dynamics is negligible. For example, the moderate sweep angle on
Sec. VIII. the LJ-25B aircraft produces a minor degree of coupling between
longitudinal and lateral/directional dynamics. Stability and
performance robustness due to these unaccounted for in the control
II. Flight Control of Aerial Systems via OBLTR Design design dynamics must be analyzed and then tested in Monte Carlo–
Driven by a pilot or a guidance logic, the OBLTR flight controller based simulations. Assuming decoupled longitudinal and lateral/
has the ability to maintain (when possible) and to enforce a graceful directional modes, the baseline pitch and roll/yaw controllers could
degradation of the desired baseline closed-loop performance, while be designed separately. Figure 2 shows the pitch axis controller
operating in the presence of significant uncertainties that may exist in designed to regulate a linear combination of the aircraft vertical
aerodynamics, propulsion, configuration, and environmental acceleration N z (gravity) and the body angular pitch rate qb
disturbances. Like other robust controllers, the proposed design (radians∕second). The resulting quantity is called the Cstar,
assumes controllability and observability of the system under
Downloaded by MCMASTER UNIVERSITY on September 30, 2019 | http://arc.aiaa.org | DOI: 10.2514/1.G004328
uncertainties. Also, it is assumed that the system control actuator C N z 12.4qb ; g (1)
position and rate limits are adequate to mitigate “unknown
unknowns.” OBLTR is designed to use available sensor
measurements only. The system goal is to achieve and maintain Reference [18] presents rationale and details on the criterion
predictable and stable operation of an aerial vehicle throughout the formulation. It is shown that regulating a linear combination of the
flight envelope. aircraft incremental vertical acceleration and the pitch rate in the form
The OBLTR controller is deterministic. It prescribes stable, of Eq. (1) directly correlates and significantly improves handling
predictable, and repeatable online adjustments of its baseline and properties of the corresponding closed-loop responses due to piloted
adaptive control parameters. By combining a robust baseline inner- inputs. To that end, the pitch rate scaling factor is selected based on
loop controller with a direct model reference adaptive augmentation, the desired crossover airspeed region, to emphasize either the vehicle
the resulting system receives shared benefits of each component. vertical acceleration or the pitch rate response.
While the baseline controller provides the required target Similarly, the OBLTR baseline roll/yaw controller (Fig. 3) is
performance with robustness and stability properties under nominal
designed to regulate the aircraft stability axis roll rate ps
conditions, the adaptive augmentation allows the system to retain the
(radians∕second),
desired closed-loop performance in the presence of aerodynamic,
propulsive, environmental, and other uncertainties that may result in
adverse conditions encountered during flight. ps pb cos α^ rb sin α^ (2)
Figure 1 shows a generic OBLTR-based control block diagram.
There are four essential elements in the system: 1) the robust baseline and the body lateral acceleration Ay (feet∕second2 ), where pb , rb ,
linear controller, 2) the state observer/closed-loop reference model, and α^ denote the aircraft body roll rate (radians∕second), body yaw
3) the direct adaptive augmentation, and 4) the piloted interface or rate (radians∕second), and the estimated angle of attack (radians),
guidance commands. correspondingly.
Feedforward connections from the pilot/guidance interface As shown in Figs 2 and 3, the aircraft true AOA and AOS signals
(shown in green) are optional and can be added to the total control are not fed back as output measurements. OBLTR relies only on their
signal in order to shape the system closed-loop augmented command estimated values. Furthermore, both pitch and roll/yaw OBLTR
responses. As a servomechanism, the OBLTR controller is designed controllers have a common generic architecture that incorporates a
to make the system regulated outputs follow external commands state observer with proportional and integral feedback connections.
generated by an operator interface or a guidance logic, for piloted or Also of note, the signals fed into the state observer as true
autonomous vehicles, respectively. The baseline control component measurements include integrated tracking errors, instead of the
is constructed assuming nominal plant dynamics. If in operation the original regulated output. It is straightforward to show that replacing
system measurements significantly differ from their estimated a measurement with its integrated value preserves the system
values, the adaptive augmentation provides an additive correction observability. From a practical perspective, using integrated linear
signal to the baseline control. The goal is to maintain the desired accelerations instead of the original signals significantly reduces
Measurements
Pitch ⎛ eˆC∗ I ⎞
Dynamic ⎜ ⎟ D p reg
αˆ ⎟
Compensator ⎜⎜ Short-Period Dynamics
qˆ ⎟
δe
(s I ) (α , qb ) C
−1
⎝ ⎠ −K
( s I n×n − Aobs )
−1 b
− Ap
*
Ccmd Bcmd lqr
n p ×np Bp p reg C∗
(e )
T q
Lv C∗ I
q
1 eC ∗ I
s C ∗ = N z + 12.4 qb , ( g )
Roll / Yaw ⎛ eˆ ps I ⎞
⎜ ⎟
⎜ eˆAy I ⎟
Dynamic ⎜ ˆ ⎟ D p reg
Compensator
⎜ β ⎟
⎜ pˆ ⎟
⎛ δ ail ⎞ Roll/Yaw Dynamics
⎜ b ⎟ ⎜ ⎟
⎝ δ rud ⎠
(s I )
⎛ ps cmd ⎞ ⎜ rˆ ⎟ −1 xp ⎛ ps ⎞
( s I n×n − Aobs )
−1
Downloaded by MCMASTER UNIVERSITY on September 30, 2019 | http://arc.aiaa.org | DOI: 10.2514/1.G004328
⎝ b ⎠
⎜ ⎟ Bcmd − K lqr − Ap Bp C p reg ⎜ ⎟
⎝ Ay cmd ⎠ n p ×n p
⎝ Ay ⎠
⎛ pb ⎞
⎜ ⎟
(e ) ⎝ rb ⎠
T
Lv ps I eAy I pb rb
1 e ps I
s eAy I
15
Fig. 3 OBLTR baseline roll/yaw controller.
system sensitivity to measurement noise that is often present in system dynamics, control gains, and the system state observer
accelerometers. related data.
All signals required for OBLTR design, analysis, and
implementation are defined in Table 1. For presentation clarity, the
longitudinal signals are listed first, followed by the lateral-directional III. Open-Loop System
parameters. Modeling for control begins with the formulation of a finite set of
The OBLTR baseline system, with the pitch and the roll/yaw linear time-invariant plants in the classical state-space form
controllers combined, forms a square 3 × 3 servomechanism
problem for tracking three commands Ccmd pscmd Aycmd using
x_p Ap xp Bp u; ymeas Cpmeas xp Dpmeas u;
three control inputs: aileron, elevator, and rudder δe δa δr ,
respectively. For a conventional piloted aircraft such as LJ-25B, the yreg Cpreg xp Dpreg u (3)
three commands would be connected to and generated by the pilot
interface: pitch/roll stick and rudder pedals. As such, the OBLTR whereby each linear system is verified to be controllable and
controller becomes a primary control augmentation system designed observable (that is minimal), while the corresponding dynamics
to interpret, augment, and modify pilot inputs in order to track approximate the evolution of the true process (the aircraft dynamics)
respective commands and at the same time to maintain the desired around its selected equilibrium (trim). In Eq. (3), xp ∈ Rnp denotes
closed-loop system stability and performance characteristics. the np - dimensional state, u ∈ Rm is the m-dimensional vector of
Other parameters shown in the OBLTR baseline control block controls, ymeas ∈ Rnymeas combines the system output measurements,
diagrams (Figs. 2 and 3) notionally describe the aircraft open-loop and yreg ∈ Rnyreg represents the regulated output. These output signals
Fig. 4 Learjet-25 flight envelope with control design and flight evaluation points.
Preg s Cpreg sI np ×np − Ap −1 Bp Dpreg (4) IV. State Feedback Linear Quadratic Regulator
OBLTR design starts with the construction of an optimal state
feedback LQR regulator for the extended system dynamics (7). The
from the control input u to the regulated output yreg , has no LQR state feedback solution is formulated as a conventional
transmission zeros at the origin. These two assumptions guarantee proportional integral (PI) servocontroller,
controllability of the extended open-loop system,
yreg − ycmd
u −K lqr x −KI eyI − K xp xp −KI − K xp
e_yI yreg − ycmd x_ p Ap xp Bp u (5) s
ycmd − yreg
xp K I − K xp xp (8)
with the integrated tracking error eyI yreg − ycmd ∕s embedded s
into the system dynamics to create a type-1 controller. Rewriting
Eq. (5) in matrix form gives The matrix of optimal feedback gains,
solution (8) is combined with a dynamic state observer in such a way p − inputs
0 ⇓
1
that the closed-loop tracking performance remains optimal while at
z}|{
the same time the LQR state feedback stability margins are recovered p − outputs ⇒ @ A B B2 A ∈ Rnp×np
at the plant input, and the overall control architecture takes the form Cmeas 0p×p
of a dynamic output feedback controller operating on available
sensor measurements only.
has its transmission zeros in the open left half complex plane C− . The
squaring-up design entails placing p − m transmission zeros at the
desired locations in C− . Constructive and numerically efficient
V. OBLTR Baseline Control algorithms for squaring up are developed in [17]. Solutions are not
The OBLTR baseline design starts with a redefinition of the system unique. A set of sufficient conditions for a solution to exist is given as
measurements. Basically, any output signal that has a direct follows.
feedforward control connection (a nonzero D matrix component) is Assumption 1:
replaced by its integrated value. Such an operation does not destroy 1. A; Cmeas is observable.
the observability of the original system pair Ap ; Cp . For LJ-25B, 2. The system is tall p > m, and rankCmeas B m.
there are two regulated outputs with nonzero D matrices: C (1) and 3. The tall system (12) has no finite transmission zeros in the closed
the aircraft lateral acceleration Ay . Treated as measurements for the right half complex plane.
observer, both signals are replaced with their integrated tracking The first condition (observability) is basic. The second assumption
errors, eC I and eAy I , as defined in Table 1, without a loss of the (tallness) is often true in all flight control applications, and the third
Downloaded by MCMASTER UNIVERSITY on September 30, 2019 | http://arc.aiaa.org | DOI: 10.2514/1.G004328
system observability. assumption is almost always satisfied for all tall systems. In essence,
Generalizing this concept, the system measurements are written as the three assumed relations are not restrictive going forward.
It can be verified that the transfer function from the elevator δe to
I m×m 0m×np eyI 0m×m the vertical acceleration N z has an unstable zero. However, one can
ymeas u Cmeas x (11) show that the corresponding tall system driven by the elevator and
0p×m Cpmeas xp Dpmeas
|{z} |{z} |{z} having two outputs represented by the vertical acceleration and the
Cmeas x Dmeas 0p×m
pitch rate satisfies the third assumption. That is, the tall system has no
finite unstable transmission zeros.
whereby the first m output components represent the system Under these three assumptions, the squaring-up design for the
integrated tracking errors eyI , and the last p − m components are extended open-loop dynamics (12) is accomplished by adding
the original system measurements without direct feedforward control pseudocontrol inputs B2 ∈ Rp−m to compose an extended B matrix
connections. One can prove the following fact: if the original system and to place (p − m) transmission zeros at the desired locations in the
matrix pair Ap ; Cp is observable, then the extended system pair left-half open complex plane C− .
A; Cmeas remains observable.
In Eq. (11), it is explicitly assumed that the extended open-loop B B; B2
dynamics are “tall,”
≠ 0 ∧ det sI − A
⇒ detCmeas B
B
0 ⇒ s ∈ C−
x_ Ax Bu Bcmd ycmd ymeas Cmeas x Dmeas u (12) Cmeas 0
that is, there are more measurements than control inputs u ∈ Rm . For (13)
flight control applications, the tall assumption is not restrictive.
For LJ-25B pitch axis OBLTR control design, there are two Next, the full-state Luenberger observer for Eq. (12) is formulated
measurements p 2 and one control input m 1: in its standard form [1,3,6]:
ymeas eC I qb T , u δe . For the roll/yaw controller, four
measurements p 4 and two control inputs m 2 are used: x_^ Ax^ Bu Bcmd ycmd Lv ymeas − y^ meas (14)
ymeas eφI eAy I pb rb T , u δa δr .
The next concept is pivotal for OBLTR. It is called the squaring-up The observer dynamics are driven by the control input u, the
design. The squaring-up problem for a nonsquare linear MIMO command ycmd , the estimated measurement
system, with m inputs and m < p outputs,
y^ meas Cmeas x^ (15)
m − inputs
0 1
⇓ and the innovation term ymeas − y^ meas feedback, with the parameter-
z}|{
p − outputs ⇒ @ A B A ∈ Rnp×nm dependent Luenberger gain,
Cmeas Dmeas
Lv Pv CTmeas R−1
v (16)
can be stated as follows [17]: given the state matrix A ∈ the Rn×n ,
input matrix B ∈ Rn×m , and the output matrices Cmeas ∈ Rp×n , In Eq. (16), Pv is the unique positive-definite symmetric
Dmeas ∈ Rp×m , with n > m; p > m, determine pseudoinput parameter-dependent solution of the ARE,
matrices B2 ∈ Rn×p−m and D2 ∈ Rp×p−m , such that the resulting
square system with p inputs and p outputs, Pv AT APv Qv − Pv CTmeas R−1
v Cmeas Pv 0 (17)
has its transmission zeros in the open left half complex plane, C− . Note that Eq. (17) is similar to the steady-state ARE in Kalman
Systems considered in this paper have no feedforward connections filtering [3,4]. The parameterization of the matrix weights (18) comes
(11), that is, Dmeas 0p×m . So, the squaring-up problem is reduced from [1,2]. It differs significantly from the LQG/LTR design
to finding a pseudo-input matrix B2 ∈ Rn×p−m , such that the p × [4,19,20], in which only the system-process-related weights Rv are
p square system, increased in order to recover stability margins. Also, note that the
2162 LAVRETSKY
u out
u in
Dp reg
Dynamic Compensator Plant
(s I )
x̂ −1 xp
( s I n×n − Aobs )
−1
ycmd Bcmd −K
np ×np − Ap Bp C p reg yreg
⎛ ey I ⎞ y p meas
Lv ⎜ ⎟ C p meas
⎝ y p meas ⎠
1 ey I
s
squaring-up solution B in Eq. (18) is employed to form the ARE control position and rate limits are not explicitly taken into
weight matrix Qv . consideration. Their effects are tested in simulations.
The OBLTR baseline control uses the estimated system state The original design concept was referred to as the observer-based
vector x^ from the observer (14) in a feedback loop, adaptive control [2]. Subsequently, in [6,7], stability proofs were
Downloaded by MCMASTER UNIVERSITY on September 30, 2019 | http://arc.aiaa.org | DOI: 10.2514/1.G004328
B C _^
B C Θ ^ Γ Φ
ProjΘ; ^ ubl ymeas − y^ meas T R−1∕2
x; WST (40)
x_ Ax Bubl BΛB uad ΘT Φx I m×m − Λ−1 ubl C Θ 0
B |{z} C
B C In [6,7], it is formally proven that for a sufficiently small positive v
B ubl C
@ I m×m − Λ−1 ΘT Θ bl ;xA
T Φu
the time derivative of Vex ; ΔΘ,
along the estimation error dynamics
|{z} Φx
ΘT |{z} (38), becomes negative semidefinite, which establishes Lyapunov
Φu
bl ;x stability of the coupled system (38), (40) equilibrium. Also asserted is
Bcmd ycmd (31) the UUB property for all internal signals. Asymptotic stability of the
estimation error follows from application of the Barbalat lemma [21].
Dynamics of the observer tracking error,
gives
e x^ − xref (41)
T Φu
x_ Ax Bubl BΛuad Θ bl ; x Bcmd ycmd (32)
are derived by subtracting Eq. (36) from Eq. (35):
Based on Eq. (32), the state observer (14) would have to be
redefined: e_ Aref e − Lv Cex (42)
then the state observer (33) becomes identical to the original one (14). and consequently, the state xt of the original system (29) tracks the
In addition, substituting the baseline control component from state xref t of the open-loop reference model (36). The reference
Eq. (30), the state observer takes the form of a CRM: model tracking performance of the OBLTR (Baseline Adaptive)
output feedback controller (30), (34), (40) is global and asymptotic.
x_^ A − BKlqr x^ Bcmd ycmd Lv ymeas − y^ meas (35) As noted earlier, the derivation of the adaptive laws (40) is largely
|{z} predicated on the ARE asymptotics (27). The latter builds a bridge
Aref between achieving loop transfer recovery properties of robust control
and the design of an adaptive output feedback for model reference
^
In Eq. (34), Θt denotes a time-varying matrix of the adaptive control. In that context, the state observer also acts as the CRM for
parameters, and Φu bl ; x ^ T is the regressor vector.
^ uTbl ΦT x adaptive control. In [15], it is shown that if rates of adaptation are
It depends on the system estimated state x^ and on the OBLTR baseline selected as ΓΘ OkLv k the transient performance in CRM-based
control ubl −Klqr x. ^ Components of the regressor vector are adaptive control is superior to that of Model Reference Adaptive
selected to enable approximation of matched uncertainties in the Control systems with open-loop reference models. Reduced
system dynamics. transients and global asymptotic tracking performance are the two
Without the measurement innovation feedback term main benefits of the CRM-based design.
Lv ymeas − y^ meas , the CRM dynamics (35) become identical to the
standard (also known as open-loop) reference model in model
reference adaptive control [13,22], VII. OBLTR Controller Design and Evaluations on
Calspan LJ-25B Aircraft
x_ ref Aref xref Bcmd ycmd (36)
Open-loop linear data for OBLTR control design were calculated
using the Calspan LJ-25B high-fidelity flight simulation
For the state estimation error, environment [24]. The aircraft flight operational envelope is shown
in Fig. 4.
ex x^ − x (37) The OBLTR baseline and adaptive control designs were performed
at the four dynamic pressure values, shown as red dashed lines in the
the corresponding estimation error dynamics can be derived [1,2]: figure. For real-time operation, control parameters were gain
2164 LAVRETSKY
Downloaded by MCMASTER UNIVERSITY on September 30, 2019 | http://arc.aiaa.org | DOI: 10.2514/1.G004328
Fig. 7 LJ-25B pitch step-input closed-loop responses with OBLTR Fig. 9 LJ-25B SISO elevator loop gains with the OBLTR baseline
baseline controller and feedforward gains. controller and actuator dynamics.
Fig. 8 LJ-25B roll/yaw step-input closed-loop responses with OBLTR Fig. 10 LJ-25B MIMO roll/yaw loop gains with the OBLTR baseline
baseline controller and feedforward gains. controller and actuator dynamics.
VIII. Conclusions [9] Gibson, T. E., Qu, Z., Annaswamy, A. M., and Lavretsky, E., “Adaptive
Output Feedback Based on Closed-Loop Reference Models,” IEEE
The main question addressed in this paper is related to the use of Transactions on Automatic Control, Vol. 60, No. 10, Oct. 2015,
robust and adaptive control methodologies to design flight critical pp. 2728–2733.
systems for piloted and autonomous aerial platforms. In particular, doi:10.1109/TAC.2015.2405295
the recently developed Observer-Based Loop Transfer Recovery [10] Balas, G., and Young, P., “Sensor Selection via Closed-Loop Control
(OBLTR) (Baseline Adaptive) dynamic output feedback control Objectives,” IEEE Transactions on Automatic Control, Vol. 7, No. 6,
architecture is presented, applied, and flight verified on the Calspan 2003, pp. 692–704.
VSS LJ-25B aircraft. In March of 2018, the OBLTR controller was doi:10.1109/87.799670
successfully implemented and flight evaluated at the U.S. Air Force [11] Doyle, J. C., Francis, B. A., and Tannenbaum, A. R., Feedback Control
Theory, Macmillan, New York, 1992.
Test Pilot School, Edwards Air Force Base, California. The flight-test [12] Aström, K. J., and Murray, R. M., Feedback Control Systems: An
plan included evaluations of the system closed-loop tracking Introduction for Scientists and Engineers, Princeton Univ. Press,
performance, stability, robustness, and the aircraft handling qualities Princeton, NJ, 2008.
under nominal and off-nominal flight conditions. The paper presents [13] Ioannou, P., and Fidan, P., Adaptive Control Tutorial, SIAM, Advances
an overview of the OBLTR design, its application to Calspan VSS LJ- in Design and Control, Soc. for Industrial and Applied Mathematics,
25 aircraft, and a brief summary of flight-test evaluation results. Philadelphia, PA, 2006.
Detailed analysis of flight-test data, lessons learned, and Cooper– doi:10.1137/1.9780898718652
Harper pilot ratings for the OBLTR controller can be found in [14] Lavretsky, E., “Reference Dynamics Modification in Adaptive
Controllers for Improved Transient Performance,” AIAA Paper 2011-
[25–27].
Downloaded by MCMASTER UNIVERSITY on September 30, 2019 | http://arc.aiaa.org | DOI: 10.2514/1.G004328
6200, 2011.
doi:10.2514/6.2011-6200
Acknowledgments [15] Gibson, T. E., Annaswamy, A. M., and Lavretsky, E., “Improved
Transient Response in Adaptive Control Using Projection Algorithms
The author would like to thank the U.S. Air Force Test Pilot School and Closed Loop Reference Models,” AIAA Paper 2012-4775, 2012.
Class 17B Team Have PACMAN, whose pilots and flight-test doi:10.2514/6.2012-4775
engineers relentlessly pursued and flawlessly executed a very [16] Thomson, C. M., Coleman, E. E., and Blight, J. D., “Integral LQG
detailed Observer-Based Loop Transfer Recovery flight evaluation Controller Design for a Fighter Aircraft,” Proceedings of AIAA
plan, over the course of two weeks at Edwards Air Force Base, Guidance, Navigation, and Control Conference, AIAA Paper 1987-
California. The author would like to also recognize Chris Cotting, Jay 2452, 1987.
Kemper, and Calspan personnel for their dedicated technical support [17] Misra, P., “Numerical Algorithms for Squaring-Up Non-Square
Systems, Part II: General Case,” Proceedings of the American Control
and engineering assistance during the program. Recommendations
Conference, IEEE Publ., Piscataway, NJ, 1988.
for Observer-Based Loop Transfer Recovery flight evaluations were [18] Malcom, L. G., and Tobie, H. N., “New Short Period Handling Quality
endorsed and sponsored by Sharon Stockbridge, Guidance, Criterion for Fighter Aircraft,” Boeing Co., Document D6-17841 T/N,
Navigation & Control engineering team lead at Eglin Air Force 1965.
Base, Fort Walton Beach, Florida. [19] Doyle, J. C., and Stein, G., “Multivariable Feedback Design: Concepts
for a Classical/Modern Synthesis,” IEEE Transactions on Automatic
Control, Vol. 26, No. 1, 1981, pp. 4–16.
References doi:10.1109/TAC.1981.1102555
[1] Lavretsky, E., and Wise, K. A., Robust and Adaptive Control with [20] Stein, G., and Athans, M., “The LQG/LTR Procedure for Multivariable
Aerospace Applications, Advanced Textbooks in Control and Signal Feedback Control Design,” IEEE Transactions on Automatic Control,
Processing, Springer–Verlag, London, 2013. Vol. 32, No. 2, 1987, pp. 105–114.
doi:10.1007/978-1-4471-4396-3 doi:10.1109/TAC.1987.1104550
[2] Lavretsky, E., “Adaptive Output Feedback Design Using Asymptotic [21] Khalil, H. K., Nonlinear Systems, 3rd ed., Prentice–Hall, Upper Saddle
Properties of LQG/LTR Controllers,” IEEE Transactions on Automatic River, NJ, 1996.
Control, Vol. 57, No. 6, 2012, pp. 1587–1591. [22] Narendra, K. S., and Annaswamy, A. M., Stable Adaptive Control,
doi:10.1109/TAC.2011.2174692 Dover, New York, 2005.
[3] Anderson, B. D. O., and Moore, J. B., Optimal Control: Linear [23] Lavretsky, E., and Gibson, T. E., Projection Operator in Adaptive
Quadratic Methods, Dover, New York, 1980. Systems, IEEE Publ., Piscataway, NJ, 2011, https://arxiv.org/abs/1112
[4] Kwakernaak, H., and Sivan, R., Linear Optimal Control Systems, .4232.
Wiley-Interscience, New York, 1972. [24] Berger, T., Tischler, M., Hagerott, S. G., Cotting, M. C., Gray, W. R.,
[5] Luenberger, D. G., “Observing the State of a Linear System,” IEEE Gresham, J., George, J., Krogh, K., and D’Argenio, A., Howland, J.,
Transactions on Military Electronics, MIL-8, Vol. 8, No. 2, 1964, et al., “Development and Validation of a Flight-Identified Full-Envelope
pp. 74–80. Business Jet Simulation Model Using Stitching Architecture,”
doi:10.1109/TME.1964.4323124 Proceedings of the AIAA SciTech Forum, AIAA Paper 2017-1550,
[6] Gibson, T. E., Annaswamy, A. M., and Lavretsky, E., “Closed–Loop 2017.
Reference Model Adaptive Control, Part I: Transient Performance,” doi:10.2514/6.2017-1550
Proceedings of the American Control Conference, IEEE Publ., [25] Alt, E., Chappell, C., Eldan, Y., Heemstra, B., Guerrero, D., and
Piscataway, NJ, 2013. Lancaster, J., “Evaluation of Observer-Based Loop Transfer Recovery
doi:10.1109/ACC.2013.6580353 Adaptive Controller (Project Have PACMAN),” Final Technical
[7] Gibson, T. E., Annaswamy, A. M., and Lavretsky, E., “Closed–Loop Information Memorandum, U.S. Air Force Test Pilot School
Reference Model Adaptive Control: Composite Control and Observer USAFTPS-TIM-17B-03, June 2018.
Feedback,” Proceedings of the 11th IFAC International Workshop on [26] Cotting, M. C., “Experimental Flight Testing of Prototype Control Law
Adaptation and Learning in Control and Signal Processing, IEEE Publ., Designs at the USAF Test Pilot School,” AIAA Scitech 2019 Forum,
Piscataway, NJ, 2013. AIAA Paper 2019-1079, 2019.
doi:10.1109/ACCESS.2013.2284005 doi:10.2514/6.2019-1079
[8] Lavretsky, E., “Transients in Output Feedback Adaptive Systems [27] Cotting, M. C., Alt, E., Chappell, C., Heemstra, B., Guerrero, D., and
with Observer-Like Reference Models,” International Journal of Lancaster, J., “Piloted Handling Qualities Evaluation of an OBLTR
Adaptive Control and Signal Processing, Vol. 29, No. 12, 2015, Prototype Flight Control Law at the USAF Test Pilot School,” AIAA
pp. 1515–1525. Scitech 2019 Forum, AIAA Paper 2019-1082, 2019.
doi:10.1002/acs.2557 doi:10.2514/6.2019-1082