Quadrotor Helicopter Flight Dynamics and Control:

Theory and Experiment

Gabriel M. Homann

Haomiao Huang

Steven L. Waslander

Claire J. Tomlin

Quadrotors are rapidly emerging as a popular platform for unmanned aerial vehicle
(UAV) research, due to the simplicity of their construction and maintenance, their ability
to hover, and their vertical take o and landing (VTOL) capability. Current designs have
often considered only nominal operating conditions for vehicle control design. This work
seeks to address issues that arise when deviating signicantly from the hover ight regime.
Aided by well established research for helicopter ight control, four separate aerodynamic
eects are investigated as they pertain to quadrotor ight. The eects result from either
translational or vertical vehicular velocity components, and cause both moments that af-
fect attitude control and thrust variation that aects altitude control. Where possible, a
theoretical development is rst presented, and is then validated through both thrust test
stand measurements and vehicle ight tests using the Stanford Testbed of Autonomous
Rotorcraft for Multi-Agent Control (STARMAC) quadrotor helicopter. The results have
enabled improved controller tracking throughout the ight envelope, including at higher
speeds and in gusting winds.
I. Introduction
Quadrotor helicopters are an emerging rotorcraft concept for unmanned aerial vehicle (UAV) platforms.
The vehicle consists of four rotors in total, with two pairs of counter-rotating, xed-pitch blades located at
the four corners of the aircraft, an example of which is shown in Figure 1. Due to its specic capabilities, use
of autonomous quadrotor vehicles has been envisaged for a variety of applications both as individual vehicles
and in multiple vehicle teams, including surveillance, search and rescue and mobile sensor networks.
1
The particular interest of the research community in the quadrotor design can be linked to two main
advantages over comparable vertical take o and landing (VTOL) UAVs, such as helicopters. First, quadro-
tors do not require complex mechanical control linkages for rotor actuation, relying instead on xed pitch
rotors and using variation in motor speed for vehicle control. This simplies both the design and mainte-
nance of the vehicle. Second, the use of four rotors ensures each individual rotor is smaller in diameter than
the equivalent helicopter rotor, relative to the size of the airframe. The individual rotors, therefore, store
signicantly less kinetic energy during ight, mitigating the risk posed by the rotors should they entrain any
objects. Furthermore, by enclosing the rotors within a frame, the rotors can be protected from breaking
during collisions, permitting ights indoors and in obstacle-dense environments, with low risk of damaging
the vehicle, its operators, or its surroundings. These added safety benets greatly accelerate the design and
test ight process by allowing testing to take place indoors, by inexperienced pilots, with a short turnaround
time for recovery from incidents.
Quadrotor vehicle dynamics are often assumed to be accurately modeled as linear, without any depen-
dence on velocity of attitude or altitude control. At slow velocities around hover, this is indeed a reasonable
assumption, but even at moderate velocities, the impact of the aerodynamic eects resulting from variation

This research was supported by ONR under the CoMotion MURI contract N00014-02-1-0720, and by NASA grant
NNAO5CS67G.

Ph.D. Candidate, Department of Aeronautics and Astronautics, Stanford University. AIAA Member. gabeh@stanford.edu

M.S. Student, Department of Aeronautics and Astronautics, Stanford University, AIAA Member, haomiao@stanford.edu

Ph.D. Candidate, Department of Aeronautics and Astronautics, Stanford University. AIAA Member. stevenw@stanford.edu

Associate Professor, Department of Aeronautics and Astronautics; Director, Hybrid Systems Laboratory, Stanford Univer-
sity. Associate Professor, Department of Electrical Engineering and Computer Sciences, University of California at Berkeley.
AIAA Member. tomlin@stanford.edu
1 of 17
American Institute of Aeronautics and Astronautics
Figure 1. STARMAC II quadrotor aircraft unmanned aerial vehicle (UAV), in ight, with autonomous attitude
and altitude control. This is a vehicle of the Stanford Testbed of Autonomous Rotorcraft for Multi-Agent
Control (STARMAC) project. Applications include search and rescue, surveillance operation in cluttered
environments, and mobile sensor networks. Operation throughout the ight envelope allows characterization
of the aerodynamic disturbance eects on the control system, caused by vehicle motion relative to the free
stream.
in air speed is signicant. Although many of the eects have been discussed in the helicopter literature,
25
their inuence on quadrotors has not been comprehensively explored. This work focuses on four main eects,
two that impact attitude control and two that impact altitude control as they pertain to quadrotors. First,
for attitude control, blade apping results from the diering inow velocities experienced by the advancing
and retreating blades and induces both roll and pitch moments applied at the blade root as well as a change
in the direction of the thrust vector away from the horizontal plane. Second, it was observed that interfer-
ence caused by the vehicle body in the downstream rotor ow resulted in unsteady thrust behavior and poor
attitude tracking. This eect was demonstrated to be signicantly reduced by locating the rotors further
apart. Third, for altitude control, the vertical component of the vehicle velocity directly impacts the thrust
produced, resulting in a negative eect when the vehicle is climbing and a positive eect while descending.
Finally, the combination of translational velocity and vehicle angle of attack can result in increases to the
expected thrust produced at a given power. For all but motor positioning, a theoretical derivation of the
eect is developed based on previous work, and the specic impact on quadrotor dynamics are dened. The
eects are then validated through thrust test stand experiments, and by ight tests of the Stanford Testbed
of Autonomous Rotorcraft for Multi-Agent Control (STARMAC).
We proceed with a brief survey of development eorts for quadrotor vehicles in Section II. Section III
presents details of the test stand apparatus and the STARMAC II testbed, and the nonlinear vehicle dynamics
for quadrotors are then summarized in Section IV. In Sections V and VI, we present analysis of each of the
four aerodynamic eects as they pertain to quadrotor vehicles, along with experimental results demonstrating
their presence in both thrust test stand experiments and in ight test recordings. Possible control techniques
are also presented to accommodate the resulting nonlinearities, and nally, ight results for outdoor hover
are presented in Section VII.
II. Background
Although the rst successful quadrotors ew in the 1920s,
4
no practical quadrotor helicopters have
been built until recently, largely due to the diculty of controlling 4 motors simultaneously. The only
manned quadrotor helicopter to leave ground eect was the Curtiss-Wright X-19A in 1963, though it lacked
a stability augmentation system to reduce pilot work load, rendering stationary hover near impossible,
6
and development stopped at the prototype stage. Recently, advances in microprocessor capabilities and in
micro-electro-mechanical-system (MEMS) inertial sensors have spawned a series of radio-controlled (RC)
quadrotor toys, such as the Roswell yer (HMX-4),
7
and Draganyer,
8
which include stability augmentation
systems to make ight more accessible for RC pilots.
2 of 17
American Institute of Aeronautics and Astronautics
Many research groups are now working on quadrotors as UAV testbeds for control algorithms for au-
tonomous control and sensing,
7, 915
consistently selecting vehicle sizes in the range of 0.3 - 4.0 kg. Several
testbeds have achieved control with external tethers and stabilizing devices. One such system,
7
based on the
HMX-4, was own with the RC gyro augmentation system active, and with X-Y motion constraints. Alti-
tude and yaw control were demonstrated using feedback linearized attitude control and backstepping control
was applied for position, while state estimation was accomplished with an oboard computer vision system.
Another tethered testbed
14
used an extensive outward facing sensor suite of IR and ultrasonic rangers to
perform collision avoidance. Control of the vehicle was achieved using a robust internal-loop compensator,
and computer vision was used for positioning. A third project
16
relied on a tether to use a POLYHEMUS
magnetic positioning system. Tight position control at slow speeds was demonstrated using a non-linear
control technique based on nested saturation for lateral control with linearized equations of motion, though
compensating for attitude tilted thrust vectors in altitude control.
Other projects have relied on various nonlinear control techniques to perform indoor ights at low ve-
locities without a tether. One such project,
11
consisting of a modied Draganyer quadrotor helicopter,
has demonstrated successful attitude and altitude control tests using a nonlinear control scheme. The OS4
quadrotor project
10
features its own vehicle design and identies dynamics of the vehicle beyond the basic
nonlinear equations of motion, including gyroscopic torque, angular acceleration of blades, drag force on
the vehicle, and rotor blade apping as being potentially signicant, though the forces are not quantied or
analyzed further. A proportional-derivative (PD) control law led to adequate hovering capability, although
the derivative of the command rate was not included in the control law to maneuver the vehicle. A Lyapunov
proof proved stability of the simplied system in hover, and successful attitude and altitude control ights
were achieved. A third project
12
achieved autonomous hover with IR range positioning to walls indoors, with
a stability proof under the assumed dynamics. The system was modied to incorporate ultrasonic sensors,
17
and later incorporated two cameras for state estimation
18
as well.
Several vehicles saw success using Linear Quadratic Regulator (LQR) controllers on linearized dynamic
models. The Cornell Autonomous Flying Vehicle (AFV)
13
designed and built a custom airframe with brush-
less motors controlled by custom circuitry to improve resolution. Position control was accomplished using
dead-reckoning estimation, with a human input to null integration error. The MIT multi-vehicle quadrotor
project
19
uses an oboard Vicon position system to achieve very accurate indoor ight of the Draganyer V
Ti Pro, and demonstrated multiple vehicles ying simultaneously. The vehicles are capable of tracking slow
trajectories throughout an enclosed area that is visible to the Vicon system. It is possible to observe, in
ight videos presented with the paper, the downwash from one vehicle disturbing another vehicle in ight,
causing a small rocking motion, possibly due to blade apping.
At Stanford, there has been considerable prior work on quadrotor helicopters as well. The STARMAC I
project was a testbed of two vehicles that performed GPS waypoint tracking using an inertial measurement
unit (IMU), an ultrasonic ranger for altitude, and an L1 GPS receiver.
9
The testbed was derived from a
Draganyer aircraft, and weighed 0.7 kg. In order to improve attitude control, this project found that frame
stiening greatly improved attitude estimation from the IMU. Also, aerodynamic disturbances were observed
with this testbed, and modeled using ight data.
20
Despite the substantial interest in quadrotor design for autonomous vehicle testbeds, very little attention
has been paid to the aerodynamic eects that result from multiple rotors, and from motion through the
free stream. Exceptions to this trend include work from a group in Velizy, France
21
which investigates drag
forces due to wind and presents a control law to handle such forces should they be estimated. Also, many
important aerodynamic phenomenon were identied in the X-4 Flyer project at the Australian National
University.
22
The project considers the eects of blade apping, roll and pitch damping due to diering
relative ascent rates of opposite rotors, as well as dynamic motor modeling. Preliminary results of the
inclusion of aerodynamic phenomena in vehicle and rotor design show promise in ight tests, although an
instability currently occurs as rotor speed increases, making untethered ight of the vehicle impossible.
15
In the following sections, this paper extends the investigation of quadrotor aerodynamics, as they pertain
to position control and trajectory tracking ight. The eects of aerodynamics on a moving quadrotor
helicopter are analyzed, through theory, and by experiment. The next section presents the test apparatus
used.
3 of 17
American Institute of Aeronautics and Astronautics
Figure 2. Thrust test stand used to capture experimental aerodynamic data. A load cell measures the force
and torque exerted on the mounting point. Battery monitoring circuitry measures motor voltage and current.
Data is captured to the computer using an Atmel processor for A/D at 400 Hz.
III. Experimental Setup
The experimental equipment consisted of two primary components: a thrust test stand and prototype
quadrotor aircraft, STARMAC II. The thrust test stand permitted research into the performance of individual
motors and rotors, in varying ight conditions, while STARMAC II permitted experiments with an actual
quadrotor vehicle through indoor and outdoor ight testing. This section presents the relevant details of the
two systems.
A. Thrust Test Stand
In order to evaluate motor and rotor characteristics, a thrust test stand was developed, shown in Figure 2.
It measures the forces exerted on a lever using a load cell. The mounting point on the lever is adjustable
to allow load sensitivity to be varied. An Atmel microprocessor board was programmed to perform motor
control through its pulse width modulation (PWM) outputs, and to acquire analog inputs from the load cell,
current sensor, and battery voltage.
The microprocessor board interfaces with a data acquisition program on the PC to perform automated
tests, making measurements at 400 samples per second, well over twice the frequency of typical rotor rota-
tions. To perform some experiments, external wind was applied using a fan. Wind speeds were measured
using a Kestral 1000 wind meter, with a rated accuracy of 3%.
B. STARMAC II Quadrotor
The STARMAC II quadrotor helicopter, part of the Stanford Testbed of Autonomous Rotorcraft for Multi-
Agent Control (STARMAC) project, was designed and built to carry computer and sensor payloads for UAV
projects, as shown in Figure 3. The vehicle was fabricated from carbon ber and berglass honeycomb, using
lessons learned from STARMAC I. The nylon rotors, with a 0.127 m radius, are driven by sensorless, brushless
motors, controlled by commercial electronic speed controllers (ESCs). The combined thrust capacity is 4.0 kg
for the 1.3 kg vehicle. It has own with up to 1.1 kg of additional payload, including a Pentium-M computer
and environment sensors.
Vehicle state sensors include a Microstrain 3DM-GX1 inertial measurement unit (IMU), an ultrasonic
ranger for altitude, and a Novatel Superstar II GPS receiver. An onboard processor computes the carrier
phase dierential GPS solution for single-centimeter level accuracy. The sensor suite is able to accurately
characterize the state of the aircraft to measure the eects of aerodynamic disturbances.
The vehicle is own in the conned space of the lab, as well as outdoors. The control capabilities currently
allow the vehicle to be own in two modes, both with altitude control. In the rst mode, attitude-controlled,
a human uses a joystick to send attitude and altitude commands to the aircraft. In doing so, the human
4 of 17
American Institute of Aeronautics and Astronautics
Figure 3. STARMAC II quadrotor aircraft unmanned aerial vehicle (UAV), part of the Stanford Testbed of
Autonomous Rotorcraft for Multi-Agent Control project. The air frame is 0.56 m diagonally from motor to
motor. The vehicle is capable of autonomous GPS waypoint tracking and ying aggressive maneuvers following
attitude commands, indoors and outdoors.
controls the position. In the second mode, position-controlled, the aircraft controls its attitude to navigate
to a commanded GPS waypoint.
IV. Inertial Dynamics
The derivation of the nonlinear dynamics is performed in North-East-Down (NED) inertial and body
xed coordinates. Let {e
N
, e
E
, e
D
} denote unit vectors along the respective inertial axes, and {x
B
, y
B
, z
B
}
denote unit vectors along the repsective body axes, as dened in Figure 4. Euler angles of the body axes are
{, , } with respect to the e
N
, e
E
and e
D
axes, respectively, and are referred to as roll, pitch and yaw.
The current velocity direction unit vector is e
v
, in inertial coordinates, and denes coordinates relative to
the c.g. referred to as longitudinal, lateral and vertical. The rotor plane does not necessarily align with the
x
B
, y
B
plane, so let {x
R
, y
R
, z
R
} denote unit vectors aligned with the plane of the rotor and oriented with
respect to the lateral, longitudinal, and vertical directions as shown in Figure 5. Let r be dened as the
position vector from the inertial origin to the vehicle center of gravity (c.g.), and let
B
be dened as the
angular velocity of the aircraft in the body frame.
Figure 4. Free body diagram of a quadrotor aircraft.
5 of 17
American Institute of Aeronautics and Astronautics
The rotors, numbered 1 4, are mounted outboard on the x
B
, y
B
, x
B
and y
B
axes, respectively,
with position vectors r
i
with respect to the c.g. Each rotor produces a thrust, T
i
, that varies with rotor
angular velocity,
i
,
T
i
= C
T
A(
i
R)
2
(1)
where C
T
is the dimensionless coecient for rotor thrust, A = R
2
is the area swept out by the rotor, is the
density of air and R is the radius of the rotor. The value for C
T
can change due to wind and vehicle motion.
Note that the thrust acts perpendicularly to the rotor plane along the z
R
axis. Each rotor also produces an
aerodynamic moment, M
i
, which is a function of the motor torque as well as various aerodynamic eects
which are described in Section V. Rotors 1 and 3 rotate in the opposite direction of rotors 2 and 4, so that
counteracting aerodynamic torques can be used for yaw control, where the aerodynamic torques are equal
to the torques applied by the motors.
Note that the thrust is proportional to the voltage, applied to the motors, squared. The derivation and
supporting experimental data will be in the nal version of this paper. Also note that in the operating
regime of the aircraft, this quadratic curve is well approximated for small variations in voltage by linearized
variation in voltage.
Figure 5. Free body diagram of the moments and forces on a single rotor.
The vehicle drag force is dened as D
v
, vehicle mass is m, acceleration due to gravity is g, and the inertia
matrix is I R
33
. A free body diagram is depicted in Figure 4, with a depiction of the rotor forces and
moments in Figure 5. The total force, F, can be summed as,
F = D
v
e
v
+mge
D
+
4

i=1
(R
Ri,I
z
R
T
i
) (2)
where R
Ri,I
is the rotation matrix from the plane of rotor i to inertial coordinates
a
. Similarly, the total
moment, M, is,
M =
4

i=1
(M
i
+r
i
(T
i
R
Ri,B
z
R
)) (3)
where R
Ri,B
is the rotation matrix from the plane of rotor i to body coordinates. The full nonlinear dynamics
can be described as,
mr = F
I
B
+
B
I
B
= M
(4)
where the total angular momentum of the rotors is assumed to be near zero, because they are counter-
rotating.
V. Attitude Control
A. Blade Flapping in Translational Flight
Initial outdoor ights with STARMAC encountered an unexpected restorative eect damping commanded
attitudes. As translational velocity increased or in the presence of wind, the vehicle tended to settle toward
a
The notation R
A,B
shall refer to rotation matrices from coordinate system A to B throughout.
6 of 17
American Institute of Aeronautics and Astronautics
0

pitch and roll, limiting the velocities which could be attained (see Figure 9a). This damping was later
identied to be the result of rotor blade apping. In the following, rst the aerodynamics of blade apping
are presented from theoretical basis. Then, a comparison is made to experimental data, measured using the
thrust test stand. This section is concluded with STARMAC II ight test results, comparing the performance
of the attitude controller under nominal, low speed conditions, to the controller under translational motion,
with the center of gravity (c.g.) at varying heights. The results demonstrate the eect of speed of blade
apping, and consequently attitude control performance, as well as the impact that the vertical position of
the c.g. has on attitude dynamics.
1. Aerodynamics of Blade Flapping
In translational ight, the advancing blade of a rotor sees a higher eective velocity relative to the air, while
the retreating blade sees a lower eective velocity. This results in a dierence in lift between the two rotor,
causing the rotor blades to ap up and down once per revolution.
5
This apping of the blades tilts the rotor
plane back away from the direction of motion, which has a variety of eects on the dynamics of the vehicle,
in particular aecting stability in attitude.
15
The backwards tilt of the rotor plane generates a longitudinal
thrust, T
lon
,
T
lon
= T sin a
1s
(5)
where a
1s
is the angle by which the thrust vector T is deected (see Figure 1). If the center of gravity of
the vehicle is not aligned with the rotor plane, this longitudinal force will generate a moment about the c.g.
For sti rotors, the tilt of the blades also generates a moment at the rotor hub
M = k

a
1s
(6)
where k

is the stiness of the rotor blade in Nm/rad.

Coning (the upward exure of the rotor blades from the lift force on each blade) causes the impinging
airow to have another unbalanced forcing of the blades which causes a lateral tilt of the rotor plane, the
details of which are developed in Prouty.
5
This lateral tilt generates moments at right angles to the velocity
vector, but because of the symmetric position of quadrotor rotors, the lateral eects cancel. For sti 2-bladed
rotors, the moments due to the coning angles are symmetric about the rotor hub and also cancel.
A distinction must be noted here in the use of the terminology ap angle and deection angle a
1s
. The
ap angle of a rotor blade is typically dened in the helicopter literature as the total deection of a rotor
blade away from the horizontal in body coordinates at any point in the rotation, and is calculated as
= a
0s
a
1s
cos +b
1s
sin (7)
where a
0s
is the blade deection due to coning, a
1s
and b
1s
are the longitudinal and lateral blade deection
amplitudes, respectively, due to apping. is the azimuth angle of the blade, and is dened as 0 at the
rear. Since coning aects both blades equally, the deection of the thrust vector is due to both longitudinal
and lateral tilts. For quadrotor vehicles, however, the moments generated by lateral deections cancel,
and generation of unbalanced moments is due entirely to the longitudinal deection, a
1s
. The longitudinal
deection gives the amplitude of the rotor tilt fore and aft ( = , 0 rad), which we will refer to from here
on as the deection angle to avoid confusion with the apping angle, .
The equation for deection angle of a apping rotor with hinged blades is presented in Pounds as:
15
a
1s
=
1
1 +

2
lon
2
4
3
_
C
T

2
3

lon

a
0
+
lon
_
(8)
where a
0
is the slope of the lift curve per radian (typically about 6.0 for conventional airfoils at low Mach
numbers according to Prouty
5
),
lon
is the longitudinal rotor advance ratio, dened as the ratio of the
longitudinal to blade tip speed,

lon
=
v
lon
v
t
(9)
and is the nondimensional Lock number, which gives the ratio aerodynamic to centrifugal forces and is
dened as
=
a
0
cR
4
I
b
(10)
7 of 17
American Institute of Aeronautics and Astronautics
Figure 6. Diagram representing the eects of rotor blade apping
5
and the modeling of sti rotor blades as
hinged blades with an eective oset and blade stiness k

.
3
The rotor plane becomes tilted, resulting in a
deection of the thrust vector, and a moment is generated at the blade root.
where I
b
is the moment of inertial of the blade about the hinge, c is the chord of the blade, and R is the
rotor radius. is the solidity ratio of the rotor, and is dened as
=
A
b
A
(11)
where A
b
is the total area of the rotor blades.
Equation (8) predicts a roughly linear relationship between velocity and deection angle in the STARMAC
II operating regime. In practice this equation over-predicts the apping seen by rotors with unhinged blades
(see Figure 7) where the stiness of the blades must accounted for.
The apping properties of a sti, xed-pitch rotor blade can be analyzed by modeling the blade as being
hinged at an eective oset e
f
from the center of rotation (expressed as a percentage of the rotor radius) and
a torsional spring with stiness k

Nm/rad at the hinge.

3
This approximates the rst bending mode of the
blade and is sucient for the small angles we are concerned with. Both e
f
and and k

can be determined
by measuring the natural frequency
n
of blade vibration and using the following relations:
3, 5

n
=
_
k

I
b
(12)
e
f
=
1
3
4
b
2
I
b
k

(13)
where b is the number of blades and I
b
is the moment of eective moment of inertia of the blade about the
hinge at e
f
. Substituting Equation (12) into Equation (13), we have
e
f
=
1
3
4
b
2

2
n
(14)
The constants k

and I
b
can be obtained by determining the force required at the tip to deect the blade
through some angle and balancing moments:
F(1 e
f
)R = k

(15)
Substituting the value for k

determined in Equation (15) back into Equation (12) yields I

b
. With these
parameters, the equilibrium apping constants can be determined by solving the following matrix equation
developed in Newman:
3
_

0 0 0

lon
(1
2

8
0
0

8
(1
2

) 0
0 0 0 1
_

_
_

_
a
0s
a
1s
b
1s
CT
a0
_

_
=
_

8
0

lon

6
0

8
0 0

lon
0

8
0
1
3
0
1
2

lon

1
2
_

_
_

avg
A
B

ver
+
i
_

_
(16)
8 of 17
American Institute of Aeronautics and Astronautics
2 4 6 8
145
150
155
160
165
170
175
time (s)
M
e
a
s
u
r
e
d

H
o
r
i
z
o
n
t
a
l

F
o
r
c
e

(
g
r
a
m
s

f
o
r
c
e
)


0 m/s
0.6 m/s
1.1 m/s
1.5 m/s
1.8 m/s
2.5 m/s
3.4 m/s
(a)
0 1 2 3 4
0
1
2
3
4
5
Wind Velocity (m/s)
D
e
f
l
e
c
t
i
o
n

A
n
g
l
e

a
1
s

(
d
e
g
)


measured
predicted (stiff blades)
predicted (hinged blades)
(b)
Figure 7. (a) Horizontal force is measured at dierent wind velocities in order to calculate the apping angles
and to see the eect of apping. (b) The measured deection angle is compared with predicted values using
equations which assume hinged, freely apping blades and more complex ones which model sti, unhinged
blades. The hinged equations greatly over-predict the apping eect if used to analyze the behavior of sti
blades.
Equation (16) depends heavily on several parameters. Once again,
lon
and
ver
are the horizontal and
vertical advance ratios, respectively (
ver
= 0 in translational ight).
avg
, A, and B are constants which
dene the rotor pitch control:
=
avg
+Acos +B sin (17)
For xed pitch propellers, A, B = 0 and
avg
is the average pitch angle.

is the ratio of the apping frequency

to the angular rate of the rotor, and for sti propellers is

dened as

(18)
and can be calculated for use in Equation (16) as

=
_
(1 +
3
2
e
f
) +
k

I
b

2
(19)
2. Static Measurements of Flapping Eect
The lateral force due to the deection of the thrust vector by apping was measured for a single rotor by
blowing air at xed velocities across a spinning rotor attached to the test stand. This data was ltered and
used to calculate the average deection angle as a function of incident wind velocity and compared to the
predictions of the apping equations developed in Newman.
3

for the apping equations was also measured

using the test stand, giving an eective hinge oset of 25%. k
beta
was measured to be 0.23Nm/rad. The
results are plotted in Figure 7.
Although the deection angles noted in Figure 7 are very small, they represent a signicant eect on
STARMAC II. The typical control eort for low-velocity attitude tracking as shown in Figure 8 is a dierential
on the order of 3 percent of hover thrust, generating a moment of 0.06Nm. At the STARMAC II hover
thrust of 3.4N per motor, a 0.75

deection of the thrust vector, seen at speeds of about 2.5m/s, results in

a moment about the c.g. of 0.022Nm, when the c.g. was located 12cm below the rotor plane. The total
moment from the blade stiness of all four rotors with deection angles of 0.75

was 0.024Nm. Therefore,

at 2.5m/s, the aerodynamic moments generated on the vehicle by apping can be as high as 75% of the
commanded torque. STARMAC II has been own at speeds of up to 3.5m/s, where the moment caused by
apping can dominate over attitude control commands.
9 of 17
American Institute of Aeronautics and Astronautics
62 64 66 68 70 72
10
5
0
5
10
15
20
25
time (s)
R
o
l
l

(
d
e
g
)

a
n
d

V
e
l
o
c
i
t
y

(
m
/
s
)


Actual Roll
Commanded Roll
Velocity
Figure 8. At low velocities, i.e. for small displacements from hover, a PID controller is sucient for good
attitude control.
3. Attitude Control in Nominal Conditions
At low velocities and with small aerodynamic disturbances (for example in indoor ight), PID control is fully
sucient for good tracking of commanded attitude since the vehicle approximates a double-integrator with
a rst-order lag from the motor dynamics. For initial test ights indoors with STARMAC, good tracking
was obtained with simple PD controllers for pitch and roll (see Figure 8) giving tracking errors on the order
of 2 3

.
4. Eect of Flapping on Attitude Control
In translational ight, the pitch and roll dynamics of a quadrotor are very sensitive to rotor blade apping.
There is a moment generated at the hub of each rotor by the aerodynamic forces creating the blade deection,
and a moment about the center of gravity from the horizontal deection of the thrust if the plane of the
rotors is not in line with the c.g. This means that position and attitude control are not completely decoupled,
as has been assumed in the past.
20
In Figure 9, the eects of the blade apping moments and c.g. location
can be seen during a step input in the pitch command. Initially, the control eort commanded by the PD
controller is sucient to bring the vehicle toward the commanded pitch. As the speed increases, the restoring
moments caused by blade apping increase until the commanded torque is insucient to hold the vehicle
at commanded pitch despite an increase in the pitch error. With the c.g. below the plane of the rotors
as in Figure 9a, the eect is especially strong, since the horizontal component of the thrust deection due
to apping adds another moment in addition to that generated directly by the blades apping. With the
c.g. relocated to be level with the rotor plane, this extra moment is eliminated. This dierence can be seen
in Figure 9b, where the pitch gets much closer to the commanded value, and the aircraft also settles at a
higher equilibrium velocity since the only restoring moment is through the aerodynamic moments acting
directly on the blades. The eect of the position of the center of gravity is strong. Flight logs show that
the commanded torque in any axis is linearly correlated to the speed in the corresponding direction. The
magnitude of the linear coecient increases as the c.g. is shifted below the plane of the rotors. Using linear
regression on ltered ight data, the coecient was found to be 0.014Nm when the c.g. was 0.02 m below
the rotor plane, and 0.023Nm when the c.g. was 0.12 m below the rotor plane. Note that typical variation
in the commanded torque in stationary hover is on the order of these values.
B. Eect of Airow Disruption on Attitude Control
1. Motor Placement and Vortex Impingement
Motor placement and vortex impingement on the aircraft body had a signicant eect on attitude stability.
Unfortunately, we did not have time to prepare results for inclusion by the paper deadline. Details will
be included in the nal version of this paper. They will include the predicted dynamics, using the inertial
10 of 17
American Institute of Aeronautics and Astronautics
57 58 59 60 61 62 63
10
0
10
20
R
o
l
l

A
n
g
l
e

(
d
e
g
)


Actual
Commanded
57 58 59 60 61 62 63
1
0
1
2
3
4
Time(s)
W
e
s
t
w
a
r
d

V
e
l
o
c
i
t
y

(
m
/
s
)
(a)
17 18 19 20 21 22 23
10
0
10
20
R
o
l
l

A
n
g
l
e

(
d
e
g
)


Actual
Commanded
17 18 19 20 21 22 23
1
0
1
2
3
4
Time(s)
W
e
s
t
w
a
r
d

V
e
l
o
c
i
t
y

(
m
/
s
)
(b)
Figure 9. The eect of rotor blade apping is shown in the vehicles response to a step input command in pitch
angle. The controller is a simple PD controller, without an integral component, in order to demonstrate the
eects of rotor blade apping on the quadrotor in translational motion. a Center of gravity c.g. 0.12 m below
the plane of the rotors. As the aircraft translates, the thrust vector is deected, causes a restoring moment
about the c.g., adding to the moment from the apping of the blades to force the vehicle to tilt back. b C.G.
0.02 m below the plane of rotors, so horizontal force from the tilted thrust vectors generates less moment. Note
the reduced undershoot and higher speed achieved for the same commanded attitude.
dynamics and a motor model,
23
and the dynamics observed experimentally, given the initial frame design.
Modications were made to the air frame to test several physical congurations for aerodynamic interference
with the rotor wash. Upon converging on a suitable frame design, a root locus analysis was done for
proportional and derivative parameters, separately factored out, to aid in online controller tuning to improve
performance.
2. Eect of Protective Shrouds on Yaw Control
Yaw control is achieved by giving dierential commands to rotors spinning in opposite directions to generate
a torque about the vertical axis which is used for control through a PID controller. We have noticed that the
torque of a motor at a given command voltage depends on the airow through the rotor. The initial design
of STARMAC II called for protective circular shrouds around each propeller. In this conguration, it was
very dicult to get good or even consistent yaw tracking performance (see Figure 10a). The shrouds were
simple protective enclosures which were not designed with aerodynamic considerations in mind. With the
shrouds removed, yaw tracking instantly improved from errors of roughly 10

to less about 3

(see Figure
10b). Subsequent analysis of the data showed that during shrouded ights, measured angular accelerations
of the vehicle did not consistently match with motor commands. The mostly likely explanation is that the
shrouds were disturbing or disrupting the ow of air through the rotors, causing the actual motor torque to
vary for a given commanded voltage level. Replacing the shrouds with xed guards away from the rotors
eliminated this problem.
VI. Altitude Control
A. Eect of Angle-of-Attack on Thrust in Translational Flight
1. Aerodynamics of Induced Power
The induced velocity, v
i
, is the change in air speed across the blades, with respect to the free stream velocity,
v
inf
, induced by the rotor blades. The induced power is the required power input to create the induced
velocity. As a rotorcraft undergoes translational motion, the induced power requirement of a rotorcraft
changes. Note that v
inf
is the total free stream velocity, not just the translational component. To derive
the eect of free stream velocity on induced power, from conservation of momentum, the induced velocity,
11 of 17
American Institute of Aeronautics and Astronautics
0 10 20 30 40 50
10
5
0
5
10
15
20
time (s)
Y
a
w

(
d
e
g
)


actual yaw
commanded yaw
(a)
0 10 20 30 40 50 60
10
5
0
5
10
15
20
time (s)
Y
a
w

(
d
e
g
)


actual yaw
commanded yaw
(b)
Figure 10. The eect of shrouds on yaw control. a) With shrouds, unless the shrouds are built to very exacting
specications yaw control performance is highly variable since instability in the air owing through the shrouds
leads to highly varying torques on the rotors. b) With the shrouds removed, performance of the yaw control
loop is instantly improved.
v
i
for an ideal vehicle can be found by solving
4
v
i
=
v
2
h
_
(v

cos )
2
+ (v

sin +v
i
)
2
(20)
for v
i
, is the angle of attack, with positive corresponding to pitching forward, and v
h
is the induced velocity
at hover conditions. From momentum theory,
4
v
h
=

T
2A
(21)
where T is the thrust produced by the rotor to remain in hover. For a quadrotor helicopter, this is equal to
1
4
T
nom
, where the nominal thrust T
nom
is equal to the weight of the vehicle.
The analytical solution for v
i
is the solution to the quartic polynomial of Equation (20), and its expression
is too large to include
b
. Using the expression for v
i
, or a numerical solution, the ideal thrust per power input
can be computed, using
T =
P
v

sin +v
i
(22)
where the denominator corresponds to the air speed across the rotors. The value of the ratio of thrust to
hover thrust, T/T
h
, is plotted for the v
h
of STARMAC II, in Figure 12. At low speeds, the angle of attack
has vanishingly little eect on T/T
h
. However, as speed increases, the ratio T/T
h
becomes increasingly
sensitive to the angle of attack, varying by a substantial fraction of the aircrafts capabilities, within the
ight envelope. Similar to an airplane, pitching up increases the lift force. The angle of attack for which
thrust is at the hover value increases with forward speed. For level ight, the power required to retain
altitude increases with the forward speed.
2. Altitude Control in Nominal Conditions
Altitude is controlled using a modied version of the Integral Sliding Mode (ISM) controller described in
previous work.
20
The modication adjusts the control input u
alt
to be
u
alt
= u
p
+u
d
+
T
nom
cos cos
(23)
b
Although the solution is too long to print here, it will be provided in a Matlab script online.
12 of 17
American Institute of Aeronautics and Astronautics
0.96
0
.9
8
0.98
1
1
1
.
0
2
1
.0
2
1
.
0
4
1.04
1
.
0
6
1.06
1
.
0
8
1
.0
8
1
.
1
1.1
1
.
1
2
1
.1
2
1
.
1
4
1
.1
4
1
.1
6
1
.
1
8
1
.
2
1
.
2
2
1
.
2
4
1
.
2
6 1
.
2
8
1
.
3
1
.
3
2
1
.
3
4
1
.
3
6
1
.
3
8
1
.
4
1
.
4
2
1
.
4
4
1
.
4
6
1
.
4
8
1
.
5
1
.
5
4
1
.
6
1
.
6
6
Flight Speed (m/s)
A
n
g
l
e

o
f

A
t
t
a
c
k

(
d
e
g
)
(T/T
h
)
P=const
for v
h
=6 m/s


1 2 3 4 5 6
20
15
10
5
0
5
10
15
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
Figure 11. The thrust generated at a constant power, Pconst, varies depending on vehicles speed and angle of
attack relative to the incoming ow (the angle of attack is positive when pitching down). As speed increases,
the variation of thrust with the angle of attack becomes more extreme, causing strong control disturbances,
particularly in attitude, as shown in Figure 12.
where u
p
and u
d
are the proportional and derivative control terms commanded by the ISM controller, and
and are the pitch and roll angles respectively. The third term is the steady state oset required to maintain
an altitude adjusted for the loss of vertical thrust due the deection of the thrust vector from rolling and
pitching.
20
3. Flight Results
In general, the ISM controller has proved to be very eective in altitude control, though performance can
be improved by better ltering of the ultrasonic altitude sensor readings (see Figure 12a). It must provide
strong active damping whenever descent velocity is encountered. Otherwise, attitude oscillations have been
observed to occur, due to an apparent drop in thrust during small descent velocities, as predicted by the
induced velocity model results, shown in Figure 13. However, with strong damping, this eect has been
reduced, as shown in Figure 12a.
Apparent degradation in the altitude control capabilities as speed increases has been found to be due
to a strong aerodynamic eect. The thrust variation created by dierent angles of attack at varying speeds
and wind conditions can have a substantial, systematic, nonlinear eect which requires wind estimation to
be compensated for, as was derived in Section 1.
These eects of forward speed are strongly felt by the altitude control loop, even with the disturbance
rejection provided by an integrator term. For slow changes in motion, although disturbances occur, their
slow speed allows the controller to reject them eectively. However, the quadrotor is able to rapidly pitch and
roll, leading to disturbances that are dicult to reject at non-zero speeds, as shown in Figure 12. The ability
to counteract this eect in control is compounded by the diculty of accurately measuring the ambient wind.
This is demonstrated in Figure 12b, where the vehicle was operating under human control of attitude
commands, and with the roll axis aligned with North/South. A positive roll was commanded to arrest
the vehicles sideways progress, but as it rotated through 0

roll while at a velocity of 3.5m/s, the thrust

increased such that the controller was unable to compensate for the upward force and caused the vehicle to
balloon high above the commanded altitude.
13 of 17
American Institute of Aeronautics and Astronautics
75 80 85 90 95
0.4
0.45
0.5
0.55
0.6
0.65
0.7
Time (s)
A
l
t
i
t
u
d
e

(
m
)


Actual
Commanded
(a)
20 21 22 23 24 25
0
20
40
R
o
l
l

(
d
e
g
)


20 21 22 23 24 25
0
0.5
1
A
l
t
i
t
u
d
e

(
m
)


20 21 22 23 24 25
2
0
2
Time (s)
E
a
s
t

V
e
l
o
c
i
t
y

(
m
/
s
)
actual
commanded
actual
commanded
(b)
Figure 12. The eect of angle of attack and velocity on thrust is demonstrated during a ight with human-
commanded attitude. (a) Altitude data from a typical ight near hover with sinusoidal roll commands, with
a usual error of 10cm. Note the sharp spikes which indicate noise in the ultrasonic altitude sensor. (b) Data
from a subsequent ight. At just past 22, s, a positive roll is commanded to reduce the vehicles westward
velocity (the vehicle is aligned such that the roll axis points north/south). As the vehicle rolls through 0

, it
still has a large non-zero velocity, giving it zero angle of attack relative to the incoming ow. This causes the
thrust to increase (see Figure 11), pushing the vehicle above the commanded altitude. Note that the altitude
reference is time varying, though not shown here.
B. Ascending and Descending Rotors
Rotorcraft have three operational modes for climb velocity,
4
v
c
,
1. Normal working state: 0
vc
v
h
2. Vortex ring state (VRS): 2
vc
v
h
< 0
3. Windmill brake state:
vc
v
h
< 2
In normal working state, air is owing down through the rotor. In windmill brake state, air is owing
up through the rotor due to rapid descent. For these two states, conservation of momentum can be used
to derive the induced velocity. For the normal working state, the hover and ascent condition, the induced
velocity is
4
v
i
=
v
c
2
+
_
_
v
c
2
_
2
+v
2
h
(24)
For the windmill braking state, rapid descent, it is
v
i
=
v
c
2

_
_
v
c
2
_
2
v
2
h
(25)
In the vortex ring state, air recirculates through the blades in a periodic and somewhat random fashion.
As a result, the induced velocity varies greatly, particularly over the domain 1.4 v
c
/v
h
0.4, reducing
aerodynamic damping.
24
An empirical model of induced velocity in vortex ring state is
4
v
i
= v
h
_
+k
1
_
v
c
v
h
_
+k
1
_
v
c
v
h
_
+k
2
_
v
c
v
h
_
2
+k
3
_
v
c
v
h
_
3
+k
4
_
v
c
v
h
_
4
_
(26)
where k
1
= 1.125, k
2
= 1.372, k
3
= 1.718, k
4
= 0.655. This model compares with the mean of
experimental results in literature, though it fails to capture the periodic nature of the vortex entrapment.
4
To model the dynamics during climbing, the power is the thrust times the speed it is applied at,
P = T(v
c
+v
i
) (27)
14 of 17
American Institute of Aeronautics and Astronautics
1 0.5 0 0.5 1
0.85
0.9
0.95
1
1.05
Vc

m
s

T
/
T
h
P
(
W

1
)
(a)
3 2 1 0 1 2
0.9
1
1.1
1.2
1.3
Velocity (m/s)
T

/

T
h
(b)
Figure 13. a) Theoretically predicted eect of climb velocity, vc, on normalized thrust, per unit of power input
and b) thrust test stand measured eect of vc on normalized thrust. The hover velocity is for a vehicle with the
same parameters as STARMAC II. In ascent, there is a virtual damping force close to linearly proportional to
speed. However, for descent at a slow velocity, there is no damping force, rather there may be a small negative
damping force. The ascent equations are based on momentum conservation of the airstream, and the descent
equations are based on published experimental curve ts.
ignoring prole power losses. Note that Tv
c
is the power consumed by the climbing motion, whereas Tv
i
is
the power transfered into the air. It is typically desirable to avoid the vortex ring state. Note that one way
to avoid this state is to maintain a substantial forward speed.
3
We can compute the thrust achieved for a given input power, as a function of climb velocity, by sub-
stituting Equation (24), Equation (25), and Equation (26) into Equation (27). For the ight conditions
experienced by STARMAC II, the ratio of the thrust to hover thrust, per power input, is shown in Figure
13, where plot (a) is the theoretical curve, using the solution to the above equations, and plot (b) shows
data from a thrust test stand experiment using a vertical wind disturbance. As shown in the plots, there
is certainly a damping eect associated with climbing, with thrust reducing nearly proportionally to the
climb velocity. However, in decent, for low speeds, there is essentially no damping. There is possibly nega-
tive damping for small descent rates, although the model of Equation (26) is an approximation based on a
compilation of experimental data in literature.
In thrust test stand experiments, the loss of thrust with an applied climb velocity was experienced. The
descent velocity experiments were much less conclusive. The average thrust remained close to the zero climb
velocity value, though it oscillated substantially, making assessment of the change in thrust dicult. A more
formal summary of the data will be included in the nal version of this paper.
VII. Flight Results
Position control is currently implemented using a PID controller which actuates the vehicles roll and
pitch as control inputs. Tilting the vehicle in any direction causes a component of the thrust vector to point
in that direction, so commanding pitch and roll is directly analogous to commanding accelerations in the
X-Y plane. The inner-loop dynamics are assumed to be fast enough not to aect position control. Figure
14 shows a typical ight using this control scheme. The vehicle is able to stay inside of a 1.5m-radius circle,
but the current control implementation has little ability to reject disturbances from wind and translational
velocity eects. A key weakness of this and similar position controllers used by other groups is the assumption
that position control and attitude control are decoupled. This is in fact only true for very small velocities.
As shown in Section V, there is signicant coupling between the velocity of the aircraft and the attitude
dynamics. The PD and LQR-derived PD attitude controllers currently used by many groups
25
would be
unable to cope with this coupling, leading to loss of performance for both attitude and position control.
Ultimately, the dynamics of a quadrotor in ight in attitude, altitude, and position are not decoupled to
the extent previously assumed.
9, 20, 25
The aerodynamics in ight cause disturbance forces to be generated
in each set of states to be controlled. If these eects are not accounted for, control performance can be
15 of 17
American Institute of Aeronautics and Astronautics
2 3 4 5 6
1
1.5
2
2.5
3
3.5
4
4.5
East (m)
N
o
r
t
h

(
m
)


trajectory
Error Circle
Desired Point
Start
Stop
Figure 14. Hover position control with simple PD control commanding attitude achieves a maximum error
of about 2m, with the integrator turned o (in poth the position and the attitude control loops) to keep
the dynamics simple. Thus, the quadrotor helicopter was very vulnerable to wind gusts, demonstrating the
disturbances experienced, without the dynamics introduced by the integrator. Tests such as these showed
what the underlying aerodynamic phenomena were that must be addressed.
greatly diminished. In order to achieve good tracking of either altitude, attitude, or position, the inner-loop
controllers must have good enough disturbance rejection to negate the coupling eects from translational
and vertical velocities as well as external disturbances like wind. Vehicle design can also reduce or eliminate
some of these eects (i.e. location of the c.g. and rotor spacing). In the nal version of the paper, we will
present results for improved autonomous hover and trajectory tracking with controllers incorporating the
aerodynamic eects discussed in earlier sections.
VIII. Conclusions
Quadrotor helicopters are popular as testbeds for small UAV development, but their aerodynamics are
complex and have not been well addressed. Although many good control results have been reported in previ-
ous work, these have focused around simple trajectories at low velocities, in a controlled indoor environment.
In this paper, we have addressed a number of issues observed in quadrotor aircraft operating at speed and
in the presence of wind disturbances. We have explored the resulting forces and moments applied to the
vehicle through these aerodynamic eects and investigated their impact on attitude and altitude control.
We have uncovered the extent of their inuence using data from static measurements and ight data from
the STARMAC II quadrotor. These results have shown that existing models and control techniques are in-
adequate for accurate trajectory tracking at speed and in uncontrolled environments. Careful consideration
of these disturbances has allowed us to improve both the physical conguration and control design of the
STARMAC II quadrotor, improving attitude and altitude tracking performance and permitting controlled,
stable ight at higher velocities and in the presence of gusting winds. This work should open the door to
improved autonomous hover and trajectory tracking in the near future, enabling most of the applications
that have been envisaged for the STARMAC testbed.
Acknowledgments
The authors would like to thank Jung Soon Jang, David Shoemaker, David Dostal, Dev Gorur Raj-
narayan, Vijay Pradeep, Paul Yu, and Justin Hendrickson, for their many contributions to the development
of the STARMAC testbed.
16 of 17
American Institute of Aeronautics and Astronautics
References
1
Homann, G. M., Waslander, S. L., and Tomlin, C. J., Distributed Cooperative Search using Information-Theoretic Costs
for Particle Filters with Quadrotor Applications, Proceedings of the AIAA Guidance, Navigation, and Control Conference,
Keystone, CO, August 2006.
2
Stepniewski, W. Z., Rotary-wing aerodynamics, Dover Publications, New York, NY, 1984.
3
Newman, S., The Foundations of Helicopter Flight, Halsted Press, New York, NY, 1994.
4
Leishman, J. G., Principles of Helicopter Aerodynamics, Cambridge University Press, New York, NY, 2000.
5
Prouty, R. W., Helicopter Performance, Stability, and Control , Krieger Publishing Company, Malabar, FL, 1990.
6
Anderson, S. B., Historical Overview of V/STOL Aircraft Technology, NASA Technical Memorandum 81280, Ames
Research Center, Moett Field, CA, March 1981.
7
Altu g, E., Ostrowski, J. P., and Taylor, C. J., Quadrotor Control Using Dual Camera Visual Feedback, In Proceedings
of the IEEE International Conference on Robotics and Automation, Taipei, Taiwan, Sept 2003, pp. 42944299.
8
DraganFly-Innovations, 2006, http://www.rctoys.com.
9
Homann, G., Rajnarayan, D. G., Waslander, S. L., Dostal, D., Jang, J. S., and Tomlin, C. J., The Stanford Testbed
of Autonomous Rotorcraft for Multi Agent Control (STARMAC), In Proceedings of the 23rd Digital Avionics Systems Con-
ference, Salt Lake City, UT, November 2004.
10
Bouabdallah, S., Murrieri, P., and Siegwart, R., Towards Autonomous Indoor Micro VTOL, Autonomous Robots,
Vol. 18, No. 2, March 2005, pp. 171183.
11
Guenard, N., Hamel, T., and Moreau, V., Dynamic modeling and intuitive control strategy for an X4-yer, In Pro-
ceedings of the International Conference on Control and Automation, Budapest, Hungary, June 2005, pp. 141146.
12
Escare no, J., Salazar-Cruz, S., and Lozano, R., Embedded control of a four-rotor UAV, Proceedings of the AACC
American Control Conference, Minneapolis, MN, June 2006, pp. 39363941.
13
Nice, E. B., Design of a Four Rotor Hovering Vehicle, Masters thesis, Cornell University, 2004.
14
Park, S., Won, D., Kang, M., Kim, T., Lee, H., and Kwon, S., RIC (Robust Internal-loop Compensator) Based Flight
Control of a Quad-Rotor Type UAV, Proceedings of the IEEE/RSJ International Conference on Intelligent Robotics and
Systems, Edmonton, Alberta, August 2005.
15
Pounds, P., Mahony, R., and Corke, P., Modelling and Control of a Quad-Rotor Robot, In Proceedings of the Aus-
tralasian Conference on Robotics and Automation, 2006.
16
Castillo, P., Dzul, A., and Lozano, R., Real-Time Stabilization and Tracking of a Four-Rotor Mini Rotorcraft, IEEE
Transactions on Control Systems Technology, Vol. 12, No. 4, 2004, pp. 510516.
17
Kendoul, F., Lara, D., Fantoni, I., and Lozano, R., Nonlinear control for systems with bounded inputs: Real-time
embedded control applied to UAVs, Proceedings of the IEEE Conference on Decision and Control , San Diego, CA, December
2006, pp. 58885893.
18
Romero, H., Benosman, R., and Lozano, R., Stabilization and location of a four rotor helicopter applying vision,
Proceedings of the AACC American Control Conference, Minneapolis, MN, June 2006, pp. 39303935.
19
Valenti, M., Bethke, B., Fiore, G., How, J. P., and Feron, E., Indoor Multi-Vehicle Flight Testbed for Fault Detection,
Isolation, and Recovery, In Proceedings of the AIAA Guidance, Navigation and Control Conference, Keystone, CO, August
2006.
20
Waslander, S. L., Homann, G. M., Jang, J. S., and Tomlin, C. J., Multi-Agent Quadrotor Testbed Control Design:
Integral Sliding Mode vs. Reinforcement Learning, In Proceedings of the IEEE/RSJ International Conference on Intelligent
Robotics and Systems 2005, Edmonton, Alberta, August 2005, pp. 468473.
21
Mokhtari, A. and Benallegue, A., Dynamic Feedback Controller of Euler Angles and Wind parameters estimation for
a Quadrotor Unmanned Aerial Vehicle, In Proceedings of the IEEE International Conference on Robotics and Automation,
New Orleans, LA, April 2004, pp. 23592366.
22
Pounds, P., Mahony, R., Gresham, J., Corke, P., and Roberts, J., Towards Dynamically-Favourable Quad-Rotor Aerial
Robots, In Proceedings of the Australasian Conference on Robotics and Automation, Canberra, Australia, 2004.
23
Franklin, G. F., Powell, J. D., and Emami-Naeini, A., Feedback Control of Dynamic Systems, Prentice Hall, Upper Saddle
River, NJ, 4th ed., 2002.
24
Johnson, W., Helicopter Theory, Princeton University Press, Princeton, NJ, 1980.
25
Hamel, T., Mahony, R., Lozano, R., and Ostrowski, J., Dynamic Modelling and Conguration Stabilization for an
X4-Flyer, Proceedings of the IFAC World Congress, Barcelona, Spain, August 2002.
17 of 17
American Institute of Aeronautics and Astronautics