Beruflich Dokumente
Kultur Dokumente
i
Acknowledgements
I would like to thank those who helped during my thesis work. Without their
support, I could have never accomplished this work.
I take this special occasion to thank my parents. I dedicate this work to my
parents. It would have been simply impossible to start, continue and complete
without the support of my parents who, unconditionally provided the resources
to me.
I am eternally indebted to my supervisor Ivan Kalaykov for all the help,
invaluable guidance and generous support throughout my thesis work. I have
been very fortunate to be associated with such a kind and good person and it
would take more than a few words to express my sincere gratitude.
I also like to thank Bo-lennart and Boyko Iliev for their enlightening sugges-
tions and advices. Their professionalism, guidance, dedication and inspirations
will always serve me as an example in my professional life.
iii
Contents
1 Introduction 1
1.1 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Concept of Quadrocopter . . . . . . . . . . . . . . . . . . . . . 2
1.3 Brief History . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Modern Quadrocopters . . . . . . . . . . . . . . . . . . . . . . 3
1.5 Advantages and disadvantages . . . . . . . . . . . . . . . . . . . 4
1.6 Outline of this document . . . . . . . . . . . . . . . . . . . . . . 5
2 Hardware Description 7
2.1 The Quadrocopter at Orebro University . . . . . . . . . . . . . 7
2.2 Onboard System . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2.1 Micro-controller . . . . . . . . . . . . . . . . . . . . . . 8
2.2.2 YGE motor controllers . . . . . . . . . . . . . . . . . . . 8
2.2.3 Motors . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.4 Propellers . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.5 Sensors on-board . . . . . . . . . . . . . . . . . . . . . . 10
3 Background 15
3.1 Different control strategies used for quadrocopters . . . . . . . . 15
3.2 Major Research/Educational Projects . . . . . . . . . . . . . . . 15
3.2.1 European Aeronautic Defense and Space Company . . . 15
3.2.2 Pennsylvania State University . . . . . . . . . . . . . . . 16
3.2.3 Middle East Technical University . . . . . . . . . . . . . 17
3.2.4 Australian National University . . . . . . . . . . . . . . 17
3.2.5 University of British Columbia Vancouver, BC, Canada . 18
3.2.6 Cornell University . . . . . . . . . . . . . . . . . . . . . 19
3.2.7 Swiss Federal Institute of Technology . . . . . . . . . . . 19
3.2.8 University of Technology in Compiegne, France . . . . . 21
3.2.9 Stanford University . . . . . . . . . . . . . . . . . . . . . 22
v
vi CONTENTS
References 55
List of Figures
vii
viii LIST OF FIGURES
ix
List of Algorithms
xi
Chapter 1
Introduction
An unmanned aerial vehicle (UAV) is an unpiloted aircraft which can either fly
autonomously or it can be remotely controlled based on program downloaded
into on-board computers. Unmanned aerial vehicles have very vast area of ap-
plications in military for missions that are too dull, dirty, or risky for human
piloted aircraft. They are also used in a growing number of civil applications
such as aireal inspection.
UAVs can be divided basically into two categories, fixed wing UAVs and ro-
tatory wing UAVs. Rotatory wing (or helicopter) UAVs have some advantages
over fixed wing UAVs that they can take off and land vertically, and that they
also can maintain their position at a fixed point in 3D space. Maintaining a po-
sition at a fixed point is called hovering. One very successful design for smaller
UAVs is an helicopter with four horizontal rotors with no tailrotor, such an
helicopter is called quadrocopter or quadrotor. Quadrocopters have some ad-
vantages over conventional single blade helicopter that they can be controlled
by changing the speed of the rotors and thus fixed-pitch blades are used that
simplifies the design and controlling of the quadrocopters. Secondly, the use of
four rotors allows each individual rotor to have a smaller diameter than the
equivally capable single blade conventional helicopter rotor, for a given vehicle
size, allowing them to attain less kinetic energy during flights. The most impor-
tant advantage is that they can make manuevers that conventional helicopters
can not make.
1.1 Objectives
The objective of this thesis is to design a fuzzy flight controller for a quadro-
copter that will control quadrocopter orientation in 3DOF namely; Pitch, Roll
and YAW. The controller is tested in Hardware In Loop (HIL) simulation. To
reach this objective:
Three gyros, one 3DOF accelerometer, a heading sensor and a GPS mod-
ule have to interfaced with the on-board micro-controller
1
2 CHAPTER 1. INTRODUCTION
A simulink model for the fuzzy controller has to be developed and down-
loaded into dSPACE computer
Figure 1.1: Quadrocopter motion, the arrow width is proportional to propeller rota-
tional speed
1.3. BRIEF HISTORY 3
is solely used to compensate for the gravity and not to counteract the torque,
because the four rotors neutralize the gyroscopic effect, so thrust is completely
used to produce the lift.
For micro quadrocopters, four rotors results in a very small rotor diameter,
which decreases the efficiency and increases energy requirment to get equal lift.
Also the dimensions and weight of a quadrocopter with equal payload is more
than a conventional helicopter. The simplifications in developing and control
which is offered by the design, makes it still a very favourable option for UAVs.
7
8 CHAPTER 2. HARDWARE DESCRIPTION
The YGE motor controllers can be programmed by moving the throttle and
waiting for acknowledgement beeps of the YGE. The featured programming
modes and options can be looked up in the YGE manual.
2.2.3 Motors
Since the motors are brushless, the phases of the AC supply have to be shifted
circularly in the three supply wires in order to rotate the motor. Since the motor
features not only three poles but 14, the motor rotation per 2 3 = 60 phase
The motors are equipped with neodymmagnets and feature high torques in
order to be used without gear box.
A motor carries out 840rpm per volt and is suitable for up to 3 LiPo cells re-
sulting in a total voltage of 12.6V and a maximum rotation speed of 10080rpm.
The maximum current drainage to the battery is 18A, where the range for best
efficiency is 8-14A. The maximum torque is approximately 0.2Nm at 18A.
2.2.4 Propellers
The propellers in use are EPP1045 propellers by Maxx Ltd. out of compos-
ite material with a length of 10 inches and a pitch of 4.5 inches per revolu-
tion. Mounted are two sets of counter rotating propellers. The shaft diameter
is 3mm.
10 CHAPTER 2. HARDWARE DESCRIPTION
Heading Sensor
A compass sensor allows absolute flight direction measurement and a perfectly
long term stability of the yaw axis. For sensing exact heading position Hon-
eywell HMC6352 compass module is used. The Honeywell HMC6352 is a
fully integrated compass module that combines 2-axis magneto-resistive sen-
sors with the neccesary analog and digital support circuits, and algorithms for
heading computation. By combining the sensor elements, processing electron-
ics, and firmware in to a 6.5mm by 6.5mm by 1.5mm LCC package, it offers a
complete, ready to use electronic compass.
Figure 2.4: Timing diagrram of reading two bytes from the HMC6352 (slave)
Gyro Sensor
A stable hovering control requires 3 gyro sensors: Pitch, Roll, and Yaw. ADXRS300
gyro sensors by Analog Devices are found to be most suitable for UAV appli-
cation, but not usable without special soldering equipment. Therefore already
mounted sensors to break-out-PCBs are used.
Acceleration Sensor
For even more hovering stability, a 3-DOF- acceleration sensor is required, in
addition to the gyro sensors. The LIS3LV02DQ is a three axes digital output
linear accelerometer that includes a sensing element and an IC interface able
to take the information from the sensing element and to provide the measured
acceleration signals to the external world through an I2 C/SPI serial interface.
12 CHAPTER 2. HARDWARE DESCRIPTION
GPS Sensor
To fly autonomously a navigation technique is required. Global Positioning
System (GPS) is found to be most simplest and appropriate for our application.
LEA-5H-008 in-door GPS module is selected for the project. An external usb to
UART converter is made to connect GPS module to on-board micro-controller.
2.2. ONBOARD SYSTEM 13
15
16 CHAPTER 3. BACKGROUND
kilogram. It can fly with in the range of one kilometer. Its IMU (Inertial Mea-
surment Unit) comprised of six inertial sensors for its six degrees of freedom. In
addition to these six sensors one GPS unit and air data sensors (gas sensors) is
also interfaced. Total IMU weighing 65 grams, consumes about three watts at
5 V. The system can be stored in a small space, as the motors can be detached
easily [20].
Figure 3.1: Quadrocopter developed by European Aeronautic Defense and Space Com-
pany
grams. Crossbow DMU-6 is also used in the prototype. It weighs about 500
grams
The pilot augmentation control system is used. A double lead compensator
is used for the inner loop. The final setup is shown in Figure 3.5.
was based on Lyapunov analysis using a nested saturation algorithm. Figure 3.9
shows the experimental setup for this research.
STARMAC I
STARMAC I has a total of 1 kilogram of thrust and can hover for not more
than ten minutes at full throttle. The onboard electronics were replaced with
PCB to have complete control over motors, power supply and sensor readings.
An upgrade to lipo batteries has increased both flight payload capacity and
hovering time and has considerably enhanced the capacity of the system. For
inertial measurement the Microstrain 3DM-G motion sensor was used, with
3-axis gyro, accelerometer and heading sensor information. For velocity and
position measurement, a Trimble Lassen Low Power GPS receiver was used.
To improve altitude information the downward-pointing SOnic Detection And
Ranging (SODAR) the Devantech SRFO8 was used, specially for difficult tasks
such as take off and landing.All of the onboard sensing is controlled through
two Microchip 40 MHz micro-controllers. Position estimation uses both GPS
position and velocity measurements, as well as orientation information in a
Extended Kalman Filter (EKF) to update the position and velocity estimates at
10 Hz. The GPS data is used to compensate integration bias of the small angle
approximated accelerations derived from the information from gyro sensors.
Attitude of the vehicle is controlled on board at 50 Hz and communicated
to a base sta- tion on the ground, communicating via a Bluetooth Class II device
with a range of over 150 feets. Designed to replace serial cable and therefore op-
erates at a bandwidth of 115200bps. This bandwidth of 115.2 kbps is shared
by all (of four) flyers. The base station on the ground performs Differential
GPS (DGPS) and waypoint tracking tasks for all four flyers, and sends appro-
24 CHAPTER 3. BACKGROUND
priate attitude values to the flyers for stable flight. Human piloted flight and
pre planned control is performed via the ground station laptop using Labview.
To control height a sliding mode control is designed and the attitude loop
was designed LQR techniques. Angular deviations were penalised more than
rate deviations. Also, the pitch and roll loops were penalised identically, with
a reduced penalty on the yaw deviations. On the STARMAC I platform also
more advanced controller types are tested, such as integral sliding mode and
reinforcement learning [21].
STARMAC II
The STARMAC II testbed is a sequel of first version with improvements at
several points:
3.2. MAJOR RESEARCH/EDUCATIONAL PROJECTS 25
Thrust capabilities: Old motors and propellers were replaced with Brush-
less motors and more rigid plastic propellers to double the output and
enhance the total thrust up to a total of 4 kilograms.
27
28 CHAPTER 4. CONTROLLER DESIGN AND SIMULINK MODEL
ulation. The article [3] reports the use of three inputs namely; acceleration,
velocities and accumulation of past commands around two horizontal axes X
and Y to maintain quadrocopter in hovering position. To design the fuzzy con-
troller we adopted a different approach than the one proposed in article [3].
We took the quadrocopter angles and the rates of angle change as input to our
fuzzy controllers. And we did not use any third input as used in article [3]. And
unlike article [3] we tested our controller on quadrocopter in HIL simulation
setup.
In design process of fuzzy controller, designers usually try to inject the exper-
tise of experienced human operator into their controller so that, their controller
control the plant in a same way as experienced human operator do. To design
the fuzzy controller for quadrocopter we found a set of guidelines presented
in article [3] that describe how human operators control the quadrocopter.
These guidelines were formulated after interviewing experienced quadrocopter
operators. These guidelines were written for how human operators perceive
the velocities and acceleration along different axes, and how they control the
quadrocopter based on their perception. Despite we are not using velocities
and accelerations as inputs, but we can use the same set of guidelines for our
controller with angles and the rates of angle of change as inputs as the veloci-
ties and accelerations are directly propotional to the angles and rates of angle
change respectively. So we can mold the presented guidlines for our controller
with angles and rates of angle change as inputs.
The molded guidelines are given below:
2. Only qualitative information of the bank angles and the rate of change of
bank angles are available to the external pilot.
3. Except in high aerobatic maneuvers roll, pitch and yaw control of the
vehicle are handled separately and independently.
4. When the observed angles of the vehicle are small (i.e. near hovering
point) more attention is paid to the angles of the quadrocopter than the
rate of change.
5. When the observed angles of the quadrocopter are large more attention
is paid to the vehicle angle change rate and more effort goes in to com-
pensating the rate than that to the angles.
6. Control commands issued by the human operator could be any of the
four quantitative commands namely roll, pitch, thrust and yaw.
4.2. QUADROCOPTER FUZZY CONTROLLER DESIGN 29
7. Observable angles and the rates of angle change are understood by the
human operator as either small, medium or big. Accumulation of control
commands is perceived as zero, small, medium and big.
We started designing the fuzzy controller for the quadrocopter in the light of
above stated guidelines. Guideline 1 defines the inputs of the controller, namely
the angles and the rate at which angles change. Guideline 7 suggest five mem-
bership functions for each input. We used triangular or trapezoidal shapes for
membership functions to avoid computational and programming complexities,
as there is a plan to translate the designed controller to C language in future.
We started with uniform distribution of membership functions as shown in
Fig. 4.1. Then in search of better performance we increased the input gains by
contracting the membership function closer to zero as shown in Fig. 4.2. But for
this setting we got oscillatory response, plus too high input gains also amplifies
the noise and can cause serious disturbances specially when the input values
are small. So, to rectify oscillations and avoid noise amplification we started re-
ducing input gains by stretching membership functions gradually. After testing
the system with several membership functions settings, the input membership
functions setting presented in Fig. 4.3 were found best. The same procedure
was repeated for the second input and we got best results at the same settings
used for the first input.
The same design process was repeated for output membership function.
Guideline 7 suggest seven membership functions for output of each controller.
We started with uniform distribution of all seven membership functions. Then
we increased output gains gradually to minmize the steady state error. But on
high output gains we got oscillatory response. So we had to find the settings
where the steady state error is minimized without oscillatory behavior. Fig-
ure 4.4 shows the settings for output membership functions where we found
such response.
30 CHAPTER 4. CONTROLLER DESIGN AND SIMULINK MODEL
Figure 4.2: input variable roll angle with high input gains
The rule base of the controller was also designed in the light of the above
guidelines. Guidelines 5, 6 and 7 suggest a subset of the rule base for the fuzzy
logic controller. We used 25 rules in our rule base that are presented in Fig. 4.5
Figure 4.6 shows the control surface produced by the rules presented in Fig
4.5.
The implemented fuzzy controller is an equivalent of position PD controller
as clear from Fig. 4.5. But it can be seen how easily it is designed, even without
mathematical model of the quadrocopter. In addition to this it has a big advan-
tage over conventional PD controller that it is non-linear. The same controller
can be regarded as equivalent to velocity PI controller.
Pitch and yaw axes controllers are designed in a same way as roll controller.
4.3 Implementation
The fuzzy logic controller and sensory data processing are modeled in Matlab
using Simulink. The Simulink model are then downloaded to dSPACE computer
before experimentation. We prefered to use Mamdani type fuzzy controller as it
is the simplest and more intuitive [19] specially when controller design is based
on a set of guidelines describing the human behavior. Guideline 3 suggests that
4.4. COMPLETE DESCRIPTION OF SIMULINK MODEL FOR
EXPERIMENTATION 31
the controller for each axes can be designed separately. So, we designed three
similar fuzzy controllers for the pitch, yaw and roll axis of the quadrocopter.
1. Data acquisition
2. De-multiplexing
4. Controllers
4.4.2 De-multiplexing
A very commonly used block "Demux" is used for de-multiplexing of data. The
block recieves 12 bytes serially, and it sends them forward in parallel fashion.
4.4.4 Controllers
Pitch and Roll controllers
In step 4, two fuzzy controllers are implemented in two subsystem blocks
namely; "Roll controller" and "Pitch controller" to control roll and pitch an-
gles, respectively. The internal working of these two blocks are exactly same,
shown in Fig. 4.11. The processed output of gyros are fed to these blocks that
is rate of angle change which is directly fed to fuzzy controller as input. The
second input is the angle that is determined by passing the rate of angle change
through integral block.
controllers are fed to motors 1 and 3 with a sign change. And the output of
Yaw controller is fed to all motors, motor 1 and 3 are fed with opposite sign as
of motor 2 and 4.
The installation of the experiment follows the scheme depicted in figure 5.1.
The wired connections between the dSPACE RTI and the QAP chassis have
been established and function correctly.
39
40 CHAPTER 5. EXPERIMENTATION AND RESULTS
the objectives of the thesis. The YGE controller for motor 1 has burned out, so
motor 1 cannot be operated. As motor 1 cannot be operated, we have to stop
motor 3. With remaining two motors it is certainly not possible to perform
experiments for any of 6 axes except roll axes. To perform experiment for roll
axis, we built an experiment setup with the intension to limit the DOF to only
one axis that is the roll axis. The setup we built for experimentation is shown
in Fig. 5.1 in which it can be seen the quadrocopter is tied with ropes from top
and from down. Only the roll axes is left free.
5.2 Results
As stated in the last section that due to the technical problems the experimenta-
tion plan was effected badly. With only two motors we could test our controller
for roll axis. We performed several tests to validate the functionality and ap-
plicability of our fuzzy controller. We are including some selected experiments
among the several we performed during thesis work.
The experimentation setup we built allows the quadrocopter to be stabi-
lized without motors in few seconds, if it is disturbed along roll axis. Here we
can give the example of pendulum. If pendulum is disturbed from its point of
stability, it stabilized back to this point after some oscillations. But there is a
difference between our system and pendulum. The point of stability for pendu-
lum is fixed, where pendulum attains minimum potential energy. But this is not
the case with our system. When we disturb the quadrocopter along roll axis,
the eight threads used to limit the movements around other axes change their
adjusments that results in a slight change in the angle of stability for the tied
quadrocopter.
We performed the first experiment to analyze, how much time it takes the
quadrocopter to get stabilized without motors. The results of this experiment
is shown in Fig. 5.2.
In Fig. 5.2, it is shown that the quadrocopter is destabilized by an external
force at around 11th second of the experiment. The quadrocopter became sta-
bilized after around 8 seconds. Here we can see the difference between the angle
of stability before the occurence of external disturbence and after it. This slight
shift in the angle of stability for the quadrocopter is due to the re-adjustment
of the threads tied to quadrocopter, as stated earlier.
After performing experiment without motors we performed an experiment
with motors to analyze how the newly designed controller works to stabilize
the quadrocopter in our constrained experimentation setup. The results of this
experiment are shown in Fig. 5.3.
In Fig. 5.3 it is shown that when the experiment was started the roll angle
of the quadrocopter was around -4 degrees, means that the quadrocopter was
tilted -4 degrees towards motor 2. At around 4th second of the experiment the
motors were started with the power of 35 percent. The controller is started at
around 7th second. The activity of the motors can be seen in the second graph
5.2. RESULTS 41
of Fig. 5.3, where the red line represents the power plus controller signal to
motor 2 and blue line represents the power plus controller signal to motor 4.
At around 13th second we disturbed the quadrocopter along roll axis to 32
degrees.After that controller stabilized the quadrocopter in less than 4 seconds
that is the half of the time it took in last experiment.
The red line in the first graph of Fig. 5.3 represents the set point for the
roll angle. The controller should stabilize the quadrocopter at the set point.
But it can be seen in Fig. 5.3 that roll angle never reached to the set point.
On the other hand, we see in the second graph of Fig. 5.3 that the controller
continously tried to overcome the error by increasing the speed of motor 2 and
by decreasing the speed of motor 4, but still the roll angle remained unchanged.
This is due to the limitations associated with our experimentation setup. With
two motors we cannot perform experiments with high motor speeds, because
the torque produced by two motors destabilize the whole setup. And at lower
motor speeds we need high controller gains to stabilize the quadrocopter at
42 CHAPTER 5. EXPERIMENTATION AND RESULTS
set point. And when we use higher controller gains, it results in uncontrolled
oscillations.
Figure 5.3: Disturbence test with 35 percent power and controller gain 1
Figure 5.4: Disturbence test with 30 percent power and multiple controller gain
Figure 5.5: Disturbence test with 60 percent power and controller gain 1
used threads to limit the movements around other axes other than the roll axis,
but this is not the real solution. These threads also interfare with the movement
along roll axis.
After performing several experiments with different power settings and con-
troller gains, we found that operating motors at 35 percent power with con-
troller gain of 1 results in best performance for our tightly constrained exper-
imental setup. To find these settings, we performed several experiments. We
started from low power settings to high, with different values for gain. The
Fig. 5.6 show the results for 15 percent power and controller gain equals to 1.
As this was one the earliest experiment, we could not classify it as good or bad.
Then we tried higher gains at lower power settings. Fig. 5.7 is an example
of it. It represents the results for the experiment performed at 15 percent power
and gain value equals to 2.
Then we performed the experiment with 15 percent power with 2.5 con-
troller gain. The results for this experiment are shown in Fig. 5.8. By compar-
ing the results of the experiments performed at 15 percent power with different
5.2. RESULTS 45
Figure 5.6: Disturbence test with 15 percent power and controller gain 1
controller gain values, we found that higher gains at lower power settings cause
higher and more oscillations.
Then we increased power to 25 percent and test our controller with gain
equals to 1. The results for this experiment is shown in Fig. 5.9.
Then we performed experiment with same power settings but with different
controller gain values. Experiment shown in Fig. 5.10 was performed with 25
percent power with controller gain value equals to 2.
Then we performed another experiment with same power setting of 25 per-
cent with controller gain value equals to 2.5. The results for this experiment
are shown in Fig. 5.11. We noticed the same trend with 25 percent power as
we noticed for 15 percent. But we found that the over all performance of the
system with 25 percent is better than performance with 15 percent power.
In the search of better power and gain settings for our system, we increased
the power to 35 percent. And we found that the power of 35 percent with
controller gain values equals to 1 result in best performance for our system.
The results of experiment with these power and gain settings are shown in
Fig. 5.3.
46 CHAPTER 5. EXPERIMENTATION AND RESULTS
Figure 5.7: Disturbence test with 15 percent power and controller gain 2
Figure 5.8: Disturbence test with 15 percent power and controller gain 2.5
Figure 5.9: Disturbence test with 25 percent power and controller gain 1
5.3. LIMITATIONS IN EXPERIMENTATION 49
Figure 5.10: Disturbence test with 25 percent power and controller gain 2
50 CHAPTER 5. EXPERIMENTATION AND RESULTS
Figure 5.11: Disturbence test with 25 percent power and controller gain 2.5
5.3. LIMITATIONS IN EXPERIMENTATION 51
Figure 5.12: Disturbence test with 35 percent power and controller gain 2
52 CHAPTER 5. EXPERIMENTATION AND RESULTS
Figure 5.13: Disturbence test with 35 percent power and controller gain 2.5
Chapter 6
Conclusions and future work
6.1 Conclusions
The objective of this thesis was to design the fuzzy flight controller for the
quadrocopter. Initially we were designing the fuzzy controller for all six de-
grees of freedom, but later the objective of the thesis was reduced to 3 degrees
of freedom. After interfacing all the required sensors and establishing com-
munication between the dSPACE computer and on-board micro-controller, the
designed controller was tested in hardware in loop setup. The experiments to
validate the functionality of the designed controller were performed in highly
constrained experimentation setup because of major technical problems. De-
spite major technical problems we managed to perform experiments and we
got satisfactory results. By analyzing the results we conclude that it is possible
to control the quadrocopter with fuzzy controller. However, it is not possible
now to compare our results with other existing control approaches as the ex-
perimentation was done in highly constrained setup. There is no guarantee that
the designed controller is optimal. We strongly believe that the further tunning
will be required once the harware problems has been fixed.
53
54 CHAPTER 6. CONCLUSIONS AND FUTURE WORK
modification has also been done on the quadrocopter to interface GPS module.
These things are to be integrated in future.
[4] E.Altug. Vision based control of unmanned aerial vehicles with applica-
tions to an autonomous four rotor helicopter, Quadrotor. PhD thesis,
University of Pennsylvania, 2010. (Cited on pages 15, 16, and 27.)
55
56 REFERENCES
[11] I. Kroo and F. Prinz. The mesicopter: A meso-scale flight vehicle - niac
phase i final report. technical report. Technical report, Stanford Univer-
sity, 2001. (Cited on page 22.)
[12] I. Kroo and F. Prinz. The mesicopter: A meso-scale flight vehicle - niac
phase ii technical proposal. technical report. Technical report, Stanford
University, 2001. (Cited on page 22.)