Author manuscript, published in "Flight Formation Control, Josep Guerrero (Editor), Rogelio Lozano (Editor) (Ed.

) (2012) 99-134"
hal-00923127, version 1 - 2 Jan 2014

UAV Flight Formation Control

Jose Alfredo G UERRERO

Rogelio L OZANO

Version 0.5,
 hal-00923127, version 1 - 2 Jan 2014

ii
 Contents

Chapter 1. Modeling and Control of Mini UAV . . . . . . . . . . . . . . . . . 1

hal-00923127, version 1 - 2 Jan 2014

G. Flores , J.A. Guerrero , J. Escareno , R. Lozano

1.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2. The Quad-Rotor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.1. Dynamical Model: Euler-Lagrange Approach . . . . . . . . . . . 4
1.3. Control of a Mini Tail-Sitter . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.1. Linear Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1.1. Roll control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.1.2. Pitch control . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3.1.3. Yaw control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.1.4. Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.1.5. Experimental Results . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2. Robust Control Considering Parametric Uncertainty . . . . . . . . 17
1.3.2.1. Pitch Subsystem . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3.2.2. Yaw Subsystem . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.3.2.3. Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.3.3. Nonlinear Control based on Nested Saturations . . . . . . . . . . . 23
1.3.3.1. Equations of Translational Motion . . . . . . . . . . . . . . . 25
1.3.3.2. Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . 30
1.4. Quad-Tilting Rotor Convertible MAV . . . . . . . . . . . . . . . . . . . 30
1.4.1. Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.4.2. Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.4.3. Control strategy for hover flight mode . . . . . . . . . . . . . . . . 41
1.4.4. Control strategy for forward flight mode . . . . . . . . . . . . . . . 43
1.4.5. Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
1.5. Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
1.6. Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

ix
 hal-00923127, version 1 - 2 Jan 2014

x
 Chapter 1

Modeling and Control of Mini UAV

hal-00923127, version 1 - 2 Jan 2014

This Chapter deals with the modeling and control of different configurations of
the UAVs, and is organized as follows. Section 1.2 gives a general overview of the
quad-rotor aerial vehicle and its operation principle. The modeling is presented using
the Euler-Lagrange approach. Sections 1.3 and 1.4 deal with the Hybrid or Con-
vertible MAVs, combining the advantages of horizontal and vertical flight. Different
approaches for non-linear control are presented using the Lyapunov theory. Finally,
some concluding remarks are presented in Section 1.5.

1.1. Introduction

The applications of mini Unmanned Aerial Vehicles (UAVs) comprise both mili-
tary and civilian, though the latter has had a lower development rate. The use of aerial
robots, specially miniature (mini and micro) UAVs (MAVs), has enhanced activities
such as surveillance of sensible areas (borders, harbors, prisons), wildlife study, nat-
ural disasters assessment, traffic surveillance, pollution monitoring, just to mention a
few. However, there are missions whose scope is beyond the capabilities of conven-
tional small UAVs designs since they require not only longer flight endurance but also
hovering/VTOL capabilities. Besides the commonly used aerial vehicles, the Hybrid
or Convertible MAVs, have been gaining popularity recently. By marrying the take-
off and landing capabilities of the helicopter with the forward flight efficiencies of
fixed-wing aircraft, the Convertible UAV promise a unique blend of capabilities at
lower cost than other UAV configurations. Two kinds of Convertible UAV vehicles
are discussed: the Bidule mAV and the Quad-plane in section 1.3 and 1.4 respectively.

Chapter written by G. Flores and J.A. Guerrero and J. Escareno and R. Lozano.

1
 2 UAV Flight Formation Control

The complete dynamics of these kind of vehicles, taking into account aero-elastic
effects, flexibility of the wings, internal dynamics of the engine and the whole set of
changing variables are quite complex and somewhat unmanageable for the purposes
of control. Therefore, it is interesting to consider a simplified model of an aircraft
formed by a minimum number of states and inputs, but retaining the main features
that must be considered when designing control laws for a real aircraft.

The development of the simplified model of the common Quadrotor will be pre-
sented and it will be used throughout the entire book.

1.2. The Quad-Rotor

The quad-rotor mini-rotorcraft is controlled by the angular speeds of four electric

motors as shown in Figure 1.1. Each motor produces a thrust and a torque, whose
hal-00923127, version 1 - 2 Jan 2014

combination generates the main thrust, the yaw torque, the pitch torque, and the roll
torque acting on the quad-rotor. Conventional helicopters modify the lift force by

Figure 1.1: The quad-rotor in an inertial frame. f1 , f2 , f3 , f4 represent the thrust of each motor, ψ, θ and φ represent the Euler angles,
and u is the main thrust.

varying the collective pitch. Such aerial vehicles use a mechanical device known
as swashplate. This system interconnects servomechanisms and blade pitch links in
order to change the rotor blades pitch angle in a cyclic manner, so as to obtain the
pitch and roll control torques of the vehicle. In contrast, the quad-rotor does not have
a swashplate and has constant pitch blades. Therefore, in a quad-rotor we can only
 Modeling and Control of Mini UAV 3

vary the angular speed of each one of the four rotors to obtain the pitch and roll control
torques.

From Figure 1.1 it can be observed that the motor Mi (for i = 1, . . . , 4) pro-
duces the force fi , which is proportional to the square of the angular speed, that is
fi = kwi2 . Given that the quad-rotor’s motors can only turn in a fixed direction,
the produced force fi is always positive. The front (M1 ) and the rear (M3 ) motors
rotate counter-clockwise, while the left (M2 )and right (M4 ) motors rotate clockwise.
With this arrangement, gyroscopic effects and aerodynamic torques tend to cancel in
trimmed flight. The main thrust u is the sum of individual thrusts of each motor. The
pitch torque is a function of the difference f1 − f3 , the roll torque is a function of
f2 − f4 , and the yaw torque is the sum τM1 + τM2 + τM3 + τM4 , where τMi is the
reaction torque of motor i due to shaft acceleration and blades drag. The motor torque
is opposed by an aerodynamic drag τdrag , such that
hal-00923127, version 1 - 2 Jan 2014

Irot ω̇ = τMi − τdrag (1.1)

where Irot is the moment of inertia of a rotor around its axis. The aerodynamic drag
is defined as
1
τdrag = ρAv 2 (1.2)
2
where ρ is the air density, the frontal area of the moving shape is defined by A, and v
is its velocity relative to the air. In magnitude, the angular velocity ω is equal to the
linear velocity v divided by the radius of rotation r
v
ω= (1.3)
r
The aerodynamic drag can be rewritten as

τdrag = kdrag ω 2 (1.4)

where kdrag > 0 is a constant depending on the air density, the radius, the shape of
the blade and other factors. For quasi-stationary manoeuvres, ω is constant, then

τMi = τdrag (1.5)

Forward pitch motion is obtained by increasing the speed of the rear motor M3 while
reducing the speed of the front motor M1 . similarly, roll motion is obtained using the
left and right motors. Yaw motion is obtained by increasing the torque of the front
and rear motors (τM1 and τM3 respectively) while decreasing the torque of the lateral
motors (τM2 and τM4 respectively). Such motions can be accomplished while main-
taining the total thrust constant. The quad-rotor model is obtained by representing the
aircraft as a solid body evolving in a three dimensional space and subject to the main
thrust and three torques: pitch, roll and yaw.
 4 UAV Flight Formation Control

1.2.1. Dynamical Model: Euler-Lagrange Approach

Let the generalized coordinates of the rotorcraft be expressed by

q = (x, y, z, ψ, θ, φ) ∈ R6 (1.6)

where ξ = (x, y, z) ∈ R3 denotes the position vector of the center of mass of the quad-
rotor relative to a fixed inertial frame I. The rotorcraft’s Euler angles (the orientation
of the rotorcraft) are expressed by η = (ψ, θ, φ) ∈ R3 , ψ is the yaw angle around
the z-axis, θ is the pitch angle around the y-axis and φ is the roll angle around the
x-axis (see [ETK 96]).An illustration of the generalized coordinates of the rotorcraft
is shown in Figure ??. Define the Lagrangian
hal-00923127, version 1 - 2 Jan 2014

L(q, q̇) = Ttrans + Trot − U (1.7)

where Ttrans = m T 1 T
2 ξ̇ ξ̇ is the translational kinetic energy, Trot = 2 Ω IΩ is the
rotational kinetic energy, U = mgz is the potential energy of the rotorcraft, z is
the rotorcraft altitude, m denotes the mass of the quad-rotor, Ω is the vector of the
angular velocity, I is the inertia matrix and g is the acceleration due to gravity. The
angular velocity vector ω resolved in the body fixed frame is related to the generalized
velocities η̇ (in the region where the Euler angles are valid) by means of the standard
kinematic relationship [GOL 83]

Ω = Wη η̇ (1.8)

where

 
− sin θ 0 1
Wη =  cos θ sin φ cos φ 0  (1.9)
cos θ cos φ − sin φ 0

then

 
φ̇ − ψ̇ sin θ
Ω =  θ̇ cos φ + ψ̇ cos θ sin φ  (1.10)
ψ̇ cos θ cos φ − θ̇ sin φ
 Modeling and Control of Mini UAV 5

Define

J = J(η) = WηT IWη (1.11)

where

 
Ixx 0 0
I =  0 Iyy 0  (1.12)
0 0 Izz

so that
hal-00923127, version 1 - 2 Jan 2014

1 T
Trot = η̇ J η̇ (1.13)
2

Thus, the matrix J = J(η) acts as the inertia matrix for the full rotational kinetic
energy of the quad-rotor, expressed directly in terms of the generalized coordinates η.

The model of the full rotorcraft dynamics is obtained from Euler-Lagrange equa-
tions with external generalized forces

   
d ∂L ∂L Fξ
− = (1.14)
dt ∂ q̇ ∂q τ

where Fξ = RF̂ ∈ R3 is the translational force applied to the rotorcraft due to main
thrust, τ ∈ R3 represents the yaw, pitch and roll moments and R denotes the rotational
matrix. R(ψ, θ, φ) ∈ SO(3) represents the orientation of the aircraft relative to a
fixed inertial frame

 
cθ cψ cψ sθ sφ − cφ sψ sφ sψ + cφ cψ sθ
R =  cθ sψ cφ cψ + sθ sφ sψ cφ s θ s ψ − cψ s φ  (1.15)
−sθ cθ sφ cθ cφ

where cθ stands for cos θ and sθ for sin θ. From Figure 1.1, it follows that

 
0
F̂ =  0  (1.16)
u
 6 UAV Flight Formation Control

where u is the main thrust directed out of the bottom of the aircraft and expressed as

4
X
u = fi (1.17)
i=1

and, for i = 1, . . . , 4, fi is the force produced by motor Mi , as shown in Figure 1.1.

Typically fi = kωi2 , where ki is a constant and ωi is the angular speed of the i-th
motor. The generalized torques are thus

   P4 
τψ i=1 τMi
τ =  τθ  ,  (f2 − f4 ))`  (1.18)
τφ (f3 − f1 ))`
hal-00923127, version 1 - 2 Jan 2014

where ` is the distance between the motors and the center of gravity, and τMi is the
moment produced by motor Mi , for i = 1, . . . , 4, around the center of gravity of the
aircraft.

Since the Lagrangian contains no cross terms in the kinematic energy combining ξ̇
with η̇, the Euler-lagrange equation can pe partitioned into dynamics for ξ coordinates
and η coordinates. The Euler-Lagrange equation for the translational motion is

 
d ∂Ltrans ∂Ltrans
− = Fξ (1.19)
dt ∂ ξ̇ ∂ξ

then

mξ̈ + mgEz = Fξ (1.20)

As for the η coordinates it can be written

 
d ∂Lrot ∂Lrot
− = τ (1.21)
dt ∂ η̇ ∂η

or

 
d ∂ η̇ 1 ∂
η̇ T J η̇ T J η̇ =


− τ (1.22)
dt ∂η 2 ∂η
 Modeling and Control of Mini UAV 7

Thus one obtains

1 ∂
J η̈ + J˙ η̇ − η̇ T J η̇


(1.23)
2 ∂η

Defining the Coriolis-Centripetal vector

1 ∂
J˙ η̇ − η̇ T J η̇


V̄ (η, η̇) = (1.24)
2 ∂η

one writes
hal-00923127, version 1 - 2 Jan 2014

J η̈ + V̄ (η, η̇) = τ (1.25)

but V̄ (η, η̇) can be expressed as

 
1 ∂
V̄ (η, η̇) = J˙ − (η̇ T J) η̇
2 ∂η
= C(η, η̇)η̇ (1.26)

where C(η, η̇) is referred to as the Coriolis terms and contains the gyroscopic and
centrifugal terms associated with the η dependence of J. This yields

mξ̈ + mgEz = Fξ (1.27)

J η̈ = τ − C(η, η̇)η̇ (1.28)

To simplify lets make

 
τ̃ψ
τ̃ =  τ̃θ  = J −1 (τ − C(η, η̇)η̇) (1.29)
τ̃φ
 8 UAV Flight Formation Control

Finally one obtains

mẍ = u(sin φ sin ψ + cos φ cos ψ sin θ) (1.30)

mÿ = u(cos φ sin θ sin ψ − cos ψ sin φ) (1.31)
mz̈ = u cos θ cos φ − mg (1.32)
ψ̈ = τ̃ψ (1.33)
θ̈ = τ̃θ (1.34)
φ̈ = τ̃φ (1.35)

where x and y are coordinates in the horizontal plane, z is the vertical position, and τ̃ψ ,
τ̃θ and τ̃φ are the yawing moment, pitching moment and rolling moment, respectively,
hal-00923127, version 1 - 2 Jan 2014

which are related to the generalized torques τψ , τθ , τφ .

1.3. Control of a Mini Tail-Sitter

Tail-sitter vehicles represent a configuration of aircraft that remains relatively

unexplored. Tail-sitters have more operational flexibility than conventional UAVs
because a vertical airframe attitude is adopted during take-off and landing, while
maintaining a horizontal airframe attitude during cruise just like conventional air-
planes. Tail-sitters have not been as widely adopted as an aircraft configuration due to
complex flight dynamics in the hover mode, making them typically very difficult to
control.

The Convair XF-Y1 and Lockheed XF-V1 were examples of experimental Tail-
sitters aircraft in the 1950s, but they were unsuccessful mostly due to the problem
caused by the awkward position of pilot required during the vertical flight phases,
which would not be relevant for UAVs. In the 1990s Boeing presented its tail-sitter
Heliwing UAV with a flight controller using cyclic-pitch rotor control for its vertical
flight phases [CAS 05], while more recently in [STO 02a], the University of Sydney’s
T-Wing UAV has an autopilot which uses control surfaces in the slipstream of fixed-
pitch propellers for control in its vertical flight phases.

In recent years, interest in Vertical Take-Off and Landing (VTOL) mini Air vehi-
cles (mAVs) have increased significantly due to a desire to operate UAVs in an urban
environment. Many concepts have been proposed globally [BLY 06]. The Bidule
mAV was developed at the University of Sydney to explore design issues related to
small flight platforms [SPO 01]. The latest version, the Bidule CSyRex, is a joint
project between the University of Sydney and the University of Technology of Com-
piègne to develop a VTOL variant of the Bidule. The vertical flight schematic of
 Modeling and Control of Mini UAV 9

this VTOL vehicle is shown in Figure ??, which is basically a fixed wing tailless
aircraft with two propellers. In hover mode, the altitude is controlled with the col-
lective thrust, this means, the lift force is generated increasing the lift force produced
by the propellers. The pitch attitude angular displacement is achieved by moving
the elevons in the same direction. The vertical yaw-attitude angular displacement is
achieved through moving the elevons in opposing direction. The vertical roll-attitude
angular displacement is controlled by changing the pitch angle of the Variable Pitch
Propeller (VPP). Typically, mAVs, such as the Planar Vertical Takeoff and Landing
(PVTOL) platforms [CAS 05] modify the speed of the DC electric motors to effect
altitude and attitude control. But when, brushless electric AC motors are used con-
trol responses have been too slow due to the time delay produced by the available
speed controllers, leading to problems utilizing motor speed for roll control. VPP is
thus being investigated as a potential solution, increasing the control response. This
allows to implement a simple flight controller without considering the time-delay in
actuators.
hal-00923127, version 1 - 2 Jan 2014

F2

F1 n

ψ
θ
nR
m vo
Ele l
nL
vo X-Axis
Y-Axis Ele (body)
(body)

Z-Axis
(body)

Figure 1.2: Vehicle schematic for vertical

flight mode of Bidule

1.3.1. Linear Control Strategy

In this section, the main purpose is to control the attitude of the VTOL in hover
flight. Therefore only the kinematic and moment equations will be used to obtain
three decoupled attitude system for the pitch, roll and yaw angular position. The
vehicle main wing has a profile NACA0008.
 10 UAV Flight Formation Control

Three decoupled stability augmentation control systems for the roll, the pitch, and
the yaw positions of the vehicle in hover flight are developed. These subsystems will
be obtained using only the kinematics and moment equations from the general model.
Several aerodynamic factors will be taken into account to obtain the transfer function
that represents the dynamic of each system.

1.3.1.1. Roll control

To obtain the roll control system, it is assumed that the pitch and yaw rates are zero.
Then, the vehicle can be analyzed in a similar manner to a PVTOL flight platform, as
in [CAS 05]. This configuration is shown in Figure ??. Therefore, using the equations
(??) and (??), the rotational dynamics for the roll angle can be represented by:

φ̈ = `/Jx (1.36)

where, the sum of moments ` can be calculated as follows:

hal-00923127, version 1 - 2 Jan 2014

` = F · d − C`φ̇ φ̇ (1.37)

and F = f1 − f2 is the force difference between the right and left rotor and d is the
distance from the center of mass to each rotor. The second term in the right side of
equation (1.37) represents an aerodynamic moment produced by the change of the roll
rate, normally opposing to the roll moment, that is why, the derivative, C`φ̇ = 0.36, is
known as roll damping derivative. Then equation (1.36) can be rewritten as follows:

φ̈ = (F · d − C`φ̇ φ̇)/Jx (1.38)

f1 f2
F = f1 − f 2

φ
d

Figure 1.3: Approach of PVTOL to control

the roll position

The lift force in each rotor can be considered as the thrust and can be calculated
by the following expression:
T = Ct ρn2 Dp4 (1.39)
where Ct is the thrust coefficient, ρ is the density of the air, n is the number of revo-
lutions per second of the motor and Dp is the diameter of the propellers. The thrust
 Modeling and Control of Mini UAV 11

coefficient is a function of the pitch angle propeller ϕ, which is shown in Figure ??.
The thrust coefficient in a linear region can be calculated by:

Ct = Ctϕ ϕ (1.40)

where Ctϕ is a derivative which represents the thrust slope with respect to the VPP
angle. This derivative has been estimated using a shareware program called JavaProp
[HEP 06]. This program uses the number of blades, the velocity of rotation, the diam-
eter of the propellers, the velocity and the power of the motor to give the value of
Ct for an operational range 5◦ ≤ ϕ ≤ 15◦ as is shown in Figure ??. Then, using
MatlabT M a first order polynomial (dashed line) can be constructed using the values
of the thrust coefficient for each ϕ angle. The dashed line slope is the derivative of
this polynomial which in fact represents the derivative Ctϕ , and its value is estimated
to be 0.0025.
hal-00923127, version 1 - 2 Jan 2014

vrotational
Bearing

Ball bearing joint

Brushless

ϕ motor

m propeller
mservo
Servo
Futaba

Figure 1.4: Schematic of Variable Pitch Pro-

peller (VPP) System

Derivative of Ct with respect to VPP angle

0.085

0.08

0.075

0.07
Ct

0.065

0.06

0.055

0.05
5 6 7 8 9 10 11 12 13 14 15

Ct at different VPP angles VPP Angle (Deg)

Then using the inertia values given in Table 1.1 and applying the Laplace trans-
form, the following transfer function for the roll angle with respect to the VPP angle
displacement is obtained.
φ(s) 5
= 2 (1.41)
ϕ(s) s + 25s
 12 UAV Flight Formation Control

Now, the VPP dynamics will be determined. In the Figure ??, it can be seen that
the aerodynamic pitch moment of the blades must be equal to the moment gener-
ated by the servo mechanism. Considering that the blade profile corresponds to the
NACA0014, then the following approximation can be used to obtain the blade pitch
moment:
ρVt2b Sb cb  
mb = Cmϕ ϕ + Cmϕ̇ ϕ̇ = ks fs δs (1.42)
2Jyb
where the subscript b denotes the blade. The term Vtb denotes the total velocity of the
propeller at the tip, it is given by:
q
2
Vtb = vaxial 2
+ vradial (1.43)

where vradial = πnD.

hal-00923127, version 1 - 2 Jan 2014

The term Cmϕ = −0.0019, is the estimated blade pitch moment coefficient slope
with respect to ϕ, being obtained using Javafoil [HEP 06], an airfoil analysis share-
ware software. The term Cmϕ̇ = 1.6 × 10−5 is a stability derivative generated by
the variation of the VPP rate. The right hand side term of equation (1.42) represents
the moment produced by the servo, where fs is the force produced by the servo, δs is
the servo displacement and ks is a mechanical reduction factor. Using the parameter
values in Table 1.1 and applying the Laplace transform, yields the VPP dynamic’s
transfer function:
ϕ(s) 120
= (1.44)
δs (s) s + 120
The actuator dynamics is given in [KAN 01] as follows:

δs (s) 0.6
= (1.45)
δc (s) 0.1s + 1
Then using the transfer functions given previously, the control loop system shown
in Figure ?? is proposed to stabilize the roll angle. The system is stable since the
characteristic equation 0.1s4 + 15.5s3 + 445s2 + 4260s + 16200 has all its roots in
the left hand side of the complex plane, the roots are located at: −6.7 ± 4.5i, −121
and −20.5.

φref PID
0.6 120 5
φ
0.1s+1 s + 120 s 2 + 25s
PD Controller Servo Roll Angle
Kp=45, Kd=3.5 VPP dynamics
Transfer Function

Figure 1.5: Roll

Control Loop.
 Modeling and Control of Mini UAV 13

1.3.1.2. Pitch control

To obtain the pitch control system the vehicle is considered to be a tailless aircraft
flying in forward flight. Assuming that the roll angle is small enough and the roll rate
is instantaneously zero, then using (??) and (??), a second order differential equation
describing the rotational dynamics for the platform pitch angle can be written as:

θ̈ = m/Jy (1.46)

where m is the pitch moment of the wing, which is given by the following expression:

1 2
m= V ρScCm (1.47)
2
where c̄ is the wing chord, S is the wing reference area, V is the airflow speed and
Cm is the pitching moment coefficient given by [ETK 96]:
hal-00923127, version 1 - 2 Jan 2014

c
Cm = Cmac + Cmα α + Cmδe δe + Cmq Q (1.48)
V
Assuming that in steady hover flight θ = α and Cmac = 0, then (1.46) can be reduced
to:
ρV 2 Sc
 
c
θ̈ = Cmα θ + Cmδe δe + Cmq θ̇ (1.49)
2Jy V
The derivative Cmα represents the variation of the pitching moment with respect to
the angle-of-attack α. This coefficient depends strongly on the airfoil profile. The
derivative Cmδe = ∂Cm /∂q represents the variation of the pitching moment with
respect to the elevator control. To estimate these parameters, a shareware software
named JavaFoil has been used. This program allows the user to analyze and design,
in a rapid and interactive way, a profile over a range of angles of attack, [HEP 06].
The vehicle main wing has a profile NACA0018 and its pitch moment curves at dif-
ferent angles of attack and elevator positions obtained with this program, are shown
in Figure ??. Then Cmα = −0.145 and Cmδe = 0.65. The derivative, Cmq = −10,
represents the aerodynamic effects due to rotations of the vehicle while the angle of
attack remains zero. Using the vehicle parameters given in Table 1.1 and the Laplace
transform, a second order transfer function representing the pitch angle dynamics is
given as follows:
θ(s) 85
= 2 (1.50)
δe (s) s + 40s + 18
Then, using the actuator dynamics given previously, a simple proportional derivative
compensator with Kp = 80 and Kd = 17, is proposed to stabilize the pitch angle.
This controller stabilizes the platform pitch angle system because the roots of the
characteristic equation 0.1s3 + 5s2 + 908.8s + 4098 are located at −22.7 ± 91.45i
and −4.62 which are in the left hand side of the complex plane.
 14 UAV Flight Formation Control

0.1

0.08
δ e =−5
0.06

0.04
∆ Cm
0.02
δ e =0

Cm
0

−0.02
∆ Cm
−0.04
δ e =5
−0.06

−0.08

−0.1
−15 −10 −5 0 5 10 15
Angle of attack

Figure 1.6: Pitch moment

coefficient curves

1.3.1.3. Yaw control

hal-00923127, version 1 - 2 Jan 2014

Now, to control the vehicle yaw position, it is assumed that the pitch and roll angles
are stabilized, then the roll rate and pitch rate vanish, then equation (??) can be written
as follows:
ψ̈ = n/Jz (1.51)
where n is the vehicle yaw moment. Notice that n is used to control yaw during
hover flight and to control roll during forward flight as shown in Figure ??. Under this
assumption, the yaw moment can be approximated by the following expression
n = ρV 2 SbCn /4 (1.52)
where b is the wing span and Cn is the yawing moment coefficient given by Cn =
Cnψ̇ ψ̇ + Cnδe δe . Then, (1.51) can be rewritten as

ψ̈ = (ρV 2 Sb)(Cnψ̇ ψ̇ + Cnδe δe )/4Jz (1.53)

where Cnδe = 0.19 represents the variation of the yaw moment with respect to the
ailerons positions. Cnψ̇ = 0.19 is the yaw damping derivative. Applying Laplace
transform and using numerical values yields:
ψ(s) 20
= 2 (1.54)
δe (s) s + 20s
Then, using the actuator dynamics given before, a closed-loop control system with a
proportional derivative controller can be proposed, where Kp = 68 and Kd = 17.
The characteristic equation is 0.1s3 + 3s2 + 224s + 20 and its roots are located at
−13 ± 44.4i and −3.8, therefore the system is stable.

1.3.1.4. Simulation Results

We have developed several simulations of the model to determine its qualities
of flight. The following graphs shows stable dynamics using a PD controller. The
 Modeling and Control of Mini UAV 15

b/2

n δe

δe

l
δe
δe Vertical Flight Horizontal Flight

Figure 1.7:
Yaw Con-
trol.

Table 1.1: Bidule-CSyRex Aircraft Parameters

Parameter Value Definition

hal-00923127, version 1 - 2 Jan 2014

S 0.18m2 Wing Reference area

c 0.3m Wing chord
b 0.6m Wing span
ρ 1.225kg/m3 Air density
V 10m/s Wind velocity (airflow)
Jx 0.0144kg.m x-axis moment of inertia
Jy 0.0254kg.m y-axis moment of inertia
Jz 0.0312kg.m z-axis moment of inertia
d 0.2m Rotor distance from the center of mass
D 0.27 Propeller diameter
n 9000RPM Rotor speed
Sb 0.006m2 Blade reference area
cb 0.3m Blade chord
Jyb 2 × 10−6 y-axis blade inertia
fs 17N Force produced by the servo

response of the roll subsystem to a unit step is shown in Figure ??. It is clear that
VPP control can stabilize the system very fast while speed control can not stabilize
this system at all.

For each control loop the step response is evaluated. First, the roll control system is
validated in simulation, this system based in the VPP mechanism has been compared
with a roll control system based on the speed variation of the rotors. Normally, a
control based on the speed variation introduces a time-delay, which is caused by the
electronic of the speed controller. This time-delay, provokes instability in the system
making the tuning of the controller parameters a very difficult task. Figure ?? shows
the comparison of the two systems. The two systems reach the desired value almost
at the same time, but in the system using speed variation there are oscillations in the
 16 UAV Flight Formation Control

steady state, while the VPP control quickly stabilizes the system. In the same way, the
step response for the pitch and yaw closed-loop control systems have been simulated.
Roll Control Response
6

Roll Angle
3

−1
0 5 10 15 20 25 30 35 40 45 50
Time

Figure 1.8: Response of the

Roll Control Loop
hal-00923127, version 1 - 2 Jan 2014

1.3.1.5. Experimental Results

In this section, qualitative results in flight test of the tail-sitter are discussed. This
vehicle in vertical flight presents a natural unstable behavior, and the manual guidance
and control is a very difficult task even if the remote human operator has an excellent
piloting skill. Figure ?? shows the vehicle crashing due to the high instability in a test
without any automatic control algorithm.

Figure 1.9: Bidule-CSyRex

with no control

As it was seen in previous sections, the control law for this vehicle is a simple PD
control, which has been chosen due that the position variables and its derivatives are
obtained directly from the IMU. The integral is avoided due to the high probability for
error in the steady state because of the signal noise in the sensors. To adjust the control
parameters several flight tests were carried out until obtaining a good performance
of the vehicle. First the Kd gains were adjusted to get a good stiffness in all the
angular displacements, then the Kp gain was adjusted to obtain a good time response
 Modeling and Control of Mini UAV 17

to changes of angular position. The stability derivatives, C`φ̇ , Cmq , Cnδe , Cnψ̇ and
Cmϕ̇ would normally be estimated using the data obtained from wind tunnel tests.
However, in the current study first the controller parameters were first obtained in
flight test, then using the values of the derivatives and the aerodynamic coefficients
estimated using Javafoil and Javaprop, the unknown derivatives were obtained. Figure
?? shows the vehicle flying stable when the linear control PD is used. Note that tethers
were used for safety purposes only, with satisfactory flight test results used only when
all the tether are slack, thus not supporting the flight platform in any way.
hal-00923127, version 1 - 2 Jan 2014

Figure 1.10: Bidule-

CSyRex with PD control

1.3.2. Robust Control Considering Parametric Uncertainty

It is usual to handle inaccurate mathematical models. In the previous section,

aerodynamic coefficients for a tail-sitter had been estimated using shareware software.
Then, mathematical model obtained for the tail-sitter vehicle may not be accurate due
to estimation errors and due to imperfections in the vehicle building process. Thus, in
this section a robust control design and analysis for a tail-sitter is developed.

Let us recall the roll subsystem

φ̈ = (Ctϕ ϕρn2 D4 · d − C`φ̇ φ̇)/Jx (1.55)

  2Jyb
Cmϕ ϕ + Cmϕ̇ ϕ̇ = ks fs δs (1.56)
ρVt2b Sb cb

Then, the roll subsystem can be expressed as follows

η˙r = Aηr + Bδc (1.57)

y = Cηr (1.58)
 18 UAV Flight Formation Control

where    
0 1 0 0 0
 0 0 1 0  ,B =  0
 
A=
 
0 0 0 1   0 
0 −30000 −4450 −155 3600
and    
φ 1
 φ̇  T
 0 
η= ,C =  
 −25φ + 5ϕ   0 
625φ̇ − 725ϕ + 600δs 0

These values were obtained using the parameters values in Table 1.1. It is impor-
tant to note that some of the aircraft parameters were obtained by experimental tests
and the remaining aircraft parameters were estimated using shareware software. In
order to take into account possible measurement errors, we will consider uncertainty
hal-00923127, version 1 - 2 Jan 2014

in the last row coefficients of the state matrix. In spite of the uncertainty structure in
the coefficients of the last row of the matrix, it is always possible to lump the uncer-
tainty such that the resulting polynomial family is a lumped version of the original
interval polynomial family. As a result of this considerations, the following matrix is
obtained
 
0 1 0 0
 0 0 1 0 
A(q) =   0 0 0 1 
 (1.59)
q0 q1 q2 q3

where q0 ∈ [−10, 10], q1 ∈ [−36000, −24000], q2 ∈ [−5340, −3560] and q3 ∈

[−186, −124].
1.3.2.1. Pitch Subsystem
The pitch angle dynamics (1.48) can also be written as

η̇p = Aξp + Bδe (1.60)

y = Cξp (1.61)

Using the vehicle parameters given in Table 1.1 and the same procedure for the roll
subsystem leads to
 
0 1
A(q) = (1.62)
−q0 −q1

with q0 = [−21.6, −14.4], q1 = [−48, −32].

 Modeling and Control of Mini UAV 19

1.3.2.2. Yaw Subsystem

The yaw angle dynamics (1.53) can also be written as

η̇y = Aηy + Bδe (1.63)

y = Cηy (1.64)

Using the vehicle parameters given in table 1.1 and the same procedure used in roll
subsystem we obtain

 
0 1
A(q) = (1.65)
−q0 −q1
hal-00923127, version 1 - 2 Jan 2014

with q0 = 0, q1 = [−24, −16].

Now on our goal is to present a robust state feedback control design to stabilize a
system with uncertain parameter values. Subsequently, the value set characterization
is used to verify the robust stability property when a time delay in the process is
considered. To do this, the Bidule CSyRex roll subsystem (1.57) and (1.59) will be
represented in the form:


η̇ = A(q − )η + Bu + BΓ (r)η
Σun , (1.66)
y = Cη

where

 
0 1 0 0
 0 0 1 0 
A(q− ) = 
 0

0 0 1 
−10 −36000 −5340 −186
 
Γ (r) = r3 r2 r1 r0

where r0 ∈ [0, 0.0172]; r1 ∈ [0, 0.4944]; r2 ∈ [0, 3.3333]; r3 ∈ [0, 0.0056]. Now,
the F matrix is defined in such a way that the following condition is satisfied:

Γ (r)T Γ (r) ≤ F ∀r ∈ R (1.67)

 20 UAV Flight Formation Control

and the nominal system will be considered as follows:


η̇ = A(q − )η + Bu
Σnom , (1.68)
y = Cη

With the above definitions, it is possible to present the following result:

P ROPOSITION 1.1.– Consider the Bidule CSyRex roll subsystem (1.66), and the fol-
lowing control law
u = −B T Sη (1.69)
where
SA(q − ) + A(q − )T S + F + I − SBB T S = 0, S > 0 (1.70)
for system (1.66), the control law is:
hal-00923127, version 1 - 2 Jan 2014

u = −10η1 − 1.5428η2 − 0.1173η3 − 0.0034η4 (1.71)

then, the closed-loop system is robustly stable.

Proof. considering the following Lyapunov candidate function:

Z ∞
V (η) = minu∈R (η T F η + η T η + uT u)dt (1.72)
0

It is possible to verify that the proposed control law (1.71) corresponds to the solution
of the LQR optimal control problem for the Σnom system (1.68), considering the
cost functional V (η), and the relative weights matrices H = F + I and R = 1.
Obviously, the above control law stabilizes the nominal system Σnom . Next, a proof
that the same control law also stabilizes the uncertain system Σun will be presented.
Using the results of the LQR optimal control problem, it is possible to obtain the
following solution to the problem in (1.72):
Z ∞
∗
V (η) = (η T F η + η T η + η T SBB T Sη)dt
0

by definition, V ∗ (η) must satisfy the Hamilton-Jacobi-Bellman equations, see

[AND 90], [LEW 95] and [LIN 07]:

η T F η + η T η + η T SBB T Sη
h iT (1.73)
+ ∂V∂η(η) A(q − )η + BB T Sη = 0

 T
∂V (η)
2η T SB + B=0 (1.74)
∂η
 Modeling and Control of Mini UAV 21

Along the trajectories of the system Σun (1.66) we have

 T
∂V (η)
V̇ (η) = η̇
∂η
 T
∂V (η)  
= A(q − )η + Bu + BΓ (r)η
∂η
 T
∂V (η) 
A(q − )η + BB T Sη

=
∂η
 T
∂V (η)
+ BΓ (r)η
∂η
hal-00923127, version 1 - 2 Jan 2014

Then (1.73) and (1.74), leads to

V̇ (η) = −η T F η − η T η − η T SBB T Sη
−2η T SBΓ (r)η
= −η T F η − η T η − η T SBB T Sη
−2η T SBΓ (r)η ± η T Γ T (r)Γ (r)η
= −η T F − Γ T (r)Γ (r) η − η T η
 

T  T
−η T B T S + Γ (r)
 
B S + Γ (r) η

from condition (1.67) it follows

V̇ (η) ≤ −η T η (1.75)

Then, V̇ (η) < 0 for all η 6= 0 and V̇ (η) = 0 if and only if η = 0, which ends the
proof.

All processes have time delays due to sensor information process, actuator time
delay, etc. Considering a time delay τ , system (1.66) can be rewritten as
 
0 1 ... 0 
0

 0 0 ... 0   0 
η̇(t) =  .  0  u(t − τ )
η(t) +  (1.76)
 
. . 
 .. .. .. 
1 
−q0 −q1 . . . −qn−1 r0
 22 UAV Flight Formation Control

P ROPOSITION 1.2.– Consider the Bidule CSyRex roll subsystem with time delay
(1.76), then this system is robustly stable if the same control law (1.71) is used and the
maximum time-delay is τmax = 1sec.

Proof. The uncertain time-delay system (1.76) has the following characteristic equa-
tion:

p(s, q, r, e−τ s ) = s4 + [124, 186]s3

+[3560, 5340]s2 + [24000, 36000]s + [−10, 10]
(1.77)
+(12s3 + 422s2
+5553s + 35990)e−[0,1]s
These kind of functions are known as quasipolynomials. It is clear that the above
characteristic equation (1.77) represents an infinite number of quasipolynomials that
hal-00923127, version 1 - 2 Jan 2014

have to be considered to verify the robust stability property. This family is defined as
follows:
p(s, q, r, e−τ s ) : q ∈ Q;
 
Pτ , (1.78)
r ∈ R; τ ∈ [0, τmax ]
where Q and R represent the set of uncertainty, see [BAR 94].

It is clear that the value set of Pτ is a set of complex numbers plotted on the
complex plane for values of q, r, ω and τ inside the defined boundaries. Next, the
zero exclusion principle is presented in order to verify the robust condition [BAR 94].

L EMMA 1.1.– Consider the characteristic equation (1.77), also called quasipolyno-
mials. Suppose that (1.77) has at least one stable member. Then the robust stability
property of the control system is guaranteed if and only if:

0∈
/ Vτ (ω) ∀ω ≥ 0 (1.79)

The results presented in [ROM 95] and [ROM 97] permit building the value set
Vτ (ω) for the characteristic equation (1.77) and is presented in Figures ?? and ??.

It can be noted that the zero is not included in the value set Vτ (ω). Then the system
(1.76) is then robustly stable.

1.3.2.3. Simulation
To investigate the behavior of the control stabilization system, several simulations
of the model have been run using Matlab SimulinkT M . This helps determine the flight
handling qualities of the vehicle. Roll control system is simulated using control law
 Modeling and Control of Mini UAV 23

4 value−set
x 10
16

14

12

10

Im 6

−2
−14 −12 −10 −8 −6 −4 −2 0 2 4
Re 4
x 10

Figure 1.11: Value set of Roll

subsystem control
hal-00923127, version 1 - 2 Jan 2014

4
x 10
12

10

0
−6 −5 −4 −3 −2 −1 0 1 2 3
4
x 10

Figure 1.12:
Value set
zoom

(1.71). Its behavior is shown in Figure ??. In the same way, the pitch and yaw closed-
loop control systems have been simulated. Their respective responses are shown in
Figures ?? and ??.

1.3.3. Nonlinear Control based on Nested Saturations

Linear and robust control approaches to attitude stabilization of a mini tail-sitter

have been introduced in sections 1.3.1 and 1.3.2. Now, a nonlinear control strat-
egy based on nested saturations to stabilize attitude and position of a mini tail-sitter
 24 UAV Flight Formation Control

Roll Angle Subsystem

1.5
x
1
x
1 2
x
3
x
4
0.5

−0.5

−1

−1.5

−2

−2.5

−3
0 1 2 3 4 5 6 7 8 9 10
Time (s)
hal-00923127, version 1 - 2 Jan 2014

Figure 1.13: Response of the

Roll control loop

Pitch subsystem

1 x1
x2
0.8

0.6

0.4

0.2

−0.2

−0.4

−0.6

−0.8

−1
0 1 2 3 4 5 6 7 8 9 10
Time (sec)

Figure 1.14: Response of the

Pitch control loop

shown in Figure ?? in hover flight is introduced. Let us recall the attitude dynamic
b
equations for mini UAVs ?? and ?? where MA,T represents the aerodynamic and
thrust moments. Then, applying a control strategy similar to the one presented for the
quadrotor vehicle, the following control input is proposed
h i
b b
J b ωb/e
b
+ H (Φ) J b τ̃ − Ḣ (Φ) ωb/e b
−1
MA/T , Ωb/e (1.80)
 Modeling and Control of Mini UAV 25

Yaw Subsystem
0.6

x
1
0.5 x
2

0.4

0.3

0.2

0.1

−0.1

−0.2

−0.3
0 1 2 3 4 5 6 7 8 9 10
Time (sec)
hal-00923127, version 1 - 2 Jan 2014

Figure 1.15: Response of the

Yaw control loop

 T
where τ̃ = τ˜φ τ˜θ τ˜ψ . Then (??) can be rewritten as

φ̈ = τ̃φ (1.81)
θ̈ = τ̃θ (1.82)
ψ̈ = τ̃ψ (1.83)

1.3.3.1. Equations of Translational Motion

The translational motion equations can be obtained using equations (??) and (??).
Then the kinematic equations of translational motion are written as

U̇ = fx /m − gsθ + RV − QW (1.84)
V̇ = fy /m + gsφcθ + RU − P W (1.85)
Ẇ = fz /m + gcφcθ + QU − P V (1.86)

Since the vehicle is a basic airfoil profile NACA0018, the aerodynamic force has
two components, lift over x-axis and drag over z-axis, the vehicle forces are given by
 26 UAV Flight Formation Control

fx = Le1 + Le2 , fy = 0 and fz = F1 + F2 − g − De1 + De2 where Le1 and Le2 are
the lift forces in each elevon, De1 and De2 are the drag forces due to the elevons and
F1 and F2 are the forces due to thrust of the motors. These forces are given as

1 2
Lei = ρV SCL (1.87)
2
1 2
Dei = ρV SCD (1.88)
2

where CL and CD are the lift and drag coefficients which can be calculated using
shareware programs as Javafoil.
hal-00923127, version 1 - 2 Jan 2014

Assuming that the translational and angular motion are decoupled, i.e. the euler
angles and angular rates are zero, then the translational motion equations can be rewrit-
ten as

     fx 
ẍ cθcψ sφsψ + cφsθcψ 0 m
 ÿ  =  cθsψ −sφcψ + cφsθsψ 0   fmz  (1.89)
z̈ −sθ cφcθ 1 −g

In this case a four integrators in cascade to stabilize roll is presented. The same
recursive algorithm is used to stabilize pitch.

Now, it is possible to introduce the following theorem :

T HEOREM 1.1.– Consider the vehicle dynamics (1.89), (1.81)-(1.83) with control
input fx , (θ − g sin θ)/ cos θ and fz , (r + g)/ cos θ with r , −k1 ż − k2 (z − zd ),

τ̃θ = −σn (χn (x) + σn−1 (χn−1 (x) + . . . + σ1 (χ1 (x)))) (1.90)

τ̃φ = −σn (χ̄n (x) + σn−1 (χ̄n−1 (x) + . . . + σ1 (χ̄1 (x)))) (1.91)
with for n = 1, ..., 4 for both θ subsystem and φ subsystem and τ̃ψ = −k3 ψ̇ −
k4 (ψ − ψhd ) for ψ subsystem. The functions
i σi are differentiable linear saturations.
Let b ≡ x, ẋ, y, ẏ, z, ż, φ, φ̇, θ, θ̇, ψ, ψ̇ . Then for any b(0) ∈ R12 , limt→∞ b(t) =
0.

Proof. The aim of the proposed nonlinear control is to stabilize the vehicle in hover
flight. Then, we develop a control for longitudinal, lateral and axial dynamics. Thus,
 Modeling and Control of Mini UAV 27

we start the proof by considering the translational dynamic equations (1.89), then the
translational dynamics is reduced to

ÿ = −fz sin φ (1.92)

z̈ = fz cos φ − g (1.93)

To stabilize the altitude, it is proposed that fz = (r + g) / cos φ where r = −k1 ż −

k2 (z − zd ) with k1 , k2 > 0 and zd is the desired altitude. Since r → 0 it can be
concluded that z → zd .

Then, equation (1.92) becomes ÿ = −g tan φ. Assuming that the vehicle evolves
in a neighborhood kφk < π/10 Therefore, the lateral dynamics reduces to four inte-
grators in cascade as follows
hal-00923127, version 1 - 2 Jan 2014

ÿ = −gφ (1.94)
φ̈ = τϕ (1.95)

Assuming that the roll angle and yaw angle are stabilized, the translational
dynamic equations (1.89) are reduced to

ẍ = cθfx /m + sθfz /m (1.96)

z̈ = −sθfx /m + cθfz /m − g (1.97)

Assuming that for a time long enough to make fz ≈ g and using the control input
θ − g sin θ
fx , (1.98)
cos θ

Then, the longitudinal model is reduced to a four integrators in cascade

ẍ = θ (1.99)
θ̈ = τθ (1.100)

Considering that the pitch and roll angles are stabilized, a simple PD controller
can be proposed to stabilize the yaw attitude. Therefore, τψ = −k3 ψ̇ − k4 ψ with
k3 , k4 > 0. This control is such that ψ → 0.
 28 UAV Flight Formation Control

Now, introducing the following variables

χ1 = χ2 + ẏ + 2φ + φ̇ (1.101)
χ2 = χ3 + φ + φ̇ (1.102)
χ3 = φ̇ + φ (1.103)
χ4 = φ̇ (1.104)

To simplify the analysis, a recursive methodology is proposed. To do this, it is

assumed that

ζn = χn (x) + σn−1 (ζn−1 (x))) (1.105)

hal-00923127, version 1 - 2 Jan 2014

ζ1 = χ1 (x) (1.106)

and
u = −σn (ζn )) (1.107)
Let us define the following positive definite function

Vn = (1/2)χ2n (1.108)

Differentiating V with respect to time, we obtain

V˙n = χn χ̇n (1.109)

from the fact that χ̇n = −σn (ζn ), we have

V˙n = χn u = −χn σn (ζn )) (1.110)

due to equation (1.105) we get

V˙n = −χn σn (χn (x) + σn−1 (ζn−1 (x)))) (1.111)

Using definition (??) and the condition from theorem (1.1), that Mn−1 < 0.5Ln , it
can be noted that if |χn | > 0.5Ln then V˙n < 0. This means that there exist a time
Tn such that |χn | ≤ 0.5Ln for ∀t > Tn which implies that |χn + σn−1 (ζn−1 (x))| ≤
0.5Ln + Mn−1 ≤ Ln .

When n = 1 we have the base case of the recursion. This case is treated a little
different, let us propose
V1 = (1/2)χ21 (1.112)
 Modeling and Control of Mini UAV 29

Differentiating V with respect to time, we obtain

V˙1 = χ1 χ̇1 (1.113)

using (1.101)-(1.104) is possible to see that χ̇1 = −σ1 (χ1 ), then we have

V˙1 = −χ1 σ1 (ζ1 )) (1.114)

due to equation (1.106) we get

V˙1 = −χ1 σ1 (χ1 ) (1.115)

As in the recursive case, it can be noted that if |χ1 | > 0.5L1 then V˙1 < 0. This
means that there exist a time T1 such that |χ1 | ≤ 0.5L1 for ∀t > T1 . It is important to
note that Tn < Tn−1 for all n > 2.
hal-00923127, version 1 - 2 Jan 2014

Since V˙1 < 0 then, from equations (1.106) and (1.112) implies that χ1 = ζ1 → 0.
It can be noted that starting from i = 2 until i = n we have the following set of
equations due to the recursion of the method

V˙2 = −χ2 σ2 (χ2 (x) + σ1 (ζ1 (x))) (1.116)

V˙3 = −χ3 σ3 (χ3 (x) + σ2 (ζ2 (x))) (1.117)
V˙4 = −χ4 σ4 (χ4 (x) + σ3 (ζ3 (x))) (1.118)

The recursion of equation (1.105) leads us to:

ζ2 = χ2 (x) + σ1 (ζ1 (x))) (1.119)

ζ3 = χ3 (x) + σ2 (ζ2 (x))) (1.120)
ζ4 = χ4 (x) + σ3 (ζ3 (x))) (1.121)

From (1.116), χ2 → 0, (1.119) implies that ζ2 → 0, in a recursive form (1.117),

χ3 → 0, from (1.120) ζ3 → 0, from (1.118), χ4 → 0, from (1.121), ζ4 → 0. This
means that, from (1.104) φ̇ → 0, from (1.103) φ → 0, from (1.102) ẏ → 0, and finally
from (1.101) y → 0.

By using a set of variables similar to χ1 − χ4 replacing φ by θ, y-position by x-

position and the same procedure described above for the x−θ subsystem, it is possible
to show the stability of this subsystem.
 30 UAV Flight Formation Control

1.3.3.2. Simulation Results

To investigate the behavior of the control stabilization systems, several simulations
of the model have been run using Matlab SimulinkT M . Their respective responses are
shown in Figures ?? and ??.
Mini rotorcraft position
3

2.5

Position (m)
1.5

0.5

0 Position X
Position Y
Altitude
−0.5
0 5 10 15 20 25 30 35 40 45 50
Time (sec)

Figure 1.16: Response of the

hal-00923127, version 1 - 2 Jan 2014

Roll Control Loop

Mini rotorcraft Attitude

25
Pitch angle
20 Roll angle
Yaw angle
15

10

5
Angle(deg)

−5

−10

−15

−20

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

Figure 1.17: Response of the

Pitch Control Loop

1.4. Quad-Tilting Rotor Convertible MAV

By marrying the take-off and landing capabilities of the helicopter with the forward
flight efficiencies of fixed-wing aircraft, the Quad-plane promises a unique blend of
capabilities at lower cost than other UAV configurations. While the tilt-rotor concept
is very promising, it also comes with significant challenges. Indeed it is necessary
to design controllers that will work over the complete flight envelope of the vehicle:
from low-speed vertical flight through high-speed forward flight. The main change in
this respect (besides understanding the detailed aerodynamics) is the large variation in
the vehicle dynamics between these two different flight regimes. Several experimen-
tal platforms have been realized with a body structure in which the transition flight is
executed by turning the complete body of the aircraft [GRE 06], [ESC 06], [STO 04],
 Modeling and Control of Mini UAV 31
hal-00923127, version 1 - 2 Jan 2014

Figure 1.18: Quad-Tilting Rotor

Convertible UAV.

[STO 02b], [ESC 07]. In [STO 04] and [STO 02b] the authors described the develop-
ment (modeling, control architecture and experimental prototype) of Two-rotor tail-
sitter. The control architecture features a complex switching logic of classical linear
controllers to deal with the vertical, transition and forward flight. [GRE 06] presents
a classical airplane configuration MAV to perform both operational modes. The hover
flight is autonomously controlled by an onboard control flight system while the tran-
sition and cruise flight is manually controlled. A standard PD controller is employed
during hover flight to command the rudder and elevator. In [ESC 06] some preliminary
results are presented for the vertical flight of a Two-rotor MAV as well as a low-cost
embedded flight control system. There are some examples to other tilt-rotor vehi-
cles with quad-rotor configurations like Boeing’s V44 [SNY 00] and the QTW UAV
[NON 07]. In [ONE 08] the authors present the progress of their ongoing project,
an aircraft with four tilting wings. A new tilt-rotor aircraft (Quad-plane Unmanned
Aerial Vehicle) that is capable of flying in horizontal and vertical modes is presented
in this section. The vehicle is driven by four rotors and has a conventional airplane-
like structure, which constitutes a highly nonlinear plant and thus the control design
should take into account this aspect. A nonlinear control strategy, consisting of a
feedback-linearizable input for altitude control and a hierarchical control (inner-outer
loop) scheme for the underactuated dynamic subsystem (x-position, pitch), is pro-
posed to stabilize the aerial robot within the hovering mode. Backstepping [KHA 02],
a Lyapunov based method is presented to stabilize the vehicle within the airplane
mode. Through the use recursive method, backstepping divides the control problem
 32 UAV Flight Formation Control
hal-00923127, version 1 - 2 Jan 2014

Figure 1.19: Opera-

tional transition.

into a sequence of designs for simpler systems. This mini aerial vehicle is one of the
first of its kind among tilt-wing vehicles on that scale range.

1.4.1. Modeling

This section presents the longitudinal equations of motion as well as the aero-
dynamics of the vehicle. Due to the flight profile of the vehicle we distinguish three
operation modes: (1) Hover Flight (HV) the aircraft behaves as a rotary-wing platform
(|γ| ≤ π6 ), (2) Slow-Forward Flight (SFF) ( π6 < |γ| ≤ π3 ) and finally (3) Fast-Forward
Flight (FFF), where the aerial robot behaves as a pure airborne vehicle ( π3 < |γ| < π2 ).
1) During the HF the 3D vehicle’s motion relies only on the rotors. Within this
phase the vehicle features VTOL flight profile. The controller for this regime disregard
the aerodynamic terms due to the negligible translational speed.
2) It is possible to distinguish an intermediate operation mode, the SFF, which
links the two flight conditions, HF and FFF. This is probably the most complex
dynamics.
3) FFF regime mode (Aft position), at this flight mode the aircraft has gained
enough speed to generate aerodynamic forces to lift and control the vehicle motion.

Kinematics
– F i denotes the inertial earth-fixed frame with origin, Oi , at the earth surface.
This frame is associated to the vector basis {ii , ji , ki }.
 Modeling and Control of Mini UAV 33
hal-00923127, version 1 - 2 Jan 2014

Figure 1.20: Coordinate systems: Inertial frame (F i ) and

Body-fixed frame (F b )

Figure 1.21: Free-body scheme showing the forces acting on

the Quad-tilting MAV.
 34 UAV Flight Formation Control

– F b denotes the body-fixed frame, with origin, Ob , at the center of gravity CG.
This frame is associated to the vector basis {ib , jb , kb }.
– F a denotes the aerodynamic frame, with origin, Ob , at the center of gravity CG.
This frame is associated to the vector basis {ia , ja , ka }.
– The orthonormal transformation matrices Rbi and Rab , respectively used to
transform a vector from F b → F i and F a → F b within the longitudinal plane (pitch
axis), are given by:

   
cos θ 0 sin θ cos α 0 sin α
Rbi = 0 1 0  , Rab =  0 1 0 
− sin θ 0 cos θ − sin α 0 cos α

Aerodynamics
hal-00923127, version 1 - 2 Jan 2014

It is important to consider these forces properly because they are fundamentally

affected by the vehicle’s motion and thus they alter the basic dynamics involved. The
analysis used in the present paper will be based on a combination of a low-order panel
method aerodynamic model coupled with a simple actuator disc model of the flow
induced by the propellers. In order to proceed with the aerodynamic analysis, it is
worth to mention the following assumptions:
– A1. The vehicle is a rigid body, i.e. the felexibility of the aircraft wings or
fuselage will be neglected.
– A2. Non-varying mass is considered (ṁ(t) = 0).
– A3. The aerodynamic center (AC) and the center of gravity (CG) are coincident.

In order to determine the aerodynamic forces exerted on the vehicle, we need to

know both the direction and velocity of the total airflow vector. We can identify three
wind vectors acting on the vehicle: the airflow speed Vp produced by the rotors,
the Vb airflow generated by the translational motion of the body (U, W ) and a third
component due to the external wind (disturbance) Ve , generally unknown. Hence, the
total wind vector in the body frame can be written as
Vtot = Vp (γ) + Vb + Ve (1.122)
where Vtot = (vu , vw )T . The total wind vector Vtot experienced by the wing varies
depending on the flight mode. Within the HF and SFF regimes the wing is not washed
by the propeller airflow Vp (Fig. ??), while in the FFF mode, it is assumed that the
wing is significatively submerged (Fig. ??) by Vp . Therefore the propeller slipstream
Vp is disregarded in HF and SFF. To include the behavior of Vp in the equations let
us introduce the following function
0 if γ ≤ π3

ξ(γ) = (1.123)
1 if γ > π3
 Modeling and Control of Mini UAV 35

Figure 1.22: Airflow profile generated by the rotors during the flight envelope. Relative wind velocity
in HF and SFF modes.
hal-00923127, version 1 - 2 Jan 2014

Figure 1.23: Relative wind

velocity in FFF mode.

The parallel wind velocity vu and the normal wind velocity vw components at
the wing encompass the velocity that the vehicle experiments through the air and the
corresponding components of Vp due the tilting of the rotors and the aleatory external
wind Ve , i.e.
Vu = (u + ξ(γ)vp sin(γ) + veu )ib (1.124)
Vw = (w + ξ(γ)vp cos(γ) + vew )kb (1.125)
Assuming purely axial flow into the propellers, simple actuator disc theory [STE 04]
gives the induced propeller velocity for the ith rotor as
s
2Ti
vpi = (1.126)
ρAp
 36 UAV Flight Formation Control

where Ap is the total disc-area of the propeller and ρ the air density. Figure ?? shows
the aerodynamic forces on a small UAV with a tilt angle γ. The forces consist of a lift
force, L, perpendicular to the total flow vector, Vtot , a drag force D parallel to Vtot ,
and the airfoil’s pitching moment, M , about the positive cartesian y-axis. The above
discussion can be summarized by:
Cl = Clα α
Cd = Cdp + Cdi
Cm = Cmα α
where these equations are standard aerodynamic non-dimensional lift, drag and
moment coefficients 1 To obtain the lift and drag forces and the pitching moment
on the aircraft it is only necessary to obtain the total wind velocity vector Vtot , see
(1.122), the angle of attack and the aerodynamic parameters Clα , Clδ , Cd , Cmδ , Cmα
which depend on the geometry of the vehicle.
hal-00923127, version 1 - 2 Jan 2014

1 2
L = 2 Cl ρVtot S
1 2
D = 2 Cd ρVtot S
1 2
M = 2 Cm ρVtot Sc̄

In these equations S and c̄ are the area and the wing chord respectively. The angle of
attack α and the magnitude of Vtot are obtained through the following equations
α = arctan(vw /vu ) (1.127)
p
2 + v2
|Vtot | = vw (1.128)
u
The lift force will depend on the velocity Vtot and the angle of attack. The figure ??
represents the different values of lift for several speed conditions:

Forces exerted on the Quad-plane

The vector that contains the set of forces applied to the Quad-plane (Fig. ?? ) is
given by
mξ̈ = Rbi T b + Rbi Rab Aa + W i (1.129)
where, ξ = (x, y, z)T is the CG’s position vector in F i , T b = (0, 0, −T )T is the
collective thrust in F b , Aa = (−D, 0, −L)T is the vector of aerodynamic forces in
F a and finally W i = (0, 0, mg)T denotes the weight of the vehicle in F i . The four
propellers produce the collective thrust T , which can be modeled as
i=4
X
T = Kl ωi2 (1.130)
i=1

1. C∗ slopes are obtained from the software XFOIL.

 Modeling and Control of Mini UAV 37

100

50
Lift [N]

−50
0
5
hal-00923127, version 1 - 2 Jan 2014

−100 10
100
50 15
0
−50 20
−100 speed [m/s]
α [deg]

Figure 1.24: Lift values for

different velocities.

where ωi is the angular velocity of ith -rotor, Kl is a lift factor depending on the
aerodynamic parameters of the propeller. Note that the vector of aerodynamic forces
Aa is not only involved in translational motion, but also in the rotational motion of
the vehicle, as is shown next.

Moments acting on the Quad-plane

The forces shifted away from the center of gravity CG induce moments causing
the rotational motion. The corresponding vectorial equation grouping the moments
exerted about CG is written as

τ b = τTb + τM
b b
+ τG (1.131)

where τT is the induced moment due the difference of thrust between T3,4 and T1,2 ,
τM is the airfoil’s pitching moment, τG is the gyroscopic moment. τT is obtained
through
τTb = l1 (−T3,4 cos γ + T1,2 cos γ)jb (1.132)

where, l1 is the distance from the CG to the rotors. The airfoil’s pitching moment τM
 38 UAV Flight Formation Control

is obtained from the airfoil’s Cm slope and the lift contribution of the elevator.
b
τM = M jb (1.133)

The gyroscopic moment τG arises from the combination of the airframe’s angular
speed Ω b = (p, q, r)T and the rotors angular speed ωi . The τG vector is then modeled
as 4i=1 (Ω b × Ip ωi ), leading to
P

b
τG = Ip [q(ω2 − ω1 + ω3 − ω4 )ib + p(ω1 − ω2 − ω3 + ω4 )jb ] (1.134)

where, Ip represents the inertia moment of the propeller. For simplicity we do not
take into account the drag torque due to the propeller drag force. Since the present
paper concentrates on the longitudinal flight of the vehicle, the corresponding scalar
equations modeling the forces and moments applied to the vehicle are written as:
hal-00923127, version 1 - 2 Jan 2014

mẌ = −T3,4 sin (θ + γ) − T1,2 sin (θ + γ) − L sin (θ − α) − D cos (θ − α)

mZ̈ = D sin (θ − α) − T1,2 cos (θ + γ) − L cos (θ − α) − T3,4 cos (θ + γ) + g
Iyy θ̈ = M + l1 (−T3,4 cos γ + T1,2 cos γ) + Ip p(ω1 − ω2 − ω3 + ω4 )
(1.135)

FFF mathematical model

In this regime the vehicle essentially behaves as an airplane, thus we can consider
the common longitudinal aircraft model [STE 92]. In addition to the body-axis equa-
tions, it is important to express the equations of motion in the wind axis, because
the aerodynamic forces act in these axis and the magnitude of Vtot (written as V
from here onwards"), α can be expressed in terms of u and w. This reference sys-
tem is used for translational equations because angle of attack and velocity are either
directly measurable or closed related to directly measurable quantities, while the body
axis velocities (u, w) are not. The equations of motion take the form

1
V̇ = [−D + Tt cos α − mg (cos α sin θ − sin α cos θ)]
m
1
α̇ = [−L − Tt sin α + mg(cos α cos θ + sin α sin θ)] + q
Vm
θ̇ = q (1.136)

The angles θ and α lie in the same vertical plane above the north-east plane (Fig. ??),
and their difference is the flight-path angle Γ = θ−α (Fig. ??). Under this definition
 Modeling and Control of Mini UAV 39

Figure 1.25: Wind-Axis

Reference Frame.
hal-00923127, version 1 - 2 Jan 2014

Figure 1.26: Flight-path

angle definition.

and from the last equation of (1.135) we obtain the next mathematical model

1
V̇ = [−D + Tt cos α − mg sin Γ ]
m
1
α̇ = [−L − Tt sin α + mg cos Γ ] + q
Vm
θ̇ = q (1.137)
1
q̇ = M
Iyy

1.4.2. Transition

The flight envelope of the vehicle encompasses different flight conditions,

achieved by means of the collective angular displacement of the rotors. Indeed, is this
 40 UAV Flight Formation Control

18
Vu α=3.02
16 Vw

14

12
α=19.1
Vu,Vw [m/s]

10

8
α=35.8
6

4
hal-00923127, version 1 - 2 Jan 2014

0
0 20 40 60 80 100
t [seg]

Figure 1.27: Behavior velocities during the

tilting of the rotors.

tilting that provides a continuous mechanism to perform the operational transition. To

illustrates this, let us consider the following scenario:
– Tt ≥ mg i.e. the vehicle flies at a stabilized altitude.
– θ ≈ 0 i.e. stabilized vertical flight.

Is clear that as γ is tilted the horizontal velocity increases, while the vertical veloc-
ity is reduced (figures ?? and ??). These facts affect proportionally to the forces
coming from the rotors and the wing.

Thus, both vertical and horizontal controllers can still be used at the same time
whose actions is controlled by γ. The vertical collective thrust is gradually reduced
inhibiting the action of vertical controller and allowing the action of the horizontal
controller and viceversa. So, for example, for larger values of γ, i.e. γ > 45, the
rotorcraft behaves more like a classical airplane. As the vehicle is gaining speed due
to rotors tilt (γ), then aerodynamic forces arise. For this reason we consider that the
control of vertical and forward flight are active during the whole flight envelope.
 Modeling and Control of Mini UAV 41

100
Attack Angle (alpha)
90 Tilting Angle (gamma)

80

70

60
α,γ [m/s]

50

40

30

20

10
hal-00923127, version 1 - 2 Jan 2014

0
0 20 40 60 80 100
t [seg]

Figure 1.28: Behavior of angles γ and α during the

tilting of the rotors.

1.4.3. Control strategy for hover flight mode

The vertical flight of the Quad-plane represents a challenging stage due to the
aircraft’s vertical dynamics are naturally unstable. In this regime, the Quad-plane
aerial robot aims to emulate the flight behavior of a Quadrotor which features and non
conventional Quadrotor design, i.e. an asymmetrical H-form structure.

Vertical flight regime encompasses two dynamic subsystems: the altitude dynam-
ics, actuated by the thrust T , and the horizontal translational motion, generated by the
pitch attitude. Taking into account the item 1) presented in section II, we can consider
a simplified model from which is derived the controller in HF regime (i.e. α ≈ 0 since
γ = 0).

For simplicity we consider that the gyroscopic moment is very small. These con-
siderations allows us to rewrite (1.135) as
 
T3,4 +T1,2
Ẍ = − (sin θ)
 m 
T3,4 +T1,2
Z̈ = − m (cos θ) + g (1.138)
 
l1
θ̈ = − Iyy (−T3,4 + T1,2 )
 42 UAV Flight Formation Control

If we rename the total thrust as Tt = T3,4 + T1,2 , and the difference of these thrusts
as Td = T1,2 − T3,4 . Then

Ẍ = − Tmt sin θ
Z̈ = − Tmt cos θ + g (1.139)
θ̈ = − Ilyy
1
Td

thus, we have derived a simple model, suitable for controller design. The altitude
(1.139b) can be stabilized via a feedback-linearizable input through the total thrust Tt
muz − mg
Tt = − (1.140)
cos(θ)

where uz = −kpz (z − z d ) − kdz ż with kpz , kdz > 0 and zd is the desired altitude.
Since the vehicle works in an area close to θ ≈ 0, the singularity is avoided. For
the subsystem 1.139a and 1.139c, a two-level hierarchical control scheme is used to
hal-00923127, version 1 - 2 Jan 2014

stabilize its dynamics. The outer-loop control stabilizes the translational motion (slow
dynamics [KOK 86]) along the x-axis, while the inner-loop control stabilizes the atti-
tude (fast dynamics). Introducing (1.140) into (1.139a) and assuming that z ≈ z d ,
namely uz → 0, leads to

ẍ ≈ −g tan θ = −g tan ux (1.141)

For the horizontal motion (1.141), θ can be considered as virtual control input ux .
However, it is a state not an actual control. Given that θ̇d is slowly time-varying, we
will assume that the x-dynamics converges slower than the θ-dynamics. The reference
for the inner-loop systems is

 
d −vx
ux = θ = arctan (1.142)
g
u̇x = θ̇d ≈ 0 (1.143)

where vx = kvx ẋ + kpx x with kvx , kpx > 0. Using the linearizing control input
(1.142) in (1.141) yields

ẍ = vx , provided that θ̃ = 0 (i.e. θ = θd )

As the previous equation shows, the success of the outer-loop controller relies directly
on the inner-loop attitude control performance, thus the inner loop controller must
guarantee the stabilization of the attitude around the reference. For this reason, the
stability analysis of the inner-loop controller is presented next. Consider the following
positive function which is an unbounded function
1
V (θ̃, θ̇) = Iyy θ̇2 + ln(cosh θ̃) (1.144)
2
 Modeling and Control of Mini UAV 43

Using (1.139) its corresponding time-derivative yields

l1
V̇ (θ̃, θ̇) = Iyy θ̇(− Td ) + θ̃˙ tanh θ̃ (1.145)
Iyy

˙
Considering θ̃ = θ̇, thus (1.145) may be rewritten as

V̇ (θ̃, θ̇) = θ̇(−l1 Td + tanh θ̃) (1.146)

Using the control input

tanh θ̃ + tanh θ̇
Td = (1.147)
l1
in (1.146) yields
V̇ (θ̃, θ̇) = −θ̇ tanh θ̇, (1.148)
hal-00923127, version 1 - 2 Jan 2014

where V̇ (θ̃, θ̇) ≤ 0. Therefore, the origin (θ̃, θ̇) is stable and the state vector remains
bounded. The asymptotic stability analysis can be obtained from LaSalle’s Theorem.
Therefore, θ̃ → 0 and θ̇ → 0 as t → ∞.

1.4.4. Control strategy for forward flight mode

In this section the flight path angle Γ will be controlled using the backstepping
algorithm taking the following approximations into consideration:
– The air speed is assumed constant, V̇ = 0 [MAT 07].
– From the definition of flight-path angle, the dynamics Γ̇ = θ̇ − α̇ yields Γ̇ =
1
mV [Tt sin α + L − mg cos Γ ].
– The thrust term Tt sin α in (1.137) will be neglected as it is generally much
smaller than lift.
– Cm = Cmδ (α)δ, since the main contribution to M is provided by the elevator.

With these considerations in mind and using the change of coordinates ζ = Γ −

1/2π, the system (1.137) may be expressed as

g cos (z + 1/2π) Clα α

ζ̇ = − +
V mV
g cos (z + 1/2π) Clα α
α̇ = − +q (1.149)
V mV
1
q̇ = Cmδ δ
Iyy
 44 UAV Flight Formation Control

Equation (1.150) is now in feed forward form for backstepping procedure. For nota-
tional simplification, let

ẋ = f (x) + ξ1
ξ˙1 = f1 (x, ξ1 ) + ξ2 (1.150)
ξ˙2 = f2 + g2 (ξ1 )u

with
 
C lα x
x = mV Γ
C lα ; f (x) = − Vg cos mV
 
g C lα x Clα ξ1
ξ1 = α; f1 (x, ξ1 ) = V cos mV − mV (1.151)
ξ2 = q; f2 = 0
hal-00923127, version 1 - 2 Jan 2014

1
u = δ; g2 (ξ1 ) = C
J y mδ

Defining the following error states as

e , x − xdes
e1 , ξ1 − ξ1,des (1.152)
e2 , ξ2 − ξ2,des

Now, following the backstepping procedure differentiating the first equation in (1.152)
yields
ė = f (x) + ξ1,des + e1 − ẋdes (1.153)
where ξ1,des is viewed as a virtual control for the last equation, choosing as ξ1,des =
−f (x) − k1 e + ẋdes . Then substituting this virtual control in (1.153) we have that

ė = −ke + e1 (1.154)

Repeating the same procedure, differentiating e1 yields

ė1 = f1 (x, ξ1 ) + e2 + ξ2,des − ξ˙1,des (1.155)

Let ξ2,des = −f1 (x, ξ1 ) − e − k1 e1 + ξ˙1,des so that

ė1 = −e − k1 e1 + e2 (1.156)

As a last step, now the real control signal is obtained in similar way. Differentiating
e2 yields
ė2 = f2 + g2 (ξ1 )u − ξ˙2,des (1.157)
 Modeling and Control of Mini UAV 45

Let
1 h i ...
u= −f2 − e1 − k2 e2 + ξ˙2,des = u( z d , z̈d , żd , zd , e, e1 , e2 ) (1.158)
g2 (ξ1 )
so that
ė2 = −e1 − k2 e2 (1.159)
It is important to ensure that g2 (ξ1 ) 6= 0, which occurs only with big enough

5
x position [m]

4
3
2 x
reference
1
0 500 1000
hal-00923127, version 1 - 2 Jan 2014

time[ms]
1
z
z position [m]

−2 reference

−5

−8
0 500 1000
time[ms]

0.4 θ
reference
θ [rad]

0.2
← disturbance
0

−0.2
0 500 1000
time[ms]

Figure 1.29: Position and attitude of the vehicle under dis-

turbance condition.

negative values of α. These values are assumed to be impossible to achieve in standard

operation of the airplane, so avoiding division by zero. Equations (1.154,1.156,1.159)
expressed in vectorial form
ė = −Ke + Se (1.160)
S = −S T satisfies eT Se = 0, ∀e, so that with the Lyapunov-candidate-function
V (e) = 12 eT e, and the time derivative evaluated in the trajectories yields

V̇ (e) = eT (−Ke + Se) = −eT Ke < 0, ∀e (1.161)

This proves that the above differential equation, is asymptotically stable about the
origin.
 46 UAV Flight Formation Control

x velocity [m/s]
x velocity
reference
0

−1
0 500 1000
time[ms]
z velocity [m/s]

−2
z velocity
−4 reference
0 500 1000
time[ms]
θ velocity [rad/s]

1
0
hal-00923127, version 1 - 2 Jan 2014

−1
θ velocity
−2
reference
−3
0 500 1000
time[ms]

Figure 1.30: Derivatives of the position and attitude of the vehicle under
disturbance condition.

1.4.5. Simulation results

HF

The performance of the nonlinear controller presented in the previous section is

evaluated on the dynamic model (1.139) in MATLAB/Simulink. We started the Quad-
plane at the position (x0 , z0 , θ0 ) = (2, 0, π8 ) and (ẋ, ż, θ̇) = (0, 0, 0). The aircraft
had the task of performing hover flight at (xd , z d , θd ) = (4, −7, 0). Figures ?? and
?? show the evolution and convergence of the states (x, ẋ, z, ż, θ, θ̇) to the desired
references with the initial conditions mentioned above. It is important to note that
position and angle references are tracked with negligible steady state errors. The
controller is robust in the presence of a perturbation in t = 600 ms with a magnitude
of 1/8π radians, as seen in figures ?? and ??.

The control inputs are depicted in the figure ??, which shows the reaction to the
disturbance.
 Modeling and Control of Mini UAV 47

5.5 T
1,2
T
3,4
4.5
Thrust

response to disturbance
3.5 ↓

2.5

1.5
0 500 1000
time[ms]

0.5
u
x
d(u )/dt
x
ux, d(ux)/dt

0
hal-00923127, version 1 - 2 Jan 2014

−0.5 ↑
response to disturbance

−1
0 500 1000
time[ms]

Figure 1.31: UAV’s control inputs and

response to disturbance.

FFF

After the vehicle experiments the transition-flight, its behavior is like an airplane.
We have considered the next initial conditions for purposes of simulation: Γ0 = 5,
α0 = 5 and θ0 = 10. The aircraft had the task of tracking a trajectory shown in the
first part of Fig. ??. This figure shows the evolution and tracking trajectory of the
states (Γ, α, θ) to the desired reference, with the initial conditions mentioned above.
It is important to note that angle references are tracked with negligible steady state
errors.

1.5. Concluding remarks

In this Chapter Euler-Lagrange approach has been applied for obtaining a simpli-
fied model of a quad-rotor rotorcraft.

The longitudinal dynamics of this aircraft including its aerodynamics are derived
at the hover and forward flight operating mode. The proposed control strategies were
evaluated, at simulation level, for the nonlinear dynamic model, obtaining satisfactory
 48 UAV Flight Formation Control

12

Γ [deg] 6

0 Γ
reference
−6
0 10 20 30 40
time[s]

10
5 α
α [deg]

0
−5
−10
−15
0 10 20 30 40
time[s]

8
4 θ
θ [deg]

−0
hal-00923127, version 1 - 2 Jan 2014

−4
−8
0 10 20 30 40
time[s]

Figure 1.32: Tra-

jectory tracking.

results. The proposed control algorithm is based on an inner-outer loop scheme since
it is suitable for implementation purposes.

For energy-saving purposes during forward flight (airplane mode), it is proposed

that the vehicles can lead their orientation towards wind velocity vector. To achieve
this objective, the vehicles could easily rotate their orientation (yaw movement in
helicopter mode) and once addressed the wind vector, switch to airplane mode. In
the Quad-plane configuration, this process could be quite simple, since the vehicle is
always maintained with the roll and pitch angles close to zero, which is not possible
in the case of a tail-sitter configuration since within vertical mode the wing surface is
highly vulnerable to wind gusts.

1.6. Bibliography

[AND 90] A NDERSON B., M OORE J., Optimal Control, Prentice Hall, 1990.
[BAR 94] BARMISH B., New Tools for Robustness of Linear Systems, Macmillan, 1994.
[BLY 06] B LYENBURG P., “UAVs systems, the global perspective”, UVS International, 2006.
[CAS 05] C ASTILLO P., L OZANO R., D ZUL A., Modelling and Control of Mini-Flying
Machines, Springer-Verlag, 2005.
 Modeling and Control of Mini UAV 49

50

25
u [deg]

−25
u

−50
0 10 20 30 40
time[s]

0.5
Tracking error
e(t) [deg]

0
hal-00923127, version 1 - 2 Jan 2014

−0.5
0 10 20 30 40
time[s]

Figure 1.33: Control and error defined

by e = Γ − Γd .

[ESC 06] E SCARENO J., S ALAZAR S., L OZANO R., “Modeling and Control of a Convertible
VTOL Aircraft”, 45th IEEE Conference on Decision and Control, 2006.
[ESC 07] E SCARENO J., S TONE H., S ANCHEZ A., L OZANO R., “Modeling and Control
Strategyfor the transition of a Convertible UAV”, European Control Conference (ECC07),
2007.
[ETK 96] E TKIN B., R EID L., Dynamics of Flight Stability and Control, John Wiley and Sons,
1996.
[GOL 83] G OLDSTEIN H., P OOLE C., S AFKO J., Classical Mechanics, Addison-Wesley,
1983.
[GRE 06] G REEN W., O H P., “Autonomous Hovering of a Fixed-Wing Micro Air Vehicle”,
International Conference on Robotics and Automation, 2006.
[HEP 06] H EPPERLE M., Aerodynamics for model aircraft, Technical report, 2006.
[KAN 01] K ANNAN N., S EETHARAMA M., “Longitudinal h1 stability augmentation system
for a thrust-vectored unmanned aircraft”, AIAA, Journal of Guidance, Control and Dynam-
ics, vol. 28, 2001.
[KHA 02] K HALIL H., Nonlinear Systems, Prentice Hall, 2002.
[KOK 86] KOKOTOVIC P., O’R EILLY P., K HALIL H., Singular Perturbation Methods in Con-
trol: Analysis and Design, Academic Press, 1986.
 50 UAV Flight Formation Control

[LEW 95] L EWIS F., S YRMOS V., Optimal Control, John Wiley and Sons, 1995.
[LIN 07] L IN F., Robust Control Design, An optimal control approach, John Wiley and Sons,
2007.
[MAT 07] M ATES D., H AGSTROM M., “Nonlinear Analysis and Synthesis Techniques for
Aircraft Control”, Lecture Notes in Control and Information Sciences, vol. 365, Springer,
2007.
[NON 07] N ONAMI K., “Prospect and Recent Research and Development for Civil Use
Autonomous Unmanned Aircraft as UAV and MAV”, Journal of System Design and
Dynamics, vol. 1, 2007.
[ONE 08] O NER K., C ETINSOY E., U NEL M., A KSIT M., K ANDEMIR I., G ULEZ K.,
“Dynamic Model and Control of a New Quadrotor Unmanned Aerial Vehicle with Tilt-
Wing Mechanism”, vol. 45, World Academy of Science, Engineering and Technology,
2008.
[ROM 95] ROMERO G., C OLLADO J., “Robust stability of interval plants with perturbed time
hal-00923127, version 1 - 2 Jan 2014

delay”, Proceedings of the IEEE American Control Conference, 1995.

[ROM 97] ROMERO G., Analysis of Robust Stability of Dynamics Systems with Time Delay,
PhD thesis, FIME - UANL, 1997.
[SNY 00] S NYDER D., “The Quad Tiltrotor: Its Beginning and Evolution”, Proceedings of
the 56th Annual Forum, American Helicopter Society, 2000.
[SPO 01] S POERRY, W ONG K., “Design and developement of a micro air vehicle
(mav)concept: Project bidule”, 9th Australian International Aerospace Congress (AIAC),
2001.
[STE 92] S TEVENS B., L EWIS F., Aircraft Control and Simulation, John Wiley and Sons Inc,
1992.
[STE 04] S TENGEL R., Flight Dynamics, Princeton University Press, 2004.
[STO 02a] S TONE H., “Aerodynamic modelling of a wing-in-slipstream tail-sitter UAV”,
AIAA Biennial International Powered Lift Conference and Exhibit, 2002.
[STO 02b] S TONE H., “Aerodynamic Modeling of a Wing-in-Slipstream Tail-Sitter UAV”,
Biennial International Powered Lift Conference and Exhibit, 2002.
[STO 04] S TONE H., “Control Architecture for a Tail-sitter Unmanned Air Vehicle”, 5th Asian
Control Conference, 2004.