See discussions, stats, and author profiles for this publication at: [Link]

net/publication/312109050

Modeling and Flight Control Simulation of a Quadrotor Tailsitter VTOL UAV

Conference Paper · January 2017

DOI: 10.2514/6.2017-1561

CITATIONS READS
32 2,759

5 authors, including:

Fu Zhang Lyu Ximin

The University of Hong Kong Sun Yat-sen University
144 PUBLICATIONS 6,724 CITATIONS 33 PUBLICATIONS 732 CITATIONS

SEE PROFILE SEE PROFILE

Haowei Gu
Hong Kong University of Science and Technology
12 PUBLICATIONS 469 CITATIONS

SEE PROFILE

All content following this page was uploaded by Fu Zhang on 31 July 2019.

The user has requested enhancement of the downloaded file.

 Modeling and Flight Control Simulation of A Quad
Rotor Tail-Sitter VTOL UAV
Fu Zhang∗, Ximin Lyu †, Ya Wang ‡, Haowei Gu ‡ and Zexiang Li §

Hong Kong University of Science and Technology, Hong Kong

This paper presents the modeling and simulation of a quad rotor tail-sitter VTOL UAV.
The VTOL UAV is configured with four rotors for full attitude control and thrust gener-
ation for both vertical and forward flight. In the modeling process, a full attitude wind
tunnel test was performed on the full scale UAV to capture the UAV aerodynamics. In our
model, we explicitly consider the degradation of motor thrust and torque at the presence
of forward speed, which is usually neglected in most multi-rotor models. Quaternion based
attitude representation is utilized to drive the dynamics of the aircraft. When monitoring
and commanding the UAV attitude, ZXY Euler angles are used to avoid the singularity
at pitching angle of 90 degrees of commonly used ZYX Euler angles. Other minor consid-
erations include motor dynamics, gyroscopic effect, etc. Such a model captures the UAV
kinematics, dynamics, aerodynamics and actuator dynamics, and serves a good reference
model for flight controller design and verification. As a testimony of the presented model,
an attitude controller with an altitude holding loop will be presented.

Nomenclature
p Position vector represented in earth frame, m
R Vehicle orientation w.r.t. earth frame
v Vehicle groundspeed vector represented in earth frame, m/s
ω Angular velocity vector represented in body frame, rad/s
m Mass, kg
It Inertia matrix of the UAV, kg · m2
Ft Force vector represented in earth frame, N
Mt Moment vector represented in body frame, N · m
CT Propeller thrust coefficient
CQ Propeller torque coefficient
w Wind speed vector represented in earth frame, m/s
u Airspeed vector represented in body frame, m/s
V Airspeed magnitude, m/s
αx Angle of attack, rad
β Sideslip angle, rad
CD Drag coefficient
CL Lift coefficient
CY Side force coefficient
Cl Rolling coefficient
Cm Pitching coefficient
Cn Yawing coefficient
S Reference area, m2
c̄ Characteristic length, m
ρ Air density, kg/m3
∗ Research Assistant Professor, Department of Electronic and Computer Engineering, HKUST, Clear Water Bay, Hong Kong
† Ph.D. Candidate, Department of Electronic and Computer Engineering, HKUST, Clear Water Bay, Hong Kong
‡ [Link] Student, Department of Electronic and Computer Engineering, HKUST, Clear Water Bay, Hong Kong
§ Professor, Department of Electronic and Computer Engineering, HKUST, Clear Water Bay, Hong Kong

1 of 13

American Institute of Aeronautics and Astronautics

 I. Introduction
Advances in the field of sensing, computing and actuation have allowed the cost-effective implementation
of small lightweight Unmanned Aerial Vehicle (UAVs). Miniature UAVs have attracted immense interests
from researchers throughout the world. New designs, configurations, and functional algorithms are being
actively developed in order to address the use of UAVs in a number of civilian applications such as aerial pho-
tography, agriculture, intelligence surveillance and reconnaissance (ISR) missions, mapping, remote sensing,
parcel delivery, and dangerous missions.
There are mainly two types of small-scale UAV platforms, namely, fixed-wing and rotary-wing UAVs.
Each type has its own advantages and drawbacks. Fixed-wing UAVs have long flight endurance and high
operational range, but at the same time, requires a runway or catapult to gain take-off speed. New take-off
methods such as ”throw-to-fly” have been developed for small, lightweight flying wings. But this creates a
major obstruction for automating UAV operations, and requires some level of operating skills. Therefore,
fixed-wing UAVs are usually used in application scenarios where the flight range is crucial and the related
facility can be made available easily. On the other hand, rotary-wing UAVs are highly maneuverable, able of
vertical take-off and landing, lending themselves ideal for autonomous operation and applications in confined
environments. Such superiority is nevertheless at the cost of higher power consumption than its fixed-wing
UAVs counterparts.
A natural way to obtain both power efficiency and maneuverability is combining the design of rotary-
wing and fixed-wing in one platform, called Vertical Take-Off and Landing (VTOL) UAV. Current lightweight
VTOL UAV implementations can be divided into four categories, namely, tilt-rotor, tilt-wing, extra-propulsion
and tail-sitter. A tilt-rotor VTOL UAV is configured with rotors that can be tilted by a servo motor, to
provide either the lift force during vertical flight (i.e. hovering, take-off and landing) or the thrust during
forward flight. Examples of tilt-rotor VTOL UAVs include FireFly6, Orange Hawk1 , IAI Panther, etc. The
concept of tilt-wing VTOL UAVs is similar to tilt-rotor except that the wing also tilts instead of the rotors
alone2–5 . When compared with tilt-rotor VTOL UAVs, the tilt-wing mechanism is more sensitive to cross
wind during vertical flight. It also requires a high torque servo motor to tilt the wing and rotors. The
third type of VTOL UAVs is referred as extra-propulsion, where it utilizes separate rotors for vertical and
forward flight6, 7 . As their actuation mechanisms for vertical and forward flights are separate, the design,
modeling and control of extra-propulsion VTOL UAVs are relatively simple. The main drawback of this
configuration is the inefficient use of the lifting rotors which will not be in operation during forward flight,
but add extra weight to the aircraft. A tail-sitter VTOL UAV differs from the extra-propulsion mechanism
in that it uses the same rotors for both vertical and forward flight, hence suffering from no such drawback
as extra-propulsion. It also differs from the tilt-rotor and tilt-wing mechanisms in that it involves no tilting
mechanisms, thus saving weight and significantly simplifying the aircraft design. However, the mechanically
simple configuration comes with difficulties in flight controller design and poor hovering performance against
cross wind. A more comprehensive review on the development of VTOL UAVs can be found in Saeed et al.8
This paper presents the modeling and simulation of a quad rotor tail-sitter VTOL UAV. As seen in
Figure 1, the VTOL UAV is configured with four rotors for full attitude control and thrust generation for
both vertical and forward flight. The four rotors are inclined by a small angle to gain more yawing (i.e.
normal to the plane of the four rotors) control. In the modeling process, a wind tunnel test was performed
on the full scale UAV (as opposed to a down-scaled model) to capture the UAV aerodynamics. In the wind
tunnel test, the static aerodynamic force and moment were measured at airspeed up to 19 m/s, angle of
attack from −180 to 180 degrees and sideslip angle from zero to 90 degrees. The raw data collected during
the wind tunnel tests is interpolated by fitting it to a group of carefully chosen basis functions, thus capturing
the full aerodynamic force and moment at any wind direction. In our model, we also explicitly consider the
degradation of motor thrust and torque at the presence of forward speed, which is usually neglected in most
multi-rotor models. When compared with conventional fixed-wing airplanes, quaternion based (as opposed
to Euler angles) attitude representation was utilized to avoid singularity. In order to avoid the singularity
at pitching angle of 90 degrees, ZXY Euler angles (as opposed to ZYX Euler angles in conventional airplane
analysis) are used for monitoring purpose and accepting attitude commands from operators if the UAV is
operating in attitude mode. Other minor considerations include motor dynamics, gyroscopic effect, etc.
Such a model captures the UAV kinematics, dynamics, aerodynamics and actuator dynamics, and serves a
good reference model for flight controller design and verification. As a testimony of the presented model, an
attitude controller with an altitude holding loop will be presented.

2 of 13

American Institute of Aeronautics and Astronautics

 Figure 1. The tail-sitter VTOL UAV prototype developed at HKUST.

II. Vehicle Dynamics

The equations of motion for the quad rotor VTOL UAV can break into four inter-related parts: airframe
aerodynamics, body kinematics, body dynamics and rotor dynamics.

A. Coordinate Systems
Coordinate systems used in this paper will take conventions from conventional fixed-wing aircrafts10, 11 . As
seen in Figure 2, the earth frame denoted by Fe (Oe , Xe , Ye , Ze ) is attached to the earth and is considered as
an inertial frame in analyzing the UAV motion. The three axes of the earth frame are respectively aligned
with the North, East and Down direction of the earth. The body frame denoted by Fb (CG, Xb , Yb , Zb )
is attached to the UAV body thus being non-inertial. The three axes of the body frame are respectively
the UAV front, right and down direction. The orientation of the body frame w.r.t. earth frame is time
varying and is denoted as R. In addition to the earth frame and body frame, a motor frame denoted as
(i) (i) (i) (i) (i)
Fm (Om , Xm , Ym , Zm ) is embedded to the i-th motor. This is necessary as the motor is inclined and is no
longer aligned with the body frame. As seen in Figure 3, the X axis of the motor frame is aligned with the
motor rotation axis and stays still with the motor stator (instead of rotating with the rotor). As a result,
(i)
the orientation of the i-th motor frame w.r.t. the body frame is fixed and can be denoted as Rm .

Xb
Xm(i)
Zb Zm(i)
CG
Ym(i)
Yb
CG
Zb Xb Yb
Oe
Ye
Xe
Ze

Figure 2. Earth frame versus body frame Figure 3. Body frame versus motor frame

3 of 13

American Institute of Aeronautics and Astronautics

B. Dynamics
The vehicle is considered to be rigid when modeling its dynamics. Vibration modes will be further considered
in designing advanced flight controllers but will be omitted in this paper. Applying the balance of linear
momentum and angular momentum theorem9 to the VTOL UAV yields its translational and rotational
dynamics:
mv̇ = Ft (1)
 (i)
  (i)

X si Ir σ̇i X si Ir σi
(i)  (i) 
It ω̇ + ω × (It ω) + Rm  0 +ω× Rm  0  = Mt (2)
 

0 0
where m is the mass; Ir is the moment of inertia of the rotor and propeller; It is the inertial matrix of
the aircraft; v is the linear velocity of the UAV center of mass relative to earth (i.e. the groundspeed) and
is represented in the earth frame; ω is the UAV angular velocity relative to earth and is represented in
body frame; σi is the rotation speed of i-th motor; si indicates the i-th motor rotating direction; Ft and
Mt are respectively the total force and moment applied on the UAV. When compared with Euler’s rotation
equations describing the rotation of a generic rigid body, Eq. (2) takes into account for the rotor gyroscopic
effect and motor inertia. It is worth mentioning that the motor inertia term will contribute a zero in the
UAV’s transfer function in yawing direction and should be compensated in designing flight controllers.
Forces exerted on the vehicle consist of aerodynamic force, propeller thrust and gravity. Assume the
thrust produced on i-th propeller is Ti , the total force can be rewritten as
     
X Ti 0
(i) 
Ft = R  Rm  0  + fa  +  0  (3)
    

0 mg

where fa is the aerodynamic force produced by the airframe. It should be noticed that fa is a vector in body
frame while Ft in earth frame. Similarly, the total moment Mt can be rewritten as
    
X Ti −si Qi
(i)  (i) 
Mt = r̂i Rm  0  + Rm  0  + Ma (4)
 

0 0

where Ma is the aerodynamic moment on the airframe, Qi is the torque produced on the i-th propeller and
ri is the position vector of each motor in body frame.
Eq. (1 – 4) is a generic model for a wide class of UAVs such as multi-copters and tail-sitter UAVs where
the rotors are arbitrarily located. It can also be extended with minor modifications to capture the model of
other UAVs and actuation mechanisms such as tilt-rotor, tilt-wing, elevons, etc.

C. Kinematics
When compared with the vehicle dynamics described in Eq. (1 - 2), the kinematics of the vehicle also breaks
into translation and rotation parts. The translation part describes the displacement (denoted as p) of the
UAV center of mass and is related to the velocity by

ṗ = v (5)

The rotation part describes the UAV attitude, which is denoted by its rotation matrix R and lies on
SO(3)12 ,
Ṙ = Rbω (6)
where the b· is an operation that converts a vector in R3 to a skew symmetric matrix in R3×3
 
0 −ωz ωy
ωb =  ωz 0 −ωx  (7)
 

−ωy ωz 0

4 of 13

American Institute of Aeronautics and Astronautics

 The rotation matrix R is also called the Direct Cosine Matrix (DCM) as each of its element is the cosine
of the angle between a unit vector and an axis13 . Parameterization using rotation matrix R is highly
redundant as it has nine parameters although the UAV attitude only requires three independent parameters
to parameterize14 . Indeed, the rotation matrix R is an element of SO(3) and is subject to

R ∈ SO(3) , {R ∈ R3×3 : RT R = I, detR = 1} (8)

A number of minimal dimensional (i.e. three) parameterization of the attitude kinematics have been
developed to eliminate the redundancy, such as Euler angles through three successive rotations, parame-
terization methods developed by Tsiotras et al 15 through two successive rotations, and angle-axis param-
eterization through one single rotation. Unfortunately, it has been proved that none of three-dimensional
parameterization methods can be singularity free14, 16 .
A good balance between singularity free and low redundancy is quaternion based parameterization, which
is singularity free while keeps only one parameter of redundancy. Readers can refer to Farrell17 for a full
description of quaternion. The attitude propagation using quaternion representation is
" #
1 −T
q̇ = ω (9)
2 ηI + b 
" #
η
where q = ∈ R4 , η ∈ R is the scalar part of the quaternion q while  ∈ R3 is the vector part of the

quaternion q.
In our simulation, the attitude is updated using Eq. (9) and converted to the DCM through

R = I + 2ηb 2
 + 2b (10)

It can be seen that the quaternion is ideal for computer simulation as it is singularity free. However, as
it keeps one parameter of redundancy, quaternion based parameterization is difficulty for visualization and
human interpretation. In conventional aircraft analysis, the airplane attitude is usually represented by ZYX
Euler angles10, 11 . Even though ZYX Euler angles are singular at pitching angle of 90 degrees, the singular
point is well avoided as conventional airplanes rarely operate in such a high pitching angle. However, the
tail-sitter VTOL UAV is designed to tilt itself from vertical direction to horizontal direction, resulting in a
large pitching angle nearly 90 degrees. Apparently, ZYX Euler angles is not a good choice in this situation.
In this paper, we propose to use ZXY Euler angles to represent the UAV attitude for the purpose of
monitoring and accepting operator’s attitude commands if the UAV is operating in attitude mode. ZXY
Euler angles refer to a sets of consecutive rotations where the initial frame first rotates along its Z axis by
angle Ψ, then rotates along the X axis of the new frame by angle Φ, and finally along the Y axis of the
newest frame by angle Θ. As each rotation is made with respect to the latest frame (i.e. intrinsic), the
resulting rotation matrix is
R = RZ (Ψ)RX (Φ)RY (Θ) (11)
It can be seen that when compared with ZYX Euler angles, ZXY Euler angles exchanged the order of
the last two rotations, which does not cause any difficulty in human interpretation but successfully shifts
the singular point from Θ = ± π2 to Φ = ± π2 . Fortunately, this singular point is rarely reached for tail-sitter
VTOL UAVs.
ZXY Euler angles is also called Pitch-Roll-Yaw (PRY) Euler angles which refer to a set of consecutive
rotations where the initial frame is first rotates along its Y axis (i.e. Pitch) by angle Θ, then rotates along
the X axis (i.e. Roll) of the initial frame by angle Φ, and finally rotates along the Z axis (i.e. Yaw) of the
initial frame by angle Ψ. As each rotation is made with respect to the initial frame (i.e. extrinsic), the
resulting rotation matrix is
R = RZ (Ψ)RX (Φ)RY (Θ) (12)
which is essentially the same with Eq. (11). ZXY Euler angles can be computed from the DCM using the
following formulas  
Θ = tan−1 −r 31
r33
Φ = sin−1 (r
32 )  (13)
−1 −r12
Ψ = tan r22

5 of 13

American Institute of Aeronautics and Astronautics

 Thrust coefficient x 10
-3 Torque coefficient
0.1 RPM:1000 RPM:1000
RPM:2000 RPM:2000
RPM:3000 RPM:3000
14
0.09 RPM:4000 RPM:4000
RPM:5000 RPM:5000
RPM:6000 RPM:6000
0.08 RPM:7000 RPM:7000
12
RPM:8000 RPM:8000
RPM:9000 RPM:9000

Torque coefficient [C Q]
0.07
Thrust coefficient [C T]

RPM:10000 RPM:10000
RPM:11000 10 RPM:11000
0.06 RPM:12000 RPM:12000
RPM:13000 RPM:13000
RPM:13999 RPM:13999
0.05 RPM:15000 8 RPM:15000
RPM:16000 RPM:16000
RPM:17000 RPM:17000
0.04 RPM:18000 RPM:18000
6
RPM:19000 RPM:19000
RPM:20000 RPM:20000
0.03

4
0.02

0.01 2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6
Advance Ratio [J] Advance Ratio [J]

Figure 4. The thrust (left) and torque (right) coefficient of APC95x45. The data is drawn from the APC
Performance Data Files22

D. Actuator Dynamics
Brushless DC Motors (BLDCs) are used to change the rotation speed of the four propellers. The model of
a typical three-phase BLDC can usually be expressed as follows.
dia
La = uVb − ia Ra − Kv σ (14)
dt
dσ
Ir = Kt ia − bω − Q (15)
dt
where La is the armature inductance, ia is the current on the armature, Ra is the armature resistance, Kv is
the back emf constant, Ir is the moment of inertia of the rotor and propeller, σ is the rotation speed in radian
per second of the rotor, Kt is the torque constant, b is the damping constant, Q is the torque produced on
the propeller, Vb is the battery voltage applied on the DC motor and u is the PWM duty inputed to the
ESC.
It can be seen that the motor dynamics is coupled with the propeller torque, which is in general nonlinear
to the rotation speed. In Brandt et al 19 , the propeller thrust and torque are respectively parameterized as
CT ρσ 2 D4
T = (16)
4π 2
CQ ρσ 2 D5
Q= (17)
4π 2
where ρ is the air density, D is the propeller diameter, CT and CQ are respectively the thrust and torque
coefficients of the propeller. As shown in Figure 4, the thrust and torque coefficient depend both on RPM
and advanced ratio, which is defined as follows
2πVf
J= (18)
σD
where Vf is the rotor forward speed (i.e. the speed perpendicular to rotor disk). In Bangura et al 20 and
Prouty21 , it shows that a H-force will be produced as a result of the rotor lateral speed. The H-force can
be expressed as
Cd
H= ρAb σDVl (19)
8
where Cd is the blade airfoil drag coefficient, Ab is the rotor blade surface area, and Vl is the rotor lateral
speed. In our simulation, we neglect the H-force but retain the effect of the forward speed on the propeller
aerodynamics. The propeller profile is preloaded to the model.

E. Aerodynamics
In preceding sections, the aerodynamic force and moment were referred as fa and Ma , which are functions
of the speed of the vehicle relative to its surrounding air mass (i.e. airpseed). Suppose that the vehicle speed

6 of 13

American Institute of Aeronautics and Astronautics

 CG

uy

L
𝛼𝑥
ux 𝛼𝑥
uz 𝛽
D
ux (body X)
𝛼𝑥

uxz
uz (body Z)
u
(a) Angle of attack and sideslip angle (b) Lift and drag

Figure 5. Aerodynamic nomenclatures

relative to earth is v (see Eq. (1)) and the wind speed relative to earth is w, the airspeed is therefore

u = RT (v − w) (20)

In Eq. (20), v and w are vectors in earth frame while u is a vector in body frame, hence the rotation
matrix R takes place. The purpose of expressing the airspeed in body frame is to ease the definition of angle
of attack and sideslip angle, which are directly related to the aerodynamic force and moment.
As seen in Figure 5, the angle of attack, denoted by αx , and sideslip angle β are defined as
q
V = u2x + u2y + u2z (21)
 
uz
αx = tan−1 (22)
ux
u 
y
β = sin−1 (23)
V
In conventional aircraft analysis10, 11 , the aerodynamic force is usually characterized by Lift force L,
Drag force D and Side force Y. While the Side force is along body Y axis, the Lift and Drag forces are
respectively along and perpendicular to the airspeed projection onto the body x − z plane, as seen in Figure
5. The resulting aerodynamic force fa , which is a vector in body frame, is therefore
  
− cos αx 0 sin αx D
fa =  0 1 0  Y  (24)
  

− sin αx 0 − cos αx L

The aerodynamic moment consisting of rolling L, pitching M and yawing moments N can be expressed
as  
L
Ma =  M  (25)
 

7 of 13

American Institute of Aeronautics and Astronautics

 (a) Front view (b) Rear view

Figure 6. Wind tunnel setup

In Bernard et al 11 , the aerodynamic forces and moments are further parameterized as

L = 12 ρV 2 SCL (26)
D = 12 ρV 2 SCD (27)
Y = 12 ρV 2 SCY (28)
L = 12 ρV 2 Sc̄Cl (29)
M = 21 ρV 2 Sc̄Cm (30)
N = 12 ρV 2 Sc̄Cn (31)

where S is the reference area, which is usually the wing area; c̄ is the characteristic length which is usually
the mean aerodynamic chord, V is the airspeed magnitude; CL , CD and CY are respectively the lift, drag
and side force coefficients; and Cl , Cm and Cn are respectively the rolling, pitching and yawing coefficients.
These coefficients are in general functions of airspeed magnitude V , angle of attack αx and sideslip angle
β. In order to characterize these coefficients, the full scale VTOL prototype was tested in a 3m×2m wind
tunnel located at HKUST.
Figure 6 shows the wind tunnel setup. The full scale VTOL UAV (propeller removed) is mounted on
a six-dimensional force sensor through a strut. The six-dimensional force sensor is capable of measuring
the three-dimensional forces and three-dimensional moments. Right underneath the UAV is a gimbal which
enables the adjustment of pitching angle (i.e. angle of attack). The force sensor is fixed on a rotation table
which enables the adjustment of sideslip angle. The airspeed is adjusted by changing the level of voltage
supplied to the fan. A pitot tube installed at the UAV shoulder is used to measure the airspeed. A number
of combinations of sideslip angle, angle of attack and airspeed were tested with sideslip angle from 0 to 90
degrees, angle of attack from -180 to 180 degrees and airspeed up to 19m/s, which is sufficiently high to
cover the UAV speed envelope. The tested data can effectively capture all the possible flight conditions of
the quad rotor tail-sitter VTOL UAV.
The raw data measured in wind tunnel test was interpolated by fitting it to a group of carefully chosen
basis functions. Figure 7 shows the fitted aerodynamic force and moment coefficients. The mean residual
error for lifting and drag coefficients are respectively 10% and 8% for all different level of airspeed and all
possible angle of attack and sideslip angle. A comparison between the raw data and the fitted data at
different airspeed level and sideslip angle can be seen in Figure 8 and 9.

III. Flight Simulator Development

A. Attitude Controller
In order to test the UAV model, an attitude controller was designed and tested in simulation. The attitude
controller is a nonlinear controller that operates on SO(3). Assume that the desired attitude trajectory is
Rd , which satisfies
Ṙd = Rd ω
bd (32)

8 of 13

American Institute of Aeronautics and Astronautics

 Lift coefficient Rolling coefficient

1.5

1
1

0.5 0.5
CL

0
0

Cl
-0.5
-0.5

-1
200
-1 150
-1.5 100 100
80
100 200 60 50
80 150 40
60 100 0
40 20
20 50 0 -50
0 0 -20
-20 -50 -40 -100
-40 -100
-60 -60 -150
-80 -150 -80
-100 -200 -100 -200

Y:  X: 
Y:  X: 

Drag coefficient Pitching coefficient

2
2.5

2
1.5
1.5

1 1
CD

Cm
0.5
0.5
0

-0.5
0
-1
200
-0.5 200 -1.5 150
100 150 100
80 100
80 100
60 60 50
40 50 40
20 20 0
0
0 0 -50
-20 -50 -20
-40 -100 -40 -100
-60 -150 -60 -150
-80 -80
-100 -200 -100 -200

Y:  X:  Y:  X: 

Side force coefficient Yawing coefficient

0.5
0.4
0.4

0.3 0.3

0.2 0.2

0.1 0.1
CY

0
Cn

0
-0.1
-0.1
-0.2
-0.2
-0.3
-0.3
-0.4 200
200 -0.4 150
-0.5 100 100
150
100 100 80 50
80 60
60 50 40 0
40 0 20
20 0 -50
0 -50 -20
-20 -100
-40 -100 -40
-60 -150 -60 -150
-80 -80
-100 -200 -100 -200

X:  Y:  X: 
Y: 

Figure 7. The interpolated aerodynamic coefficients

The attitude error is defined as

1
∨
RdT R − RT Rd eR = (33)
2
where the vee map ∨ is the inverse of the hat map b·. And the angular velocity error is defined as

eω = ω − RT RdT ωd (34)

The proposed attitude controller is a PD controller defined as follows

b RT Rd ωd − RT Rd ω̇d − Ma


u = −kR eR − kω eω + ω × (It ω) − It ω (35)

where kR and kω are scalars.

The attitude controller described in Eq. (35) is similar to Lee et al 18 except that an additional term
−Ma is used to cancel the aerodynamic moment. In Lee et al,18 it showed that the proposed controller is
exponential stable if the initial condition satisfies

1  λmax (It )keω (0)k2

tr I − RdT R < min{2, 4 −

} (36)
2 kR

9 of 13

American Institute of Aeronautics and Astronautics

  =0 2.9 m/s  =30 2.9 m/s
1.5 4.4 m/s 1.5 4.4 m/s
6.0 m/s 6.0 m/s
1 7.6 m/s 1 7.6 m/s
9.2 m/s 9.2 m/s
0.5 10.9 m/s 0.5 10.9 m/s
12.5 m/s 12.5 m/s

CL

CL
0 14.1 m/s 0 14.1 m/s
15.7 m/s 15.7 m/s
-0.5 17.3 m/s -0.5 17.3 m/s
18.9 m/s 18.9 m/s
-1 -1

-1.5 -1.5
-200 -150 -100 -50 0 50 100 150 200 -200 -150 -100 -50 0 50 100 150 200
 

 =60  =90
1 0.4
2.9 m/s 2.9 m/s
6.0 m/s 0.3 6.0 m/s
0.5 7.6 m/s 7.6 m/s
0.2
9.2 m/s 9.2 m/s
10.9 m/s 0.1 10.9 m/s
CL

CL
0 12.5 m/s 12.5 m/s
14.1 m/s 0 14.1 m/s
15.7 m/s 15.7 m/s
-0.1
-0.5 17.3 m/s 17.3 m/s
-0.2 18.9 m/s

-1 -0.3
-200 -150 -100 -50 0 50 100 150 200 -200 -150 -100 -50 0 50 100 150 200
 

Figure 8. The interpolated aerodynamic lift coefficient

 =0 drag
2.9 m/s  =30 2.9 m/s
2 4.4 m/s 2 4.4 m/s
6.0 m/s 6.0 m/s
7.6 m/s 7.6 m/s
1.5 9.2 m/s 1.5 9.2 m/s
10.9 m/s 10.9 m/s
12.5 m/s 12.5 m/s
CD

CD

1 14.1 m/s 1 14.1 m/s

15.7 m/s 15.7 m/s
17.3 m/s 17.3 m/s
0.5 18.9 m/s 0.5 18.9 m/s

0 0
-200 -150 -100 -50 0 50 100 150 200 -200 -150 -100 -50 0 50 100 150 200
 

 =60  =90 2.9 m/s

1.4 2.9 m/s 0.3
6.0 m/s 6.0 m/s
1.2 7.6 m/s 0.2 7.6 m/s
9.2 m/s 9.2 m/s
1 0.1 10.9 m/s
10.9 m/s
12.5 m/s 12.5 m/s
0.8 0
14.1 m/s
CD

CD

14.1 m/s
0.6 15.7 m/s -0.1 15.7 m/s
17.3 m/s 17.3 m/s
0.4 -0.2 18.9 m/s

0.2 -0.3

0 -0.4
-200 -150 -100 -50 0 50 100 150 200 -200 -150 -100 -50 0 50 100 150 200
 

Figure 9. The interpolated aerodynamic drag coefficient

B. Altitude Hold Controller

Besides the attitude controller, an altitude hold controller is developed to maintain the UAV altitude at the
desired one. Recall that the UAV translation motion can be described as follows
     
X Ti 0
(i) 
p̈ = R  Rm  0  + fa  +  0  (37)
    

0 mg

10 of 13

American Institute of Aeronautics and Astronautics

 Figure 10. The Simulink model for VTOL UAV

In our prototype, the four motors are symmetrically inclined such that
   
X Ti Tx
(i) 
Rm =
 0   0 (38)
  

0 0

where Tx is the total thrust along body X axis.

The altitude (denoted as h) dynamics is therefore
  
Tx
ḧ = r3T  0  + fa  + mg (39)
  

where R = [r1 , r2 , r3 ]. Assume the desired altitude is hd , the designed altitude actuation is computed from
a PID controller as follows
  
Tx Z
T  d
r3  0  + fa  + mg = kP (h − hd ) + kI (h − hd ) dt + kD (h − hd ) (40)
 
dt
0

The resulted thrust is therefore

d
(h − hd ) − mg − r3T fa
R
kP (h − hd ) + kI (h − hd ) dt + kD dt
Tx = (41)
r31
where r31 denotes the first entry of the vector r3

C. Flight Simulator
The UAV model, attitude and altitude holding controllers developed in proceeding sections are simulated in
MATLAB Simulink. As seen in Figure 10, the model is highly modularized. It consists of four sub blocks:
vehicle dynamics, flight controller and mixer, command block and visualization block. The command block
can accept either software generated attitude commands such as step, sine, etc. or attitude commands from

11 of 13

American Institute of Aeronautics and Astronautics

a joystick that is plugged into the computer via USB. The visualization block collects all the flight states
and plots them for post analysis. Oscilloscopes are also placed at places of interest in order to display the
flight states in real time.

IV. Results and Discussion

In order to test the VTOL model and the proposed attitude controller, different attitude commands were
fed to the simulated model. The attitude commands are the Pitch, Roll and Yaw angles in earth frame
and are interpreted as PRY Euler angles as defined in Eq. (12). Figure 11 shows the system responses
(upper part) and motor PWM duty (lower part) when the input command is 10 degrees pitching angle (i.e.
80 degrees of tilting from the initial hovering state). It can be seen that the actual attitude exponentially
converges to the attitude set-point while the altitude is successfully maintained at the desired one. The
simulated altitude drop during the transition is about 0.1m.

100
Pitc h (deg)

Set Point
Response
50

0
0 5 10 15 20 25 30 35 40 45 50

0.05
Roll (deg)

-0.05
0 5 10 15 20 25 30 35 40 45 50

0.2
Yaw ( deg)

0.1

-0.1
0 5 10 15 20 25 30 35 40 45 50

-9.8
Altitude ( m)

-9.9

-10

-10.1
0 5 10 15 20 25 30 35 40 45 50
Time (s)

50
Motor 1

40

30
0 5 10 15 20 25 30 35 40 45 50

50
Motor 2

40

30
0 5 10 15 20 25 30 35 40 45 50

40
Motor 3

30

20

0 5 10 15 20 25 30 35 40 45 50

40
Motor 4

30

20

0 5 10 15 20 25 30 35 40 45 50
Time (s)

Figure 11. System response when commanded a pitching angle of 10 degrees

Figure 12 shows the flight speed and required motor thrust for different pitching angles. It can be seen
that as the pitching angle increases, the required thrust first decreases and then increases. The minimal
thrust is achieved at around 12 degrees of pitching angle. On the other hand, the flight speed decreases
monotonically as pitching angle increases. Unlike conventional aircrafts, with thrust-weight-ratio lager than
one, the UAV is capable of flying at extremely high angle of attack where stall occurs. During this regime,
the motor thrust provides the main lift. Although flying at this regime is not necessarily fuel optimal, it
considerably extends the UAV’s speed envelope.

12 of 13

American Institute of Aeronautics and Astronautics

 20 15

Velocity
Thrust

10 10
Velocity (m/s)

Thrust (N)
0 5

-10 0
0 10 20 30 40 50 60 70 80 90
Pitch (deg)

Figure 12. The required motor thrust and vehicle speed when flying at different pitching angle

References
1 Carlson,
Stephen. ”A hybrid tricopter/flying-wing vtol uav.” American Institute of Aeronautics and Astronautics (2014).
2 Holsten,
J., T. Ostermann, and D. Moormann. ”Design and wind tunnel tests of a tiltwing UAV.” CEAS Aeronautical
Journal 2.1-4 (2011): 69-79.
3 Cetinsoy, Erturul, et al. ”Design and development of a tilt-wing UAV.” Turkish Journal of Electrical Engineering &

Computer Sciences 19.5 (2011): 733-741.

4 Muraoka, Koji, Noriaki Okada, and Daisuke Kubo. ”Quad tilt wing VTOL UAV: aerodynamic characteristics and prototype

flight test.” AIAA Infotech@ Aerospace Conference and AIAA Unmanned Unlimited Conference. 2009.
5 Dickeson, Jeffrey J., et al. ”Robust LPV H-infinity gain-scheduled hover-to-cruise conversion for a tilt-wing rotorcraft in

the presence of CG variations.” Decision and Control, 2007 46th IEEE Conference on. IEEE, 2007.
6 Jump-20. Arcturus UAV. Available at [Link]
7 Amazon. Prime Air. Available at [Link]/b?node=8037720011
8 Saeed, Adnan S., et al. ”A review on the platform design, dynamic modeling and control of hybrid UAVs.” Unmanned

Aircraft Systems (ICUAS), 2015 International Conference on. IEEE, 2015.

9 O’Reilly, Oliver M. Engineering Dynamics: A Primer. Springer Science & Business Media, 2010.
10 Nelson, Robert C. Flight stability and automatic control. Vol. 2. WCB/McGraw Hill, 1998.
11 Etkin, Bernard, and L. Duff Reid. Dynamics of flight. Vol. 2. New York: Wiley, 1959.
12 Murray, Richard M., et al. A mathematical introduction to robotic manipulation. CRC press, 1994.
13 Marktey, F. L. ”Three-Axis Attitude Determination Methods.” Spacecraft Attitude Determination and Control. Vol. 1.

D. Reidel, Dordrecht, Holland, 1978.

14 Schaub, Hanspeter. Analytical mechanics of space systems. Aiaa, 2003.
15 Tsiotras, Panagiotis, and James M. Longuski. ”A new parameterization of the attitude kinematics.” Journal of the

Astronautical Sciences 43.3 (1995): 243-262.

16 Fraiture, Luc. ”Regularization of Minimum Parameter Attitude Estimation.” Journal of Guidance, Control, and Dynam-

ics 32.3 (2009): 1029-1034.

17 Farrell, Jay. Aided navigation: GPS with high rate sensors. McGraw-Hill, Inc., 2008.
18 Lee, Taeyoung, Melvin Leoky, and N. Harris McClamroch. ”Geometric tracking control of a quadrotor UAV on SE (3).”

Decision and Control (CDC), 2010 49th IEEE Conference on. IEEE, 2010.
19 Brandt, John B., and Michael S. Selig. ”Propeller performance data at low reynolds numbers.” 49th AIAA aerospace

sciences meeting. 2011.

20 Bangura, Moses, Marco Melega, Roberto Naldi, and Robert Mahony. ”Aerodynamics of Rotor Blades for Quadrotors.”

arXiv preprint arXiv:1601.00733 (2016).

21 Prouty, Raymond W. Helicopter performance, stability, and control. 1995.
22 APC. APC Performance Data Files. Available at [Link]/v/PERFILES WEB/[Link]

13 of 13

American Institute of Aeronautics and Astronautics

View publication stats