ABSTRACT
Title of dissertation: EXPERIMENTAL INVESTIGATION OF
A MAV-SCALE CYCLOCOPTER
Elena Shrestha
Doctor of Philosophy, 2018
Dissertation directed by: Professor Inderjit Chopra
Department of Aerospace Engineering
The development of an efficient, maneuverable, and gust tolerant hovering concept
with a multi-modal locomotion capability is key to the success of micro air vehicles
(MAVs) operating in multiple mission scenarios. The current research investigated
performance of two unconventional cycloidal-rotor-based (cyclocopter) configura-
tions: (1) twin-cyclocopter and (2) all-terrain cyclocopter. The twin-cyclocopter
configuration used two cycloidal rotors (cyclorotors) and a smaller horizontal edge-
wise nose rotor to counteract the torque produced by the cyclorotors. The all-terrain
cyclocopter relied on four cyclorotors oriented in an H-configuration.
Objectives of this research include the following: (1) develop control strategies
to enable level forward flight of a cyclocopter purely relying on thrust vectoring, (2)
identify flight dynamics model in forward flight, (3) experimentally evaluate gust
tolerance strategies, and (4) determine feasibility and performance of multi-modal
locomotion of the cyclocopter configuration.
The forward flight control strategy for the twin-cyclocopter used a unique
combination of independent thrust vectoring and rotational speed control of the
cyclorotors. Unlike conventional rotary-winged vehicles, the cyclocopter propelled
in forward flight by thrust vectoring instead of pitching the entire fuselage. While
the strategy enabled the vehicle to maintain a level attitude in forward flight, it was
accompanied by significant yaw-roll controls coupling and gyroscopic coupling. To
understand these couplings and characterize the bare airframe dynamics, a 6-DOF
flight dynamics model of the cyclocopter was extracted using a time-domain system
identification technique. Decoupling methods involved simultaneously mixing roll
and yaw inputs in the controller. After implementing the controls mixing strategy in
the closed-loop feedback system, the cyclocopter successfully achieved level forward
flight up to 5 m/s.
Thrust vectoring capability also proved critical for gust mitigation. Thrust
vectoring input combined with flow feedback and position feedback improved gust
tolerance up to 4 m/s for a twin-cyclocopter mounted on a 6-DOF test stand. Flow
feedback relied on a dual-axis flowprobe attached to differential pressure sensors
and position feedback was based on data recorded by the VICON motion capture
system. The vehicle was also able to recover initial position for crosswind scenarios
tested at various side-slip angles up to 30◦.
Unlike existing multi-modal platforms, the all-terrain cyclocopter solely relied
on its four cyclorotors as main source of propulsion, as well as wheels. Aerial and
aquatic modes used aerodynamic forces generated by modulating cyclorotor rota-
tional speeds and thrust vectors while terrestrial mode used motor torque. In aerial
mode, cyclorotors operated at 1550 rpm and consumed 232 W to sustain hover. In
terrestrial mode, forward translation at 2 m/s required 28 W, which was an 88%
reduction in power consumption required to hover. In aquatic mode, cyclorotors
operated at 348 rpm to achieve 1.3 m/s translation and consumed 19 W, a 92% re-
duction in power consumption. With only a modest weight addition of 200 grams for
wheels and retractable landing gear, the versatile cyclocopter platform achieved sus-
tained hover, efficient translation and rotational maneuvers on ground, and aquatic
locomotion.
EXPERIMENTAL INVESTIGATION OF
A MAV-SCALE CYCLOCOPTER
by
Elena Shrestha
Dissertation submitted to the Faculty of the Graduate School of the
University of Maryland, College Park in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
2018
Advisory Committee:
Dr. Inderjit Chopra, Chair/Advisor
Dr. Anubhav Datta
Dr. James Baeder
Dr. David Akin
Dr. Nikhil Chopra
©c Copyright by
Elena Shrestha
2018

Dedication
To my family for giving me the opportunity to pursue my degree and supporting
me unconditionally throughout the entire journey.
ii
Acknowledgments
I would like to thank my advisor, Dr. Inderjit Chopra for taking the risk in hiring
an inexperienced Sophomore in 2010 for the challenging cyclocopter project. The
past 8 years under his supervision have molded me into a better researcher. I
really appreciate the open research environment and autonomy that he provided for
students in the Micro Air Vehicle (MAV) Lab.
I would also like to thank my close mentors: Dr. Derrick Yeo, Dr. Vikram
Hrishikeshavan, and Dr. Moble Benedict for providing me with guidance and sup-
port throughout all the ups and downs involved in research. Dr. Yeo not only piloted
the cyclocopters, he also helped guide me through the process of articulating my
research ideas and provided constant moral support. Dr. Hrishikeshavan and his
magical custom autopilot made it possible to scale-down and even explore different
cyclocopter configurations. Finally, I am extremely grateful to Dr. Benedict for
introducing me to the unforgiving (yet satisfying) world of vertical axis rotors and
instilling in me the value of a strong work ethic.
I am very thankful to my committee members: Dr. Anubhav Datta, Dr. James
Baeder, Dr. Dave Akin, and Dr. Nikhil Chopra for all their technical guidance and
feedback on my dissertation. My sincere thanks to Dr. Datta for his continued
encouragement and insightful conversations.
Throughout my graduate degree, I had the pleasure of working closely with
many exceptional undergraduate research assistants, and I am very grateful to all
of the students. I want to especially thank Brian Davis and Benjamin Williams for
iii
their contributions to the development of the all-terrain cyclocopter. I hope we get
an opportunity to work together again in the future.
Graduate studies would be incomplete without the friends and colleagues that
make the entire experience worthwhile. Special thanks to Steven Hill for his unwa-
vering support throughout all these years. I feel that it is necessary to dedicate an-
other sentence to Steven. I am also very thankful to Joseph Lemich, Jonathan Cham-
bers, James Lankford, Lex Chi, Lem Carpenter, Snehesh Shrestha, Lina Maria,
Vera Klimchenko, Wanyi Ng, Michael Van Order, Angela Maki, Peter Ryseck, Eric
Solomon, and all the students housed in the Manufacturing building.
Finally, I am extremely grateful to my family for their unconditional support
and patience all these years. My parents, Udaya and Romee, have sacrificed so much
to provide me with an opportunity to obtain my degree. I am forever indebted to
them. My siblings, Urusha and Alex, have kept me grounded and provided endless
emotional support.
It is impossible to acknowledge every single person, and I apologize to those
I have inadvertently left out. Thank you to everyone that I have shared special
memories with during my studies at the University of Maryland!
iv
Table of Contents
List of Tables x
List of Figures xi
List of Abbreviations xvi
1 Introduction 1
1.1 Micro Air Vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Technical Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.1 Low Reynolds Number Aerodynamics . . . . . . . . . . . . . . 3
1.2.2 Size, Weight,and Power (SWaP) Constraints . . . . . . . . . . 6
1.2.3 Flight Stability and Control . . . . . . . . . . . . . . . . . . . 8
1.3 Existing Micro Air Vehicle Configurations . . . . . . . . . . . . . . . 10
1.3.1 Fixed-wing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.2 Rotary-wing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.3 Flapping-wing . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.4 Cycloidal Rotor Concept . . . . . . . . . . . . . . . . . . . . . 14
1.4 Previous Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4.1 Existing Cyclocopter Research . . . . . . . . . . . . . . . . . . 15
1.4.2 Gust Tolerance Studies of Rotary-wing MAVs . . . . . . . . . 24
1.4.3 MAVs with Multi-mode Locomotion Capability . . . . . . . . 27
1.5 Research Objective and Technical Approach . . . . . . . . . . . . . . 28
1.5.1 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . 30
2 Vehicle Design 33
2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Twin-Cyclocopter Configuration . . . . . . . . . . . . . . . . . . . . . 33
2.3 Cyclorotor Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3.1 Blade Pitching Mechanism . . . . . . . . . . . . . . . . . . . . 39
2.4 Blade Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4.1 Aerodynamic Design . . . . . . . . . . . . . . . . . . . . . . . 43
2.4.2 Structural Design . . . . . . . . . . . . . . . . . . . . . . . . . 47
v
2.4.2.1 Carbon Fiber Blade . . . . . . . . . . . . . . . . . . 47
2.4.2.2 Exposed Foam Blade . . . . . . . . . . . . . . . . . . 48
2.4.2.3 Mylar Wrapped Monolithic Blade . . . . . . . . . . . 48
2.4.2.4 Multi-Component Rib and Spar Blade . . . . . . . . 50
2.5 Cyclorotor Performance . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.6 Nose Rotor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.7 Scalability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3 Avionics and Control System Development 63
3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.2 MAV-scale Autopilot . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.2.1 GINA Mote . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.2.2 ELKA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.3 Sensors for Flight Stability and Control . . . . . . . . . . . . . . . . . 68
3.3.1 Inertial Measurement . . . . . . . . . . . . . . . . . . . . . . . 68
3.3.2 Attitude Estimation . . . . . . . . . . . . . . . . . . . . . . . 71
3.3.2.1 Quaternion Representation . . . . . . . . . . . . . . 72
3.3.2.2 Gyroscope . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3.2.3 Accelerometer . . . . . . . . . . . . . . . . . . . . . . 75
3.4 Sensors for Flight Performance Diagnostics . . . . . . . . . . . . . . . 78
3.4.1 RPM Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.4.2 Proximity Sensors . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.4.3 Barometer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.4.4 Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.5 Sensors for Guidance and Navigation (Autonomy) . . . . . . . . . . . 81
3.5.1 Global Navigation Satellite Systems (GNSS) . . . . . . . . . . 81
3.5.2 Visual Odometry (VO) . . . . . . . . . . . . . . . . . . . . . . 85
3.5.2.1 Passive Cameras . . . . . . . . . . . . . . . . . . . . 89
3.5.2.2 Active Cameras . . . . . . . . . . . . . . . . . . . . . 90
3.5.2.3 Event-Based Cameras . . . . . . . . . . . . . . . . . 91
3.5.3 Simultaneous Localization and Mapping (SLAM) . . . . . . . 92
3.5.3.1 LIDAR . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.5.3.2 Visual SLAM (V-SLAM) . . . . . . . . . . . . . . . . 95
3.6 MAV-Scale Onboard Computer . . . . . . . . . . . . . . . . . . . . . 97
3.6.1 Qualcomm Snapdragon Flight . . . . . . . . . . . . . . . . . . 97
3.6.2 ODROID XU-4 . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.6.3 NVIDIA Jetson TX2 . . . . . . . . . . . . . . . . . . . . . . . 100
3.7 Telemetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.7.1 Wireless Communication (Wi-Fi) . . . . . . . . . . . . . . . . 100
3.7.2 Automatic Dependent Surveillance-Broadcast (ADS-B) . . . . 103
3.8 Closed-Loop Feedback System . . . . . . . . . . . . . . . . . . . . . . 104
3.8.1 Attitude Stabilization . . . . . . . . . . . . . . . . . . . . . . . 105
3.8.2 Ziegler-Nichols PID Tuning in Hover . . . . . . . . . . . . . . 108
3.8.3 Hover Flight Tests . . . . . . . . . . . . . . . . . . . . . . . . 110
vi
3.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4 Flight Control Strategy 113
4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.2 Rigid Body Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.2.1 Forces and Moments . . . . . . . . . . . . . . . . . . . . . . . 116
4.2.2 Gyroscopic Coupling . . . . . . . . . . . . . . . . . . . . . . . 117
4.3 Cyclorotor Coordinate System . . . . . . . . . . . . . . . . . . . . . . 119
4.3.1 Forward Flight Orientation . . . . . . . . . . . . . . . . . . . . 119
4.3.2 Virtual Camber and Virtual Incidence Effect . . . . . . . . . . 121
4.4 Attitude Control in Hover . . . . . . . . . . . . . . . . . . . . . . . . 122
4.5 Attitude Control in Forward Flight . . . . . . . . . . . . . . . . . . . 122
4.6 Wind Tunnel Studies . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
4.6.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.6.2 Effect of Free-Stream Velocity . . . . . . . . . . . . . . . . . . 126
4.6.3 Effect of Phasing . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.6.4 Control Inputs for Trimmed Forward Flight . . . . . . . . . . 128
4.7 Roll-Yaw Decoupling Strategy . . . . . . . . . . . . . . . . . . . . . . 129
4.7.1 Controller Design . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.8 Forward Flight Test Results . . . . . . . . . . . . . . . . . . . . . . . 132
4.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
5 System Identification in Forward Flight 139
5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.2.1 Model Structure . . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.2.2 Time-Domain Approach . . . . . . . . . . . . . . . . . . . . . 143
5.2.3 Stepwise Regression . . . . . . . . . . . . . . . . . . . . . . . . 144
5.2.4 Equation-Error Method . . . . . . . . . . . . . . . . . . . . . 146
5.3 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.3.1 Data Acquisition Setup . . . . . . . . . . . . . . . . . . . . . . 149
5.3.2 Sampling Rate . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.3.3 Bare-Airframe Dynamics . . . . . . . . . . . . . . . . . . . . . 151
5.3.4 Pilot Excitation . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.3.5 Vehicle Response . . . . . . . . . . . . . . . . . . . . . . . . . 154
5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.4.1 Model Structure Determination . . . . . . . . . . . . . . . . . 156
5.4.2 Longitudinal Mode . . . . . . . . . . . . . . . . . . . . . . . . 158
5.4.3 Coupled Lateral-Yaw Modes . . . . . . . . . . . . . . . . . . . 159
5.4.3.1 Gyroscopic Coupling . . . . . . . . . . . . . . . . . . 160
5.4.3.2 Controls Coupling . . . . . . . . . . . . . . . . . . . 162
5.4.4 Stability analysis in forward flight . . . . . . . . . . . . . . . . 163
5.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
vii
6 Gust Rejection Study 167
6.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.2 Gust Rejection Strategies . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.2.1 Control Input . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.2.2 Feedback System . . . . . . . . . . . . . . . . . . . . . . . . . 169
6.3 Real-time Flow Measurement . . . . . . . . . . . . . . . . . . . . . . 170
6.4 Preliminary Wind Tunnel Studies . . . . . . . . . . . . . . . . . . . . 175
6.4.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
6.4.2 Results: Longitudinal Axis . . . . . . . . . . . . . . . . . . . . 176
6.4.3 Results: Lateral Axis . . . . . . . . . . . . . . . . . . . . . . . 182
6.5 Tethered Experiments (6-DOF) . . . . . . . . . . . . . . . . . . . . . 183
6.5.1 Experimental Stand . . . . . . . . . . . . . . . . . . . . . . . . 183
6.5.2 Synthetic Gust Generation Device . . . . . . . . . . . . . . . . 184
6.6 Gust Rejection Controller . . . . . . . . . . . . . . . . . . . . . . . . 190
6.7 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
6.7.1 Pitch Control vs. Thrust Vectoring . . . . . . . . . . . . . . . 191
6.7.2 Position Feedback vs. Flow Feedback . . . . . . . . . . . . . . 195
6.7.3 Crosswind Scenario . . . . . . . . . . . . . . . . . . . . . . . . 199
6.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
7 Multi-Mode Capability 204
7.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
7.2 Vehicle Integration and Design . . . . . . . . . . . . . . . . . . . . . . 205
7.2.1 Cyclorotor Design . . . . . . . . . . . . . . . . . . . . . . . . . 206
7.2.2 Cyclowheel Design . . . . . . . . . . . . . . . . . . . . . . . . 210
7.2.3 Retractable Landing Gear . . . . . . . . . . . . . . . . . . . . 211
7.3 Closed-loop Feedback System . . . . . . . . . . . . . . . . . . . . . . 213
7.4 Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
7.4.1 Aerial Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
7.4.2 Terrestrial Mode . . . . . . . . . . . . . . . . . . . . . . . . . 217
7.4.3 Aquatic Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
7.5 Multi-Mode Testing Results . . . . . . . . . . . . . . . . . . . . . . . 219
7.5.1 Aerial Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
7.5.2 Terrestrial Mode . . . . . . . . . . . . . . . . . . . . . . . . . 224
7.5.3 Aquatic Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
7.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
8 Conclusion and Future Work 229
8.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
8.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
8.2.1 Vehicle Development . . . . . . . . . . . . . . . . . . . . . . . 232
8.2.2 Flight Control Strategy . . . . . . . . . . . . . . . . . . . . . . 233
8.2.3 System Identification in Forward Flight . . . . . . . . . . . . . 235
8.2.4 Gust Rejection Study . . . . . . . . . . . . . . . . . . . . . . . 237
8.2.5 Multi-Mode Capability . . . . . . . . . . . . . . . . . . . . . . 238
viii
8.3 Recommendations for Future Work . . . . . . . . . . . . . . . . . . . 240
Bibliography 243
ix
List of Tables
2.1 Weight distribution comparison of previous generation twin-cyclocopters
(60–550g). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.2 Cyclorotor design parameters for the 550 g cyclocopter. . . . . . . . . 42
2.3 Blade properties (60g-scale cyclocopter). . . . . . . . . . . . . . . . . 58
3.1 Comparison between GINA Mote, ELKA, and ELKA-R [78]. . . . . . 66
3.2 Summary of popular MAV-scale commercial GPS modules. . . . . . . 83
3.3 Comparison between types of cameras for VO. . . . . . . . . . . . . . 88
3.4 Existing LIDAR sensors for UAVs. . . . . . . . . . . . . . . . . . . . 94
3.5 Comparison of available visual SLAM framework [67,68]. . . . . . . . 96
3.6 Comparison of available light-weight onboard computer [69]. . . . . . 98
3.7 Ziegler-Nichols Frequency Response Method. . . . . . . . . . . . . . . 108
3.8 Gains used in feedback system. . . . . . . . . . . . . . . . . . . . . . 110
5.1 Comparison between frequency-domain and time-domain methods [84].142
5.2 Parameter estimation results. . . . . . . . . . . . . . . . . . . . . . . 155
5.3 Eigenvalue Comparison . . . . . . . . . . . . . . . . . . . . . . . . . . 163
7.1 Key cyclorotor design parameters. . . . . . . . . . . . . . . . . . . . . 208
7.2 Comparison of cyclowheel designs. . . . . . . . . . . . . . . . . . . . . 209
7.3 Weight distribution of the all-terrain cyclocopter. . . . . . . . . . . . 212
7.4 Performance of the all-terrain cyclocopter. . . . . . . . . . . . . . . . 219
x
List of Figures
1.1 Commercial applications for MAVs. . . . . . . . . . . . . . . . . 2
1.2 Key design considerations for the next generation MAV. . . . 3
1.3 Effect of Reynolds number on maximum sectional lift-drag
ratio [2–4]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Components required for MAVs. . . . . . . . . . . . . . . . . . . 6
1.5 Sub-system weights for commercial quadcopters. . . . . . . . . 7
1.6 CFD cases of wind stream-lines at 1.5 m above ground [11]. . 9
1.7 Endurance vs. vehicle mass of current MAVs. . . . . . . . . . 10
1.8 Cyclorotor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.9 Four-bar linkage pitching mechanism. . . . . . . . . . . . . . . 13
1.10 Rendering of an early Samoliot cyclocopter concept by E.P.
Sverchkov. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.11 Kirsten’s cycloplane featured in Popular Science, model in
the wind tunnel, and Kirsten-Boeing experimental test rig
[18,19]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.12 Voith-Schneider propeller used by tugboats [20, 21]. . . . . . . 17
1.13 Strandgren’s cyclogyros [22]. . . . . . . . . . . . . . . . . . . . . 17
1.14 Wheatley’s experimental setup [23]. . . . . . . . . . . . . . . . . 18
1.15 Cyclocopters developed by the Seoul National University
[24–28].) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.16 Early twin-cyclocopter work conducted at the University of
Maryland [29]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.17 Hover-capable cyclocopter MAVs developed at the Univer-
sity of Maryland. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.18 Power loading (thrust/power) vs. disk loading (conventional
micro-rotors vs. optimized cyclorotor) . . . . . . . . . . . . . . 21
1.19 Power versus forward speed for constant rotational speed of
1740 rpm (13.88 m/s) for level, steady flight [36,37]. . . . . . 22
1.20 System identification conducted on a twin-cyclocopter by
the University of Maryland [39]. . . . . . . . . . . . . . . . . . . 23
xi
1.21 Synthetic gust generation device and experimental setup as-
sembled by Georgia Institute of Technology [42]. . . . . . . . . 25
1.22 Experimental setup for gust study of a ducted-fan MAV con-
ducted by the University of Maryland [43]. . . . . . . . . . . . 26
1.23 Existing multi-modal micro air vehicles [44–46]. . . . . . . . . 27
2.1 Twin-cyclocopters developed at University of Maryland. . . . 34
2.2 550 g twin-cyclocopter. . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3 Weight distribution of twin-cyclocopters are different scales. 36
2.4 Cyclorotor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.5 Four-bar linkage pitching mechanism. . . . . . . . . . . . . . . 38
2.6 Pitching mechanism implemented on the 550g twin-cyclocopter.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.7 Cyclorotor coordinate system. . . . . . . . . . . . . . . . . . . . 40
2.8 Blade pitch angle as a function of rotor azimuth for each
cyclorotor at trim. . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.9 Virtual camber and incidence in a curvilinear flow . . . . . . . 44
2.10 Linear approximation to measure virtual camber and virtual
incidence [31,47]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.11 Previous iterations of blade design. . . . . . . . . . . . . . . . . 46
2.12 Rib and spar blade design. . . . . . . . . . . . . . . . . . . . . . . 49
2.13 Blade mold for constructing new blades. . . . . . . . . . . . . . 49
2.14 Nano-17 (force/torque transducer) experimental setup. . . . 51
2.15 Cyclorotor thrust vs. rotational speed . . . . . . . . . . . . . . 51
2.16 Cyclorotor electrical power vs. thrust. . . . . . . . . . . . . . . 52
2.17 Cyclorotor power loading vs. rotational speed. . . . . . . . . . 52
2.18 Commercial propellers tested for the nose rotor. . . . . . . . . 54
2.19 Setup to measure thrust produced by nose rotor. . . . . . . . 54
2.20 Nose rotor CT vs. CP . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.21 Nose rotor thrust vs. rotational speed for 9050-3 propeller. . 56
2.22 Nose rotor electrical power vs. thrust for 9050-3 propeller. . 56
2.23 550g and 60g twin-cyclocopters. . . . . . . . . . . . . . . . . . . 57
2.24 Thrust vs. rpm for the 60g cyclorotor blades. . . . . . . . . . . 59
2.25 Electrical power loading of old design vs. new blade design. . 60
3.1 Components of an autopilot. . . . . . . . . . . . . . . . . . . . . 64
3.2 Comparison between MAV-scale autopilot used on the cy-
clocopter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.3 Notation for body rates and attitude of the cyclocopter. . . . 68
3.4 Measurement from magnetometer showing large bias errors
[50]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.5 Sensor suite for flight performance diagnostics and atmo-
spheric measurement. . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.6 GPS waypoint following using the Pixhawk GPS module. . . 84
xii
3.7 Variation in number of reachable satellites in an urban en-
vironment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
3.8 Landmarks tracked by ROVIO to construct estimated cam-
era poses in 3-D [57]. . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.9 Camera concepts used for visual odometry. . . . . . . . . . . . 87
3.10 Comparison of data transmission rate between frame-based
and event-based cameras [60]. . . . . . . . . . . . . . . . . . . . . 90
3.11 Captured events and projected gradient map [63]. . . . . . . . 91
3.12 Telemetry implemented for the cyclocopter. . . . . . . . . . . . 101
3.13 PWM signals used by actuators (f = frequency). . . . . . . . . 102
3.14 MAV-scale ADS-B sensors developed by uAvionix. . . . . . . 103
3.15 Schematic of closed-loop feedback system with the incorpo-
rated roll-yaw decoupling strategy. . . . . . . . . . . . . . . . . . 104
3.16 Comparison of attitude tracking controller vs. fixed attitude
controller (previous implementation). . . . . . . . . . . . . . . . 107
3.17 Twin-cyclocopter in steady hover (position data on right). . 110
3.18 Attitude data from ELKA during hover. . . . . . . . . . . . . . 111
4.1 Notation for forces and moments on the cyclocopter. . . . . . 114
4.2 Gyroscopic coupling due to net angular momentum. . . . . . 114
4.3 Cyclorotor coordinates showing measurement of absolute
phase angle (γabs), lift force, and propulsive force in forward
flight. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.4 Schematic showing hover condition with γ ◦abs > 0 and Φ=0
◦. 118
4.5 Cyclorotor coordinates in forward flight where γabs > 90
◦ and
Φ=90◦. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.6 Control strategy for twin-cyclocopter. . . . . . . . . . . . . . . 120
4.7 Lift and propulsive thrust for cyclocopter in level forward
flight. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.8 Wind tunnel testing setup. . . . . . . . . . . . . . . . . . . . . . 124
4.9 Variation of forces with velocity and phasing at 1400 RPM. . 126
4.10 Oscillations in pitch, roll, and yaw dynamics caused by cross-coupling. 127
4.11 Control inputs required to obtain trimmed free flight (lift =
weight, thrust = drag). . . . . . . . . . . . . . . . . . . . . . . . 128
4.12 Controls mixing for forward flight. . . . . . . . . . . . . . . . . . 130
4.13 Closed-loop feedback system with feedforward mixing of con-
trol inputs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.14 First successful forward flight of a twin-cyclocopter (V = 3
m/s). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.15 Airspeed data taken shows forward velocity of 3.5 m/s. . . . 132
4.16 Phase angle (Φ = 35◦) and rotational speed (1400 RPM)
during forward flight at 3.5 m/s. . . . . . . . . . . . . . . . . . 133
4.17 Control inputs at 5 m/s. . . . . . . . . . . . . . . . . . . . . . . . 133
4.18 Attitude data at 5 m/s. . . . . . . . . . . . . . . . . . . . . . . . . 134
4.19 Attitude data at 3.5 m/s. . . . . . . . . . . . . . . . . . . . . . . 135
xiii
5.1 System Identification using time-domain approach. . . . . . . 140
5.2 Time history data for pilot excitation and vehicle response. . 141
5.3 Vicon motion capture testing setup. . . . . . . . . . . . . . . . . 148
5.4 Synchronization between ELKA and Vicon measurements
using a fiducial perturbation in pitch. . . . . . . . . . . . . . . . 149
5.5 Data flowchart between ELKA and Vicon. . . . . . . . . . . . . 150
5.6 Time history data for pilot excitation and vehicle response. . 153
5.7 Model structure for cyclocopter in forward flight. . . . . . . . 156
5.8 Estimated model output compared to flight data. . . . . . . . 157
5.9 Gyroscopic cross-coupling between roll and yaw degrees of
freedom. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5.10 Controls cross-coupling between roll and yaw degrees of free-
dom occurring at high phase angles. . . . . . . . . . . . . . . . . 162
5.11 Pole location for the cyclocopter in forward flight. . . . . . . . 164
6.1 Pitch control vs. thrust vectoring control. . . . . . . . . . . . . 168
6.2 Multi-hole flowprobe system. . . . . . . . . . . . . . . . . . . . . 170
6.3 Flowprobe calibration compared to hot-wire and rotary anemome-
ter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
6.4 Power spectral density estimate via Welch (contributions
from the flow field). . . . . . . . . . . . . . . . . . . . . . . . . . . 173
6.5 Power spectral densities at different airspeeds (hover, 1 m/s,
and 5 m/s). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
6.6 Filtered pressure signal statistics. . . . . . . . . . . . . . . . . . 174
6.7 1-DOF pitch stand. . . . . . . . . . . . . . . . . . . . . . . . . . . 175
6.8 Gust perturbation along longitudinal axis results in pitch-up
moment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
6.9 Variation of nose rotor thrust with rotational speed at vari-
ous freestream velocities. . . . . . . . . . . . . . . . . . . . . . . . 177
6.10 Variation of pitching moment with rotational speed at vari-
ous freestream velocities. . . . . . . . . . . . . . . . . . . . . . . . 178
6.11 1-DOF pitch response for 10 m/s longitudinal ramp gust
perturbation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
6.12 1-DOF roll stand . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.13 Gust perturbation along lateral axis results in roll-right mo-
ment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.14 1-DOF roll response for a 6.5 m/s lateral ramp gust pertur-
bation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
6.15 1-DOF roll response for a 7 m/s lateral ramp gust pertur-
bation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
6.16 6-DOF experimental stand. . . . . . . . . . . . . . . . . . . . . . 184
6.17 Synthetic gust generation device with shutter system. . . . . 185
6.18 Shutter system mechanism provides step gust inputs. . . . . . 185
6.19 Custom mechanism to survey flow field at 66 discrete points
in test cross-section. . . . . . . . . . . . . . . . . . . . . . . . . . . 186
xiv
6.20 Magnitude of step gust input for one vs. two rows of Dyson
fans. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
6.21 Variations in magnitude and duration of gust profiles using
the shutter mechanism. . . . . . . . . . . . . . . . . . . . . . . . . 187
6.22 Closed-loop feedback system with gust rejection controller. . 189
6.23 Onboard flow sensing mounted on the cyclocopter. . . . . . . 189
6.24 Displacement and pitch control input for 2.8 m/s step gust. . 192
6.25 Displacement and thrust vector input for 2.8 m/s step gust. 193
6.26 Displacement vs. duration of gust for pitch control and
thrust vectoring inputs. . . . . . . . . . . . . . . . . . . . . . . . . 194
6.27 Displacement vs. gust durations for 2.8 m/s and 4 m/s. . . . 195
6.28 Definition of settling time. . . . . . . . . . . . . . . . . . . . . . . 196
6.29 Settling time vs. gust durations for 2.8 m/s and 4 m/s. . . . 196
6.30 Setup for crosswind gust scenarios (β > 0). . . . . . . . . . . . . 198
6.31 Attitude response for step gust input at 30 degrees azimuth
from longitudinal axis. . . . . . . . . . . . . . . . . . . . . . . . . 198
6.32 Feedforward roll input to counteract gust effects. . . . . . . . 199
6.33 Displacement vs. gust duration for crosswinds using com-
bined controller. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
6.34 Settling time vs. gust duration for crosswinds using com-
bined controller. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
7.1 All-terrain cyclocopter equipped with four cyclorotors. . . . 205
7.2 Modified cyclorotor blade pitching mechanism for the all-
terrain cyclocopter. . . . . . . . . . . . . . . . . . . . . . . . . . . 206
7.3 Averaged thrust vs. RPM for cyclorotors. . . . . . . . . . . . . 207
7.4 Averaged power vs. RPM for cyclorotors. . . . . . . . . . . . 207
7.5 Iterations of cyclowheel designs. . . . . . . . . . . . . . . . . . . 209
7.6 Retractable landing gear design. . . . . . . . . . . . . . . . . . 211
7.7 Closed-loop feedback system for the all-terrain cyclocopter. 213
7.8 Orientation of cyclorotors in aerial mode. . . . . . . . . . . . . 214
7.9 Control strategy for aerial mode. . . . . . . . . . . . . . . . . . . 215
7.10 Control strategy for terrestrial mode. . . . . . . . . . . . . . . . 217
7.11 Control strategy for aquatic mode. . . . . . . . . . . . . . . . . 218
7.12 Demonstration of aerial, terrestrial, and aquatic modes. . . . 219
7.13 Onboard attitude data streamed from ELKA-R during hover.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
7.14 Power required for terrestrial and aerial modes. . . . . . . . . 222
7.15 Input required for forward and backward translation in ter-
restrial mode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
7.16 Minimum turn radius for terrestrial locomotion. . . . . . . . 223
7.17 Experimental setup to evaluate maximum ground speed. . . 223
7.18 Power vs. velocity in terrestrial mode. . . . . . . . . . . . . . 225
xv
List of Abbreviations
A cyclorotor projected area
b blade span
c blade chord
CG center of gravity
dp moment arm of propeller
dx, dy, dz moment arm of cyclorotor
Dx, Dy, Dz X, Y, Z drag components of cyclorotor
fs sampling rate
Fp Z component force from propeller
Fx, Fy, Fz X, Y, Z lift components of cyclorotor
g gravity
H angular momentum of the cyclorotor
Ix, Iy, Iz moment of inertia in body reference frame
kd coefficient for the drag force
K gains used for PD controller
L,M,N X, Y, Z component of the moment in body reference frame
m mass
p, q, r roll, pitch, yaw rates in body reference frame
PL power loading
Qcyclo torque of cyclorotor
Qprop torque of propeller
R radius
u, v, w X, Y, Z component of velocity vector in body reference frame
γabs absolute phase angle
γhover hover phase angle
Φ phase angle with respect to hover
φ, θ, ψ roll, pitch, yaw Euler angles
ω angular velocity of the cyclorotor
ρ air density
σ rotor solidity
BEMT Blade Element Momentum Theory
CFD Computation Fluid Dynamics
DARPA Defense Advanced Research Projects Agency
FM Figure of Merit
MAV Micro Air Vehicle
PID Proportional-Integral-Derivative
PIV Particle Image Velocimetry
Re Reynolds Number
RPM Revolutions Per Minute
UAV Unmanned Aerial Vehicle
VTOL Vertical Take-off and Landing
xvi
Chapter 1: Introduction
1.1 Micro Air Vehicles
Growing interest in highly portable versatile flying platforms and recent advance-
ments in microelectronics have led to the development of a scaled-down class of
Unmanned Aerial Vehicles (UAVs) known as Micro Air Vehicles (MAVs). MAVs
were formally defined as aircraft with maximum dimension of 15.24 cm (6 in) and
maximum weight of 100 g (0.22 lbs) by the Defense Advanced Research Projects
Agency (DARPA) in 1997 [1]. The size constraints were selected to define a robust
MAV platform that is portable, operational in confined spaces, and undetectable
due to its low noise signatures. In addition, the target performance set forth by
DARPA included payload capacity of 20 g, endurance up to 1 hr, range of 10 km,
and maximum speed between 10 to 20 m/s.
While DARPA envisioned MAVs for military applications to increase situa-
tional awareness and reduce unit exposure times, MAVs have since been widely
adopted for civilian and commercial applications. Some of these applications in-
clude agricultural monitoring, search and rescue (disaster relief), package delivery,
aerial photography, and infrastructure inspection (Fig. 1.1). With advent of com-
mercial autopilot kits and accessible additive manufacturing, MAVs have also been
1
Figure 1.1: Commercial applications for MAVs.
assimilated as a recreational activity. Considering both commercial and recreational
applications, the Federal Aviation Administration forecasts that the number of op-
erating MAVs could reach 442,000 by 2021. Therefore, it is important to understand
the technical barriers in designing a robust MAV platform suitable for multiple mis-
sion scenarios.
1.2 Technical Challenges
For many mission scenarios, high endurance, maneuverability, and the ability to
tolerate and overcome environmental disturbances such as wind gusts are critical
requirements for MAVs. Designing the next generation micro air vehicle that is also
aerodynamically efficient and capable of vertical take-off and landing (VTOL), high
speed forward flight, and multi-mode locomotion involves overcoming an array of
technical challenges further exacerbated at the MAV-scale (Fig. 1.2).
2
Figure 1.2: Key design considerations for the next generation MAV.
1.2.1 Low Reynolds Number Aerodynamics
One key challenge is understanding the governing aerodynamics of the flow regimes
at which MAVs operate. Compared to their full-scale counterparts, MAVs operate
at Reynolds numbers atleast two order of magnitudes lower. The Reynolds number
(Re) is a non-dimensional parameter used to quantify scaling effects and is effectively
a ratio of inertial forces to viscous forces that act on fluid elements in the flow:
ρvl
Re = (1.1)
µ
The Reynolds number is proportional to the fluid density (ρ), velocity of the
object moving relative to the fluid (v), characteristic length of the object (l), and
inversely proportional to the dynamic viscosity of the fluid (µ). At high Reynolds
3
Figure 1.3: Effect of Reynolds number on maximum sectional lift-drag
ratio [2–4].
numbers (Re > 106), inertial forces dominate and fluid momentum has a greater
influence than viscous resistance. The resulting flow is mostly characterized by tur-
bulent boundary layers. MAVs typically operate at low Reynolds number (104 –
105) because of their reduced size and low operational velocity. At low Reynolds
numbers, viscous forces are dominant and the resulting flow is mostly characterized
as laminar. As a result, the upper surface on the airfoil experiences large adverse
pressure gradient around the leading edge. With laminar flow over the airfoil sur-
face, the flow separates into a shear layer that reattaches as a turbulent boundary
layer. The process of separation and reattachment creates a laminar separation bub-
ble (LSB). The resulting flow is characterized by a turbulent boundary layer that
remains attached throughout the surface.
The aerodynamic performance of any lifting surface is sensitive to the transi-
tion and separation of flow on the boundary layer of its airfoil section. As Reynolds
4
number decreases, the separated shear layer reattaches closer towards the trail-
ing edge. Consequently, the size of both the laminar separation bubble (15%–40%
chord) and thickness of the turbulent boundary layer increases, resulting in reduced
maximum lift capability and increased parasitic drag [3]. At even lower Reynolds
numbers (10,000 < Re < 50,000), the separation of shear layer occurs closer towards
the trailing edge and is unable to reattach itself as a turbulent boundary layer. The
phenomenon results in increased pressure drag due to the airfoil operating in trailing
edge stall. Overall, flow physics lead to degraded maximum sectional lift-to-drag
ratios at low Reynolds numbers (Fig. 1.3).
Figure. 1.3 shows that alternatives to conventional smooth airfoils, such as
rough airfoils and flat plates, can result in increased lift-to-drag ratios at low Reynolds
numbers. Early studies on low Reynolds number (20,000 < Re < 200,000) aero-
dynamics found that thin plate airfoils performed better than conventional thick
cambered airfoils such as N60 [5]. In addition, a sharpened leading edge prevented
flow separation and reduced viscous wake and profile losses [6, 7]. Recently, a sys-
tematic parametric study found that a 2-bladed rotor using carbon fiber blades with
1.2% thickness-to-chord ratio flat plate, 6.1% camber, large chord (21.4 mm), high
taper (0.25-0.5), and negative blade twist of -10◦ to -20◦ achieved 34% increase in
aerodynamic efficiency over a baseline NACA0012 rotor at Re < 5× 104 [4].
Overall, understanding low Reynolds number aerodynamics and its effect on
MAV design is still a subject of ongoing research and one that proves critical for
developing next generation platforms. While these studies focused on a conventional
rotor (e.g main rotor), there is a potential to achieving higher performance with novel
5
Figure 1.4: Components required for MAVs.
rotary-wing configurations. Such configurations are explored in this research.
1.2.2 Size, Weight,and Power (SWaP) Constraints
Aside from aerodynamic considerations, MAVs are also constrained by size, weight,
and power (SWaP) requirements. Increasing endurance and range capability largely
depends on efficiency of the propulsion system and the overall size and weight of
the vehicle. The total mass of the vehicle should be minimized to effectively reduce
power consumption. In particular, the minimum power required (ideal power) to
hover for a rotary-winged platform with a single main rotor can be expressed as:
3/2
P = √W (1.2)
2ρA
6
Figure 1.5: Sub-system weights for commercial quadcopters.
Minimum power required is dependent on vehicle weight (W), air density (ρ),
and rotor area (A). As evident from the equation, power can be approximated by
W 3/2 such that doubling the weight increases the power required by a factor of
approximately 2.8. For a typical MAV, weight contribution can be divided into
sub-systems containing propulsion, avionics, electronics, structure, and power plant
(Fig. 1.4). The selection of avionics, batteries, and onboard electronics for the
transmission system is severely limited by commercial availability.
Figure 1.5 presents a breakdown of sub-system contributions to total vehicle
weight of commercial quadcopters [4]. Components not individually labeled (e.g
avionics) are grouped together in the last category. As weight decreases by an
order of magnitude, contribution of motors increases due to limited availability of
commercially off-the-shelf components. Commercial availability of microelectronics
plays a significant role in designing the propulsion system and even sets power
requirements for MAVs. Overall, SWaP contraints require careful consideration
of the MAV design in terms of mechanical complexity, commercial limitations of
microelectronics, and power budget.
7
1.2.3 Flight Stability and Control
While reduced size and weight of MAVs provide operational benefits, the result-
ing low inertia coupled with limited aerodynamic damping contributes to degrading
flight stability and control performance. Compared to full-scale vehicles, MAVs
typically have higher thrust-to-inertia ratios that result in increased control sensi-
tivity. The difference in flight dynamics is evident when considering scaling effects.
A geometrically and dynamically scaled model can be compared with its full-scale
counterpart using the following simplified relations for length (L), weight (W), and
moment of inertia (I) [8, 9]:
Lm = La/N (1.3)
W 3m = Wa/N (1.4)
Im = Ia/N
5 (1.5)
Dynamic scaling equates the Froude number (ratio of inertia-to-gravity forces)
of the model (m) to the full-scale aircraft (a). A 1th scaled model will have
1
N N3
the weight and 15 the moment of inertia of its full-scale counterpart. Compari-N
son between the Yamaha R-50 UAV and UH-1H confirmed the scaling laws [10].
Vehicle frequencies remain the same, but as the vehicle size is reduced, the subse-
quent decrease in weight and inertia results in increased sensitivity to control inputs.
Therefore, stability augmentation is critical to achieve stable controlled flights for
8
Figure 1.6: CFD cases of wind stream-lines at 1.5 m above ground [11].
MAVs.
Because of low inertia and operational speeds, MAVs are highly susceptible to
environmental disturbances such as gusts. Flow fields in urban settings are especially
prone to spatial and temporal variations arising from scattered layouts of buildings
and trees (Fig. 1.6). These complex and unpredictable gusts can lead to critical
perturbations in vehicle attitude, stall, and ultimate loss of control. For an edgewise
gust, MAVs experience both drag force in the direction of the flow and pitching
moment that rotates the vehicle away from the source. Possible methods to alleviate
gust effects include increasing the vehicle’s control authority and improving the
closed-loop feedback system. Later sections discuss current literature in MAV gust
studies and the unique control capability that increased gust tolerance capability of
the unconventional MAV explored in this research.
9
Figure 1.7: Endurance vs. vehicle mass of current MAVs.
1.3 Existing Micro Air Vehicle Configurations
Within the past decade, numerous successful MAVs have been developed that ad-
dress many of these technical challenges. Existing vehicles can be classified into
three major categories: fixed-wing, rotary-wing, and flapping-wing configurations.
1.3.1 Fixed-wing
Fixed-wing MAVs are the most prevalent due to their higher lift-to-drag ratio, me-
chanical simplicity, and ease of fabrication. Figure 1.7 presents flight endurance
versus vehicle mass of current MAVs while highlighting the different configurations.
As evident, fixed-wing MAVs typically have higher endurance than rotary-wing or
flapping-wing configurations and are the closest to reach DARPA’s definition of the
10
target MAV with weight less than 100 g and minimum endurance of 60 mins. One
particular example is the AeroEnvironment Wasp AE RQ-12A that weighs 1300
grams and has an endurance up to 50 mins. The vehicle has a wing span of 102 cm
(3.3 ft) and cruise speed of 10.3 m/s (23 mph). Typical cruise speed of fixed-wing
MAVs range from 6 m/s to 20 m/s [12]. Although highly efficient, fixed-wing MAVs
are unable to hover and are often limited to outdoor missions that require high
endurance, long range, and high-speed flights.
1.3.2 Rotary-wing
Due to their hover to low-speed flight capability, rotary-wing MAVs tend to have
higher mission performance in scenarios requiring indoor operation or in spatially
restricted environments. Conventional rotary-wing MAVs rely on multi-rotor, single
main rotor/tail rotor, or co-axial rotor designs. One of the most prominent recre-
ational quadcopter is the DJI Phantom. The current iteration (DJI Phantom 4)
weighs 1380 g, endurance up to 28 mins, and a maximum flight speed of 20 m/s.
Compared to fixed-wings, rotary-wing MAVs typically have lower endurance because
hovering and low-speed flight modes are states of high power consumption (Fig. 1.7).
The situation is further exacerbated by the degraded performance of conventional
airfoils at the low Reynolds number range (10,000 – 50,000) at which these vehicles
operate. The maximum achievable figure of merit for rotary-wing MAVs is currently
0.65, compared to the 0.85 achieved by their full-scaled counterparts [13,14].
11
1.3.3 Flapping-wing
Technical challenges ensuing from low Reynolds number aerodynamics, SWaP con-
straints, and flight stability and control have prompted for unconventional, ‘out-
of-the-box’ MAV design solutions. One such solution is flapping-wing MAVs that
are bio-inspired and categorized into avian-based and insect-based kinematics [15].
The flapping motion for avian-based platforms operate on the vertical plane and
enable efficient high-speed flights. On the other hand, insect-based platforms op-
erate on the horizontal plane using a reciprocating wing kinematics that achieve
both translation and rotational motions. The reciprocating wing kinematics enable
insect-based flapping-wing MAVs to hover.
While flapping-wing MAVs offer highly maneuverable and gust-tolerant solu-
tions, existing platforms achieve low hover efficiencies. The AeroVironment Hum-
mingbird Nano is a two-wing platform weighing 19 g with a tip-to-tip wing span
of 16 cm (6.5 in). The vehicle has demonstrated continuous hover endurance of
8 mins, forward flight up to 4.9 m/s (11 mph), and tolerated gusts up to 2 m/s
(5 mph) [16]. A key challenge in developing flapping-wing MAVs is that they are
mechanically complex and experience heavy fatigue due to their high frequency flap-
ping kinematics. In addition, much of the research into understanding the unsteady
aerodynamics and aeroelasticity of flapping wings are still in the incipient stages.
As a result, only a few flapping-wing MAVs have been successfully developed in the
past decade.
12
Figure 1.8: Cyclorotor.
Figure 1.9: Four-bar linkage pitching mechanism.
13
1.3.4 Cycloidal Rotor Concept
Unconventional rotary-wing configurations such as the cyclocopter could be an alter-
nate solution to developing a hover-capable, maneuverable, and gust tolerant MAV.
The cyclocopter relies on cycloidal rotors (cyclorotors) as main source of propulsion.
The cyclorotor is a horizontal axis propulsion system where the blades span parallel
to the axis of rotation and perpendicular to direction of flight (Fig. 1.8). The unique
arrangement of the cyclorotor blades with a pitching mechanism enables a passive
cyclic blade pitching around the rotor azimuth. The pitching mechanism is designed
such that the blades have a positive geometric angle of attack at both the top and
bottom halves of the circular trajectory (Fig. 1.9), thereby producing a net resultant
thrust.
Since the span-wise sections of the cyclorotor blade operate at similar aero-
dynamic conditions (e.g flow velocity, angle of incidence, Reynolds number, etc.),
the cyclorotor can be easily optimized for maximum power loading (thrust/power).
Another advantage of the cyclorotor is its thrust vectoring capability by changing
the phase of cyclic pitching. Rotating the phase angle rotates the location of the
maximum blade pitch angle along the rotor azimuth. As a result, both the magni-
tude and direction of the thrust vector can be adjusted by varying the blade pitch
amplitude and phasing, respectively. The current research is focused on exploring
improvements in flight performance of a cyclocopter configuration using thrust vec-
toring capability. Detailed information on the cyclorotor operating principles and
cyclocopter vehicle development are provided in Chapter 2.
14
Figure 1.10: Rendering of an early Samoliot cyclocopter concept by E.P.
Sverchkov.
1.4 Previous Studies
The following sections are categorized into literature review of (1) cyclocopter re-
search and (2) MAV research on the experimental studies related to the current
work. The latter includes studies pertaining to other rotary-wing configurations.
1.4.1 Existing Cyclocopter Research
While many breakthroughs in cyclorotor research have occurred in recent years, at-
tempts to develop a cyclocopter date back to early 20th century (1909-1914) [17].
Numerous full-scaled models intended to seat one pilot were developed, but none
of the attempts were successful in achieving flight. One of the earliest known cy-
clocopter was the “Samoliot” (wheel ornithopter) developed in 1909 by E. P. Sver-
chkov, a Russian military engineer. The manned cyclocopter was estimated to have
an empty weight of 200 kg (Fig. 1.10). Another early cyclocopter was developed be-
tween 1909-1914 in France. However, the invention did not have any documentation
15
Figure 1.11: Kirsten’s cycloplane featured in Popular Science, model in
the wind tunnel, and Kirsten-Boeing experimental test rig [18, 19].
aside from a video of the failed flight test showing the rotors disintegrating.
In the 1920s, series of research on cycloidal propulsion systems for aerial and
marine applications were published by Professor Kurt Kirsten from the University of
Washington [18,19]. A collaboration with W. E. Boeing led to the development of the
“Kirsten-Boeing” propeller, an early version of the cycloidal propeller (Fig. 1.11).
Kirsten’s idea was to equip an airship and a “cycloplane” with the propeller to
provide three-dimensional control. The first model of the “cycloplane” was built
with a small 26 cm (10.2 in) propeller and 15 cm (5.9 in) blade span. After wind
tunnel results of the model showed satisfactory results, a larger model was built and
six propellers were selected to power the USS Shenondoah airship. However, the
airship crashed before the propellers were installed, which ultimately halted further
development.
Kirsten also modified the propellers for maritime application and sold his
patent to the Voith-Schneider Corp. Replacing the rudder, the experimental pro-
16
Figure 1.12: Voith-Schneider propeller used by tugboats [20, 21].
Figure 1.13: Strandgren’s cyclogyros [22].
peller with its thrust vectoring capability could provide the necessary propulsion
and control for boats. A variant of Kirsten’s propeller was developed in 1931 by
Ernst Schneider, an Austrian inventor [20,21]. The Voith-Schneider propeller (VSP)
is still successfully being used for commercial applications (Fig. 1.12).
From 1924-1933, Swedish-French engineer Strandgren conducted series of ex-
periments with the cyclogyro model and ultimately published the first scientific
study on cycloidal rotors [22]. The publication included a simplified quasisteady
aerodynamic analysis of a cyclogyro to understand how lift and propulsive thrust
17
Figure 1.14: Wheatley’s experimental setup [23].
are generated. The analysis showed that varying the amplitude and phasing of the
blade pitch kinematics could change the resultant magnitude and direction of the
rotor’s thrust. He also explored the feasibility of the concept and built a 600 kg
model (Fig. 1.13).
John B. Wheatley at the National Advisory Committee for Aeronautics (NACA)
followed and continued Strandgren’s analysis. In 1933, Wheatley published a simpli-
fied aerodynamic theory using blade element momentum theory (BEMT) to predict
forward flight performance of the cyclogyro [23]. He also performed a case study
to investigate the feasibility of a 1360 kg (3000 lbs) cyclorotor-based aircraft. In
1935, Wheatley and Windler conducted experimental studies of a cyclorotor in for-
ward flight to compare against his analysis. The study was performed on a 4-bladed
cyclorotor with a diameter and span of 2.4 m and blade chord of 0.095 m. While
18
Figure 1.15: Cyclocopters developed by the Seoul National University [24–
28].)
the results qualitatively agreed with model, the relative magnitudes significantly
differed because of all the simplifying assumptions used to derive the model.
With a lapse in cyclocopter research up to 1998, modern work largely focused
on UAV and MAV-scale vehicle development. Advent of microelectronics, compos-
ites (e.g carbon fiber), rapid prototyping (additive manufacturing), and improved
understanding of aerodynamics led to renewed interest in the cyclocopter config-
uration. Since 2003, many UAV-scale versions of the cyclocopter were developed
at the Seoul National University in South Korea [24–28]. Work completed by Kim
et. al include both computational and experimental studies in order to improve the
design of a 12.8 kg quad-cyclocopter and a 110 kg twin-cyclocopter UAV (Fig. 1.15).
The quad-cyclocopter used four elliptical blades with a blade span and rotor diam-
eter of 0.5 m, NACA0018 airfoil, and operating speed of 1100 rpm. The vehicle’s
flight stability was augmented with a commercial attitude heading reference system
(AHRS) which enabled it to demonstrate stable flight. Pilot controls included throt-
tle, pitch, roll, and yaw, which were all achieved by modulating rotational speeds of
19
Figure 1.16: Early twin-cyclocopter work conducted at the University of
Maryland [29].
the cyclorotors. The UAV-scale twin-cyclocopter used 4-bladed cyclorotors with 2 m
diameter, blade span of 1.5 m, NACA0018 airfoil, and operating speed of 420 rpm.
An additional anti-torque horizontal tail rotor with diameter of 0.574 m operated
at 5320 rpm and also provided pitch control.
In 2006, Sirohi, Parsons, and Chopra conducted feasibility studies of MAV-
scale cyclocopter at the University of Maryland [29,30]. Experimental studies were
conducted on a cyclorotor with a 15 cm diameter and blade blade span. The effect
of rotational speed, blade pitch amplitude, and number of blades on rotor thrust and
power consumption in hover were systematically studied. Analysis using a double-
multiple-streamtube model was developed and compared against the experimental
results. Results from the study was used as a basis for the conceptual design of a
twin-cyclorotor solely using two cyclorotors (Fig. 1.16).
From 2008 to 2012, Benedict et. al performed comprehensive experimental and
analytical studies on multiple MAV-scale cyclocopters at the University of Mary-
land [31–35]. An extensive parametric study was conducted to determine blade
airfoil, rotor span, radius, rotor solidity, number of blades, blade chord length,
20
Figure 1.17: Hover-capable cyclocopter MAVs developed at the University
of Maryland.
0.25  
Optimized Cyclorotor
Conventional micro−rotor
0.2
0.15
0.1
0.05  
0 20 40 60 80
Disk loading (N/m2)
Figure 1.18: Power loading (thrust/power) vs. disk loading (conventional
micro-rotors vs. optimized cyclorotor)
21
Power loading (N/W)
Figure 1.19: Power versus forward speed for constant rotational speed of
1740 rpm (13.88 m/s) for level, steady flight [36,37].
blade planform, blade pitching amplitude, phase, and blade kinematics for an op-
timized cyclorotor. Compared to a conventional edge-wise rotor at the same disk
loading (thrust/disk area), the optimized cyclorotor achieved higher aerodynamic
power loading (Fig. 1.18). The results led to the successful development of multi-
ple hover-capable cyclocopter MAVs (Fig.1.17). The twin-cyclocopter configuration
used two cyclorotors and an additional horizontal tail rotor that counteracted the
reaction torque from the cyclorotors and also provided pitch control. These vehicles
ranged from 100 grams to 550 grams and relied on both rotational speed and thrust
vectoring to achieve controlled hover. Since 2014, Benedict et. al continued work on
the cyclocopter configuration at Texas A&M University and developed a meso-scale
cyclocopter weighing only 29 grams [38].
Wind tunnel studies conducted by Jarugumilli, Benedict, and Chopra on an
isolated cyclorotor have shown that an aircraft using cyclorotors could efficiently
reach very high forward speeds without any lift augmenting devices/surfaces [36,37].
22
Figure 1.20: System identification conducted on a twin-cyclocopter by the
University of Maryland [39].
It was observed that the power required for a cyclorotor to maintain a steady, level
flight significantly drops with forward speed (Fig. 1.19) up to almost an advance
ratio of 1, due to increased lift producing efficiency of the cyclorotors. Unlike a
conventional helicopter, the forward flight of a cyclocopter would be performed
purely by utilizing thrust vectoring (varying cyclic pitch phasing) instead of pitching
the entire vehicle forward. Thrust vectoring enables the cyclocopter to maintain a
level pitch attitude in forward flight. Unlike the transition for a tilt-rotor, transition
from hover to forward flight for the cyclocopter is possible without undergoing any
configuration changes. Overall, the study showed that thrust vectoring provides a
mechanically simple and power efficient method of achieving forward flight.
To evaluate the vehicle’s tolerance to external gust disturbances, Hrishikesha-
van, Benedict, and Chopra experimentally derived a comprehensive flight dynamics
model in hover using time domain system identification techniques [39]. The study
was conducted using a 500 g twin-cyclocopter tested inside a motion capture system
23
(Fig. 1.20). It was the first vehicle performance study and system identification
conducted on the cyclocopter MAV. The experiments involved series of input exci-
tations that excited different dynamic modes. Throughout the experiments, vehicle
response was recorded by both the motion capture system and onboard autopilot.
The results provided insights into the existence of a destabilizing gyroscopic cou-
pling between roll and yaw degrees of freedom, which required implementation of
decoupling methods in the controls system. When compared to other rotary-based
MAV platforms, the cyclocopter had higher longitudinal and lateral gust tolerances
in hover (7.9 m/s and 17 m/s, respectively). The flight dynamics model from the
study revealed many key insights into the vehicle’s maneuverability in hover condi-
tions and led to an improvement of the closed-loop feedback control system.
1.4.2 Gust Tolerance Studies of Rotary-wing MAVs
While flow fields in urban environments are complex and unpredictable, research
studies have attempted to study the effects of gusts and turbulence on MAV per-
formance. Galway et al. [40] modeled the effects of building wakes on a Yamaha
R-50 R/C helicopter analytically using CFD and estimated that the turbulent flow
fields cause up to 10◦ variations in attitude for wind up to 8 knots. There has also
been numerous studies that took an experimental approach and either flight tested
MAVs in front of a wind tunnel or a custom gust generation device. Johnson and
Jacob [41] assembled a gust and shear wind tunnel using a bank of independently
controlled computer fans and a flow straightener. The fan bank generated up to 5
24
Figure 1.21: Synthetic gust generation device and experimental setup as-
sembled by Georgia Institute of Technology [42].
m/s freestream velocity and a slow gust ramp input lasting 3s. The study tested a
tethered R/C fixed-wing biplane MAV with a nose propeller in order to characterize
its wake.
Zarovy et al. [42] assembled a gust generation system using a bank of eight
rotary fans that are simultaneously controlled to achieve up to 2.6 m/s (Fig.1.21).
The study specifically looked at scaling effects on disturbance rejection capability
of single-main-rotor MAVs and found that the Lama 400, the largest tested R/C
helicopter, was capable of withstanding 2.6 m/s gusts. In the experiments, the
MAVs vertically took-off and slowly translated into the flow-field region generated
by the fans. The metric used for characterizing performance is the spherical error
probable (SEP) which covers 50% of the vehicles translational trajectory in gust
and a smaller magnitude indicates higher gust tolerance. Similarly, Hrishikeshavan
et al. used a comparable metric to study gust effects on a shrouded rotor MAV and
found that the vehicle is capable of tolerating gusts up to 3 m/s [43]. The study
25
Figure 1.22: Experimental setup for gust study of a ducted-fan MAV con-
ducted by the University of Maryland [43].
used the circle error probable (CEP) which restricts the radius of vehicle’s trajectory
within the X-Y plane. The gust generation device used a large rotary fan with a
honeycomb flow-straightener to test the MAV in free flight. In addition, the vehicle
was first mounted on a gimbal stand in front of an open-jet wind tunnel in order to
measure the vehicle’s control moments against a gust perturbation.
While there has not been any prior experimental studies investigating the re-
sponse of a cyclocopter MAV to gust perturbation, Hrishikeshavan et al. conducted
a maneuverability and disturbance rejection analysis using a linear 6-DOF flight dy-
namics model of the cyclocopter in hover [39]. As previously mentioned, the study
estimated that a 550 g cyclocopter had longitudinal and lateral gust tolerances of 7.9
m/s and 17 m/s, respectively. Chapter 6 investigates the gust tolerance of the 550 g
cyclocopter through wind tunnel studies and tethered flight tests using a synthetic
gust generation device.
26
Figure 1.23: Existing multi-modal micro air vehicles [44–46].
1.4.3 MAVs with Multi-mode Locomotion Capability
Multi-mode mobility is prevalent in biological systems where organisms such as
birds and insects efficiently switch between different modes of locomotion to conserve
energy, traverse long distances, and maneuver through confined spaces. Biologically-
inspired multi-modal MAV such as the Morphing Micro Air-Land Vehicle (MMALV)
achieves flight using a combination of nose propeller and morphing flexible wings
[44]. The MMALV weighs 120 g and has a maximum dimension of 30 cm (wingspan)
7.12. The vehicle achieves terrestrial locomotion using rotating wheel-legs (Mini-
WhegsTM) that are independently driven by a motor and inspired by locomotion
principles of the cockroach. While the vehicle is capable of flight, it is unable to
hover and requires either a 3–5 m runway or a 6 m vertical drop in order to take-off.
Rotary-winged multi-modal MAV such as the Hybrid Terrestrial and Aerial
Quadrotor (HyTAQ) relies on conventional edgewise rotors with fixed-pitch pro-
pellers as main source of propulsion [45]. The 570 g platform includes a quadcopter
suspended inside a rolling cage that enables terrestrial locomotion and also functions
27
as the landing gear 7.12. Although the multi-modal design is mechanically simple,
rolling motion is attained aerodynamically using propulsive thrust from the rotors
instead of directly harnessing the torque of the motors.
The Multi-field Universal Wheel for Air-land (MUWA) quadcopter is capable
of aerial, terrestrial, and aquatic locomotion [46]. Rotors are independently driven
by brushless motors and use variable pitch propellers that are actuated by rotary
servos. Terrestrial and aquatic modes rely on the ring-shaped polystyrene foam
body that operates as a mono-wheel, buoy, and landing gear. In terrestrial mode,
the MUWA platform acrobatically flips onto its foam body and applies rotational
control by differentially modulating torque of the four motors (Fig.7.12). For aquatic
locomotion, only a fraction of the total resultant lift is applied to achieve translation
since the vehicle has to pitch to generate propulsive thrust. Overall, to function in
multiple modes of locomotion, the MUWA platform uses eight different actuators
and an external foam body. Along with its foam outer-ring, MUWA weighs 2100 g
and spans 0.91 m in diameter and 0.12 in body thickness.
1.5 Research Objective and Technical Approach
Previous work included series of experimental and analytical work on the cycloro-
tor propulsion system, as well as vehicle development for both twin-cyclocopter
and quad-cyclocopter configurations. However, previous studies mainly focused on
demonstrating stable flight in hover and understanding flight dynamics to improve
vehicle performance in that limited flight envelope. Apart from performance studies
28
and flow field measurements of an isolated cyclocopter in forward flight, none of
the previous studies integrated thrust vectoring capability as a key control input in
achieving stable level forward flight. As a result, there is a lack of understanding of
cyclocopter flight dynamics in forward flight and the required control strategies to
achieve level power-efficient transition from hover to level forward flight.
Moreover, the only gust tolerance study of the cyclocopter configuration was
the analytical grammian-based maneuverability and disturbance rejection analysis
using the experimentally derived linear 6-DOF flight dynamics model of the vehicle
in hover. While the results of that study provide a foundation for gust tolerance
assessment of the cyclocopter, the results still require experimental validation. In
addition, previous studies focused on demonstrating controlled flight with a basic
avionics package and closed-loop feedback system for flight stability and control.
Additional sensors for flight performance diagnostics and a supplementary gust re-
jection controller could augment cyclocopter’s performance against external distur-
bances.
Furthermore, previous studies explored ways of increasing cyclocopter’s en-
durance through systematic aerodynamic optimization of the cyclorotors. An al-
ternative and unconventional method of conserving energy is switching between
different modes of locomotion. Such multi-mode locomotion capability has already
been demonstrated for fixed-wing and rotary-wing MAVs.
With the limitations of the previous studies in mind, the key objectives of the
present work include:
29
1. Develop control strategies required to achieve power-efficient level forward
flight for the twin-cyclocopter configuration using a combination of thrust
vectoring and rotational speeds of the two cyclorotors and a horizontal nose
rotor.
2. System Identification of a 6-DOF flight dynamics model in forward flight and
conduct performance studies in the wind tunnel in order to understand ef-
fects of cross-coupling dynamics and thrust vectoring capability on vehicle’s
stability and control.
3. Experimentally evaluate gust tolerance strategies based on control inputs (pitch
control vs. thrust vectoring) and type of gust feedback controller (flow-based
vs. position-based) to increase cyclocopter’s performance against external dis-
turbances.
4. Determine feasibility and performance of cyclocopter’s multi-mode capabil-
ity by demonstrating efficient aerial, terrestrial, and aquatic locomotion with
seamless transition between modes. In particular, explore a quad-cyclocopter
configuration unaffected by cross-coupling dynamics in hover.
1.5.1 Thesis Organization
The current chapter provided an overview of technical challenges affecting MAV de-
sign and summarized existing MAV configurations (fixed-wing, rotary-wing, flapping-
wing). In addition, the chapter also presented a thorough literature review of pre-
vious work on cyclorotors and cyclocopter vehicle developement.
30
Chapter 2 provides a detailed overview of the twin-cyclocopter configuration
developed in this research. The chapter also discusses the basic operating principle
of the cyclorotor and the mechanical design of the four-bar linkage-based pitching
mechanism. The aerodynamic and structural design of cyclorotor blades are also
discussed, along with the resulting performance of the cyclorotor and the horizontal
nose rotor.
Chapter 3 includes an extensive review of avionics including autopilots, sen-
sors, and onboard computers for executing computationally intensive tasks such as
image processing. The chapter first reviews two custom autopilots implemented
on the cyclocopter and then surveys sensors and onboard computers commercially
available for MAVs. These sensors are categorized for flight stability and control,
flight performance diagnostics, and guidance and navigation (autonomy). Finally,
the chapter concludes with an overview of the closed-loop feedback system imple-
mented on the cyclocopter.
Chapter 4 focuses on the control strategies developed for the cyclocopter in
hover and forward flight. Sources of cross-couplings across all flight modes are also
discussed in details. Results from wind tunnel studies conducted using a 5-DOF
experimental stand are used to develop a control strategy to achieve decoupled
dynamics. Furthermore, effects of free-stream velocity, rotational speed, and phasing
angle on the roll-yaw cross-couplings are closely investigated and used to determine
control inputs to achieve zero pitch attitude at various airspeed.
Chapter 5 presents a 6-DOF flight dynamics model of the cyclocopter that
is experimentally determined using time-domain system identification technique.
31
The existence of the longitudinal control derivative in translation mode is used to
evaluate the control authority of thrust vectoring to command forward speed. The
model is also used to identify sources of roll-yaw coupling and develop decoupling
methods to achieve stable flight.
Chapter 6 examines the disturbance rejection capability of the cyclocopter
through series of experiments in the wind tunnel and using a synthetic gust genera-
tion device. The results from the wind tunnel studies are used to evaluate response
to gust perturbation along the longitudinal and lateral axis. The chapter also com-
pares control authorities of pitch control versus thrust vectoring and feedback control
schemes of position feedback versus flow feedback. Results of the comparative study
of gust rejection strategies are provided for the 6-DOF experiments in front of a syn-
thetic gust generation device for step gust inputs along the longitudinal axis and
crosswinds.
Chapter 7 discusses design modifications to a quad-cyclocopter configuration
to achieve multi-mode locomotion capability. Key structural components used to
enable aerial, terrestrial, and aquatic modes are discussed. The chapter also presents
the control strategies for each mode and multi-mode testing results. Results include
onboard attitude data from a custom auto-pilot, power consumption for all modes,
and performance in terrestrial mode evaluated using a 1-DOF test rig and VICON
motion capture system.
Chapter 8 concludes this dissertation with summaries of previous chapters
and suggestions of recommended future work. Overall, the chapter provides a final
discussion of key contributions from this research.
32
Chapter 2: Vehicle Design
2.1 Overview
In order to realize improvements in aerodynamic performance, power-efficient for-
ward flight, gust tolerance, maneuverability, and multi-mode capability, an uncon-
ventional configuration such as the cyclocopter is explored in this work. This chapter
covers design and development of the cyclocopter with a focus on cyclorotor de-
sign. The key operating principle of cyclorotor and the associated mechanisms are
discussed. Aerodynamic and structural variations in blade design are thoroughly
explored with emphasis on weight reduction and hover performance. The chap-
ter also summarizes parametric test results used to select the nose rotor propeller
and presents scalability study on developing a meso-scale cyclorotor. Overall, this
chapter discusses the development of the 550 g twin-cyclocopter that is used as the
testbed and demonstrator for the remaining chapters.
2.2 Twin-Cyclocopter Configuration
The twin-cyclocopter relies on two cycloidal rotors (cyclorotors) as main source
of propulsion and an additional horizontal anti-torque nose rotor. The nose rotor
33
Figure 2.1: Twin-cyclocopters developed at University of Maryland.
counteracts the torque resulting from both cyclorotors spinning in the same direc-
tion. Anti-torque capability from the nose rotor is also augmented by increasing the
horizontal offset between the rotor and vehicle’s center of gravity (c.g). The offset
reduces the thrust required from the nose rotor which subsequently minimizes its
size. The nose rotor also provides pitch control by modulating its rotational speed.
Work from this research has produced twin-cyclocopters ranging from 60 grams to
550 grams (Fig. 2.1).
The rest of the chapters will focus on the 550 g twin-cyclocopter. The vehicle
has a lateral dimension of 0.381 meters (1.25 feet), longitudinal dimension of 0.457
meters (1.5 feet), and a height of 0.305 meters (1 foot) (Fig. 2.2). Each cyclorotor
has a diameter of 0.152 meters (6 inches), blade span of 0.171 meters (6.75 inches),
a chord of 0.051 meters (2 inches), and is independently driven by a 1400kV AXI
2204/54 out-runner motor through a 7.5:1 single-stage transmission. The horizontal
34
Figure 2.2: 550 g twin-cyclocopter.
Table 2.1: Weight distribution comparison of previous generation twin-cyclocopters
(60–550g).
60g 130g 235g 550g
System % % % %
Cyclorotors 23 33 34 27
Electronics 16 17 14 11
Propeller System 8 16 6 9
Battery 16 14 18 19
Structure 13 9 9 12
Transmission 12 7 17 21
Avionics 12 4 2 1
Total 100% 100% 100% 100%
nose rotor is directly driven by a 1200kV eRC BL400 brushless outrunner motor
and is fitted with a 3-bladed GWS 9050 propeller. All three rotors are powered by
a single 3S 11.1 volt 850 mAh Lithium-Polymer battery.
Weight distribution of various twin-cyclocopters is compared in Table 7.3. All
of the vehicles are equipped with rotors individually driven by a brushless out-runner
motor and two servos that rotate the thrust vectors of the cyclorotors. Additional
onboard components include avionics and power supply. From the weight distribu-
35
Figure 2.3: Weight distribution of twin-cyclocopters are different scales.
tion, it is evident that the cyclorotors make up the majority of the empty weight
fraction of the vehicle. Aside from the 60g vehicle, cyclorotors consistently consti-
tute 27–34% of the total gross weight and the range largely varies based on blade
design (Fig. 2.3). Therefore, significant effort was focused on designing lightweight
and structurally robust blades and pitching mechanism in order to reduce total ve-
hicle weight and increase endurance. Another noticeable trend is that the weight
contribution of the transmission system (cyclorotor motors, gears, bearings, and
associated screws) increases with vehicle scale.
As expected, the weight contribution from electronics (servos, ESCs, and as-
sociated wires) increase at smaller scales. That is because the size of the selected
electronics is limited by what is commercially available. A similar trend is observed
36
Figure 2.4: Cyclorotor.
in the avionics (processor, breakout board, and satellite receiver) since all of these
vehicles use the lightest available onboard processor. The LiPo batteries compose
14–19% of the total vehicle weight and is also limited by commercial availabil-
ity. Overall, the most significant improvements to the vehicle design can be made
through the cyclorotors, which directly affects the transmission system, anti-torque
propeller size, and structural weight (landing gear, fuselage, etc.).
2.3 Cyclorotor Design
The blades on the cyclorotor follow a circular trajectory about a horizontal
axis of rotation (Fig. 2.4). The cyclorotor blade span is parallel to the horizontal
axis and perpendicular to the direction of flight. A pitching mechanism cyclically
varies individual blade pitch angle along a circular trajectory. In hover, the blades
achieve a positive geometric pitch angle at both top and bottom portions of the
trajectory, yielding a net resultant thrust. Both the magnitude and direction of the
thrust vector can be adjusted by varying the pitch amplitude and phasing of the
blade pitching kinematics.
37
Figure 2.5: Four-bar linkage pitching mechanism.
Since all spanwise blade sections of the cyclorotor operate under similar aero-
dynamic conditions (i.e., flow velocity, angle of incidence, Reynolds number, etc.),
the blades can be easily optimized to achieve best aerodynamic efficiency [31]. In
addition, the uniform spanwise distribution of aerodynamic forces enables the cy-
clorotor to operate at lower tip speeds, thereby improving the acoustics signature.
Another advantage of the cyclorotor is its unique thrust vectoring capability that is
achieved by varying the phase of the cyclic pitching. By thrust vectoring instead of
pitching forward, the cyclocopter can achieve level forward flight in a power efficient
manner. Thrust vectoring strategy for forward flight is further discussed in Chapter
4. Effect of thrust vectoring on maneuverability and gust tolerance is also explored
in Chapter 6.
38
Figure 2.6: Pitching mechanism implemented on the 550g twin-
cyclocopter.
2.3.1 Blade Pitching Mechanism
The schematic of the blade pitching mechanism is depicted in Fig. 2.5 where
the four bars of the linkage system are labeled L1, L2, L3 and L4. L1, also referred
to as rotor radius, is the distance between the blade pitching axis and the horizontal
axis of rotation. The pitch links (of length L3) are connected to the end of the offset
link on one end and the other end is connected to point B which is at a distance L4
behind the pitching axis. The connections at both ends of the pitch link are through
pin joints to allow the rotational degree of freedom. With this arrangement, as the
rotor rotates, the blades automatically pitch cyclically, where the pitching amplitude
depends on the offset length, L2, when the other linkage lengths remains fixed. The
rotation of the offset link changes the phasing of the cyclic pitching and thereby
changes the direction of the thrust vector.
39
Figure 2.7: Cyclorotor coordinate system.
Figure 2.6 shows the actual pitching mechanism implemented on the 550g
cyclorotor. The current pitching mechanism requires the offset link (L2) to be in-
stalled at the tip of shaft in a non-rotating frame. In order to reduce mechanical
complexities, the distance L2 is kept constant. With a fixed maximum blade pitch-
ing amplitude, the magnitude of thrust is modulated during flight by varying the
rotational speed.
The direction of the thrust vector can be manipulated by rotating the offset
link, which changes the phasing of the harmonic pitching schedule. For the current
coordinate system (Fig. 2.7), the azimuthal position of the blade (Ψ) is measured
clockwise from the horizontal axis of rotation and is equivalent to zero when the
blades are at the extreme left of the circular trajectory. The time-varying aero-
40
Figure 2.8: Blade pitch angle as a function of rotor azimuth for each cy-
clorotor at trim.
dynamic forces produced by the cyclorotor can be resolved into the vertical and
horizontal directions with a positive x-axis extending towards the right and a posi-
tive z-axis towards the top. In the figure, the phasing angle is denoted as Φ, which
is zero when aligned with the z-axis and positive with a clockwise rotation.
Even though they produce identical values of thrust at each rpm, the cycloro-
tors have different pitch phasings in order to counteract the clockwise reaction torque
from the tail rotor (Fig. 2.8). As a result, the cyclorotor on the right has a higher
phasing which tilts the thrust vector forward and produces a net counterclockwise
yaw moment of the vehicle. The thrust vectoring mechanism on the cyclocopter
that is used to rotate the offset link and provide a phasing input is composed of two
BMS-373MG servos that are directly connected to the pitching mechanism through
a 2:1 gear ratio. The setup allows for ±60◦ of thrust vectoring.
The pitch links are manufactured out of Delrin acetal resin using a CNC
milling machine. Compared to carbon fiber linkages used on the smaller vehicles,
Delrin linkages for the 550 g cyclocopter are pliable and resistant to abrupt fractures.
41
Table 2.2: Cyclorotor design parameters for the 550 g cyclocopter.
Parameter Cyclorotor
Rotor Diameter 0.152 meters (6 inches)
Blade span 0.171 meters (6.75 inches)
Blade chord 0.051 meters (2 inches)
Blade airfoil NACA 0015
Blade pitch amplitude ±45◦
Blade pitch axis 45% from LE
Pliability is important considering the large centrifugal loading on the blades and
subsequently on the linkages which will be discussed in later sections.
2.4 Blade Design
As seen from the weight distributions, the cyclorotors compose up to 27% of the
vehicle’s gross weight. To reduce the empty weight fraction, the cyclorotor blades
must be carefully designed to be lightweight and structurally robust. The rotor
structural design must be able to withstand the dominant centrifugal force because
each gram of blade weight amounts to approximately 0.4 Newton of centrifugal load
on the blade. In addition, the blades also need to have high strength-to-weight
ratio to minimize the bending and torsional deformations caused by the transverse
centrifugal loading. The cyclorotor performance can be attributed to four main
design aspects: (1) blade aerodynamic design, (2) blade structural design, (3) rotor
geometry, and (4) blade kinematics.
42
2.4.1 Aerodynamic Design
Selection of key aerodynamic blade parameters such as blade pitching amplitude,
location of pitch axis, rotor radius, blade airfoil, chord, span, and planform were
based on empirical studies. The parameters of the present cyclorotor for the 550
g vehicle have been optimized through series of systematic experimental studies
conducted in the past [31]. Each blade uses a NACA 0015 airfoil, a symmetric
pitching amplitude of 45◦, a chord/radius of 0.625, and aspect ratio of approximately
3.38. All of the design parameters of the cyclorotor are summarized in Table 2.2.
Leveraging the insights gained from these studies, the present cyclorotor is designed
for improved thrust-to-power ratio (power loading) in hover and power efficiency in
forward flight. Emphasis was also placed on the blade and rotor structural design
to reduce the overall rotor weight.
A critical parameter in cyclorotor performance regarding blade and rotor ge-
ometry is the solidity (σ):
Nbc
σ = (2.1)
2πR
The solidity is affected by the number of blades (Nb) and chord-to-radius
(c/R) ratio. The radius of the cyclorotor is measured using its circular trajectory,
unlike the blade span measurement used for conventional rotors. On a conventional
rotor, there is an optimum solidity that is directly influenced by the total blade area
(Nbcb) at each thrust value. However, for the cyclorotor, changing the solidity by
43
Figure 2.9: Virtual camber and incidence in a curvilinear flow
varying either Nb or c/R yields completely different results because of the different
phenomenon involved. For a fixed c/R, changing the number of blades alters the
total blade area and subsequently, the solidity, which results in a similar effect on
cyclorotor performance as on the conventional rotor. For a fixed number of blades,
changing c/R not only affects the blade area, but also changes the flow curvature
effect on the cyclorotor [36].
The blades on the cyclorotor experience a virtual camber effect because of the
curvilinear nature of the flow, which introduces a chordwise variation in velocity.
Therefore, a symmetric airfoil at 0◦ pitch behaves as a virtually cambered airfoil at
a virtual incidence (Fig. 2.9) in a rectilinear flow. The blades experience a negative
virtual camber and incidence at the top of the trajectory (Cl < 0) and a positive
camber and incidence at the bottom portion of the trajectory (Cl > 0). The study
found that virtual camber and incidence was proportional to the chord-to-radius
ratio at higher values and location of the blade pitching axis. After systematically
44
Figure 2.10: Linear approximation to measure virtual camber and virtual
incidence [31,47].
measuring the power loading (thrust/power) with disk loading (thrust/disk area) at
various solidities and at 40◦ pitching amplitude, the maximum performance resulted
from a four-bladed rotor with a c/R ratio of approximately 0.83. While aerodynam-
ically efficient, the c/R of 0.83 would result in heavy blades with low aspect ratios
(AR) and lead to increased structural loads on the blades. However, it should be
noted that power loading did not drastically change when c/R was within 0.6 to 0.8.
A linear approximation shows that the virtual camber for selected blade pa-
rameters is 3.5% of chord and virtual incidence is 5◦. The pitching axis locations
(x/c) of all blades were centered around the chordwise c.g. location, which resulted
in x/c of 0.45. The pitching axis aligned with the chordwise location of blade c.g.
in order to to reduce the pitching moments on the blade caused by the centrifugal
force. In addition, it reduces the loads on the pitching mechanism and ultimately on
the lightweight thrust vectoring servos with limited torque capacity (1.58 kg-cm).
45
Figure 2.11: Previous iterations of blade design.
46
2.4.2 Structural Design
There have been several iterations of blade designs from the composite sandwich
structure of foam wrapped with carbon fiber to multi-component rib and spar as-
sembly (Fig. 2.11). The main purpose of the design iterations was to reduce overall
blade weight while increasing bending and torsional stiffness. The blades undergo
significant centrifugal loading both perpendicular and transverse along blade span.
The centrifugal loading can be calculated using equation 2.2 where the force in-
creases linearly with mass and rotor radius and quadratically with rotational speed:
CF = mΩ2R (2.2)
While radius and rotational speed are set by aerodynamic parameters of the blade,
mass can be reduced based on improved manufacturing techniques.
2.4.2.1 Carbon Fiber Blade
The first iteration of blade design was a composite structure with rohacell foam
interior wrapped with -45◦/45◦ carbon fiber fabric. The direction of fiber and NACA
0015 profile contributed to torsional stiffness while the closed foam and carbon fiber
contributed to bending stiffness. A bent titanium spar sandwiched inside the foam
provided additional bending stiffness and an attachment point for blade pitching
mechanism. Due to multiple components, the manufacturing process for this design
was complicated and required multiple steps. Both the foam core and final carbon
47
fiber wrap needed to be baked at 350◦ for 30 minutes and 75 minutes respectively,
using an aluminum mold. The blades then connected to the rotor end plates through
a blade attachment adaptor held on by two screws at both spanwise boundaries of
the blade. While this over-designed blade reduced structural deformations, the 10
grams blade was too heavy to use on the cyclocopter.
2.4.2.2 Exposed Foam Blade
In order to reduce the mass, the next iteration of blade design included a light-weight
foam structure with a 1/16 inch titanium inner spar and strips of carbon fiber at the
leading edge and spanwise bounds. The titanium spar and foam provided bending
stiffness while the closed NACA 0015 profile provided torsional stiffness. Similar
to the previous iteration, manufacturing was labor-intensive because of the curing
process required for carbon fiber and foam. While this lightweight design provided
80% reduction in weight, the blade was too flexible and resulted in large bending
deformations. In addition, the foam structure was not robust enough to survive
vehicle crashes during flight testing.
2.4.2.3 Mylar Wrapped Monolithic Blade
A compromise between lightweight and stiff design was reverting back to the initial
composite sandwich structure but removing multiple rectangular cross-sections to
reduce weight. The cut-out blade would then be wrapped with Mylar film to main-
tain a closed structure. The Mylar film was 50 micron thick and provided a smooth
48
Figure 2.12: Rib and spar blade design.
Figure 2.13: Blade mold for constructing new blades.
blade surface. To simplify the blade structure, the inner titanium spar was removed
such that the pitching mechanism connected to the blade solely through the blade
attachment. With the changes, the final blade weight was 7.4 g. The simple mono-
lithic design also reduced probability of structural failure during flight testing. The
pitching axis and linkage attachment point were on two separate delrin blade at-
tachment pieces mounted on spanwise ends of the blade and contributed to the total
blade weight. Besides the weight, a significant drawback of the design was incon-
sistent placement of the blade attachment pieces during the manufacturing process,
which consequentially led to significant blade twist and decreased performance.
49
2.4.2.4 Multi-Component Rib and Spar Blade
With the advent of additive manufacturing and commercially available 3D printers,
the blade design could be further improved using multi-component rib and spar
structure (Fig. 2.12). The ribs are milled out carbon fiber in the NACA 0015 profile
with through-holes for two 1/16 inch carbon fiber spars that attached directly to
the end plates and pitching mechanism. To maintain the NACA profile, 0.1 cm
diameter stringers outlined the boundaries of blade and an additional strip of foam
provide a smooth leading edge shape. The multi-component blade is assembled
using a 3D-printed blade mold (Fig. 2.13). Bending stiffness is provided by two
carbon fiber spars and torsional stiffness is from the closed structure. Removing the
need for blade attachment also improved the consistency of the blade design since
the pitching axis location was determined by the spars and did not vary. This had
a previously unaccounted effect of reducing blade twist at higher rotational speeds
and improved cyclorotor performance. Overall, the blade only weighed 6 grams
and provided high bending and torsional stiffness to mitigate effects of centrifugal
loading.
2.5 Cyclorotor Performance
Various blades were tested using Nano17, a commercial force-torque sensor (Fig. 2.14)
and an Ohaus Explorer precision force balance. A systematic performance sweep
was conducted from 800 to 1600 rpm until the cyclorotor reached maximum power
rated for the AXI motor. Figures 2.15 to 2.17 show thrust, power, and power loading
50
Figure 2.14: Nano-17 (force/torque transducer) experimental setup.
Figure 2.15: Cyclorotor thrust vs. rotational speed
51
Figure 2.16: Cyclorotor electrical power vs. thrust.
Figure 2.17: Cyclorotor power loading vs. rotational speed.
52
comparisons between the monolithic and rib-spar blade designs. Each data point is
an average of multiple trials.
The aerodynamic parameters of the blade tested are summarized in Table 2.2.
Both cyclorotors produced 270 grams of thrust at 1600 rpm. With each cyclorotor
weighing 27 grams, the thrust-to-weight (T/W) ratio is approximately 10 at the
operating rotational speed. Slight variation of power consumption was observed
between the two designs. At the operational rpm, the monolithic blade design
consumed 71 watts of electrical power while the rib-spar design consumed 67 watts.
Because of blade construction, the monolithic design requires two 3D printed blade
attachments that connect the blade to the endplates. The rib-spar design uses a
carbon fiber spar inside the blade at the pitching axis to connect to the endplates.
The reduction in power can be attributed to lower parasitic drag. The discrepancy
in power resulted in a 18% increase in power loading from the monolithic design.
Overall, both blade designs were implemented on a cyclocopter. The 550 g twin-
cyclocopter uses the monolithic blade design while the quad-cyclocopter discussed
in Chapter 7 uses the rib-spar design with a slight modification of the pitching axis
location.
2.6 Nose Rotor
The nose rotor used on the cyclocopter was selected through series of experiments
comparing power loading of commercially available propellers. The propellers are
directly driven and mounted on the eRC BL400 brushless motor shaft. Five dif-
53
Figure 2.18: Commercial propellers tested for the nose rotor.
Figure 2.19: Setup to measure thrust produced by nose rotor.
54
Figure 2.20: Nose rotor CT vs. CP .
ferent plastic propellers with variations in diameter, pitch, and number of blades
were tested (Fig. 2.18). A 1045 carbon fiber propeller was also considered, but not
included in the results because of additional mounts required to attach the propeller
to the motor.
Throughout the experiment, thrust was measured using the Ohaus Explorer
precision force balance (Fig. 2.19). The nose rotor was mounted on top of a stand-
off tower for enough clearance to provide unobstructed airflow to the rotor. The
propeller was also inverted such that downwash was directed upwards, away from
the force balance, to prevent any potential ground effects. The motor was driven
using AFRO 12 amps multi-rotor electronic speed controller and powered by a BK
Precision DC Regulated Power Supply at 11.2 V. A reflective marker was placed on
the outer casing of the motor in order to measure the operating rotational speed
55
Figure 2.21: Nose rotor thrust vs. rotational speed for 9050-3 propeller.
Figure 2.22: Nose rotor electrical power vs. thrust for 9050-3 propeller.
56
Figure 2.23: 550g and 60g twin-cyclocopters.
using the Monarch Laser Tachometer.
Propellers were tested at various rotational speeds that produced thrust from
50 to 350 grams at increments of 50 grams. Based on CT vs. CP measurements,
the 9050 3-bladed propeller was selected for the 550 g cyclocopter (Fig. 2.20). Fig-
ures 2.21 and 2.22 show the thrust and electrical power measurements for the selected
propeller. The nominal contribution of the nose rotor is 33% of vehicle weight, but
the actual thrust is dictated by the pitch trim in hover. In-flight measurement of
nose rotor operating condition showed the rotational speed to be around 5000 rpm,
producing 250 g to compensate for pitching moment from the cyclorotors.
2.7 Scalability
The lightweight and aerodynamically efficient multi-component rib and spar
blade design made it possible to develop a 60 g meso-scale cyclocopter (Fig. 2.23).
The vehicle has a lateral dimension of 0.178 meters (7 inches), longitudinal dimension
of 0.127 meters (5 inches), and a height of 0.152 meters (6 inches). Each cyclorotor
57
Table 2.3: Blade properties (60g-scale cyclocopter).
Parameter Blade #1 Blade #2 Blade #3
Weight (g) 0.50 1.50 0.45
Chord-to-radius (c/R) 0.57 0.47 0.47
Pitching axis (x/c) 0.25 0.50 0.25
Chord (in) 0.85 0.70 0.70
Span (in) 1.7 2.0 2.0
Virtual camber (%c) 7 5 5
Virtual incidence (deg) 8 6 6
has a diameter of 0.076 meters (3 inches), blade span of 0.043 meters (1.70 inches),
a chord of 0.022 meters (0.85 inches), and uses a NACA 0015 airfoil. The blades on
the cyclorotors achieve symmetric pitching with a maximum pitching amplitude of
45◦.
Three different blades with the same planform area, but variations in c/R ratio,
blade geometry, and structure were tested for maximum power loading. Blade #1
weighs 0.50 g, has a lower AR of 2 (c = 0.85 in, b = 1.70 in), and a higher c/R of 0.57.
Blade #2 uses the initial design and weighs 1.50 g with an aspect ratio of 2.86 (c =
0.70 in, b = 2.0 in) and c/R ratio of 0.47. To observe the effects of blade structure
on performance, a newly designed blade (#3) with identical blade geometry and
c/R was created and weighed approximately 0.45 g. The virtual chamber for blade
#1 is approximately 7% of the blade chord while 5% of chord for blades #2 and
#3. In addition, blade #1 experiences an 8◦ virtual incidence while blades #2 and
#3 experience 6◦ of incidence. The pitching axis location (x/c) is 0.25 for the new
blade designs and x/c of 0.50 for the old design. Table 3 provides a full summary
58
Figure 2.24: Thrust vs. rpm for the 60g cyclorotor blades.
of all the blade properties.
To determine the optimum blade design, both thrust and power loading were
systematically measured for each blade from 1500 to 3000 rpm in increments of 100
rpm. When comparing the thrust, increased virtual chamber and incidence (blade
#1) resulted in a higher thrust value at each rpm (Fig. 2.24). Interestingly, even
though blades #2 and #3 had the same blade geometry, the lighter 0.45 g blade (#3)
produced more thrust at higher rotational speeds. One possibility for discrepancies
in rotor performance could be that blade #3 experiences a lower centrifugal loading
compared to the heavier 1.50 g blade since centrifugal force varies linearly with mass
and square of blade velocity (mV 2/R). However, previous studies have shown that
moving the pitching axis location closer to the leading edge steadily increases the
thrust produced by the cyclorotor [31]. When comparing the power loading, the
59
Figure 2.25: Electrical power loading of old design vs. new blade design.
optimum pitching axis location should be between 35% to 45% of blade chord at
higher disk loadings (thrust/disk area).
A comparison of power loading (thrust/electrical power) shows an increase of
up to 35% of aerodynamic efficiency from blade #2 to blade #1 (Fig. 2.25). The
trend lines from these experiments agree with the previous studies on the effects of
flow curvature on the cyclorotor. While increasing the c/R ratio appears to improve
rotor efficiency, the overall blade geometry is also restricted by weight. There is a
noticeable difference in efficiency between blades #2 and #3 caused by an increase
in blade mass which proportionally leads to an increase in the centrifugal loading.
However, the c/R ratio seems to have the most significant impact on cyclorotor
performance.
Based on the results of the parametric study, the optimized cyclorotor on the
60
60 g twin-cyclocopter utilizes four 0.50 g blades with the new blade design, c/R
ratio of 0.57, and pitching axis location at 25% of blade chord. Along with other
structural components, the optimized cyclorotor weighs 7 g and has a thrust-to-
weight (T/W) ratio of approximately 3 at the operating rotational speed of 3000
rpm (T = 23 g). This is a significant improvement to previous design iterations at
the meso-scale as evidenced by thrust and power loading trends.
2.8 Summary
The 550 g twin-cyclocopter described in this chapter relied on two cyclorotors as
main source of propulsion and an additional horizontal anti-torque nose rotor. The
vehicle had a lateral dimension of 0.381 meters (1.25 feet), longitudinal dimension of
0.457 meters (1.5 feet), and a height of 0.305 meters (1 foot). Each cyclorotor had
a diameter of 0.152 meters (6 inches), blade span of 0.171 meters (6.75 inches), and
a chord of 0.051 meters (2 inches). In addition, the nose rotor used a three-bladed
9050 propeller with a 0.229 m diameter (9 inches). Each rotor was independently
driven by a motor and two additional rotary servos were used to provide thrust
vectoring capability for the cyclorotors.
Since the cyclorotors comprised a significant portion (23-34%) of the empty
weight fraction of the vehicle, its blades were designed to be lightweight with high
bending and torsional stiffness. Cyclorotors using Mylar wrapped monolithic blades
and multi-Component rib-spar blades were tested. Both cyclorotors produced 270
grams of thrust at 1600 rpm, achieving a thrust-to-weight (T/W) ratio of approx-
61
imately 10. Slight variation of power consumption was observed between the two
designs because of differences in parasitic drag due to blade construction. At the
operational rpm, the monolithic blade design consumed 71 watts of electrical power
while the rib-spar design consumed 67 watts, resulting in a 18% increase in power
loading. Overall, both blade designs were implemented on the final cyclocopter.
62
Chapter 3: Avionics and Control System Development
3.1 Overview
Micro air vehicles are generally constrained by their empty weight fraction which
comprises of vehicle structure, power source (e.g batteries), electronics, and onboard
avionics. The previous chapter discussed the vehicle development of the cyclocopter
and provided an overview of structural design, electronics, and selected LiPo bat-
teries. In order to achieve stable flight, the vehicle must also be equipped with
avionics and a closed-loop feedback system. Avionics include autopilot, sensors,
and an onboard computer for executing computationally intensive tasks such as im-
age processing. This chapter will first review two custom autopilots implemented
on the cyclocopter and then survey sensors and onboard computers commercially
available for MAVs.
Sensors can be categorized into three main applications: (1) flight stability
and control, (2) flight performance diagnostics, and (3) guidance and navigation.
Cyclocopters developed in this work relies on a combination of (1) and (2). Sensors
used for flight stability and control provide measurements of vehicle attitude angle
and body rates. Sensors for flight performance diagnostics provide both vehicle state
(e.g RPM, altitude) and atmospheric conditions (e.g pressure, gust, temperature).
63
These sensors can be used to collect mission-specific atmospheric data, as well as
diagnose vehicle performance issues during flight. Finally, guidance and naviga-
tion sensors enable autonomous navigation using measurements of vehicle position,
velocity, attitude, and trajectory of path.
Figure 3.1: Components of an autopilot.
3.2 MAV-scale Autopilot
While structural components and power systems can be customized for different
vehicle configurations, the choice of avionics is often limited to commercial avail-
ability. In addition, only few autopilots are open-source and permit developers to
modify original source code to adapt to multiple vehicle configurations. An au-
topilot is generally comprised of a microprocessor with built-in memory (FLASH,
RAM, etc.), input and output peripherals such as serial ports, wireless communi-
64
cation components, and an assortment of sensors (Fig. 3.1). The following sections
compare two light-weight and open-source MAV-scale autopilots implemented on
the cyclocopter. Physically, these particular autopilots are only integrated with an
inertial measurement unit (IMU) onboard. However, sensors for flight performance
measurement and navigation can be supplemented.
Figure 3.2: Comparison between MAV-scale autopilot used on the cyclo-
copter.
3.2.1 GINA Mote
GINA Mote is an existing autopilot weighing 1.6 grams and originally developed by
the University of California, Berkeley [48]. The board is relatively small (25mm x
25mm) and houses IMU and wireless communication components (Fig. 3.2). The
principal components of this board are a TI MSP430 microprocessor for onboard
computation tasks, ITG3200 tri-axial gyros, KXSD9 tri-axial accelerometer, and an
ATMEL radio and antenna for wireless communication. The wireless communication
has a latency less than 20-30 milliseconds such that the critical inner loop feedback
occurs at an update rate of 3 milliseconds.
65
Table 3.1: Comparison between GINA Mote, ELKA, and ELKA-R [78].
GINA Mote ELKA ELKA-R
CPU TIMSP430 ARM Cortex-M3 ARM Cortex-M4
Clock Speed (MHz) 16 72 168
Update Rate (Hz) 333 500 1000
Bits 16 32 32
Flash Memory (KB) 116 128 1024
Actuators 6 8 12
Weight (g) 1.6 1.3 1.7
Ground Control Station Required Optional Optional
Breakout Board Required Required Optional
The 16-bit microcontroller supports attitude stabilization at 333 Hz and was
able to sufficiently stabilize the cyclocopter in hover. Due to limited flash memory
(116 KB), most of the feedback control system computations were performed off-
board on a nearby ground control station running LabVIEW. For example, the final
actuator commands (PWM) were computed on LabVIEW after receiving vehicle
attitude and body rates from the autopilot. Communication between the autopilot
and remote station is achieved using a wireless IEEE 802.15.4 data link. Updates
to the autopilot firmware are contributed by the original developers and others
using the GINA Mote’s open-source software. In addition, pilot inputs were directly
transmitted to the ground control station through a serial connection via an Arduino
Uno.
3.2.2 ELKA
Due to minimal updates in the past few years and inherent limitations on attitude
stabilization rate and inability to perform onboard computations, the GINA Mote
66
was unable to sufficiently stabilize the cyclocopter beyond hover. As a result, a cus-
tom autopilot with higher computation power, flash memory, and updated sensors
was developed in-house for the cyclocopter (Table 3.1).
The Embedded Lightweight Kinematic Autopilot (ELKA) weighs 1.3 grams
and is powered by a single 1-cell 3.7 volt 125 mAh Li-Po battery [78]. ELKA
houses a STM32 microprocessor with a 32-bit ARM Cortex-M3 core for high-end
onboard computation tasks. The InvenSense MPU-9150 IMU integrated on the
board includes tri-axial gyroscopes, accelerometers, and magnetometers. Wireless
communications are serviced by an onboard nRF24L01 chip, a low-power 2.4 GHz
RF transceiver. ELKA has a sensor update rate of 500 Hz and is capable of streaming
vehicle attitude data and actuator controls data to a remote base station with a 30
milliseconds latency.
ELKA was also refined (ELKA-R) with a faster ARM Cortex-M4 processor
that doubled the clock speed and update rate of the inner-loop (Table 3.1). With
only an increase of 0.4 grams, ELKA-R is able to support up to 12 actuators. Unlike
GINA Mote and ELKA, ELKA-R does not require an additional breakout board to
connect to actuator pins. The flash memory was also increased by 8 times (1024
KB) and enabled all computations to be handled onboard. This capability provided
the option of operating the cyclocopter without a ground control station for outdoor
flight tests.
67
Figure 3.3: Notation for body rates and attitude of the cyclocopter.
3.3 Sensors for Flight Stability and Control
Without additional attitude stabilization, the cyclocopter was found to be unstable
due to limited inherent aerodynamic damping [39]. At the MAV-scale, instability is
further exacerbated by low vehicle inertia and faster dynamics. Artificial damping
can be achieved using arrays of sensors integrated with a closed-loop feedback sys-
tem. Inertial sensors used for stability augmentation include gyroscopes, accelerom-
eters, and magnetometers. Raw measurements from these sensors are typically prone
to drift and sensor noise. However, there are different filters and algorithms that
are widely applied to achieve accurate estimates. The following sections provide a
comprehensive overview of state measurement and attitude estimation techniques.
3.3.1 Inertial Measurement
Many of the recent advancements in development of MAVs can be attributed to
microelectromechanical system (MEMS) sensors. MEMS sensors are manufactured
68
using a combination of semiconductor and microfabrication technologies to create
a silicon substrate housing electronic and mechanical components. MEMS sensors
are typically integrated into a single IMU that can measure linear acceleration along
the X-Y-Z axes (ax, ay, az), angular velocities (p, q, r), and attitude (φ, θ, ψ). The
resulting 9-DOF IMU houses tri-axial accelerometer, gyroscope, and magnetome-
ter. As mentioned in the previous section, the IMU used on the cyclocopter is the
InvenSense MPU-9150 integrated on ELKA-R.
MEMS-based accelerometer measures static acceleration due to gravity along
X, Y, and Z axes in the body frame and also dynamic acceleration due to mo-
tion or vibrations. Similar to an inclinometer, the rotation angle and individual
component of gravity along X, Y, and Z axes can be calculated if the sensor is
tilted while stationary in space (Fig. 3.3). The actual sensing mechanism relies on
micro-machined capacitive devices that compute changes in capacitance of a moving
mass. The measurement is then converted to a digital signal via an analog-to-digital
converter (ADC). The measured acceleration is integrated with respect to time in
order to compute linear velocity and position. As a result, any constant bias error
in acceleration leads to a linear error in velocity and quadratic error in position.
Accelerometers are also sensitive to vibrations, especially at high frequencies.
MEMS-based gyroscope measures angular velocity of the vehicle along all axes
in the body frame. When the gyroscope is rotated, a resonating mass inside dis-
places because of the Coriolis force. The displacement shifts the capacitance on the
electrodes. Similar to accelerometers, gyroscopes output low-current electrical sig-
nals that are converted to digital signals. Due to fabrication techniques and dynamic
69
Figure 3.4: Measurement from magnetometer showing large bias errors
[50].
physical properties of MEMS, gyroscopes experience drift and noise in measurement
over time [51]. Techniques to reduce measurement errors are discussed in the later
sections.
Aside from integrating gyroscope measurement along the Z-axis, vehicle’s ori-
entation can also be measured using a magnetometer. Magnetometers detect mag-
netic field along X, Y, and Z axes and provide relative orientation with respect to
the Earth’s magnetic north. The sensor outputs voltage that is proportional to the
magnitude and polarity of the detected magnetic field. While magnetometers are
not subject to drifts in measurement, they are affected by electromagnetic inter-
ference from motors (Fig. 3.4). Due to large bias errors, measurements from the
magnetometer were not used for attitude estimation.
70
3.3.2 Attitude Estimation
Raw signals from tri-axial gyroscope and accelerometer implemented on the cy-
clocopter output linear accelerations (ax, ay, az) and angular velocities (p, q, r) in
the vehicle’s body frame B = (xB, yB, zB) (Fig. 3.3). However, attitude angles
(φ, θ, ψ) are typically provided with respect to the fixed frame or inertial frame,
I = (xI , yI , zI). Conversion between inertial and body frame is calculated using a
conventional 3-2-1 rotation sequence using Euler angles [φ, θ, ψ]. The first rotation is
about zB by yaw angle (ψ), second rotation about yB by pitch angle (θ), and finally
the third rotation about zB by roll angle (φ). Rotational transformation between
the two frames is summarized by the following [52]:
 
 cos θ cosψ cos θ sinψ − sin θ
[R ] = 

BG 

sinφ sin θ cosψ − cosφ sinψ sinφ sin θ sinψ + cosφ cosψ sinφ cos θ 
cosφ sin θ cosψ + sinφ sinψ cosφ sin θ sinψ − sinφ cosψ cosφ cos θ
(3.1)
The angular velocities are calculated using:
  φ̇ 
  
  1 sinφ tan θ cosφ tan θ  

  
 p 
 θ̇  =  0 cosφ −sinφ  q  (3.2)
ψ̇ 0 sinφ sec θ cosφ sec θ r
From the transformation matrix, it is evident that singularity occurs when
71
θ = ±90◦. The transformation matrix is a square matrix (n) with orthonormal
columns and generally referred to as the special orthogonal group (SO(n)). The
inverse transformation (RGB) can be calculated by taking a transpose of the matrix.
3.3.2.1 Quaternion Representation
One disadvantage of relying on Euler angles is existence of singularities, which can
lead to gimbal lock or loss of a degree of freedom. For an aircraft, singularity at
pitch attitude of 90◦ (nose up or nose down) is consequential. To avoid this issue, a
quaternion-based attitude representation is implemented on the onboard autopilot.
A quaternion is a point on the four dimensional hypersphere and is governed by the
following equations where i, j, and k denote the complex numbers [54]:
i2 = j2 = k2 = ijk = −1 (3.3)
ij = k = −ji (3.4)
jk = i = −kj (3.5)
ki = j = −ik (3.6)
The four parameters (q0, q1, q2, q3) represent quaternion rotation about an ar-
bitrary axis. The parameter, q0, is the scalar component while q1, q2, and q3 are the
distance along i, j and k. For a particular orientation, the relationship between the
four parameters is given by:
72
q2 + q20 1 + q
2
2 + q
2
3 = 1 (3.7)
Quaternions can be written in terms of Euler angles:
ψ θ φ ψ θ φ
q0 = cos cos cos + sin sin sin (3.8)
2 2 2 2 2 2
ψ θ φ ψ θ φ
q1 = cos cos sin − sin sin cos (3.9)
2 2 2 2 2 2
ψ θ φ ψ θ φ
q2 = cos sin cos + sin cos sin (3.10)
2 2 2 2 2 2
ψ θ φ ψ θ φ
q3 = − cos sin sin + sin cos cos (3.11)
2 2 2 2 2 2
Quaternion rates can also be expressed in terms of angular rates in the body frame:
1
q̇0 = − (q1p+ q2q + q3r) (3.12)
2
1
q̇1 = (q0p+ q2r + q3q) (3.13)
2
1
q̇2 = (q0q + q3p+ q1r) (3.14)
2
73
1
q̇3 = (q0r + q1q + q2p) (3.15)
2
Conversion between quaternions and Euler angles is typically performed when
selecting desired pitch, roll, or yaw angle through the ground control station. Euler
angles are intuitive and easier to translate into a physical representation. However,
because of the singularities, state estimation using gyroscopes and accelerometer is
performed using quaternions.
3.3.2.2 Gyroscope
Angular velocities (ω) from the gyroscope in the body frame are converted to the
quaternion derivative (q̇k) describing the rate of change between the earth (E) and
body frame (B) using the following equations [53,54]:
  0 

ω
ω =  x 


(3.16)
 ωy 
ωz
1
q̇k = qk−1 ⊗ ωk (3.17)
2
qk = qk−1 + q̇k 4 t (3.18)
The quaternion derivative (Eq. 3.17) is computed at discrete time steps (k =
74
0, 1, 2...n). The quaternion orientation (Eq. 3.18) from the gyroscope is then esti-
mated by integrating the quaternion derivative at the sampling period, 4t. The ⊗
symbol represents quaternion multiplication where:
  q 0
p0 − q1p1 − q2p2 − q3p3
 

  q0p1 − q1p0 − q  2
p3 − q3p2 
q ⊗ p =   (3.19) q0p2 − q1p3 − q2p0 − q 3p1 
q0p3 − q1p2 − q2p1 − q3p0
3.3.2.3 Accelerometer
As previously mentioned, integrating gyroscope measurements leads to errors that
increase over time. Correcting the gyroscope drift involves fusing sensor mea-
surements with the accelerometer. The accelerometer measures the acceleration
(ax, ay, az) in the body frame, but does not provide the rotation of the body frame
relative to the earth frame. Therefore, a unique orientation is calculated by taking
the shortest rotation between the gravity vector in the earth frame (g) and accel-
eration in the body frame (a). To avoid trigonometric functions, the rotation is
calculated by minimizing an optimization function for k = 1,2...n:
f(qk, g, ak) = q
∗
k ⊗ g ⊗ qk − ak (3.20)
where q∗ is the quaternion conjugate and the accelerometer measurements are:
75
  0    a a =  x 
 (3.21)
 ay 
az
The solution to the objective function (Eq. 3.20) is found using the gradient
descent algorithm. In the gradient descent algorithm, the quaternion orientation
estimate (q̂k+1) is based on an initial estimate (q̂0) and a step-size of µ:
− Of(qk, g, ak)q̂k+1 = q̂k µ (3.22)‖f(qk, g, ak)‖
The gradient of the objective function (Of(qk, g, ak)) is defined in terms of its Ja-
cobian (J) to compute the direction of error on the solution surface:
Of(qk, g, ak) = J
T (qk, g)f(qk, g, ak) (3.23)
Since the gravity vector in the earth frame only has a Z component, the objective
function and Jacobian can be simplified to:
 
 2(q2q4 − q1q3)− ax 
f(q , a k k) =  2(q1q2 + q3q4)− a  (3.24)y
2(1 − q22 − q23)− a2 z
76
 
 −2q 3
2q4 −2q1 2q2 
J(qk) =  2q2 2q −2q 2q  (3.25)1 4 3
0 −4q2 −4q3 0
The optimization problem requires multiple iterations (k) for each estimated
orientation and sensor measurements. However, it can also be solved through a
single iteration per sample time if the convergence rate is equal to or greater than
rate of change of the physical orientation.
Figure 3.5: Sensor suite for flight performance diagnostics and atmospheric
measurement.
77
3.4 Sensors for Flight Performance Diagnostics
Whether operating autonomously or by a pilot, vehicle’s flight performance and
operating conditions can be monitored using an array of sensors. Figure 3.5 shows
the sensor suite integrated on a commercial data logging system. The following
sections cover sensors used to measure vehicle’s state and atmospheric conditions.
Sensors for airspeed measurement will be covered in Chapter 6 along with a detailed
overview of a custom sensor developed for the cyclocopter gust study.
3.4.1 RPM Sensors
RPM sensors that are non-contact can be categorized into electrical, magnetic, and
optical sensors. Electrical RPM sensors measure back electromotive force (EMF)
feedback from motors to determine approximate speed. Back EMF is commonly
measured as the voltage opposing the induced current flow resulting from the move-
ment of motor coils in the magnetic field. As motor RPM increases, back EMF
also proportionally increases in order to oppose the changing magnetic field. Since
back EMF is equivalent to the motor RPM and voltage constant, it can be used
to directly calculate RPM. However, certain combinations of motors and electronic
speed controllers (ESCs) can lead to issues with properly syncing these sensors.
Hall effect sensors operate by measuring changes in the magnetic field. The
sensor is composed of a thin piece of rectangular semi-conductor (e.g gallium ar-
senide) that carries a continuous current. When placed within a magnetic field,
electrons in the semi-conductor are influenced by the Lorentz force and deflect from
78
their linear path to either sides of the semi-conductor. The resulting magnetic flux
creates a voltage differential called the Hall Voltage. A hall effect sensor placed near
a rotating motor equipped with a small magnet on its outer casing can measure the
variations in magnetic field and estimate the RPM. Existing hall effect sensors are
inexpensive, compact, and highly reliable. These sensors are also able to operate
under severe conditions (e.g outdoors, temperature variations, dusty setting, etc.).
Optical RPM sensors, such as tachometers, transmit infrared light beam and
measure rate at which light beam is reflected. The laser beam is pulsed against a
rotating element that has a reflective surface. Because of its reliance on infrared
light, optical RPM sensors are limited to indoor applications. Similar mechanism is
also used to measure proximity to nearest object.
3.4.2 Proximity Sensors
Proximity sensors are used to measure altitude and distance to nearest object for
sense-and-avoid capabilities. These sensors are typically light-weight and low-cost
since they only require a source and a receiver. Proximity sensors available for
MAVs are typically for non-contact detection and emit either ultrasonic or infrared
light to measure distance. A high performance ultrasonic range finder, such as the
Maxbotix MB1000, has a 2.54 cm resolution, wide range from 15.24 cm to 645.16
cm, and a 20 Hz update rate. High-frequency ultrasonic pulses are transmitted
back-and-forth using a single transducer in order to relay proximity information.
Distance is calculated using lapsed time between sending and receiving reflections
79
from the sound waves. Ultrasonic range finders can be used in low lighting conditions
and are unaffected by dark colors and dust. However, these sensors are not suitable
for applications on soft materials (e.g carpets) since sound waves are not properly
reflected.
Infrared (IR) range finders use optical sensors and estimate distance using
triangulation. Sensors such as the SHARP GP2Y0A21YK operate using IR LEDs
that emit a narrow light beam and measure distance between 10 cm to 80 cm. Unlike
ultrasonic sensors, infrared range finders require both a transmitter and a receiver.
Reflected light waves are received by a position-sensible photo detector (PSD). In
addition, these sensors are restricted to indoor applications due to interference from
sunlight.
3.4.3 Barometer
Another method of estimating altitude is by using a barometric pressure sensor
(barometer). A barometer measures atmospheric pressure and estimates altitude
based on a model relating altitude to pressure. MEMS-based barometers, such as
the STMicroelectronics LPS25H, contain a suspended membrane inside a mono-
silicon substrate used to measure pressure. The LPS25H is capable of acquiring
high resolution measurements (±1 Pa) with an absolute pressure range of 260 mbar
to 126 0mbar. Barometers are typically calibrated to a standard atmosphere at
mean sea level, which equates to pressure of 1013.25 mbars and temperature of 15◦
Celsius.
80
3.4.4 Temperature
Temperature can be monitored using a non-contact infrared sensor or a contact-
based thermistor. Infrared temperature sensors convert thermal energy (0.7µm to
20µm wavelength) into electrical energy and are limited to indoor applications.
Thermistor is a type of semiconductor that changes its electrical resistance based
on temperature. Compared to resistance temperature detectors (RTD), thermistors
provide a higher resolution because they undergo a higher resistance change per
degree of temperature. Thermistor-based temperature sensors are generally used to
monitor motor or battery temperature during flight.
3.5 Sensors for Guidance and Navigation (Autonomy)
The current sensor suite on the cyclocopter facilitates controlled piloted flight. For
greater autonomy, the vehicle requires guidance and navigation sensors that enable
localization. While inertial sensors like the accelerometer can be used to calculate
displacement by numerically integrating acceleration measurements, the resulting
estimate is often inaccurate and prone to drift. The following sections summarize
prominent localization techniques and sensors currently employed on MAVs.
3.5.1 Global Navigation Satellite Systems (GNSS)
A common approach for outdoor navigation is to rely on real-time geo-spatial posi-
tion triangulated by global navigation satellite systems (GNSS). Current GNSS sys-
tems include the Global Positioning System (GPS) by the United States, GLONASS
81
by Russia, Galileo by the European Union, BeiDou by China, Navigation Indian
Constellation (NavIC) by India, and Quasi-Zenith Satellite System (QZSS) by Japan.
These systems typically rely on medium Earth orbit (MEO) satellites that orbit at
20,000 km.
82
83
Table 3.2: Summary of popular MAV-scale commercial GPS modules.
U-blox LEA-6H U-blox NEO-M8N DJI D-RTK GNSS
Compatible Board Pixhawk/DJI NAZA Pixhawk/DJI NAZA DJI A3 Series
Horizontal Accuracy (m) 2.5 2.5 0.01 + 1 ppm
Vertical Accuracy (m) 0.02 + 1 ppm
Update Rate (Hz) 5 10 10
Interface DDC, UART, USB, SPI DDC, UART, USB, SPI CAN, UART, USB
Min. Battery (V) 3.3 3.3 12
Weight (g) 16.8 32 139.5
Figure 3.6: GPS waypoint following using the Pixhawk GPS module.
These satellites are arranged into six equally-spaced regions such that four are
accessible at any location. The GPS receiver analyzes radio signals from several
of these satellites and determines position, speed, and elevation using trilateration
techniques. Unlike triangulation which uses angles, trilateration uses the intersection
of radial distances (surface for 3-D geometry) between satellites to calculate the
approximate geo-location.
Light-weight and low-cost GPS modules compatible with various flight con-
trollers are widely available. Table 3.2 shows popular commercial GPS modules
used on UAVs. These modules are capable of communicating with multiple GNSS
and interfacing through various serial communication protocols. Due to FAA reg-
ulations, they are programmed to limit the maximum above ground level (AGL)
altitude based on operating location.
84
Figure 3.7: Variation in number of reachable satellites in an urban envi-
ronment.
Figure 3.6 shows the autonomous waypoint following trajectory taken by a
quadcopter equipped with Pixhawk and the u-blox LEA-6H module described in
table 3.2. The u-blox GPS was tested outdoors in an open field and as a result,
was able to accurately navigate the quadcopter to the selected waypoints within 2.5
meters of error. Accuracy declines when operating in urban environments composed
of buildings and large structures because of propagation effects of signals diffracting
off the surfaces. As a result, receivers on the GPS modules are unable to discern
the different signals transmitted by satellites (Fig. 3.7). An alternative means of
calculating position indoors and in urban settings is by relying on visual odometry.
3.5.2 Visual Odometry (VO)
Visual odometry (VO) is a specific form of structure from motion (SFM) technique
that estimates real-time pose by identifying visual changes in captured images with
85
Figure 3.8: Landmarks tracked by ROVIO to construct estimated camera
poses in 3-D [57].
respect to a reference frame [56]. VO generates a local position estimate that is
incrementally updated based on detected features. Common VO methods focus on
one particular technique, from feature matching or feature tracking, to optic flow.
Feature tracking identifies the selected feature across all frames, whereas feature
matching only uses consecutive frames. Therefore, the feature tracking technique
results in lower drift in the position estimate.
Instead of using features (e.g edge, color etc.), the optic flow technique relies
on spatio-temporal image gradients or pixel intensities in specific regions between
consecutive frames. Traditional optic flow algorithms such as the Lucas-Kanade
assume that pixels in a local neighborhood have constant intensity, spatial coherence,
and undergo small motions. A least-squares fit is then applied to the neighboring
pixels to obtain their 2-D velocity components (u, v) across consecutive frames.
Overall, VO can be separated into four main parts: (1) image sequence, (2) fea-
ture detection, (3) motion estimation, and (4) local bundle adjustment [56]. Motion
86
Figure 3.9: Camera concepts used for visual odometry.
estimation is typically performed using random sample consensus (RANSAC), an it-
erative method for model fitting using data set containing outliers. Using RANSAC,
transformation between camera images can be recalculated using only inlier points.
Local bundle adjustment is the final step and involves refining 3-D coordinates and
parameters of relative motion.
Several open-source VO software frameworks are available and have been suc-
cessfully demonstrated on aerial platforms [56–58]. Robust Visual Inertial Odometry
(ROVIO) is a framework developed by ETH Zürich that combines visual odometry
with inertial measurements using iterated extended Kalman filter (EKF). The re-
sulting algorithm exhibits high levels of robustness and enables quick start-up by
omitting initialization procedures. Figure 3.8 shows a snapshot of the monocular
camera-based ROVIO extracting landmark data to construct 3-D estimated camera
poses and their distance uncertainties. In terms of hardware, visual odometry can
be implemented using passive, active, and event-based cameras.
87
88
Table 3.3: Comparison between types of cameras for VO.
Advantage Disadvantage
Monocular Inexpensive and light-weight No depth information
Simple calibration
Stereo Depth information (3-D) Complex calibration
Observable image scale Degrades to monocular (2-D) in some cases
Omni-directional Wide field of vision (360◦) Expensive and heavy
Rotational invariance Complex algorithm (synchronization)
Time-of-flight Inexpensive and light-weight Vulnerable to high ambient light
Depth information (3-D) Interference from other ToF devices
Fast processing No auto-focus capability
RGB-D High depth image resolution Expensive and heavy
Provides colored depth points Low frame rate
Operational in low-light conditions Limited to indoor applications (IR)
Event-based Asynchronous (variable time intervals) Requires new computer vision algorithms
Fast visual measurements with low latency No intensity information
High dynamic range Low image resolution
Low computational cost
3.5.2.1 Passive Cameras
Passive cameras rely on an external light source (e.g sunlight) to illuminate the scene
and measure naturally emitted reflections. These cameras are typically simpler
in mechanical design and include monocular, stereo, and omni-directional vision
technologies.
Monocular visual odometry relies on a single camera while stereo uses two or
more cameras to capture multiple images of the same scene. As a result, stereo
visual odometry method provides depth information. The 3-D position of an object
is reconstructed using triangulation by simultaneously tracking features in succes-
sive frames captured by multiple cameras that are separated by known distances.
The algorithm uses a simple pin-hole camera model (Fig. 3.9) where depth (z) is
calculated using measured angles (α, β):
x
z = 1 (3.26)+ 1
tanα tanβ
Complications arise when a corresponding feature or point is not identified by
all cameras. If captured images lack variations in color and intensity (e.g uniformly
colored wall), the stereo vision algorithm would be unable to detect features. Omni-
directional cameras have a wider field of vision (up to 360◦) and provides larger
image frame for feature detection [59]. While they can provide a well refined 3-D
model of the environment, calibrating and synchronizing multiple cameras consume
high computation power. Table 3.3 provides a comparison of different types of
89
cameras typically used for visual odometry.
Figure 3.10: Comparison of data transmission rate between frame-based
and event-based cameras [60].
3.5.2.2 Active Cameras
While passive cameras use natural light, active cameras illuminate the images with
artificial light to measure depth. In particular, time-of-flight (ToF) cameras use a
CMOS pixel array and a solid-state laser or LED in the near-infrared range [61]. The
light source is pulsed as a sinusoidal electromagnetic wave with constant amplitude
and frequency. Between the pulses, the reflected light is detected by an imaging
sensor and converted from photonic energy to electrical current (Fig. 3.9). One
disadvantage of using ToF cameras is that the reflections can be distorted by high
ambient light. As a result, measurements have reduced signal-to-noise ratio, which
diminishes the accuracy of depth calculation.
Instead of modulated light, RGB-depth (RBG-D) cameras, such as Microsoft’s
90
Kinect, use an infrared structured light that operates by projecting a pattern onto
the scene (Fig. 3.9) and then analyzing the pattern distortion to calculate depth [62].
Typical RGB-D cameras use a digital light processing (DLP) projector and a high
resolution color camera. The resulting measurement has high depth image resolution
and also colored depth points (colored point clouds). Since the structured light
algorithm relies on detecting pattern disparities between frames, it operates at a
lower frame rate compared to ToF sensors.
Figure 3.11: Captured events and projected gradient map [63].
3.5.2.3 Event-Based Cameras
A significant limitation of frame-based cameras such as ones described in previous
sections is that frames are captured at fixed time intervals [64]. Each frame cap-
tures all the pixels in order to track features, which results in an inherent limitation
on data rate and latency. Bio-inspired event-based cameras such as the Dynamic
Vision Sensor (DVS) transmit asynchronously where each event is only registered
when the sensor detects change in the image (Fig. 3.10). Instead of capturing stan-
91
dard intensity frames, DVS detects disparities in pixel-level brightness. In general,
conventional cameras record the scene while DVS records how the scene changes.
Figure 3.11 shows example of captured events and the projected gradient map of
the scene. Compared to conventional cameras, event-based cameras have latency in
the order of microseconds, high dynamic range (130 dB vs. 60 dB), and consume
less power [60].
Since the event-based camera captures events instead of conventional frames,
it requires new computer vision algorithms. In the past few years, new algorithms
have been developed for visual-inertial odometry (VIO) and 3-D visual simultaneous
localization and mapping (SLAM) using the DVS128 camera. Using a single event-
based camera, these algorithms are able to quickly track 6-DOF camera motion and
reconstruct 3-D scenes [65].
3.5.3 Simultaneous Localization and Mapping (SLAM)
While VO and VIO incrementally estimate the local pose (position and orientation)
of an object using camera images, SLAM involves estimating the global pose of the
vehicle on a map of the environment that is simultaneously generated. With each
new measurement, the estimated pose and map of environment are incrementally
updated. However, it should be noted that visual SLAM incorporates VO and
VIO into the algorithm to estimate pose from matched features between images.
Main components of SLAM include: (1) sensor measurement, (2) landmark/feature
extraction, (3) data association, (4) state update, and (5) state prediction, where
92
the update and prediction steps form a closed loop [56].
Fundamentally, the concept can be defined as a probabilistic problem that
is solved using Bayes’ Theorem [66]. Physical implementation of SLAM requires
observation and motion models to computationally calculate the probabilities. Ex-
isting statistical techniques used to approximate the SLAM solution can be catego-
rized into filtering approaches and smoothing approaches. Filtering approaches solve
SLAM real-time with each available sensor measurement and typically rely on an
Extended Kalman Filter (EKF-SLAM) or a particle filter. Smoothing approaches,
such as GraphSLAM and RBG-D SLAM, use least-squares error minimization to
solve the full SLAM problem
93
94
Table 3.4: Existing LIDAR sensors for UAVs.
Leddertech Vu8 (Wide) Velodyne VLP-16 Velodyne HDL-32E
Weight (g) 129 830 1300
Range (m) 34 100 80–100
Accuracy (cm) ±5 ±3 ±2
Horizontal FOV (◦) 100 360 360
Angular Resolution (◦) 0.25 0.1–0.4 1.33
Dimensions (mm) 73(H) x 40(W) x 65(D) 103(Diam) x 72(H) 86(Diam) x 150(H)
Typical Power Consumption (W) 2 8 24
Cost (USD) 450 7999 29,900
over the entire map. These solutions rely on sensor measurements from various
sources such as Light Imaging And Detection (LIDAR) scanners and cameras.
3.5.3.1 LIDAR
A LIDAR scanner operates by pulsing laser beams and analyzing the reflected light
to measure traveled distance. Since the sensor uses lasers with a specific wavelength
(typically 905 nm), it avoids interference from other sources [66]. Compared to
cameras discussed in the previous section, LIDAR sensors have a wider field of view
(FOV) and operate at high frame rates. Table 3.4 summarizes existing LIDAR
sensors suitable for UAVs due to their weight and capabilities. These sensors are
capable of taking high accuracy measurements (±5 cm) with considerable range of
up to 100 meters. Upgrading from Leddertech Vu8 to Velodyne VLP-16 in order
to acquire a full 360◦ FOV amounts to a significant increase in weight, quadrupled
power consumption, and more than 17 times the increase in cost.
3.5.3.2 Visual SLAM (V-SLAM)
Recent advancements in CPU and GPU technologies have resulted in surge of
camera-based SLAM algorithms, referred to as visual SLAM (V-SLAM). Visual
SLAM has been demonstrated using monocular, stereo, omni-directional, RGB-D,
and even event-based cameras. Table 3.5 compares existing V-SLAM algorithms
ordered by sensor type and map density method. A sparse map density algorithm
only uses a small region of pixels while a dense map algorithm uses most or all the
95
96
Table 3.5: Comparison of available visual SLAM framework [67,68].
PTAM ORB-SLAM VISLAM ORB-SLAM2 DTAM
Sensor Monocular Monocular Stereo Monocular, Stereo, RGB-D RGB-D
Method (Direct/Indirect) Indirect Indirect N/A Direct Direct
Map Density Sparse Sparse N/A Semi-dense Dense
Global Optimization Yes Yes N/A Yes No
Loop Closure No Yes N/A Yes No
pixels in an image frame. The resulting map from sparse methods are point clouds
while dense methods provide a detailed scene. As expected, dense methods require
higher computation power and typically run on GPUs. Indirect methods such as
PTAM and ORB-SLAM extract features first and then use the features to localize
while direct methods use pixel intensities directly instead of features. To reduce
sensor drift, algorithms such as ORB-SLAM and ORB-SLAM2 employ loop clo-
sure techniques where previously detected features are used to update the current
pose estimate. While most existing algorithms are open-source and developed by
public institutions, emerging proprietary algorithms such as VISLAM developed by
Qualcomm are becoming accessible.
3.6 MAV-Scale Onboard Computer
VIO and SLAM algorithms for autonomous navigation require computation power
beyond capabilities of autopilots discussed in the previous section. MAVs must
instead rely on onboard computers with dedicated processors (CPU, GPU) and
adequate data processing power. A significant constraint for selecting onboard com-
puters is limitation on weight. Table 3.6 compares commercial off-the-shelf (COTS)
processors available for MAVs due to their light weight (≤100g) and compact size.
3.6.1 Qualcomm Snapdragon Flight
Qualcomm Snapdragon Flight board serves as both an onboard computer and an
auto-pilot because of its integrated sensors and stereo camera. The current version
97
98
Table 3.6: Comparison of available light-weight onboard computer [69].
Snapdragon Flight ODROID XU-4 NVIDIA Jetson TX2
Weight (g) 26 65 85
Dimensions (cm) 7.9 x 6.1 x 0.8 9.4 x 7.0 x 1.8 8.6 x 5.1 x 0.5
CPU Krait (2.26GHz) Exynos5422 Cortex-A15 (2GHz) ARM Cortex-A57 (2GHz)
+ Cortex-A7 (octa-core) + NVIDIA Denver2 (2GHz)
GPU Adreno 330 (578MHz) ARM Mali-T628 MP6 (600MHz) 256-core Pascal (1300MHz)
DSP QDSP6 V5A (801MHz) - -
Operating System Linaro Linux Linux LTS Linux4Tegra
Integrated Sensors Yes No No
Integrated Camera Yes No No
Cost (USD) 400 59 500
of the board is built on the Snapdragon 801 series processor commonly found in
smartphones. Weighing only 26 grams, the board is integrated with a quad-core 2.26
GHz central processing unit (CPU), a dedicated digital signal processor (DSP), and
Qualcomm’s 378 MHz graphics processing unit (GPU) that runs their proprietary
VISLAM algorithm. VISLAM uses the 4K high-resolution Sony IMX135 forward-
facing camera and the downward-facing Omnivision OV7251 optic flow camera.
Additional sensors include the Telit Jupiter SE868 V2 GPS module, Invensense
MPU-9250 IMU, and the Bosch BMP280 barometer. Researchers have previously
demonstrated autonomous hover using the Snapdragon Flight kit interfaced with
the ELKA autopilot [69, 70]. While the Snapdragon Flight 801 series was recently
discontinued, a new version armed with the Snapdragon 802 series processor will be
introduced in the near future.
3.6.2 ODROID XU-4
ODROID XU-4 is an inexpensive board-only alternative, equipped with two CPUs
(Samsung Cortex-A15 and Cortex-A7 octa-core) and a GPU (Mali-T628 MP6).
With open-source support, the board is capable of running multiple Linux operating
systems such as Ubuntu 16.04, Android 4.4 Kitkat, 5.0 Lollipop, and 7.1 Nougat.
ODROID also offers variety of boards with different processing power, from the
Exynos5 Octa series to ESP32 series. Researchers have previously supplemented a
Samsung Galaxy S5 smartphone with ODROID XU-4 to demonstrate autonomous
navigation [71].
99
3.6.3 NVIDIA Jetson TX2
The NVIDIA Jetson TX2 is integrated with the 64-bit hex-core ARMv8 CPU com-
plex, which is a stack of ARM Cortex-A57 (quad-core, 2GHz) and NVIDIA Denver2
CPUs. With its 256-core NVIDIA Pascal GPU and NVIDIA cuDNN and TensorRT
libraries, the board supports deep neural network (DNN) architectures, recurrent
neural networks (RNNs), long short-term memory networks (LSTMs), and online re-
inforcement learning. Aside from machine learning applications, the Jetson TX2 has
been integrated with the Bebop 2 quadcopter kit to demonstrate active vision-based
gap detection for quadcopters [72].
3.7 Telemetry
Whether piloted or autonomous, UAVs maintain remote communication with a
nearby ground control station (GCS). Noncritical tasks with high computational
costs can be processed on the GCS instead of overloading onboard avionics. In
particular, missions such as surveillance and search-and-rescue especially require
real-time data streams and vehicle diagnostics. Telemetry between the cyclocopter
and GCS was enabled through wireless communication.
3.7.1 Wireless Communication (Wi-Fi)
Communication between ELKA and the ground control station is achieved via Lab-
VIEW interface through a nRF wireless transceiver module (Fig. 3.12). The board
receives pilot inputs through a commercial Spectrum receiver which independently
100
Figure 3.12: Telemetry implemented for the cyclocopter.
connects to the transmitter through a wireless 2.4 GHz radio link. A separate
connection between the Spectrum receiver and transmitter ensures that the pilot
will still have command over the vehicle if connection is lost to the ground control
station.
The ground control station allows the user to modify the gains of the feedback
system, change sensitivity of pilot inputs, and record attitude data transmitted by
ELKA. All the data processing, feedback control calculations, and roll-yaw decou-
pling methods are performed onboard by ELKA. It is also able to serially connect
with auxiliary sensors such as a motion capture system and simultaneously record
data from all connected sensors.
High level stick inputs from the pilot are transmitted to the vehicle through
the radio control (RC) transmitter and received by the satellite receiver connected to
ELKA. The JR X9303 transmitter sends actuator commands via Digital Spectrum
Modulation (DSM). Compared to traditional radio systems, DSM is less susceptible
to radio interference or crosstalk because it does not require a crystal set to a
specific frequency band and channel. Instead, DSM broadcasts signals using either
101
Figure 3.13: PWM signals used by actuators (f = frequency).
Frequency Hopping Spread Spectrum (FHSS) or Direct Sequence Spread Spectrum
(DSSS). FHSS is a broadcasting method where the controller transmits narrow band
signals that consistently hop between multiple frequencies within the 2.4 GHz band.
DSSS is a more common method, and used by JR transmitters, where the system
operates on a single frequency but on a very wide band.
The satellite receiver translates inputs from the radio controlled (RC) trans-
mitter into pulse width modulated (PWM) signals. These inputs are in microseconds
and represent the duty cycle of the PWM signal (Fig. 3.13). PWM signals range
from 1000–2000 microseconds and denote the time voltage is provided to the actua-
tor, effectively turning it on. Final inputs to the feedback control system include 5
PWM values for the three motors and two servos used by the twin-cyclocopter. The
frequency of pulse can also be adjusted to match actuator requirements. Typically,
motors require a higher pulse frequency (> 500 Hz) while analog servos prefer lower
pulse frequency (< 100 Hz) to prevent servo noise.
102
Figure 3.14: MAV-scale ADS-B sensors developed by uAvionix.
3.7.2 Automatic Dependent Surveillance-Broadcast (ADS-B)
Aside from communication between the UAV and its GCS, communication with
neighboring aircraft is critical for safe integration of UAVs into the National Airspace
System. An automatic dependent surveillance-broadcast (ADS-B) is a surveillance
equipment that broadcasts the location of an aerial vehicle in real-time using a
digital data link. Ground stations serve as nodes by relaying ADB-S data to ve-
hicles in the airspace. The digital data link broadcasts in the 1090MHz band and
includes flight identification number, unique aircraft code, position, altitude, rate
of climb/descent, and velocity data. Recently, uAvionix released a FCC approved
miniature ADS-B system for UAVs. The ping2020i is a light-weight full range ADS-
B transceiver weighing at 20 grams (Fig. 3.14). Integrated with both GPS and
barometer, ping2020i is capable of detecting aircraft within 100 statute mile radius.
103
Figure 3.15: Schematic of closed-loop feedback system with the incorpo-
rated roll-yaw decoupling strategy.
3.8 Closed-Loop Feedback System
While the previous sections covered sensors for both piloted and autonomous nav-
igation, the cyclocopter is primarily equipped with sensors for flight stability and
control such as gyroscopes and accelerometers. Therefore, the closed-loop feedback
system implemented on the vehicle is a linear single-input single-output (SISO)
system divided into an inner-loop for attitude stabilization and an outer-loop for
position control by a pilot (Fig. 3.15). The vehicle relies on manual control because
of limitations of estimating position with the current setup of onboard sensors in an
indoor setting.
Chapter 6 discusses modifications to the outer-loop to achieve autonomous
navigation using external sensors. One method of estimating position and veloc-
ity is through dead-reckoning, a process of integrating accelerometer and gyroscope
104
measurements. Due to noise levels and biases, accuracy of the integrated data deteri-
orates over time. For stationary accelerometer measurements, the error in estimated
√
position is α T where α depends on standard deviation of the accelerometer noise
and sampling frequency of data and T is the integration time [73].
As mentioned previously, alternatives to integrating inertial sensors include
using cameras and estimating position through optic flow. These techniques are
beyond the scope of the present study and are not considered at this time. Au-
tonomous navigation can also be achieved using external sensors such as a motion
capture system. Later chapters will discuss methodologies involved in controlling
the cyclocopter using such system.
The outer-loop implemented on the vehicle updates at 45 Hz while the inner-
loop updates at 1000 Hz. The outer-loop rate is limited due to the commercial
transmitter. Desired states of the closed-loop feedback system include pilot com-
mands through the RC transmitter, which associates to a desired pitch and roll
Euler angles (φ, θ) and the angular velocities in the body frame (p, q, and r). The
final control inputs to the vehicle actuators are the individual rotational speeds of
the rotors and the two servo inputs for thrust vectoring.
3.8.1 Attitude Stabilization
Typical SISO systems use a proportional-integral-derivative (PID) control scheme
for attitude stabilization. The proportional (P) compensator outputs a value that is
a gain (Kp) multiplied by the error between the system’s output and desired input.
105
The integral (I) compensator reduces the steady-state error of the feedback system
by using the integral of error over time multiplied by a gain (Ki). Finally, the
derivative (D) compensator provides damping and improves the transient response
of the system by using the derivative of the error signal multiplied by a gain (Kd).
The PID controller is represented by the following equation [74]:
∫ t de
u(t) = Kpe(t) +Ki e(τ)dτ +Kd (3.27)
0 dt
where e(t) is the error between system’s measured output, ŷ(t), and desired input,
ydes(t):
e(t) = ydes(t)− ŷ(t) (3.28)
Within PID, the integral controller compensates for environmental distur-
bances or modeling errors. It slowly reduces the steady-state errors by increasing
the feedback inputs, u(t). A large disturbance or immediate change in the set-
point can result in an undesirably large feedback input which can cause the system
to overshoot the setpoint and destabilize the vehicle. In addition, the cyclocopter
experiences inherent gyroscopic couplings along the roll-yaw axis. The gyroscopic
coupling and the corresponding decoupling strategy is intensively discussed in Chap-
ter 5. Complexities from the decoupling strategy led to selection of a PD control
law for attitude stabilization of the cyclocopter.
The inner-loop can be further broken down into cascaded control loops com-
posed of attitude and rate controllers. The inner-loop stabilizes around a desired
106
Figure 3.16: Comparison of attitude tracking controller vs. fixed attitude
controller (previous implementation).
input commanded by the pilot. These inputs to the system are roll and pitch atti-
tude angles (φ, θ). Heading angle (ψ) is provided as a feedforward term, where the
final servo output is calculated based on the given pilot command and independent
of the attitude controller. Both the attitude and rate controllers operate at 1000
Hz, but in the current setup, the attitude controller has a dynamic setpoint while
the rate controller always stabilizes around zero angular velocities along each axis.
The modification is known as derivative kick and is typically used to reduce large
spikes originating from taking derivative of the setpoint.
The inner-loop control schematic achieved improved setpoint tracking (Fig. 3.16)
compared to previous system [39]. The desired states (φ, θ, p, q, r) are dependent on
107
Table 3.7: Ziegler-Nichols Frequency Response Method.
Controller Kp Ti Td
P 0.5 Ku ∞ 0
PI 0.4 Ku 0.8 Tu 0
PID 0.6 Ku 0.5 Tu 0.125 Tu
the pilot command and the mixed output from decoupling strategy. Previously, the
desired states were specific to the hover condition and set to zero in order to negate
any perturbation from trim. In addition, the modified controller incorporates roll-
yaw decoupling scheme in the inner-loop and as a feedforward term (Fig. 3.15). The
integration to the inner-loop reduced the dependence on the feedforward correction,
umix(t), which improved overall tracking.
3.8.2 Ziegler-Nichols PID Tuning in Hover
Equation 3.27 shows the PID controller in the time domain. However, the equation
is better represented in the frequency domain (s-domain) for analysis and when
designing the initial system. The PID controller in s-domain is written in terms of
integration time (Ti) and derivative time (Td):
1
u(s) = Kp(1 + + Tds) (3.29)
Tis
The relationship between Kd and Ki with integration and derivative times can be
described by the following:
108
Kp
Ki = (3.30)
Ti
Kd = KpTd (3.31)
Ziegler-Nichols is a traditional method of experimentally tuning PID gains.
The tuning process involves either using open-loop step response or by character-
izing frequency response of the system. The first method uses the following to
determine the gains: (1) time delay between the desired step-change in output and
the measured output and (2) reaction rate, which is calculated using intersection
between tangent of the step response and the coordinate axes. The second method,
using frequency response, is an empirical tuning process based on locating the point
of intersection between the system’s Nyquist curve and the negative real axis. Exper-
imentally, this is determined by incrementally increasing the proportional gain until
the system achieves sustained oscillations. Values of the gain (Ku) and period of
oscillation (Tu) are then subsequently used to determine the PID gains (Table. 3.7).
Prior to free-flight tests, the Ziegler-Nichols method was used to find the pro-
portional and derivative gains along the pitch, roll, and yaw axes through series of
isolated 1-DOF and 3-DOF perturbation experiments. The vehicle was suspended
by its nose rotor spar along two ball bearings encased in a 3-D printed housing
for both the roll and pitch stands. The setup restricted the vehicle’s rotation to
±20◦ either along the longitudinal or lateral axis. Afterwards, the cyclocopter was
mounted on a 3-DOF gimbal stand that enabled up to ±10◦ of rotation along all
109
Table 3.8: Gains used in feedback system.
Feedback System Gain Value Unit
Attitude K 1pφ 0.4 rad
Attitude Kpθ 0.3
1
rad
Attitude Kdφ 0.6
s
rad
Attitude Kdψ 0.7
s
rad
Attitude K sdθ 0.7 rad
Figure 3.17: Twin-cyclocopter in steady hover (position data on right).
three body axes. The stand composed of a gimbal bearing attached to a 3-D printed
housing, which coincided at the vehicle’s center of gravity (C.G).
3.8.3 Hover Flight Tests
The gains were further fine-tuned in flight due to discrepancies between gimbal stand
testing and actual flight conditions. Friction from the gimbal bearing contributed
additional damping to the vehicle’s dynamics, which resulted in selecting lower gains.
110
20
0
−20
5 10 15
20
0
−20
5 10 15
100
0
−100
5 10 15
Time (s)
Figure 3.18: Attitude data from ELKA during hover.
In addition, there was a shift in C.G since the vehicle used a LiPo battery in flight
instead of directly connecting to a power supply.
Flight tests were systematically conducted by first providing a pure throttle
command and observing the vehicle response. The cyclocopter was found to be
sensitive to ground effect. If the throttle was increased incrementally such that
the cyclorotors spooled up in a slow rate, the vehicle would drift laterally and fail
to gain altitude. To mitigate ground effect, throttle was increased in a step-wise
manner to quickly achieve the operating RPM. Any additional forward translation
out of ground effect was counteracted by either uniformly reducing the phasing of
the cyclic blade pitch or varying the tail propeller RPM.
After fine-tuning the gains, the twin-cyclocopter successfully demonstrated
stable piloted hover within 1 m2 area (Fig. 3.17). Position data was taken using the
VICON motion capture system. With the modified closed-loop feedback system,
111
Yaw rate (deg/s) Roll, φ (deg) Pitch, θ (deg)
onboard attitude data showed that the controller achieved satisfactory attitude an-
gles and rates (Fig. 3.18). The next step is to explore control strategies for forward
flight using the cyclocopter’s thrust vectoring capability.
3.9 Summary
This chapter provided a comprehensive overview of sensors equipped on both piloted
and autonomous MAVs. Due to weight restrictions, MAVs are confined to using
light-weight and compact avionics consisting of autopilot, sensors, and an onboard
computer. Sensors discussed in this chapter are used for flight stability and control,
flight performance diagnostics, and guidance and navigation (autonomy). Sensors
required for visual odometry and SLAM were also compared along with available
open-source algorithms. Finally, the chapter included an introduction to closed-loop
feedback system and the tuning method used for the cyclocopter.
In particular, the cyclocopter achieved stable piloted hover using a PD-based
closed-loop feedback system and ELKA-R, a custom light-weight autopilot. Im-
provements to the inner-loop attitude stabilization resulted in improved tracking.
Onboard sensors include an IMU with tri-axial gyroscopes and accelerometers. An
additional pressure transducer was used to mitigate perturbation from gusts and will
be discussed in a later chapter. Communication between ELKA-R and a nearby
ground control station was facilitated using a nRF wireless transceiver. In addi-
tion, a 2.4 GHz radio link connected the pilot’s transmitter to a Spektrum receiver
onboard ELKA-R in order to maintain an independent link to the vehicle.
112
Chapter 4: Flight Control Strategy
4.1 Overview
The previous chapter discussed the development of twin-cyclocopter subsystems
ranging from cyclorotor propulsion system to the closed-loop feedback system. In
this chapter, flight control strategies for the cyclocopter are developed by using a
combination of independent thrust vectoring and rotational speed control of the
two cyclorotors and nose propeller. Advantages of the developed strategy include
seamless transition from hover to forward flight, and capability to maintain a power-
efficient level attitude in flight. Such a strategy also led to roll-yaw cross-couplings
that inhibited onboard stabilization. Sources of cross-couplings across all flight
modes are discussed in details. Results from wind tunnel studies conducted using
a 5-DOF experimental stand were used to develop a control strategy to achieve
decoupled dynamics. In addition, effects of free-stream velocity, rotational speed,
and phasing angle on the roll-yaw cross-couplings were closely investigated and
used to determine control inputs to achieve zero pitch attitude at various airspeed.
Finally, the developed flight control strategy was successfully demonstrated up to 5
m/s.
113
Figure 4.1: Notation for forces and moments on the cyclocopter.
Figure 4.2: Gyroscopic coupling due to net angular momentum.
114
4.2 Rigid Body Dynamics
The first step in developing a flight control strategy is to understand the rigid body
dynamics and evaluate trim equations for the forces and moments acting on the vehi-
cle. Figure 4.1 shows the notation used for forces (X, Y, Z) and moments (L,M,N)
along the vehicle’s body axes. Translational velocities [u, v, w]T and angular rates
[p, q, r]T are typically provided in the body frame, B = (xB, yB, zB). However, posi-
tion [x, y, z]T is typically provided with respect to the fixed frame or inertial frame,
I = (xI , yI , zI). Conversion between inertial and body frame is calculated using a
conventional 3-2-1 rotation sequence using Euler angles [φ, θ, ψ]. The first rotation
is about ez by yaw angle (ψ), second rotation about ey by pitch angle (θ), and finally
the third rotation about ex by roll angle (φ). Rotational transformation between
the two frames is summarized by the following [52]:
 
 cos θ cosψ cos θ sinψ − sin θ [RBG] =  sinφ sin θ cosψ − cosφ sinψ sinφ sin θ sinψ + cosφ cosψ sinφ cos θ 
cosφ sin θ cosψ + sinφ sinψ cosφ sin θ sinψ − sinφ cosψ cosφ cos θ
(4.1)
Using the transformation matrix, body rates can be expressed in terms of Euler
angles:
p = φ̇− ψ̇ sin θ (4.2)
115
q = θ̇ cosφ+ ψ̇ sinφ cos θ (4.3)
r = ψ̇ cosφ cos θ − θ̇ sinφ (4.4)
4.2.1 Forces and Moments
The rigid body (X, Y, Z) and moments (L,M,N) can be represented as series of
ordinary differential equations as follows:
∑
X = m(u̇+ qw − rv) +mg sin θ + (FL + FR) sin Φ +−Dx (4.5)
∑
Y = m(v̇ + ru− pw)−mg sinφ cos θ −Dy (4.6)
∑
Z = m(ẇ + pv − qu)−mg cosφ cos θ − Fz +Dz (4.7)
∑
L = Ixxṗ− (Iyy − Izz)qr + (FL − FR)dy +Dydz +Hyr (4.8)
∑
M = Iyy q̇ − (Izz − Ixx)pr − (FL + FR)dx + Ftaildtail +Dxdz −Qcyclo (4.9)
∑
N = Izz ṙ − (Ixx − Iyy)pq + (FL + FR)dy sin Φ +Dydy −Qprop −Hyp (4.10)
In these equations, Fx, Fz, and Ftail are the total lift produced by the cycloro-
tors and propeller while FL and FR are the forces produced by the left and right
cyclorotors (Fig 4.1). The Dx, Dy, and Dz are drag terms while the dx, dy, dz, and
dtail terms are the moment arms. During forward flight, the forces produced by the
116
Figure 4.3: Cyclorotor coordinates showing measurement of absolute phase
angle (γabs), lift force, and propulsive force in forward flight.
cyclorotor are influenced by its phase angle (Φ), rotational speed, and freestream
velocity. The products of inertia terms (Ixy, Ixz, Iyz) have been removed because
the cyclocopter is quasi-symmetric about the X-Y and X-Z planes.
4.2.2 Gyroscopic Coupling
The dominant gyroscopic moments produced by the cyclorotors are expressed in
terms of angular momentum (Hy) in the respective axis. Angular momentum is a
cross-product of rotational moment of inertia (Ir) and rotational speed (ω) of the
cyclorotors. From the rigid body equations of motion, it is evident that roll and
yaw dynamics are cross-coupled due the gyroscopic moment terms. The magni-
tude of cross-coupling increases with rotational speed, which further complicates
stabilization. This inertial coupling is addressed in this chapter.
117
Figure 4.4: Schematic showing hover condition with γabs > 0
◦ and Φ=0◦.
Figure 4.5: Cyclorotor coordinates in forward flight where γabs > 90
◦ and
Φ=90◦.
118
4.3 Cyclorotor Coordinate System
The coordinate system used for the cyclorotor is shown in (Fig. 4.3). The time-
varying aerodynamic forces produced by the cyclorotor can be resolved into the
vertical and horizontal directions. The azimuthal position of the blade (Ψ) is mea-
sured counter-clockwise from the horizontal axis of rotation and is equivalent to zero
when the blades are rightmost side of the circular trajectory. The blade pitch angle
(θ) is the angle between the blade chord and the line tangent to blade path. The
absolute phase angle (γabs) is defined as the relative azimuthal location of the max-
imum blade pitch angle with respect to the -Z-axis. When γabs = 0
◦, the maximum
blade pitch occurs at the top of the circular trajectory and when γabs = +90
◦, the
maximum occurs +90◦ clockwise towards the direction of incoming flow.
The lift and propulsive force components are defined as the net aerodynamic
forces produced along the -Z-axis and +X-axis. For the present cyclorotor, the pitch
amplitude is kept constant and the magnitude of the net resultant thrust vector is
controlled by varying the cyclorotor rotational speed. The direction of the thrust
vector is rotated through cyclic pitch phasing. To achieve a stable hover, the phase
angle must be rotated such that the resultant thrust vector is purely vertical.
4.3.1 Forward Flight Orientation
For the remainder of the dissertation, the forward flight phase angle (Φ) of the
cyclorotor, distinct from the absolute phase angle of each blade, is defined with
respect to the phase in hover (γhover).
119
(a) Definition of pitch, roll, and yaw degrees of (b) Pitch control.
freedom.
(c) Roll control. (d) Yaw control.
Figure 4.6: Control strategy for twin-cyclocopter.
Φ = γabs − γhover (4.11)
A forward flight phase angle of Φ = 0◦ corresponds to the hover condition
when there is a net lift component and zero propulsive force (Fig. 4.4). As shown
in (Fig. 4.5), the phase angle in forward flight is increased (Φ ≥ 0◦) such that the
blades have a non-zero pitch angle along the horizontal axis (Ψ = 0◦ and Ψ = 180◦).
The resulting blade kinematics produces a net force vector with both a vertical lift
component and a horizontal propulsive thrust component.
The phasing of cyclic blade pitch plays a significant role in trimming the
cyclorotor in forward flight and this will be discussed in further detail in the next
section.
120
4.3.2 Virtual Camber and Virtual Incidence Effect
As mentioned in chapter two, the blades on the cyclorotor experience virtual camber
effect because of the curvilinear nature of the flow, which introduces a chordwise
variation of velocity direction [80]. Therefore, a symmetric airfoil at 0◦ pitch behaves
as a virtually cambered airfoil at a virtual incidence in a rectilinear flow. The blades
experience a negative virtual camber and incidence at the top of the trajectory
(Cl < 0) and a positive camber and incidence at the bottom portion of the trajectory
(Cl > 0).
In forward flight, the advancing side of the blades are at the bottom portion
of the trajectory while retreating at the top of the trajectory. The blades on the
advancing side experience an increase in the net tangential velocity and a reduction
in the retreating side. As a result, the blades produce a small downward lift force
while operating in the upper half of the trajectory and large upwards lift force on
the lower half, leading to an overall net positive lift force. Since virtual camber
is influenced by the chord/radius ratio of the rotor and virtual incidence is also
dependent on the chordwise location of the blade pitching axis, the magnitude of
lift depend on these parameters and the direction of rotation.
While the direction of rotation is not important in hover, it is critical to
the lift and thrust producing capability of the cyclorotor in forward flight. In the
present vehicle, both the cyclorotors spin in the same direction along the +Y axis
(Fig. 4.1), resulting in an unbalanced clockwise torque that must be counteracted by
the conventional propeller. To obtain a positive resultant lift force, the cyclorotors
121
are mounted aft of the propeller system such that the blades are retreating at the
top and advancing at the bottom of the trajectory.
4.4 Attitude Control in Hover
Figure 5.10(a) shows the definition of pitch, roll, and yaw degrees of freedom for
the twin-cyclocopter. Stable hover, transition, and forward flight require a unique
combination of independent rotational speed of all three motors and thrust vectoring
of the cyclorotors for each mode of flight.
A positive roll is produced by increasing the rpm of the left rotor and decreasing
the rpm of the right rotor (Fig. 5.10(c)). Finally, yaw is controlled through differen-
tial rotation of the two cyclorotor thrust vectors (Fig. 5.10(d)). The cyclorotors spin
in the same direction along the +Y-axis so there is a net angular momentum that
induces gyroscopic coupling between the roll and yaw degrees of freedom. To elim-
inate the gyroscopic coupling, an onboard control mixing was implemented where
the roll and yaw inputs were appropriately combined to give decoupled motion. The
decoupling strategy is presented with details towards the end of this chapter.
4.5 Attitude Control in Forward Flight
In forward flight, the combined forward flight phase actuation of cyclic blade pitch
is used to control the forward velocity of the vehicle. Providing a cyclic pitch
phasing for both cyclorotors rotates the resultant thrust vectors by the same phase
angle. Essentially, an increase of the combined forward flight phase angle would
122
Figure 4.7: Lift and propulsive thrust for cyclocopter in level forward flight.
result in a higher propulsive force contribution from the cyclorotors. As in hover,
pitch, roll, and yaw moments in forward flight are controlled through propeller
rpm, differential cyclorotor rpm, and differential thrust vectoring, respectively. The
pilot controls now include throttle (collective cyclorotor rpm), pitch (propeller rpm),
roll (differential cyclorotor rpm), yaw (differential phasing), and forward velocity
(collective phasing). Pitch control is available to the pilot because during flight, as
the battery voltage drops, the pitch trim also changes due to discrepancies between
the cyclorotors and nose rotor rotational speeds.
For the cyclocopter, transitioning from hover to forward flight does not involve
any configuration changes. By slowly increasing the phase angle (Φ), the vehicle is
able to steadily increase forward velocity and transition into forward flight. The
rate of transition from hover to forward flight depends entirely on the phasing input
(Fig. 4.7).
One of the major challenges in developing the forward flight control strategy
123
(a) 5-DOF experimental stand for wind tun-
(b) Wind tunnel testing setup.
nel testing.
Figure 4.8: Wind tunnel testing setup.
is the existence of a strong coupling between the roll and yaw control that increases
with the forward tilt of the thrust vector. Furthermore, this coupling is in addition
to the inertial gyroscopic roll-yaw coupling that is experienced even for the hover
case. Extensive wind tunnel experiments were conducted in order to understand the
source of the roll-yaw control cross-coupling.
4.6 Wind Tunnel Studies
The objectives of the wind tunnel testing of the cyclocopter are as follows: (1)
understand the effect of forward velocity and phase angle on vehicle dynamics, (2)
systematically find the appropriate mixing ratios required to decouple the roll-yaw
couplings at various operating forward velocities, and (3) quantify the control inputs
required to maintain a trimmed level forward flight (total lift = weight, thrust =
124
drag).
One of the challenges of accurately simulating forward flight conditions in the
wind tunnel was to develop an experimental setup that would enable 5-DOF motions
without significantly introducing additional damping into the system. The vehicle
was mounted on a spherical gimbal at CG, which allowed free rotation in pitch, roll,
and yaw (Fig. 4.8(a)). The entire setup was then augmented with a horizontal and a
vertical slider to enable forward translation and heave (Fig. 4.8(b)). Only the lateral
translation was restricted due to the physical constraints set by the wind tunnel.
4.6.1 Methodology
The open-jet wind tunnel consists of a square test section dimensions of 0.56
meters (22 inches) in length and width. The vehicle was restricted to maximum
horizontal translation from reference trim location (center of slider) of±0.178 meters
(7 inches) and a positive heave of 0.051 meters (2 inches). Rotations about pitch,
roll, and yaw were mechanically restricted to ±45◦.
The variables in this experimental study were airspeed (V∞), phase angle (Φ),
and rotational speed (Ω). A systematic velocity sweep from 2 to 5 m/s was con-
ducted at intervals of 1 m/s. At each velocity, the phase angles were varied from 0◦
to 50◦ in increments of 5◦. The rotational speed was varied such that the cyclocopter
always maintained a positive displacement on the vertical slider. This represents
the trim condition where lift is equivalent to total vehicle weight. Throttle (collec-
tive cyclorotor rpm), pitch (propeller rpm), roll (differential cyclorotor rpm), yaw
125
(a) Variation of lift. (b) Variation of thrust.
Figure 4.9: Variation of forces with velocity and phasing at 1400 RPM.
(differential phasing), and forward velocity (collective phasing) inputs were given
so that the vehicle maintained its reference trim position on the horizontal slider
without pitch, roll, or yaw moments.
4.6.2 Effect of Free-Stream Velocity
An extensive wind tunnel testing was previously conducted on an isolated cyclorotor
in which the effects of airspeed, phase angle, and advance ratios on lift and thrust
were measured [37]. The results of the study at the constant operational speed of
1400 rpm are plotted showing the variation of lift and thrust against airspeed and
phase angle (Fig. 4.9(a), 4.9(b)). Lift produced by the cyclorotor increases with
free-stream velocity and decreases with phase angle while thrust inversely decreases
with velocity and increases with phase angle.
A significant finding from the current experiment is that for a constant phas-
ing, varying the free-stream velocity did not affect the magnitude of roll-yaw cross-
coupling. In addition, the cyclocopter was able to maintain pitch trim within 3◦ for
126
100
0
−100
4 6 8 10 12
100
0
−100
4 6 8 10 12
100
0
−100
4 6 8 10 12
Time (s)
Figure 4.10: Oscillations in pitch, roll, and yaw dynamics caused by cross-coupling.
a step change in forward velocity up to 5 m/s with a PD-based feedback control
system. While free-stream velocity affects the lift produced by the cyclocopter, the
feedback control system is able to stabilize imbalance in the longitudinal dynamics.
4.6.3 Effect of Phasing
To understand the effect of phasing on the roll-yaw coupling, the rotational speed
and airspeed are kept constant as phasing increases. In one case, the cyclorotor
rotational speed is kept at a constant 1200 rpm and the airspeed is maintained at
2 m/s. Starting from Φ=0◦, the phasing is incrementally increased by Φ=5◦. As
the phasing approaches Φ=15◦, there is an observable oscillation in all three degrees
of freedom (Fig. 4.6.2). Throughout the process, the feedback control system is
attempting to stabilize the vehicle attitude.
127
Yaw rate (deg/s) Roll rate (deg/s) Pitch rate (deg/s)
Increasing the phasing, which tilts the thrust vector, results in a noticeable in-
crease in roll-yaw coupling when at a constant forward velocity. The coupling is due
to the increase in magnitude of the propulsive forces at high phase angles. Therefore,
thrust vectoring has a larger impact on the roll moments and subsequently, changes
in cyclorotor rotational speed have a greater influence on yaw moments. Since the
roll-yaw coupling is only dependent on phase, mixing ratios for roll and yaw inputs
vary only with phase angle and not forward velocity.
1300
1250
1200
1150
1100
1 2 3 4 5
60
40
20
0
1 2 3 4 5
Forward Velocity (m/s)
Figure 4.11: Control inputs required to obtain trimmed free flight (lift =
weight, thrust = drag).
4.6.4 Control Inputs for Trimmed Forward Flight
A sweep of forward velocities and phasing was conducted to find the appropriate
control inputs needed to achieve steady level flight (thrust = drag, lift = weight) at
each velocities. At a constant forward velocity and phasing, increasing the rotational
speed results in an increase of both lift and thrust and consequently, an increase
128
RPM
Phase, Φ (deg)
in power. Increasing phasing of the cyclorotors, while keeping both the forward
velocity and rpm constant, results in an increase in thrust and a decrease in lift.
Therefore, to increase steady level flight speed from a trimmed condition, both the
phasing and rpm are simultaneously adjusted.
As mentioned earlier, a phase input produces a differential change in phasing
of both cyclorotor thrust vectors while a throttle input uniformly adjusts the rota-
tional speeds. Figure 4.11 shows the phasing and throttle inputs required to main-
tain vehicle trim (lift = weight, thrust = drag) at different freestream velocities.
As the forward velocity increases, the required phasing increases almost linearly.
The required rotational speed of the cyclorotor also increases until 3 m/s and then
decreases at higher airspeed. Further experiments will be conducted in the future
to measure vehicle performance past 5 m/s. Overall, the insights gained from the
wind tunnel experiments show that the forward flight control strategy and feedback
control system is suitable for free flight testing.
4.7 Roll-Yaw Decoupling Strategy
Decoupling the roll and yaw degrees of freedom in forward flight involves mixing
the roll and yaw inputs of both the pilot and the feedback system. The finalized
forward flight control strategy is depicted in (Fig. 4.12). A positive roll (right) input
results in an increase in rpm of the left rotor and a decrease in the right rotor. At
higher phases, this also leads to positive yaw (clockwise). If the positive roll input
is accompanied by a negative yaw (counterclockwise), it would negate the effects of
129
(a) Mixed roll right and counterclock-
(b) Pure roll right response.
wise yaw input.
(c) Mixed clockwise yaw and roll right (d) Pure clockwise yaw response.
input.
Figure 4.12: Controls mixing for forward flight.
Figure 4.13: Closed-loop feedback system with feedforward mixing of con-
trol inputs.
130
the rpm on yaw and result in a pure roll output. Likewise, a positive yaw input
would be paired with a negative roll input to accomplish a pure yaw output.
Both the gyroscopic and control couplings induced the same positive roll re-
sponse for a positive yaw input. However, for a positive roll input, gyroscopic
coupling induced a negative yaw (counterclockwise) response while the control cou-
pling led to a positive yaw (clockwise) response. Therefore, decoupling strategy uses
eight different mixing gains (a, b). The mixing scheme for gyroscopic coupling is:
δφa = [aφ δlat lat + aφ δ ] (4.12)ped ped
δψa = [aψ δlat + aψ δped] (4.13)lat ped
The mixing ratios aφ , aφ , aψ and aψ were determined using a gimballat ped lat ped
stand at the operational rpm. For a given roll or yaw input, the ratios were in-
crementally increased until the vehicle showed a decoupled response. The mixing
scheme for controls coupling is as follows:
δφ = [bb φ δlat lat + bφ δped ped]sinΦ (4.14)
δψ = [bψ δlat + bψ δped]sinΦ (4.15)b lat ped
The mixing ratios bφ , b , b and blat φped ψlat ψ were identified in the wind tunnel.ped
The mixing ratios were iterated until the vehicle dynamics were decoupled at each
phasing input. The mixed inputs are combined with the output of the PD controller
and sent to the vehicle (Fig. 4.13).
131
Figure 4.14: First successful forward flight of a twin-cyclocopter (V = 3
m/s).
4
3
2
1
0
20 25 30 35
Time (s)
Figure 4.15: Airspeed data taken shows forward velocity of 3.5 m/s.
4.7.1 Controller Design
4.8 Forward Flight Test Results
During the flight tests, the proportional and derivative gains were tuned using the
Ziegler-Nichols approach. The gains that offered acceptable stiffness and damping
to reject external disturbance with minimal oscillations were selected. Gains and
control inputs for trimmed flight varied slightly from the ones obtained through
wind tunnel testing because the cyclocopter was previously constrained in a gimbal
132
Velocity (m/s)
40
20
0
20 22 24 26 28 30
1600
1400
1200
1000
20 22 24 26 28 30
Time (s)
Figure 4.16: Phase angle (Φ = 35◦) and rotational speed (1400 RPM)
during forward flight at 3.5 m/s.
40
20
0
13 14 15 16 17 18
5
4
3
2
1
0
13 14 15 16 17 18
1500
1450
1400
1350
1300
13 14 15 16 17 18
Time (s)
Figure 4.17: Control inputs at 5 m/s.
133
RPM
Phase angle (deg)
RPM Velocity (m/s) Phase angle (deg)
20
0
−20
13 14 15 16 17 18
20
0
−20
13 14 15 16 17 18
100
50
0
−50
−100
13 14 15 16 17 18
Time (s)
Figure 4.18: Attitude data at 5 m/s.
setup that provided additional damping in pitch, roll, and yaw. Additional variation
was caused by shifts in the moments of inertia because the cyclocopter is equipped
with a single 3S 11.1 volt 850 mAh Lithium-Polymer battery weighing 75 grams.
Flight tests were systematically conducted by first providing a pure throttle
command and observing the vehicle response. Any forward translation was coun-
teracted by either reducing the phasing of the cyclic blade pitch or varying the
propeller rpm. After trimming the cyclocopter, level forward flight is then initiated
solely through thrust vectoring of the cyclorotors.
The present vehicle successfully demonstrated steady level forward flight up
to a maximum of 5 m/s forward velocity. Figure 4.14 shows level flight achieved at
3 m/s. As seen from figures 4.17 and 4.18, a satisfactory level forward flight was
134
Yaw rate (deg/s) Roll, φ (deg) Pitch, θ (deg)
100
40
0
20
0 −100
18 20 22
17 18 19 20 21 22 23
5 20
4
3 0
2
1 −20
0 18 20 22
17 18 19 20 21 22 23
100
1500
50
1450
0
1400
−50
1350
−100
1300 17 18 19 20 21 22 23
17 18 19 20 21 22 23 Time (s)
Time (s)
(b) Decoupled roll right response for a roll
(a) Control inputs and velocity.
input.
100
40
0
20
0 −100
10 12 14 16 18
10 12 14 16 18
5 20
4
3 0
2
1 −20
0 10 12 14 16 18
10 12 14 16 18
100
1500
50
1450
0
1400
−50
1350
−100
1300 10 12 14 16 18
10 12 14 16 18 Time (s)
Time (s)
(d) Decoupled counterclockwise yaw re-
(c) Control inputs and velocity.
sponse for a yaw input.
Figure 4.19: Attitude data at 3.5 m/s.
135
RPM Velocity (m/s) Phase angle (deg) RPM Velocity (m/s) Phase angle (deg)
Yaw rate (deg/s) Roll, φ (deg) Yaw input (µs) Yaw rate (deg/s) Roll, φ (deg) Roll input (µs)
obtained with the pitch and roll angles held close to zero and minimum offsets in
yaw rate.
Fig. 4.19 show flight test data at 3.5 m/s and Φ = 35◦ with the rotational
speed maintained near 1400 rpm. At these control settings, roll and yaw inputs
were given to demonstrate decoupled roll-yaw moments. Fig. 4.19(b) shows the
vehicle response for a positive roll input (roll right) plotted against the attitude
rates and roll attitude angle. The roll step input is given at time = 21 seconds. An
immediate perturbation in the roll rate is observed without significant changes in
the yaw rate. Interestingly, the feedback overshoots when trying to stabilize the roll
moment, but quickly manages to recover the trim state. In order to get desirable
performance with minimal overshoot, controller gains were fine-tuned in hover since
it was found to be challenging to tune the gains in forward flight without potentially
crashing the vehicle.
The mixing ratios found in the wind tunnel were again applied and tuned
during free flight tests. There was a notable difference in the ratios, which can be
attributed to the additional damping provided by the test stand and changes in
vehicle inertia when mounting batteries. While the ratios found in the wind tunnel
were similar for both roll-yaw and yaw-roll couplings, the ratios for yaw-roll coupling
was much greater in flight. At higher phase angles, the yaw-roll mixing ratio was
observed to be 35% higher since the vehicle is much more susceptible to yaw-roll
coupling. This is due to the fact that both the gyroscopic coupling and the controls
coupling induce a similar response for a yaw input, which adds the effect of the
couplings, and opposite responses in roll.
136
The 550 gram twin-cyclocopter is currently the first forward flight capable
cyclocopter in literature that has successfully used thrust vectoring to gain forward
velocity instead of pitching the vehicle forward. Thrust vectoring in forward flight
has unique aerodynamic advantages in that it produces both lift and thrust forces
with a measured drop in required power [37]. As a result, the maximum forward
flight capability can be potentially increased over a system that solely relies on
tilting the vehicle to achieve translational motion.
4.9 Summary
In this chapter, the flight control strategy that enabled the cyclocopter to expand it’s
flight envelope to steady level attitude in power-efficient forward flight was discussed.
The control strategy used a combination of independent pitch phasing and rotational
speed control of the two cyclorotors, along with the propeller system, without relying
on any additional control by means of a traditional empennage system. While the
strategy facilitated power-efficient level forward flight, it was accompanied by a
strong yaw-roll control cross-coupling. These were in addition to the gyroscopic
couplings that are experienced even in hover. Existence of controls coupling in
forward flight was attributed to increased magnitude of propulsive forces at high
phase angles. As a result, thrust vectoring had a larger impact on roll moments and
subsequently, changes in cyclorotor rotational speeds had a greater influence on yaw
moments.
Decoupling the roll and yaw degrees of freedom in forward flight involved
137
mixing the roll and yaw inputs of both the pilot and the feedback system. Wind
tunnel studies showed that the mixing ratio was dependent on phasing of the blade
cyclic pitch, especially at high phase angles. If a positive roll input was accompanied
by a negative yaw (counterclockwise), it would negate the effects of the rpm on yaw
and result in a pure roll output. Likewise, a positive yaw input paired with a positive
roll input resulted in a pure yaw output. Both the gyroscopic and control couplings
induced the same positive roll response for a positive yaw input. However, for a
positive roll input, gyroscopic coupling induced a negative yaw (counterclockwise)
response while the control coupling led to a positive yaw (clockwise) response.
Overall, the capability to control forward flight speed using only thrust vec-
toring (phasing of the cyclic pitch) was successfully demonstrated up to 5 m/s. The
rate of transition from hover to forward flight entirely depended on the magnitude
of phasing input. Flight tests demonstrated that the cyclocopter was able to imme-
diately halt to hover from 5 m/s flight. In addition, the cyclocopter achieved the
desired airspeed within a second after a given phasing input.
138
Chapter 5: System Identification in Forward Flight
5.1 Overview
The previous chapter discussed the flight control strategy that enabled the cyclo-
copter to maintain a level attitude in forward flight. Even though such a strategy
could facilitate power-efficient forward flight, it is accompanied by a strong yaw-
roll cross coupling, which is in addition to the inherent gyroscopic coupling that is
present in hover. To understand these couplings and characterize the bare-airframe
dynamics, a 6-DOF flight dynamics model of the cyclocopter was extracted using
time-domain system identification technique. The existence of the longitudinal con-
trol derivative in translation mode was used to evaluate the control authority of
thrust vectoring to command forward speed. The model was also used to identify
sources of roll-yaw coupling and develop decoupling methods to achieve stable flight.
5.2 Methodology
System identification is performed using the NASA Langley’s System Identification
Programs for Aircraft (SIDPAC) [81]. SIDPAC is a software package containing
MATLAB scripts and functions of system identification tools. It is capable of per-
139
Figure 5.1: System Identification using time-domain approach.
forming both frequency-domain and time-domain analysis of flight data. In addition,
it contains additional tools for real-time parameter estimation, model structure de-
termination, input design, and data analysis. The first step in system identification
is applying engineering simplifications and prior knowledge of the dynamic system
to predict the model structure (Fig. 5.1).
5.2.1 Model Structure
As discussed in the previous chapter, a state-space model is considered in the general
form:
ẋ = Ax+Bu (5.1)
Matrix A is composed of the stability derivatives that describe the homogeneous
system dynamics and matrix B contains the control derivatives. System Identifica-
tion techniques provide an estimation of these stability and control derivatives. The
state vector is denoted by x = [u v w p q r φ θ] while the input vector is denoted
140
200 50
0
-200 0
11 11.5 12 12.5 13 13.5 14 18 19 20 21 22 23 24
50 50
0 0
-50 -50
11 11.5 12 12.5 13 13.5 14 18 19 20 21 22 23 24
100 100
0 0
-100 -100
11 11.5 12 12.5 13 13.5 14 18 19 20 21 22 23 24
3 3
2 2
1 1
0 0
11 11.5 12 12.5 13 13.5 14 18 19 20 21 22 23 24
Time (s) Time (s)
(a) Longitudinal doublet input (b) Phasing input
200 200
0 0
-200 -200
11 11.5 12 12.5 13 13.5 14 11 11.5 12 12.5 13 13.5 14 14.5 15
50 100
0 0
-50 -100
11 11.5 12 12.5 13 13.5 14 11 11.5 12 12.5 13 13.5 14 14.5 15
50 1
0 0
-50 -1
11 11.5 12 12.5 13 13.5 14 11 11.5 12 12.5 13 13.5 14 14.5 15
Time (s) Time (s)
(c) Decoupled longitudinal mode (d) Effect of δlat on lateral velocity
200 200
0 0
-200 -200
11 11.5 12 12.5 13 13.5 14 14.5 15 13 13.5 14 14.5 15 15.5 16 16.5 17
50 50
0 0
-50 -50
11 11.5 12 12.5 13 13.5 14 14.5 15 13 13.5 14 14.5 15 15.5 16 16.5 17
20 100
0 0
-20
-100
11 11.5 12 12.5 13 13.5 14 14.5 13 13.5 14 14.5 15 15.5 16 16.5 17
100 50
0 0
-100 -50
11 11.5 12 12.5 13 13.5 14 14.5 15 13 13.5 14 14.5 15 15.5 16 16.5 17
Time (s) Time (s)
(e) Coupled response for lateral doublet input (f) Coupled response for yaw doublet input
Figure 5.2: Time history data for pilot excitation and vehicle response.
by u = [δlat δlon δthrottle δped δphase]. States include velocities (u, v, w), angular rates
(p, q, r), and Euler attitude angles (φ, θ). δlat refers to the differential rpm of the
141
Yaw Heading Roll Lateral Heading Roll Longitudinal Pitch Pitch Longitudinal
rate (deg/s) angle (deg) angle (deg) input (µs) angle (deg) angle (deg) input (µs) Longitudinal rate (deg/s) angle (deg) input (µs)
velocity (m/s)
Roll Yaw Heading Yaw Roll Lateral Pitch
rate (deg/s) input (µs) Lateral
Pitch
angle (deg) angle (deg) rate (deg/s) input (µs) Longitudinal rate (deg/s)
Phase angle
velocity (m/s) angle (deg)velocity (m/s) input (deg)
Table 5.1: Comparison between frequency-domain and time-domain methods [84].
Frequency-domain methods Time-domain methods
Frequency responses (derived from Time-history data
time-history data)
Frequency sweep with long flight test Multistep (eg. doublets, 3-2-1-1) with
records shorter record lengths
Bias effects of noise eliminated from Requires noise model (e.g equation er-
analysis ror or output error)
Direct identification of time delays Not identified directly
No biases needed to be identified Must be identified and can be corre-
lated with aerodynamics parameters
two cyclorotors, δlon is the rpm control of the propeller, δthrottle collectively varies
rpm of all three rotors, δped is differential thrust vectoring, and δphase is the phasing
of the cyclic pitch. These inputs are in microseconds and represent the duty cycle
of the PWM signal transmitted to the actuators. PWM signals range from 1000–
2000 µsec. Pilot inputs are combined with initial PWM values associated with trim
condition such that δlat, δlon, and δped ∈ [−300, 300]µsec, δthrottle ∈ [0, 1500]µsec,
and δphase ∈ [0, 60]µsec. For comparison with other existing flight dynamics model
of MAV platforms, inputs have been scaled to u ∈ [−1, 1]. As a result, the control
derivatives can be compared across vehicle platforms [82,83].
The most significant simplification made for the model is that the cyclorotors
are governed by rigid-body dynamics. In addition, the longitudinal and heave modes
are independent and decoupled from the rest of the vehicle dynamics. Due to re-
strictions in the test volume, system identification was not performed to determine
heave dynamics in forward flight.
142
5.2.2 Time-Domain Approach
Table 5.1 compares frequency-domain and time-domain approaches in system iden-
tification. The time-domain approach was used because of restrictions in maneuver
time and limitations in performing frequency sweeps. The frequency-domain ap-
proach using spectral estimation generally requires a large amount of flight data
with pilot input excitation across multiple frequency ranges. In addition, since the
time-domain approach uses time-history data, the results can be directly interpreted
and applied to the physical system. While the advantages of using frequency-domain
methods include elimination of noise, direct identification of time delays, and ap-
plicability to unstable systems, performing Fourier transforms to acquire the initial
frequency response data may introduce errors and reduce accuracy of modeling re-
sults.
Within the time-domain, there are two main parameter estimation techniques:
(1) Equation-error method and (2) Output-error method. The equation-error mini-
mizes the sum of squared differences between the measured and modeled state time
derivatives while output-error minimizes the sum of weighted squared differences
between measured and modeled outputs [81]. One disadvantage of the output-error
method is that it involves integrating differential equations for each iteration and
may lead to divergence in parameter estimation for unstable systems. The Equation-
error method is used in this research because it does not require integration of the
state equations and will converge to some parameter estimation. Before using the
equation-error method, an initial model structure was identified using Stepwise re-
143
gression. Both methods are summarized in the following sections.
5.2.3 Stepwise Regression
System identification techniques used in this research results in a reduced linear
regression model where one or more regressors are removed from the model [81].
The output of the linear system (y) can be represented by columns of regressors
(X) that are a function of the state vectors (x) and input vectors (u), and the
parameter vectors (θ) that are calculated through system identification:
y = ẋ = X(x, u)θ (5.2)
Measurement of the linear system is represented by the measured output (z) and
random noise (v):
z = y + v (5.3)
z = Xθ + v (5.4)
The relevance of each candidate regressor for the reduced linear regression
model is evaluated by either adding or deleting regressors one at a time. The
technique of adding regressors, while starting without any in the model, is called
forward selection. The technique of deleting regressors, while starting with the entire
pool of candidate regressors, is called backward elimination. Stepwise regression
is a computation technique that combines both forward selection and backward
144
elimination so that all regressors in the model are continuously re-evaluated. As a
result, regressors that become redundant can be removed from the model.
The method starts by constructing a pool of candidate regressors. Adjusting
for the mean value or bias (X0), the first regressor selected is the one with the highest
simple correlation with the dependent variable (z). The correlation coefficient (rjz)
for each regressor (ξ) is calculated using the following equation for j = 1,2,...n and
N number of data points:
∑N
i=1[ξj(i)− ξ̄j][z(i)− z̄]rjz = √ (5.5)
SjjSzz
where Sjj is the centered sum of squares for the regressor (ξj) and Szz is the centered
sum of squares for the measured response (z):
∑N
S 2jj = [ξj(i)− ξ̄j] (5.6)
i∑=1N
Szz = [z(i)− z̄]2 (5.7)
i=1
After selecting the first regressor (ξ1), the regressor is removed from the pool
of candidate regressors and the model is used to fit the measured response through
ordinary least-squares parameter estimation. The metric that quantifies the close-
ness of estimated model values (ŷ) to measured values (z) is called the coefficient of
determination, R2:
2 SSRR = (5.8)
SST
145
where 0 ≤ R2 ≤ 1, SSR is the regression sum of squares, and SST is the total sum
of squares:
∑N
SSR = [y(ˆi)− z̄]2 (5.9)
i=1
SST = z
T z −Nz̄2 (5.10)
Based on the equation, each new model term increases the R2 value. If the
value of R2 changes by less than 0.05, then the regressor is deemed statistically
insignificant and is omitted from the model. Additional regressors are added to the
model if they satisfy the partial F statistic criterion for each estimated parameter
vector (θ̂) for j = 1,2,...n:
θ̂2
Fp = (5.11)
V ar(θj)
The Fp values indicate the significance of an individual parameter to the overall
model. For a 95% confidence interval, regressors with Fp < 20 are removed from
the model.
5.2.4 Equation-Error Method
After selecting the column of regressors (X), the model is refined using the equation-
error method, which calculates the potential correlation of each regressor based on
a linear least-squares formulation. The parameters (θ) are estimated by minimizing
the sum of squared differences between the model and measurements in the following
cost function:
146
1
J(θ) = (z −Xθ)T (z −Xθ) (5.12)
2
The best parameter estimate (θ̂) is one that minimizes the cost function and satisfies
the following:
θ̂ = (XTX)−1XT z (5.13)
The estimated linear model output (ŷ) is then calculated by:
ŷ = Xθ̂ = X(XTX)−1XT z (5.14)
The difference between the estimated value (ŷ) and measured value (z) is called the
residual (v):
v = z − ŷ = z −Xθ̂ (5.15)
Residuals are orthogonal to the column of regressors (X). If the columns are linearly
dependent, the matrix inverse (XTX)−1 and the least-squares solution will not exist.
Therefore, the model structure and the experiment should be carefully executed.
5.3 Experimental Setup
Characterizing the flight dynamics model using system identification techniques re-
quires a series of pilot input excitation experiments and corresponding vehicle mea-
surements. During the experiment, multiple lateral, longitudinal, and yaw doublet
inputs were provided in separate flights. The amplitude of the doublets were se-
lected such that the vehicle response has an acceptable signal-to-noise ratio without
147
causing a critical crash-inducing response. Over 60 flight tests were conducted in
a 525m3 test volume. Maneuvers were only initiated after the vehicle achieved a
steady state of 2 m/s forward speed and a level attitude (zero pitch Euler angle).
Capturing pilot inputs and vehicle responses within the limited test volume
was challenging as the inputs were provided only after achieving a steady cruise.
Inputs were also limited to doublets instead of a frequency sweep because of: (1)
limited flight testing time in the test space, (2) difficulty for the cyclocopter to
maintain steady flight conditions at low frequency inputs, and (3) poor signal-to-
noise ratio at lower frequencies. The following sections will provide a description of
the experimental methodology and sensor instrumentation.
Figure 5.3: Vicon motion capture testing setup.
148
5.3.1 Data Acquisition Setup
The first step towards identifying the flight dynamics model is to systematically
collect flight data. Both onboard instrumentation and external sensors are simulta-
neously used to collect position, velocity, and attitude data. The ViconTM Motion
Capture System consists of eight high speed infrared cameras that track the retro-
reflective markers placed on the vehicle (Fig. 5.3). The motion capture system
triangulates the location of the markers inertially within the three-dimensional test
volume. These infrared cameras capture the translational and rotational states of
the vehicle at a frequency of 100 Hz.
Figure 5.4: Synchronization between ELKA and Vicon measurements us-
ing a fiducial perturbation in pitch.
As discussed in chapter 2, the onboard instrumentation consists of ELKA, an
in-house developed micro embedded processor-sensor board weighing 1.3 grams [78].
149
Figure 5.5: Data flowchart between ELKA and Vicon.
ELKA houses a STM-32 microprocessor with a 32-bit ARM Cortex M3 core for high-
end onboard computation tasks and performs all the data processing, vehicle stabi-
lization, and roll-yaw decoupling computations. The MPU-9150 IMU integrated on
the board includes tri-axial gyroscopes, accelerometers, and magnetometers. Wire-
less communications are serviced by an onboard nRF24L01 chip, a low-power 2.4
GHz RF transceiver. ELKA has a sensor update rate of 500 Hz and is capable of
streaming vehicle attitude data and actuator controls data to the base station with
a latency upper bound of 30 ms. Pitch (q), roll (p), and yaw (r) attitude rates are
provided by tri-axial gyroscopes and combined with accelerometer data through a
gradient descent method to produce precise attitude measurements [54].
Measurements from ELKA are streamed to the ground station through a wire-
less IEEE 802.15.4 data link while measurements from Vicon are streamed through
a network TCP/IP port. At the end of each experiment, a fiducial pitch pertur-
bation was manually provided to the vehicle on the ground in order to synchronize
the Vicon and ELKA data streams (Fig. 5.4). Both data sets were then allocated
150
a common time stamp after post-processing. System identification measurements
include translational states (x, y, z) from Vicon and rotational states (p, q, r, θ, φ)
from ELKA (Fig. 5.5).
5.3.2 Sampling Rate
The general rule for selecting the sampling rate is [81]:
fs = 25fmax (5.16)
where fmax is the maximum rigid-body frequency. For the cyclocopter, the rigid-
body dynamic modes occur at 4 Hz so the sampling frequency was selected as 100
Hz. As previously mentioned, the ELKA onboard processor measures the vehicle
states at 500 Hz compared to 100 Hz from the Vicon motion capture system. As
part of post-processing, measured states from ELKA were down-sampled in order
to synchronize with Vicon measurements.
5.3.3 Bare-Airframe Dynamics
System identification is typically conducted on an open-loop system to determine
the bare-airframe characteristics of a vehicle. However, since the cyclocopter is
dynamically unstable, the closed-loop feedback system remained implemented with
low control gains during the experiments. Furthermore, any decoupling techniques
and mixing ratios were eliminated during the experiment in order to capture true
vehicle dynamics in forward flight.
There are two conventional methods of extracting the bare-airframe dynamics
151
from a closed-loop system: (1) having a priori knowledge of control law to deduce the
closed-loop dynamics matrix or (2) directly measuring actuator inputs to the rotors
as a result of both pilot and onboard stabilization. Since the actuator inputs were
directly available from onboard measurements, the second approach was taken. Data
collected from a closed-loop system could potentially introduce correlation between
input excitation and output noise [84]. To circumvent this issue and improve signal-
to-noise ratio, input excitations were restricted to low amplitudes.
5.3.4 Pilot Excitation
During the experiment, all decoupling methods were removed in order to observe
the vehicle’s inherent roll-yaw coupled responses. The longitudinal and heave modes
were assumed to be independent and decoupled. In addition, the heave mode was as-
sumed to be identical to that in hover conditions. These assumptions were validated
from the flight test results (Fig. 5.6).
Flight tests were systematically conducted by first providing a pure throttle
command and observing the vehicle response. Any forward translation was coun-
teracted by either reducing the phasing of the cyclic blade pitch or varying the
propeller rpm. After trimming the cyclocopter, level forward flight is then initiated
solely through thrust vectoring of the cyclorotors.
All maneuver inputs were provided after the cyclocopter achieved a steady
state of 2 m/s flight at a steady, level attitude (θ, φ, ψ=0). From the time history
data, increasing the phasing to Φ = 45◦ with a δphase step input (time = 20 seconds)
152
200 50
0
-200 0
11 11.5 12 12.5 13 13.5 14 18 19 20 21 22 23 24
50 50
0 0
-50 -50
11 11.5 12 12.5 13 13.5 14 18 19 20 21 22 23 24
100 100
0 0
-100 -100
11 11.5 12 12.5 13 13.5 14 18 19 20 21 22 23 24
3 3
2 2
1 1
0 0
11 11.5 12 12.5 13 13.5 14 18 19 20 21 22 23 24
Time (s) Time (s)
(a) Longitudinal doublet input (b) Phasing input
200 200
0 0
-200 -200
11 11.5 12 12.5 13 13.5 14 11 11.5 12 12.5 13 13.5 14 14.5 15
50 100
0 0
-50 -100
11 11.5 12 12.5 13 13.5 14 11 11.5 12 12.5 13 13.5 14 14.5 15
50 1
0 0
-50 -1
11 11.5 12 12.5 13 13.5 14 11 11.5 12 12.5 13 13.5 14 14.5 15
Time (s) Time (s)
(c) Decoupled longitudinal mode (d) Effect of δlat on lateral velocity
200 200
0 0
-200 -200
11 11.5 12 12.5 13 13.5 14 14.5 15 13 13.5 14 14.5 15 15.5 16 16.5 17
50 50
0 0
-50 -50
11 11.5 12 12.5 13 13.5 14 14.5 15 13 13.5 14 14.5 15 15.5 16 16.5 17
20 100
0 0
-20
-100
11 11.5 12 12.5 13 13.5 14 14.5 13 13.5 14 14.5 15 15.5 16 16.5 17
100 50
0 0
-100 -50
11 11.5 12 12.5 13 13.5 14 14.5 15 13 13.5 14 14.5 15 15.5 16 16.5 17
Time (s) Time (s)
(e) Coupled response for lateral doublet input (f) Coupled response for yaw doublet input
Figure 5.6: Time history data for pilot excitation and vehicle response.
also causes a step change in velocity (Fig. 5.6(b)). During the acceleration phase,
the initial response is large and acceleration begins to decrease with time. When
153
Yaw Heading Roll Lateral Heading Roll Longitudinal Pitch Pitch Longitudinal
rate (deg/s) angle (deg) angle (deg) input (µs) angle (deg) angle (deg) input (µs) Longitudinal rate (deg/s) angle (deg) input (µs)
velocity (m/s)
Roll Yaw Heading Yaw Roll Lateral Pitch Pitch
angle (deg) rate (deg/s) angle (deg) input (µs)
Lateral rate (deg/s) input ( s) Longitudinal rate (deg/s) Phase angleµ
velocity (m/s) angle (deg)velocity (m/s) input (deg)
the phase input is returned, the flight speed returns to the original forward velocity
at the same rate.
During the experiment, multiple lateral, longitudinal, and yaw doublet inputs
were provided in separate flights. Low amplitude doublets were selected to keep
resulting vehicle dynamics within the assumed linear range of operation and to
achieve acceptable signal-to-noise ratio without causing a critical crash-inducing
response. Higher amplitude responses generally tend to become more nonlinear,
which affects the linear coherence between the input and output data.
5.3.5 Vehicle Response
Another observation from flight test results is the existence of a dominating roll-
yaw coupling for both lateral and yaw inputs. The coupling also causes a secondary
response from the feedback system, which leads to data collinearity. The adverse
effects of data collinearity can be suppressed by omitting the secondary regressors
from the regressor pool. For example, a positive roll input induces a clockwise yaw
response, which leads to a counterclockwise yaw actuator input from the feedback
system. Therefore, δyaw would be present in the regressor pool even though only a
roll input was initially actuated. In these experiments, secondary regressors were
be manually omitted to improve accuracy of parameter estimation.
154
Table 5.2: Parameter estimation results.
Parameter Value F-ratio
Longitudinal Mode
Xu -0.021 51
Xphase 1.55 125
Mu 1.97 31
Mq 0.009 22
Mlon -83.1 96
Lateral and Yaw Modes
Yv -0.036 68
Yp -0.001 115
Yr 0.003 242
Lv 4.29 54
Lp 0.097 53
Lr 0.056 194
Llat 76.4 31
Lrud 114 91
Nv 1.14 109
Np 0.051 32
Nr 0.018 149
Nlat -53.9 34
Nrud 19.5 23
155
Figure 5.7: Model structure for cyclocopter in forward flight.
5.4 Results
The following sections discuss results of the system identification from Vicon experi-
ments, including model structure determination, parameter estimation, and stability
analysis.
5.4.1 Model Structure Determination
The linear model structure was first determined using stepwise regression and the
corresponding stability and control derivatives were identified in the matrix form
provided in Fig. 5.7. From the pool of regressors, the identified model parameters
were selected based on the coefficient of determination (R2) and F-ratio (Fp). The
R2 values exist within [0,1] and increase with the addition of regressors. If the
value changes by only 0.05, then the regressor is deemed statistically insignificant
and is omitted from the model. All of the identified modes had acceptable R2
values between 0.65–0.93. The Fp values indicate the significance of an individual
parameter to the overall model. For a 95% confidence interval, regressors with
Fp < 20 were not considered. In addition, the parameters were corrected for colored
residuals. Overall, the final model parameter estimates using the equation-error
156
(a) Angular acceleration in pitch (b) Longitudinal acceleration
(c) Angular acceleration in roll (d) Lateral acceleration
(e) Angular acceleration in yaw
Figure 5.8: Estimated model output compared to flight data.
method are provided in Table. 5.2.
Fig. 5.8 shows correlation between model predicted results and a different set
157
of flight test data. Limitations in model input and total maneuver flight time were
from the restricted motion capture testing volume and length of time it took to reach
the steady state. Considering these challenges, the estimated model parameters are
able to adequately characterize the vehicle dynamics.
5.4.2 Longitudinal Mode
The longitudinal dynamics are identified as:
    

u̇   Xu 0 −g  u 
  q̇ 

=      Mu 0 0  q  (5.17)
θ̇ 0 1 0 θ
  0 X   phase + M 0 

lon 
δlon
 

δphase
0 0
Compared to the flight dynamics model derived in hover conditions, the only new
model parameter in forward flight is the control derivative, Xphase. As expected,
Xphase is positive because phasing input leads to longitudinal translation. The
strength of the control derivative confirms that forward speed is highly sensitive
to phasing input. Similar response was observed during flight tests. The negative
longitudinal translation derivative (Xu) indicates positive aerodynamic damping.
Therefore, the vehicle is able to stabilize a velocity perturbation in the longitudinal
direction.
158
The longitudinal coupling term (Mu) is stabilizing. A perturbation increas-
ing the longitudinal velocity will induce a nose-up pitching moment. The positive
pitching moment would further decrease the forward velocity, stabilizing the cyclo-
copter. Likewise, a velocity perturbation from the opposite direction would cause
the cyclocopter to pitch-down and increase forward speed. Compared to the other
longitudinal stability derivatives, Mq is almost insignificant in magnitude and re-
moved from the model. Therefore, the aerodynamic pitch damping is non-existent
and the vehicle requires feedback damping to stabilize the pitch attitude rate.
5.4.3 Coupled Lateral-Yaw Modes
The lateral and yaw dynamics are identified as:
    

v̇
 
Y 0 Y g v
 ṗ  
 v r  L L
  v p
Lr 0 
  =  
 p 

    
 (5.18)
 ṙ   Nv Np Nr 0  r 
φ̇ 0 0 0 1 φ
  0 0 
 Llat Lrud 
δlat 
+    Nlat Nrud  δyaw
0 0
Similar to the longitudinal translation derivative, the lateral translation derivative
(Yv) is stabilizing. However, the lateral and yaw coupling terms, Lv and Nv are
159
destabilizing. In forward flight, a velocity perturbation would induce both roll and
yaw responses. In particular, perturbation to the right induces a clockwise yaw
and a roll-right moment, which would further increase the velocity. Based on Yr, a
clockwise yaw results in a translation to the right, further destabilizing the vehicle.
In addition, both Lp and Nr are unstable.
The most significant observation from the model structure is the contribution
of Lr and Lrud to the lateral state and Np and Nlat to the yaw state. This is the
dominating cross-coupling effect unique to the cyclocopter. The roll-yaw couplings
and their respective decoupling methods are described in details in the following
subsections.
5.4.3.1 Gyroscopic Coupling
As seen from the model and flight tests, the cyclocopter dynamics are dominated
by gyroscopic roll-yaw cross-coupling. This coupling is caused by the unbalanced
angular momentum from the cyclorotors spinning in the same direction. For a
positive roll input, the vehicle responds in both roll right and counterclockwise yaw
(Fig. 5.9). Likewise, for a positive yaw input, the vehicle responds in clockwise yaw
and also rolls to the right.
In the present vehicle, the cross-coupling is decoupled by mixing roll and yaw
inputs in the controller. To achieve a pure roll right response, a roll right input
is combined with a clockwise yaw input to cancel out the inherent cross-coupling
response. Likewise, for clockwise yaw, the input is appropriately combined with a
160
(a) Roll right control input. (b) Roll-yaw response.
(c) Yaw control input. (d) Yaw-roll response.
Figure 5.9: Gyroscopic cross-coupling between roll and yaw degrees of
freedom.
roll left input. The mixing ratio is only applied to the open-loop pilot input and
directly implemented on the onboard controller.
The gyroscopic mixing terms were experimentally found by suspending the
cyclocopter about c.g. on a 3-DOF gimbal stand unrestrained in pitch, roll, and
yaw. Starting with zero mixing, the constants for both roll and yaw inputs were
slowly increased until the vehicle achieved a decoupled response. Experimental
results showed a stronger coupling for a roll input, which results in higher mixing
ratios. One reason for a stronger roll-yaw coupling could be that the moment of
inertia about the x-axis is greater than that about the z-axis.
161
(a) Roll right control input. (b) Roll-yaw response.
(c) Yaw control input. (d) Yaw-roll response.
Figure 5.10: Controls cross-coupling between roll and yaw degrees of free-
dom occurring at high phase angles.
5.4.3.2 Controls Coupling
Existence of controls coupling at high phase angles is summarized in Fig. 5.10. The
phenomenon can be physically explained by considering the magnitude of vertical
(lift) and horizontal (thrust) forces generated by the cyclorotor at high phase angles.
As the phase angle increases, the thrust vector is tilted towards the +X-axis and
thus thrust increases while lift decreases. A differential change in rotational speed
would have a greater effect on yaw than in roll. Likewise, thrust vectoring would
effectively cause a roll response.
162
Table 5.3: Eigenvalue Comparison
Mode Eigenvalues
Hover Forward Flight
Longitudinal Translation -3.83 -2.69
Longitudinal Rotation 1.66±3.05i 1.33±2.32i
Lateral-Yaw Translation -2.29 0.133
1.48 1.01
Lateral-Yaw Rotation 0.28±8.02i -0.027±0.046i
As mentioned in section 4.7, the mixing scheme for controls coupling is as follows:
δφ = [bφ δlat + bφ δped]sinΦ (5.19)b lat ped
δψ = [bψ δlat + bψ δped]sinΦ (5.20)b lat ped
The mixing ratios were identified in the wind tunnel. In the previous experimental
study, the mixing ratios were iterated until the vehicle dynamics were decoupled at
each phasing input. After implementing the decoupling strategy, the vehicle was
able to achieve decoupled response in forward flight.
5.4.4 Stability analysis in forward flight
The eigenvalues of the system provide a better understanding of the vehicle flight
dynamics and have been summarized in Table. 5.3. The eigenvalues in hover were
computed in a previous study using the same system identification techniques [39].
It is interesting to note that the eigenvalues for the longitudinal and lateral-yaw
163
4 Longitudinal Translation
Longitudinal Rotation
Lateral-Yaw Translation
2 Lateral-Yaw Rotation
0
-2
-4
-4 -2 0 2 4
Real Axis
Figure 5.11: Pole location for the cyclocopter in forward flight.
translation modes are similar between the hover and forward flight cases.
The cyclocopter shows highly oscillatory and unstable longitudinal and lateral-
yaw response in hover and forward flight. The lateral-yaw rotation mode is more
unstable in forward flight possibly due to the combined roll-yaw cross-couplings
previously mentioned. Unique to forward flight, the cyclocopter experiences both
gyroscopic and controls coupling. Based on the locations of the system poles, the
overall bare airframe dynamics of the cyclocopter is inherently unstable (Fig. 5.11).
Therefore, the cyclocopter requires feedback regulation and roll-yaw controls mixing
to achieve trimmed flight.
5.5 Summary
In this chapter, a flight dynamics model in forward flight was developed using sys-
tem identification techniques. The 6-DOF model allowed assessment of the vehicle’s
susceptibility to velocity perturbation (gust) along the x-axis. The flight dynamics
164
Imaginary Axis
model was extracted using the time-domain approach from SIDPAC system identi-
fication tools. The model was based on measurements from doublet inputs across
various modes. The model structure was initially identified using stepwise regression
and the parameters were estimated using the equation-error method. All maneuver
inputs were provided after the cyclocopter achieved a steady state of 2 m/s flight at
a steady, level attitude (θ, φ, ψ=0).
The existence of the control derivative, Xphase showed that phasing input com-
manded forward speed. From flight tests, it was observed that increasing the phasing
with a δphase step input directly resulted in a step change in velocity. During the
acceleration phase, the initial response was large and acceleration decreased with
time until the vehicle reached steady state after two seconds. When the phase in-
put was removed, the flight speed returned to the original forward velocity at the
same rate. The negative longitudinal translation derivative, Xu indicated positive
aerodynamic damping. Therefore, the vehicle was able to stabilize a velocity pertur-
bation in the longitudinal direction. However, the aerodynamic pitch damping (Mq)
was non-existent and the vehicle required feedback damping to stabilize the pitch
attitude rate. In general, the longitudinal mode was decoupled and independent of
the rest of the modes. The longitudinal coupling term, Mu was stabilizing. A per-
turbation that increased the longitudinal velocity would induce a nose-up pitching
moment. The positive pitching moment would further decrease the forward velocity,
stabilizing the cyclocopter.
The lateral translation derivative, Yv was also stabilizing. However, the lateral
and yaw coupling terms, Lv and Nv were unstable. In forward flight, a velocity per-
165
turbation would induce both roll and yaw responses. Based on Yr, a clockwise yaw
resulted in a translation to the right, further destabilizing the vehicle. In addition,
both Lp and Nr were unstable. The model was able to validate the existence of the
inherent roll-yaw coupling in forward flight, which was identified by contributions
of Lr and Lrud to the lateral state and Np and Nlat to the yaw state. These are the
result of combined gyroscopic coupling and controls coupling at high phase angles.
Overall, the gyroscopic coupling is caused by the unbalanced angular momentum
from the cyclorotors spinning in the same direction. The controls coupling is due
to the increase of magnitude of propulsive forces at high phase angles. As a result,
thrust vectoring has a larger impact on the roll moments and subsequently, changes
in rotor RPM have a greater influence on yaw moments.
166
Chapter 6: Gust Rejection Study
6.1 Overview
The previous chapter presented a linear state-space model of the cyclocopter in
forward flight extracted through series of system identification experiments. The
flight dynamics model showed that phasing input (thrust vectoring) had a high
control authority over airspeed. In addition, a perturbation in the longitudinal
velocity induced a stabilizing nose-up pitching moment. The positive pitching mo-
ment decreased the vehicle’s airspeed back to desired condition. In this chapter, the
disturbance rejection capability of the cyclocopter is investigated through series of
experiments in the wind tunnel and using a synthetic gust generation device. The
cyclocopter was first mounted on 1-DOF experimental stand in front of an open-jet
wind tunnel to evaluate response to gust perturbation along the longitudinal and
lateral axis. Since the cyclorotor has the unique capability of providing both lift
and propulsive thrust by phasing the blade pitch cyclic, it is possible to reject gust
disturbance isolated along the longitudinal axis using thrust vectoring. In addition,
feedback control schemes of position feedback using a motion capture system and
flow feedback using onboard flow sensing are also evaluated. Results of the com-
parative study of gust rejection strategies are provided for the 6-DOF experiments
167
Figure 6.1: Pitch control vs. thrust vectoring control.
in front of a synthetic gust generation device for step gust inputs up to 4 m/s and
6 seconds in duration. Results also include performance for crosswinds up to 30
degrees from the longitudinal axis.
6.2 Gust Rejection Strategies
Due to their compact size and low inertia, MAVs are highly susceptible to gusts
when operating in urban settings. For an edgewise gust, MAVs experience both
drag force in the direction of the flow and pitching moment that rotates the vehicle
away from the source. Possible methods to alleviate gust effects include increasing
the vehicle’s control authority and improving the closed-loop feedback system. The
gust rejection strategies are evaluated through series of tethered experiments in the
wind tunnel and in front of a custom gust generation device.
6.2.1 Control Input
Conventional rotary-winged MAVs like quadcopters rely on differential rpm to re-
cover from gust perturbation. Through this control input, the vehicle achieves a
168
non-zero pitch attitude angle, thereby increasing its effective surface area and po-
tentially reducing its gust rejection capability. For the cyclocopter configuration,
there are two possible gust rejection control strategies for disturbance along the
longitudinal axis: (1) pitch control by actuating nose rotor rotational speed and (2)
thrust vectoring the cyclorotors (Fig. 6.1).
As introduced in previous chapters, the cyclocopter has the unique capability
of thrust vectoring by phasing the blade pitch cyclic. The resulting thrust vector
has both a vertical lift component and a horizontal propulsive thrust component.
By vectoring the resultant thrust to increase its propulsive thrust component, the
cyclocopter is able to maintain a zero pitch attitude angle in forward flight and
independently translate forward and backward without adjusting the vehicle’s pitch
angle.
6.2.2 Feedback System
An intuitive control scheme to negate gust effects is to rely on inertial measurements
coupled with closed-loop position feedback. The drag force and pitching moment
experienced by the cyclocopter would affect body angular rates and displacement.
Inertial measurements are then used to calculate control inputs required to recover
initial position and attitude. With position feedback, the gust controller only comes
into effect after the vehicle has been perturbed.
An alternative control scheme is to rely on real-time flow measurement in order
to directly measure disturbances. The flow feedback controller would then predict
169
Figure 6.2: Multi-hole flowprobe system.
impending vehicle response and provide control inputs to negate gust effects. Unlike
position feedback, flow feedback could potentially enable the vehicle to maintain its
initial attitude throughout the gust perturbation. However, the control scheme
is sensitive to the choice of onboard miniature flow sensor and requires a priori
knowledge of control inputs suitable for measured disturbances.
6.3 Real-time Flow Measurement
A number of approaches were considered to measure freestream velocity for low
operational range. A hot-wire or ultra-sonic anemometer offers high resolution and
3D measurement of velocity, but has a high weight penalty making them unsuitable
for MAVs. A mechanical approach that uses miniature rotating vanes or a turbine
would provide adequate resolution at low speeds, but will be mechanically complex
and sensitive to vibrations observed on the cyclocopter. A pressure-probe based
approach such as traditional pitot-probes could be easily implemented at the MAV-
scale but are prone to poor signal-to-noise ratios at low operating airspeed.
170
Figure 6.3: Flowprobe calibration compared to hot-wire and rotary
anemometer.
An alternative pressure-probe based approach relies on measuring differential
pressure. A traditional pitot-static probe includes a forward-facing port for total
pressure and a ring of inter-connected static ports perpendicular to the flow. The
dynamic pressure is measured as the difference between the total pressure and the
static pressure. While this allows for a probe that is less directional, a relatively
large forward velocity is needed before a clear signal is generated. The current
setup uses a two-component flowprobe design with one opening facing towards the
freestream and other against the flow in order to provide a differential pressure
measurement. Any forward velocity causes a degree of pressure loss on the aft-
facing port, enabling the probe to provide an amplified measurement of dynamic
pressure with higher sensitivity at low speeds.
171
Freestream velocity was continuously measured throughout the experiment
using an analog Honeywell HSCDRR series differential pressure sensor. The sensor
measures pressure range of ±249.08Pa and has an input voltage of 3.33 V. The
analog voltage measurements from the differential pressure sensor are converted to
ADC (analog-to-digital) values on ELKA at 1000 Hz with a 32-bit resolution. Dual
ports on the sensor are individually connected to an opening on the flowprobe.
The pressure measurements from the sensor are converted to velocity using a scaled
version of pitot static equations [85]. A multi-probe setup with flowprobes along
X-Y axes (Fig 6.2) was used to enable 2D velocity measurement. The components of
velocity are calculated using calibration gains (Ku, Kv, Kb), pressure measurements
(X1, X3, Y2, Y4), and density (ρ):
√
2Kb(X1 −X3)
u = Ku (6.1)
ρ
√
2Kb(Y2 − Y4)
v = Kv (6.2)
ρ
Symmetry of the bi-directional flow probe resulted in equivalent calibration
gains (Ku = Kv). With this setup, the total velocity magnitude (V) and side-slip
angle (β) can be calculated using:
√
V = u2 + v2 (6.3)
v
β = tan−1 (6.4)
V
172
Figure 6.4: Power spectral density estimate via Welch (contributions from
the flow field).
Figure 6.5: Power spectral densities at different airspeeds (hover, 1 m/s,
and 5 m/s).
Flowprobe measurements were calibrated against measurements from both a
hot-wire anemometer and rotary anemometer in the wind tunnel from 1 m/s to
8 m/s. Figure 6.3 shows velocity measurements of all three sensors. Additional
calibration was experimentally performed with the probes mounted on the vehicle
173
to account for the flow field disturbances specific to a cyclocopter. In order to
filter high frequency noise, the wind tunnel pressure data is first characterized by its
power spectral density (PSD). By comparing the power spectrum with the cyclorotor
powered down and with the rotors spinning, contributions from the flow field and
the vehicle can be independently identified. Fig. 6.4 shows the PSD of the pressure
data with the cyclorotors completely powered down. From the figure, it is evident
that there are no discernible modes in the low-turbulence freestream because the
flow far downstream of the wind tunnel drive system does not have any identifiable
characteristic frequencies. Therefore, any observed modes in the power-on case are
caused solely by the noise from the cyclorotors. The PSDs for pressure data at three
different velocities are shown in Fig 6.5.
Figure 6.6: Filtered pressure signal statistics.
Harmonics are noted at 135 Hz and 200 Hz, which suggests that either the pres-
174
Figure 6.7: 1-DOF pitch stand.
sure transducer measurements are being affected more by probe/vehicle structural
response than actual flow or by the forward mounted propeller rotor. In addition,
a harmonic near 35 Hz was noted with an incident free stream, suggesting that this
might be flow field dependent. As a result, a low pass filter with a cut-off freq of 40
Hz was chosen. The filtered measurements are compared with raw measurements in
Fig. 6.6.
6.4 Preliminary Wind Tunnel Studies
One method of evaluating the vehicle’s control authority is by conducting tethered
experiments in the open-jet wind tunnel (0.56 m × 0.56 m). The objective of the
wind tunnel experiments is to determine the cyclocopter’s response to an edgewise
flow and identify the maximum gust perturbation along both longitudinal and lat-
eral axis that the cyclocopter can overcome before reaching control saturation. In
particular, these experiments will evaluate the effectiveness of cyclocopter’s pitch
175
control.
6.4.1 Methodology
In order to isolate either the pitch or roll degree of freedom, the cyclocopter was
mounted on a 1-DOF gimbal stand at c.g (Fig. 6.7). Rotations about pitch and roll
were restricted to ±8◦. For each experiment, the control inputs are solely provided
by the PD feedback control discussed in Chapter 2 to recover the desired attitude.
At the beginning of each experiment, the cyclorotors operate at 1400 rpm while
the nose rotor operates at 4000 rpm. Trim values are continuously adjusted through
each experiment to achieve stable hover (θ,φ = 0). The inner-loop PD gains were
selected to minimize settling time after a gust perturbation. While the settling time
is dependent on the feedback control system, the maximum magnitude of the gust
input a vehicle can tolerate before experiencing control saturation is dependent on
its configuration. For each experiment, the magnitude of the gust input is varied by
increments of 0.5 m/s up to 10 m/s. In addition, freestream velocity is continuously
measured using the differential pressure sensor discussed in the previous section.
6.4.2 Results: Longitudinal Axis
A gust perturbation along the longitudinal axis resulted in a pitch-up moment on
the cyclocopter (Fig. 6.8). The pitch-up moment may be possibly attributed to
the following: (1) increase in thrust produced by the nose rotor combined with
increase in pitching moment and (2) drag force on the nose rotor combined with
176
Figure 6.8: Gust perturbation along longitudinal axis results in pitch-up
moment.
Figure 6.9: Variation of nose rotor thrust with rotational speed at various
freestream velocities.
177
Figure 6.10: Variation of pitching moment with rotational speed at various
freestream velocities.
Figure 6.11: 1-DOF pitch response for 10 m/s longitudinal ramp gust per-
turbation.
178
a low vehicle c.g. As the freestream velocity increases, the nose rotor thrust also
increases at each rotational speed (Fig. 6.9). Interestingly, the pitching moment
caused by blade flapping is non-negligible and also increases with freestream velocity
(Fig. 6.10). Yeo et al. [85] and Hoffmann et al. [86] also observed similar trends for
MAV-scale commercial propellers and accounted for blade flap in their feedback
controllers. Since the nose rotor is operating at 4000 rpm, its pitching moment at
higher freestream velocities will influence the cyclocopter’s pitch response to a gust
perturbation along the longitudinal axis.
Since the wind tunnel velocity is controlled by switching the power to the
motor, the resulting gust profiles are ramp inputs. The cyclocopter was able to
withstand a 10 m/s ramp gust perturbation and recover desired pitch attitude within
12 seconds before reaching control saturation (Fig. 6.11). As previously mentioned,
pitch control is solely provided by regulating the rotational speed of the horizontal
nose rotor. The feedback controller initially reduces the nose rotor rpm to counteract
the induced pitch-up moment. However, the cyclocopter overshoots the desired
pitch attitude and is pitched down towards the 10 m/s freestream. Because of the
increased drag force on the vehicle, the feedback controller peaks at 90% of positive
pitch input to stabilize the cyclocopter. Throughout the experiment, PD gains were
consistently adjusted to reduce overshoot and settling time.
179
Figure 6.12: 1-DOF roll stand
Figure 6.13: Gust perturbation along lateral axis results in roll-right mo-
ment.
180
Figure 6.14: 1-DOF roll response for a 6.5 m/s lateral ramp gust pertur-
bation.
Figure 6.15: 1-DOF roll response for a 7 m/s lateral ramp gust perturba-
tion.
181
6.4.3 Results: Lateral Axis
For the 1-DOF roll experiments, the cyclocopter was positioned in front of the wind
tunnel such that the freestream velocity aligned with the +Y-axis of the body frame
(Figs. 6.12). With this setup, a gust perturbation along the lateral axis results in a
roll-right moment (Fig. 6.13). The positive roll moment may be caused by the drag
force on the cyclorotor combined with a vertical c.g offset.
The threshold for the maximum gust perturbation is much lower along the
lateral axis (Fig. 6.14). The cyclocopter was able to withstand up to 6.5 m/s of
gust before experiencing control saturation. The cyclocopter achieves roll control
by differentially varying the rotational speed of both cyclorotors, which have a lower
maximum achievable rpm compared to the nose rotor. Before applying a gust per-
turbation, both the cyclorotors operate near 1400 rpm while the maximum possible
rotational speed is 1600 rpm. At 7 m/s, the feedback controller inputs the maximum
available negative roll input which increases the operating speed of the right cycloro-
tor to 1600 rpm and decreases the left cyclorotor to 1200 rpm (Fig. 6.15). Similar
to the 1-DOF pitch experiment, the cyclocopter overshoots the desired roll attitude
and the increased drag force on the vehicle combined with control saturation leads
to roll oscillation about the lateral axis (Fig. 6.15). Based on these preliminary
results, the cyclocopter has a higher control authority in pitch compared to roll for
gust disturbance rejection.
182
6.5 Tethered Experiments (6-DOF)
Preliminary wind tunnel experiments showed that a gust perturbation along the
longitudinal axis leads to a pitch up moment, which consequentially results in back-
wards translation. Pitch control enabled the cyclocopter to recover level attitude
for ramp gust profiles up to 10 m/s with a settling time of 12 seconds. However,
the 1-DOF experiments do not account for translation or secondary effects on roll
and yaw body rates. In addition, a slow ramp gust profile enables the controller to
respond earlier at a lower gust magnitude instead of the selected magnitude.
Tethered experiments using a custom gust generation device and 6-DOF test
stand were conducted to evaluate effectiveness of pitch control and thrust vectoring
control for a gust along the longitudinal axis. Both position feedback and flow
feedback control schemes were tested for step gust profiles with magnitudes up to 4
m/s and duration of 1, 3, and 6 seconds. The following sections provide information
on the 6-DOF test stand, custom gust generation device, and the gust rejection
controller.
6.5.1 Experimental Stand
In order to emulate free-flight conditions, the cyclocopter was mounted on a 6-DOF
stand (Fig. 6.16) assembled using three linear ball-bearing sliders and a gimbal
bearing. The gimbal mount enables up to ±16◦ in pitch, ±20◦ in roll, and ±16◦
in yaw while the linear sliders permit translation of 0.53 m across x-axis, 0.27 m
across y-axis, and 0.15 m across z-axis. The vehicle’s main mount is attached to a
183
Figure 6.16: 6-DOF experimental stand.
pulley system and a counterweight of 300 grams to offset weight of the mount. As
a result, the cyclocopter is able to operate at 1400 rpm through these experiments.
The coefficient of friction for the ball-bearing sliders is 0.001. However, the setup
does have a considerable amount of inertia due to weight of the aluminum carriages.
Therefore, experiments conducted using 6-DOF setup is to evaluate effectiveness of
different gust rejection strategies.
6.5.2 Synthetic Gust Generation Device
To produce repeatable step gust perturbations, a custom gust generation setup was
developed in this study (Fig. 6.17). These bladeless fans output a stream of air from
small ducts outlining its boundaries. A set of three fans were grouped together and
184
Figure 6.17: Synthetic gust generation device with shutter system.
Figure 6.18: Shutter system mechanism provides step gust inputs.
185
Figure 6.19: Custom mechanism to survey flow field at 66 discrete points
in test cross-section.
placed 0.089 meters apart. Another set of fans were stacked directly behind the first
row but with a 0.10 meters horizontal offset in order to augment the airflow. The
setup was placed inside the testing volume (3.0 m x 2.4 m x 1.8 m) enclosed by the
VICON Motion Capture System.
The fans are arranged behind a shutter system that is actuated using two
Hitec HS-1000SGT rotary servos. The shutter mechanism is controlled through the
ground control station and is automated to open and close for the desired dura-
tion (Fig. 6.18). With this setup, the Dyson fans are always operating in order to
achieve a step gust profile instead of the ramp profile produced in the wind tunnel.
Variable parameters of the gust setup include fan power, distance between adjacent
fans, distance between front and back rows of fans, location of shutter system, and
distance to the test area. These parameters were adjusted in order to achieve steady
uniform flow throughout the test region.
186
Figure 6.20: Magnitude of step gust input for one vs. two rows of Dyson
fans.
Figure 6.21: Variations in magnitude and duration of gust profiles using
the shutter mechanism.
187
An important characteristic of the gust generation setup is that the flow field
within the test region must have a steady unidirectional profile so that the vehicle
is not subjected to additional aerodynamic effects from turbulence. A systematic
flow field survey of the test region, 1.32 m from the fans, was conducted using a dif-
ferential pressure sensor mounted on an automated carriage mechanism (Fig. 6.19).
The survey measured the average freestream velocity at 11 points evenly distributed
along the cross-stream location (0-80cm) and at 6 different flow probe heights (0-
16.8cm). A sweep of the flow field shows that the mean freestream velocity is uniform
across the 80 cm by 16.8 cm test region. Two rows of fans generated a maximum
freestream velocity of 4 m/s with variance of 0.15 (Fig. 6.20).
√√√√∑n Vi − V̄Vrms = (6.5)
n− 1
i=1
Vrms
T = × 100 (6.6)
V̄
Using equations 6.5 and 6.6, the turbulence at the maximum generated freestream
velocity is approximately 11.4%. Wind tunnels typically have a small turbulence in-
tensity of less than 0.7% [87]. Overall, the synthetic gust generation device is capable
of producing step gust inputs of 2.8 m/s and 4 m/s at various step durations. The
setup can also generate sinusoidal gust profiles by actuating the shutters. Through-
out the experiments, the duration of gust input was varied at 1, 3, or 6 seconds
(Fig. 6.21).
188
Figure 6.22: Closed-loop feedback system with gust rejection controller.
Figure 6.23: Onboard flow sensing mounted on the cyclocopter.
189
6.6 Gust Rejection Controller
The closed-loop feedback system was modified with a PID controller for autonomous
position control. Position data from the VICON motion capture system were trans-
mitted to the onboard autopilot via a LabVIEW interface on the ground control
system. The position trim setpoints (xdes, ydes, zdes) were initialized at the start of
each experiment. The position controller outputs desired attitude angles (φ, θ, ψ)
and are the setpoints for the attitude stabilization inner-loop controller. The inner-
loop uses a PD control scheme and modulates the rotational speeds (throttle, pitch,
and roll) and differential thrust vector (yaw) inputs to the vehicle.
A gust controller with a flow feedback scheme was added to the inner-loop
and only provided a collective thrust vector input to the vehicle. Unlike differen-
tial thrust vectoring, collective thrust vectoring results in forward and backward
translation of the cyclocopter instead of a change in orientation. Inputs to the gust
controller includes X-Y velocity components (u, v) from a miniature dual-probe flow
sensor (Fig. 6.23). The appropriate phase angle control input for a given velocity
was determined a priori in the wind tunnel and from forward flight experiments
discussed in Chapter 4:
Φ = KΦV + Φ0 (6.7)
The nominal phase angle command (Φ) is equivalent to measured velocity
(V ) multiplied by a gain (KΦ) that was fine-tuned during these experiments. Since
190
the experiments were conducted in hover, the initial phase angle is zero (Φ0 = 0).
However, the controller can be adapted to forward flight if the cyclocopter’s flight
speed is subtracted from measured velocity.
6.7 Results
While the 1-DOF wind tunnel studies explored both longitudinal and lateral gust
response, 6-DOF experiments focused on longitudinal gust response in order to
evaluate effectiveness of thrust vectoring control. Initial experiments assessed pitch
control and thrust vectoring control against a step gust perturbation. These experi-
ments used only position feedback as means to detect disturbances. At the beginning
of each experiment, the vehicle maintained a steady hover at approximately 0.1 m
longitudinal offset from the slider bounds while centered laterally. Afterwards, a
step gust input was provided with magnitudes of 2.8 m/s and 4 m/s and durations
of 1, 3, and 6 seconds. The selected performance metric was displacement from
the desired hover position. After selecting the control input, additional experiments
were conducted to specifically evaluate performance of position feedback and flow
feedback controllers.
6.7.1 Pitch Control vs. Thrust Vectoring
Effectiveness of position feedback using pitch control was investigated because pre-
liminary wind tunnel results showed that the cyclocopter is able to maintain pitch
control authority in a 10 m/s gust perturbation. However, the 1-DOF experiment
191
Figure 6.24: Displacement and pitch control input for 2.8 m/s step gust.
only assessed the vehicle’s ability to recover pitch attitude and not the hover trim.
For this experiment, the attitude stabilization for pitch used a PD controller and po-
sition feedback used a PID controller. Therefore, the cyclocopter will not maintain
a level attitude (θ = 0◦) during a gust perturbation but instead use the non-zero
pitch attitude to recover the trim position on the horizontal slider (thrust = drag).
This control strategy is similar to how typical rotary MAVs such as quadcopters
recover from gust perturbation.
Figure 6.24 shows the 6-DOF response to a 2.8 m/s gust perturbation using
pitch control to recover trim. The plots display displacement along both lateral
and longitudinal axis, step gust input, and the controller’s response in terms of
resulting pitch angle. The dashed line indicates maximum achievable pitch angle
due to constraints on the gimbal stand. As evident from the figure, a step gust
input causes the cyclocopter to pitch-up and subsequently translate backwards by
0.34 meters. In addition, gust along the longitudinal axis does not have a significant
impact on lateral translation. The feedback controller is ultimately able to recover
192
Figure 6.25: Displacement and thrust vector input for 2.8 m/s step gust.
desired position on the slider with a 16◦ pitch down attitude when operating at 1400
rpm.
Because of its thrust vectoring capability, the cyclocopter is able to provide
both lift and propulsive thrust by phasing the blade pitch cyclic. Position feedback
using thrust vectoring resembles the forward flight control strategy previously im-
plemented on the vehicle. Therefore, it is entirely possible for the cyclocopter to
maintain both a level attitude (θ = 0◦) and hover at the setpoint during a gust
perturbation.
Figure 6.25 shows a comparison of the resulting displacement between pitch
control and thrust vectoring. A phasing input of 26◦ counteracts the drag force from
the 2.8 m/s gust perturbation and follows a similar step profile. Even with a phasing
input, the cyclocopter still experiences a pitch-up moment, but it is stabilized by the
inner-loop PD controller. The cyclocopter translates only 0.1 m along the longitu-
dinal axis since the propulsive thrust was increased by instantaneously rotating the
193
Figure 6.26: Displacement vs. duration of gust for pitch control and thrust
vectoring inputs.
thrust vector instead of incrementally modulating the nose rotor’s rotational speed.
Figure 6.26 summarizes the displacement results for both pitch control and
thrust vectoring for 2.8 m/s at 1, 3, and 6 seconds durations. During the exper-
iments, the linear slider enabled 0.53 m of translation along the perturbed axis.
Accounting for the vehicles initial position, displacements of 0.4 m or above were
categorized as a failed controller (i.e vehicle was not able to tolerate gust). Each
data point on the graph was an averaged value of multiple trials after selecting the
final gains of the position feedback controller.
Pitching forward to counteract a longitudinal gust perturbation resulted in a
maximum gust tolerance of 2.8 m/s with a duration of 3 seconds. The maximum
pitch attitude (θ = 16◦) at 1400 rpm on the 6-DOF stand did not produce adequate
propulsive thrust to counteract the drag on the vehicle. While the vehicle was able to
194
Figure 6.27: Displacement vs. gust durations for 2.8 m/s and 4 m/s.
output a pitching moment to overcome 10 m/s gust perturbation in pitch attitude,
it was not able to recover the longitudinal translation trim using the same control
strategy. Thrust vectoring resulted in reduced displacement for each gust profile
and was therefore selected as the control input for the gust controller.
6.7.2 Position Feedback vs. Flow Feedback
Figure 6.27 shows resulting displacement of a gust controller using thrust vectoring
control coupled with (1) position feedback, (2) flow feedback, and (3) combined
position and flow feedback at both 2.8 m/s and 4 m/s step gust inputs. The results
show that flow feedback is more effective than position feedback in minimizing the
vehicle’s response to gust. In flow feedback, the controller provides an input to
negate predicted response of vehicle based on measured velocity instead of correcting
195
Figure 6.28: Definition of settling time.
Figure 6.29: Settling time vs. gust durations for 2.8 m/s and 4 m/s.
196
after the vehicle responds to the disturbance. The combined flow feedback and
position feedback controller was able to withstand 4 m/s step gust input with one
second duration with only 0.01 m displacement. For 4 m/s magnitude and durations
of 3 s and higher, the combined controller was unable to counteract the pitching
moment due to gust and failed to recover position.
The significance of position feedback becomes evident when considering set-
tling time. Settling time is defined as the time to recover desired position. Since the
position feedback controller has a 3 cm buffer from the setpoint, settling time is only
measured after the vehicle passes this point (Fig. 6.28). The buffer or dead zone
prevents the gust controller from oscillating about the setpoint since the vehicle is
resting on a ball bearing carriage. In addition, settling time of 10 seconds or greater
was set as the time limit for the controller to recover trim. It should be noted that
settling time is dependent on control system gains and is being evaluated in these
experiments to highlight significance of position feedback.
Figure 6.29 shows that the combination of flow feedback and position feedback
results in a lower settling time. Intuitively, the settling time result is sensible because
only the position feedback controller uses displacement error in its control scheme.
The intention of the flow feedback controller is to prevent any change in the vehicle’s
inertial state since it relies on flow measurements. If a disturbance were to effect
a large displacement, the controller would not be able to recover position since the
only available sensor measurement is of the flow velocity.
197
Figure 6.30: Setup for crosswind gust scenarios (β > 0).
Figure 6.31: Attitude response for step gust input at 30 degrees azimuth
from longitudinal axis.
198
Figure 6.32: Feedforward roll input to counteract gust effects.
6.7.3 Crosswind Scenario
In a realistic scenario, MAVs can encounter gust perturbation from any direction in
flight. Therefore, additional experiments were conducted with a gust perturbation
experienced at 10, 20, and 30 degrees azimuthal variation from cyclocopters longitu-
dinal axis (Fig. 6.30). The gust controller relied on thrust vectoring and combined
flow feedback and position feedback controller in order to reduce settling time. The
acceptable maximum displacement was lowered to 0.3 m to keep the vehicle within
the Dyson fan test cross-section and the maximum settling time was increased to
15 seconds.
Unlike the longitudinal gust experiments where pitch response was isolated,
crosswind tests affect both pitch and roll angles (Fig. 6.31). The cyclocopter re-
sponds in a roll right and pitch up reaction to a step gust perturbation. In addition
199
Figure 6.33: Displacement vs. gust duration for crosswinds using combined
controller.
to the inner-loop attitude stabilization, an additional feedforward input was supple-
mented to the gust controller. The roll input depends on the y-component flow mea-
surement (vy) and uses a similar controller input calculation as discussed previously
for thrust vectoring. Because the magnitude of the y-component flow measurement
(vy) is lower than the resultant velocity (V ), the measurements are sensitive to noise
from vibrations. Therefore the feedforward roll inputs were non-zero above a set
saturation point, paralleling the step gust profiles (Fig. 6.32).
Figures 6.33 and 6.34 show that the cyclocopter was able to tolerate step gusts
of 2.8 m/s in magnitude and 3 seconds duration with crosswind components at 30
degrees from the longitudinal axis. The decline in performance at higher azimuth
angle can be attributed to the cyclocopter’s lower control authority in roll. Similar
results were observed in the 1-DOF wind tunnel experiments.
200
Figure 6.34: Settling time vs. gust duration for crosswinds using combined
controller.
6.8 Summary
The gust tolerance capability of the cyclocopter was investigated through series of
experiments in the wind tunnel and using a synthetic gust generation device. Both
pitch control by actuating nose rotor rotational speed and thrust vectoring control
were investigated. The cyclocopter was first mounted on 1-DOF experimental stand
in front of an open-jet wind tunnel to evaluate response to gust perturbation along
the longitudinal and lateral axis. A gust perturbation along the longitudinal axis
resulted in a pitch-up moment on the cyclocopter. 1-DOF pitch stand results showed
that the cyclocopter was able to withstand 10 m/s gust perturbation using pitch
control. The threshold for the maximum gust perturbation was much lower along
the lateral axis. 1-DOF roll stand results showed that the cyclocopter was able to
201
withstand up to 6.5 m/s of gust when operating at 1400 rpm before experiencing
control saturation. The cyclocopter achieved roll control by differentially varying
the rotational speed of both cyclorotors, which have a lower maximum achievable
rpm compared to the nose rotor.
While the 1-DOF wind tunnel studies explored both longitudinal and lateral
gust response, 6-DOF experiments focused on longitudinal gust response in order
to evaluate effectiveness of thrust vectoring control. A synthetic gust generation
device was assembled using six Dyson fans. A thorough sweep of the flow field
showed that the mean freestream velocity was uniform across the 80 cm by 16.8
cm test region. The setup generated 2.8 m/s and 4 m/s gust with variance of 0.15
and a step gust profile using an automated shutter system. A gust controller with a
flow feedback scheme was added to the closed-loop feedback system to augment the
outer-loop PID position controller. Inputs to the gust controller included position
measurements from the VICON motion capture system and flow measurements from
a miniature dual-probe flow sensor.
For a step gust input along the longitudinal axis, thrust vectoring was more
effective than pitch control in reducing displacement from gust. Pitch control using
the position feedback controller resulted in a maximum gust tolerance of 2.8 m/s
with a duration of 3 seconds while thrust vector control using the combined flow
feedback and position feedback controller was able to withstand 4 m/s step gust
input with one second duration with only 0.01 m displacement. The results also
showed that flow feedback is more effective than position feedback in minimizing
the vehicle’s response to gust. While the settling time results measured effectiveness
202
of the gains used in the controller, the displacement results demonstrated that thrust
vectoring coupled with onboard flow sensing increased gust tolerance capability of
the cyclocopter.
203
Chapter 7: Multi-Mode Capability
7.1 Overview
The previous chapter explored potential gust tolerance benefits of the cyclocopter
configuration. Another advantage is the cyclocopter’s adaptability for multi-mode
locomotion. Multi-mode mobility is prevalent in biological systems where organisms
efficiently switch between different modes of locomotion to conserve energy, traverse
long distances, and maneuver through confined spaces. Challenges include imple-
menting unique control strategies for each mode of locomotion and overcoming the
vehicle’s high empty-weight fraction penalty from additional structural components
required to successfully enable all three modes. The chapter initially discusses de-
sign modifications to the pitching mechanism implemented on the cyclorotor, as well
as overall vehicle design specifications. Afterwards, key structural components used
to enable terrestrial and aquatic modes are discussed. Finally, the developed control
strategies and multi-mode testing results are presented. Results include demonstra-
tion of stable hover based on onboard attitude data from a custom auto-pilot, power
consumption for all modes, and performance in terrestrial mode evaluated using a
1-DOF test rig and VICON motion capture system.
204
Figure 7.1: All-terrain cyclocopter equipped with four cyclorotors.
7.2 Vehicle Integration and Design
The all-terrain cyclocopter uses four cyclorotors configured into two pairs at the
fore and aft of the vehicle (H-configuration) (Fig.7.1). Both pairs rotate in opposite
directions in order to counteract the resulting torque. The vehicle spans 0.51 meters
(20 inches) in width, 0.43 meters (17 inches) in length, and 0.30 meters (12 inches)
in height. The all-terrain cyclocopter weighs approximately 1010 grams with the
cyclorotors contributing up to 33% of the total gross take-off weight.
The cyclorotors are independently driven by a 75 W Axi out-runner motor
through a 7.5:1 single-stage gear transmission. Thrust vectoring for each rotor is
actuated using a rotary BMS-373MG servo that controls the blade pitching mecha-
nism. Overall, there are eight actuators that command lift and propulsive thrust in
aerial, terrestrial, and aquatic modes of locomotion. The motors are powered by a
3S 11.1V LiPo battery with 1300 mAh capacity while the servos are powered by a
205
Figure 7.2: Modified cyclorotor blade pitching mechanism for the all-
terrain cyclocopter.
single-cell 3.7 V LiPo battery with 500 mAh capacity. The following sections elabo-
rate on cyclorotor design, additional structural components required for terrestrial
and aquatic modes, and the control strategy implemented on the vehicle.
7.2.1 Cyclorotor Design
Because a cyclorotor rotates about its horizontal axis, it can be adapted into a wheel
for maneuvering on ground. The vehicle can be driven directly using torque from its
four motors, similar to cars with an all-wheel drive transmission. One complication
with the cyclorotor design used on the twin-cyclocopter configuration is that the
main rotor shaft is not connected to the gear transmission and instead connects to
the blade pitching mechanism (Fig. 7.2). As a result, both endplates freely rotate
and rest on ball bearings. The inner-most endplate is driven by the motor and
rotates the outer endplate because of its connection through the four blades. In
flight, the blade attachments points are strong enough to prevent any lag between
the two endplates. In terrestrial mode, additional friction results in a noticeable lag
206
Figure 7.3: Averaged thrust vs. RPM for cyclorotors.
Figure 7.4: Averaged power vs. RPM for cyclorotors.
207
Table 7.1: Key cyclorotor design parameters.
Parameter Cyclorotor
Rotor Diameter 0.152 meters (6 inches)
Blade span 0.171 meters (6.75 inches)
Blade chord 0.051 meters (2 inches)
Blade airfoil NACA 0015
Blade pitch amplitude ±45◦
Blade weight 5.8 grams
between the endplates, which can damage blades at the attachment points.
One solution is to combine the endplates and main shaft into one rotating
body and include a separate inner-shaft that interfaces with the pitching mechanism.
The all-terrain cyclocopter uses a titanium inner-shaft channeled inside the hollow
carbon fiber main shaft (Fig 7.2). The titanium shaft rests on several ball bearings
that enable it to freely rotate. One end of the titanium shaft is connected to the
eccentric piece on the blade pitching mechanism while the other end is attached to
a small spur gear. In the redesigned cyclorotor, both endplates are attached to the
main shaft using a Delrin hub and are directly driven by the transmission.
As mentioned in Chapter 2, the chord/radius of 0.67 is achieved with cyclorotor
radius of 0.076 m (3 in) and blade chord of 0.051 m (2 in). The blades span 0.171 m
(6.75 in) and use a NACA 0015 airfoil profile and a symmetric pitching amplitude of
45◦. The parameters are again summarized in Table 7.1. A linear approximation [47]
shows that the virtual chamber is approximately 7% while the virtual incidence
is 8◦. The pitching axis locations (x/c) of the blades were centered around the
chordwise center of gravity (CG) location, which resulted in x/c of 0.25. The pitching
axis aligned with the chordwise location of blade CG in order to to reduce the
208
Figure 7.5: Iterations of cyclowheel designs.
Table 7.2: Comparison of cyclowheel designs.
Iteration Advantages Disadvantages
1st Strong aluminum wheel Heavy (27 grams)
Large surface area Difficult to manufacture
2nd Simple to manufacture Heavy (22 grams)
Reduced surface area
3rd Lightweight (8 grams) Reduced surface area
Simple to manufacture Flexible
pitching moments on the blade caused by the centrifugal force. In addition, it
reduces the loads on the pitching mechanism and ultimately on the rotary servos.
Both thrust and power were systematically measured for each rotor from 800 to
1600 rpm. Figures 7.3 and 7.4 shows that each cyclorotor is able to produce 270
grams (total of 1080 grams) while drawing 65 W at the operating speed of 1600
rpm. Along with other structural components, the optimized cyclorotor weighs 83
grams and has a thrust-to-weight ratio of approximately 3.25 at 1600 rpm.
209
7.2.2 Cyclowheel Design
There were multiple iterations of cyclowheel design for the all-terrain cyclocopter
(Fig. 7.5). The overall goal was to reduce the total cyclowheel weight and simplify
the manufacturing process. Three iterations using wheel diameter of 0.23 m (9.25
in) are summarized in Table. 7.2. The initial design treated the wheel component
as an independent structure attached to the main shaft. The wheel composed of a
thin aluminum sheet attached to four carbon fiber spokes mounted on ABS plastic
connectors. The wheel added 27 grams to the total cyclorotor weight, largely due
to the 0.74 m (2.43 ft) long aluminum sheet with 0.635 cm (0.25 in) width. The
weight resulted in an additional 107 grams to the overall vehicle weight. Another
disadvantage of the design was the complex manufacturing process of bending the
aluminum sheet in order to maintain a uniform radius.
The second iteration replaced the aluminum wheel with carbon fiber, reducing
the added weight per cyclorotor to 22 grams and 88 grams to overall vehicle weight.
Similar to the the first design, the wheel is still independent of the cyclorotor and
is composed of two 0.79 mm (0.031 in) carbon fiber plates stacked together. The
plates connect with a Delrin hub attachment on the main shaft. While it simplified
the manufacturing process, the carbon fiber wheels resulted in less surface area for
rubber coating. The coating improves traction and prevents the wheels from slipping
on smooth surfaces.
In the final design, the wheels were efficiently integrated into the carbon fiber
rotor endplates (Fig. 7.5). The diameter of the endplates was extended from 0.15
210
Figure 7.6: Retractable landing gear design.
m (6 in) to 0.23 m (9.25 in). Each cyclowheel contained two 0.8 mm (0.03 in) thick
carbon fiber wheels that were rigidly attached to the main shaft. Compared to
the initial design, the wheels were flexible and had reduced surface area for rubber
coating. However, the integrated wheels were simple to manufacture and contributed
only 8 grams to the rotor and 32 grams to the total vehicle weight.
7.2.3 Retractable Landing Gear
One of the key components of the all-terrain cyclocopter is the retractable landing
gear system. Cyclorotors operate at drastically different rotational speeds for terres-
trial locomotion compared to in flight. Therefore, to vertically take-off, cyclorotors
need to first eliminate contact with the ground and then spool up to the operating
rpm. The retractable landing gear system is driven by two Firgelli PQ-12 micro lin-
ear servos (Fig. 7.6). At full extension, the landing gear system provides up to 7 cm
(2.75 in) of extension for vertical take-off and completely retracts until the wheels
contact the ground. During operation, the pilot is able to retract and deploy the
211
Table 7.3: Weight distribution of the all-terrain cyclocopter.
System Weight (g) % Total
Cyclorotors 333 33
Fuselage 210 21
Transmission 152 15
Landing Gear 130 13
Batteries 115 11
Rotary Servos 45 4
Additional sensors (power logger) 20 2
Avionics 5 <1
Total 1010 grams 100%
landing gear through a switch on the transmitter. For aerial and terrestrial modes,
the landing gear rests on lightweight carbon fiber spars that are attached using 3-D
printed ABS mounts. Including additional carbon fiber spars along the vertical axis
and carbon fiber plates used to secure the landing gear onto the vehicle’s fuselage
frame, the entire landing gear system weighs 100 grams.
Aquatic mode requires additional polystyrene foam pontoons mounted to the
landing gear system. Pontoons provide enough buoyancy to enable the 1010 grams
all-terrain cyclocopter to safely float on the surface on water. They contribute
an additional 30 grams of weight to the system (total 130 grams). In order to
safely operate near water, additional fuselage structures were added to shield del-
icate onboard electronics (avionics, battery, etc.) from exposure. Electronic speed
controllers (ESCs) and connectors were also thoroughly masked with waterproof
coatings.
Overall, structures required to permit terrestrial and aquatic locomotions in-
clude cyclowheels, retractable landing gear system, additional fuselage structure to
212
Figure 7.7: Closed-loop feedback system for the all-terrain cyclocopter.
shield electronics, and layers of waterproof coating for ESCs and connectors. These
components contribute close to 200 grams, which account for 20% of total vehi-
cle weight. Table 7.3 provides a weight breakdown of all the subsystems on the
all-terrain cyclocopter.
7.3 Closed-loop Feedback System
The closed-loop feedback system implemented on the vehicle is similar to the scheme
used by the twin-cyclocopter (Fig. 7.7). As discussed in Chapter 3, onboard attitude
stabilization is implemented using ELKA-R. The inner-loop attitude stabilization
uses a proportional-derivative (PD) controller, which was sufficiently able to stabilize
the all-terrain cyclocopter to sustain stable hover. Inputs to the feedback are the
four individual motor rotational speeds, four thrust vectoring positions of the servos,
213
Figure 7.8: Orientation of cyclorotors in aerial mode.
and two inputs for the linear servos on the landing gear. The feedback states are
aircraft pitch and roll Euler angles (φ, θ) and the attitude rates (p, q and r).
Onboard IMU data is streamed to a LabVIEW interface, which sends input
to switch between different modes, permits data logging, and transmits settings
for the closed-loop feedback system. All the data processing, control strategies
for different locomotion modes, and feedback control calculations are performed
onboard by ELKA-R. A separate connection between a commercial receiver and
transmitter ensures a constant and fail-safe connection with the pilot.
7.4 Control Strategy
Aerial, terrestrial, and aquatic modes require three different control strategies that
rely on combinations of motor torque and thrust vectoring. In aerial mode, cy-
clorotor rotational speeds and thrust vectors are individually modulated to sustain
stable hover. A similar control strategy relying on aerodynamic forces generated by
214
Figure 7.9: Control strategy for aerial mode.
cyclorotors was also implemented for aquatic locomotion while terrestrial locomo-
tion only uses motor torque. Actuator controls mixing of all operational modes is
performed onboard the flight controller while the desired mode is selected through
the ground control station.
215
7.4.1 Aerial Mode
As previously mentioned, the fore and aft pairs of cyclorotors spin in opposite di-
rections to counterbalance the resulting torque. On the current vehicle, cyclorotors
spin outwards such that the front pair rotates clockwise along the +Y-axis while the
back rotates counterclockwise (Fig. 7.8). The orientation of the rotors was selected
by considering the pitch control strategy for aerial mode and the resulting torque.
Figure 7.9 provides an overview of the control strategy used in hover. Pitch control
is provided by modulating the rpm of the fore and aft pairs of cyclorotor 7.9(a).
For a pitch down input, the rotational speed of the front rotors decreases while
the back increases. The resulting torque induces a beneficial pitch down reaction
moment. If the orientations were reversed, the reaction moment would oppose the
initial pitching moment.
Furthermore, roll control is achieved by differentially varying rotational speeds
of the side pairs 7.9(b) and yaw control is possible by differentially rotating the thrust
vectors of the side pairs 7.9(c). An advantage of the quad-cyclocopter configuration
over the twin-cyclocopter is decoupled dynamics. Since the angular momentum of
cyclorotors are balanced, the vehicle is unaffected by gyroscopic coupling. Therefore,
mixing actuator inputs between the operational modes only involves disabling the
feedback control system and reversing the front pair of cyclorotors in terrestrial
mode. Chapter 4 provides additional information about the roll-yaw cross coupling.
216
Figure 7.10: Control strategy for terrestrial mode.
7.4.2 Terrestrial Mode
Complications of switching between aerial and terrestrial modes is that the direction
of rotation of two cyclorotors must be reversed so that the rotors do not oppose the
resulting translational motion. One possible solution is to simply rely on a two-
wheel drive system such that only a single pair of cyclorotors would be powered.
This strategy would result in unnecessary power consumption since the cyclorotors
would have to operate at a higher rotational speed in order to support the other
cyclorotors.
An alternative solution is using programmable electronic speed controllers
(ESCs) that can electronically reverse the direction of the motors based on trans-
mitter input. The all-terrain cyclocopter uses four Afro 12A multi-rotor ESCs re-
programmed with a reversible motor setting. The direction of rotation of two cy-
clorotors is reversed when the throttle input is below the adjustable neutral throttle
point. Since the throttle setting differs between the pairs of cyclorotors, appropriate
adjustments to the transmitter throttle stick input and trim settings have been ap-
217
Figure 7.11: Control strategy for aquatic mode.
plied through the flight controller so that rotors initialize and operate on the same
throttle curve.
The all-terrain cyclocopter can effectively function as both two-wheel and all-
wheel drive system in terrestrial mode. To translate forward, the direction of rota-
tion of the aft pair of cyclorotors is reversed so that all cyclorotors spin clockwise
about the +Y-axis (Fig. 7.10). However, backward translation is only achieved
by powering the aft pair in its aerial mode configuration (counterclockwise rotation
about +Y-axis). Differentially varying the rotational speeds of the side pairs enables
the vehicle to turn with a small radius.
7.4.3 Aquatic Mode
In aquatic mode, the all-terrain cyclocopter relies on propulsive thrust generated by
thrust vectoring the cyclorotors. For a constant throttle input, collectively thrust
vectoring the cyclorotors either forward or backward results in translation in the
218
Figure 7.12: Demonstration of aerial, terrestrial, and aquatic modes.
Table 7.4: Performance of the all-terrain cyclocopter.
Mode Maneuver Speed (m/s) Average RPM Power Required (W)
Aerial Hover 0 1550 232
Aquatic Translation 1.3 348 19
Terrestrial Translation 2 83 28
same direction (Fig. 7.11). Translational velocity can be increased by either in-
creasing the throttle input or further rotating the thrust vector. The thrust vectors
on the current vehicle can be rotated up to ±60◦ from the vertical axis. Similar to
the terrestrial mode, the vehicle is unable to translate laterally and instead must
change its orientation. The yaw control strategy uses differential thrust vectoring
of the side pairs, which is the same strategy used in flight.
7.5 Multi-Mode Testing Results
The all-terrain cyclocopter was first systematically tested in each mode and then in
hybrid aerial and terrestrial mode where the vehicle transitioned using the landing
gear system. Figure 7.12 shows video stills of the vehicle successfully achieving all
219
three modes of locomotion. The following sections discuss performance in each mode
with accompanying sensor measurements from the onboard autopilot, commercial
power logger, and motion capture system. Performance results are summarized in
Table 7.4.
7.5.1 Aerial Mode
Prior to flight testing, initial gains and trim settings for hover were obtained through
bench-top experiments. The vehicle was mounted on a 3-DOF gimbal stand, which
allowed up to ±20◦ of rotation in pitch, roll, and yaw. Gimbal testing demonstrated
that the all-terrain cyclocopter’s dynamics, such as gyroscopic interactions, are de-
coupled unlike for its twin-cyclocopter counterpart. This is due to the inherent
symmetry of the vehicle about an axis parallel to the rotation axis of each cycloro-
tor. Therefore, no additional modifications or actuator mixing was implemented
onboard the flight controller.
Flight tests were systematically conducted by first providing a pure throttle
command and observing the vehicle response. Any forward translation was counter-
acted by either reducing the phasing of the cyclic blade pitch or varying the motor
rpm. Slight inconsistencies during rotor assembly resulted in variations of operating
rotational speed of the cyclorotors. Therefore, individual rpm settings were applied
for each cyclorotor by adjusting pitch and roll trims. Proportional and derivative
gains were fine-tuned using the Ziegler-Nichols approach. The gains that offered
acceptable stiffness and damping to reject external disturbance with minimal oscil-
220
Figure 7.13: Onboard attitude data streamed from ELKA-R during hover.
lations were selected. Gains and control inputs for trimmed flight varied slightly
from the ones obtained through gimbal testing because the gimbal setup provided
additional damping along each rotation axis.
The all-terrain cyclocopter successfully demonstrated stable hover (Fig. 7.12).
As seen from Fig. 7.13, a sustained hover was obtained with pitch and roll angles
held close to zero with minimum fluctuations from the desired yaw rate. Power
measurements recorded by a commercial Eagle Tree Systems eLogger showed that
the vehicle requires 232 W to sustain hover while operating at 1550 rpm. When
the vehicle is at idle, onboard microelectronics (servos, avionics, power logger, etc.)
draw approximately 4.3 W (Fig. 7.14). At full extension, the landing gear system
draws an additional 4 W because it supports the entire vehicle weight.
221
Figure 7.14: Power required for terrestrial and aerial modes.
Figure 7.15: Input required for forward and backward translation in ter-
restrial mode.
222
Figure 7.16: Minimum turn radius for terrestrial locomotion.
Figure 7.17: Experimental setup to evaluate maximum ground speed.
223
7.5.2 Terrestrial Mode
Switching into terrestrial or aquatic modes requires additional mode selection through
the ground control station. In addition, extraction and retraction inputs for the
landing gear are manually provided by the pilot either through the transmitter or
the ground control station. With the current setup, seamless transition between
aerial and terrestrial modes is accomplished within 5 seconds. Compared to hover,
forward translation in terrestrial mode requires 28 W, which is a 88% reduction in
power consumption (Fig. 7.14). Because terrestrial locomotion is directly controlled
using motor torque, the vehicle is able to operate at a reduced rpm. This significant
reduction in power can result in increased endurance and range.
Position and velocity measurements of terrestrial locomotion were recorded
using the VICON Motion Capture System. The setup consists of eight high speed
infrared cameras that track the retro-reflective markers placed on the vehicle. These
infrared cameras capture the translational and rotational states of the vehicle at a
frequency of 100 Hz and triangulate the location of the markers inertially within
the three-dimensional test volume. A throttle stick input of 27% results in forward
translation of 2 m/s (Fig. 7.15). In the rpm subfigure, negative throttle input
corresponds to the reversal in cyclowheel rotation. When translating backwards, the
all-terrain cyclocopter functions as a two-wheel drive system, as opposed to all-wheel
drive, and only relies on the aft pair of cyclowheels. As a result, cyclowheels must
operate at a higher rpm with a throttle stick input of 38% to translate backwards
at 1.5 m/s. VICON experiments also shows that using differential rpm control to
224
Figure 7.18: Power vs. velocity in terrestrial mode.
change direction of orientation results in a minimum turn radius of 0.2 m (Fig. 7.16).
An experimental setup using rotating pulleys was designed to evaluate the
minimum load (no wind resistance and low friction) achievable rpm or ground speed
for the all-terrain cyclocopter (Fig. 7.17). Each cyclowheel rested on four pulleys
with ball bearing mounts attached to aluminum shafts. The vehicle was suspended
in place using 3-D printed mounts connecting to the fore and aft spars of the fuselage.
The setup enabled the vehicle to freely rotate its wheels while remaining stationary.
During testing, a constant voltage of 11.1 V, equivalent to the 3S LiPo battery
onboard the vehicle, was provided using an external power supply. The power
supply also displayed the current drawn, which was used to calculate total power
consumption. Velocity was measured using an external laser tachometer. Finally,
vehicle trim was adjusted such that the cyclorotors maintained similar rotational
speeds.
225
Under minimum load conditions (no wind resistance and low friction), the all-
terrain cyclocopter demonstrated ground speed of approximately 10 m/s (22 mph)
while consuming 116 W of power (Fig. 7.18). Testing on the experimental setup
was limited because of significant vibrations causing the cyclowheels to lose contact
with the pulleys. To prevent the vehicle from taking off, phase angles were rotated
to 90 degrees so that they were parallel to the horizontal axis. At 10 m/s (22 mph),
the cyclorotors operated at approximately 1000 rpm. Compared to aerial mode, the
cyclowheels consumed more power (factor of 2.9) at 800 rpm because of additional
friction forces from contacting the pulleys.
7.5.3 Aquatic Mode
Translational velocity in aquatic mode can be varied using the throttle input or
further rotating the thrust vector to increase propulsive thrust. The vehicle was
tested on calm water to demonstrate maneuvering capability. By increasing the
rotational speed to 348 rpm, the vehicle achieved 1.3 m/s in forward translation.
The maneuver consumed 19 W, which is only 8% of the total power required to
hover.
7.6 Summary
The chapter provided an overview of the development of an all-terrain cyclocopter
capable of aerial, terrestrial, and aquatic locomotion with seamless transition be-
tween the modes. Unlike existing multi-modal quadcopter platforms, the all-terrain
226
cyclocopter solely relied on its four cyclorotors as source of propulsion, as well as
wheels. Structures required to permit terrestrial and aquatic locomotions included
wheels, retractable landing gear system, additional fuselage structure to shield elec-
tronics, and layers of waterproof coating. These components contributed close to
200 grams, which accounted for 20% of total vehicle weight.
The all-terrain cyclocopter weighing 1010 grams successfully demonstrated
sustained hover, translation and rotational maneuvers on ground, and aquatic lo-
comotion. Control strategy for each mode relied on combinations of motor torque
and thrust vectoring. Cyclorotor rotational speeds and thrust vectors were indi-
vidually modulated to sustain stable hover. A similar control strategy relying on
aerodynamic forces generated by cyclorotors was also implemented for aquatic loco-
motion. Because cyclorotors rotate about the horizontal axis, wheels were efficiently
integrated into the carbon fiber rotor endplates and terrestrial locomotion was com-
manded directly by motor torque.
In aerial mode, the cyclorotors operated at 1550 rpm and consumed 232 W to
sustain hover. Each optimized cyclorotor weighed 83 grams and contributed up to
33% of total gross take-off weight. At the operating rotational speed, each optimized
cyclorotor produced 270 grams of lift, which resulted in thrust-to-weight ratio of
approximately 3.25. Forward translation of 2 m/s in terrestrial mode required 28 W,
which was a 88% reduction in power consumption required to hover. Using a pulley-
based experimental setup, the cyclocopter demonstrated ground speed of 10 m/s (22
mph) under minimum load conditions (no wind resistance and low friction). Motion
capture measurements also showed that using differential rpm control to change
227
direction of orientation resulted in a minimum turn radius of 0.2 m. Translational
velocity in aquatic mode can be varied using the throttle input or further rotating
the thrust vector. In aquatic mode, cyclorotors operated at 348 rpm and consumed
19 W in order to translate forward at 1.3 m/s.
228
Chapter 8: Conclusion and Future Work
8.1 Overview
Growing interest in highly portable versatile flying platforms and recent advance-
ments in microelectronics led to the development of a scaled-down class of Unmanned
Aerial Vehicles (UAVs) known as Micro Air Vehicles (MAVs). Applications include
agricultural monitoring, search and rescue (disaster relief), package delivery, aerial
photography, and infrastructure inspection. For many mission scenarios, high en-
durance, maneuverability, and the ability to tolerate and overcome environmental
disturbances such as wind gusts are critical requirements for MAVs. Designing the
next generation micro air vehicle that is also aerodynamically efficient and capable
of vertical take-off and landing (VTOL), high speed forward flight, and multi-mode
locomotion involves overcoming an array of technical challenges further exacerbated
at the MAV-scale.
Compared to their full-scale counterparts, MAVs experience degraded perfor-
mance of conventional airfoils at the low Reynolds number range (10,000 – 50,000)
at which these vehicles operate. Reduced aerodynamic efficiency of blades at the
MAV-scale is caused by high viscous forces, which leads to increased boundary layer
thickness, increased profile drag, and large induced losses. Aside from aerodynamic
229
considerations, MAVs are also constrained by size, weight, and power (SWaP) re-
quirements. Increasing endurance and range capability largely depends on efficiency
of the propulsion system and the overall size and weight of the vehicle. However, se-
lection of avionics, batteries, and onboard electronics for the transmission system is
severely limited by commercial availability. While reduced size and weight of MAVs
provide operational benefits, the resulting low inertia coupled with limited aerody-
namic damping contributes to degrading flight stability and control performance.
Compared to full-scale vehicles, MAVs typically have higher thrust-to-inertia ratios
that result in increased control sensitivity. MAVs are also highly susceptible to en-
vironmental disturbances such as gusts due to their low inertia and low operational
speeds.
The current research investigated performance of the twin-cyclocopter and
quad-cyclocopter configurations. The twin-cyclocopter configuration used two cy-
clorotors and a smaller horizontal edge-wise nose rotor to counteract the torque
produced by the cyclorotors. The all-terrain cyclocopter relied on four cyclorotors
oriented in an H-configuration. Control strategies were developed for the twin-
cyclocopter in hover and forward flight. Roll-yaw cross-couplings resulting from
gyroscopic coupling and controls coupling were also discussed in details. Results
from wind tunnel studies conducted using a 5-DOF experimental stand were used
to develop a control strategy to achieve decoupled dynamics. Furthermore, effects
of free-stream velocity, rotational speed, and phasing angle on the roll-yaw cross-
couplings were closely investigated and used to determine control inputs to achieve
zero pitch attitude at various airspeed. In addition, a linear 6-DOF flight dynamics
230
model of the twin-cyclocopter in forward flight was experimentally determined us-
ing time-domain system identification technique. The existence of the longitudinal
control derivative in translation mode was used to evaluate the control authority of
thrust vectoring to command forward speed. The model was also used to identify
sources of roll-yaw cross-coupling.
Moreover, disturbance rejection capability of the cyclocopter was investigated
through series of experiments in the wind tunnel and using a synthetic gust gener-
ation device. Disturbance rejection strategies compared control authorities of pitch
control versus thrust vectoring and feedback control schemes of position feedback
versus flow feedback. Furthermore, key structural components used to enable aerial,
terrestrial, and aquatic modes of locomotion were explored for the all-terrain cyclo-
copter. Design modifications and control strategies for each mode were developed.
Results included onboard attitude data from a custom auto-pilot, power consump-
tion for all modes, and performance in terrestrial mode evaluated using a 1-DOF
test rig and VICON motion capture system.
The experimental studies in forward flight, gust tolerance, and multi-mode
capability provided a foundation to guide additional work towards improving cyclo-
copter performance across all flight modes.
8.2 Conclusions
The following are specific conclusions from this work.
231
8.2.1 Vehicle Development
1. The 550 g twin-cyclocopter relied on two cyclorotors as main source of propul-
sion and an additional horizontal anti-torque nose rotor. The vehicle had a
lateral dimension of 0.381 meters (1.25 feet), longitudinal dimension of 0.457
meters (1.5 feet), and a height of 0.305 meters (1 foot). Each cyclorotor had a
diameter of 0.152 meters (6 inches), blade span of 0.171 meters (6.75 inches),
and a chord of 0.051 meters (2 inches). In addition, the nose rotor used a
three-bladed 9050 propeller with a 0.229 m diameter (9 inches).
2. Each rotor was independently driven by a motor and two additional rotary ser-
vos were used to provide thrust vectoring capability for the cyclorotors. Thrust
vectoring capability was enabled using a pitching mechanism that cyclically
varied individual blade pitch angle along a circular trajectory. The direction
of the thrust vector is manipulated by rotating a key component of the mech-
anism, which changes the phasing of the harmonic pitching schedule.
3. Since the cyclorotors comprised a significant portion (23-34%) of the empty
weight fraction of the vehicle, its blades were designed to be lightweight with
high bending and torsional stiffness. Selection of key aerodynamic blade pa-
rameters such as blade pitching amplitude, location of pitch axis, rotor radius,
blade airfoil, chord, span, and planform were based on previous empirical stud-
ies. Each blade uses a symmetric pitching amplitude of 45◦, pitch axis 45%
from leading edge, rotor radius of 7.6 cm (3 in), chord/radius of 0.625, span of
232
17.1 cm (6.75 in), and a NACA 0015 airfoil.The direction of the thrust vector
can be manipulated by rotating the offset link, which changes the phasing of
the harmonic pitching schedule
4. The blades on the cyclorotor experience a virtual camber effect because of
the curvilinear nature of the flow, which introduces a chordwise variation in
velocity. A symmetric airfoil at 0◦ pitch behaves as a virtually cambered airfoil
at a virtual incidence in a rectilinear flow. A linear approximation shows that
the virtual camber for the selected blade parameters is 3.5% of chord and
virtual incidence is 5◦.
5. Cyclorotors using Mylar wrapped monolithic blades and multi-Component
rib-spar blades were tested. Both cyclorotors produced 270 grams of thrust
at 1600 rpm, achieving a thrust-to-weight (T/W) ratio of approximately 10.
Slight variation of power consumption was observed between the two designs
because of differences in parasitic drag due to blade construction. At the
operational rpm, the monolithic blade design consumed 71 watts of electrical
power while the rib-spar design consumed 67 watts, resulting in a 18% increase
in power loading. Overall, both blade designs were implemented on the final
cyclocopter.
8.2.2 Flight Control Strategy
1. Flight control strategy for the cyclocopter used a combination of independent
pitch phasing and rotational speed control of the two cyclorotors, along with
233
the propeller system, without relying on any additional control by means of a
traditional empennage system.
2. While the strategy facilitated power-efficient level forward flight, it was ac-
companied by a strong yaw-roll control cross-coupling. These were in addition
to the gyroscopic couplings that are experienced even in hover. Existence of
controls coupling in forward flight was attributed to increased magnitude of
propulsive forces at high phase angles. As a result, thrust vectoring had a
larger impact on roll moments and subsequently, changes in cyclorotor rota-
tional speeds had a greater influence on yaw moments.
3. Decoupling the roll and yaw degrees of freedom in forward flight involved
mixing the roll and yaw inputs of both the pilot and the feedback system.
Wind tunnel studies showed that the mixing ratio was dependent on phasing
of the blade cyclic pitch, especially at high phase angles. If a positive roll input
was accompanied by a negative yaw (counterclockwise), it would negate the
effects of the rpm on yaw and result in a pure roll output. Likewise, a positive
yaw input paired with a positive roll input resulted in a pure yaw output.
4. Both the gyroscopic and control couplings induced the same positive roll re-
sponse for a positive yaw input. However, for a positive roll input, gyroscopic
coupling induced a negative yaw (counterclockwise) response while the control
coupling led to a positive yaw (clockwise) response.
5. The capability to control forward flight speed using only thrust vectoring
234
(phasing of the cyclic pitch) was successfully demonstrated up to 5 m/s. The
rate of transition from hover to forward flight entirely depended on the magni-
tude of phasing input. Flight tests demonstrated that the cyclocopter was able
to immediately halt to hover from 5 m/s flight. In addition, the cyclocopter
achieved the desired airspeed within a second after a given phasing input.
8.2.3 System Identification in Forward Flight
1. A linear 6-DOF flight dynamics model in forward flight was developed using
system identification techniques. The flight dynamics model was extracted us-
ing the time-domain approach from SIDPAC system identification tools. The
model was based on measurements from doublet inputs across various modes.
The model structure was initially identified using stepwise regression and the
parameters were estimated using the equation-error method. All maneuver
inputs were provided after the cyclocopter achieved a steady state of 2 m/s
flight at a steady, level attitude (θ, φ, ψ=0).
2. The existence of the control derivative, Xphase showed that phasing input com-
manded forward speed. From flight tests, it was observed that increasing the
phasing with a δphase step input directly resulted in a step change in velocity.
During the acceleration phase, the initial response was large and acceleration
decreased with time until the vehicle reached steady state after two seconds.
When the phase input was removed, the flight speed returned to the original
forward velocity at the same rate.
235
3. The negative longitudinal translation derivative, Xu indicated positive aerody-
namic damping. Therefore, the vehicle was able to stabilize a velocity pertur-
bation in the longitudinal direction. However, the aerodynamic pitch damping
(Mq) was non-existent and the vehicle required feedback damping to stabilize
the pitch attitude rate. In general, the longitudinal mode was decoupled and
independent of the rest of the modes.
4. The longitudinal coupling term, Mu was stabilizing. A perturbation that
increased the longitudinal velocity would induce a nose-up pitching moment.
The positive pitching moment would further decrease the forward velocity,
stabilizing the cyclocopter.
5. The lateral translation derivative, Yv was also stabilizing. However, the lat-
eral and yaw coupling terms, Lv and Nv were unstable. In forward flight, a
velocity perturbation would induce both roll and yaw responses. Based on Yr,
a clockwise yaw resulted in a translation to the right, further destabilizing the
vehicle. In addition, both Lp and Nr were unstable.
6. The model was able to validate the existence of the inherent roll-yaw coupling
in forward flight, which was identified by contributions of Lr and Lrud to the
lateral state and Np and Nlat to the yaw state. These are the result of combined
gyroscopic coupling and controls coupling at high phase angles.
7. The gyroscopic coupling is caused by the unbalanced angular momentum from
the cyclorotors spinning in the same direction. The controls coupling is due to
236
the increase of magnitude of propulsive forces at high phase angles. As a result,
thrust vectoring has a larger impact on the roll moments and subsequently,
changes in rotor RPM have a greater influence on yaw moments.
8.2.4 Gust Rejection Study
1. The gust tolerance capability of the cyclocopter was investigated through se-
ries of experiments in the wind tunnel and using a synthetic gust generation
device. Both pitch control by actuating nose rotor rotational speed and thrust
vectoring control were investigated.
2. A gust perturbation along the longitudinal axis resulted in a pitch-up moment
on the cyclocopter. 1-DOF pitch stand results from the wind tunnel showed
that the cyclocopter was able to withstand 10 m/s gust perturbation using
pitch control. The threshold for the maximum gust perturbation was much
lower along the lateral axis. 1-DOF roll stand results showed that the cyclo-
copter was able to withstand up to 6.5 m/s of gust when operating at 1400 rpm
before experiencing control saturation. The cyclocopter achieved roll control
by differentially varying the rotational speed of both cyclorotors, which has a
lower maximum achievable rpm compared to the nose rotor.
3. 6-DOF experiments focused on longitudinal gust response in order to evaluate
effectiveness of thrust vectoring control. A synthetic gust generation device
was assembled using six Dyson fans. A thorough sweep of the flow field showed
that the mean freestream velocity was uniform across the 80 cm by 16.8 cm
237
test region. The setup generated 2.8 m/s and 4 m/s gust with variance of 0.15
and a step gust profile using an automated shutter system.
4. A gust controller with a flow feedback scheme was added to the closed-loop
feedback system to augment the outer-loop PID position controller. Inputs
to the gust controller included position measurements from the VICON mo-
tion capture system and flow measurements from a miniature dual-probe flow
sensor.
5. For a step gust input along the longitudinal axis, thrust vectoring was more
effective than pitch control in reducing displacement from gust. Pitch control
using the position feedback controller resulted in a maximum gust tolerance
of 2.8 m/s with a duration of 3 seconds while thrust vector control using the
combined flow feedback and position feedback controller was able to withstand
4 m/s step gust input with one second duration with only 0.01 m displacement.
6. The results also showed that flow feedback is more effective than position
feedback in minimizing the vehicle’s response to gust. While the settling time
results measured effectiveness of the gains used in the controller, the displace-
ment results demonstrated that thrust vectoring coupled with onboard flow
sensing increased gust tolerance capability of the cyclocopter.
8.2.5 Multi-Mode Capability
1. All-terrain cyclocopter capable of aerial, terrestrial, and aquatic locomotion
with seamless transition between the modes was developed. Unlike existing
238
multi-modal quadcopter platforms, the all-terrain cyclocopter solely relied on
its four cyclorotors as source of propulsion, as well as wheels.
2. Structures required to permit terrestrial and aquatic locomotions included
wheels, retractable landing gear system, additional fuselage structure to shield
electronics, and layers of waterproof coating. These components contributed
close to 200 grams, which accounted for 20% of total vehicle weight.
3. The all-terrain cyclocopter weighing 1010 grams successfully demonstrated
sustained hover, translation and rotational maneuvers on ground, and aquatic
locomotion. Control strategy for each mode relied on combinations of motor
torque and thrust vectoring. Cyclorotor rotational speeds and thrust vectors
were individually modulated to sustain stable hover. A similar control strategy
relying on aerodynamic forces generated by cyclorotors was also implemented
for aquatic locomotion. Because cyclorotors rotate about the horizontal axis,
wheels were efficiently integrated into the carbon fiber rotor endplates and
terrestrial locomotion was commanded directly by motor torque.
4. In aerial mode, the cyclorotors operated at 1550 rpm and consumed 232 W to
sustain hover. Each optimized cyclorotor weighed 83 grams and contributed
up to 33% of total gross take-off weight. At the operating rotational speed,
each optimized cyclorotor produced 270 grams of lift, which resulted in thrust-
to-weight ratio of approximately 3.25. Forward translation of 2 m/s in terres-
trial mode required 28 W, which was a 88% reduction in power consumption
required to hover. In aquatic mode, cyclorotors operated at 348 rpm and
239
consumed 19 W in order to translate forward at 1.3 m/s.
5. Using a pulley-based experimental setup, the cyclocopter demonstrated ground
speed of 10 m/s (22 mph) under minimum load conditions (no wind resistance
and low friction). Motion capture measurements also showed that using dif-
ferential rpm control to change direction of orientation resulted in a minimum
turn radius of 0.2 m.
8.3 Recommendations for Future Work
The feasibility and performance of the cyclocopter configuration was explored in
this research. However, these experimental studies only provide a foundation to
guide additional work towards improving vehicle performance across all flight modes.
Suggestions for future work focused on the vehicle instead of an isolated cyclorotor
include:
1. Performance Studies: Both the twin-cyclocopter and quad-cyclocopter con-
figurations were investigated in this work. Particularly for the quad-cyclocopter
configuration, additional studies could explore changing key geometric param-
eters (e.g offset between the fore and aft pairs of cyclorotors) to improve per-
formance in hover and forward flight. The study entails understanding aerody-
namic performance of the cyclorotors, specifically characterizing its downwash.
A possible CFD/CSD coupled study combined with PIV on cyclorotor-to-
cyclorotor interaction could provide insights into the optimal geometric con-
figuration for forward flight. Another important study is evaluating the cyclo-
240
copter’s aeroacoustics. Previous chapters mentioned that the cyclorotor has
reduced acoustic signature compared to a geometrically similar conventional
rotor because of its lower operating tip speeds. However, a proper experi-
mental validation would quantify the vehicle’s acoustic signature and allow
comparisons with other rotary-wing configurations.
2. Autonomous Flight Capability: The current work mostly focused on pi-
loted flights and closed-loop position controller using an external motion cap-
ture system. The SISO PID controller allowed the feedback control system to
be easily modified for each of the experimental studies. However, future work
can focus on achieving autonomous control using a multivariable controller de-
sign combined with guidance and navigation sensors discussed in Chapter 3.
Multivariable controller could provide improved performance for disturbance
rejection. Instead of external position sensors, onboard sensors would allow
field testing of the gust rejection controller.
3. Non-Linear 6-DOF Flight Dynamics Model: A relatively simple linear
6-DOF flight dynamics model in forward flight was derived in this work. An
important aspect of future work should be to improve the model fidelity by
including higher order effects across all flight modes. Understanding higher
order effects could potentially improve the gust rejection controller, especially
if these effects are included in the closed-loop feedback system.
4. Continuation of Gust Rejection Study: Free-flight experiments in front
of the synthetic gust generation device and ultimately outdoors in realistic
241
conditions should be considered for future work. In conjunction with the
previous suggestion, the effects of gust loads can be incorporated into the flight
dynamics model and would require an aerodynamic model of the cyclorotor.
Moreover, systematic gust trade studies can be conducted with varying gust
profiles (magnitude, duration), turbulence levels, shape (e.g sinusoidal), and
frequencies. Performance could also be compared against other rotary-wing
configurations.
242
Bibliography
[1] McMichael, J.M, and Francis, USAF (Ret.), C. M. S., “Micro Air Vehicles:
Toward a New Dimension in Flight,” U.S Department of Defense Weapon
Systems Technology Information Analysis Center (WSTIAC) Newsletter, Vol.
1, No. 1-3, August 7, 1997.
[2] McMasters, J. H., Henderson, M. L., “Low Speed Single-Element Airfoil Syn-
thesis,” NASA Langley Research Center, The Science and Technology of Low
Speed and Motorless Flight, Hampton, VA, March 29-30, 1979.
[3] Mueller, T., J., “Aerodynamic Measurements at Low Reynolds Numbers for
Fixed Wing Micro-Air Vehicles,” RTO AVT/VKI Special Course on Devel-
opment and Operation of UAVs for Military and Civil Applications, VKI,
Belgium, September 13-17, 1999.
[4] Winslow, J., “Understanding of Low Reynolds Number Aerodynamics and
Design of Micro Rotary-wing Air Vehicles,” M.Sc Thesis, Department of
Aerospace Engineering, University of Maryland, College Park, 2016.
[5] Schmitz, F. W., “Aerodynamik des Flugmodell,” C. J. E. Volckmann Nachf.
E. Wette, Berlin-Charlottenburg, 1942.
[6] Laitone, E.V., “Wind Tunnel Tests of Wings at Reynolds Numbers Below
70000,” Experiments in Fluids Vol. 23, 1997, pp 405-409.
[7] Ramasamy, M., Johnson, B., and Leishman, J. G., “Understanding the Aero-
dynamic Efficiency of a Hovering Micro-Rotor,” Journal of American Heli-
copter Society, Vol. 53, No. 4, October 2008, pp. 412428.
[8] Burk, S. M., Wilson, C. F, Jr., “Radio-Controlled Model Design and Test-
ing Techniques for Stall/Spin Evaluation of General-Aviation Aircraft.” SAE
National Business Aircraft Meeting, 1975.
243
[9] Mettler, B., Identication Modeling and Characteristics of Small Scale Rotor-
craft, Kluwer Academic Publishers, 2003.
[10] Mettler, B., Tischler, M., and Kanade, T., “System Identification of Small-Size
Unmanned Helicopter Dynamics,” Proceedings of the American Helicopter
Society 55th Forum, Montreal, Quebec, Canada, May 25–27, 1999.
[11] Fadl, M., and Karadelis, J., “CFD Simulation for Wind Comfort and Safety in
Urban Area: A Case Study of Coventry University Central Campus,” Inter-
national Journal of Architecture, Engineering and Construction, Vol. 2, (2),
June 2006, pp. 131–143.
[12] Hassanalian, M., and Abdelkefi, A., ”Classifications, Applications, and Design
Challenges of Drones: A Review,” Progress in Aerospace Sciences, May 2017.
[13] Pines, D., and Bohorquez, F., “Challenges Facing Future Micro-Air-Vehicle
Development,” Journal of Aircraft, Vol. 43, (2), March/April 2006, pp. 290–
305.
[14] Leishman, J.G., Principles of Helicopter Aerodynamics (Cambridge Univ.
Press, New York, 2000).
[15] Shyy, W., Lian, Y., Tang, J., Viieru, D., and Liu, H., Aerodynamics of Low
Reynolds Number Flyers, Cambridge University Press, New York, NY, 2008.
[16] Keennon, M., Klingebiel, K., Won, H., and Andriukov, A., “Development of
the Nano Hummingbird: A Tailless Flapping Wing Micro Air Vehicle,” 50th
AIAA Aerospace Sciences Meeting, , Nashville, TN, January 912, 2012.
[17] Shavrov, V., History of Aircraft Construction in the USSR to 1938, Mashinos-
troenie (Moscow), Vol. 1, pp. 3940, 1985.
[18] Sachse, H., “Kirsten-Boeing Propeller,” Technical Memorandum 351, National
Advisory Committee for Aeronautics, January 1926.
[19] Kirsten, F. K., “Cycloidal Propulsion Applied to Aircraft,” Transactions of
the American Society of Mechanical Engineers, Vol. 50, (12), 1928, pp 2547.
[20] Schneider, E., “Blade Wheel Propeller Particularly for Watercraft,” Patent
2,291,062, 28 July 1942.
244
[21] Bartels, Jens-Erk, and Dirk Jrgens, The Voith Schneider Propeller: Current
applications and new developments, Technical report, Voith Turbo Marine
GmbH, 2006.
[22] Strandgren, C. B., “The Theory of the Strandgren Cyclogiro,” NACA Tech-
nical Report Vol. 41, No. 7, Washington D.C., July 1933.
[23] Wheatley, J., “Simplied Aerodynamic Analysis of the Cyclogiro Rotating
Wing System,” Technical Report 467, National Advisory Committee for Aero-
nautics, August 1933.
[24] Kim, S., Yun, C., Kim, D., Yoon, Y., and Park, I., “Design and Perfor-
mance Tests of Cycloidal Propulsion Systems,” Paper AIAA-2003-1786, 44th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Mate-
rials Conference, Norfolk, Virginia, April 710, 2003.
[25] Yun, C. Y., Park, I. K., Hwang, I. S., and Kim, S. J., “Thrust Control Mech-
anism of VTOL UAV Cyclocopter with Cycloidal Blades System,” Journal
of Intelligent Material Systems and Structures, Vol. 16, No. 1112, December
2005, pp. 937943.
[26] Yun, C. Y., Park, I. K., Lee, H. Y., Jung, J. S., Hwang, I. S., and Kim,
S. J., “Design of a New Unmanned Aerial Vehicle Cyclocopter,” Journal of
American Helicopter Society, Vol. 52, (1), January 2007, pp. 2435.
[27] Hwang, I. S., Min, S. Y., Lee, C. H., and Kim, S. J., “Development of a
Four-Rotor Cyclocopter,” Journal of Aircraft, Vol. 45, (6), December 2008,
pp.2151–2157.
[28] Min., S. Y., Lee. C. H., Sung, M. H., Kim, Y. S., Hur, C. M., and Kim, S. J.,
“Experimental Study of a Quadrotor Cyclocopter,” Proceedings of the 69th
American Helicopter Society Annual Forum, Phoenix, AZ, May 21–23, 2013.
[29] Sirohi, J., Parsons, E., and Chopra, I., “Hover Performance of a Cycloidal
Rotor for a Micro Air Vehicle,” Journal of American Helicopter Society, Vol.
52, (3), July 2007, pp. 263-279.
[30] Parsons, E., “Investigation and Characterization of a Cycloidal Rotor for Ap-
plication to a Micro-Air Vehicle,” M.S. Thesis, University of Maryland, College
Park, MD, 2005.
[31] Benedict, M., “Fundamental Understanding of the Cycloidal-Rotor Concept
for Micro Air Vehicle Applications,” Ph.D. Thesis, University of Maryland,
College Park, MD December 2010.
245
[32] Benedict M., Ramasamy, M., and Chopra, I., “Improving the Aerodynamic
Performance of Micro-Air-Vehicle-Scale Cycloidal Rotor: An Experimental
Approach,” Journal of Aircraft, Vol. 47, No. 4, July–August 2010, pp. 1117–
1125.
[33] Benedict, M., Jarugumilli, T., and Chopra, I., “Experimental Optimization of
MAV-Scale Cycloidal Rotor Performance,” Journal of the American Helicopter
Society, Vol. 56, No. 2, April 2011, pp. 1–11.
[34] Benedict, M., Gupta, R., and Chopra, I., “Design, Development, and Open-
Loop Flight Testing of a Twin-Rotor Cyclocopter Micro Air Vehicle,” Journal
of the American Helicopter Society, Vol. 58, No. 4, October 2013, pp. 1–10.
[35] Benedict, M., Shrestha, E., Hrishikeshavan, V., and Chopra, I., “Development
of Micro Twin-Rotor Cyclocopters Capable of Autonomous Hover,” Journal
of Aircraft, Vol. 51, (2), March–April 2014, pp.672-676.
[36] Benedict, M., Jarugumilli, T., and Chopra, I., “Effect of Rotor Geometry
and Blade Kinematics on Cycloidal Rotor Hover Performance,” Journal of
Aircraft, Vol. 50, No. 5, 2013, pp. 1340–1352.
[37] Jarugumilli, T., Benedict, M. and Chopra, I., “Wind Tunnel Studies on a
Micro Air Vehicle-Scale Cycloidal Rotor,” Journal of the American Helicopter
Society, Vol. 59, No. 2, April 2014, pp. 02208 (1–10).
[38] Runco, C., Benedict, M., and Coleman, D., “Design and Development of a
Meso-Scale Cyclocopter,” AIAA SciTech Forum, (AIAA 2016-1255).
[39] Hrishikeshavan, V., Benedict, M., and Chopra, I., “Identification of Flight
Dynamics of a Cylcocopter Micro Air Vehicle in Hover,” Journal of Aircraft,
Vol. 52, No. 1 (2015), pp. 116-129.
[40] Galway, D., Etele, J., and Fusina, G., “Modeling of Urban Wind Field Effects
on Unmanned Rotorcraft Flight,” Journal of Aircraft, Vol. 48, No. 5, 2011.
[41] Johnson, E. and Jacob, J., “Development and Testing of a Gust and Shear
Tunnel for NAVs and MAVs,” 47th AIAA Aerospace Sciences Meeting Includ-
ing The New Horizons Forum and Aerospace Exposition, January 2009.
[42] Zarovy, S., Costello, M., and Mehta, A., “Experimental method for studying
gust effects on micro rotorcraft,” Journal of Aerospace Engineering, Vol. 227,
No. 4, April 2013, pp. 703-703.
246
[43] Hrishikeshavan, V. and Chopra, I., “Performance, Flight Testing of Shrouded
Rotor Micro Air Vehicle in Edgewise Gusts,” Journal of Aircraft, Vol. 49, No.
1, Jan. 2012, pp. 193–205.
[44] Boria F., Bachmann J., Ifju P., Quinn R., Vaidyanathan R., Perry C., and
Wagener J., “A sensor platform capable of aerial and terrestrial locomotion,”
IEEE/RSJ International Conference on Intelligent Robots and Systems, 2005,
Edmonton, Canada.
[45] Kalantari A. and Spenko M., “Design and Experimental Validation of Hy-
TAQ, a Hybrid Terrestrial and Aerial Quadrotor,” Proceedings of the IEEE
International Conference on Robotics and Automation, 2013, Karlsruhe, Ger-
many.
[46] Kawasaki K., Zhao M., Okada K., Inaba M., “MUWA: Multi-field universal
wheel for air-land vehicle with quad variable-pitch propellers,” Proceedings
of the 26th IEEE/RSJ International Conference on Intelligent Robots and
Systems: New Horizon, 2013, Tokyo, Japan.
[47] Migliore, P.G., Wolfe, W.P., and Fanuccif, J.B., “Flow Curvature Effects on
Darrieus Turbine Blade Aerodynamics,” Journal of Energy, Vol. 4, (2), 1980,
pp. 49–55.
[48] Mehta, A., and Pister, K. “WARPWING: a complete open source control
platform for miniature robots,” Proceedings of IEEE/ RSJ International Con-
ference on Intelligent Robots and Systems (IROS), Taipei, Taiwan, October
2010.
[49] Hrishikeshavan V., and Chopra I., “Refined lightweight inertial navigation
system for micro air vehicle applications,” American Helicopter Society In-
ternational 6th International Specialists’ Meeting On Unmanned Rotorcraft
System, 2015, pp. 200–210.
[50] Hrishikeshavan, V. “Experimental Investigation of a Shrouded Rotor Micro
Air Vehicle in Hover and in Edgewise Gusts,” Ph.D Thesis, Department of
Aerospace Engineering, University of Maryland, College Park, 2011.
[51] Diao, Z., Quan, H., Lan, L., and Han, Y., “Analysis and Compensation of
MEMS Gyroscope Drift,” Proceedings of the 7th International Conference on
Sensing Technology, 2013.
[52] Kasdin, J., and Paley, D. Engineering Dynamics: A Comprehensive Introduc-
tion. Princeton University Press, 2011.
247
[53] Kuipers, J. B. Quaternions and Rotation Sequences: A Primer with Appli-
cations to Orbits, Aerospace and Virtual Reality. Princeton University Press,
1999.
[54] Madgwick, SO. An efficient orientation filter for inertial and inertial-magnetic
sensor arrays. UK: Internal report and University of Bristol, 2010.
[55] Thong, Y., Woolfson, M., Crowe, J., Hayes-Gill, B., and Challis, R., “Depen-
dence of inertial measurements of distance on accelerometer noise,” Measure-
ment Science Technology, Vol. 13, pp. 1163–1172, 2002.
[56] Yousif, K., Bab-Hadiahsar, A., and Hoseinnezhad, R., “An Overview to Visual
Odometry and Visual SLAM: Applications to Mobile Robotics,” Intelligent
Industrial Systems, Vol. 1, No. 4, 2015, pp. 289-311.
[57] Sa et. al, “Build Your Own Visual-Inertial Drone: A Cost-Effective and Open-
Source Autonomous Drone,” IEEE Robotics and Automation Magazine, Vol.
25, No. 1, March 2018, pp. 89–103.
[58] Delmerico, J., and Scaramuzza, D., “A Benchmark Comparison of Monocular
Visual-Inertial Odometry Algorithms for Flying Robots,” Proceedings of IEEE
International Conference on Robotics and Automation (ICRA), May 2018.
[59] Aqel, M., Marhaban, M., Saripan, I., and Ismail, N., “Review of visual odom-
etry: types, approaches, challenges, and applications,” SpringerPlus, Vol. 5,
No. 1, 2016, pp. 1897.
[60] Mueggler, E., Gallego, G., Rebecq, H., and Scaramuzza, D., “Continuous-
Time Visual-Inertial Odometry for Event Cameras,” Proceedings of IEEE
Transactions on Robotics and Automation, June 2018.
[61] Li, L., “Time-of-Flight Camera - An Introduction,” Sensing Solutions, May
2014.
[62] Aleny, G., Foix, S., and Torras, C., “Using ToF and RGBD Cameras for
3D Robot Perception and Manipulation in Human Environments,” Intelligent
Service Robotics, Vol. 7, 2014, pp. 211–220.
[63] Kim, H., Handa, A., Benosman, R., Ieng, S., and Davison, A., “Simultaneous
Mosaicing and Tracking with an Event Camera,” Proceedings of the British
Machine Vision Conference, BMVA Press, September 2014.
248
[64] Everding, L., and Conradt, J., “Low-Latency Line Tracking Using Event-
Based Dynamic Vision Sensors,” Frontiers in Neurobotics, Vol. 12, No. 4,
2018.
[65] Pijnaker Hordijk B.J., Scheper K.Y.W., de Croon GCHE., “Vertical Landing
for Micro Air Vehicles using Event-based Optical Flow. J Field Robotics,”
2017;00:1-22. https://doi.org/10.1002/rob.21764.
[66] Coppejans, H., and Myburgh, H., “A Primer on Autonomous Aerial Vehicle
Design,” Sensors, Vol. 15, 2018, pp. 30033–30061.
[67] Taketomi, T., Uchiyama, H., and Ikeda, S., “Visual SLAM algorithms: a
survey from 2010 to 2016,” IPSJ Transactions on Computer Vision and Ap-
plications, 2017, pp. 9–16. https://doi.org/10.1186/s41074-017-0027-2.
[68] Gaoussou, H., and Dewei, P., “Evaluation of the Visual Odometry Methods
for Semi-Dense Real-Time,” Advanced Computing: An International Journal
(ACIJ), Vol. 9, No. 2, March 2018, pp. 1–14.
[69] Solomon, E., Hrishikeshavan, V., and Chopra, I., “Autonomous Quadrotor
Control and Navigation with Snapdragon Flight,” Proceedings of the Ameri-
can Helicopter Society 74th Annual Forum, Phoenix, AZ, May 14–17, 2018.
[70] Solomon, E., Vorwald, C., Hrishikeshavan, V., and Chopra, I., “Visual Odom-
etry Onboard a Micro Air Vehicle Using Snapdragon Flight,” Proceedings
of 7th American Helicopter Society Technical Meeting on VTOL Unmanned
Aircraft Systems and Autonomy, Mesa, AZ, Jan 24–26, 2017.
[71] Loianno, G., Mulgaonkar, Y., Brunner, C., Ahuja, D., Ramanandan, A., Mu-
rali, C., Diaz, S., Kumar, V., “Smarpthones power ying robots,” IEEE/RSJ
International Conference on Intelligent Robots and Systems (IROS), Septem-
ber 2015, pp. 1256-1263.
[72] Sanket, N., Singh, C., Ganguly, K., Fermuller, C., and Aloimonos, Y.,
“GapFlyt: Active Vision Based Minimalist Structure-less Gap Detection For
Quadrotor Flight,” Published in IEEE Robotics and Autonomation Letters,
2018.
[73] Thong, Y., Woolfson, M., Crowe, J., Hayes-Gill, B., and Challis, R., “Depen-
dence of Inertial Measurements of Distance on Accelerometer Noise,” Mea-
surement Science Technology, Vol. 13, pp. 1163–1172, 2002.
[74] Astrom, K., and Murray, R., “Feedback Systems: An Introduction for Scien-
tists and Engineers,” Princeton, NJ, 2008.
249
[75] Pamadi, B. Performance, Stability, Dynamics, and Control of Airplanes.
American Institute of Aeronautics and Astronautics, 1998.
[76] Benedict, M., Jarugumilli, T., Lakashminaryayan, V. and Chopra, I., “Effect
of Flow Curvature on Forward Flight Performance of a Micro-Air-Vehicle-Scale
Cycloidal-Rotor,” AIAA Journal, Vol. 52, No. 6, June 2014, pp. 1159-1169.
[77] Zachary, H., A., Benedict, M., Hrishikeshavan, V., and Chopra, I., “Design,
Development, and Flight Test of a Small-Scale Cyclogyro UAV Utilizing a
Novel Cam-Based Passive Blade Pitching Mechanism,” International Journal
of Micro Air Vehicles, Vol. 5, No. 2, June 2013, pp. 145–162.
[78] Petricca, L., Hrishikeshavan, V., Ohlckers, P., Chopra, I., “Design, fabrication
and test of an embedded lightweight kinematic autopilot (ELKA),” Interna-
tional Journal of Intelligent Unmanned Systems, Vol. 2 (2014), pp.140–150.
[79] Yeo, D., Atkins, E., Bernal, L., and Shyy, W., “Fixed-Wing Unmanned Air-
craft In-Flight Pitch and Yaw Control Moment Sensing,” Journal of Aircraft,
Vol. 52, No. 2 (2015), pp. 403–420.
[80] Benedict, M., Jarugumilli, T., Lakashminaryayan, V. and Chopra, I., “Effect
of Flow Curvature on Forward Flight Performance of a Micro-Air-Vehicle-Scale
Cycloidal-Rotor,” AIAA Journal, Vol. 52, No. 6, June 2014, pp. 1159-1169.
[81] Klein, V., and Morelli, E., Aircraft System Identication: Theory and Practice.
AIAA Education Series, Reston, Virginia, 2006.
[82] Conroy, J., Humbert, J., and Pines, D., “System Identication of a Rotary
Wing Micro Air Vehicle,” Journal of the American Helicopter Society, Vol.
56, No. 2, April 2011, pp. 25001.
[83] Grauer J., Conroy J., Hubbard J. Jr., and Humbert J., “System Identification
of a Miniature Helicopter,” Journal of Aircraft, Vol. 46, No. 4, 2009, pp. 1260-
1269.
[84] Tischler, M., and Remple, R., Aircraft and Rotorcraft System Identification:
Engineering Methods with Flight Test Examples. AIAA Education Series, Re-
ston, Virginia, 2012.
[85] Yeo, D., Sydney, N., and Paley, D., “Onboard Flow Sensing for Rotary-Wing
UAV Pitch Control in Wind,” AIAA Guidance, Navigation, and Control Con-
ference, AIAA SciTech, Jan. 2016.
250
[86] Hoffmann, G., Huang, H., Waslander, S., and Tomlin, J., “Autonomous
quadrotor helicopter testbed design, control, and experiments,” AIAA Guid-
ance, Navigation and Control Conference, AIAA SciTech, Aug. 2007.
[87] Manshadi, D M., “The Importance of Turbulence Reduction in Assessment of
Wind Tunnel Flow Quality,” Wind Tunnels and Experimental Fluid Dynamics
Research, 2011.
251