Advanced Autopilot Design With Extremum-Seeking Control For Aircraft Control

Open Engineering 2024; 14: 20240044
Regular Article
Haci Baran* and Ismail Bayezit
Advanced autopilot design with extremum-

seeking control for aircraft control
https://doi.org/10.1515/eng-2024-0044 stability and control. This novel method reduces operator
received November 29, 2023; accepted May 09, 2024 workload and ensures reliable and efficient autonomous
Abstract: The aim of this research is to enhance adaptive flight capabilities. Additionally, the adaptability of the
autopilots for the effective management of aircraft sys- Autopilot-ESC to changing environmental conditions make
tems, control maintenance, and the rejection of external it well-suited in aircraft UAVs. This upgraded version of
disturbances. To achieve this objective, we propose the autopilot surpasses other robust controllers, such as Linear
design of an autopilot integrated with the extremum- Quadratic Gaussian (LQG) regulator and Model Predictive
seeking control (ESC) algorithm. Although autopilots pro- Control (MPC), as it can effectively address ramp, sinusoidal,
ficiently manage the lateral and longitudinal modes of and step disturbances, which LQG and MPC cannot handle.
aircraft control, they lack filtering or adaptive capabilities, Keywords: flight, pitch attitude hold autopilot, altitude
thereby exposing the system to significant external threats. hold autopilot, yaw autopilot, extremum seeking control,
To mitigate these risks, the ESC method is employed. This disturbance, disturbance rejection
adaptive approach can operate in a disturbance rejection
manner by adjusting parameters for unknown inputs and
restoring the system to its original controlled response. ESC
represents a versatile control method suitable for effec- 1 Introduction
tive application in simulations or experimental models.
Through the incorporation of this method, the pitch atti- Unmanned aerial vehicles (UAVs) present numerous advan-
tude hold autopilot, altitude hold autopilot, and yaw tages over conventional aircraft. By lacking human passen-
autopilot acquire advanced disturbance rejection capabil- gers, accidents involving these compact aircraft do not
ities with adaptive ESC features. The novelty of the pro- result in human fatalities. Furthermore, UAVs obviate the
posed method lies in providing advanced disturbance necessity for a pilot, thereby enabling them to proficiently
rejection properties to conventional autopilots, thereby undertake arduous and perilous tasks. Additionally, UAVs
rendering them innovative and superior disturbance rejec- are characterized by their lightweight construction and the
tion controllers. The newly developed autopilots are capable capacity to operate continuously for prolonged durations
of eliminating severe disturbances from the system response, without interruption. In recent years, UAVs have undergone
including ramp, sinusoidal, and step disturbances. The inte- significant advancements and are now extensively deployed
gration of autopilots with ESC offers significant advantages, across various domains including military operations, sur-
such as superior disturbance rejection properties for the air- veillance, transportation, wildfire suppression, and more,
craft unmanned aerial vehicle (UAV) system. The proposed particularly in hazardous environments. The efficacy of
method successfully eliminates severe disturbances, as UAV missions crucially depends on the implementation of
demonstrated in simulation results, surpassing previous effective and dependable control methodologies. However,
methods in effectiveness. Furthermore, the Autopilot-ESC developing a control system for such aerial applications
method enhances aircraft operation even under distur- presents a formidable challenge due to their extensive and
bances, minimizing energy consumption and ensuring irregular flight envelopes. Safety emerges as a paramount
concern in the aeronautical industry, particularly for UAVs,
which are increasingly employed in critical missions.
 Small UAVs have garnered significant popularity owing
* Corresponding author: Haci Baran, Department of Aeronautical
to their portability, cost-effectiveness, and capacity to under-
Engineering, Istanbul Technical University, 34000 Istanbul, Turkey,
e-mail: baranh16@itu.edu.tr
take challenging tasks, notably critical location monitoring.
Ismail Bayezit: Department of Aeronautical Engineering, Istanbul Nonetheless, their diminutive size, limited strength, and low
Technical University, 34000 Istanbul, Turkey weight render them susceptible to diverse external and
Open Access. © 2024 the author(s), published by De Gruyter. This work is licensed under the Creative Commons Attribution 4.0 International License.
2  Haci Baran and Ismail Bayezit
internal factors such as faults, disturbances, and sensor mal- nonlinear robust controller for a non-minimum phase
functions. These factors can adversely affect the flight per- model of an air-breathing hypersonic vehicle is presented
formance, stability, and overall quality of UAVs. Designing a by Fiorentini and Serrani [11]. A study of nonlinear robust
flight control system for small UAVs constitutes a complex adaptive control of flexible air-breathing hypersonic vehi-
endeavor requiring the mitigation of external disturbances cles is reported by Fiorentini et al. [12]. A robust nonlinear
and consideration of the impact of unknown factors. Given control method for a hypersonic aircraft is presented by
that small UAVs can operate for extended durations without Wang and Stengel [13]. Seferian and Bazzi [14] proposed a
human pilots, the development of disturbance rejection con- method based on extremum-seeking control (ESC) for the
trol systems is imperative to ensure safe and reliable opera- detection of four major types of faults that occur in induc-
tions, particularly crucial during winter when rapid climatic tion motors. An integrated approach to the fault-tolerant
changes can pose challenges during prolonged flight mis- control (FTC) of a quadcopter UAV with incipient actuator
sions. Numerous studies have been conducted to investigate faults is presented by Kantue and Pedro [15]. Vandermeulen
the influence of external disturbances on the flight perfor- et al. [16] proposed a discrete-time version of ESC which can
mance of miniature UAVs. By comprehensively analyzing be implemented by several agents communicating over a
the effects of various factors, including disturbances, the network. In the study by Sahneh et al. [17], an extremum
design and implementation of robust control systems can seeking algorithm is proposed for mappings where the
augment the reliability and effectiveness of small UAVs optimal point is time-varying. Oliveira et al. addressed the design
across diverse operational scenarios. Shen and Xu [1], to and analysis of multi-variable extremum seeking for static maps
enable the unmanned helicopter to fly autonomously in pre- subject to arbitrarily long time delays [18]. Xie et al. considered
cise paths and reduce the influence of internal and external the problem of designing nonlinear robust formation controllers
unknown disturbances of the unmanned helicopter, proposed on a team of UAV using off-the-shelf autopilots [19]. Binetti et al.
the adaptive radial basis function neural network-based active presented a comprehensive design procedure based on
disturbance rejection controller (ADRC). In the study by extremum seeking for minimum power demand formation
Labbadi and Cherkaoui [2], to achieve fast speed and high flight, the first with performance guarantees [20]. Lavretsky
accuracy performances for the quadrotor UAV, an adaptive et al. focused on an adaptive output-tracking problem using
fractional-order nonsingular fast terminal sliding mode on-line extremum seeking command generation [21]. An ESC
controller (SMC) is proposed. Liu et al. [3] focused on path‐ algorithm was defined for discrete-time systems which imple-
following control for small fixed‐wing UAVs under wind mented to a group of plants [22]. A type 2 fuzzy logic PID con-
disturbances. Lai and Le [4] presented the development of troller is proposed for twin rotor mimo system control problem
an autonomous flight control system for a small-scale in the study by Zeghlache et al. [23]. Krause reported on the
unmanned helicopter based on an online adaptive learning- latest results concerning the active noise control approach using
based observer and model predictive control (MPC). He et al. net flow of acoustic energy [24]. Gheni et al. explored the use of
[5] proposed a sliding mode-based nonlinear control scheme DRL for controlling vibrations in building structures. Specifically,
for the hovering flight of a tilt tri-rotor UAV, consisting of they focused on the problem of reducing vibrations induced by
position control, attitude control, and control allocation. In external sources such as wind or earthquakes [25].
the study by Lanzon et al. [6], the problem of designing a Disturbance rejection control in UAVs stands as a fore-
control law in case of rotor failure in quadrotor vehicles is front research area essential for enhancing UAV autonomy,
addressed. In the study by Kumar et al. [7], the conceptual resilience, and mission effectiveness. UAVs operate in diverse
design and flight controller of a novel kind of quadcopter are and dynamic environments, encountering disturbances such
presented. This design is capable of morphing the shape of the as gusts, wind shear, and atmospheric turbulence that can
UAV during flight to achieve position and attitude control. significantly affect their stability and performance. Current
Shin et al. [8] presented nonlinear disturbance observer-based research focuses on developing sophisticated control algo-
standoff tracking guidance for multiple small fixed-wing UAVs rithms and adaptive strategies to effectively counteract dis-
in the presence of wind. Furthermore, Eliker and Zhang [9] turbances and maintain desired flight trajectories. In recent
presented a reliable and novel quadrotor flight control years, significant strides have been made in the utilization of
system designed to enhance trajectory tracking perfor- robust control techniques such as adaptive control, predictive
mance, robustness, and adaptiveness against the uncertain control, H-infinity control, SMC, and active disturbance rejec-
parameters and the external wind disturbance. Wu and tion control, offering disturbance rejection capabilities.
Mueller [10] proposed a method for finding the optimal Moreover, the integration of machine learning and data-driven
speed and heading of a multicopter when flying a given methods has emerged as a promising frontier, enabling the
path to achieve the longest flight range. The design of a development of adaptive controllers capable of learning
Advanced autopilot design with extremum-seeking control  3
from real-time data streams to adaptively reject disturbances. autopilots. This newly designed autopilots show superiority in
The following journal descriptions mention some studies of last terms of disturbance rejection compared with other previous
3 years about disturbance rejection of UAVs. controllers that have disturbance rejection properties.
The authors present an integral backstepping active This study presents a comprehensive design approach
disturbance rejection control strategy for quadrotor UAV for an advanced autopilot system. The autopilot system
attitude control, featuring a novel smooth function newfal employed in this research comprises two distinct control
(∙), an improved expanded state observer, and recursion- components, each fulfilling a specific function. The first
based nonlinear backstepping integral control, resulting in component aims to ensure system stability and control
enhanced tracking precision, dynamic response, and anti- by effectively regulating the system’s behavior and main-
interference capabilities [26]. The article explores the applica- taining the desired performance criteria. In contrast, the
tion of ADRC in enhancing trajectory tracking and disturbance second component is dedicated to adapting the system para-
rejection for multirotor UAVs, highlighting its robustness meters in response to external disturbances that may affect
against parameter variations while maintaining performance the system’s performance. This adaptive parameter adjust-
integrity without the need for constant control gains tuning ment mechanism enables the autopilot to promptly identify
[27]. The authors present a novel mass adaptive control and mitigate the effects of such disturbances, thereby ensuring
method integrating robust SMC and linear active disturbance that the system response remains within acceptable bounds
rejection control (LADRC) for quadrotor-loaded UAVs with and restoring the desired performance characteristics.
varying mass, addressing centroid position changes for The novelty of the proposed method lies in its integra-
improved model accuracy, enhancing robustness through tion of the ESC algorithm with conventional autopilots to
SMC to counteract disturbances and compensate for control provide them advanced disturbance rejection properties.
precision limitations, while incorporating adaptive control Unlike traditional autopilots, which lack adaptive, distur-
within LADRC to adjust parameters in real-time, validated bance rejection capabilities, this approach utilizes ESC’s dynamic
through simulation comparisons demonstrating superior parameter adjustment to effectively mitigate unknown inputs
performance in trajectory tracking and disturbance rejec- and restore the system to its original controlled response,
tion [28]. Wang and Zhao presented a novel LADRC strategy thereby enhancing resilience against external threats. By incor-
combined with adaptive control to effectively address atti- porating ESC features into autopilot systems such as pitch atti-
tude control challenges in a quadrotor UAV, enhancing its tude hold autopilot, altitude hold autopilot, and yaw autopilot,
ability to counteract external disturbances while simplifying the resulting integration yields innovative disturbance rejection
parameter tuning and ensuring stability, as validated through controllers capable of eliminating severe disturbances, such as
simulation results [29]. Azid et al. presented an innovative ramp, sinusoidal, and step disturbances. Notably, this integration
approach utilizing an unknown input-state observer employing offers superior disturbance rejection compared to previous
the Lipschitz method to estimate internal states and wind dis- methods, as demonstrated by simulation results, thereby
turbances in a UAV quadrotor, enabling compensation through enhancing aircraft operation under disturbances while
rotor velocity adjustments and achieving robustness and accu- minimizing energy consumption and reducing operator
racy in position and attitude control, validated via simulation in workload. For instance, the Linear Quadratic Gaussian Regulator
Matlab/Simulink and practical implementation on the Parrot (LQG) and MPC fail to mitigate harsh disturbances, whereas our
Mambo mini quadrotor [30]. The study explores enhancing proposed method successfully rejects these disturbances, as evi-
the performance of small fixed-wing UAVs by segmenting denced by the simulation results.
aileron control surfaces and deploying multiple MPCs, validated This work follows a systematic structure. Section 1
through wind tunnel experiments demonstrating significant begins with an introductory section that provides an over-
improvements in roll attitude control and disturbance rejection view of the research topic and its significance in the field.
under turbulent flight conditions [31]. The study proposes a Section 2 is dedicated to clearly defining the problem state-
robust controller for quadrotor UAVs that effectively decouples ment, establishing specific challenges, and outlining the
position and attitude control, utilizing MPC for position control objectives addressed in the study. The mathematical formu-
with motor constraints, and a nonlinear robust control law for lation of the nonlinear equations governing the dynamics of
attitude control in SO (3) space, showcasing superior accuracy the aircraft UAV is presented in Section 3, laying the ground-
in tracking aggressive trajectories amidst severe external dis- work for further analysis. Progressing to Section 3.3, the
turbances through simulation validation [32]. In these cited arti- autopilot structure is detailed, offering insights into the
cles, the recent approaches about disturbance rejection of UAVs design principles and control strategies applied. Section 5
are described. In our proposed method, we provide adaptive, presents simulations conducted to validate and assess the
advanced disturbance rejection properties for the conventional performance of the proposed autopilot system. Finally,
Section 6 engages in the discussion of findings and results, the velocity components that govern its motion. The figure
drawing conclusions based on the analysis conducted also includes an illustration emphasizing the potential impact
throughout the study in Section 7. of disturbances, which have the capacity to adversely affect
the overall performance and operation of the aircraft system.
2 System model
2.2 Nonlinear system dynamics
2.1 Problem statement
This section offers a synopsis of the nonlinear and linear
equations governing the dynamics of small UAVs. Due to
Managing small UAVs during extended and high-risk mis-
their substantial number, a detailed analysis of each equa-
sions can present challenges, given the potential hazards of
tion is not undertaken here. However, for readers seeking a
collisions and severe disturbances. Hence, it is imperative
deeper understanding, reference to Chapter 3 of the study
to develop control systems that exhibit robustness and
by Beard and McLain [33] is recommended, as it extensively
reliability. In this study, an autopilot system has been
explores these dynamics. Furthermore, Chapter 4 of their
developed to adeptly regulate the dynamics of the UAV
study [33] specifically focuses on equations related to forces
and maintain stability. However, recognizing the challen-
and moments, while Chapter 5 delves into the intricacies
ging environmental conditions, additional measures may
of the trim and linearization process, providing valuable
be necessary to optimize the system’s performance. To
insights into the dynamics of small UAVs.
address these challenges, an ESC approach has been imple-
mented, aiming to mitigate the risks associated with crashes
and adverse weather. This adaptive autopilot system can
2.2.1 Forces and moments equations
adjust to inputs and varying system parameters. Through
the implementation of a cascaded control system, the pri-
The equations for forces and moments can be succinctly
mary objective is to ensure the stability and precise control
summarized in the following matrix system. To repre-
of the UAV throughout its operations.
sent the entire equation in a single horizontal line, the
Figure 1 illustrates the structural features of a fixed-
force equations in [33] are expressed using the following
wing UAV model, providing a visual representation of its
equality:
key characteristics. Additionally, it graphically depicts the
external forces and moments affecting the UAV, along with ⎡ − mg sin θ ⎤
A1 = ⎢ mg cos θ sin ϕ ⎥,
⎢⎣ mg cos θ cos ϕ ⎥⎦
⎡ c ⎤
CX (a ) + CX (a ) q
⎢ 2Va ⎥
1 2 ⎢ b b ⎥
B1 = ρVa S ⎢CY0 + CYββ + CYp p + CYr r ,
2 2Va 2Va ⎥
⎢ c ⎥
⎢⎣ CZ (a ) + CZq (a ) q ⎥⎦
2Va
⎡ C Xδe (a )δe ⎤ ⎡Tp(δtVa ) ⎤

1 2 ⎢
C1 = ρVa S CYδaδa + CYδrδr ⎥ + ⎢ 0 ⎥,
2 ⎢ ⎥ ⎢ ⎥
⎣ CZδe (a )δe ⎦ ⎣ 0 ⎦
⎡ fx ⎤
⎢ fy⎥ = A + B + C , (1)
1 1 1
⎢ ⎥
f
⎣ z⎦
where A1 , B1 , and C1 are chosen symbols to denote a part

of the longitudinal aerodynamic force matrix system of the
Figure 1: Fixed-wing UAV model exposed to disturbance. UAV to be able to write this matrix system in a single line.
CX (a ) ≜ − CD(a ) sin a + CL(a ) sin a , A2 andB2 are chosen symbols to denote a part of the long
C Xq ≜ − CDq cos a + CLq sin a , aerodynamic moment matrix system of the UAV to be able
to write this matrix system in a single line. Cl is the rolling
C Xδe ≜ − CDδe cos a + CLδe sin a , moment coefficient, Cl0 is the Cl at zero sideslip angle, Clβ is
CZ (a ) ≜ − CD(a ) sin a − CL(a ) cos a , the Cl rate with respect to sideslip angle, Clp is the Cl rate
with respect to roll rate, Clr is the Cl rate with respect to yaw
CZq (a ) ≜ − CDq sin a − CLq cos a , rate, b is the wing span, c is the mean aerodynamic chord,
CZδe (a ) ≜ − CDδe sin a − CLδe cos a , S is the wing area, Cm is the pitching moment coefficient,
Cm0 is the Cm at zero angle of attack, Cmδe is the Cm rate with
CL(a ) = Cl0 + Claa, respect to elevator deflection, Cma is the Cm rate at an angle
CD(a ) = CD0 + CDaa, of attack, Cmq is the Cm rate with respect to pitch rate, Cn is
the yaw moment coefficient, Cn0 is the Cn at zero sideslip
where ρ is the air density, Va is the airspeed, S is reference, angle, Cnβ is the Cn rate with respect to sideslip angle, Cnp is
Cl is the roll moment coefficient, Cl0 is the lift coefficient at 0 the Cn rate with respect to roll rate, Cnr is the Cn rate with
angle of attack, Cla is the lift coefficient at a angle of attack, respect to yaw rate, Clδa is the Cl rate with respect to aileron
p is the roll rate, q is the pitch rate, r is yaw rate, β is the deflection, Clδr is the Cl rate with respect to rudder deflec-
sideslip angle, a is the angle of attack, δe is the elevator tion, Qp is the propeller thrust, Cnδa is the Cn rate with
deflection, δa is the aileron deflection, δr is the rudder respect to aileron deflection, Cr δr is the Cn rate with respect
deflection, δt is the throttle deflection, Tp is the propeller to rudder deflection.
thrust, CX , CY , and CZ are the aerodynamic force coefficients
in x , y , and z directions, respectively. CYβ is the CY rate with
respect to the sideslip angle. CY0 is the lateral force coeffi- 2.2.2 Equations of motion
cient when β = p = r = δa = δr = 0, CYr is the CY rate with
respect to yaw rate, CYp is the CY rate with respect to roll The matrix system presented below provides a concise
rate, CZq is the CY rate with respect to pitch rate, C Xδe is the representation of the aircraft’s equations of motion, incor-
force coefficient in x direction with respect to elevator porating a total of 12 aircraft states, as evident from the
deflection, CYδa is the force coefficient in y direction with formulation.
respect to aileron deflection, CZδe is the force coefficient in z
direction with respect to elevator deflection, fx , f y , and fz ⎡ cθcψsϕsθcψ − cϕsψ cϕsθcψ − sϕsψ ⎤
are the aerodynamic forces in x, y, and z directions, A3 = ⎢ cθsψsϕsθsψ − cϕcψ cϕsθsψ − sϕcψ ⎥
⎢⎣ − sθsϕcθ cϕcθ ⎥⎦
respectively.
To express the moment equations in a concise and
⎡ ṗ n ⎤
single horizontal line, the following equality is utilized: ⎡u ⎤
⎢ ṗ e ⎥ = A3 v , (3)
⎢⎣ w ⎥⎦
⎢ ṗ ⎥
⎡ ⎡ b b ⎤ ⎤ ⎣ d⎦
⎢ b⎢⎣Cl0 + Clββ + Clp 2Va p + Clr 2Va r ⎥⎦ ⎥
⎢ ⎥ ⎡ fx ⎤
1 ⎡ c ⎤ ⎡ u̇ ⎤ ⎡ rv − qw ⎤ 1⎢ ⎥
A2 = ρVa2S ⎢ c⎢Cm0 + Cmaa + Cmq q⎥ ⎥, ⎢ v̇ ⎥ = ⎢ pw − ru ⎥+ fy , (4)
2 ⎣ 2Va ⎦
⎢ ⎥ ⎣ ẇ ⎦ ⎣ qu − pv ⎦ m ⎢ f ⎥
⎢ ⎡ ⎣ z⎦
b b ⎤⎥
⎢⎣ b⎣⎢Cn0 + Cnββ + Cnp 2Va p + Cnr 2Va r ⎦⎥ ⎥⎦ 1 sin ϕ tan θ cos ϕ tan θ ⎤
⎡ ϕ̇ ⎤ ⎡
⎢ θ̇ ⎥ = ⎢ 0 cos ϕ − sin ϕ ⎥⎡ u ⎤
⎡ b[Clδaδa + Clδrδr ] ⎤ ⎡Qp(δtVa ) ⎤ v , (5)
1 2 ⎢ ⎥+⎢ ⎢ ̇⎥ ⎢ sin ϕ cos ϕ ⎥⎢⎣ w ⎥⎦
B2 = ρVa S c[Cmδeδe] 0 ⎥, ⎣ψ ⎦ ⎢⎣0 ⎥
2 ⎢ ⎥ ⎣⎢ cos θ cos θ ⎦
⎣ b[Cnδaδa + Cnδrδr ]⎦ 0 ⎥⎦
⎡ Γ3l + Γ4n ⎤
⎡l⎤ ⎡ ṗ ⎤ ⎡ Γ1 pq − Γ2qr ⎤ ⎢ 1 ⎥
(2) ⎢ q̇ ⎥ = ⎢ Γ5 pr − Γ6( p2 − r 2 ) ⎥ + ⎢ m , (6)
⎢ m ⎥ = A2 + B2 , Jy ⎥
⎣n⎦ ⎢⎣ ṙ ⎥⎦ ⎢⎣ Γ7 pq − Γ1qr ⎥⎦ ⎢ ⎥
⎣ Γ4l + Γ8n ⎦
where the variables l , m , and n represent the external
moments acting about the x, y, and z axes, respectively. where
Ixz(Iz − I y + Iz) Iz(Iz − I y ) + Ixz 2 behavior, enabling control engineers to develop feedback
Γ1 = , Γ2 = ,
Γ Γ controllers that stabilize and regulate the aircraft’s motion.
Iz Ixz Iz − Ix Ixz By integrating appropriate control strategies, the UAV can
Γ3 = , Γ = , Γ5 = , Γ6 = , achieve desired flight trajectories and effectively perform
Γ 4 Γ Iy Iy
various tasks, such as aerial surveillance, mapping, pay-
(Ix − I y )Ix + Ixz 2 Ix load delivery, and environmental monitoring. As the UAV
Γ7 = , Γ8 = , Γ = Ix Iz − Ixz 2 .
Γ Γ industry advances, the development of sophisticated con-
In this specific context, the symbols Γ are utilized to trol algorithms and navigation systems will play a pivotal
represent functions associated with moments and products role in enhancing the safety, efficiency, and versatility of
of inertia, whereas the symbol I is used to denote the UAVs across a wide range of applications.
moment of inertia and m denotes the UAV mass.
pn , pe , pd are inertial positions in north, east, and down
directions, respectively. u, v , w are velocities in x-axis,
y-axis, and z-axis and l , m , n are moments in x-axis, 2.2.3.1 Trim conditions
y-axis, and z-axis respectively. Ix , I y , Iz are moments of The illustration of nonlinear systems is depicted in the
inertia in x-axis, y-axis, and z-axis, respectively. ϕ , θ , ψ following manner:
are roll, pitch, and yaw angles, respectively. ẋ = f (ẋ , u̇ ) . (7)
The variables ẋ and u̇ represent the first derivatives

2.2.3 Model linearization with respect to time of the state variables and control
inputs, respectively. These derivatives offer insights into
The equations governing an aircraft’s motion in six degrees the rates of change and dynamics of the system.
of freedom inherently display nonlinearity, introducing The equilibrium condition is expressed as follows: This
complex dynamics that can pose challenges in control condition delineates the stable state wherein the system’s
system design. Nevertheless, it is a common practice to state variables and inputs persist without change over
linearize the nonlinear model of fixed-wing UAVs for the time, thereby indicating a state of balance or stability.
ease of designing and analyzing control systems. This lin- f (x *, u*) = 0. (8)
earization procedure transforms the system into a linear
model, simplifying control design and rendering it more In this context, the variables x* and u* represent the
amenable to classical control techniques. trim states and trim inputs of the UAV, respectively. These
The dynamics of a fixed-wing UAV can be categorized trim values correspond to stable operating points where
into two primary modes: the longitudinal mode and the the UAV maintains steady flight conditions with balanced
lateral mode. The longitudinal mode focuses on variables forces and moments.
such as airspeed, pitch angle (theta), and altitude, which Under the trim condition, the UAV achieves a state of
are crucial for controlling the UAV’s forward motion and equilibrium with balanced forces and moments. And the
climb/descent. Conversely, the lateral mode involves roll subsequent relation is fulfilled, indicating the successful
angle (phi) and yaw angle (psi), determining the UAV’s attainment of stable flight conditions.
lateral movements and heading changes. In the field of ẋ * = f (x *, u*) . (9)
flight dynamics, the concept of trim plays a significant
role in UAV control. Trim entails achieving a state of force In the calculation process to determine trim states, it is
and moment equilibrium, where the sum of all forces and imperative to satisfy and uphold the prescribed set of con-
moments acting on the aircraft is balanced, resulting in a ditions. These conditions are crucial for ensuring the sta-
stable flight path without constant pilot input. This trim bility and equilibrium of the analyzed system.
condition is essential for various flight phases, including • Constant Speed Va*,
straight-and-level flight, climbing, and turning. By attaining • Climbing at constant flight path angle γ*,
trim, the UAV can maintain a desired flight attitude and path, • Constant orbit of radius R*.
thereby enhancing overall flight stability and efficiency.
In the process of designing control systems, it is crucial The positions, velocities, Euler angles, and roll, pitch,
to comprehend and consider both the longitudinal and and yaw rates are individually presented according to the
lateral modes, as well as the concept of trim. The linearized study by Beard and McLain [33]. This representation elu-
model provides a simplified representation of the UAV’s cidates the specific values corresponding to each of the
mentioned state variables, facilitating a comprehensive ⎡ Xδe Xδe ⎤

understanding of the system’s dynamics. ⎢ Zδe 0 ⎥
x ≜ ( pn , pe , pd , u, v , w , ϕ , θ , ψ , p , q , r )T . (10) B4 = ⎢ Mδe 0 ⎥,
⎢ 0 ⎥
0
The control surfaces, namely, elevator, throttle, aileron, ⎢ 0 ⎥⎦
⎣ 0
and rudder deflections, specific to the fixed-wing UAV, are
individually depicted as provided below. These representa- ⎡ u̅̇ ⎤ ⎡ u̅ ⎤
tions offer a comprehensive view of the orientation and ⎢ w̅̇ ⎥ ⎢ w̅ ⎥
⎢ q̅̇ ⎥ ⎡ δ̅ e ⎤
configuration of each control surface, which is critical for = A4 ⎢ q̅ ⎥ + B4⎢ ⎥. (16)
⎢ ⎥ ⎢ θ̅ ⎥ ⎣ δ̅ t ⎦
understanding the UAV’s control system. ̇
⎢ θ̅ ⎥ ⎢⎣ h̅ ⎥⎦
u ≜ (δe , δt , δa , δr )T . (11) ⎣ h̅̇ ⎦
As a consequence of the trim conditions, the following The lateral state space model, as presented by Beard
equalities, as referenced in the study by Beard and McLain and McLain [33], is illustrated below. This model provides a
[33], are utilized. These equalities play a significant role in detailed mathematical depiction of the UAV’s lateral dynamics,
establishing and maintaining the stable and balanced state enabling a comprehensive examination of its motion and sta-
of the UAV during operation. bility characteristics in the lateral plane.
u̇ * = v̇ * = ẇ * = 0, (12) ⎡ Yv Yp Yr − gc θ * ⎤
0
⎢L Lp Lr − gs θ * 0⎥
ϕ̇ * = θ̇ * = ṗ * = q̇ * = 0. (13) v
⎢ Nr 0 ⎥
A5 = Nv Np 0 ,
The turn rate constant, as referenced in the study by ⎢ c ϕ*tθ * q*c ϕ*tθ * − r *s ϕ*tθ * 0⎥
0 1
Beard and McLain [33], is explicitly presented below. This ⎢ 0⎥
p*c ϕ*secθ * p*c ϕ*secθ * − r *s ϕ*secθ *
⎣ 0 0 ⎦
critical constant determines the UAV’s ability to perform
turning maneuvers and plays a pivotal role in the analysis ⎡ Yδa Yδr ⎤
of its turning dynamics. ⎢ Lδa L δr ⎥
Va* B5 = ⎢ Nδa Nδr ⎥,
ϕ̇ * = cos γ* → ṙ * = 0. (14) ⎢ ⎥
R* 0 0
⎢ ⎥
⎣ 0 0 ⎦
As indicated by Beard and McLain [33], the climb rate
constant is denoted below. This essential constant governs ⎡ v̅̇ ⎤ ⎡ v̅ ⎤
the rate of vertical ascent or descent of the UAV, playing a ⎢ p̅̇ ⎥ ⎢ p̅ ⎥
pivotal role in assessing its vertical flight characteristics. ⎢ ̇⎥ ⎡ δ̅ a ⎤
r̅ ⎢ r̅ ⎥ (17)
⎢ ⎥ = A5 ⎢ ⎥ + B5⎢⎣ δ̅ r ⎥⎦,
ḣ * = Va*cos γ*. (15) ̇ ϕ̅
⎢ ϕ̅ ⎥ ⎢ ψ̅ ⎥
As mentioned by Beard and McLain [33], the state vari- ⎢⎣ ψ̅̇ ⎥⎦ ⎣ ⎦
able ẋ * can be determined based on the given parameters
Va*, γ*, and R*. These specified parameters are essential in where
computing the time derivative of x at its trim condition. sϕ = sin ϕ , sθ = sin θ , sψ = sin ψ , cϕ = cos ϕ ,
As indicated by Beard and McLain [33], the longitu-
cθ = cos θ , cψ = cos ψ , scθ = sec θ , tθ = tan θ.
dinal state space model is presented below. This model
provides a detailed mathematical representation of the
UAV’s longitudinal dynamics, serving as a valuable tool
for analyzing its motion and stability characteristics in
the longitudinal plane. 3 ESC
⎡ Xu Xw Xq − g sin θ * 0⎤ ESC is an adaptive control algorithm that has been pre-
⎢ Zu Zw Zq − g cos θ * 0⎥ viously discussed in the literature. This method proves
A4 = ⎢ Mu Mw Mq 0 0 ⎥, particularly advantageous when confronted with scenarios
⎢ 0⎥
0 0 1 0 that lack precise knowledge of system dynamics and the
⎢ 0⎥
⎣ sin θ * cos θ * 0 u*sin θ * + w*sin θ * ⎦ mappings between control parameters and an objective
Figure 2: ESC block diagram.
function. The versatility of extremum seeking extends ability to efficiently search for the optimal operating point
beyond a single application area, as it can effectively opti- without necessitating extensive system modeling or exhaus-
mize parameters in dynamic systems and tackle static optimi- tive exploration of the parameter space. Instead, it relies on
zation problems alike. The core strength of ESC lies in its local measurements and feedback to iteratively update con-
adaptive nature, enabling autonomous convergence towards trol parameters, effectively seeking the extremum (max-
an optimal solution through iterative adjustments of control imum or minimum) of a given objective function.
parameters based on real-time measurements and feedback. In summary, ESC emerges as a powerful and versatile
This inherent adaptability renders it well-suited for sys- technique adept at adapting to diverse applications and
tems with time-varying dynamics, where traditional con- problem domains. Its capability to optimize control para-
trol approaches may prove insufficient or impractical. meters in dynamic systems renders it a valuable tool for
Moreover, ESC has found applications in various domains, enhancing performance, efficiency, and stability across a
spanning from aerospace and robotics to power systems and broad spectrum of engineering and scientific disciplines.
industrial processes. In aerospace, for instance, it optimizes The process of tuning a parameter value using the
aircraft control parameters, enhancing performance and extremum seeking algorithm comprises four distinct stages.
fuel efficiency. In robotics, ESC fine-tunes robotic manipu- These stages constitute a systematic approach to iteratively
lator parameters, improving task performance and preci-
sion. Additionally, it optimizes the operation of distributed
energy resources and stabilizes grid dynamics in power sys-
tems. The effectiveness of extremum seeking stems from its
Figure 3: The application of extremum seeking aiming to optimize a Figure 4: The utilization of ESC to optimize a particular section of the
rising segment of the curve representing the objective function. objective function curve.
optimize the parameter and attain the desired extremum multiplication by a sinusoidal signal with the same fre-
(maximum or minimum) of the objective function. quency as that of the modulation signal. The aim of this
Stage #1. Within the extremum seeking algorithm, a step is to leverage sinusoidal modulation and extract valu-
specific stage, referred to as modulation, involves the able information about the system’s response to perturba-
application of a low-amplitude sinusoidal signal to perturb tions. To alleviate potential bias in the objective function
the parameter value under optimization. This modulation signal, a high-pass filter can be optionally employed, thereby
process aids in gathering information about the system’s enhancing the accuracy of the optimization process.
response to perturbations, facilitating the determination of Stage #4. Parameter Update – During the updating pro-
the optimal parameter value corresponding to the desired cess, the parameter value undergoes adjustment through an
extremum of the objective function. integrator that receives the demodulated signal. The state of
Stage #2. System Response – As the optimized system the integrator can be toggled between on and off, thereby
encounters perturbations in the parameter, it responds accord- influencing the rate of parameter adaptation. In some cases,
ingly, leading to a corresponding alteration in the value of the a low-pass filter is employed to counteract high-frequency
objective function. This interaction between the perturbations noise present in the demodulated signal, enhancing the
and the objective function offers valuable feedback to the accuracy of the parameter tuning.
extremum seeking algorithm, aiding it in refining the para-
meter value towards the optimal extremum.
Stage #3. Demodulation – In this stage of the extremum 3.1 Optimization of static systems
seeking algorithm, the objective function signal undergoes
Figure 2 illustrates the application of extremum seeking for
addressing a static optimization problem. This involves opti-
mizing certain parameters within the system to achieve the
desired extremum (maximum or minimum) of the objective
function, thus demonstrating the effectiveness of the extremum
seeking technique in such scenarios.
In the subsequent sections, the ESC problem under
consideration will be formally defined, based on the fol-
lowing set of assumptions. These assumptions lay the foun-
dation for the subsequent analysis and investigation of the
problem at hand.
A.1. In this context, the estimation of the parameter
value is denoted as θ̂ , while the signal utilized for modula-
tion is represented as θ . These representations facilitate a
clear distinction between the estimated parameter and the
modulating signal in the extremum seeking algorithm.
A.2. The output of the function being maximized,
referred to as the objective function, is represented by
Figure 5: Utilizing ESC on a level section of the objective function curve.
Figure 6: Sinusoidal perturbation based analog optimization.

the equation y = J (θ ) . This expression delineates the rela- modulated signal to the objective function J (θ ), we obtain
tionship between the output ( y ) of the objective function a modified objective function with a phase identical to that
and the parameter value (θ ) under consideration within of the modulation signal. Subsequently, the modified objec-
the context of extremum seeking problem. tive function is multiplied by the demodulation signal,
A.3. The frequency at which both the modulation and resulting in a positive signal. Through integration, this posi-
demodulation signals are applied is denoted as ω. This tive signal influences the value of θ, causing it to increment
frequency serves as a crucial parameter governing the and approach the maximum point in the objective function.
rate of perturbation and information extraction during This iterative process continues, guiding the extremum seeking
the extremum seeking process. algorithm towards the optimal parameter value θ*, which cor-
A.4. The modulating signal is denoted as b sin(ωt ) , responds to the maximum of the objective function. The con-
while the demodulating signal is represented as a sin(ωt ). cept of modulating and demodulating signals, coupled with the
These sinusoidal signals play a vital role in the extremum iterative adaptation of parameters, underpins the efficacy of
seeking algorithm, facilitating perturbation and informa- extremum seeking in converging towards optimal solutions in
tion extraction during the optimization process. scenarios characterized by incomplete knowledge of system
A.5. The learning rate of the autopilot is symbolized as dynamics or mappings. This renders extremum seeking a valu-
k . This parameter governs the rate at which the autopilot able tool for addressing a broad spectrum of optimization
adapts and adjusts its control actions based on real-time problems across engineering and scientific disciplines.
feedback, thereby influencing the convergence and stabi- In Figure 4, we present the practical implementation of
lity of the control system. extremum seeking, demonstrating its utility in optimizing
A.6. The optimal parameter value, denoted as θ*, cor- a decreasing segment of the objective function curve. By
responds to the highest value of the objective function employing the modulation signal alongside the current
J (θ ), indicating the parameter setting that yields the best parameter estimate, we derive the modulated signal θ . Sub-
performance or solution within the optimization context. sequently, the objective function J (θ ) is employed to gen-
In cases where multiple parameters require optimization, erate a modified objective function, which, in this instance,
employing a distinct tuning component for each parameter is exhibits a phase shift of 180° with respect to the modulation
essential. This approach ensures that each parameter is inde- signal. Multiplying the modified objective function by the
pendently adjusted and fine-tuned to reach its optimal value demodulation signal yields a signal in the negative direction.
within the system, thereby contributing to a comprehensive Through integration, this negative signal influences the
and efficient optimization process. value of θ , facilitating its decrease and approach towards
In Figure 3, we depict the practical implementation of the maximum point in the objective function. The iterative
extremum seeking, emphasizing a rising segment of the process persists, guided by the modulation and demodula-
objective function curve. The modulation signal, in con- tion signals, propelling the extremum seeking algorithm
junction with the current parameter estimate, yields the towards the optimal parameter value θ*, corresponding to
modulated signal denoted as θ. Upon application of this the maximum of the objective function within this specific
Figure 7: Perturbation based ESC.

section. This mechanism underscores the adaptability and for ESC. Renowned for its efficacy, this method facilitates
versatility of extremum seeking in addressing both increasing rapid adaptation and straightforward implementation, par-
and decreasing objective function segments, thus establishing ticularly when addressing gradient-type problems. The pri-
it as a robust and effective approach for parameter optimiza- mary foundation for this discourse can be traced back to
tion in diverse real-world scenarios. Chapter 3 of the study by Zhang and Ordóñez [34], which
In Figure 5, we illustrate the practical application of serves as a fundamental resource for exploring and com-
extremum seeking on a plateau of the objective function prehending the intricacies of the sinusoidal perturbation
curve, representing a vicinity near the maximum. In this technique. The application of this method assumes a pivotal
scenario, employing J (θ ) results in a perturbed objective role in numerous fields, including aerospace, robotics, and
function that approaches zero, indicating the system’s power systems, where optimization and control are of para-
close proximity to the optimal parameter value, θ*. Conse- mount importance. By harnessing continuous sinusoidal
quently, when this perturbed function is multiplied by the modulation and sophisticated algorithms, ESC with sinu-
demodulation signal and integrated, its impact on the soidal perturbation presents a versatile and robust tool for
value of θ is negligible. This observation confirms that θ addressing complex optimization challenges.
is already near its optimal point, θ*, and further adjust- Consider the optimization process of a performance
ments are minimal. The extremum seeking algorithm effi- function represented as y = J (θ ) . In Figure 6, we encounter
ciently identifies the vicinity of the optimal solution and a fundamental continuous optimization method based on
converges towards θ* with reduced adaptation, conserving sinusoidal perturbation, where the multiplication symbol
computational resources and ensuring stability around the ⊗ is utilized to denote the operation of multiplication. By
desired maximum of the objective function. This cap- introducing the perturbation signal a sin(ωt ) into the func-
ability to recognize and adapt to regions near the tion, valuable gradient information about J (θ ) can be gleaned,
optimal solution underscores extremum seeking’s aiding in the determination of the optimal parameter
robustness and reliability for parameter optimization value. The insights provided by Zhang and Ordóñez [34]
in practical applications. further underscore the essential characteristics of the per-
turbation-based extremum seeking approach, as depicted in
Figure 6. These key traits encompass the ability to iteratively
3.2 Perturbation based ESC perturb the parameter value and utilize the resulting gra-
dient information to efficiently converge towards the desired
In this section, we examine the sinusoidal perturbation extremum of the objective function. The sinusoidal pertur-
method, which stands as the most widely utilized approach bation-based extremum seeking technique serves as a valu-
able tool for optimizing various performance functions in
diverse fields of engineering, science, and technology. Its
adaptability and effectiveness render it well-suited for
real-world optimization challenges, thereby contributing
to enhanced system performance and efficiency across a
wide array of applications.
For the extremum-seeking problem, the following assump-
tions are posited. These assumptions function as foundational
premises for the subsequent analysis and investigation of the
optimization process.
Assumption 1. There exists a smooth function

l : → n
such that
f (x , a (x , θ )) = 0, if and only if x = l (0). (18)
Assumption 2. There exists θ* ∈ such that

Figure 8: Flowchart for Autopilot-ESC integration control under
(Jol)′ (θ *) = 0, (19)
disturbances.
Figure 9: The improved pitch attitude hold autopilot block diagram.
(Jol)″ (θ *) < 0. (20) with a period of 2π /ω. This particular solution fulfills the
specified conditions and elucidates the stability and dynamic
Hence, we deem the output equilibrium map as a fun-
behavior of the system.
damental assumption for this study. This assumption forms
the basis for the subsequent analysis:
∼ 2π 2π a2J ″ 1
y = J (l (θ )) (21) θ ω (t ) + θ ω (t ) − ≤ O⎛ ⎞ , ∀t ≥ 0. (22)
4 ⎝ω⎠
has a maximum at θ = θ*.
The theorem presented in this context establishes a
Theorem 1. Provided the appropriate system size, Figure 6 property of local convergence in continuous perturbation-
illustrates a singular exponentially stable periodic solution based optimization of single-parameter functions. This
Figure 10: The improved altitude hold autopilot block diagram.

Figure 11: The improved yaw autopilot block diagram.
Figure 12: Theta angle control under white noise disturbances.

Figure 13: Theta angle control under sinusoidal and white noise disturbances.
result signifies the system’s capacity to converge toward a This vicinity diminishes in scale as the adaptation gain, fre-
local optimal solution through continuous perturbations quency, and amplitude of the periodic signal employed for
and is crucial for analyzing the effectiveness and reliability extremum seeking increase, adhering to an inverse relation-
of such optimization techniques. The output y = J * con- ship. Although not a mandatory component, the incorpora-
ω
⎛1 ⎞ tion of a low-pass filter s + lω effectively filters out the
verges to J ″ O ω2 + a2 . The velocity of convergence is l
⎝ ⎠ cos(2ωt ) signal following the demodulator (multiplier),
1
related to ω , ω, a , k , and J ′ and this particular convergence further refining the optimization process. Through meticu-
outcome has a second-order degree. lous selection of the design parameters outlined below, we
In this investigation, Figure 6 provides an illustrative enhance the performance of the ESC system, thereby guaran-
representation of the ESC scheme, a methodology predicated teeing robust and dependable operation within a wide
upon analog optimization and reliant upon perturbations. spectrum of practical applications across diverse indus-
In order to broaden the applicability of perturbation-based tries. The comprehensive strategy of integrating pertur-
ESC to moderately unstable systems, particularly addressing bation-based ESC with advanced techniques and filtering
the challenge of autonomous vehicle source seeking, we mechanisms renders it a potent and adaptive instrument
derive valuable insights from the findings outlined in Chapter for tackling optimization challenges within intricate and
3 of the study by Zhang and Ordóñez [34]. By employing dynamic systems.
sophisticated methodologies including averaging and sin-
ωh = ωωH = ωδωH′ = O(ωδ ) , (23)
gular perturbation, our research demonstrates the conver-
gence of closed-loop system solutions toward a narrow vici- ωl = ωωL = ωδωL′ = O(ωδ ) , (24)
nity encompassing the extremum of the equilibrium map. k = ωK = ωδK′ = O(ωδ ) . (25)
Figure 14: Theta angle control under step and white noise disturbances.
In the context of this study, we denote ω and δ as time period τ = ωt , the system in Figure 7 can be
small positive constants, whereas ωH′ , ωL′ , and K′ are O(1) expressed in a more convenient form for further ana-
constants with positive values. It will be apparent that lysis and mathematical treatment. These transforma-
employing a small value for a is a prerequisite. The cut- tions facilitate a systematic approach for investigating
off frequencies of the filters should be set below the per- the system’s dynamics, aiding in the derivation of cri-
turbation signal frequency, necessitating both ωH < 1 and tical insights and control strategies for the perturbation-
ωL < 1 to be less than 1, as demonstrated in equations (23) based ESC scheme.
and (24). Furthermore, the adaptation gain k should also be dx ∼ (26)
chosen to be small to effectively attain the desired system ω = f (x , a (x , θ * + θ + a sin τ )) ,
dτ
performance. These considerations are pivotal for the suc-
cessful implementation and performance of the perturbation- ⎡ θ∼ ⎤ ⎡ K′ ξ ⎤
d
based ESC approach and are crucial for the optimization pro- ⎢ ξ ⎥ = ⎢− ωL + ωL′ (J (x ) − Jol(θ *) − η∼)a sin τ ⎥.
ξ
(27)
dτ ⎢ ∼⎥ ⎢ ⎥
cess in moderately unstable systems and autonomous vehicle ⎣η ⎦ ⎣ − ωH η∼ + ωH′ (J (x ) − Jol(θ *))
ξ
⎦
source-seeking applications.
The initial analysis focuses on the static case, wherein The subsequent theorem establishes a formal mathe-
the variable ξ is fixed at its equilibrium, and the averaging matical proposition.
method is utilized. Subsequently, the technique of singular
perturbation is applied to the entire system depicted in Theorem 2. [34]: Assuming that the feedback system (22)
Figure 6. To simplify the analysis, new coordinates are intro- and (23) satisfies Assumptions 1–2, it becomes feasible to
∼ identify a set of initial conditions centered at the point
duced, specifically θ = θˆ − θ * and η∼ = η − Jol (θ*), where
η̇ = − ωLη + ωLy as defined in Figure 7. By considering the (x , θˆ, ξ , η) = (l (θ *) , θ *, 0, Jol(θ *)) and incorporating ω̅ , δ̅ ,
Figure 15: Theta angle control under ramp and step disturbances.
and a̅ . This identification facilitates exponential conver- successfully integrated autopilot systems with the ESC algo-
gence of the solution (x (t ) , θ (t ) , ξ (t ) , η(t )) approaching a rithm, introducing unique disturbance rejection properties
neighborhood around the mentioned point with a size to the autopilots. This integration marks a pioneering advance-
of O(ω + δ + a ). Furthermore, the function y(t) tends to ment in disturbance rejection. Moreover, our method exhibits
approach an O(ω + δ + a )-neighborhood surrounding superior disturbance rejection capabilities compared to other
Jol(θ *), indicating the level of proximity to the optimal value existing approaches. Specifically, our proposed method can
of J (θ ) as influenced by the adaptation gain and perturba- effectively reject severe disturbances that other methods
tion amplitude on the convergence behavior. These findings struggle to address. The efficacy and superiority of our
significantly contribute to the understanding of the stability approach over previous methods are evident in the com-
and performance of the feedback system and its application parative analysis presented in Section 5. Our method is rig-
in ESC across various practical scenarios. orously compared with well-established methods such as
LQG and MPC. Examining the error functions depicted in
the figures, it is apparent that our proposed method demon-
3.3 Advanced autopilot strates significantly lower error levels when compared to
the LQG and MPC methods. Additionally, the tables pre-
In this section, we designed novel, adaptive autopilots by senting Root mean square error (RMSE) values further sub-
integrating ESC algorithm to them. This new autopilot stantiate the effectiveness and superiority of our proposed
structure has innovative disturbance rejection properties. method. The RMSE values for our method are notably
The innovative structure of the autopilots we have smaller than those associated with other methods, under-
developed represents a novel approach in the realm of scoring the enhanced performance and robustness of our
autopilots and disturbance rejection controllers. We have approach.
Figure 16: Theta angle control under step, ramp, and white noise disturbances.
Figure 8 shows fixed-wing UAV flight operation by uti- Validation, on the other hand, entails comparing the inte-
lizing autopilot-ESC integration under external disturbances. grated autopilot-ESC system with previous control methods
The fixed-wing UAV is controlled by conventional autopilot possessing disturbance rejection properties through simula-
under normal conditions. However, when it is exposed to tions to verify its functionality under various disturbances.
external disturbances ESC algorithm is utilized to reject In this study, we compared our proposed method with LQG
them and provide safety during flight operations. and MPC for validation. As evident from the comparative
In the research methodology, the calibration and validation analysis of the simulation results, the proposed method per-
of the tools, specifically the autopilots integrated with the ESC forms effectively and exhibits superiority over LQG and MPC.
algorithm, represent pivotal stages in ensuring the effectiveness Ultimately, the calibration and validation processes contri-
and reliability of the proposed system. Calibration entails the bute to the development of advanced disturbance rejection
precise adjustment of autopilot parameters and the ESC algo- autopilots capable of ensuring safe and reliable autonomous
rithm to optimize their performance and ensure compatibility. flight operations in UAVs.
This procedure guarantees that the proposed autopilots accu- As the proposed method is compared with LQG and
rately respond to control inputs and external disturbances, MPC in this study, defining the calibration/validation of
thereby augmenting their adaptive capabilities. Parameters of the tools employed in the research methodology necessi-
both the autopilots and the ESC algorithm were meticulously tates the provision of the following adjusted controller
adjusted to ensure stability, control, and disturbance mitigation. parameters:
Detailed information regarding the tuned parameters is pro- LQG tuned parameters for longitudinal theta angle
vided in Sections 4, 4.1, and 4.2. control were determined as follows:
Figure 17: Altitude control under white noise disturbances.
For LQG : Q = 1, R = 0.0005, Upper limits for plant outputs are named as scale
N = 0, initial states(x 0) = 0, factor: Scale factor{60,60} .
MPC weights are dimensionless and applied to the
K = [44.7420 102.3557 106.1251 41.0130]
scaled MV and OV values. The weight scales are as follows:
LQG tuned parameters for longitudinal altitude con- input[0 0], input rate[0.1 0.1], Output limits[200 10],
trol were determined as follows:
where MV represents the manipulated variables and OV
For LQG : Q = 1, R = 0.002,
denotes the output variables.
N = 0, initial states(x 0) = 0, In the subsequent subsections, explanations of pitch atti-
K = [22.2721 48.5148 52.1189 18.8181] tude hold, altitude hold, and yaw autopilots are provided.
Subsequently, the integration of the ESC algorithm with these
LQG tuned parameters for lateral psi angle control autopilots is illustrated in Figures 9–11, respectively.
were determined as follows:
For LQG : Q = 1, R = 0.004,

N = 0, initial states(x 0) = 0, 4 Adaptive pitch attitude hold
K = [9.0411 33.4947 18.9663 60.6987]
autopilot
For longitudinal and lateral control, tuned parameters
of MPC were determined as given below: The pitch attitude hold autopilot, also known as the pitch
Elevator and rudder were scaled as Min{ − 25, − 25} , hold mode or altitude hold mode, represents a ubiquitous
Max{25, 25} . feature within aircraft autopilot systems. Its principal role
Figure 18: Altitude control under sinusoidal and white noise disturbances.
is to autonomously sustain a specified pitch attitude or aircraft types, including commercial airliners, business jets,
flight level, thereby enabling pilots to mitigate their work- and general aviation aircraft. They offer significant advantages,
load and concentrate on other operational tasks. Upon particularly during cruise flight, where precise control of pitch
activation, the pitch attitude hold autopilot system employs attitude or altitude is essential for optimizing fuel efficiency,
various sensors, including an attitude indicator, altimeter, ensuring passenger comfort, and adhering to air traffic con-
and airspeed indicator, to ascertain the present pitch atti- trol instructions. It is crucial to emphasize that the operation
tude and altitude of the aircraft. Subsequently, it issues of the autopilot system, including the pitch attitude hold
directives to the pertinent control surfaces, typically the mode, must strictly adhere to the aircraft’s operating manual
elevators, to maintain the desired pitch attitude or altitude. and be conducted under appropriate flight conditions. Pilots
Upon engagement, the autopilot system maintains the rela- must remain vigilant and be prepared to disengage the autop-
tionship θ = α + γ at its current value. This control mechanism ilot, assuming manual control when necessary.
ensures that the pitch angle (θ) remains equivalent to the sum The diagram depicted in Figure 9 illustrates the
of the angle of attack (α) and the flight path angle (γ), recently designed autopilot. The internal component is
thereby contributing to stable flight conditions, as discussed utilized to stabilize systems that are inherently unstable,
in Chapters 7 and 8 of the study by Nelson [35] and Chapter 4 thereby ensuring effective control. Furthermore, the
of the study by Stevens et al. [36]. external component integrates ESC to facilitate the sys-
The primary function of the pitch attitude hold autopilot is tem’s adaptation to disturbances. This methodology guar-
to stabilize the aircraft at a specific pitch angle, directly influ- antees that, even in the presence of adverse external
encing its ascent, descent, or level flight. By doing so, it con- influences, the system can reliably return to a controlled
tributes to maintaining a consistent angle of attack, thereby state.
reducing the need for continuous pilot intervention. Pitch atti- We used the transfer function between q and δe as
tude hold autopilot systems are widespread across various aircraft model to design our proposed method.
We used the following parameter values for ESC for component within aircraft autopilot systems. Its principal
adaptive pitch attitude hold autopilot design: function is to automatically maintain a predetermined alti-
Gain k = 0.00000015, initial value = 0.00001, wl = 1 , tude, thereby enabling the pilot to alleviate their workload
wh = 5, w = 1 , a = 1 . and focus on other flight-related tasks. Upon activation, the
Transfer function for pitch attitude hold and altitude altitude hold autopilot system leverages various sensors,
hold autopilots are as follows: notably the altimeter and vertical speed indicator, to accu-
rately determine the aircraft’s current altitude and vertical
q 8.434s 3 + 17.27s 2 + 9.618s + (4.23 × 10−16)
= . speed. Subsequently, it adjusts the aircraft’s control sur-
δe s 4 + 2.267s 3 + 8.159s 2 + 2.808s + 3.876
faces and, where applicable, throttle settings to uphold
We used the following parameter values for pitch atti- the desired altitude, as elaborated in chapters 7 and 8 of
tude hold autopilot design: [35] and chapter 4 of [36]. The primary aim of the altitude
−20
Actuator transfer function = s + 20
, kq = 0.333 , Rate gyro hold autopilot is to sustain the aircraft at a specified altitude,
=1 , Attitude gyro = 0.524, Gc = kp = 0.925 typically established by the pilot or programmed into the
flight management system. This continuous maintenance of
altitude facilitates compliance with air traffic control direc-
tives and diminishes the pilot’s workload during level flight.
4.1 Adaptive altitude hold autopilot Altitude hold autopilot systems integrate with other
autopilot modes and sensors to achieve precise altitude con-
The altitude hold autopilot, also referred to as altitude hold trol. These modes may include vertical speed control, flight
mode or altitude hold function, constitutes a fundamental level change, or vertical navigation modes, allowing for more
Figure 19: Altitude control under step and white noise disturbances.
Figure 20: Altitude control under ramp and step disturbance.
sophisticated altitude management during climbs, descents, As seen in Figure 10, we developed a new autopilot
or transitions between altitudes. Additionally, certain altitude structure. We used the transfer function between q and
hold autopilot systems incorporate features such as altitude δe as aircraft model to design our proposed method.
capture and altitude pre-select. Altitude capture facilitates a We used the following parameter values for ESC for
seamless transition from the current altitude to a selected one adaptive altitude hold autopilot design:
by automatically adjusting pitch and thrust. Conversely, alti- Gain k = 0.000005, initial value = 0.00005, wl = 1
tude pre-select enables the pilot to input a desired altitude wh = 5, w = 1 , a = 1 ,
into the autopilot system, which the system will then capture Transfer function for Altitude Hold is the same as pitch
and maintain upon engagement. attitude hold autopilot and is given below:
Altitude hold autopilot systems are prevalent in a wide
q 8.434s 3 + 17.27s 2 + 9.618s + (4.23x10−16)
array of aircraft, ranging from smaller general aviation = 4
δe s + 2.267s 3 + 8.159s 2 + 2.808s + 3.876
planes to larger commercial airliners. Their utility is parti-
cularly pronounced during cruise flight, where the ability We used the following parameter values for Altitude
to maintain a specific altitude is paramount for considera- Hold Autopilot design:
−20
tions such as fuel efficiency, compliance with air traffic Actuator transfer function = s + 20
, kq = 0.333, GF
control instructions, and overall flight management. = 1 , kp = 0.7 ,
For Gc we used a PID controllers with parameter yaw effects under specific flight conditions, yet they do not
values P = 0.000405, I = 0, D = 0.002 fully automate the aircraft’s yaw control. Autopilot systems
primarily focus on pitch and roll control, which encompasses
altitude maintenance, airspeed, and navigation guidance.
Nevertheless, they indirectly influence yaw control by main-
4.2 Adaptive yaw autopilot taining the desired heading or track through coordinated roll
inputs and sustaining a consistent bank angle during turns.
Autopilot systems generally lack a specific “yaw autopilot” In conclusion, while autopilot systems typically do not
mode primarily because the aircraft’s yaw control is tradi- incorporate a specialized “yaw autopilot” mode, they none-
tionally managed by the rudder, predominantly operated theless play a significant role in maintaining coordinated
by the pilot or through coordinated flight control inputs. flight and indirectly influence yaw control through their
Yaw denotes the rotation of the aircraft around its vertical diverse operational modes, as elaborated in chapters 7 and
axis, influencing its heading or direction. Yaw control is vital 8 of the study by Nelson [35] and chapter 4 of the study by
for sustaining coordinated flight, particularly during turns or Stevens et al. [36].
when contending with crosswinds. Although certain aircrafts As illustrated in Figure 11, we developed a new adap-
feature yaw stability and control augmentation systems, these tive autopilot structure. We used the transfer function
are not typically termed “yaw autopilot” systems. Instead, between ψ and δa as aircraft model to design our pro-
they assist in preserving yaw stability and mitigating adverse posed method.
Figure 21: Altitude control under step, ramp, and white noise disturbances.
Figure 22: Psi angle control under white noise disturbance.
We used the following parameter values for ESC for k

θˆ = ξ , (30)
the newly adaptive yaw autopilot design: s
Gain k = 0.0003, initial value = 0.00015, wl = 2 , wh = 5, where v can be considered as a disturbance.
w = 1, a = 1. Figures 9–11 depict the implemented autopilot employing
Transfer function for Yaw Autopilot is as follows: the ESC algorithm. The incorporation of ESC enhances the
ψ 12.46s 3 + 96.3s 2 + 480.8s + 930.3 autopilot’s effectiveness in mitigating external disturbances.
= 4 . The newly advanced autopilots exhibit exceptional perfor-
δa s + 13.01s 3 + 90.54s 2 + 344.1s − 28.29
mance in both optimal and adverse conditions, as evidenced
We used the following parameter values for Yaw by the simulation figures.
Autopilot design:
10
Aileron transfer function = s + 10 .
For Gc , we used a PID controller with parameter values
P = 16.346, I = 118.36, D = 0.5378. 5 Simulation results
s
y−η= (Y + v ) , (28) Within this section, the proposed method is utilized to
s + ωh
govern the longitudinal and lateral modes of the UAV aircraft,
ωl as illustrated in Figures 9–11. In these figures, external nega-
ξ= (( y − η)α sin(ωt )) , (29)
s + ωl tive input v yields hazardous disturbances in the output Y .
Figure 23: Psi angle control under sinusoidal and white noise disturbances.
To showcase the efficacy of the proposed adaptive autopilots, tool was pivotal in ensuring the accuracy and reliability of the
various combinations of disturbances are introduced to the results presented in this work [40].
system response. As evidenced by the simulation results, The following simulation results were obtained by col-
noise and harsh disturbances such as step, ramp, and sinu- lecting figures of generations of our study. MATLAB was
soidal disturbances are applied to the longitudinal and lateral utilized for gathering these figures. The figures are original
outputs of the UAV [37–39]. It is apparent in the simulation work outputs.
figures that the proposed adaptive autopilots exhibit signifi-
cant effectiveness and demonstrate robust disturbance rejec-
tion properties.
The analysis and processing of data in this study 5.1 Longitudinal results
were conducted using MATLAB (MathWorks, 2021), a widely
employed software environment for scientific computing, 5.1.1 Adaptive pitch attitude hold results
numerical analysis, and data visualization. The selection of
MATLAB was grounded in its robust capabilities for managing In this section, noise, step, ramp, and sinusoidal distur-
and analyzing complex datasets, performing statistical bances were applied to the pitch angle of the fixed-wing
analyses, and facilitating result visualization. MATLAB was UAV. Simulation results demonstrate that, whereas the
specifically employed for statistical tests, plot creation, and conventional pitch attitude hold autopilot fails to reject
regression analyses. The choice of MATLAB as the analytical the noise and disturbances, the proposed autopilot method
Figure 24: Psi angle control under step and white noise disturbances.
successfully mitigates them. Step, ramp, and sinusoidal dis- Step, sinusoidal, and white noise disturbances are applied
turbances pose significant challenges for current control through integration in Figures 12–14, and the resulting out-
methods. Nonetheless, the proposed method exhibits notable comes are observed. As depicted in the figures, the pitch
performance improvements and effectively mitigates these attitude hold autopilot proves incapable of effectively
harsh disturbances. rejecting disturbances. However, employing pitch attitude
The poles set of transfer function is [−1.0139 + 2.4703i − hold autopilot and ESC in integration yields favorable out-
1.0139 − 2.4703i − 0.1196 + 0.7275i − 0.1196 − 0.7275i]. Given comes. The first channel of Figure 9 corresponds to the result
that all poles of the transfer function concerning the theta of Yv , where v is constructed by the addition of external
angle reside in the left half-plane, the system is deemed sinusoidal, step, and white noise disturbances, while Y repre-
stable. The values for maximum overshoot, undershoot, sents the output theta as depicted in Figure 9.
and rise time are provided below. It is evident that Figure 15 depicts the management of the theta angle in
due to the low-rise time, there is a notable increase in response to step disturbance and ramp disturbance. The
overshoot. diagrams clearly illustrate that the pitch attitude hold
Overshoot: 24.627%, undershoot: −1.704%, and rise autopilot fails to attain satisfactory results when external
time: 0.649488 s. ramp and step disturbances are introduced. However,
Figures 12–14 depict the management of the theta employing the pitch attitude hold autopilot and ESC sepa-
angle in the presence of disturbances. The diagrams illus- rately results in positive outcomes. In Figure 15, the first
trate that while the pitch attitude hold autopilot technique channel represents the result of Yv , where v is generated by
operates effectively under normal conditions, it fails to yield adding external ramp and step disturbances, while Y
satisfactory results in the presence of external disturbances. represents the output theta as depicted in Figure 9.
Figure 25: Psi angle control under step and ramp disturbances.
Figure 16 illustrates the management of the theta upon this framework, the subsequent section extends these
angle in response to step disturbance and white noise dis- disturbances to the altitude control of the fixed-wing UAV.
turbance. The diagram clearly indicates that while the Simulation outcomes indicate that the conventional alti-
pitch attitude hold autopilot technique operates effectively tude hold autopilot fails to mitigate the influence of such
under normal conditions, it fails to produce satisfactory disturbances. Conversely, our proposed autopilot method
results when external step and white noise disturbances demonstrates robustness against these disturbances, effec-
are introduced. However, employing the pitch attitude tively attenuating their effects. Step, ramp, and sinusoidal
hold autopilot and ESC separately leads to a favorable out- disturbances present considerable challenges for conventional
come. In Figure 16, the first channel represents the result of control methodologies. Nonetheless, our proposed approach
Yv , where v is generated by adding external step and white exhibits marked performance improvements and effectively
noise disturbances, while Y represents the output theta as mitigates these disruptive influences.
depicted in Figure 9. The poles set of transfer function is [−1.0139 + 2.4703i
−1.0139 −2.4703i −0.1196 + 0.7275i − 0.1196 − 0.7275i]. Given
that all poles of the transfer function governing altitude
5.1.2 Adaptive altitude hold results reside in the left half-plane, the system is deemed stable.
The maximum overshoot, undershoot, and rise time values
In Section 5.1.1, we introduced and applied various distur- are provided below. Notably, the elevated rise time corre-
bances, including noise, step, ramp, and sinusoidal pertur- sponds to a diminished overshoot.
bations to the pitch angle of the fixed-wing UAV. Building Overshoot: 0.505%, undershoot: 1.999 %, and rise time: 25.168 s.
Figure 26: Psi angle control under ramp and white noise disturbances.
Figures 17–19 illustrate altitude management amidst ramp and step disturbances. Nonetheless, employing the
disturbances. Under normal conditions, the altitude hold auto- altitude hold autopilot and ESC separately yields positive
pilot technique exhibits efficacy. However, its performance outcomes. In Figure 20, the first channel illustrates the Yv
falters when confronted with external disturbances. The outcome, where v denotes the inclusion of external ramp
diagrams depict the ramifications of sinusoidal and white and step disturbances, while Y represents the output alti-
noise disturbances, as well as step disturbances introduced tude as depicted in Figure 10.
via integration. It is evident from the figures that the altitude Figure 21 illustrates altitude control in the presence of
hold autopilot struggles to adequately counteract and reject step, ramp, and white noise disturbances. The diagram
these disturbances. Nonetheless, employing the altitude hold distinctly shows that the altitude hold autopilot tech-
autopilot and ESC separately yields favorable results. In nique performs effectively under normal conditions;
Figure 17, the first channel depicts the Yv result, where v however, it fails to produce satisfactory results when
denotes the incorporation of external step, sinusoidal, and external step, ramp, and white noise disturbances are
white noise disturbances into the altitude output Y , as illu- introduced. Nonetheless, employing the altitude hold
strated in Figure 10. autopilot and ESC separately leads to favorable out-
Figure 20 presents an overview of altitude manage- comes. In Figure 21, the first channel depicts the Yv result,
ment amidst step and ramp disturbances. The diagrams where v is generated by the inclusion of step, ramp, and
distinctly illustrate that the altitude hold autopilot fails to white noise disturbances, while Y represents the output
attain satisfactory results upon the introduction of external altitude as depicted in Figure 10.
Figure 27: (a) Theta angle control under noise. (b) Error functions of LQG, MPC, and the proposed method under noise.
5.2 Lateral results proves effective under normal conditions, it falls short
in delivering satisfactory results when encountering
5.2.1 Adaptive yaw autopilot results external disturbances. Step, sinusoidal, and white noise dis-
turbances are introduced through integration in Figures
The poles set of transfer function is [−7.1547 − 2.9679 + 22–24, with subsequent recording of the observed outcomes.
6.34495i − 2.9679 − 6.34495i 0.0805]. The instability of the It is apparent from the figures that the yaw autopilot fails
system is indicated by the presence of a positive pole in the to sufficiently attenuate these disturbances. Conversely,
transfer function of the psi angle. The specific metrics char- employing the yaw autopilot and ESC separately yields
acterizing system response are detailed below, encom- favorable outcomes. In Figure 22, the first channel depicts
passing maximum overshoot, undershoot, and rise time. the result of Yv , where v is obtained by adding external noise
Notably, the brief rise time is associated with a relatively to the output psi Y , as illustrated in Figure 11.
elevated overshoot level. Figure 25 provides a visual representation of how the psi
Overshoot: 10.556%, undershoot: −5.071 %, and rise angle is managed in the presence of step and ramp distur-
time: 0.019133 s. bances. The diagrams vividly illustrate that the yaw autopilot
Figures 22–24 present an overview of the management fails to achieve satisfactory results when external ramp and
of the psi angle in the presence of disturbances. The dia- step disturbances are introduced. However, employing the yaw
grams elucidate that while the yaw autopilot technique autopilot and ESC separately yields favorable outcomes. In
Figure 27: (Continued)
Figure 25, the first channel represents the result of Yv , where v introduced. However, employing the yaw autopilot and ESC
is obtained by adding external ramp and step disturbances, separately leads to a favorable outcome. In Figure 26, the first
while Y denotes the output psi as depicted in Figure 11. channel represents the result of Yv , where v is generated by
Figure 26 presents a visual representation of how adding step, ramp, and white noise disturbances, while Y
the psi angle is managed in the presence of step, ramp, represents the output psi as depicted in Figure 11.
and white noise disturbances. The diagram clearly indicates
that while the yaw autopilot technique operates effectively
under normal conditions, it fails to achieve satisfactory results 5.3 Comparison of the proposed method
when external step, ramp, and white noise disturbances are with MPC and LQG methods to show the
superiority
Table 1: RMSE values
In this section, we conduct a comparative analysis between
Controller RMSE value
our proposed method and previously established techniques
renowned for their disturbance rejection capabilities,
LQG 0.0568 namely, the LQG Regulator and MPC. These methodo-
MPC 0.3260
logies embody well-established controller designs. We
Proposed method 2.0568 × 10‒14
have devised and implemented noise and disturbance
Figure 28: (a) Theta angle control under white noise disturbance. (b) Error functions of LQG, MPC, and the proposed method under white noise
disturbance.
scenarios to assess the performance of these methods. output of each controller, and the resulting error is ana-
The subsequent figures present a visual depiction of the lyzed. Figure 27b illustrates that MPC exhibits the highest
superior performance of our proposed method compared error amplitudes, followed by LQG with lower error ampli-
to both LQG and MPC. tudes, while the proposed method demonstrates the lowest
error amplitudes. Consequently, in a broader assessment,
MPC displays the least favorable performance, whereas
5.3.1 Adaptive pitch attitude hold comparison results our proposed method stands out as the most proficient in
noise rejection.
Figure 27a depicts the implementation of theta angle con- Table 1 presents the RMSE values for the various con-
trol amidst noise interference. Observable noise is evident trollers, serving as a metric to assess the effectiveness of
in the output of both LQG and MPC. LQG demonstrates our method in mitigating noise. As depicted in Table 1, the
superior noise rejection performance compared to MPC. RMSE value for MPC is the highest, followed by a lower
However, the proposed approach consistently yields the RMSE value for LQG. In contrast, the RMSE value for the
most optimal results in terms of noise rejection. proposed method is the smallest. This signifies the superior
To enable a more comprehensive comparison among noise rejection capability of our proposed method com-
controllers, the clean output is subtracted from the faulty pared to both LQG and MPC.
Figure 28a and b illustrate a discernible trend wherein 5.3.2 Adaptive altitude hold comparison results
the level of noise introduced to the system’s output under-
goes an increase compared with Figure 27a and b. Notably, Figure 29a elucidates the administration of altitude control
the magnitude of noise amplification is considerably higher in the presence of ramp and sinusoidal disturbances. It is
in the systems controlled by LQG and MPC. Conversely, the apparent that both LQG and MPC exhibit substantial dis-
increment in noise for the system governed by our proposed turbance contributions in their outputs. LQG demonstrates
method is notably modest. This particular scenario becomes superior disturbance rejection performance compared to
evident when examining the relevant information provided MPC. However, the recommended approach consistently
in Figure 28a and b. produces the most favorable outcomes in terms of distur-
Table 2 presents a tabulated representation of the bance rejection.
RMSE values associated with the different controllers, ser-
ving as a quantitative measure to evaluate the efficacy of
our method in mitigating noise. As illustrated in Table 2,
MPC exhibits the highest RMSE value, while LQG displays
a comparatively lower RMSE value. In sharp contrast, Controller RMSE value
the RMSE value associated with our proposed method is
the most minimal. This underscores the heightened noise LQG 0.1796
MPC 1.6703
rejection proficiency of our proposed method compared to
Proposed method 2.0567 × 10‒14
both LQG and MPC.
Figure 29: (a) Altitude control under ramp and sinusoidal disturbances. (b)Error functions of LQG, MPC, and the proposed method under ramp and
sinusoidal disturbances.
To enable a more rigorous comparison among the con- boasts the lowest RMSE value. This firmly underscores the
trollers, the clean output is subtracted from the faulty superior disturbance rejection capabilities of our proposed
output of each controller, and the resulting errors are exam- method compared to both LQG and MPC.
ined. As depicted in Figure 29b, MPC exhibits the largest error
amplitudes, LQG displays smaller error amplitudes, and the
proposed method exhibits the lowest error amplitudes. There- 5.3.3 Adaptive yaw autopilot comparison results
fore, in a comprehensive assessment, MPC is characterized by
the poorest performance, while our proposed method demon- Figure 30a illustrates the management of psi angle control
strates the most superior performance in the context of dis- under the influence of a step and white noise disturbances.
turbance rejection. It is evident that both LQG and MPC approaches exhibit a
Table 3 provides the RMSE values for the controllers, level of noise in their outputs. LQG demonstrates superior
serving as a metric to convey the superiority of our method in performance in rejecting both step disturbance and white
mitigating ramp and sinusoidal disturbances. As evidenced in noise when compared to MPC. Remarkably, the recom-
Table 3, MPC exhibits the highest RMSE value, followed by mended approach consistently yields the most favorable
LQG with a lower RMSE value, while the proposed method outcomes concerning noise rejection.
To enable a more rigorous comparison among the con- 6 Discussion

trollers, we subtract the clean output from the faulty
output of each controller and analyze the resulting error. In this section, we scientifically present our method by
As illustrated in Figure 30b, MPC displays the largest error framing it within the subsequent paragraph. This approach
amplitudes, while LQG exhibits smaller error amplitudes, ensures a structured and coherent explanation of our
and the proposed method showcases the lowest error methodology.
amplitudes. Therefore, in a comprehensive evaluation, a) Main findings of the present study: The current study
MPC demonstrates the poorest performance, whereas our focuses on developing autopilots for stability, control,
proposed method emerges as the top performer in terms of and the rejection of external disturbances. The proposed
noise rejection.
Table 4 illustrates the RMSE values for the controllers, pro-
viding evidence of the superior performance of our method in
mitigating step disturbance and white noise. As evident in Table
4, the RMSE value for MPC is the highest, while LQG demon- Controller RMSE value
strates a lower RMSE value than MPC. Importantly, the RMSE
LQG 140.1356
value for our proposed method is the lowest among the three.
MPC 69.3386
This unequivocally demonstrates that our proposed method Proposed method 2.7697 × 10‒13
surpasses both LQG and MPC in terms of white noise rejection.
Figure 30: (a) Psi angle control under step and white noise disturbances. (b) Error functions of LQG, MPC, and the proposed method under step white
noise and disturbances.
method provides significant disturbance rejection proper- c) Implication and explanation of findings: The effective-
ties by rejecting severe disturbances. In the comparison ness and benefits of the proposed method can be shown
part of the simulation results, the proposed method exhi- by explaining the findings of the proposed work. The
bits superiority over previous methods having distur- proposed method demonstrates excellent performance
bance rejection properties. For instance, LQG and MPC both in normal conditions and under disturbances. The
fail to mitigate harsh disturbances, and do not provide proposed autopilot is able to reject harsh disturbances.
acceptable responses. Overall, providing novel, advanced, This feature provides superiority for the proposed
and superior disturbance rejection properties are the method in the field of disturbance rejection control
main findings for this study. because other previous controllers cannot reject these
b) Comparison with other studies: In comparison to existing external factors. The explanation of the findings can
studies, the integration of ESC with autopilots presents a be made with these expressions. The findings of this
novel approach to enhance disturbance rejection in air- study bear significant implications for the field of
craft systems. Unlike traditional autopilots lacking adap- autonomous flight control in aircraft UAVs. Since the
tive capabilities, the proposed method provides superior proposed method rejects severe disturbances, it can be
disturbance rejection, surpassing robust controllers such implicated that this method provides robustness when
as LQG Regulator and MPC by effectively rejecting severe severe environmental condition occurs during flight.
disturbances. The Autopilot-ESC integration method yields a novel
outcome for aircraft control that provides significant newly designed proposed autopilots demonstrate
safety for UAV systems during flight. Enhanced perfor- robust disturbance rejection properties, effectively
mance during specific missions is achievable with rejecting harsh disturbances. In this regard, they out-
reduced reliance on remote control, facilitated by perform previous methods with disturbance rejection
this novel and robust disturbance rejection controller. capabilities. However, the study primarily focuses on
The proposed method aids in minimizing the risk of disturbance rejection and may not address all aspects
accidents or damage to the UAV and its surroundings, of autopilot performance comprehensively. Another
ensuring safe and reliable operation in various envir- significant drawback of the proposed method is its
onments. The proposed approach enables the UAV reliance on a single optimization criterion, typically
to adjust its behavior in real-time, and enhance the
overall mission success rates, even in challenging
situations where disturbances may occur.
d) Strengths and limitations: One strength of this study lies
Controller RMSE value
in its innovative and novel approach to enhance distur-
bance rejection in aircraft autopilots, which can pro- LQG 16.1546
MPC 123.8637
foundly impact the safety and reliability of autonomous
Proposed method 2.3107 × 10 14
flight operations under external disturbances. The
based on the system’s performance metric. In addition, external disturbances, several limitations should be noted.
ESC might suffer from slow convergence rates, poten- Although the study emphasizes the superiority of the upgraded
tially impairing its effectiveness in disturbance detec- autopilots over robust controllers like LQG and MPC in hand-
tion and recovery. ling various disturbance types, the comparison may not fully
capture the performance differences under all possible sce-
narios. Additionally, the complexity introduced by ESC integra-
tion may pose challenges to computational efficiency. Finally,
7 Conclusion the study primarily focuses on the pitch attitude hold, altitude
hold, and yaw autopilots, potentially limiting the generaliz-
This study introduces enhanced autopilots with distur- ability of the findings to other autopilot functionalities or air-
bance rejection capabilities, specifically focusing on pitch craft configurations. Prospects for future research regarding
attitude hold autopilot, altitude hold autopilot, and yaw the proposed method include flying a UAV formation with
autopilot. The augmentation of disturbance rejection fea- disturbance rejection properties. Moreover, optimization of
tures was achieved by integrating the ESC algorithm into the trajectory of a UAV group is another prospect for future
these autopilots. This autopilot is adaptable to both stable investigation.
and unstable systems. The innovative approach involves Key results: The proposed approach provides key results
the construction of two distinct parts, each serving a spe- as disturbance rejection capabilities, adapting parameters for
cific purpose. The inner part utilizes the autopilot for unknown inputs, and restoring controlled responses. For
optimal stability and control under normal conditions, instance, the proposed method significantly provides strong
while the outer part incorporates ESC for disturbance disturbance handling for autopilots by successfully elimi-
rejection by mitigating severe external disturbances. ESC nating various external factors. Moreover, it enhances air-
ensures that the system maintains the desired control level craft operation, reduces operator workload, and ensures
despite adverse disturbances. The simulation results illus- reliable autonomous flight capabilities. Furthermore, the
trate that the autopilot is highly effective in achieving optimal adaptability of the Autopilot-ESC to the changing environ-
flight conditions, while ESC proves essential for mitigating mental conditions makes it suitable to be employed in aircraft
harsh disturbances. The novelty of this approach lies in the UAVs. Finally, the proposed method shows superiority prop-
elimination of severe disturbances, resulting in a unique dis- erties to compare with other robust controllers like LGQ reg-
turbance rejection adaptive autopilot through the integration ulator and MPC in severe disturbance elimination.
of ESC. Significance of key results: The key results highlight
In conclusion, the integration of ESC with conventional the significance of the proposed method in maintaining
autopilots represents a significant advancement in enhan- control and stability under severe disturbances, under-
cing the disturbance rejection capabilities of aircraft UAV scoring its essential role in achieving reliable and safe
systems. By incorporating ESC into pitch attitude hold flight operations. By effectively mitigating disturbances,
autopilot, altitude hold autopilot, and yaw autopilot, these the autopilot ensures optimal flight conditions even in
systems acquire adaptive features capable of effectively challenging environments. In addition, enhancing aircraft
managing external disturbances. Through simulations, it operation, decreasing operator workload and providing
has been demonstrated that the autopilot-ESC integration safety are very crucial for the flight field. Moreover, key
surpasses the capabilities of previous methods like LQG results highlight superiority of the proposed method that is
and MPC. The theoretical contributions of this study lie very significant for minimizing the effects of disturbances
in the development of a robust and effective approach by on UAVs. Overall, this innovative approach represents a
utilizing the ESC algorithm together with conventional unique solution for enhancing aircraft safety and perfor-
autopilots for managing aircraft systems, stability, control, mance through newly designed disturbance rejection adap-
and rejecting external disturbances, ultimately advancing tive autopilot systems.
the field of autonomous flight control. However, the novel The integration of the ESC algorithm into these autop-
contribution in this research is to provide advanced dis- ilots marks a novel approach, enabling adaptability to both
turbance rejection properties to conventional autopilots by stable and unstable systems. We enabled a novel and
integrating ESC to them, thereby transforming them into superior disturbance rejection property for the conventional
innovative and superior controllers. autopilots in this method. By constructing a dual-part system,
While the study proposes an innovative method of inte- where the inner part ensures stability and control while the
grating the ESC algorithm with conventional autopilots to outer part provides adaptability for disturbance rejection
enhance disturbance rejection capabilities and adaptability to through ESC, the autopilot demonstrates remarkable
resilience against adverse external factors. This is a superior [7] Kumar R, Deshpande AM, Wells JZ, Kumar M. Flight control of
contribution in this field. The application of ESC with autopi- sliding arm quadcopter with dynamic structural parameters. IEEE/
lots in response to external factors showcases its efficacy in RSJ International Conference on Intelligent Robots and Systems
(IROS). IEEE; 2020. p. 1358–63. doi: 10.1109/IROS45743.2020.
maintaining the desired control level and restoring system sta-
9340694.
bility. Ultimately, this research contributes new knowledge by [8] Shin D, Song Y, Oh J, Oh H. Nonlinear disturbance observer-based
presenting a unique and superior disturbance rejection adap- standoff target tracking for small fixed-wing UAVs. Int J Aeronaut
tive autopilot system, eliminating disturbances, and enhancing Space Sci. 2021;22:108–19. doi: 10.1007/s42405-020-00275-6.
aircraft safety and performance in challenging operational [9] Eliker K, Zhang W. Finite-time adaptive integral backstepping fast
terminal sliding mode control application on quadrotor UAV. Int J
environments.
Control Autom Syst. 2020;18:415–30. doi: 10.1007/s12555-019-0116-3.
[10] Wu X, Mueller MW. In-flight range optimization of multicopters
Funding information: Authors state no funding involved. using multivariable extremum seeking with adaptive step size.
IEEE/RSJ International Conference on Intelligent Robots and
Author contributions: All authors have accepted responsi- Systems (IROS). IEEE; 2020. p. 1545–50. doi: 10.1109/IROS45743.
2020.9340762.
bility for the entire content of this manuscript and con-
[11] Fiorentini L, Serrani A. Adaptive restricted trajectory tracking for a
sented to its submission to the journal, reviewed all the
non-minimum phase hypersonic vehicle model. Automatica.
results and approved the final version of the manuscript. 2012;48(7):1248–61. doi: 10.1016/j.automatica.2012.04.006.
HB managed the research, methodology determination, [12] Fiorentini L, Serrani A, Bolender MA, Doman DB. Nonlinear robust
conceptualization and design of the study, and preparation adaptive control of flexible air-breathing hypersonic vehicles.
of the manuscript. IB reviewed and edited the manuscript, J Guidance Control, Dyn. 2009;32(2):402–17. doi: 10.2514/1.39210.
[13] Wang Q, Stengel RF. Robust nonlinear control of a
managed analysis, interpretation, and provided feedback
hypersonic aircraft. J Guid Control Dyn. 2000;23(4):577–85. doi: 10.
to reach the publication standards. 2514/2.4580.
[14] Seferian V, Bazzi AM. Comprehensive induction motor fault diag-
Conflict of interest: Authors state no conflict of interest. nosis using extremum seeking control. IEEE Transportation
Electrification Conference & Expo (ITEC). IEEE; 2022. p. 1011–4.
Data availability statement: The data are not publicly doi: 10.1109/ITEC53557.2022.9813752.
[15] Kantue P, Pedro JO. Integrated fault-tolerant control of a quad-
available due to privacy or ethical restrictions.
copter UAV with incipient actuator faults. Int J Appl Maths Comput
Sci. 2022;32(4):601–17. doi: 10.34768/amcs-2022-0042.
[16] Vandermeulen I, Guay M, McLellan PJ. Discrete-time distributed
extremum-seeking control over networks with unstable dynamics.
IEEE Trans Control Netw Syst. 2017;5(3):1182–92. doi: 10.1109/TCNS.
References 2017.2691464.
[17] Sahneh FD, Hu G, Xie L. Extremum seeking control for systems with
[1] Shen S, Xu J. Adaptive neural network-based active disturbance time-varying extremum. Proceedings of the 31st Chinese Control
rejection flight control of an unmanned helicopter. Aerosp Sci Conference. IEEE; 2012. p. 225–31.
Technol. 2021;119:107062. doi: 10.1016/j.ast.2021.107062. [18] Oliveira TR, Krstić M, Tsubakino D. Extremum seeking for static
[2] Labbadi M, Cherkaoui M. Adaptive fractional-order nonsingular maps with delays. IEEE Trans Autom Control. 2016;62(4):1911–26.
fast terminal sliding mode based robust tracking control of quad- doi: 10.1109/TAC.2016.2564958.
rotor UAV with Gaussian random disturbances and uncertainties. [19] Xie F, Zhang X, Fierro R, Motter M. Autopilot-based nonlinear UAV
IEEE Trans Aerosp ElectrSyst. 2021;57(4):2265–77. doi: 10.1109/TAES. formation controller with extremum-seeking. Proceedings of the
2021.3053109. 44th IEEE Conference on Decision and Control. IEEE; 2005.
[3] Liu C, McAree O, Chen WH. Path‐following control for small fixed‐ p. 4933–8. doi: 10.1109/CDC.2005.1582943.
wing unmanned aerial vehicles under wind disturbances. Int J [20] Binetti P, Ariyur KB, Krstic M, Bernelli F. Control of formation flight
Robust Nonlinear Control. 2013;23(15):1682–98. doi: 10.1002/ via extremum seeking. Proceedings of the 2002 American Control
rnc.2938. Conference. (IEEE Cat. No. CH37301). Vol. 4. IEEE; 2002. p. 2848–53.
[4] Lai YC, Le TQ. Adaptive learning-based observer with dynamic doi: 10.1109/ACC.2002.1025221.
inversion for the autonomous flight of an unmanned helicopter. [21] Lavretsky E, Hovakimyan N, Calise A. Adaptive extremum seeking
IEEE Trans Aerosp ElectrSyst. 2021;57(3):1803–14. doi: 10.1109/TAES. control design. Proceedings of the 2003 American Control
2021.3050653. Conference. Vol. 1. IEEE; 2003. p. 567–72. doi: 10.1109/ACC.2003.
[5] He G, Yu L, Huang H, Wang X. A nonlinear robust sliding mode 1239079.
controller with auxiliary dynamic system for the hovering flight of a [22] Choi JY, Krstic M, Ariyur KB, Lee JS. Extremum seeking control for
tilt tri-rotor UAV. Appl Sci. 2020;10(18):6551. doi: 10.3390/ discrete-time systems. IEEE Trans Autom Control.
app10186551. 2002;47(2):318–23. doi: 10.1109/9.983370.
[6] Lanzon A, Freddi A, Longhi S. Flight control of a quadrotor [23] Zeghlache S, Kara K, Saigaa D. Type-2 fuzzy logic control of a 2-DOF
vehicle subsequent to a rotor failure. J Guid Control Dyn. helicopter (TRMS system. Open Eng. 2014;4(3):303–15. doi: 10.2478/
2014;37(2):580–91. doi: 10.2514/1.59869. s13531-013-0157-y.
[24] Krause U. Active noise control experiments using sound energy flu. with multiple ailerons. Int J Adapt Control Signal Process.
Open Eng. 2015;5(1):157–65. doi: 10.1515/eng-2015-0021. 2023;37(5):1072–101. doi: 10.1002/acs.3561.
[25] Gheni EZ, Al-Khafaji HM, Alwan HM. A deep reinforcement learning [32] Xu Z, Fan L, Qiu W, Wen G, He Y. A robust disturbance-rejection
framework to modify LQR for an active vibration control applied to controller using model predictive control for quadrotor UAV in
2D building models. Open Eng. 2024;14(1):20220496. doi: 10.1515/ tracking aggressive trajectory. Drones. 2023;7(9):557. doi: 10.3390/
eng-2022-0496. drones7090557.
[26] Yu S, Fan X, Qi J, Wan L, Liu B. Attitude control of quadrotor UAV [33] Beard RW, McLain TW. Small unmanned aircraft: Theory and
based on integral backstepping active disturbance rejection con- practice. New Jersey: Princeton University Press; 2012.
trol. Trans Inst Meas Control. 2024;46(4):703–15. doi: 10.1177/ [34] Zhang C, Ordóñez R. Extremum-seeking control and applications: a
0142331223118542. numerical optimization-based approach. New York: Springer
[27] Chebbi J, Briere Y. Robust active disturbance rejection Science & Business Media; 2011.
control for systems with internal uncertainties: Multirotor UAV [35] Nelson RC. Flight stability and automatic control. Vol. 2. New York:
application. J Field Robot. 2022;39(4):426–56. doi: 10.1002/rob. WCB/McGraw Hill; 1998.
22058. [36] Stevens BL, Lewis FL, Johnson EN. Aircraft control and simulation:
[28] Wang Z, Zhao T. Based on robust sliding mode and linear active dis- dynamics, controls design, and autonomous systems. New Jersey:
turbance rejection control for attitude of quadrotor load UAV. Nonlinear John Wiley & Sons; 2015.
Dyn. 2022;108(4):3485–503. doi: 10.1007/s11071-022-07349-y. [37] Ahi B, Haeri M. Linear active disturbance rejection control from the
[29] Wang Z, Zhao T. Adaptive-based linear active disturbance rejection practical aspects. IEEE/ASME Trans Mechatron. 2018;23(6):2909–19.
attitude control for quadrotor with external disturbances. Trans doi: 10.1109/TMECH.2018.2871880.
Inst Meas Control. 2022;44(2):286–98. doi: 10.1177/ [38] Ahmed N, Chen M, Shao S. Disturbance observer based
01423312211031781. tracking control of quadrotor with high-order
[30] Azid SI, Ali SA, Kumar M, Cirrincione M, Fagiolini A. Precise trajec- disturbances. IEEE access. 2020;8:8300–13. doi: 10.1109/ACCESS.
tory tracking of multi-rotor UAVs using wind disturbance rejection 2020.2964013.
approach. IEEE Access. 2023;11:91796–806. doi: 10.1109/ACCESS. [39] Chen Z, Jia H. Design of flight control system for a novel tilt-rotor
2023.3308297. UAV. Complexity. 2020;2020:1–4. doi: 10.1155/2020/4757381.
[31] Sattar A, Wang L, Ansari S, Hoshu AA, Khan SK. Disturbance rejec- [40] MATLAB. MATLAB version:9.11 [R2021b]. 2021 MathWorks. https://
tion enhancement using predictive control for the fixed‐wing UAV www.mathworks.com.

Advanced Autopilot Design With Extremum-Seeking Control For Aircraft Control

Uploaded by

Copyright:

Available Formats

Advanced Autopilot Design With Extremum-Seeking Control For Aircraft Control

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Advanced Autopilot Design With Extremum-Seeking Control For Aircraft Control

Uploaded by

Copyright:

Available Formats

Open Engineering 2024; 14: 20240044

Haci Baran* and Ismail Bayezit

Advanced autopilot design with extremum-

⎡ C Xδe (a )δe ⎤ ⎡Tp(δtVa ) ⎤

where A1 , B1 , and C1 are chosen symbols to denote a part

The variables ẋ and u̇ represent the ﬁrst derivatives

mentioned state variables, facilitating a comprehensive ⎡ Xδe Xδe ⎤

Figure 2: ESC block diagram.

Figure 6: Sinusoidal perturbation based analog optimization.

Figure 7: Perturbation based ESC.

Assumption 1. There exists a smooth function

Assumption 2. There exists θ* ∈ such that

Figure 9: The improved pitch attitude hold autopilot block diagram.

Figure 10: The improved altitude hold autopilot block diagram.

Figure 11: The improved yaw autopilot block diagram.

Figure 12: Theta angle control under white noise disturbances.

Figure 17: Altitude control under white noise disturbances.

For LQG : Q = 1, R = 0.004,

Figure 20: Altitude control under ramp and step disturbance.

Figure 22: Psi angle control under white noise disturbance.

We used the following parameter values for ESC for k

Figure 27: (Continued)

Figure 28: (Continued)

Figure 29: (Continued)

To enable a more rigorous comparison among the con- 6 Discussion

Figure 30: (Continued)

You might also like