Nonlinear Model Predictive Control of

Tiltrotor Quadrotors with Feasible Control

arXiv:2406.06130v2 [cs.RO] 21 Jun 2024

Allocation
Zeinab Shayan1 , Jann Cristobal1 , Mohammadreza Izadi1 , Amin
Yazdanshenas1 , Mehdi Naderi2 , and Reza Faieghi1,*
1
Autonomous Vehicles Laboratory, Department of Aerospace Engineering,
Toronto Metropolitan University, 350 Victoria St., M5B2K3, Toronto, Ontario,
Canada
2 School of Production Engineering and Management, Technical University of

Crete, Chania, Greece

* Corresponding author: Email: [email protected]

Abstract
This paper presents a new flight control framework for tilt-rotor
multirotor uncrewed aerial vehicles (MRUAVs). Tiltrotor designs of-
fer full actuation but introduce complexity in control allocation due
to actuator redundancy. We propose a new approach where the allo-
cator is tightly coupled with the controller, ensuring that the control
signals generated by the controller are feasible within the vehicle actu-
ation space. We leverage nonlinear model predictive control (NMPC)
to implement the above framework, providing feasible control signals
and optimizing performance. This unified control structure simultane-
ously manages both position and attitude, which eliminates the need
for cascaded position and attitude control loops. Extensive numerical
experiments demonstrate that our approach significantly outperforms
conventional techniques that are based on linear quadratic regulator
(LQR) and sliding mode control (SMC), especially in high-acceleration
trajectories and disturbance rejection scenarios, making the proposed
approach a viable option for enhanced control precision and robust-
ness, particularly in challenging missions.

1
 Keywords: Uncrewed aerial vehicles, quadrotor, omnidirectional
quadrotor, nonlinear model predictive control, control allocation

1. Introduction
In conventional quadrotors and similar multirotor uncrewed aerial vehicles
(MRUAVs), all rotors are configured in the same or parallel planes, making
them structurally simple [1, 2, 3]; however, this results in under-actuation.
Each vehicle has six degrees of freedom (DOFs) to be controlled, but its
control inputs are limited to the moments around the three axes of the body
frame, and the thrust that is only alongside the vertical axis. As a result,
the longitudinal and lateral dynamics are coupled with rotational dynamics,
making it difficult to perform maneuvers that require the vehicle to control
its position or orientation independently from one another.
Tiltrotor MRUAVs address the above challenge by incorporating extra
actuation mechanisms that dynamically tilt the rotors’ planes, allowing the
vehicle to independently control six DOFs. Such vehicles can hover at a
fixed position while maintaining a constant non-zero attitude, a capability
that conventional MRUAVs cannot achieve [4, 5, 6, 7, 8]. Tiltrotor MRUAVs
are divided into two categories: single-axis [9, 10] and dual-axis [11, 12, 13]
tiltrotor. In the single-axis setup, each rotor tilts around only one axis of the
body frame, while in the dual-axis, rotors tilt around two axes.
The addition of tilting mechanisms often leads to redundancy in MRUAV
actuators. Therefore, an important question that arises is how to optimally
distribute the desired control effort among the actuators, a topic that is
widely studied in the field of control allocation [14] where the control system
typically consists of a controller and an allocator. The controller generates
commands, otherwise known as virtual control inputs, for the allocator. Next,
the allocator translates the virtual control inputs into specific commands for
each actuator, attempting to construct the desired control effort prescribed
by the controller.
A particular challenge to this setup is that the virtual control inputs com-
puted by the controller may not be feasible for the actuators. Our strategy
to address this is to tightly couple the controller and allocator, incorporating
the actuator constraints directly in the controller formulation, and generating
virtual control inputs that are feasible for the actuators.
Driven by the above factors, we aim to design a tightly coupled controller-

2
allocator for flight control of tiltrotor MRUAVs. The developed algorithm
generates control signals that are feasible in the actuation space of the vehi-
cle, optimizing the distribution of control efforts among the actuators, and
ultimately, enhancing the maneuverability of the vehicle. In our approach,
we leverage nonlinear model predictive control (NMPC) formulation, as it
allows us to account for the actuator constraints explicitly. Despite the full-
actuation of tiltrotor MRUAVs, many existing methods still rely on the sepa-
ration of position and attitude control, using cascaded control architectures.
In the proposed strategy, instead of using the cascaded architecture, we solve
6-DOF flight control in one NMPC computing frame. This enables optimiz-
ing for position and attitude control at the same time, while simplifying the
control architecture, and potentially reducing calibration efforts and compu-
tational load. The key features of our method include:

• A tightly coupled controller-allocator framework for tiltrotor MRUAVs

that ensures feasible control inputs.

• NMPC-based flight controller for tiltrotor MRUAVs, for the first time.

• Eliminating the cascaded control architecture for tiltrotor MRUAVs, si-

multaneously optimizing position and attitude control.

The above features enable more efficient and effective flight control for tiltro-
tor MRUAVs compared to the existing results, as will be shown in our com-
parative studies in Section 5.

2. Related Work
2.1. Flight Controllers for Tiltrotor MRUAVs
Flight control of tiltrotor MRUAVs has gained increasing interest in recent
years. Notable works include linear quadratic regulator (LQR) with an inte-
gral action controller [15] and linear model predictive control (LMPC) [16].
The latter intends to design a generic control framework for conventional and
tiltrotor MRUAVs; thus, proposes a cascaded control structure that is not
necessary for fully actuated MRUAVs. The methods developed in [15, 16]
provide optimal performance in simple maneuvers. However, since they rely
on a linearized model of the vehicle, they struggle in aggressive maneuvers or

3
in the presence of significant external disturbances, as the linear model may
not accurately capture the vehicle dynamics in such scenarios.
Furthermore, several studies have explored nonlinear control methods.
Sliding mode control (SMC) has been a popular choice, perhaps due to its
simplicity and robustness, being applied for controlling tiltrotor quadrotors
[17, 18] and trirotors [19, 20]. Other nonlinear methods include feedback
linearization [21, 9], adaptive control [22, 23], and backstepping [24]. The
main drawback of these approaches is the loose coupling between the con-
troller and the allocator. The controller can generate virtual control inputs
that are infeasible for the actuators. This leads to unwanted actuator satu-
ration which may lead to performance degradation, as will be shown in our
numerical experiments, or even instability in severe conditions.
In addition to these theoretical advancements, significant progress has
been made in enabling tiltrotor MRUAV control with open-source autopilot
programs. For instance, in [8, 25], a modified version of the PX4 Autopilot
was developed to implement flight control strategies on real-world single-axis
tiltrotor quadrotors.
Of note, despite the recent advances in NMPC for MARUV control [26,
27, 28, 29], we did not find a study that has explored the application of
NMPC for tiltrotor MRUAVs. Our NMPC implementation here addresses
the drawbacks of the aforementioned linear optimal and nonlinear methods,
at the cost of computational load. However, using efficient solvers such as
ACADOS [30] alongside our relatively simple control architecture ensures
real-time performance for our NMPC approach.

2.2. Control Allocation

Control allocation is an important part of flight controllers for tiltrotor MRU-
AVs. A comprehensive survey of primary control allocation algorithms is
available in [14]. The simplest algorithm is the pseudo-inverse method. This
involves building a control effectiveness matrix that maps control inputs to
virtual control inputs and computing the pseudo-inverse of that matrix to
find the inverse map needed for control allocation [14]. This method has
proven effective in many applications, especially when the virtual control in-
put is feasible within the system actuation space [31, 32]. However, if the
control effectiveness matrix is ill-conditioned or singular, the pseudo-inverse
may not provide a reliable solution. Moreover, the pseudo-inverse on its own
does not incorporate a mechanism to monitor or prevent actuator saturation.

4
There exist iterative and constrained optimization methods that address the
above concerns, e.g., redistributed pseudo-inverse [33], daisy chain [34], direct
allocation [35], linear programming [36], and quadratic programming [37].
Existing studies in tiltrotor MRUAV control have used the above control
allocation algorithms interchangeably. In this work, since we ensure the
virtual control inputs are feasible within the vehicle actuation space, we use
the simple pseudo-inverse approach; however, it can be readily switched with
more sophisticated algorithms.

3. Modeling
The key difference between the flight dynamics of tiltrotor MRUAVs and
conventional ones is how propulsive forces and moments are generated. Since
our focus is on single-axis tiltrotor quadrotors, we present the flight dynamics
model for this particular class of tiltrotor MRUAVs; however, our modeling
approach can be generalized to dual-axes tiltrotor MRUAVs with an arbitrary
number of rotors.
Let I = {xI , yI , zI } denote the inertial coordinates frame, B = {xB , yB , zB }
the body frame, as shown in Fig. 1. Let ξ = [x, y, z]T denote the vehicle posi-
tion, and η = [ϕ, θ, φ]T denote the attitude, where −π < ϕ ≤ π, − π2 ≤ θ ≤ π2 ,
and −π < ψ ≤ π are the Euler angles representing roll, pitch, and yaw in
the yaw-pitch-roll sequence. Let υ = [u, v, w]T denote the linear velocity,
ω = [p, q, r]T the angular velocity, and I RB the rotation matrix from B to I.
The transnational dynamics take the following form
1 B
(1)


ξ̈ = g + I R B f P − AT υ + f D ,
m
where g = [0, 0, 9.81]T m/s2 is the gravity vector, m is the vehicle mass,
AT is the transnational drag coefficient, and fPB and fD represent propulsive
and disturbance forces. The superscript B is to highlight fP is expressed in
B. This will become useful in the formulation of the propulsive forces and
moments to be detailed shortly.
The rotational dynamics of the vehicle take the following form
η̇ = H (η) ω,
−1 (2)
−ω × Jω + τ BP − AR ω + τ d ,


ω̇ = J
where × indicates the cross product, J is the vehicle inertia matrix, AR is the
rotational drag coefficient, τ BP and τ D represent propulsive and disturbance

5
 𝑇2 𝐳R3
𝑇2 cos(𝛽2′ )
𝑇3 cos(𝛽3 )
𝐲R2 𝛽2′ 𝐲R3 𝐱 R3
𝑇2 sin(𝛽2′ )
𝛽3 𝑇3
𝛽2
𝛽2′ 𝛽3

𝑙 𝐱B
𝐱 R2 𝑇3 sin(𝛽3 )
𝐳R2
𝑇4
𝐳R4 𝐲B 𝑇1 cos(𝛽1′ )
𝑇4 cos(𝛽4 )
𝑇4 sin(𝛽4 ) 𝛽4 𝛽1′ 𝐱 R1
𝑇1
𝛽1 𝛽1′
𝛽4 𝐳B
𝑇1 sin(𝛽1′ ) 𝐲R1
𝐱 R4 𝐲R4 𝐱I

𝐲I 𝐳R1
𝐳I
Figure 1: Coordinates frames and parameters setup in our modeling approach for
single-axis tiltrotor quadrotor

moments, and
 
1 sin ϕ tan θ cos ϕ tan θ
H (η) =  0 cos ϕ − sin ϕ  . (3)
0 sin ϕ/cos θ cos ϕ/cos θ
To formulate the propulsive forces and moments fP and τ P , we start by
defining the i-th rotor coordinates frames Ri with the following orientation
with respect to B, also illustrated in Fig. 1,

B RR1 = diag (1, 1, 1) ,

B RR2 = diag (−1, −1, 1) ,
(4)
B RR3 = diag (1, −1, −1) ,
B RR4 = diag (−1, 1, −1) .
Next, we represent the thrust, torque, and tilting angle of the i-th rotor by

6
Ti , Qi , and βi . We note that
Ti ≈ kT Ω2i , Qi ≈ kQ Ω2i , (5)
where kT and kQ are rotor coefficients, and Ωi is the angular velocity of rotor
with opposite signs Ω1,2 > 0 and Ω3,4 < 0. The propulsive force due to Ti ,
denoted by fPi is expressed in Ri as follows
fPRi i = [Ti sin (βi ) , 0, Ti cos (βi )]T , i = 1, 2,
Ri T (6)
fPi = [0, −Ti sin (βi ) , Ti cos (βi )] , i = 3, 4.
Therefore, the total propulsive force expressed in B is
4
X
fPB = B
B RRi fi . (7)
i=1

To construct τ BP , let l represent the distance of rotor from the vehicle

centre of mass. Then, the moments due to the rotors’ thrusts and torques
take the following form
τPBx = l f1Bz − f2Bz ,

τPBy = l −f3Bz + f4Bz ,

k
τPBz = kQT −f1Bz − f2Bz
+ f3Bz + f4Bz


(8)
+l −f1Bx + f2Bx
+l f3By − f4By .

Having established fP and τ P , it is easy to verify that when Ti ̸= 0 and

βi ̸= 0, the vehicle can exert forces and moments in all its DOFs. This
eliminates the underactuation of conventional quadrotors. Therefore, it is no
longer necessary to control position and attitude in a cascaded architecture,
a feature that we will leverage in our control design in the next section.
It is noteworthy that our systematic approach in constructing the propul-
sive forces and moments can be easily extended to other vehicle configurations
with an arbitrary number of rotors and single-axis or dual-axes rotor tilting
mechanisms.

4. Flight Control
Figure 2 presents an overview of the proposed flight control framework. As
mentioned earlier, a key feature in our design is to eliminate the cascaded

7
 External
Disturbance
𝑑
𝑓𝑃𝑥 Ω1𝑐
Controller Ω 2𝑐
𝑓𝑃𝑦
Ω 3𝑐
prediction 𝑓𝑃𝑧 Ω 4𝑐
Reference Optimizer
𝜏𝑃𝜙 Allocator 𝛽1𝑐
Trajectory 𝛽2𝑐
control 𝜏𝑃𝜃
actions Nonlinear 𝛽3𝑐
𝜏𝑃𝜓 𝛽4𝑐
Model

𝑇
𝑝, 𝑞, 𝑟, 𝜙, 𝜃, 𝜓, 𝑢, 𝑣, 𝑤, 𝑥, 𝑦, 𝑧

Figure 2: Overview of the proposed flight control framework for single-axis tiltrotor
quadrotor

position and attitude control loops, capitalizing on the extended actuation

space of tiltrotor MRUAVs to simplify the control system, and optimize for
14
position and attitude control at the same time. Therefore, at each computing
frame, the controller generates fPc and τ Pc , collectively called virtual control
T
signals v = fPTc , τ TPc ∈ R6 . The subscript c indicates command signals.


The vehicle actuators include four rotors and four servomotors for tilting
the rotors. Therefore, a control allocation algorithm needs to map the virtual
control input v ∈ R6 to control input
u = [Ω1c , Ω2c , Ω3c , Ω4c , β1c , β2c , β3c , β4c ]T ∈ R8 . (9)
This is where the second key feature of our design comes to play. In
many existing designs, it is not guaranteed that the virtual control signals,
i.e. v, generated by the controller leads to a control signal vector u that is
feasible for the actuators. To ensure v is feasible within the vehicle actuation
space, we tightly couple the controller and allocator by accounting for the
physical constraints of actuators in the controller design. One suitable control
framework to achieve this is NMPC.
However, our NMPC formulation is different than existing NMPC for-
mulations for quadrotors [16, 28, 29]. The existing approaches apply MPC
in cascaded control architecture, often only for position control or attitude

8
control; however, our NMPC approach controls the vehicle position and at-
titude at the same time. Having only one controller for both position and
attitude control presents several advantages inclusive of simplified control
architecture, lower computational load, easier calibration, and also simulta-
neous optimization of position and attitude control.

4.1. Controller
T
Let us start with defining the state vector x = ξ T , η T , υ T , ω T . The goal

T
is to design the virtual control signal v = fPT , τ TP such that x reaches a


desired trajectory xd . To this end, we adopt the standard NMPC formulation

X−1
k+N
min ℓ (x (i) , v (i)) + Φ (x (k + N )) (10)
i=k

subject to
x (0) = x (k) , x (i + 1) = f (x (i) , v (i)) ,
(11)
x (i) ∈ X , v (i) ∈ V, ∀i ∈ {k, . . . , k + N − 1},
where ℓ (x (i) , v (i)) is the running cost, Φ (x (k + N )) is the terminal cost,
x (i + 1) = f (x (i) , v (i)) is the discretized transnational and rotational dy-
namics (1) and (2), X and V are the sets of allowable x and v, and N is the
prediction horizon.
For our implementations, we set
ℓ(x, v) = (x − xd )T Qx (x − xd ) + vT Qv v, (12)
where Qx ≥ 0 and Qv > 0, and Φ(x(k + N )) = 0.
A distinct feature of our approach is how the constraint v (i) ∈ V is set
up. Instead of explicitly imposing constraints on v, we map v into u and use
the physical constraints of actuators to find the admissible u. This ensures
that v computed from (10) and (11) is feasible for the allocator.
To elaborate, let umin and umax represent the lower and uppper bounds
of admissible u. At each sample time k, we use the current state of actuators
and their rate of change to update umin and umax , and set
umin (k) ≤ h (v) ≤ umax (k) , (13)
where h (·) is a nonlinear function that maps the virtual control input v to
control input u. The development of h (·) roots into the allocator discussed
below.

9
4.2. Allocator
T
The allocator receives the virtual control input v = fPTc , τ TPc ∈ R6 from


the controller and computes the control input u in (9).

To this end, let us probe the propulsive forces of each rotor expressed in
(6). It follows that fPiy = 0 for i = 1, 2, and fPix = 0 for i = 3, 4. Therefore,
the remaining components of propulsive forces that can be used for control
allocation is
h iT
u′ = fP1x , fP1z , fP2x , fP2z , fP3y , fP3z , fP4y , fP4z . (14)

Using (7) and (8), we can write

v = Bu′ , (15)

where  
1 0 1 0 0 0 0 0

 0 0 0 0 1 0 1 0 

0 1 0 1 0 1 0 1
(16)
 
B=
 0 l 0 −l 0 0 0 0
.

 
 0 0 0 0 0 −l 0 l 
−kQ −kQ k kQ
−l kT
l kT
l kQT −l kT

−1
Multiplying both sides of (15) with BT B yields

u′ = B† v, (17)

where B† is pseudo-inverse of B.
Once u′ is computed, Ωic and βic can be obtained from (5) and (6) as
follows r
|fP1 |
f 
P1x
Ω1c = 30
π kT
, β 1c = arctan fP 1
,
r z

|fP2 |
f 
P2x
Ω2c = 30
π kT
, β 2c = arctan fP 2
,
r z
(18)
|fP3 |
 fP 
30 3y
Ω3c = − π , β3c = − arctan fP ,
r kT 3z

|fP4 |
 fP 
Ω4c = − 30
π kT
, β 4c = − arctan fP
4y
,
4z

where |·| indicates the vector magnitude. Computing (18) will constitute the
control input u which can be applied to the vehicle. At the same time, the

10
combination of (17) and (18) forms h (·) to be used in the constraint (13)
of the NMPC computations, ensuring v that is fed to the allocator leads to
feasible u.

5. Results
This section presents numerical experiments to evaluate the effectiveness of
the proposed flight control algorithm. We compare our method with linear
quadratic regulator (LQR) and sliding mode control (SMC), both carefully
fine-tuned to achieve their best performance. Note that both methods control
the position and attitude in the same control loop.
The vehicle parameters include m = 0.468[kg], J = diag (4.856, 4.856, 8.801)×
10−3 [kgm2 ], l = 0.225[m], kT = 1.22 × 10−5 , kQ = 1.689 × 10−7 , AT =
diag (0.3, 0.3, 0.25) [kg/s], and AR = 0.2I3 [kg/s]. The physical constraints
correspond to a custom-built single-axis tiltrotor quadrotor in our group with
a maximum rotor rotational speed of 10, 000[rpm] and −45◦ ≤ βi ≤ 45◦ .
We selected the NMPC parameters as N = 5, Qv = 5 × 10−4 I6 , and
Qx = diag (0.04, 0.04, 0.04, 8, 8, 8, 1, 1, 4, 65, 65, 70).

Scenario 1: Sluggish lemniscate trajectory tracking

We begin with a straightforward mission: tracking a lemniscate trajectory
T
defined by ξ d = 4 sin 20πt
, sin 2πt and η d = [0, 0, 0]T in the absence



20
, −4
of external disturbances (Fig. 3). The maximum acceleration of this tra-
jectory is approximately 0.1[m/s2 ], making it a relatively simple task for all
controllers.
Figures 4 and 5 present the position and attitude tracking errors. All
controllers achieve small steady-state errors; however, only NMPC reaches
zero steady-state error in all states. Despite our efforts to fine-tune LQR, a
steady-state error in the z direction persisted. SMC suffers from a fluctuating
error in the x and y directions, and slow convergence to zd compared to
NMPC.
Figures 6 and 7 present the control inputs during this mission. While
different controllers exhibit distinct patterns, all controllers keep the control
inputs well within the physical limits of the actuators after a transient state.
While NMPC demonstrates superior tracking performance, the differ-
ences between the three methods are subtle, primarily because the desired

11
 1.5

0.5

-0.5

-1

-1.5
-5 -4 -3 -2 -1 0 1 2 3 4 5

Figure 3: Top view of vehicle position in sluggish trajectory tracking

trajectory did not require aggressive maneuvers that would push the vehicle
to its actuation limits. The advantages of NMPC become more pronounced
as the mission becomes more challenging, such as the one explained below.

Scenario 2: Agile lemniscate trajectory tracking

We increased the speed of the lemniscate trajectory, reaching to accelerations
around 5 m/s2 , and added a composite sinusoidal signal similar to the one
in [3] as an external disturbance (Fig. 8).
Figures 9 and 10 present the position and attitude tracking errors. Again,
NMPC exhibits superior performance, with the lowest magnitude of errors
across all states. LQR exhibits significantly larger errors in all states, and a
steady-state error in the z direction. SMC performs poorly in the z direction,
and its attitude tracking error is also much larger than that of the other
controllers.
NMPC’s superior performance is attributed to its ability to account for
the physical limitations of the actuators when generating virtual control sig-
nals. As shown in Figs. 11 and 12, the NMPC control inputs exhibit shorter
episodes of actuator saturation and less fluctuations in the βi signals, high-
lighting better control allocation that is achieved by the tight coupling of the
controller and allocator, ultimately, leading to improved performance.

12
 0.05

-0.05

0 5 10 15 20 25 30 35 40

0.02

0.01

-0.01

-0.02
0 5 10 15 20 25 30 35 40

0
0 5 10 15 20 25 30 35 40

Figure 4: Position error in sluggish trajectory tracking

Scenario 3: Attitude tracking in hover

One advantage of tiltrotor quadrotors over conventional ones is their ability
to control pitch and roll while holding a position. We explore the performance
of the controller in conducting such maneuvers. The vehicle starts from the
origin, flying to [4, 4, −4]T and holding position despite external disturbances.
Then, it tracks different attitude set-points as shown in Figs. 13–16.
Similar to the previous scenarios, NMPC achieves superior position and
attitude tracking errors in all states. Interestingly, this is accomplished with
smaller control inputs compared to the other controllers. The SMC control
input presents undesirable fluctuations that could be attributed to chattering
phenomenon.
Overall, these results highlight the superiority of our proposed algorithm

13
 0.6

0.4

0.2

0 5 10 15 20 25 30 35 40

0.05

-0.05

-0.1
0 5 10 15 20 25 30 35 40

0.5

-0.5

-1

-1.5
0 5 10 15 20 25 30 35 40

Figure 5: Attitude error in sluggish trajectory tracking

for flight control of tiltrotor MRUAVs compared to some of the common

existing approaches.

6. Conclusion
This paper presented a new flight control framework for tiltrotor MRU-
AVs. Our approach tightly couples control allocation with the controller
and leverages NMPC formulation to effectively handle actuator constraints
while maintaining low tracking error. Our comparative study demonstrated
the superiority of the proposed method over conventional techniques based
on LQR and SMC. This highlights the viability of our approach for enhanced
control precision and robustness, especially in challenging missions.

14
 10000

5000

0
0 0.5 1 1.5 2 2.5 3
10000

5000

0
0 0.5 1 1.5 2 2.5 3

-5000

-10000
0 0.5 1 1.5 2 2.5 3

-5000

-10000
0 0.5 1 1.5 2 2.5 3

Figure 6: Ωi in sluggish trajectory tracking

We utilized the simplest control allocation approach, the pseudo-inverse

method, yet our flight control algorithm excelled in all test scenarios. How-
ever, the use of more advanced control allocation algorithms could potentially
further enhance the algorithm performance. Some advanced control alloca-
tion algorithms require solving optimization problems in real-time. Given
the computational load of NMPC, integrating advanced control allocation
poses a challenge, opening up new research questions on how to effectively
implement them in resource-constrained vehicles and conduct hardware ex-
periments. This is an area for future exploration.
Overall, our findings underscore the potential of the proposed approach
as a robust and efficient control strategy for tiltrotor MRUAVs, paving the
way for more advanced and reliable autonomous flight operations.

15
 200

180

160

140
0 0.5 1 1.5 2 2.5 3

220

200

180

0 0.5 1 1.5 2 2.5 3

40

20

0 0.5 1 1.5 2 2.5 3

-20

-40
0 0.5 1 1.5 2 2.5 3

Figure 7: βi in sluggish trajectory tracking

References
[1] M. Maaruf, M. S. Mahmoud, A. Ma’arif, A survey of control methods
for quadrotor uav, Int. J. Robot. Control Syst. 2 (4) (2022) 652–665.

[2] R. Amin, L. Aijun, S. Shamshirband, A review of quadrotor uav: con-

trol methodologies and performance evaluation, Int. J. Autom. Control
10 (2) (2016) 87–103.

[3] M. Izadi, R. Faieghi, High-gain disturbance observer for robust trajec-

tory tracking of quadrotors, Control Eng. Pract. 145 (2024) 105854.

16
1.5

0.5

-0.5

-1

-1.5
-5 -4 -3 -2 -1 0 1 2 3 4 5

Figure 8: Top view of vehicle position in agile trajectory tracking

[4] Z.-Y. Lv, Y. Wu, Q. Zhao, X.-M. Sun, Design and control of a novel
coaxial tilt-rotor uav, IEEE Trans. Ind. Electron. 69 (4) (2021) 3810–
3821.

[5] W. Hao, B. Xian, T. Xie, Fault-tolerant position tracking control design

for a tilt tri-rotor unmanned aerial vehicle, IEEE Trans. Ind. Electron.
69 (1) (2021) 604–612.

[6] M. Hamandi, F. Usai, Q. Sablé, N. Staub, M. Tognon, A. Franchi, Design

of multirotor aerial vehicles: A taxonomy based on input allocation, Int.
J. Robot. Res. 40 (8-9) (2021) 1015–1044.

[7] P. Yu, An Over-Actuated Multi-Rotor Aerial Platform and Iterative

Learning Control Applications, Univ. California, Los Angeles, 2022.

[8] S. Marcellini, J. Cacace, V. Lippiello, A px4 integrated framework for

modeling and controlling multicopters with til table rotors, in: Proc.
2023 Int. Conf. Unmanned Aircr. Syst., IEEE, 2023, pp. 1089–1096.

[9] M. Ryll, H. H. Bülthoff, P. R. Giordano, A novel overactuated quadro-

tor uav: Modelling, control and experimental validation, IEEE Trans.
Control Syst. Technol. 23 (2) (2015) 510–556.

17
 1

0.5

-0.5

-1
0 5 10 15 20 25 30 35 40

0.4

0.2

-0.2

-0.4
0 5 10 15 20 25 30 35 40

-5

-10
0 5 10 15 20 25 30 35 40

Figure 9: Position error in agile trajectory tracking

[10] M. Kamel, S. Verling, O. Elkhatib, C. Sprecher, P. Wulkop, Z. Taylor,

R. Siegwart, I. Gilitschenski, The voliro omniorientational hexacopter:
An agile and maneuverable tiltable-rotor aerial vehicle, IEEE Robot.
Autom. Mag. 25 (4) (2018) 34–44.

[11] D. De Martini, G. V. Gramazio, A. Bertini, C. Rottenbacher,

T. Facchinetti, Design and modeling of a quadcopter with double axis
tilting rotors, Unmanned Syst. 5 (3) (2017) 169–180.

[12] A. Bin Junaid, A. Diaz De Cerio Sanchez, J. Betancor Bosch, N. Vitzi-

laios, Y. Zweiri, Design and implementation of a dual-axis tilting quad-
copter, Robot. 7 (4) (2018) 65.

18
 4

-2
0 5 10 15 20 25 30 35 40

-2

-4
0 5 10 15 20 25 30 35 40

-5
0 5 10 15 20 25 30 35 40

Figure 10: Attitude error in agile trajectory tracking

[13] N. Von Klemperer, Dual-axis tilting quadrotor aircraft: Dynamic mod-

elling and control of dual-axis tilting quadrotor aircraft, Master’s thesis,
Fac. Eng. Built Environ. (2018).

[14] T. A. Johansen, T. I. Fossen, Control allocation—a survey, Automatica

49 (5) (2013) 1087–1103.

[15] M. Allenspach, K. Bodie, M. Brunner, L. Rinsoz, Z. Taylor, M. Kamel,

R. Siegwart, J. Nieto, Design and optimal control of a tiltrotor micro-
aerial vehicle for efficient omnidirectional flight, Int. J. Robot. Res.
39 (10-11) (2020) 1305–1325.

19
 10000

5000

0
0 5 10 15 20 25 30 35 40

10000

5000

0
0 5 10 15 20 25 30 35 40

-5000

-10000
0 5 10 15 20 25 30 35 40

-5000

-10000
0 5 10 15 20 25 30 35 40

Figure 11: Ωi in agile trajectory tracking

[16] A. Eskandarpour, M. Mehrandezh, K. Gupta, A. Ramirez-Serrano,

M. Soltanshah, A constrained robust switching mpc structure for tilt-
rotor uav trajectory tracking problem, Nonlinear Dyn. 111 (18) (2023)
17247–17275.

[17] S. Sridhar, R. Kumar, G. Gupta, M. Kumar, K. Cohen, Nonlinear con-

trol of a novel class of tilt-rotor quadcopters using sliding mode method:
Theory and hardware implementation, J. Aerosp. Eng. 35 (3) (2022)
04022017.

[18] L. J. Mpanza, J. O. Pedro, J. Roberts, Control-allocated sliding mode

control for a single-axis tilting quadrotor uav, IFAC-PapersOnLine
54 (21) (2021) 204–209.

20
 220
200
180
160
140
0 5 10 15 20 25 30 35 40

220
200
180
160
140
0 5 10 15 20 25 30 35 40
50

-50
0 5 10 15 20 25 30 35 40
50

-50
0 5 10 15 20 25 30 35 40

Figure 12: βi agile trajectory tracking

[19] L. Yu, G. He, S. Zhao, X. Wang, L. Shen, Immersion and invariance-

based sliding mode attitude control of tilt tri-rotor uav in helicopter
mode, Int. J. Control Autom. Syst. 19 (2021) 722–735.

[20] Z. Yu, J. Zhang, X. Wang, Thrust vectoring control of a novel tilt-rotor

uav based on backstepping sliding model method, Sensors 23 (2) (2023)
574.

[21] A.-W. A. Saif, Feedback linearization control of quadrotor with tiltable

rotors under wind gusts, Int. J. Adv. Appl. Sci. 4 (10) (2017) 150–159.

[22] F. Şenkul, E. Altuğ, Adaptive control of a tilt-roll rotor quadrotor uav,

in: Proc. 2014 Int. Conf. Unmanned Aircr. Syst., IEEE, 2014, pp. 1132–

21
 4

0
0 5 10 15 20 25 30 35 40

0
0 5 10 15 20 25 30 35 40

-1

-2

-3

-4

0 5 10 15 20 25 30 35 40

Figure 13: Position in attitude tracking during hover

1137.

[23] C. Ding, L. Lu, A tilting-rotor unmanned aerial vehicle for enhanced

aerial locomotion and manipulation capabilities: Design, control, and
applications, IEEE/ASME Trans. Mechatronics 26 (4) (2020) 2237–
2248.

[24] A.-W. A. Saif, A. Aliyu, M. A. Dhaifallah, M. Elshafei, Decentralized

backstepping control of a quadrotor with tilted-rotor under wind gusts,
Int. J. Control Autom. Syst. 16 (2018) 2458–2472.

[25] S. D’Angelo, F. Pagano, F. Longobardi, F. Ruggiero, V. Lippiello,

Efficient development of model-based controllers in px4 firmware: A

22
 20

15

10

0
0 5 10 15 20 25 30 35 40

10

-5

-10
0 5 10 15 20 25 30 35 40

10

-5
0 5 10 15 20 25 30 35 40

Figure 14: Attitude in attitude tracking during hover

template-based customization approach, in: 2024 International Confer-

ence on Unmanned Aircraft Systems (ICUAS), IEEE, 2024, pp. 1155–
1162.

[26] S. Sun, A. Romero, P. Foehn, E. Kaufmann, D. Scaramuzza, A compar-

ative study of nonlinear mpc and differential-flatness-based control for
quadrotor agile flight, IEEE Trans. Robot. 38 (6) (2022) 3357–3373.

[27] F. Nan, S. Sun, P. Foehn, D. Scaramuzza, Nonlinear mpc for quadrotor

fault-tolerant control, IEEE Robot. Autom. Lett. 7 (2) (2022) 5047–
5054.

23
 10000

5000

0
0 5 10 15 20 25 30 35 40

10000

5000

0
0 5 10 15 20 25 30 35 40

-5000

-10000
0 5 10 15 20 25 30 35 40

-5000

-10000
0 5 10 15 20 25 30 35 40

Figure 15: Ωi in attitude tracking during hover

[28] D. Wang, Q. Pan, Y. Shi, J. Hu, C. Zhao, Efficient nonlinear model

predictive control for quadrotor trajectory tracking: Algorithms and
experiment, IEEE Trans. Cybern. 51 (10) (2021) 5057–5068.

[29] B. B. Carlos, T. Sartor, A. Zanelli, G. Frison, W. Burgard, M. Diehl,

G. Oriolo, An efficient real-time nmpc for quadrotor position control
under communication time-delay, in: Proc. 2020 16th Int. Conf. Control
Autom. Robot. Vision (ICARCV), IEEE, 2020, pp. 982–989.

[30] R. Verschueren, G. Frison, D. Kouzoupis, J. Frey, N. v. Duijkeren,

A. Zanelli, B. Novoselnik, T. Albin, R. Quirynen, M. Diehl, acados—a
modular open-source framework for fast embedded optimal control,
Math. Program. Comput. 14 (1) (2022) 147–183.

24
 220
200
180
160
140
0 5 10 15 20 25 30 35 40

220
200
180
160
140
0 5 10 15 20 25 30 35 40
50

-50
0 5 10 15 20 25 30 35 40
50

-50
0 5 10 15 20 25 30 35 40

Figure 16: βi in attitude tracking during hover

[31] M. Naderi, A. Khaki Sedigh, T. A. Johansen, Guaranteed feasible con-

trol allocation using model predictive control, Control Theory Technol.
17 (3) (2019) 252–264.

[32] M. Naderi, T. A. Johansen, A. K. Sedigh, A fault tolerant control scheme

using the feasible constrained control allocation strategy, Int. J. Autom.
Comput. 16 (2019) 628–643.

[33] J. Jin, Modified pseudoinverse redistribution methods for redundant

controls allocation, J. Guid. Control Dyn. 28 (5) (2005) 1076–1079.

[34] J. M. Buffington, D. F. Enns, Lyapunov stability analysis of daisy chain

control allocation, J. Guid. Control Dyn. 19 (6) (1996) 1226–1230.

25
[35] W. C. Durham, Constrained control allocation, J. Guid. Control Dyn.
16 (4) (1993) 717–725.

[36] M. Bodson, Evaluation of optimization methods for control allocation,

J. Guid. Control Dyn. 25 (4) (2002) 703–711.

[37] O. Harkegard, Efficient active set algorithms for solving constrained least
squares problems in aircraft control allocation, in: Proc. 41st IEEE Conf.
Decis. Control, Vol. 2, IEEE, 2002, pp. 1295–1300.

26