A Self-Adjusting Multi-Objective Control Approach
A Self-Adjusting Multi-Objective Control Approach
H O S T E D BY
Alexandria University
ORIGINAL ARTICLE
a
Department of Mechanical Engineering, Faculty of Engineering, Alexandria University, Alexandria 21544, Egypt
b
Department of Electromechanical Engineering, Faculty of Engineering, Alexandria University, Alexandria 21544, Egypt
c
Faculte´ d’administration, Universite´ de Moncton, Moncton, E1A 3E9 NB, Canada
d
College of Technology, Architecture & Applied Engineering, Bowling Green State University, Bowling Green, 43402 OH, USA
KEYWORDS Abstract The quadrotor represents a class of highly nonlinear dynamic systems and its controlla-
Unmanned aerial vehicles; bility features are challenging. Hence, it serves as an ideal unmanned aerial vehicle platform to val-
PID control design; idate many artificial intelligence-based research investigations. Nonetheless, most of the offline
Multi-objective optimiza- tuning approaches devote efforts to find near optimal control gains to regulate individually the
tion; decoupled motion directions. This work adopts a multi-objective self-adjusting search mechanism
Metaheuristics; to actuate the motions of a quadrotor via deciding the control gains of the interacting loops simul-
Nonlinear threshold accept- taneously. This algorithm employs a first order low pass filter transfer function as an accepting
ing functions approach for the tunning mechanism. The proposed approach is compared with a Genetic Algo-
rithm and another nonlinear Proportional-Integral-Derivative approach to highlight the usefulness
of the proposed mechanism. It was founded that quadrotor follows the desired trajectory with a
small tracking error of less than 2% in the X-Y plane and less than 1 % error tracking error in
the altitude Z. Also, it is recorded that MONLTA can overcome the simulated wind disturbances
of 0.1 N.m as a disturbance torque.
Ó 2023 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Alexandria
University This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).
1. Introduction
Nomenclatures
payload transfer, crop dusting, filming, medical aid delivery, affiliated to the computational intelligence to develop efficient
and parcels [2]. The quadrotor is one of the most popular and innovative metaheuristics to ensure robust PID perfor-
UAVs because it has a simple low-cost structure, easy to main- mance. Ziegler–Nichols is one of the classical tuning tech-
tain, and flexible maneuverability. Hence, it provides compet- niques for PID control loops, but it is ineffective for systems
itive advantages when compared with fixed-wing aircraft and with coupled control loops such as rotorcraft UAVs. Meta-
helicopters in applications such as dense fields, cluttered areas, heuristics such as Genetic Algorithm (GA) as well as iterative
and hazardous industrial areas like offshore oil rigs and approaches have been to solve optimization problems in power
nuclear plants [3–5]. The quadrotor becomes the most signifi- systems applications [21–23]. Many research studies have been
cant among various types of UAVs so much research has been conducted to reach near-optimal settings to tune the PID con-
conducted in this field. The challenges pertaining to developing trollers using (GA), Particle Swarm(PSO), Colony Optimiza-
UAVs involve their high nonlinearity, multivariable multi- tion (ACO)[24–26], and LQR [27–29]. The range of the
input, multi-output relations, and high coupling with a larger tunned gains and the structure of the objective cost function
number of dynamics that be modeled implicitly. Further, the are shown to be critical and specific to the engineering applica-
difficulty to stabilize the UAVs subject to severe disturbances tion. Barjuei and Ortiz [30] presented a complete comparison
without external corrective forces. All of this challenge the between controllers in terms of control position and mechani-
control and autonomous operation of UAVs [6]. The recent cal vibration of a 3 Degrees-of-Freedom (DOF manipulator.
advances in enhancing the communication bandwidth and The controllers were (i) LQR, (ii) model predictive (iii) H1
computational power of systems provide the opportunity to loop shaping, and (iv) l-synthesis. Batista et al. [31] introduced
accomplish further improvements in the control of complex a comparison between the (PID) and the (LQR) controllers
unmanned systems [7]. applied to the speed control of one cylindrical joint. They
Many control approaches have been adopted, such as self- founded that the PID was better than the LQR in terms of
tunned control loops based on Proportional–Integral–Deriva the rise time. Also, they concluded that PID controllers are
tive (PID) structures [8]. These approaches required a linear robust and stable for the frequency range from 10 to
approximation of the rotorcraft (UAV) around a designated 100 rad/s. The importance of a multi objective optimization
operating region, since the PID controller is a single-input of an industrial robot arm was conducted [32], and the impor-
single-output (SISO) system. Even though, PID-based control tance of optimization of robotic systems was highlighted
schemes have been successfully developed rotorcraft flight con- [33,34], used an LQR with integral error action for the linear
trol systems. The recent advances in flight control adopt mod- model of the quadrotor [35].
ern control approaches such as H1 [9] and Linear Quadratic Recently different methods were considered to improve the
Regulators (LQR) [10,11]. However, these approaches were performance of UAV controllers. Idrissi et al. [36] developed a
faced by many technical challenges in the implementation control mechanism that tackles the dynamic variations of a
phase [8] Other control techniques have been adopted such UAV during flight. The Vehicle Routing Problem with Drone
as nonlinear inverse dynamics [11], involve feedback lineariza- (VRPD) is solved using ACO [37]. Zeng et al. [38] addressed
tion [12], adaptive control [13], sliding mode controller[14], the coordination and routing of a drone-truck pairing. Yan
control using neural network [15], fuzzy control[16], control et al. [39] developed an attitude control for UAVs using an
using fuzzy integral sliding mode control [17], control using adaptive fuzzy PID method. This research managed to solve
integral backstepping [18], and robust adaptive control [19]. the reliant on expert knowledge to determine the membership
Similarly, these approaches have been challenged with the values for the fuzzy system and the control effect. Sonugür [40]
complexity and hard-to-implement control strategies. The con- presented a comprehensive review to innovative approaches
trol design of UAVs using PID structures proved to be practi- control approached of quadrotor. Xu et al. [41] presented atti-
cal and dominated other techniques in many engineering tude control for quadrotor with time-varying effectiveness
applications due to its simple operation principle [20]. The tun- fault in terms of adaptive dynamic programming. Single-
ing of PID controllers attracts the attention of research groups objective optimization approaches based on heuristics such
A self-adjusting multi-objective control approach for quadrotors 545
as PSO, GA, Cuckoo Search (CS), and deep learning algo- Accepting Algorithm is explained in Section 3. The multi-
rithms were extensively employed for different autonomous objective optimization mechanism is demonstrated in Section 4.
systems [24,26–27]. Sections 5 and 6 present the validation results and analyze the
Herein, a unique Multi-Objective Non-Linear Threshold robustness of the proposed solution. Finally, some concluding
accepting Algorithm (MONLTA) [42] is adopted to determine remarks are highlighted in Section 7.
the near-optimal gains of a UAV control scheme. This algo-
rithm compromises between the desired overshoot percentage, 2. Mathematical modeling and control strategy
rise time, and settling time characteristics. NLTA is employed
to solve some NP hard problems with one objective function In this section, the mathematical modeling of the 6-Degrees-of-
and tune the PID control gains of voltage/frequency control Freedom (6DOF) quadrotor is presented. Furthermore, the
loops in power system applications [43,44]. Herein, the MON- underlying PID control strategy is elaborated.
LTA will be used for the first time to design the control loops
for a quadrotor. The paper applies a multi-objective optimiza- 2.1. Mathematical modeling
tion approach based on a nonlinear accepting function to
decide the different control strategies of a quadrotor for the
first time, to the best of our knowledge. The control scheme The adequate modeling of a quadrotor implies a precise con-
employs six concurrent and interactive loops. Herein, multi- trol design and ensures an optimized performance. In this
objective criteria based on Integration of Absolute Error research, the quadrotor has 6-DOF as shown in Fig. 1. Najm
(IAE) and Integration of Time Square Errors (ITSE) will be and Ibraheem [45], developed a nonlinear mathematical model
used to tune the control strategies (i.e., PID control gains of for a 6-DOF quadrotor system. Their dynamical model consid-
the different loops). The developed approach is compared with ered the acceleration and velocity vectors. It resulted in an
a nonlinear PID control mechanism dictated by an analytical accurate nonlinear dynamical model for the 6-DOF quadrotor
process. Furthermore, a comparative analysis is conducted to system. They derived the nonlinear model using the Euler-
find the best tuning outcomes attempting different metaheuris- Newton formulation to represent the 3D motion of the rigid
tic approaches. This analysis tests the robustness of the pro- body. The adopted model and simulation results are consid-
posed approach under harsh disturbances and unstructured ered to validate the efficiency of the proposed multi-objective
dynamical environment. Hence, different trajectory-tracking control mechanism. The axes and angles used to develop the
scenarios are considered (a single individual trace for each mathematical model are shown in Fig. 1.
degree of freedom, diamond, spiral, and helical paths) to test Assumptions introduced in the Model are as follows:
the effectiveness of MONLTA. So, the paper introduces excel-
lent control and tracking technique without complicity like the 1. The Quadrotor Is a Rigid Body.
nonlinear PID controller and the other complicated con-
trollers. It was founded that the quadrotor follows the desired
trajectory with a small tracking error of less than 2% in the X-
Y plane and less than 1 % error tracking error in the altitude
Z. Also, it is recorded that MONLTA stabilizes the system fas-
ter than those achieved using GA and linear PID approaches.
This highlights the significance of the MONLTA optimization
method, taking into consideration that it is simpler than the
NLPID and its behavior has an overall small settling time
associated with a small overshoot. Table 1 Compares MON-
LTA with PID with similar controllers.
The paper is organized as follows; Section 2 briefly intro-
duces the mathematical modeling of a quadrotor along with
the underlying control strategy. The Nonlinear Threshold Fig. 1 Angles of rotation of 4-motor quadrotor.
546 S.A. Kouritem et al.
2. The mass and moments of inertia are constant and the Cen- 2.2. Control strategy
ter of gravity is at a fixed location (Except in tests that
include external disturbances or payload variation). The PID control strategy is widely adopted in the industrial
3. The only forces affecting the quadrotors are gravity and applications as well as the unmanned systems, especially in
propeller thrust. drones. The 6-DoF quadcopter mathematical model is highly
4. Propellers with an odd index rotate clockwise and others non-linear where four actuators are considered. This imposed
rotate counter-clockwise. challenges to achieve the control objectives. The overall con-
5. All propellers are on the same level and are pointing in the trol scheme employs two control levels. The lower level con-
same direction. tains four PID control loops one for the altitude (z), and
6. Aerodynamic drag and blade flapping are neglected due to three for each Euler direction (i.e., the roll, pitch, and yaw
small size and low airspeed. directions). The higher level contains two PID control loops
for the translations in the X and Y directions. The control deci-
The 6-DOF quadrotor has 4 motors that produce four sions in these loops rely on the desired velocity of the drone
angular speeds (X1, X2, X3, and X4). These speeds are com- and their outputs are directed to the pitch and roll control
bined to develop a force ft and torques ðsx ; sy ; andsz Þ as in loops as depicted from Fig. 3.
(1). Further, the force ft is used to control the altitude z, while
the torques are used to control the angles ð£; ½03B8; anduÞ as 2.3. The optimization goals
can be depicted from Fig. 2. The overall mathematical model is
described by (2) and (3) [46,47]. Hence, the measurements
The objective of the control problem is to regulate the
ðx; y; z; £; ½03B8; anduÞ are needed to regulate the 6-DOF
reference-tracking errors (i.e., follow the desired trajectories).
quadrotor system [46].
This involves a minimization and a tradeoff between main
ft ¼ b X21 þ X22 þ X23 þ X24 dynamic parameters such as the rise time (tr Þ, setting time
(ts Þ; peak time (tp Þ; and Maximum overshoot amplitude
sx ¼ bl X23 X21 (Mp Þ. A multi-objective optimization procedure is needed to
compromise between the performances of the different control
sy ¼ bl X24 X22 loops. Accordingly, this procedure decides the control gains of
the different control loops as shown in Fig. 3. A multi-
sz ¼ d X22 þ X24 X22 X23 ð1Þ objective mechanism based on a nonlinear threshold accepting
function is considered to select the control gains to regulate the
80 1 0 10 1
>
> x_ cðwÞcðhÞ ½cðwÞsð/ÞsðhÞ cðuÞsðwÞ ½sð/ÞsðwÞ þ cðuÞcðwÞsðhÞ u altitude, yaw, roll, and pitch directions of the quadrotor. This
>
> B y_ C ¼ B cðwÞsðhÞ ½cð/ÞcðwÞ þ sð/ÞsðwÞsðhÞ ½cð/ÞsðwÞsðhÞ cðwÞsð/Þ CB v C
>
> @ A @ A@ A; solution solves for a near optimal set of control gains simulta-
>
>
>
>
>
> z_ sðhÞ cðhÞsð/Þ cðhÞcð/Þ w neously for all degrees of freedom (i.e., find the best control
>
> 0 1 0 10 1 0 1 0 1
>
> u_ 0 r q u sðhÞ fwx
>
> gains for the interactive different control loops collectively).
>
> B C B CB C B C B C
>
> @ v_ A ¼ @ r 0 p A@ v A þ g@ sð/ÞsðhÞ A þ m1 @ fwy A;
>
> There are different multi-objective criteria one is to minimize
>
> _ p ð hÞc ð / Þ f
< w q 0 w c wzft
0 1 0 10 1 conflicting dynamic variables such as the rise-time tr ,
> /_ 1 sð/ÞtðhÞ cð/ÞtðhÞ p
>
> B _ C B0 settling-time ts , peak-time tp , and maximum percentage over-
>
>
> @hA¼@ cð/Þ sð/Þ C B C
A@ q A;
>
>
>
>
> w_ 0 sð/Þ cð/Þ
r shoot Mp . Hence, the goal of this optimization form is to min-
>
>
cð hÞ cð hÞ
>
>
> 0 1 0 Iy Iz 10 1 0 sy þswx 1 imize the tuple (tr , ts , tp , Mp ). The second minimizes together
>
> p_ rq
>
> B C B
I
Iy Iz C
x
B x C
J
>
> @ q_ A ¼ B CB C B sy þswy C the Integration of Absolute Error (IAE)
>
>
> @ Iy A@ pr A þ @ Jy A: Rt
>
: r_ Ix Iy pq sz þswz (IAE = 0 f jerrorðtÞjdt; errorðtÞis the difference error between
Iz Jz
ð2Þ the reference and measured signals of the PID control loops
0 1 0 1 0 1 and tf is the final time of simulation) and Integral Time Square
Rt
x_ sð/ÞsðwÞ þ cð/ÞcðwÞsðhÞ 0 Error (ITSE) (ITSE = 0 f ðt ðerrorðtÞÞ2 Þdt) of the interacting
B C ft B C B C
@ y_ A ¼ @ cð/Þcð/ÞsðhÞ cðwÞsð/Þ A þ @ 0 A ð3Þ PID control loops using MONLTA and GA optimization
m
z_ cð/ÞcðhÞ g algorithms. Thus, the aim of this optimization problem is to
minimize the control performance given the tuple defined by 3.1. Nonlinear threshold accepting algorithm
(IAE and ITSE).
The NLTA heuristic approach is developed to provide solu-
3. Nonlinear threshold accepting mechanism tions for some single-objective NP-Hard problems [44]. The
multi-objective procedure decides the gains of the different
In the sequel, the self-adjusting optimization solution is control loops of the quadrotor by evaluating multiple objective
detailed out. This solution is implemented using the Nonlinear functions simultaneously. The underlying heuristic approach
Threshold Accepting Algorithm (NLTA). uses a nonlinear accepting function that takes the magnitude
form of a Low Pass Filter (LPF). Hence, this can be repre- 4. Multi-objective optimization
sented by a transfer function H such that
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 This section introduces the structure of the multi-objective
l optimization solution along with its computational structure.
HðlÞ ¼ 1 þ
l0
4.1. Multi-objective optimization: a structure
where the nominal and angular frequencies of the LPF are
denoted by l and l0 , respectively.
This function provide means to decide the best solution and The multi-objective optimization solution employs a variable-
at the same time explore new positions or solutions. The opti- length archive to temporarily memorize acceptable dominated
mization process optimizes an overall performance index func- sets as well as the non-dominated solutions. The length of the
tion c that compromises all the subobjectives together to search
for better solutions. A solution is defined to be a combination
of all the PID gains of the control loops. The optimization
scheme compares between a current solution Cs (i.e., the solu-
tion that is already accepted) with a neighboring solution Ns
(i.e., the next exploited or explored solution, given the feasible
space of the control gains). The solution Ns is accepted at either
one of two situations; First, an exploited solution is accepted
when it is found to be more valuable than the current solution
Cs (i.e., cðNÞ cðCs Þ). Second, an explored solution may be
accepted not fall in local minima and better represent the search
space (i.e., cðNs Þ HðlÞ cðCs Þ)). Fig. 4 shows a flow chart of
the NLTA solution mechanism.
Fig. 6 Altitude response using the MONLTA optimization
algorithm and classic tuning method.
Table 5 Tuning of the roll control loop using MONLTA after 15 episodes.
Episode kp ki kd tr ts (2%) tp Mp ISTE_£ IAE_£
1 3.161 0.191 2.785 2.602 6.860 4.760 1.043 5.286 27.754
2 3.089 0.250 1.335 0.849 8.022 2.020 1.033 1.727 18.335
3 2.646 0.107 1.196 0.971 3.180 2.760 1.018 1.560 17.206
4 2.820 0.096 1.808 2.105 6.012 5.040 1.020 2.460 20.999
5 2.835 0.271 1.933 1.775 13.300 3.390 1.052 3.831 24.828
6 3.889 0.179 1.830 1.426 4.521 2.930 1.020 1.522 17.775
7 2.118 0.117 1.317 1.658 10.700 3.450 1.031 2.794 21.694
8 3.617 0.271 1.635 1.185 8.200 2.560 1.031 1.703 18.419
9 3.267 0.452 1.611 1.106 9.860 2.480 1.057 2.311 20.612
10 3.933 0.068 2.919 2.995 6.254 4.900 1.012 2.732 21.340
11 2.280 0.213 1.155 1.060 10.340 2.217 1.044 2.416 20.581
12 2.798 0.166 1.396 1.319 8.034 2.579 1.027 1.929 19.023
13 1.578 0.007 0.994 2.070 na na Na 2.235 17.907
14 3.966 0.291 1.607 1.048 6.555 2.355 1.027 1.430 17.192
15 3.817 0.006 1.478 1.700 na na Na 0.935 13.172
Table 6 The PID control gains of the 6DOF quadcopter using MONLTA and GA.
Gains MONLTA GA
Kp Ki Kd Kp Ki Kd
pitch and roll 3.696 0.214 1.934 2.592 0.966 1.056
yaw 14.953 0.003 4.430 11.590 0.043 4.994
x-y 0.013 2.402 0.075 0.306 2.759 0.081
z 8.161 5.490 17.475 9.998 2.845 7.051
A self-adjusting multi-objective control approach for quadrotors 551
Fig. 11 Trajectory tracking of the quadcopter using MONLTA (a) 2D diamond motion, (b) 3D spiral motion, (c) 3D exponential helix
motion.
6. Results and discussion The simulation results depicted by Fig. 10 revealed that, the
response of MONLTA has the fastest settling time compared
This section investigates the robustness of the MONLTA solu- with the GA or Linear PID control. The values used in
tion employing various challenging scenarios. The perfor- describing the response are the rise time, peak time, overshoot
mance of the MONLTA solution is challenged by imposing percentage, and settling time at both 2% and 5% steady-state
alterations in the dynamic parameters of the drone. Further, error bands. It is observed from the comparison that, the
the effect of the external disturbances and noise on the system NLPID has the smallest overshoot and the fastest rise time
performance will be tackled. In the simulations, the archive’s compared with the LPID and GA. The reason behind this
initial size is set to 10. The initial solutions (i.e., the PID con- behavior is that, the gains of the LPID controller remain
trol gains) are randomly generated using feasible initial guess.
The PID control gains are picked randomly from an interval
[0.1,50]. The simulations are performed using MATLAB-
Simulink on a server with a 20 GHz processor. Table 4 shows
the Quadrotor dynamic parameters used in this study [45].
unchanged for all error values while the gains of the NLPID The third reference trajectory is an exponential helix path.
are adjusted depending on the error values [45]. The structure The radius of the circular path is varying in the X-Y plane
of the NLPID is more complicated in terms of the design and from zero to 5 m as the altitude Z varies from zero to 4 m.
modeling. However, the response for the MONLTA is as com- Table 9 lists the functions employed to generate the reference
petitive or outperforms the NLPID approach at some simula- trajectory. The quadrotor is able to follow the reference trajec-
tions. This highlights the significance of the MONLTA tory with small offset tracking errors. The results show that,
optimization method, taking into consideration that it is sim- the tuning approach using MONLTA provides robust perfor-
pler than the NLPID and its behavior has an overall small set- mances to different reference trajectory-tracking commands.
tling time associated with a small overshoot.
Results shown in Figs. 8-10 emphasize that the behaviors of 6.4. Tracking with disturbances and measurements of noise
the 6-DOF quadrotor tuned using MONLTA and NLPID can
be stabilized faster than those achieved using GA and LPID The performance of the MONLTA tuned control loops is val-
approaches. The MONLTA solution can achieve the same idated after exposure to external disturbance and measurement
response or outperform the NLPID for some dynamic aspects. noise. The wind disturbances affecting the quadrotor can be
The reason for that is the number of episodes was used to simulated using torque disturbance of 0.1 N.m during 80 Sec-
achieve the minimum value of the functions ISTE and IAE onds. The tuned control loops using MONLTA are able to
which generate a PID’s gains and have a response like the com- reject the simulated wind disturbances as shown in Fig. 12.
plicated NLPID. It is observed that, the 6-DOF quadrotor reacts in controlled
fashion to the disturbances with a small overshoot. Further,
6.3. Trajectory tracking Gaussian noise with zero mean is imposed on the measured
signals to test the performance of the MONLTA-tuned control
The performance of the quadrotor adopting the tuned control loops. Fig. 13 shows the performance of the quadrotor under
gains using MONLTA solution is tested for different trajecto- this scenario.
ries, given zero initial conditions. Fig. 11 shows different tra-
jectories responses (diamond, spiral, and exponential helix). 6.5. Payload variation
The first trajectory takes the form of a diamond shape
where the vertices challenge the performance of the controllers The robustness of the proposed solution is tested by varying
with sudden changes. This reference trajectory is generated the payload of the quadrotor. This represents a situation where
using the state functions described in Table 7. It is noticed that
the 6-DOF quadrotor follows the desired trajectory with an
error of 5% at each vertex of the diamond trajectory. The
tracking errors at the different vertices are the same in the x-
y plane with no change in the attitude z. This is due to the sym-
metry in the X/Y control loops as discussed in Section 6.2.
The second trajectory takes a spiral form. In this case, the
quadrotor flies in a circular path with a varying altitude Z.
This reference trajectory is generated using the state functions
described in Table 8. The quadrotor follows the desired trajec-
tory with a small tracking error offset that varies between 0%
and 2% in the X-Y plane, given the variation in the altitude Z.
Further, the reference tracking error in the altitude Z does not Fig. 14 X-tracking response after payload drop.
exceed 1%.
the dynamics of the quadrotor can change drastically. The PID References
gains of the different control loops are kept fixed and the
weight of the quadrotor is forced to drop after 5 Seconds. [1] Y. Huang, M. Zhu, Z. Zheng, M. Feroskhan, Fixed-time
The reference command is a step X-motion command. autonomous shipboard landing control of a helicopter with
Fig. 14 shows the response of the quadcopter due to a 30% external disturbances, Aerosp. Sci. Technol. 84 (2019) 18–30,
wight-drop against the no-drop or nominal case. It is observed https://doi.org/10.1016/j.ast.2018.07.032.
that, the weight-drop along with changing the X-motion com- [2] A. Mohiuddin, T. Tarek, Y. Zweiri, D. Gan, A survey of single
and multi-UAV aerial manipulation, Unman. Syst. 08 (02)
mand after five seconds will lead to an increase in the over-
(2020) 119–147, https://doi.org/10.1142/s2301385020500089.
shoot when compared with the no drop situation. As the [3] L. Marconi, R. Naldi, Aggressive control of helicopters in
weight drop percentage increases the overshoot increases. As presence of parametric and dynamical uncertainties,
log as the quadrotor is stabilizable, the tuned control loops Mechatronics 18 (7) (2008) 381–389, https://doi.org/10.1016/j.
using MONLTA are able to regulate the variations in the mechatronics.2007.10.004.
payload. [4] G. Hoffmann, H. Huang, S. Waslander, C. Tomlin, Quadrotor
helicopter flight dynamics and control: theory and experiment,
7. Conclusion in: AIAA Guidance, Navigation and Control Conference and
Exhibit, (Guidance, Navigation, and Control and Co-located
Conferences: American Institute of Aeronautics and
A multi-objective optimization scheme that employs a nonlin- Astronautics, 2007.
ear threshold accepting function is adopted to control a non- [5] X. Zhang, X. Li, K. Wang, Y. Lu, A survey of modelling and
linear 6-DOF quadrotor. This approach is used to tune the identification of quadrotor robot, Abstr. Appl. Anal. (2014)
gains of 6 interacting control loops. The proposed solution is 320526, 2014/10/20 2014, doi: 10.1155/2014/320526.
validated against other control approaches based on genetic [6] I. Raptis, K. Valavanis, Linear and nonlinear control of
algorithm, linear PID control, and non-linear PID control small-scale unmanned helicopters (International series on
intelligent systems, control and automation: science and
schemes. The comparisons were conducted to minimize
engineering ; v. 45). Dordrecht ;: Springer Science+Business
multi-objective criteria such as rise time, settling time, and
Media B.V., 2011.
overshot percentage. Further, the integration of absolute error [7] C. Chi, X. Yan, R. Chen, P. Li, Analysis of low-speed height-
and integral time square error optimization criteria were con- velocity diagram of a variable-speed-rotor helicopter in one-
sidered as objective functions for the underlying optimization engine-failure, Aerosp. Sci. Technol. 91 (2019) 310–320, https://
problem. The simulation results revealed the effectiveness of doi.org/10.1016/j.ast.2019.05.003.
the MONLTA optimization solution under challenging [8] A. Budiyono, Advances in unmanned aerial vehicles
dynamic situations such as flight paths, payload variation, technologies, in: International symposium on intelligent
wind simulated torque disturbances, and measurements noise. unmanned system, 2008, pp. 1-13.
This solution is shown to outperform outperformed the genetic [9] M. H. Khalesi, H. Salarieh, M. Saadat Foumani, System
identification and robust attitude control of an unmanned
algorithm in terms of dynamic characteristics such as rise time,
helicopter using novel low-cost flight control system, in:
settling time, and the overshot. The structure of the NLPID is
Proceedings of the Institution of Mechanical Engineers, Part I:
more complicated in terms of the design and modeling. How- Journal of Systems and Control Engineering, vol. 234, no. 5, pp.
ever, the response for the MONLTA is as competitive or out- 634-645, 2020/05/01 2019, doi: 10.1177/0959651819869718.
performs the NLPID approach at some simulations. The [10] A. Budiyono, S.S. Wibowo, Optimal tracking controller design
quadrotor was recorded following the desired trajectory with for a small scale helicopter, J. Bionic Eng. 4 (4) (2007) 271–280,
a small tracking error of less than 2% in the X-Y plane and https://doi.org/10.1016/S1672-6529(07)60041-9.
less than 1 % in the altitude Z. Also, it is recorded that MON- [11] G. Scholz, G.F. Trommer, Model based control of a quadrotor
LTA stabilizes the system faster than those achieved using GA with tiltable rotors, Gyroscopy Navig. 7 (1) (2016) 72–81,
and linear PID approaches. It was founded that the 6-DOF https://doi.org/10.1134/S2075108716010120.
[12] A. Nemati, M. Kumar, Non-linear control of tilting-quadcopter
quadrotor follows the desired trajectory with an error of 5%
using feedback linearization based motion control, in: HRI
at each vertex of the diamond trajectory. From the simulation
2014, 2014.
results, it was recorded that MONLTA can overcome the sim- [13] P.N. Chikasha, C. Dube, Adaptive model predictive control of a
ulated wind disturbances of 0.1 N.m as a disturbance torque. quadrotor, IFAC-PapersOnLine 50 (2) (2017) 157–162, https://
doi.org/10.1016/j.ifacol.2017.12.029.
8. Future work [14] O. Halbe, M. Hajek, Robust helicopter sliding mode control for
enhanced handling and trajectory following, J. Guid. Control
Dynam. 43 (10) (2020) 1805–1821, https://doi.org/10.2514/1.
The MONLTA optimization of quadcopter flight control
G005183.
results may be compared with the linear quadratic controller [15] T. Dierks, S. Jagannathan, Output feedback control of a
(LQR) method based on the extended Kalman Filter (EKF) quadrotor UAV using neural networks, IEEE Trans. Neural
and the learning-based predictive control (LBMPC) method Netw. 21 (1) (2010) 50–66, https://doi.org/10.1109/
results. The numerical results of this investigation should be TNN.2009.2034145.
verified experimentally in future work. [16] B. Erginer, E. Altuğ, Design and implementation of a hybrid
fuzzy logic controller for a quadrotor VTOL vehicle, Int. J.
Declaration of Competing Interest ControlAutom. Syst. 10 (1) (2012) 61–70, https://doi.org/
10.1007/s12555-012-0107-0.
[17] K.Z. Meguenni, M. Tahar, M.R. Benhadria, Y. Bestaoui, Fuzzy
The authors declare that they have no known competing
integral sliding mode based on backstepping control synthesis
financial interests or personal relationships that could have for an autonomous helicopter, Proc. Inst. Mech. Eng. Part G: J.
appeared to influence the work reported in this paper.
556 S.A. Kouritem et al.
Aerospace Eng. 227 (5) (2012) 751–765, https://doi.org/10.1177/ arms, 2nd International Conference on Electrical, Computer,
0954410012442119. Communications and Mechatronics EngineeringICECCME’22
[18] S. Bouabdallah, R. Siegwart, Full control of a quadrotor, in: (2022/11/16).
2007 IEEE/RSJ International Conference on Intelligent Robots [34] S.A. Kouritem, W.A. Altabey, N. Nahas, M.I. Abouheaf,
and Systems, 29 Oct.-2 Nov. 2007 2007, pp. 153-158, doi: Simplified torque modelling for different planer robots size, 2nd
10.1109/IROS.2007.4399042. International Conference on Electrical, Computer,
[19] K. A. Danapalasingam, M. A. M. Basri, A. R. Husain, Robust Communications and Mechatronics EngineeringICECCME’22
chattering free backstepping sliding mode control strategy for (2022/11/16.).
autonomous quadrotor helicopter, 2014. [35] A.S. Elkhatem, S.N. Engin, Robust lqr and lqr-pi control
[20]
K.J. Aström, T. Hägglund, The future of PID control, Control strategies based on adaptive weighting matrix selection for a uav
Eng. Pract. 9 (11) (2001) 1163–1175, https://doi.org/10.1016/ position and attitude tracking control, Alex. Eng. J. 61 (8) (2022)
S0967-0661(01)00062-4. 6275–6292, https://doi.org/10.1016/j.aej.2021.11.057.
[21] S.A. Kouritem, M.M.Y.B. Elshabasy, Tailoring the panel [36] M. Idrissi, M. Salami, F. Annaz, Modelling, simulation and
inertial and elastic forces for the flutter and stability control of a novel structure varying Quadrotor, Aerosp. Sci.
characteristics enhancement using copper patches, Compos. Technol. 119 (2021) 107093.
Struct. 274 (2021), https://doi.org/10.1016/ [37] S. Huang, Y. Huang, C.A. Blazquez, C. Chen, Solving the
j.compstruct.2021.114311 114311. vehicle routing problem with drone for delivery services using an
[22] M.M.Y.B. Elshabasy, S.A. Kouritem, Thickening of optimally ant colony optimization algorithm, Adv. Eng. Inf. 51 (2022)
selected locations on panels subjected to unyawed flow for 101536.
substantial delay of the panel flutter, Alex. Eng. J. 59 (6) (2020) [38] F. Zeng, Z. Chen, J. Clarke, D. Goldsman, Nested vehicle
5031–5044, https://doi.org/10.1016/j.aej.2020.09.026. routing problem: Optimizing drone-truck surveillance
[23] K. Mohamed, H. Elgamal, S.A. Kouritem, An experimental operations, Transp. Res. C 139 (2022) 103645.
validation of a new shape optimization technique for [39] L. Yan, J.L. Webber, A. Mehbodniya, B. Moorthy, S. Sivamani,
piezoelectric harvesting cantilever beams, Alex. Eng. J. 60 (1) S. Nazir, M. Shabaz, Distributed optimization of heterogeneous
(2021) 1751–1766, https://doi.org/10.1016/j.aej.2020.11.024. UAV cluster PID controller based on machine learning,
[24] L. Sheng, W. Li, Optimization design by genetic algorithm Comput. Electr. Eng. 101 (2022) 108059.
controller for trajectory control of a 3-RRR parallel robot, [40] G. Sonugür, A Review of quadrotor UAV: Control and SLAM
Algorithms 11 (1) (2018) 7. https://www.mdpi.com/1999-4893/ methodologies ranging from conventional to innovative
11/1/7. approaches, Rob. Auton. Syst. 161 (2023) 104342.
[25] V. Gomez, N. Gomez, J. Rodas, E. Paiva, M. Saad, R. Gregor, [41] J. Xu, L. Wang, Y. Liu, et al, Finite-time prescribed
Pareto optimal PID tuning for Px4-based unmanned aerial performance optimal attitude control for quadrotor UAV,
vehicles by using a multi-objective particle swarm optimization App. Math. Model. 120 (2023) 752–768.
algorithm, Aerospace 7 (6) (2020) 71. https://www.mdpi.com/ [42] N. Nahas, M. Abouheaf, M.N. Darghouth, A. Sharaf, A multi-
2226-4310/7/6/71. objective AVR-LFC optimization scheme for multi-area power
[26] S.A. Kouritem, W.A. Altabey, Ultra-broadband natural systems, Electr. Pow. Syst. Res. 200 (2021), https://doi.org/
frequency using automatic resonance tuning of energy 10.1016/j.epsr.2021.107467 107467.
harvester and deep learning algorithms, Energ. Conver. [43] N. Nahas, M. Abouheaf, A. Sharaf, W. Gueaieb, A self-
Manage. 272 (2022) 116332. adjusting adaptive AVR-LFC scheme for synchronous
[27] L. J. Mpanza, J. O. Pedro, Nature-inspired optimization generators, IEEE Trans. Power Syst. 34 (6) (2019) 5073–5075,
algorithms for sliding mode control parameters tuning for https://doi.org/10.1109/TPWRS.2019.2920782.
autonomous quadrotor, in: 2019 IEEE Conference on Control [44] N. Nahas, M. Nourelfath, Nonlinear threshold accepting meta-
Technology and Applications (CCTA), 19-21 Aug. 2019 2019, heuristic for combinatorial optimisation problems, Int. J.
pp. 1087-1092, doi: 10.1109/CCTA.2019.8920474. Metaheurist. 3 (4) (2014) 265–290.
[28] S. A. Kouritem, M. M. Elshabasy, H. A. El-Gamal, Optimum [45] A.A. Najm, I.K. Ibraheem, Nonlinear PID controller design for
location/area of PZT actuators for flutter damping using norm a 6-DOF UAV quadrotor system, Eng. Sci. Technol. Int. J. 22
feedback control gain-based iterative method, in: The 2015 (2019) 1087–1097.
World Congress on Advances in Structural Engineering and [46] F. Sabatino, Quadrotor control: modeling, nonlinear control
Mechanics (ASEM15), Incheon, South Korea, 2015. design, and simulation, Master’s thesis, Royal Institute of
[29] S. A. Kouritem, M. M. Elshabasy, H. A. El-Gamal, FE meshing Technology, 2015. <https://www.kth.se/polopoly_fs/1.588039!/
scheme for accurate placement/area of PZT actuators for flutter Thesis KTH - Francesco Sabatino.pdf>.
damping using LQR method, in: The 2015 World Congress on [47] R.W. Beard, Quadrotor Dynamics and Control, Brigham
Advances in Structural Engineering and Mechanics (ASEM15), Young University (June), 2008, pp. 1–47.
Incheon, South Korea, 2015, p. https://www.researchgate.net/ [48] S. Chatterjee, P. Banerjee, Hill-climbing approach for
publication. optimizing receiver bandwidth, in: International Conference on
[30] E. Barjuei, J. Ortiz, A comprehensive performance comparison of Electronics, Communication and Instrumentation (ICECI),
linear quadratic regulator (LQR) controller, model predictive India, 2014. doi:10.1109/ICECI32914.2014.
controller (MPC), H1 loop shaping and -synthesis on spatial [49] Goswami et al., Filter-Based Feature Selection Methods Using
compliant link-manipulators, Int. J. Dyn. Control 9 (2021) 121–140. Hill Climbing Approach, Natural Computing for Unsupervised
[31] J. Batista et al., Performance Comparison Between the PID and Learning, book, pp. 213–234.
LQR Controllers Applied to a Robotic Manipulator Joint, 978- [50] P. Corke, Robotics, Vision & Control, Springer, Second
1-7281-4878-6/19/$31.00 Ó2019 IEEE. Edition, 2017.
[32] S.A. Kouritem, M.I. Abouheaf, N. Nahas, M. Hassan, A multi [51] Z. He, L. Zhao, A simple attitude control of quadrotor
objective optimization design of industrial robot arms, Alex. helicopter based on ziegler-nichols rules for tuning PD
Eng. J. 61 (2022) 12847–12867. parameters, Sci. World J. (2014).
[33] S.A. Kouritem, W.A. Altabey, N. Nahas, M.I. Abouheaf, New
design of minimized torque and actuators for industrial robot