This article has been accepted for publication in a future issue of this journal, but has not been
fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access
IEEE ACCESS
Nonlinear Robust Flight Mode Transition Control
for Tail-Sitter Aircraft
Zhaoying Li, Wenjie Zhou, Hao Liu, Member, IEEE, Lixin Zhang, and Zongyu Zuo, Senior Member, IEEE
Abstract—In this paper, the nonlinear robust control problem
for tail-sitter aircraft in flight mode transitions is addressed. The
problem is challenging due to the nonlinearities and uncertainties
including parametric uncertainties, unmodeled uncertainties, and
external disturbances involved in the vehicle dynamics during
the mode transitions. The proposed controller is designed in
two steps: first, the dynamic inversion technique is adopted to
establish a tracking error model; then, for the tracking error
model, a robust controller consisting of a nominal controller
and a robust compensator is designed. The nominal controller
is designed to obtain the desired tracking performances for the
nominal system. The compensator is introduced to restrain the
effects of uncertainties. Tracking errors of the closed-loop system
are proven to converge into a neighborhood of the origin in a
finite time. Simulation results are presented to demonstrate the
advantages of the proposed controller, compared to the controller
based on the standard loop-shaping method.
Index Terms—Tail-sitter aircraft, robust control, nonlinear
control, uncertainties.
I. I NTRODUCTION
O VER the past decades, unmanned aerial vehicles (UAVs)
have drawn considerable attention due to their wide
applications, such as search and rescue, reconnaissance, mete-
orological monitoring, and wildfire tracking, as shown in [1].
UAVs are typically classified as helicopters [2], [3], fixed-wing
aircraft [4], [5], and tail-sitter aircraft [6], [7]. The tail-sitter
vehicles are introduced due to their advantages of combining
the vertical take-off and landing capability of the helicopters
and the high forward flight speed of fixed-wing aircraft, as
shown in [8]. Therefore, much interest has been aroused in
the automatic control circle for the tail-sitter aircraft. Several
tail-sitter aircraft flight controllers have been implemented
successfully in recent years, as depicted in [9]–[13].
Many researches have been done in the past decades for de-
signing flight mode transition controllers. In [14], a transition
controller including a proportional-integral-derivative (PID)
This work was supported by the National Natural Science Foundation of
China under Grants 61873012, 61503012, and 61673034, and the Fundamental
Research Funds for the Central Universities under Grants YWF-18-BJ-Y-81
and YWF-17-BJ-Y-86.
Z. Li is with School of Astronautics, Beihang University, Beijing, 100191,
P.R. China (e-mail: lizhaoying@buaa.edu.cn)
W. Zhou is with School of Astronautics, Beihang University, Beijing,
100191, P.R. China (e-mail: zwjerry@buaa.edu.cn)
H. Liu is with the School of Astronautics, Beihang University, Beijing
100191, P.R. China, and also with the University of Texas at Arlington
Research Institute, University of Texas at Arlington, Fort Worth, Texas 76118,
USA (e-mail: liuhao13@buaa.edu.cn)
L. Zhang is with School of Astronautics, Beihang University, Beijing,
100191, P.R. China (e-mail: buaazlx@buaa.edu.cn)
Z. Zuo is with Department of Seventh Research Division, Beihang Univer-
sity, Beijing, 100191, P.R. China (e-mail: zzybobby@buaa.edu.cn)
2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1
feedback controller, a feedforward controller, and controllers
based on the gain scheduling approach was designed to achieve
the desired tracking performances. The robust controller based
on state estimator was present in [15] of nonlinear systems
with parametric uncertainties and noisy outputs for quadrootor.
A tracking controller based on smooth function was introduced
in [16] for autonomous tail-sitter aircraft with bounded in-
puts. In [17], a nonlinear double integrator was introduced
to obtain desired tracking performance by the acceleration
measurements. In [18], an integrated altitude control design for
a tail-sitter UAV equipped with turbine engines was introduced
in achieving quasistationary flight. An observer was designed
to reconstruct system state variables based on the system
model and delayed outputs, as depicted in [19]. A controller
based on the Pade approximation technique and full order
observers was designed in [20]. Nonlinear controllers based
on delayed-output observers were proposed in [21] to track a
given reference trajectory. The influences of uncertainties were
not fully considered in the stability analysis by the above-
mentioned control methods.
In fact, the tail-sitter aircraft flight mode transition controller
design is challenging due to the nonlinearities and uncer-
tainties including parametric uncertainties, unmodeled uncer-
tainties, and external disturbances. Therefore, robust control
approaches have been studied to restrict the influences of
these uncertainties. PID control method was adopted in [22] to
increase stability against large disturbances. In [23], the gain
scheduled linear quadratic regulator (LQR) was adopted to
design controllers under external disturbances. Three different
transition control strategies were studied in [24] to realize the
optimal transition. A control strategy based on the backstep-
ping method was presented to perform the transition of a tail-
sitter aircraft from vertical to forward flight mode in [25]. In
these works, the effects of the external disturbances caused
by wind and aerodynamic perturbations have been explored,
but further studies on rejecting the influences of uncertainties
were not sufficiently investigated.
Furthermore, studies have been done to restrict the influ-
ences of the uncertainties on the closed-loop system for the
tail-sitter aircraft. In [26], a nonlinear controller was intro-
duced with finite-time convergent observer based on Lyapunov
function to estimate the unknown nonlinearities, the uncertain-
ties, and external disturbances during mode transition. Unmod-
eled uncertainties and external aerodynamic disturbances were
considered, but the influences of parametric uncertainties were
not discussed fully in [26].
This paper aims to propose a nonlinear robust controller
for flight mode transition control of the tail-sitter aircraft.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access
IEEE ACCESS
Fig. 1. X-hound aircraft
The proposed controller is composite of a nonlinear controller
based on the dynamic inversion technique, a linear feedback
controller, and a robust compensator. The dynamic inversion
technique is firstly utilized to generate a linear error model
of the tail-sitter aircraft. The linear error model is considered
as a nominal model with equivalent disturbances. Then, based
on the tracking error model, a robust controller consisting of
a nominal controller and a robust compensator is designed.
The nominal controller based on the eigenvalue assignment
approach is designed to obtain the desired tracking perfor-
mances for the nominal system. The compensator is introduced
to restrain the effects of parametric uncertainties, unmodeled
uncertainties, and external disturbances.
Compared with previous studies, the proposed control
method can restrain the effects of equivalent disturbances
including parametric uncertainties, unmodeled uncertainties,
and external disturbances. In this current paper, it is proven
that if the initial conditions of the tracking states are bounded,
then the tracking errors are bounded and can converge to a
neighborhood of the origin in a finite time. In addition, the
eigenvalue assignment method leads to linear time-invariant
state feedback controllers, and it is comparatively easy to be
employed in practical applications. In the current paper, a
time-invariant robust controller is proposed for a time-varying
nonlinear system. Furthermore, it do not need to switch the
controller structure or the controller parameters for different
flight status in practical applications. Therefore, the designed
controller is comparatively easy to be implemented in practical
applications. The simulation results show the advantages of the
proposed method for improving tracking performances of the
closed-loop system, compared to the loop-shaping controller.
This paper is organized as follows: the dynamic model
of the tail-sitter is shown in Section II; in Section III, the
nonlinear controller based on dynamic inversion technique is
synthesized; in Section VI, the robust stability and the tracking
properties are proven; simulation results are shown in Section
V; in Section VI, conclusions are drawn.
II. TAIL - SITTER AIRCRAFT MODEL
The tail-sitter vertical take-off and landing unmanned aerial
vehicle studied in this paper is developed by AOS Company
called ”X-hound”, as shown in Fig. 1. The mode transition
from hovering to level flight is depicted in Fig. 2. In this sec-
tion, firstly, the schematic of the tail-sitter aircraft is described.
Secondly, the coordinates and frames have been defined. Then,
2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
2
Fig. 2. Mode transition from hovering to level
the mathematical equations of motion of the tail-sitter aircraft
have been given.
A. Schematic of the tail-sitter aircraft
For the tail-sitter UAV, it can be divided into three flight
modes: the fly forward state, the hover state, and the mode
transition. In the hover state, the tail-sitter aircraft can be
regarded as a special quadrotor, and the aerodynamic analysis
can be implemented based on the body fixed frame. In the fly
forward state, the tail-sitter aircraft is similar to a general fixed-
wing aircraft. The transition mode is similar to both the hover
state and the fly forward state in some sense, and it is difficult
to analyze the aerodynamic forces and torques. In this case, the
body fixed frame is chosen to express the aerodynamic forces
and torques in the mode transition. Therefore, to facilitate
analytical calculations, the aerodynamic forces are established
in the body coordinate system. The schematic of the tail-sitter
aircraft is depicted in Fig. 3. The tail-sitter aircraft consists of
fuselage, two wings and V-tails as shown in Fig. 4. Thrust is
generated by the propellers driven by the four motors mounted
on the aircraft wings and the V-tails. Rotor 1 and Rotor 3
rotate counterclockwise, while Rotor 2 and Rotator 4 rotate
clockwise. The pitch movement is achieved by the difference
between the torques generated by the rotors (Rotor 1 and Rotor
2) mounted on the wings and the rotors (Rotor 3 and Rotor 4)
on the V-tails. Vanes installing on wings and V-tails are utilized
to generate yaw and roll movements by the deflections.
B. Definition of the coordinates and frames
© ª
In Fig. 4, E = E x © E y Ez denotes
ª an earth
fixed inertial frame. Eb = Exb Eyb Ezb defines a
body fixed frame £and C is the ¤T mass center of the tail-sitter
aircraft. Let P = x y z be the position of the mass
center
£ of the tail-sitter
¤T aircraft relative to the inertial frame,
v = vx vy vz denote the velocity of the mass center
£ ¤T
relative to the inertial frame, and ωb = ωbx ωby ωbz
represent the angular rates on each axis relative to the body
frame. The symbol cθ is used for cos θ and sθ for sin θ, the
rotation matrix from the body frame to the inertial frame is
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access

IEEE ACCESS 3

mv̇ = Ft − mgRbe , (1)

Jt ω̇b + ωb× Jt ωb = Tt , (2)

where m denotes the total mass of the tail-sitter aircraft,

g is the gravity constant, Jt indicates £ the inertial ¤matrix
T
of the tail-sitter aircraft, and Ft = Fx Fy Fz and
£ ¤T
T t = Tx T y T z are the total force and torque on
the aircraft in the body frame, respectively. The cross-product
operation is a skew-symmetric matrix and can be defined as
follows:
 
0 −ωbz ωby
ωb× = S (ωb ) =  ωbz 0 −ωbx  , (3)
−ωby ωbx 0
and the inertial matrix Jt is given as follows:
 
Ix −Ixy −Ixz
Fig. 3. The schematic of the tail-sitter aircraft. 1: fuselage, 2: wings, 3: Jt =  −Ixy Iy −Iyz  .
V-tail, 4: Vane 1, 5: Vane 2, 6: Vane 3, 7: Vane 4, 8: Rotor 1, 9: Rotor 2, 10:
Rotor 3, 11: Rotor 4. −Ixz −Iyz Iz
The relationship between the angular rates and the attitude
angles of the aircraft is given by the following equation:
   
ωbx −ψ̇ sin θ cos φ + φ̇ cos θ
 ωby  =  ψ̇ sin φ + θ̇ ,
ωbz ψ̇ cos φ cos θ + φ̇ sin θ
The total force Ft contains the thrust Fr produced by
the four rotors, the aerodynamic force Fw generated by the
fixed wings, and the force Fd including uncertainties and
external disturbances. The total force Ft can be obtained by
transforming the forces expressed in the body frame above as
follows:

Ft = Rbe (Fr + Fw + Fd ) . (4)

Fig. 4. Coordinates and frames
The aerodynamic force Fw has the following expression:
   2 
expressed as follows: Fwx ρ kvk2 SCD /2
  Fw =  Fwy  =  ρ kvk22 SCS /2  , (5)
cψ cθ cψ sθ sφ − sψ cφ sψ sθ cφ + sψ sφ Fwz 2
ρ kvk2 SCL /2
R(φ, θ, ψ) =  sψ cθ sψ sθ sφ + cψ cφ sψ sθ cφ − cψ sφ  .
−sθ cθ sφ cθ cφ where
q ρ indicates the reference atmospheric density, kvk2 =
where φ, θ , and ψ describe the Euler angles: the roll angle vx2 + vy2 + vz2 the reference speed of the aircraft, and S
φ , the pitch angle θ, and the yaw angle ψ, respectively. The the reference aircraft pneumatic area, respectively. CD , CS ,
rotation matrix between angular rates relative to the body fixed and CL denote, respectively, the drag force aerodynamic
frame and the time derivative of the attitude angles in the coefficient, the side force aerodynamic coefficient, and the
inertial frame is defined as follows: lift force aerodynamic coefficient. The total torque Tt mainly
  consists of the aerodynamic torque Ta produced by the four
1 0 − sin θ vanes, the torque Tr generated
Rv =  0 cos φ cos θ sin φ  . £ ¤T by the four rotors, and the
torque Td = Tdx Tdy Tdz caused by the uncertainties
0 − sin φ cos θ cos φ and external disturbances. It can be written as follows:

C. Mathematical model Tt = Ta + Tr + Td . (6)

As a rigid body, the mathematical equations of motion of The aerodynamic moment Ta can be modeled as follows:
the tail-sitter aircraft can be derived by Newton-Euler’s law    2 
Tax ρ kvk2 cSCR /2
and described as follows:
Ta =  Tay  =  ρ kvk22 cSCM /2  , (7)
Taz 2
Ṗ = v, ρ kvk2 cSCN /2

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access

IEEE ACCESS 4

where CR , CM , and CN denote the aerodynamic coefficients.

The torque Tr produced by the four rotors is shown as follows:
   
Trx 0
Tr =  Try  =  Fr1 l1 + Fr2 l2 − Fr3 l3 − Fr4 l4  , (8)
Trz −Fr1 l1 + Fr2 l2 + Fr3 l3 − Fr4 l4
where Fri (i = 1, 2, 3, 4) are the thrusts produced by the four
rotors, li (i = 1, 2, 3, 4) are the distances from the mass center
of the tail-sitter aircraft to the center of the i−th (i = 1, 2, 3, 4)
rotor. The four rotors are symmetrically mounted to avoid
the roll torque caused by the rotors, because these moments
act in the opposite direction relative to the rotation rate of
the rotor and thereby the roll torque from each rotor can
be counteracted. As shown in Fig. 5, it can be obtained the
position of the each rotor and the direction of the torque.
Furthermore, it can be seen that the force generated by the Fig. 5. Direction of the forces and torques during the mode transition
rotor is parallel to the axis Bx in the body fixed frame as the
mounted angle error neglected. According to the relationship
between the wings and the fuselage, the expression of the and Euler angles can be obtained by the known quaternions
vector r can be obtained, which determines the distance and as shown follows:
direction of each rotor. Because the vector Fr and r are ¡ ¢
φ = arctan 2 (q0 q1 + q2 q3 ) /(q02 − q12 − q22 + q32 ) ,
obtained, the direction of the torque can be given which is θ = arcsin (2¡ (q0 q2 − q3 q1 )) ,
corresponding to the axis in Tr . The force generated by the ¢
ψ = arctan 2 (q0 q3 + q2 q1 ) /(q02 − q12 − q22 + q32 ) ,
four rotors are given by
then according to the given desired Euler angles, the quater-
Fri = kri ωi2 , i = 1, 2, 3, 4, (9) nion form can be obtained. Let
£ ¤ £ ¤
where kri (i = 1, 2, 3, 4) define the force coefficients of rotors, q0 q1 q2 q3 = q0 q ,
and ωi (i = 1, 2, 3, 4) are the angular rates of the rotors. The
aerodynamic model structure adopted in the current paper is then the expressions can be given (see, for example, [27])
generated based on the aerodynamic database that established q̇ = ((S (q) + q0 I3 )) ωb /2,
by the computational fluid dynamics technique at different an- (10)
q̇0 = −q T ωb /2,
gles of attack, sideslip angles and vane deflections conditions.
The aerodynamic model is shown as follows: where In represents an n × n identity matrix. In the current
paper, the definition of the angle of attack and the sideslip
CD = CD0 + CDα α + CDδa δa +CDδv δv , angle are shown as follows:
CS = CS0 β,
CL = CL0 + CLα α + CLδa δa +CLδv δv , α = θ − arctan (vz /vx ) ,
CM = CM0 + CMα α + CMδa δa +CMδv δv , β = arcsin (vby /vb ) ,
CR = CR0 + CRβ β + CRδa δa +CRδv δv , q
where vb = vbx 2 + v 2 + v 2 represents the velocity of the
CN = CN0 + CNβ β + CNδv δv , by bz
aircraft relative to the body frame.The tail-sitter UAV involves
where α, β, δa , and δv are angle of attack, sideslip angle,
a large pitch maneuver during the mode transition, which may
deflection angle of Vanes 1, 2, and Vanes 3, 4, respectively.
result in the stall constraint. The angle of attack is an important
Unit quaternion is adopted to describe the attitude of the
factor to observe the motion state of the aircraft. When the
aircraft in order to avoid the drawback of using Euler angle
flight velocity is in a low speed state, the angle of attack may
which may result in the singularity problem. Simulation results
suffer from the measurement noise and therefore it is difficult
are described by Euler angles to analysis the robust stability
to determine the aerodynamic forces and torques. Therefore,
and robust tracking properties of the closed-loop systems.
the angle of attack is defined as α = θ−arctan(vzr /vxr ) in the
Quaternions have the following expressions for the given Euler
current manuscript, where vzr and vxr represent the horizontal
angles:
and longitudinal velocity references in the body fixed frame.
q0 = cos (φ/2) cos (θ/2) cos (ψ/2) In this case, the angle of attack is equal to the pitch angle θ
+ sin (φ/2) sin (θ/2) sin (ψ/2) , minus the reference trajectory angle i.e., arctan(vzr /vxr ) to
q1 = sin (φ/2) cos (θ/2) cos (ψ/2) avoid the drastically change during the transition mode.
+ cos (φ/2) sin (θ/2) sin (ψ/2) , Remark 1: From (5) and (7)-(9), it can be seen that there
q2 = cos (φ/2) sin (θ/2) cos (ψ/2) exist couplings and nonlinearities in the model of the tail-
+ sin (φ/2) cos (θ/2) sin (ψ/2) , aircraft. The couplings and nonlinearities existed in the model
q3 = cos (φ/2) cos (θ/2) sin (ψ/2) address a challenging controller design problem for the tail-
+ sin (φ/2) sin (θ/2) cos (ψ/2) , sitter aircraft.

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access

IEEE ACCESS 5

III. N ONLINEAR ROBUST CONTROLLER DESIGN The angular velocity ωb can be computed from (10) as follows:
The proposed nonlinear robust control laws in the current ωb = 2(qo q̇ − q̇q0 ) − 2S(q)q̇.
paper consist of three parts and are designed in three steps in
this section. First, a nonlinear controller based on the dynamic From [36], one can obtain the quaternion tracking error as
inversion technique is designed and is applied to counteract the eq = q0d q − q0 qd + S(q)qd ,
known nonlinear terms to derive a linear model of the tail- eq0 = q0 q0d + q T qd .
sitter aircraft. Second, the eigenvalue assignment technique is
applied to design a state feedback linear controller to achieve Then one can rewrite (10) as follows:
stability and desired tracking performance specifications with- ėq = ((S (q) + q0 I3 )) eωb /2,
out considering equivalent disturbances. Then, a linear robust (13)
ėq0 = − 21 eq T eωb /2,
compensator is designed for the closed-loop nominal linear
system in the presence of equivalent disturbances to restrain Let M (q) = S (q) + q0 I3 , then from (12), where and are
the effects of uncertainties. Therefore, the proposed controller tracking errors. Let , then from Eq. (12), we can rewrite the
is composite of a nonlinear controller based on the dynamic model (13) as follows:
inversion technique, a linear feedback controller, and a robust ėq = M (q) eωb /2,
compensator. (14)
ėq0 = −eTq eωb /2.
A similar result about the formula derivation process can be
A. Nominal controller design seen in [27]. The position errors and the Euler angle errors are
Nominal controllers design consist two parts: nominal atti- defined as ei (i = 1, 2, 3, 4, 5, 6), where e1 , e2 , and e3 represent
tude controller design and nominal position controller design. the three attitude errors based on the quaternions, and e4 , e5 ,
A nonlinear controller based on the dynamic inversion tech- and e6 are relative to the position errors. From [35], one can
nique is designed without taking into account the equivalent obtain the attitude tracking errors e1 , e2 , and e3 based on the
disturbances in this part. From (2) and (3), the system can be quaternions as follows:
rewritten as follows:    
ea1 −q0r q1 + q1r q0 + q2r q3 − q3r q2
£ ¤  ea2  = ∆ ×  −q0r q2 − q1r q3 + q2r q0 + q3r q1  ,
ω̇b = Jt−1 (Ta + Tr ) − ωb× Jt ωb . (11)
ea3 −q0r q3 + q1r q2 − q2r q1 + q3r q0 (15)
For the desired attitude angles φd , θd , and ψd , let ωbd = P3
£ ¤T ∆ = 2sgn( qir qi ),
Rbe φ̇d θ̇d ψ̇d be the given desired commands of the i=0
angular rates in the body fixed frame, where Rbe represent where qir
represent the reference quaternions signals which
the rotation matrix between the angular rates relative to the can be obtained by the transformation of the reference Euler
body fixed frame and the derivative of the attitude angles in angles. The position tracking errors e4 , e5 , and e6 have the
the inertial frame. The tracking errors of the angular rates can following expression:
be defined as follows:    r 
ep4 p1 (t) − p1 (t)
eωb = ωb − ωbd , (12)  ep5  =  pr2 (t) − p2 (t)  ,
ep6 pr3 (t) − p3 (t)
For the desired attitude angle φd , θd , and ψd , one can obtain
the desired attitude in the quaternion expression as: where pri (t) are the reference longitude position, lateral posi-
tion, and vertical position signals. The Euler angle are not used
q0 = cos (φ/2) cos (θ/2) cos (ψ/2)
to describe the attitude to avoid the singularity problem. For
+ sin (φ/2) sin (θ/2) sin (ψ/2) ,
the nominal attitude controller design, one can define the error
q1 = sin (φ/2) cos (θ/2) cos (ψ/2)
vector ea (t) = [eai (t)]3×1 . Then, ea is the attitude tracking
+ cos (φ/2) sin (θ/2) sin (ψ/2) ,
error vector and eωb is rotational velocity tracking error. One
q2 = cos (φ/2) sin (θ/2) cos (ψ/2)
can obtain that
+ sin (φ/2) cos (θ/2) sin (ψ/2) ,
q3 = cos (φ/2) cos (θ/2) sin (ψ/2) ėa = e¡ωb , £ ¤ ¢
+ sin (φ/2) sin (θ/2) cos (ψ/2) , ėωb = Jt−1 (Ta + Tr ) − ωb× Jt ωb − ω̇bd .
£ ¤ By defining the control input as Tr , the following expression
The desired unit quaternion is Qd = q0d qd which
satisfies can be obtained based on the dynamic inversion technique as:

qdT qd + q0d
2
= 1. Tr = ωb× Jt ωb + Jt ω̇bd − Ta + Jt ua .

As shown in [27], the desired unit quaternion is related to where ua = [uai ]3×1 is virtual control input that can be
the desired angular velocity ωb by the following dynamic adopted for the linear systems generated by the dynamic
equation: inversion technique. Therefore, one can obtain the following
expression:
q̇d = ((S (qd ) + q0d I3 )) ωbd /2,
q̇0d = −q T ωbd /2. ëa = [uai ]3×1 . (16)

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access

IEEE ACCESS 6

Remark 2: It should be noted that the designed nonlinear

control law based on the dynamic inversion technique is
mainly utilized to counteract the nonlinear terms and coupling
parts of the aircraft model to generate a linear error model. In
this section, the uncertainties and external disturbances have
not been considered. Quaternion tracking errors rather than
Euler angle tracking errors are applied to avoid the singularity
that may be caused by Euler angle in the mode transition
process.
As for the nominal position controller design, one can
define the error vector ep (t) = [epi (t)]3×1 , which represent
the position tracking error vector. Then it can be obtained
following expression:
ėp = ev , Fig. 6. The block diagram of the robust control system.
(17)
ėv = (Rbe (Fr + Fw + Fd ) /m) − gEz − v̇bd ,
where ev is the velocity tracking error and vbd represent the Remark 3: It should be noted that the state feedback con-
reference velocity in the body fixed frame. Let the total thrust trollers ui,con (i = 1, 2, 3, 4, 5, 6) are designed to guarantee
Fr produced by the four rotors be the control input, then the stability of the nominal plant, and the robust tracking
the following control law is selected based on the dynamic properties of the closed-loop system when uncertainties and
inversion technique without taking equivalent disturbances into external disturbances are considered are guaranteed by the
account: introduced robust compensators ui,com (i = 1, 2, 3, 4, 5, 6).
Fr = m(up − Fw )/Rbe + gEz + v̇bd , The block diagram depicted in Fig. 6 shows the overall
control system of the tail-sitter aircraft. It can be seen that
where up = [upi ]3×1 is virtual control input that can be the overall control system consists of two channels including
adopted for the linear error systems resulting by the dynamic attitude channel and position channel. The attitude channel is
inversion technique: designed to guarantee the rotational dynamics for desired atti-
ëp = [upi ]3×1 . (18) tude tracking, and translational dynamics for position tracking
is governed by the designed position channel controllers.
Actually, the dynamic system can be regarded as the nominal
system utilized for designing state feedback controllers. The B. State feedback controller design
other unknown parts of the system dynamics are treated as a
part of equivalent disturbances. Then the linear error system For linear time-invariant system, various techniques, such
state-space form can be written as follows: as linear quadratic regulation [28], eigenstructure assignment
method [29]–[32], are available for designing controllers. In
Ėi = Ai Ei + Bi ui + di , this paper, the widely utilized approach eigenstructure assign-
(19)
yi = Ci Ei , i = 1, 2, 3, 4, 5, 6, ment method is adopted to design state feedback controllers for
£ ¤T the linear error systems in (19). The eigenstructure assignment
where Ei = ei ėi ,
method is conducted by redisposing the poles of the closed-
· ¸ · ¸ · ¸T loop system with the designed controller in the s-plane. Ac-
0 1 0 1
Ai = , Bi = , Ci = , (20) cording to the given stability and performance specifications,
0 0 1 0
the desired poles of the closed-loop system which determine
and ei describes the position errors and the Euler angle errors the responses of a plant can be pre-determined, and then
of the aircraft as: according to the poles of the original system and the desired
       
e1 ėa1 e4 ėp1 poles, the state feedback controllers can be designed. From
 e2  =  ėa2  ,  e5  =  ėp2  . (20), the linear error systems have the following forms without
e3 ėa3 e6 ėp3 taking equivalent disturbances into consideration:
di are equivalent disturbances including the parametric uncer- Ėi = Ai Ei + Bi ui,con , i = 1, 2, 3, 4, 5, 6.
tainties, unmodeled uncertainties, external disturbances, and
nonlinear dynamics that cannot be counteracted accurately by The state feedback controllers based on the eigenvalue as-
the dynamic inversion technique. signment scheme are designed with the linear error system
The designed linear¤ control laws which can be expressed (19), and the designed controllers based on the state feedback
£ T approach are shown as follows:
as ui = uai upi (i = 1, 2, 3, 4, 5, 6) include the state
feedback controllers ui,con (i = 1, 2, 3, 4, 5, 6) and the robust ui,con = −Ki,con Ei , i = 1, 2, 3, 4, 5, 6,
compensators ui,com (i = 1, 2, 3, 4, 5, 6), then the virtual con-
trol laws have the following expressions: where Ki,con (i = 1, 2, 3, 4, 5, 6) are the feedback control
gains determined by the eigenstructure assignment method.
ui = ui,con + ui,com , i = 1, 2, 3, 4, 5, 6. It can be obtained that the closed-loop systems of the plant

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access

IEEE ACCESS 7

TABLE I
have the desired stability and performance specifications by NOMINAL PARAMETERS OF THE TAIL-SITTER AIRCRAFT
substituting the proposed feedback controllers into (19):
Parameters Values Parameters Values
Ėi = (Ai − Bi Ki,con ) Ei + Bi ui,con , i = 1, 2, 3, 4, 5, 6. (21) Ix 0.263kg · m2 Ixy 6.832e−5 kg · m2
Iy 0.087kg · m2 Iyz 8.967e−4 kg · m2
From (19) and (21), one can obtain that: Iz 0.333kg · m2 Izx 1.548e−5 kg · m2
ρ 1.225kg · m2 c 1m
Ėi = Ai,con Ei + Bi ui,com + di , S 1m2
(22)
yi = Ci Ei , i = 1, 2, 3, 4, 5, 6,
where Ai,con = Ai − Bi Ki,con (i = 1, 2, 3, 4, 5, 6). Since
Since the values of the equivalent disturbances cannot
feedback control gains Ki,con (i = 1, 2, 3, 4, 5, 6) are properly
be measured directly, the robust compensator can be recon-
selected to guarantee the matrices Ai,con (i = 1, 2, 3, 4, 5, 6)
structed by the following expressions:
have eigenvalues with negative real parts, then the matrices
Ai,con (i = 1, 2, 3, 4, 5, 6) are Hurwitz [34], [37]–[40]. di (s) = (sIi − Ai,con )Ei (s) − Bi vi,com (s),
(26)
i = 1, 2, 3, 4, 5, 6.
C. Robust compensator design
From (25) and (26), the robust compensating output control
The state feedback control laws are designed without con- laws can be expressed by the outputs yi (i = 1, 2, 3, 4, 5, 6) as
sidering the effects of equivalent disturbances, therefore robust shown follows:
compensators are introduced to guarantee that the closed-
loop system can achieve desired robust stability and robust vi,com (s) = −Fi (s)(1 − Fi (s))−1 Gi −1 (s)yi (s),
(27)
tracking properties in the presence of equivalent disturbances i = 1, 2, 3, 4, 5, 6.
di (i = 1, 2, 3, 4, 5, 6).

Gi (s) = Ci (sI2 − Ai, con )−1 Bi , i = 1, 2, 3, 4, 5, 6, IV. ROBUST STABILITY AND TRACKING PROPERTY
ANALYSIS
then the outputs of the system can be shown as bellow:
From (21) and (27), the linear error system can be rewritten
yi (s) = Ci (sIi − Ai,con )−1 (Ei (0) + di (s)) as follows:
+ Gi (s)vi,com (s), (23)
i = 1, 2, 3, 4, 5, 6. Ėi = Ai,con Ei + (1 − Fi ) di ,
(28)
yi = Ci Ei , i = 1, 2, 3, 4, 5, 6.
From (22) and (23), the equivalent disturbances can be coun-
teracted if the control inputs are chosen as follows: Remark 5: Since the uncertain parameters fluctuate be-
tween lower and upper bounds, therefore, assume the param-
vi,com (s) = −Gi −1 (s)Ci (sIi − Ai, con )−1 di (s), eter uncertainty terms bounded are reasonable. This is also
(24)
i = 1, 2, 3, 4, 5, 6. reasonable to be applied to external disturbances.
The robust the compensators are introduced to attenuate the According to [33], the equivalent disturbances are assumed
influence of equivalent disturbances and the following expres- to have finite norm bounds as follows:
sions deduced from (24) are given to describe the introduced
robust compensators: kdi (t)k∞ ≤ ξEi kEi (t)k∞ + ξi , i = 1, 2, 3, 4, 5, 6,

vi,com (s) = −Fi (s) Gi −1 (s)Ci (sIi − Ai, con )−1 di (s),
(25)
i = 1, 2, 3, 4, 5, 6,
where Fi (s) (i = 1, 2, 3, 4, 5, 6) are called robust filters. The
robust filters Fi (s) (i = 1, 2, 3, 4, 5, 6) have the following ex-
pressions:
fi2
Fi (s) = 2,i = 1, 2, 3, 4, 5, 6,
(s + fi )
where fi (i = 1, 2, 3, 4, 5, 6) are positive constants and deter-
mined by the specified prescribed conditions and performance
specifications.
Remark 4: If fi (i = 1, 2, 3, 4, 5, 6) are sufficiently
large enough, then the gains of the robust filters
Fi (s) (i = 1, 2, 3, 4, 5, 6) are approximately equal to one
and the designed compensators are asymptotic with the
ideal condition, which means more effects of equivalent
disturbances di (i = 1, 2, 3, 4, 5, 6) can be counteracted.
where ξi and ξEi (i = 1, 2, 3, 4, 5, 6) are positive constants.
Remark 6: It should be noted that with the boundedness of
the equivalent disturbances including parametric uncertainties,
unmodeled uncertainties, and external disturbances, the robust
stability and robust tracking properties of the closed-loop
control system can be guaranteed. The proof has been given
in the following parts.
Theorem 1: For any given positive constant ε and given
bounded initial state tracking errors Ei (0)(i = 1, 2, 3, 4, 5, 6),
positive constants fu and tu can be found such that for
any fi ≥ fu (i = 1, 2, 3, 4, 5, 6), the state errors Ei (t)(i =
1, 2, 3, 4, 5, 6) are bounded and the output tracking errors
|yi (t)| ≤ ε, ∀t ≥ Tu (i = 1, 2, 3, 4, 5, 6) are guaranteed.
Proof: From (28), it can be obtained that
Rt
E (t) = eAi,con t E (0) + eAi,con (t−τ ) (1 − Fi ) di dτ ,
0 (29)
i = 1, 2, 3, 4, 5, 6.

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access

IEEE ACCESS 8

TABLE II
From (29), it can be deduced that PARAMETERS OF AERODYNAMIC COEFFICIENTS
° °
° t Rt t−τ °
kE (t)k∞ = °eA E (0) + eA F1 di dτ °
°
°
Parameters Values Parameters Values
0 ∞ CD0 −2.2176 CM0 0.1212
≤ keA k∞ kE (0)k∞ C Dα 0.0295 CMα 0.1179
C D δα −0.0018 CMδα 0.0069
Rt
+ ketA k∞ keτ k∞ kF1 k∞ kdi k∞ dτ C Dδ v 0.4000 CMδv 0.1309
0 C L0 −3.2570 C R0 −0.0509
≤ keA k∞ kE (0)k∞ C Lα −1.7491 C Rβ −0.0233
Rt C Lδ α −0.1418 C Rδ α −0.0224
+ ketA k∞ keτ k∞ kF1 k∞ ξEi kEi (t)k∞ dτ (30) CLδv 0.1309 C Rδ v −0.0114
0 CS0 −1.1722 C N0 0.0083
Rt C Nβ 0.0504 C N δv 0.0293
+ keA k∞ keτ k∞ kF1 k∞ ξi dτ ,
0
eA = eAi,con t ,
eτ = eAi,con −τ , From (35) and (36), the following expression can be achieved:
F1 = 1 − Fi ,
eλm t kE (t)k∞ ≤ kE (0)k∞ + m
i = 1, 2, 3, 4, 5, 6.
Rt (37)
Let + e−λm τ ξEi kF1 k∞ kEi (t)k∞ dτ ,
0

−λm = maxRe (λi ) , i = 1, 2, (31) Let

where λi < 0 (i = 1, 2) indicate the eigenvalues of C = kE (0)k∞ + m, (38)

Ai,con (i = 1, 2, 3, 4, 5, 6) and −λm is the maximum real part
then from (38), one can rewrite (37) as follows:
of the eigenvalues of Ai,con (i = 1, 2, 3, 4, 5, 6). From (31),
one can obtain the following inequality: eλm t kE (t)k∞ ≤ C
Rt (39)
° A ° ° ° + 0 e−λm τ ξEi kF1 k∞ kEi (t)k∞ dτ ,
°e i,con t ° ≤ °e−λm t ° = e−λm t , i = 1, 2, 3, 4, 5, 6. (32)
According to the Bellman-Gronwall Lemma, the following
From (32), one can rewrite (30) as follows: conclusion can be deduced from (39):
kE (t)k∞ ≤ e−λm t kE (0)k∞ Rt
ξEi k(1−Fi )k∞ dτ
Rt
+ e−λm t e−λm τ kF1 k∞ Eξ dτ , eλm t kE (t)k∞ ≤ Ce 0 , i = 1, 2, 3, 4, 5, 6, (40)
0 (33)
Eξ = ξEi kEi (t)k∞ + ξi , then
i = 1, 2, 3, 4, 5, 6.
kE (t)k∞ ≤ Ce(ξEi k(1−Fi )k∞ −λm )t , i = 1, 2, 3, 4, 5, 6. (41)
The following expression can be deduced from (33):
From (25), the following condition can be guaranteed when
λm t
Rt −λm τ fi (i = 1, 2, 3, 4, 5, 6) are chosen large enough.
e kE (t)k∞ ≤ kE (0)k∞ + e kF1 k∞ Eξ dτ ,
0 (34)
i = 1, 2, 3, 4, 5, 6, ξEi k(1 − Fi )k∞ − λm < 0, i = 1, 2, 3, 4, 5, 6. (42)

then since ξi (i = 1, 2, 3, 4, 5, 6) are positive constants, there- Then it can be obtained conclusion that, for a given bounded
fore, from (34), the following expression can be obtained: initial tracking error E (0), the tracking error E (t) would be
bounded if fi (i = 1, 2, 3, 4, 5, 6) are selected as fi ≥ fu (i =
eλm t kE (t)k∞ ≤ kE (0)k∞ 1, 2, 3, 4, 5, 6).
Rt Remark 7: It should note that the robust compensator pa-
+ e−λm τ ξEi kF1 k∞ kEi (t)k∞ dτ
(35) rameters fi (i = 1, 2, 3, 4, 5, 6) are not needed to be chosen
R0 t
+ 0 e−λm τ kF1 k∞ ξi dτ , sufficiently large and can be tuned online in practical ap-
i = 1, 2, 3, 4, 5, 6. plications. The parameters fi (i = 1, 2, 3, 4, 5, 6)can be set to
certain initial values to run in the whole closed-loop control
Since k1 − Fi k∞ (i = 1, 2, 3, 4, 5, 6) are bounded, and systems. If the desired tracking performances are not satisfied,
ξi (i = 1, 2, 3, 4, 5, 6) are positive constants, then the follow- the parameters fi (i = 1, 2, 3, 4, 5, 6) can be reset to larger
ing equation can be obtained: values until the desired tracking performances can be achieved.
R t −λ τ Rt
e m kF1 k∞ ξi dτ =kF1 k∞ ξi 0 e−λm τ dτ ,
0 R t −λ τ ¡ ¢ V. S IMULATION RESULTS
kF1 k∞ ξi 0 e m dτ = λξmi kF1 k∞ 1 − e−λm t .
¡ ¢± The values of the nominal plant parameters are given in
As λm > 0, then ξi k(1 − Fi )k∞ 1 − e−λm t λm are Table I. The parameters of aerodynamic coefficients are shown
bounded and let in Table II. Table III shows the values of the external distur-
ξi ¡ ¢ bances, where dni (i = 1, 2, 3, 4, 5, 6) indicate high frequency
m= k(1 − Fi )k∞ 1 − e−λm t , i = 1, 2, 3, 4, 5, 6. (36) noise signals.
λm

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access

IEEE ACCESS 9

TABLE III
EXTERNAL DISTURBANCES
10

Roll angle (deg)

Reference roll angle signal
Roll angle by loopdshaping controller
Parameters Values 5
Roll angle by proposed robust controller
d1 0.5 sin(0.001πt) + dn1 0
d2 0.1 sin(0.001πt) + dn2
d3 0.3 sin(0.001πt) + dn3 −5
0 2 4 6 8 10 12 14 16 18 20
d4 0.1 sin(0.001πt) + dn4 Time (s)
(a) Roll angle response from hover to level
d5 0.05 sin(0.001πt) + dn5
10
d6 0.05 sin(0.001πt) + dn6

Yaw angle (deg)

Reference yaw angle signal
5 Yaw angle by loopdshaping controller
Yaw angle by proposed robust controller
0

Singular Values −5
0 2 4 6 8 10 12 14 16 18 20
200 Time (s)
Loopshaping controller
(b) Yaw angle response from hover to level
Proposed controller
100

Pitch angle (deg)

80 Reference pitch angle signal
Pitch angle by loopdshaping controller
0
60
Pitch angle by proposed robust controller
Singular Values (dB)

40
20
−100
0
0 2 4 6 8 10 12 14 16 18 20
−200 Time (s)
(c) Pitch angle response from hover to level

−300

−400
Fig. 9. Attitude responses by loop-shaping controller and proposed controller
for nominal model.
−500 −10 −5 0 5 10
10 10 10 10 10
Frequency (rad/s)

Horizontal velocity (m/s)

Fig. 7. Singular values in pitch angle channel. 20

Reference horizontal velocity signal

10
Velocity by loopshaping controller
Velocity by proposed robust controller
0
In this paper, the process of flight mode transitions from 0 2 4 6 8 10
Time (s)
12 14 16 18 20

hovering to level flight of the tail-sitter aircraft has been (a) Horizontal velocity response from hover to level
Lateral position (m)

studied. The mode transitions require that the flight path 40 Reference lateral position signal
Lateral position by loopshaping controller
angle changes from 90◦ to zero while the roll and yaw angle 20 Lateral position by proposed robust controller
maintain zero. The final desired level fight velocity is set to
0
25 m/s and height is 30 m. Theoretically, the selected con- 0 2 4 6 8 10 12 14 16 18 20
troller gain vectors Ki,con (i = 1, 2, 3, 4, 5, 6) can guarantee Time (s)
(b) Lateral position response from hover to level
the matrices Ai,con (i = 1, 2, 3, 4, 5, 6) are Hurwitz. In practi-
Vertical position (m)

30
cal design process, the gain vectors Ki,con (i = 1, 2, 3, 4, 5, 6)
20
are determined by the desired poles of the nominal closed- Reference vertical position signal
loop system selected based on the desired time domain per- 10 Vertical position by loopshaping controller
Vertical position by proposed robust controller
formances including steady-state error, overshoot, rising time, 0
0 2 4 6 8 10 12 14 16 18 20
and settling time. The desired poles of the closed-loop system Time (s)
(c) Vertical position response from hover to level
are selected as

Fig. 10. Velocity and position responses by loop-shaping controller and

proposed controller for nominal model.
Singular Values
300
Proposed controller with f1=1
4
200 Proposed controller with f1=10

Proposed controller with f =10

8 £ ¤
100
1
λ1 = £ −0.5 + 4.4j −0.5 − 4.4j , ¤
Singular Values (dB)

0 λ2 = £ −0.25 + 3.2j −0.25 − 3.2j¤ ,

−100
λ3 = £ −0.5 + 3.8j −0.5 − 3.8j ¤ ,
λ4 = −0.3 + 2.2j −0.3 − 2.2j ,
−200
λ5 = [−0.01
£ + 0.7j −0.01 − 0.7j] ,¤
−300 λ6 = −0.3 + 2.2j −0.3 − 2.2j .
−400
Numerical simulation results of loop-shaping controllers
−500 −10 −5 0 5 10
designed based on the linearized model (22), as depicted in
10 10 10 10 10
Frequency (rad/s) [34], has been shown to compare with the proposed nonlinear
robust controller to demonstrate the advantages of the pro-
Fig. 8. Singular values comparison in horizontal velocity channel. posed nonlinear robust controller.

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access

IEEE ACCESS 10

Horizontal position (m)

1
Roll angle (deg) Reference roll angle signal 400 Horizontal position by proposed robust controller
Roll angle by proposed robust controller
0 200

−1 0
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (s) Time (s)
(a) Roll angle response from hover to level (a) Horizontal position response from hover to level

Lateral velocity (m/s)

1 5
Yaw angle (deg)

Reference yaw angle signal Velocity by proposed robust controller

Yaw angle by proposed robust controller
0 0

−1 −5
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (s) Time (s)
(b) Yaw angle response from hover to level (b) Lateral velocity response from hover to level

Vertical velocity (m/s)

10
Pitch angle (deg)

80 Reference pitch angle signal Velocity by proposed robust controller

Pitch angle by proposed robust controller 5
60
40
0
20
0 −5
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (s) Time (s)
(c) Pitch angle response from hover to level (c) Vertical velocity response from hover to level

Fig. 11. Attitude responses under uncertainties by proposed controller. Fig. 13. Position and velocity responses by proposed controller for nominal
model.
Horizontal velocity (m/s)

Horizontal position (m)

20 400 Horizontal position by proposed robust controller

10 200
Reference horizontal velocity signal
Velocity by proposed robust controller
0
0 2 4 6 8 10 12 14 16 18 20 0
Time (s) 0 2 4 6 8 10 12 14 16 18 20
Time (s)
(a) Horizontal velocity reponse from hover to level
(a) Horizontal position response from hover to level
Lateral position (m)

1
Lateral velocity (m/s)

Reference Lateral position signal 10

Velocity by proposed robust controller
Lateral position by proposed robust controller
0
0

−1
0 2 4 6 8 10 12 14 16 18 20 −10
Time (s) 0 2 4 6 8 10 12 14 16 18 20
Time (s)
(b) Lateral position response from hover to level
(b) Lateral velocity response from hover to level
Vertical position (m)

Vertical velocity (m/s)

30
15
Velocity by proposed robust controller
20 10
5
10 Reference Vertical position signal
Vertical position by proposed robust controller 0
0
0 2 4 6 8 10 12 14 16 18 20 −5
Time (s) 0 2 4 6 8 10 12 14 16 18 20
Time (s)
(c) Vertical position response from hover to level
(c) Vertical velocity response from hover to level

Fig. 12. Velocity and position responses under uncertainties by proposed Fig. 14. Position and velocity responses under uncertainties by proposed
controller. controller.

Comparison results of singular values of the closed-loop different values of fi (i = 1, 2, 3, 4, 5, 6) have been shown
transfer functions from disturbances to the outputs in dif- to demonstrate the improvement of disturbances rejection
ferent channels are presented in Fig. 7. The results show properties by increasing the values of fi (i = 1, 2, 3, 4, 5, 6).
that the singular values in low frequency of the proposed Simulations of loop-shaping controllers and the proposed
robust controllers are smaller, which means the proposed robust controllers have been conducted for the nominal non-
robust controllers have better properties in rejecting the distur- linear model and uncertain nonlinear model, respectively.
bance effects than the loop-shaping controllers for the closed- Figs. 9-10 depict the attitude, velocity and position tracking
loop system, especially in low frequency. The disturbances performance responses of the nominal nonlinear model by
rejection properties of the proposed robust controllers can designed loop-shaping controllers and the proposed nonlinear
be improved significantly by increasing the values of the robust controllers. From these figures, it can be obviously
fi (i = 1, 2, 3, 4, 5, 6). In Fig. 8, singular values of the closed- obtained that both controllers can achieve desired tracking
loop transfer functions from disturbances to the outputs with performance for the nominal nonlinear model. However, the

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access

IEEE ACCESS 11

Angle of attack (deg)

Control input force

20
Angle of attack by proposed robust controller 20 The longitudinal control input force

0 10

0
−20
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (s) Time (s)
(a) Attack angle response from hover to level (a) The longitudinal control input force response from hover to level
Sideslip angle (deg)

Control input force

0.5 1
Sideslip angle by proposed robust controller The lateral control input force

0 0

−0.5 -1
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (s) Time (s)
(b) Sideslip angle response from hover to level (b) The lateral control input force response from hover to level
Flight path angle (deg)

Control input force

100
Flight path angle by proposed robust controller The vertical control input force
40

50 20

0
0
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (s) Time (s)
(c) Flight path angle response from hover to level (c) The vertical control input force response from hover to level

Fig. 15. Angle of attack, sideslip angle, and flight path angle responses by Fig. 17. The control input force response from hover to level by proposed
proposed controller for nominal model. controller.

Control input torque

Angle of attack (deg)

20 5
Angle of attack by proposed robust controller The roll control input torque

0 0

−20 -5
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (s) Time (s)
(a) Attack angle response from hover to level (a) The roll control input torque response from hover to level
Control input torque
Sideslip angle (deg)

1 5
Sideslip angle by proposed robust controller The pitch control input torque

0 0

−1 -5
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (s) Time (s)
(b) Sideslip angle response from hover to level (b) The pitch control input torque response from hover to level
Flight path angle (deg)

Control input torque

100 5
Flight path angle by proposed robust controller The yaw control input torque

50 0

0 -5
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time (s) Time (s)
(c) Flight path angle response from hover to level (c) The yaw control input torque response from hover to level

Fig. 16. Angle of attack, sideslip angle, and flight path angle responses Fig. 18. The control input torque response from hover to level by proposed
under uncertainties by proposed controller. controller.

simulation results in Figs. The uncertain nonlinear model proposed robust controller. Fig. 14 shows that the response
cannot achieve desired tracking performance by the designed of the aircraft fluctuates in a small scope near the nominal
loop-shaping controllers when the uncertainty of the parame- response of the aircraft, which demonstrates the robustness
ters are selected to be 20% larger than the nominal one while of the proposed controller. Attack angle, slide angle and
the proposed robust controllers can, as shown in Figs. 11- flight path angle responses in the mode transitions of the
12. The results obviously demonstrate the advantages of the nominal nonlinear system are described in Fig. 15. Taking
proposed controller. the uncertainties and external disturbances into account, the
Figs. 13-14 depict the movement of the horizontal posi- responses of the angle of attack, the slide angle, and the flight
tion, the velocity responses of the nonlinear model in lateral path angle are shown in Fig. 16. From these figures, it can be
direction and vertical direction with and without equivalent seen that the proposed robust closed-loop system of the tail-
disturbances, respectively. It can be seen that, in Fig. 13, sitter aircraft can track the prescribed references. One can see
the nominal system of the tail-sitter aircraft is stable by the the response of the control input from the proposed controller

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access
IEEE ACCESS
from Fig. 17 and Fig. 18, and the amplitude of the control
inputs by the proposed controller is within a reasonable range.
VI. C ONCLUSION
In the paper, the problem of mode transitions from hovering
to level flight of a tail-sitter aircraft with external disturbances
and uncertainties including parametric uncertainties, nonlin-
ear dynamics, and unmodeled uncertainties is dealt by the
proposed nonlinear robust controller. The proposed nonlinear
robust controller consisting of a dynamic inversion controller,
a state feedback controller, and a robust compensator can guar-
antee the tail-sitter aircraft can achieve desired robust stability
and robust tracking properties. The proof of the robust stability
and robust tracking properties has been given in this paper.
The comparison simulation results are presented to show
the effectiveness and advantages of proposed controller.The
proof of the robust stability and robust tracking properties has
been given in this paper. The comparison simulation results
are presented to show the effectiveness and advantages of
proposed controller.
R EFERENCES
[1] Xu, Z., Nian, X., Wang, H., Chen, Y.: Robust guaranteed cost tracking
control of quadrotor UAV with uncertainties. ISA Transactions. 69(1),
157-165 (2017)
[2] Liu, H., Ma, T., Lewis, F.L., Wan, Y.: Robust formation control for mul-
tiple quadrotors with nonlinearities and disturbances. IEEE Transactions
on Cybernetics. PP(99), 1-10 (2018)
[3] Zhao, S., Hu, Z., Yin, M., Ang, K.Z.Y., Liu, P., Wang, F., Dong, X.,
Lin, F., Chen, B.M., Lee, T.H.: A robust real-time vision system for
autonomous cargo transfer by an unmanned helicopter. IEEE Transactions
on Industrial Electronics. 62(2), 1210-1219 (2015)
[4] Liu, H., Zhao, W., Zuo, Z., Zhong, Yi.: Robust control for quadrotors
with multiple time-varying uncertainties and delays. IEEE Transactions
on Industrial Electronics. 64(2), 1303-1312 (2017)
[5] Zhao, B., Xian, B., Zhang, Y., Zhang, X.: Nonlinear robust adaptive
tracking control of a quadrotor UAV via immersion and invariance
methodology. IEEE Transactions on Industrial Electronics. 62(5), 2891-
2902 (2015)
[6] Naldi, R., Furci, M., Sanfelice, R.G., Marconi, L.: Robust global trajec-
tory tracking for underactuated VTOL aerial vehicles using inner-outer
loop control paradigms. IEEE Transactions on Automatic Control. 62(1),
97-112 (2017)
[7] Hou, X., Mahony, R.: Dynamic kinesthetic boundary for haptic teleopera-
tion of VTOL aerial robots in complex environments. IEEE Transactions
on Systems, Man, and Cybernetics: Systems. 46(10), 1352-1363 (2016)
[8] Kita, K., Konno, A., Uchiyama, M.: Hovering control of a tail-sitter
VTOL aerial robot. Journal of Robotics and Mechatronics. 21(2) 277-
283 (2009)
[9] Lin, F., Zhang, W., Brandt, R.: Robust hovering control of a PVTOL
aircraft. IEEE Transactions on Control Systems and Technology. 7(3),
343-351 (1999)
[10] Stone, R.H.: The T-wing tail-sitter unmanned air vehicle: from design
concept to research flight vehicle. Journal of Aerospace Engineering.
218(6), 417-433 (2006)
[11] Escareno, J., Salazar, S., Lozano, R.: Modelling and control of a
convertible VTOL aircraft. in Proceedings of the 45th IEEE Conference
on Decision and Control, San Diego, CA, USA. 69-74 (2006)
[12] Hauser, J., Sastry, S., Meyer, G.: Nonlinear control design for slightly
non-minimum phase systems: application to V/STOL aircraft. Automat-
ica. 28(4), 665-679 (1992)
[13] Martin, P., Devasia, S., Paden, B.: A different look at output tracking:
control of a VTOL aircraft. Automatica. 32(1), 101-107 (1996)
[14] Kita, K., Konno, A., Uchiyama, M.: Transition between level flight
and hovering of a tail-sitter vertical take off and landing aerial robot.
Advanced Robotics. 24(5), 763-781 (2010)
[15] Alberto, M. and Jesus.: On the estimation and control of nonlinear
systems with parametric uncertainties and noisy outputs. IEEE Access,
6, 31968-31973 (2018)
2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
12
[16] Ailon, A.: Simple tracking controllers for autonomous VTOL aircraft
with bounded inputs. IEEE Transactions on Automatic Control. 55(3),
737-743 (2010)
[17] Wang, X.: Take off/landing control based on acceleration measurements
for VTOL aircraft. Journal of The Franklin Institute. 350(10), 3045-3063
(2013)
[18] Li, H., Li, C., Li, H.: An integrated altitude control design for a tail-
sitter UAV equipped with turbine engines. IEEE Access, 5, 10941-10952
(2017)
[19] He, Q., Liu, J.: An observer for a velocity-sensorless VTOL aircraft
with time-varying measurement delay. International Journal of Systems
Science. 47(3), 652-661 (2015)
[20] Su, S., Lin, Y.: Output tracking control for a velocity-sensorless VTOL
aircraft with measurement delays. International Journal of Systems Sci-
ence. 46(5), 885-895 (2015)
[21] Zhu, B., Wang, X., Cai, K.: Tracking control for a velocity-sensorless
VTOL aircraft with delayed outputs. Automatica. 45(12), 2876-2882
(2009)
[22] Matsumoto, T., Kita, K., Suzuki, R., Osedo, A., Go, K., Hoshino,
Y., Konno, A., Uchiyama, M.: A hovering control strategy for a tail-
sitter VTOL UAV that increases stability against large disturbance. in
Proceedings of the IEEE International Conference on Robotics and
Automation, Anchorage, AL, USA. 54-59 (2010)
[23] Stone, R.H., Anderson, P., Hutchison, C., Tsai, A., Gibbens, P., Wong,
K.C.: Flight testing of the T-wing tail-sitter unmanned air vehicle. Journal
of Aircrafts. 45(2), 673-685 (2008)
[24] Abiko, S., Konno, A., Uchiyama, M.: Optimal transition from hovering
to level-flight of a quadrotor tail-sitter UAV. Autonomous Robots. 41(5),
1143-1159 (2017)
[25] Escareno, J., Stone, R.H., Sanchez, A., Lozano, R.: Modeling and control
strategy for the transition of a convertible tail-sitter UAV. in Proceedings
of the European Control Conference 2007, Kos, Greece. 3385-3390
(2007)
[26] Wang, X., Chen, Z., Yuan, Z.: Modeling and control of an agile tail-sitter
aircraft. Journal of the Franklin Institute. 352(12), 5437-5472 (2015)
[27] Wen, J.T., Kreutz-Delgado, K.: The attitude control problem. IEEE
Transactions on Automatic Control. 36(10), 1148-1162 (1991)
[28] Liu, H., Li, D., Zuo, Z., Zhong, Y.: Robust three-loop trajectory tracking
control for quadrotors with multiple uncertainties. IEEE Transaction on
Industrial Electronics. 63(4), 2263-2274 (2016)
[29] Dao, M.N., Noll, D., Apkarian, P.: Robust eigenstructure clustering by
non-smooth optimization. International Journal of Control. 88(8), 1441-55
(2015)
[30] Rubio, J.J., Garcia, E., Aguilar, C., Torres, C.: Stabilization of the robotic
arms. IEEE Latin America Transactions. 13(8), 2567-2573 (2015)
[31] Huang, M., Wang, X., Wang, Z.: Multiple model self-tuning control for
a class of nonlinear systems. International Journal of Control. 88(10),
1984-1994 (2015)
[32] Li, Z., Zhou, W., Liu, H.: Nonlinear robust control of hypersonic aircrafts
with interactions between flight dynamics and propulsion systems. ISA
Transactions. 64(9), 1-11 (2016)
[33] Liu, H., Xi, J., Zhong, Y.: Robust attitude stabilization for nonlinear
quadrotor systems with uncertainties and delays. IEEE Transaction on
Industrial Electronics. 64(7), 5585-5594 (2017)
[34] Liu, H., Wang, X., Zhong, Y.: Quaternion-based robust attitude control
for uncertain robotic quadrotors. IEEE Transactions on Industrial Infor-
matics. 11(2) 406-415, (2015)
[35] Zhou, K., Doyle, J.C., Glover, K.: Robust and optimal control, New
Jersey, USA: Prentice Hall. 217-221 (1995)
[36] Braganza, D., Dixon, W.E., Xian, B.: Tracking control for robot manip-
ulators with kinematic and dynamic uncertainty. International Journal of
Robotics and Automation. 23(2), 117-126 (2008)
[37] Xi, J., He, M., Liu, H., Zheng, J.: Admissible output consensualization
control for singular multi-agent systems with time delays. Journal of the
Franklin Institute. 353(16), 4074-4090 (2016)
[38] Xi, J., Fan, Z., Liu, H., Zheng, T.: Guaranteed-cost consensus for
multiagent networks with Lipschitz nonlinear dynamics and switching
topologies. International Journal of Robust and Nonlinear Control. 28(7),
2841-2852 (2018)
[39] Xi, J., Wang, C., Liu, H., Wang, Z.: Dynamic output feedback
guaranteed-cost synchronization for multiagent networks with given cost
budgets. IEEE Access. 6(1), 28923-28935 (2018)
[40] Xi, J., Wang, C., Liu, H., Wang, L.: Completely distributed guaranteed-
performance consensualization for high-order multiagent systems with
switching topologies. IEEE Transactions on Systems, Man, and Cyber-
netics: Systems. PP(99), 1-11 (2018)
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2018.2878722, IEEE Access

IEEE ACCESS 13

Zhaoying Li received the B.E. degree in detection Zongyu Zuo (M’14-SM’18) received his B.Eng.
guidance and control from the Beihang Univer- degree in automatic control from Central South
sity, Beijing, China, in 2005, the Ph.D. degree in University, Hunan, China, in 2005, and Ph.D. degree
aerospace engineering from the Beihang University, in control theory and applications from Beihang
Beijing, China, in 2011. In 2008, she was a visiting University (BUAA), Beijing, China, in 2011. He was
student in the department of aerospace engineering, an academic visitor at the School of Electrical and
Iowa State University. Since 2011, she has been Electronic Engineering, University of Manchester
with the School of Astronautics, Beihang University, from September 2014 to September 2015 and held
Beijing, China, where she is currently a Lecturer. an inviting associate professorship at Mechanical
Her research interests include flight control, path Engineering and Computer Science, UMR CNRS
planning and trajectory optimization. 8201, Université de Valenciennes et du Hainaut-
Cambrésis in October 2015 and May 2017. He is currently an associate
professor at the School of Automation Science and Electrical Engineering,
Beihang University. His research interests are in the fields of nonlinear system
control, control of UAVs, and coordination of multi-agent system. Dr. Zuo
serves as an associate editor of the International Journal of Aeronautical &
Space Sciences, and the International Journal of Digital Signals and Smart
Systems.

Wenjie Zhou received his B.S. degree in Automa-

tion from Henan Polytechnic University, Henan,
China, in 2015, and M.E. degree in Guidance,
Navigation, and Control from Beihang University,
Beijing, China, in 2018. His research interest is pri-
marily in robust control, flight control, and quadrotor
control.

Hao Liu (S’13-M’14) received the B.E. degree in

control science and engineering from the Northwest-
ern Polytechnical University, Xian, China, in 2008,
the Ph.D. degree in automatic control from the Ts-
inghua University, Beijing, China, in 2013. In 2012,
he was a visiting student in the Research School of
Engineering, Australian National University. Since
2013, he has been with the School of Astronautics,
Beihang University, Beijing, China, where he is
currently an Associate Professor. He has also been
with the University of Texas at Arlington Research
Institute, Fort Worth, USA. He received the best paper award on IEEE ICCA
2018. His research interests include formation control, reinforcement learning,
robust control, nonlinear control, unmanned aerial vehicles, unmanned under-
water vehicles, and multi-agent systems. He serves as an associate editor of
the Transactions of the Institute of Measurement and Control.

Lixin Zhang received the B.S. degree in Guidance,

Navigation, and Control from Beihang University,
Beijing, China, in 2016. He is currently working to-
ward the M.E. degree in the School of Astronautics,
Beihang University, Beijing, China. His research
interest is primarily in robust control, nonlinear
control, and unmanned aerial vehicle control.

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.