Bioinspiration & Biomimetics

ACCEPTED MANUSCRIPT

Attitude control system for a lightweight flapping wing MAV

To cite this article before publication: Sjoerd Tijmons et al 2018 Bioinspir. Biomim. in press https://doi.org/10.1088/1748-3190/aab68c

Manuscript version: Accepted Manuscript

Accepted Manuscript is “the version of the article accepted for publication including all changes made as a result of the peer review process,
and which may also include the addition to the article by IOP Publishing of a header, an article ID, a cover sheet and/or an ‘Accepted
Manuscript’ watermark, but excluding any other editing, typesetting or other changes made by IOP Publishing and/or its licensors”

This Accepted Manuscript is © 2018 IOP Publishing Ltd.

During the embargo period (the 12 month period from the publication of the Version of Record of this article), the Accepted Manuscript is fully
protected by copyright and cannot be reused or reposted elsewhere.
As the Version of Record of this article is going to be / has been published on a subscription basis, this Accepted Manuscript is available for reuse
under a CC BY-NC-ND 3.0 licence after the 12 month embargo period.

After the embargo period, everyone is permitted to use copy and redistribute this article for non-commercial purposes only, provided that they
adhere to all the terms of the licence https://creativecommons.org/licences/by-nc-nd/3.0

Although reasonable endeavours have been taken to obtain all necessary permissions from third parties to include their copyrighted content
within this article, their full citation and copyright line may not be present in this Accepted Manuscript version. Before using any content from this
article, please refer to the Version of Record on IOPscience once published for full citation and copyright details, as permissions will likely be
required. All third party content is fully copyright protected, unless specifically stated otherwise in the figure caption in the Version of Record.

View the article online for updates and enhancements.

This content was downloaded from IP address 128.111.121.42 on 16/03/2018 at 01:55

Page 1 of 13 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1

1
2
3
4
Attitude control system for a lightweight flapping wing MAV

pt
5
6 Sjoerd Tijmons, Matěj Karásek, Guido de Croon
7
8
9

cri
Abstract— Robust attitude control is an essential aspect of
10 research on autonomous flight of flapping wing Micro Air
11 Vehicles. The mechanical solutions by which the necessary
12 control moments are realised come at the price of extra weight
13 and possible loss of aerodynamic efficiency. Stable flight of these
14 vehicles has been shown by several designs using a conventional
tail, but also by tailless designs that use active control of the
15 wings. In this study a control mechanism is proposed that
16

us
provides active control over the wings. The mechanism improves
17 vehicle stability and agility by generation of control moments
18 for roll, pitch and yaw. Its effectiveness is demonstrated by
19 static measurements around all the three axes. Flight test
results confirm that the attitude of the test vehicle, including
20 a tail, can be successfully controlled in slow forward flight
21 conditions. Furthermore, the flight envelope is extended with
22
23
24
25
26
27
28
robust hovering and the ability to reverse the flight direction
using a small turn space. This capability is very important
for autonomous flight capabilities such as obstacle avoidance.
Finally, it is demonstrated that the proposed control mechanism
allows for tailless hovering flight.

I. I NTRODUCTION
an
dM
As the application areas for Micro Air Vehicles (MAVs)
29 are expanding, more concepts for platform designs are being
30 explored. While fixed-wing concepts were the dominant
31 platform for decades, multirotor concepts have become enor- Fig. 1. DelFly II flapping wing MAV with wing vectoring control system
32 mously popular over the last few years. A major trend that as developed and analysed in this study. The image shows the vehicle pose
33 can be observed is that these concepts are combined in hybrid
that corresponds to hovering flight.
34 designs [1]–[4]. This enables MAVs to take off and land
35 vertically while also being able to perform long endurance
36 forward flight. forward flight [7]. On smaller systems, that have a weight of
37 Flapping wing MAV research is focused on enabling a few tens of grams, the main research focus is on performing
38 the same combination of vertical and horizontal flight by stabilized hovering and steering control, which is especially
pte

39 drawing inspiration from nature. While hybrid concepts use challenging for designs without a tail [8]–[13] because of
40 different types of devices to produce thrust forces (rotors), their intrinsic instability. In these studies, different methods
41 lift forces (wings) and control moments (rotors and control are proposed for using the wings as control devices. Other
42 surfaces) flapping wing MAVs can generate both forces systems use a tail [14]–[18], which provides (partial) inherent
43 and control moments by their wings only. This approach is stability. In these approaches, control surfaces are added to
44 inspired by the flight methods of various birds and insects the tail to allow active stabilization and to enable steering
45 that show a wide range of capabilities: extremely precise commands.
46 hovering, agile manoeuvring, efficient forward flying and While there is a clear stability advantage when using a tail,
ce

47 gliding, and fast transitioning between flying, hovering and the use of control surfaces has an important drawback. The
48 landing. DelFly II, shown in Fig. 1, is a lightweight flapping wing
49 Various flapping wing MAV designs have been demon- MAV that originally uses a tail with rudder and elevator [19],
50 strated with varying flight capabilities. Some studies focus and later versions also use aileron surfaces close behind the
51 only on forward flight, mainly using relatively heavy bird- wings for improved heading control [18]. Previous research
52 like systems that have a weight of a few hundred grams [5], on this platform has shown that at very low speeds and
Ac

53 [6]. In the same weight class an imitation of a dragonfly has especially during hovering, the effectiveness of the control
54 been demonstrated that can perform both hovering flight and surfaces diminishes drastically [20]. It was observed that this
55 limits the turn rate significantly and results in poor attitude
56 All authors are with the Control and Simulation Section, Faculty of control performance in case of disturbances.
Aerospace Engineering, Delft University of Technology, The Netherlands
57 (e-mail: s.tijmons@tudelft.nl, g.c.h.e.decroon@tudelft.nl) In this study a wing control mechanism is proposed that
58
59
60
 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1 Page 2 of 13

1
2 is added to the flapping mechanism of the DelFly II MAV forward and backward, which affects the angle of attack of
3 [21]. It allows for independent vectoring of the wing pairs the wing. This can be done either symmetrically (for pitch)
4 on both sides. The mechanism can also control the tension in or differentially (for yaw). Similar wing twist modulation

5 the two wing pairs. The proposed wing control mechanism
6 primarily replaces the function of the ailerons and elevator
7 to allow controlled hovering flight. The mechanism allows
8 the DelFly to fly at low velocities, which is very useful for
9
autonomous flight. For instance, situations can occur where
pt
and rotation modulation effects are obtained using a different
mechanism on the KUBeetle [12]. This tailless platform of
21 g demonstrates 14 seconds of stable hovering flight.
Control by wing kinematics modulation has been shown
in real (tethered) test flights on the extremely small 0.1 g

cri
the vehicle needs to turn around immediately to prevent a Robobee [9]. The wings are individually actuated by piezo-
10
collision. This requires the ability to hover and turn on the electric actuators. By changing the wing stroke amplitudes,
11
spot. It is shown in the current study that the proposed wing roll control is obtained. Moving the mean stroke angle of the
12
control mechanism improves the hovering performance and wings forward or backward results in pitching torques. Yaw
13
allows for fast turns in this flight condition. In addition to control is implemented by modulating the stroke velocity
14
these capabilities, this study also shows that the proposed within a wing beat. Similar control methods were tested on a
15
wing control mechanism can be used to control all three at- larger system driven by a single DC motor [10]. Using a me-
16

us
titude angles required for hovering. The mechanism provides chanical solution in the flapping mechanism this system can
17
sufficient control authority to enable tailless hovering. control the wing stroke amplitude (for roll) and mean stroke
18
offsets (for pitch). Tests were performed in a static setup,
19 II. R ELATED W ORK showing that the system provides adequate torques. Another
20
This section discusses studies in the field of flapping wing study shows real test flights with a Robotic Hummingbird
21
MAVs that relate to control of such vehicles. First, differ- platform that uses wing kinematics modulation [11]. As in
22
23
24
25
26
27
28
ent studies on attitude control mechanisms are discussed.
Second, work on autonomous attitude control on board of
an
such vehicles is discussed. Third, the few studies concerning
autonomous flight tasks using onboard systems are discussed.

A. Wing control techniques for attitude control

the previous two studies, roll control is obtained by changing
the stroke amplitude of the left and right wing. Furthermore,
a hinge tilting mechanism is used to control the angle of
the leading edge stroke planes of the left and right wing
individually, which allows for pitch and yaw control. This
design weighs 62 g, which is relatively heavy compared to
dM
A straightforward method for control of flapping wing the other systems that were discussed so far, which have a
29
MAVs is to use a tail with control surfaces [14]–[17], similar weight of 20 g or less. A different mechanism for changing
30
to fixed-wing aircraft. The wings purely generate thrust and the leading edge stroke plane is presented which makes use
31
lift in such setups. Extra aileron surfaces can be applied of rotating leading edge flapping slots [23]. This allows for
32
to obtain improved yaw torques for heading control [18]. generation of pitch and yaw commands. In combination with
33
By using a tail (or another form of damping surface), the a trailing edge shift mechanism, roll control is also obtained.
34
vehicle obtains passive stability in forward flight. However, This mechanism is tested on a force balance, showing that
35
the stabilizing effect of additional surfaces decreases signif- all three control moments can be generated.
36
icantly at low speeds. Furthermore, a downside of adding Two heavier systems, both weighing more than 100 g,
37
these surfaces is that the vehicle becomes more susceptible obtain control via two other systems. The BionicOpter
38
to wind disturbances. (175 g) uses four wings that are driven by a central motor
pte

39
Flapping wing MAV research focuses on flight control [7]. For each wing the stroke amplitude and the stroke
40
methods that are inspired by nature. Most insects mainly plane tilt angle can be controlled. This system is capable
41
rely on their wings for flight control, especially in (near) of hovering and transitioning to forward flight. The Robo
42
hover conditions. Wings can be used for stabilization and Raven is a bird-like platform of 312 g [6]. It uses two servos
43
steering by deforming their shape and by modulating their to drive its two wings and an extra servo to actuate the
44
angle of attack and their kinematics. Different combinations tail as a ruddervator. This platform performs forward and
45
of these techniques have been implemented and tested on gliding flight. Furthermore, a new design concept, the quad-
46
various flapping wing MAVs. thopter [24], has been introduced which consists of four
ce

47
It was shown experimentally that controlling the feathering independently controlled flapping mechanism. By flapping
48
angle and camber of a wing by pulling at a point on the these mechanisms at individual frequencies, attitude control
49
wing can produce a pitching moment [22]. This effect was can be performed in a similar way as for normal quadrotors.
50
only demonstrated on a static setup. Fully tailless control Flight times of 9 minutes and more have been realized with
51
was demonstrated on the 19 g Nano Hummingbird that this 33 g tailless platform.
52
uses a combination of wing twist modulation and rotation
Ac

53 modulation [8]. In this system, wing twist is controlled B. Autonomous attitude control
54 by pulling the wing root away from or towards the wing Autonomous attitude control was realized on the DelFly
55 membrane. This regulates the tension in the wing, and II in a previous study [20]. In that study, the vehicle design
56 indirectly twists its camber, allowing the vehicle to roll. includes a tail with active rudder and elevator surfaces.
57 Rotation modulation is obtained by tilting the wing root Aileron surfaces are used in this design for heading control
58
59
60
Page 3 of 13 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1

1
2 during hover. This study demonstrates the low effectiveness
3 of the control surfaces during hover.
4 On tailless platforms, several studies focus on realizing

pt
5 attitude stabilization and control. Attitude rate stabilization
6 was shown on the Nano Hummingbird [8] using gyroscopes.
7 Attitude control was still performed manually. Similarly, gy-
8 roscope feedback stabilization has been demonstrated on the
9 Robobee [25], allowing the vehicle to keep its hover attitude

cri
10 for several seconds. On both platforms attitude estimation
11 using accelerometers was not possible due to platform vibra- Original Proposed

12 tions. In another study pitch and yaw control was realized on

13 the Robobee using magnetometer feedback [26]. Full attitude Fig. 2. Comparison of wing attachment for original DelFly II design
stabilization and control was demonstrated in test flights with (left) and the design as proposed in this study (right). The left image shows
14 how the top and bottom wings for the original DelFly II are attached to
15 the 62 g Robotic Hummingbird [11]. In this study, attitude the fuselage using a hook on the trailing edge. The right image shows how
16 estimation was realized using feedback from accelerometers. the left and right wing are folded around the left and right wing roots,

us
The tailless 21 g KUBeetle [12] demonstrated autonomous respectively. In this implementation the top left and bottom left wing parts
17 are connected and form a single left wing. The same goes for the right
18 attitude control by pitch, roll and yaw rate control based on wing.
19 feedback from gyroscopes. The tailless quad-thopter platform
20 [24] is able to perform full attitude control autonomously by
21 pitch, roll and yaw angle control based on gyroscope and
accelerometer feedback.
22
23
24
25
26
27
28
C. Autonomous flight capabilities
Some of the discussed studies show autonomous control
capabilities of flapping wing MAVs. Full autonomous flight
has been shown in previous work with the Delfly Explorer
[27], which in terms of actuation is similar to the DelFly
an
dM
29 II. This is a system with a tail and aileron surfaces. It uses
30 gyroscope feedback for stability enhancement of pitch and
31 yaw and a barometer for height control. Furthermore it has
32 a 4 g stereo camera system to perform onboard obstacle
33 avoidance, enabling sustained autonomous flight in unknown
34 environments for up to 9 minutes. Target-seeking using an
35 onboard Wii-mote infrared camera has been demonstrated
36 with a tailed platform [28]. In that study, pitch and yaw angle
37 control are performed autonomously. The bird-like Robo
38 Raven platform [6] demonstrated GPS/IMU-based loitering.
pte

39 This entails heading control to steer the vehicle to a specified

40 GPS-waypoint.
41
42 III. C ONTROL M ECHANISM
43
A new mechanism for wing actuation is presented that
44
creates control moments for attitude stabilization and control.
45
This mechanism is specifically suitable for four-winged
46
flapping wing MAVs such as the DelFly. The mechanism
ce

47
is combined with the original flapping mechanism of the
48
DelFly, which has proven to be a reliable and efficient design.
49
The kinematics of the wing leading edges are therefore the
50
same as in the original DelFly II design. However, the wings
51
are attached in a different way. This is illustrated in Fig. 2.
52
In the original design, a bottom and a top wing are used
Ac

53 that are fixed to the fuselage at the centers of their trailing

Fig. 3. Visualization of the proposed control mechanism for different
control inputs. The small (black) arrows indicate the direction in which the
54 edges. In the proposed design, use is made of a left wing wing roots are deflected. The large (red) arrows indicate the control moment
55 and a right wing. These wings are not connected to a fixed generated by the control inputs around the body axes which are indicated
56 point, but are folded around two separate carbon rods that
by the dashed lines. The names of the axes (pitch, yaw and roll) correspond
to the situation in hover flight.
57 act as wing roots.
58
59
60
 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1 Page 4 of 13

1
2
3
4

pt
5
6
7
8
9

cri
10
11
12
13
14
15
16

us
17
18
19 Fig. 4. Left: Image of the control mechanism as implemented on the DelFly. Two carbon rods function as the wing roots which are placed inside the left
20 and right wings (indicated by black arrows). The rods can rotate and translate (red arrows) around hinges (dashed lines) which are close to the leading
21 edges of the wings. Two linear servos (visible at the bottom of the left image) actuate the hinges via carbon push rods. Bottom-right: Detailed image
of the hinges, showing the individual motions of the push rods (red arrows) for the rotational motions, and their connection with the wing root hinges.
22
23
24
25
26
27
28
with the wing root hinges.

Control moments are generated by actuating the two wing

roots individually. Both wing roots have two degrees of
freedom around the same axis: a rotation and a translation.
an
Top-right: Detailed image of the hinges, showing the individual motions of the push rods (red arrows) for the translational motions, and their connection

changes the thrust vector. The top image shows the situation
when the wings are almost closed. The bottom image shows
the situation when the wings are opened. This demonstrates
dM
29 By combining these degrees of freedoms all three attitude how the shape of the wing deforms during the first part of
30 control moments can be generated, which is shown in Fig. 3. the flapping cycle: the upper part of the left wing (white)
31 The implementation of the mechanism on the vehicle is becomes smaller, and the lower part of the left wing (cyan)
32 shown in Fig. 4. becomes larger. Because of this effect, the shape of the wing
33 This method of actuating the wing roots shows similarities changes, but the tension in the wing does not change during
34 with the systems of the Nano Hummingbird [8] and the the flapping cycle. In other words, the wings are not stretched
35 KUBeetle [12]. This is especially true for creating roll by the mechanism. A benefit of this property is that very
36 moments, as the same principle is used: by moving the wing large root deflections can be realized.
37 roots inward and outward the tension in the wing is regulated, As Fig. 3 clearly shows, the benefit of the proposed wing
38 which influences the thrust generated by the wings. The only control system is its simplicity. Both wing roots have two
pte

39 difference is that in our proposed mechanism the wing root degrees of freedom, rotating and sliding, using a single joint
40 is translated, instead of rotated. (see left image of Fig. 4). The joint is made of thin walled
41 A larger difference is found for the pitch and yaw deflec- steel tubes that form a clearance fit with minimal play. As
42 tions. Even though the wing roots are deflected in a similar these tubes are made from syringe needles, the tube walls are
43 way as on the Nano Hummingbird and the KUBeetle, the very smooth and can easily rotate and slide with respect to
44 effect of these deflections on the wings is different. Instead each other. An extra supporting carbon rod is used to obtain a
45 of changing the wing tension during the flap cycle (which is stiff connection between the joint and the wing root, forming
46 very effective at large flapping amplitudes), our mechanism a triangular structure. Furthermore, the joints of this structure
ce

47 vectors the airflow around the wings, and thereby the thrust are reinforced by balsa wood.
48 generated by the wings is vectored. As a result, the pitch Apart from these components, only servos and pushrods
49 and yaw moments generated by the proposed mechanism are needed for actuation. As Fig. 4 shows (bottom-right),
50 are opposite to those generated by the mechanisms of the two pushrods are connected to the hinges for the rotational
51 Nano Hummingbird and the KUBeetle. The magnitude and motions, driven by two servos. A third servo can be added for
52 direction of the pitch moment depends on the location of the translating the hinges. This is shown in the top-right image
Ac

53 center of gravity. of this figure.

54 When deflecting the wing roots for pitch and yaw inputs,
55 the tension of the wings is not affected. This is visualized
IV. F LAPPING M ECHANISM
56 by Fig. 5. The top image shows that if the left wing root The flapping mechanism as used in this study is fully
57 is deflected, the left wing is tilted as a whole. Thereby it based on the original mechanism of the DelFly II [21],
58
59
60
Page 5 of 13 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1

1
2
3
4

pt
5
6
7
8
9

cri
10
11
12
13
14 Fig. 7. Image of the flapping mechanism as used in this study, which does
15 not include a dihedral angle.
16

us
TABLE I
17 Fig. 5. Schematic illustration of the effect of wing root deflection on C OMPARISON OF FLAPPING MECHANISM PARAMETERS ( AS IN F IG . 6)
18 the wing shape. A side view of the DelFly is presented with the wings
FOR ORIGINAL AND PROPOSED DESIGN
almost closed (left) and fully opened (right). In both cases, the wing root
19 of the left wing is deflected upwards. The left image shows that the wing
20 root deflection makes the wing rotate upwards as a whole and the root is Design L1 [mm] L2 [mm] L3 [mm] D[mm] Ψmax [◦ ]
21 parallel to the center line of the wing. The right image shows that the wing original
proposed
4.55
4.55
17.22
15.09
12.07
13.08
21.49
21.49
88
83
root has shifted upwards with respect to the wing center line (dashed line).
22
23
24
25
26
27
28
the top and bottom parts of the wing change their shape continuously.

an
During flapping, the wing slides over the wing root every cycle whereby

is used, which shown in Fig. 7. Originally the wing hinges

have a dihedral angle of 12◦ , which enhances yaw stability in
slow forward flight. This dihedral angle is undesirable in the
current study since the wings are used for control purposes.
dM
The dihedral angle would result in asymmetry and coupling
29 effects of yaw, pitch and roll inputs. Furthermore, it would
30 limit the maximum deflections of the control mechanism
31 which uses angle of attack modulation.
32 The definition of the characteristic dimensions of the
33 four-bar flapping mechanism are indicated in Fig. 6. These
34 dimensions, together with the stroke amplitude, are specified
35 in Table I for both the original and current design. Because
36 L3 has been made longer to maintain smooth rotations, the
37 stroke amplitude Ψmax is smaller in the current design. This
38 influences the relation between motor speed and generated
pte

39 thrust, and the overall efficiency of the flapping system.

40 Fig. 6. Schematic drawing of the original DelFly II flapping mechanism,
The plastic components of the original flapping mecha-
41 which includes a dihedral angle of 12◦ . The dimensions of the four-bar
mechanism are defined, as well as the wing flap angle. nism are injection molded. For the current mechanism, the
42 pushrods and the wing hinges have been redesigned and
43 are 3D-printed, using a UV-curable acrylic resin (Shapeways
44 [29]. The platform has four wings that are arranged in a Frosted Ultra Detail).
45 biplane configuration. The wings are coupled crosswise, as The wings in the current system are similar to the wings in
46 illustrated in Fig. 6. As a result the top and bottom wings the original system, with a wingspan of 28 cm, mean chord
ce

47 flap in anti-phase which is beneficial for limiting platform length of 8 cm and aspect ratio of 1.75. They are made of
48 vibrations. Furthermore, the system benefits from a clap-and- Mylar foil and carbon rods. The leading edges are attached to
49 peel [30] effect which is present at the end of the down-stroke the wing hinges. An extra layer of tape is added to the center
50 and the start of the upstroke, where the two wings meet. parts of the wings, as these areas of the wings make contact
51 The flapping mechanism consists of two symmetric four-bar with the carbon wing roots. The tape acts as a protective and
52 mechanisms that are coupled. As a result, the wings flap fully smooth layer that prevents wear of the Mylar foil.
Ac

53 synchronized, which is also beneficial for limiting platform

54 vibrations. The mechanism is driven by a single DC motor. V. S TATIC PERFORMANCE TESTS
55 Its speed is reduced by a ratio of 1 : 21.33 via a two stage The forces and moments generated by the proposed flap-
56 gearbox. ping system were analysed using a static test setup. In this
57 In this study, an adapted version of the flapping mechanism setup, the whole mechanism including wings is attached
58
59
60
 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1 Page 6 of 13

1
2 4 4
3 Mx M
x
4 3 My 3

pt
My

Generated control moments [Nmm]

Generated control moments [Nmm]

5 Mz Mz
2 2
6
7 1 1
8
9 0 0

cri
10
11 -1 -1
12
-2
13 -2

14 -3 -3
15
16 -4

us
-4
-20 -10 0 10 20
17 pitch command [deg]
-20 -10 0 10 20
18 yaw command [deg]
19 Fig. 8. Generated moments around all three axis for a range of pitch inputs. Fig. 9. Generated moments around all three axis for a pure yaw input.
20 A positive pitch command means that both wings are deflected upwards. A positive yaw command means that the left wing is deflected downwards,
The flapping frequency in this test is ∼ 15 Hz. All measurements are time-
21 averaged over 29 flapping cycles. Linear fits through these measurements
and the right wing is deflected upwards. The flapping frequency in this test
is ∼ 15 Hz. All measurements are time-averaged over 29 flapping cycles.
22
23
24
25
26
27
28
are shown. The error bars show the minimum and maximum time-averaged
measurement over a single flapping cycle.

to an ATI Nano 17 (force resolution 3.125 mN, moment

resolution 15.625 mNmm) transducer that measures 6-DOF
forces and moments. A combination of a hall sensor switch
an
Linear fits through these measurements are shown. The error bars show the
minimum and maximum time-averaged measurement over a single flapping
cycle.

yaw commands on the Y-axis (pitch) is shown in Fig. 11.

This shows that the generated pitch control moment mainly
dM
29 and a magnet in the flapping mechanism was used to measure depends on the pitch command, and is significantly less
30 the flapping frequency. dependent on the yaw command, as is expected.
31 The effect of giving a pure roll input is shown in Fig. 12.
The effect of giving a pure pitch input (symmetric de-
32 The command is defined as the translational displacement
flection of wing roots, Fig. 3 second from top) is shown
33 of one of the two wing roots as explained in Fig. 3. A
in Fig. 8. The results show that the pitching moment MY
34 positive command means that the right wing root is shifted
can be approximated by a linear fit. Furthermore, the results
35 outwards. As expected, the roll input mainly has an effect
show that the effect on the other two moments is significantly
36 on roll moment MZ . In this test, the maximum variation
smaller. For the tested range of inputs (±25 deg) a maximum
37 in Mz that can be generated is about 1.4 N mm. Note in
variation in pitch moment of approximately 6 N mm is
38 Fig. 10 that a comparable variation in MZ is measured as
observed. The measured magnitude of the pitching moment
the coupling effect between pitch and yaw inputs (Fig. 10).
pte

39 depends on the location of the centre of gravity. In this test,

40 This indicates that the stability margin for roll will be small
the location of the centre of gravity corresponds with Fig. 15.
41 under certain conditions. Note that the roll input also has
The effect of giving a pure yaw input (differential deflec- an effect on the yaw moment MX which is caused by an
42 tion of wing roots) is shown in Fig. 9. In this case a linear
43 imperfect differential deflection of the left and right wing
variation in yawing moment MX is observed, as is expected. roots.
44 The effect on the other two axis is again significantly smaller.
45 For the tested range of inputs (±25 deg) a variation in pitch VI. ATTITUDE E STIMATION AND C ONTROL
46 moment of approximately 5 N mm is observed. A lightweight custom made autopilot board of 1 gram
ce

47
A coupling effect on the Z-axis (roll) does occur when a is used to enable attitude estimation and control. Fig. 13
48
combination of pitch and yaw commands is given. This is shows the board attached to the vehicle. It features an
49
visible from the results in Fig. 10. Note that the maximum Atmel ATmega328P RISC-based 8-bit microcontroller that
50
input commands are only 50% of the maximum pitch and receives input from multiple sensors: 6-axis gyroscope and
51
yaw inputs, but that the sum of these inputs also covers accelerometer readings from an InvenSense MPU9150, 3-
52
the combination with maximum coupling effect. The results axis magnetometer readings from a Honeywell HMC5883L,
Ac

53 do not show perfect symmetry, but clearly show the depen- and barometric pressure readings from a Bosch BMP180.
54 dency on the combination of the two inputs. The variation The autopilot board also features a transceiver for two-
55 in roll moment is approximately 1.5 N mm, and clearly way communication (remote control and telemetry). The
56 influences the maximum yawing inputs that can be given. microcontroller runs an attitude estimation and control al-
57 For comparison, the result of the combinations of pitch and gorithm based on these inputs and controls the vehicle via
58
59
60
Page 7 of 13 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1

1
2 4
3 M
x
4 3 My

pt
Generated control moments [Nmm]
5 Mz
2
6
7 1
8
9 0

cri
10
-1
11
12 -2
13
14 -3
15
16 -4

us
-4 -3 -2 -1 0 1 2 3 4
17 Fig. 10. Generated roll moments (MZ ) as a coupling effect from roll command [mm]
18 combinations of pitch and yaw commands. The commands for both inputs
ranges to a maximum of %50, such that the summed inputs never exceed
19 %100. The flapping frequency in this test is ∼ 15 Hz.
Fig. 12. Generated moments around all three axis for a pure roll input. A
positive roll command means that the right wing root is shifted outwards.
20 The flapping frequency in this test is ∼ 15 Hz. All measurements are time-
21 averaged over 29 flapping cycles. Linear fits through these measurements
22 are shown. The error bars show the minimum and maximum time-averaged
23
24
25
26
27
28
an measurement over a single flapping cycle.
dM
29
30
31
32
33
34
35 Fig. 13. Damping structure as used to reduce the effect of mechanical
vibrations on the sensors of the autopilot. It consists of a (white) rigid foam
36 Fig. 11. Generated pitch moments (MY ) for combinations of pitch and yaw box, with (black) damping foam inside which encloses the fuslage. Both
37 commands. The commands for both inputs ranges to a maximum of %50, the battery and the autopilot board are firmly attached to the foam box such
38 such that the summed inputs never exceed %100. The flapping frequency that they form a rigid mass.
in this test is ∼ 15 Hz.
pte

39
40
frequency components that exceed the Nyquist frequency.
41 the integrated electronic speed controller and external servos. The result for the test with damping system shows that the
42
Initial tests reveal that the autopilot board cannot be frequency peak can now be found at the flapping frequency
43 attached to the main structure of the vehicle. The mechanical (∼12.5 Hz in this test).
44 vibrations from the system contain frequencies that exceed
45 the Nyquist frequency of the accelerometer sensors. There-
A. Attitude estimation
46 fore the board is attached to a substructure that includes foam The attitude is represented by Euler angles: roll (φ),
ce

47 parts for mechanical damping. The battery is also attached pitch (θ) and yaw (ψ). The accelerometer measurement
48 on it to lower the natural frequencies of the substructure.
T
a = [ax , ay , az ] is used to estimate the roll and pitch
49 This can also be seen in Fig. 13. angles; the magnetometer measurement m = [mx , my , mz ]
T

50 is incorporated to estimate the yaw angle. For the definition

To show the importance of this attachment, a comparative
51 of the Euler angles, the 2 − 1 − 3 rotation sequence (pitch-
test was performed to show the effect of using the substruc-
52 roll-yaw) is used. The angles are defined as:
ture. Fig. 14 shows the power spectral density (PSD) for az
Ac

53 for test flights with and without the damping system. The
54 sampling frequency is 500 Hz in both tests. The result for θe = atan( −a
az )
x
55 the stiff attachment shows noisy measurements that contain −a
56 φe = atan( √ y ) (1)
high frequency components close to the Nyquist frequency a2x +a2z
57 of the sensors (250 Hz). This indicates that there are also ψe = atan(
sin(φ) sin(θ)mx +cos(φ)my +sin(φ) cos(θ)mz
)
cos(θ)mx +sin(θ)mz
58
59
60
 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1 Page 8 of 13

1 Single-Sided Amplitude Spectrum of az (t)

2 Accelerometer and gyroscope fusion: The moving average
0.14
3 filter results in smoother accelerometer measurements that
no damping (stiff)
4 0.12 damping enhance the angle estimates, but at the cost of adding delay to

pt
5
damping+BW-filter the signal. These estimates are therefore fused with the faster
damping+MA-filter T
6 0.1 gyroscope measurements g = [gx , gy , gz ] using a first-order
7 complementary filter. The function of the filter for the pitch
8 0.08 angle is given by:
|az (f)|

cri
0.06
10 e − τ ) + (θ̂ + (cos(φ̂)gy + cos(θ̂) sin(φ̂)gz )∆t)τ (2)
θ̂ = θ(1
11
0.04
12 where θ̂ is the pitch estimate, θe is the measured pitch
13 0.02
angle based on Eq. 1, ∆t is the time step and τ is a weight
14 factor which is set at 0.9. The roll and yaw angle are filtered
15 0
in the same way using their corresponding components of
16 0 50 100 150 200 250 gyroscope measurements.

us
Frequency (Hz)
17 B. Attitude control
18 Fig. 14. Power spectral densities of the measured accelerations on the
19 X-axis (ax ). These are shown for four configurations. For a stiff connection The estimated attitude angles are used as feedback in
20 between autopilot and the frame, for a soft connection using a damping the attitude control loop. In this study the focus is on
structure, for a sift connection with Butterworth (BW) filter (4th order, hover and slow forward flight. Therefore the Euler angle
21 10 Hz cutoff), and for a soft connection with moving average (MA) filter
22 estimates are used in a straightforward manner to compute
23
24
25
26
27
28
(window length corresponds to measured flapping frequency)

In this equation, θ,
e φe and ψe are estimates of the pitch,
roll and yaw angle respectively. These are based on the ac-
celerometer measurement a and magnetometer measurement
an the error with the reference angles. It is thus assumed that the
definition of pitch, roll and yaw in Eqs. 1 and 2 corresponds
to the definitions in Fig. 3. PD-control is used to achieve
stabilisation. The trim settings for all angles are manually
tuned.
dM
m. This definition of the Euler angles ensures that the pitch VII. C HARACTERISTICS OF PLATFORM WITH TAIL
29 angle is defined also for high pitch angles (> 90◦ ) that occur The first DelFly design that is proposed in this study
30 during hover. Furthermore, the chosen rotation sequence also includes a tail. An active rudder surface is used in this
31 leads to more stable estimations of the roll and pitch angle. design because the type of servo used for roll is not powerful
32 The reason for this is that the DelFly (normally) flies with enough to implement it for wing-based roll control. A visual
33 small roll angles, which means that the gravitational force is overview of this DelFly design is shown in Fig. 15. The
34 expected to be predominately present on the X axis, around figure indicates the locations of the main components, as
35 hover, and also on the Z axis, in slow forward flight. From well as the location of the centre of gravity. Apart from
36 Equation 1 it can be seen that the roll and pitch angle the control and flapping mechanisms, the main components
37 estimates are not sensitive to noise measured on the Y axis. of the system are the autopilot board, the servos and the
38 The CORDIC algorithm [31] is used on board to calculate battery. Detailed information about the autopilot board is
pte

39 all the necessary trigonometric functions. It is important to discussed in Sec. VI. A SuperMicro Systems - Double Linear
40 mention that in this definition of the roll angle, the relation Servo (2.4 g) is used to actuate the left and right wing
41 between the roll angle and the body axes depends on the roots individually. The rudder is actuated by a Sub Micro
42 pitch angle. The definition of the roll and yaw body axes in LZ servo (0.5 g). A 180 mAh LiPo battery is used in
43 Fig. 3 corresponds to the hover condition (90◦ pitch). the experiments. A mass breakdown of the total system
44
Accelerometer data filtering: As is shown in Fig. 14 the is given in Table II. Furthermore an overview of vehicle
45
accelerometer measurements are mainly disturbed by the characteristics is presented in Table III.
46
periodic flapping motion of the wings (peak around 12.5 A second detail that is to be noted is the negative pitch
ce

47
Hz). To filter out this frequency and its harmonics, the angle of the horizontal tail section. In the current design this
48
accelerometer measurements are individually filtered per axis is necessary to allow stable forward flight, while previous
49
using a moving average filter. The window of the filter DelFly designs did not need this adjustment. This is due
50
is defined by the flapping frequency such that it exactly to the wing dihedral angle which is removed in the current
51
matches the period of the flapping cycle. The effect of this design. As a result, the thrust vector produced by the wings
52
filter is shown in Fig. 14. Furthermore, the effect of using a is almost collinear with the body X-axis (see Fig. 15). In the
Ac

53 Butterworth filter (4th order low-pass, 10 Hz cutoff) is also original design including the dihedral angle, the thrust vector
54 shown to demonstrate that such a filter does not completely acts further away from the body, which creates an additional
55 remove the peak around 12.5 Hz. Because of this specific pitch down moment (around the Y-axis). The interaction
56 approach, the attitude estimation function of the MPU9150 between this pitching moment and the aerodynamic moments
57 sensor is not used. acting on the wing and the tail results in a stable forward
58
59
60
Page 9 of 13 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1

1 TABLE III
2 D ESIGN AND PERFORMANCE CHARACTERISTICS
3
4 Total mass 21.1 g

pt
Wing span 28 cm
5 Wing area 224 cm 2
6 Wing aspect ratio 8 -
7 Flap amplitude 83 deg
Flight Speeds 0-0.5 m/s
8 Minimum turn radius 0.1 m
9

cri
10
11 was previously tested on a DelFly II using elevator and
12 aileron surfaces [20]. In these tests it was shown that the
13 attitude control loop enabled the stabilisation of the pitch
14 angle. However, the tests showed that the effectiveness of
15 the ailerons is insufficient to control the yaw angle during
16

us
hovering. The effectiveness of the ailerons depends mainly
17 on the flow due to the forward velocity of the vehicle. While
18 hovering, the forward velocity is zero. The flow induced by
19 the wings has a much smaller influence. The hovering tests
20 in the current study are performed to show that the control
Fig. 15. Overview of the DelFly II with wing vectoring system in hover
21 attitude (90 deg pitch). This design is used for the flight tests described in mechanism allows both pitch and heading control in this
22
23
24
25
26
27
28
mechanism is shown in Fig. 7. A detailed image of the control mechanism

TABLE II
M ASS BREAKDOWN OF DESIGN WITH TAIL
an
Sec. VIII. The location of the center of gravity is indicated, as well as the
definition of the (right-handed) body-axes. A detailed image of the flapping

is shown in Fig. 4. The damping structure attachment of the battery and the
autopilot board is more clearly visualised in Fig. 13.
condition.
roll error [°]

10

0
onboard estimate
external tracking
dM
29 Component Mass (g) −5
30 Motor 1.0
−10
Flapping mechanism 2.7
31 Wings (2) 1.4
100 105 110 115 120

32 Control mechanism 1.5 10

33 Servos(3) 2.9
pitch error [°]

Autopilot board 1.0 5

34 Fuselage 1.0
35 Damping structure 0.8
0

36 Tail 2.3 −5
Battery 5.5
37 Wiring and glue 1.0 −10
38 Total mass 21.1
100 105 110 115 120
pte

39 10
40
heading error [°]

5
41 flight pitch angle. For the original design, the pitch angle
42 is between 70◦ and close to 90◦ (almost hover). For the
0

43 current design this stable pitch angle is very close to 90◦ , −5

44 which restricts forward speed. The negative pitch angle of the −10
45 horizontal tail section in the current design, in combination
100 105 110
flight time [s]
115 120

46 with a negative pitch angle of the wings, results in a stable

ce

47 forward flight motion. In this case the pitch angle is still close
48 to 90◦ , but the wing and tail vectoring results in an additional Fig. 16. Hovering test results showing the attitude tracking performance
49 forward force. Another method that will be explored in the of the vehicle for static reference angles (pitch 90◦ and roll 0◦ ) such that
the vehicle hovers. Both onboard attitude estimates and externally tracked
50 future is to move the center of gravity along the direction of attitude values are shown for comparison.
51 the body Z-axis by repositioning heavy components.
52
The slow forward flight condition is also tested as this is an
Ac

53 VIII. F LIGHT T ESTS OF PLATFORM WITH TAIL

important flight condition for the DelFly II when performing
54 Flight tests were performed to demonstrate the effective- autonomous obstacle avoidance and navigation in indoor
55 ness and performance of the proposed control mechanism. A environments. In a previous study aileron surfaces were used
56 distinction is made between two different flight conditions: to enable smooth turns that allow the vehicle to perform
57 hovering and slow forward flight. The hovering condition visual navigation at the same time [18]. However, situations
58
59
60
 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1 Page 10 of 13

1
2 Position change Heading change Forward flight Hover flight 180−turn
1.2
3 0.2 100
1
4

y−position [m]
heading error [°]
y−position [m]

pt
5 0 0.8

6 0.1 0.6

7 −100 0.4

8 0.2

9 0
0 0.1 0.2
−200
734.3 734.8 735.3
0
−1.5 −1 −0.5 0 0.5 1 1.5

cri
10 x−position [m] flight time [s] x−position [m]

11 0.2 100 Fig. 18. Horizontal flight track (top view) of the vehicle during a stop-and-
12
heading error [°]
turn manoeuvre. Initially the vehicle flies forward, and is then commanded to
y−position [m]

13 0 transition to hover attitude. Once stable hover has been reached the vehicle
14 0.1
is instructed to perform a 180-turn.
15 −100
16 onboard estimate external tracking

us
17 0 −200 20
0 0.1 0.2 792.4 792.9 793.4
18 10

roll error [°]

x−position [m] flight time [s]
19 0
0.2 100
20 −10
heading error [°]

21
y−position [m]

0 −20
22
23
24
25
26
27
28
0.1

0
0 0.1
x−position [m]
0.2
−100

−200
800.6 801.1
flight time [s]
an
801.6
pitch error [°]

−10

−20
58

20

10

0
59 60 61 62 63 64
dM
58 59 60 61 62 63 64
29
20
30 Fig. 17. Position and heading responses of the vehicle during three 180-
heading error [°]

turns. The turns were performed while starting from the hovering condition.
31 10

32 0

33 occur where the vehicle needs to turn around quickly to avoid −10
34 collisions with obstacles that were not detected sufficiently −20
35 ahead. The flight tests in this study show that in such a case
58 59 60 61
time [s]
62 63 64

36 the vehicle is capable of smoothly transitioning from forward

37 flight to hover, and that it can perform a fast turn to instantly
38 reverse its heading.
Fig. 19. Attitude tracking performance of the vehicle in slow forward flight.
Both onboard attitude estimates and externally tracked attitude values are
pte

39 The tests are performed in a 10×10×7 m flight arena shown for comparison.
40 equipped with an OptiTrack Motion tracking system consist-
41 ing of 24 Flex13 cameras. This data is used for post-flight
42 analysis of the flight performance. heading responses show similar results, the lateral motion of
43 the vehicle is more random.
Results from the hovering test are shown in Fig. 16.
44
The tracking performance of the roll, pitch and yaw angles In Fig. 18 the flight path of the vehicle is shown for a
45 are shown for 20 seconds of flight. It can be observed test where the vehicle initially flies with a forward speed
46 that the onboard attitude estimates correlate well with the of 29 cm/s and is then commanded to hover and turn
ce

47 externally observed attitude angles. Furthermore the results 180 degrees. The figure shows the horizontal motion of the
48 show that during this test the maximum attitude errors for vehicle during these three different phases. This result shows
49 all three attitude angles are in the order of ±5◦ . This clearly that the vehicle is able to perform a stop-and-turn manoeuvre
50 illustrates the effectiveness of the wing control mechanism during slow forward flight that requires the vehicle to move
51 in controlling the pitch and yaw angles. over a distance of less than 25 cm. The manoeuvre takes
52
Attitude and position data results from fast 180◦ turns 2.5 s. The attitude tracking performance of the vehicle during
Ac

53 during hover are shown in Fig. 17. The results show that the the forward flight phase of this test is shown in Fig. 19.
54 vehicle is able to smoothly turn around in one second while These results show that the attitude tracking performance
55 requiring a turn radius of less than 0.1 m. From the position of heading, pitch and roll are worse compared to hovering
56 plots in the figure it can be observed that the response of flight. Since the tracking performance is worse for all three
57 the vehicle during the turn is different each time. While the attitude angles it is not possible to identify a main reason for
58
59
60
Page 11 of 13 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1

1 TABLE IV
2 this result. However, the poor pitch stability in this test seems
M ASS BREAKDOWN OF TAILLESS DESIGN
3 to point out that the current tail configuration, as discussed in
4 Sec. VII, does not provide the same forward flight stability Component Mass (g)

pt
benefits that were observed in flight tests with the DelFly Motor 1.0
5 Flapping mechanism 2.7
6 Explorer [18]. Wings (2) 2.5
Control mechanism 1.5
7 IX. C HARACTERISTICS OF PLATFORM WITHOUT TAIL Servos(3) 3.9
8 Autopilot board 1.0
In this section, we present the proposed mechanism with-
9 Fuselage 0.5

cri
out the additional actuator for roll. The main difference of Damping structure 0.1
10 Battery 5.5
this design, shown in Fig. 21, is that it has no tail. This has
11 Wiring and glue 1.0
several consequences for the total mass of the system, which
12 Total mass 19.7
is reduced to 19.7 g. A mass breakdown of this design can
13
be found in Table IV. A significant amount of mass is saved
14
by removing the tail (2.3 g). As a consequence, the length
15
of the carbon fuselage rod can be reduced, saving 0.5 g.
16

us
Instead of the two linear servos, two HobbyKing HK-5330S
17
Ultra-Micro Digital Servos (1.7 g) are used in combination
18
with the Sub Micro LZ servo (0.5 g). The damping structure
19 t=4
used in the first design is replaced by a more minimalistic
20
version that uses only a foam layer around the fuselage rod
21 t=2
that prevents direct contact between the fuselage and the
22
23
24
25
26
27
28
carbon wing root rod from wearing the mylar foil. Replacing
this wing section with only tape to save weight is not an
an
battery-autopilot combination. This solution is visualised in
the bottom left image in Fig. 21. In this design the wings are
heavier, 1.3 g each. This is the result of adding an extra layer
of tape on the middle section of the wing. This prevents the
t=0

Fig. 20. Hovering flight demonstration of the tailless design. The image
shows the DelFly at three instances in time during 4 seconds. The three
instances show that the DelFly slowly drifts to the left and a bit forward,
and is stable on all axes.
dM
option, as the inelastic property of the mylar foil is needed
29
to maintain the required wing tension.
30
The two HK-5330S servos are used in this design for flight times, over 5 minutes, are expected to be possible in
31
their fast response rate. According to the manufacturers terms of battery capacity, but are hindered by instabilities
32
specifications: 60◦ in 0.04 s at 4.2 V. In comparison, the of the system. Instability occurs due to disturbing wind
33
linear servos used in the design with the tail have a specified conditions, and are mainly an issue for roll stability. As the
34
speed of 0.12 s at 4.2 V. The rotary servos are also more results from Sec. V indicate, the proposed mechanism is the
35
powerful, which also influences the response time of the least effective in roll.
36
mechanism in flight. For stable flight it is required to control Pitch instability also forms an issue using this wing
37
the pitch and roll angles. The middle right image in Fig. 21 control mechanism. The reason for this is that the location
38
shows how one HK-5330S servo is used to actuate the roll of the center of gravity (CoG) influences the pitch control
pte

39
mechanism. The middle left image in Fig. 21 shows how effectiveness of the mechanism. This effectiveness increases
40
the second HK-5330S servo is implemented to control the if the CoG is located further away from the leading edge of
41
pitch mechanism of both wing roots. The small Sub Micro the wing. At the same time, the CoG is ideally located at
42
LZ servo is integrated in this mechanism. This servo is the aerodynamic center (AC) of the wing, such that flight
43
not able to deliver the force and speed required for pitch speed has no influence on the pitching moment coefficient
44
stabilisation, but forms a lightweight solution to introduce of the wing. This forms a conflict, because the pitch control
45
an offset between the left and right wing roots for enabling effectiveness of the mechanism is too small when the CoG
46
yaw control. To allow more aggressive control on the pitch is located at the AC. Therefore the CoG is located further aft
ce

47
and yaw axes, the two wing roots can also be actuated by in the current vehicle configuration. This allows only very
48
two individual HK-5330S servos. This adds approximately limited flight speeds, as pitch instabilities occur otherwise.
49
1 g to the total mass of the system. The vehicle can thus be trimmed for hovering flight and also
50
51 for slow forward flight, as can be seen in the first two videos
X. F LIGHT T ESTS OF PLATFORM WITHOUT TAIL
52 of the associated playlist1 . Furthermore, the roll control
Flight tests with the system in Fig. 21 were performed to
Ac

53 mechanism allows the vehicle to perform lateral maneuvers.

test whether the proposed wing control mechanism allows the More detailed tests and design improvements are required
54 flapping wing vehicle to perform stable flight without having
55 to explore the stability limits of the system. The effectiveness
a tail. Fig. 20 shows part of a test flight, demonstrating a of the wing control mechanism can be improved by several
56 stable, slowly drifting vehicle. Flight times of over 1 minute
57 of continuous hovering flight have been realized. Longer 1 https://goo.gl/vN6KP3
58
59
60
 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1 Page 12 of 13

1
2
3
4

pt
5
6
7
8
9

cri
10
11
12
13
14
15
16

us
17
18
19
20
21
22
23
24
25
26
27
28
an
dM
29
30
31
32
33
34
35 Fig. 21. Tailless DelFly design that uses the proposed wing control mechanism. Top: Aft view showing a clear overview of the vehicle and illustrating
36 that all components are compactly concentrated around the fuselage. Middle left: Detailed view showing the mechanism for pitch and yaw control. Middle
37 right: Detailed view showing the mechanism for roll control. Bottom left: Side view of the vehicle to clearly show where all components are placed. Note
how the foam is used in the connection between the battery and the fuselage. Bottom right: Overview of the vehicles’ appearance when the wings are
38 opened.
pte

39
40
41 design changes. Current experiments also indicate that even show that pitch and yaw moments of sufficient magnitude can
42 a small tail can improve the vehicle stability considerably be created for hover stabilisation. The magnitude of the pro-
43 (third video of the playlist). A combination of a right-sized duced roll moments is smaller, and about the same order as
44 tail and the proposed wing control mechanism is therefore the maximum roll moments that are produced by the coupling
45 expected to deliver a high reliability with the ability to effect of combined pitch and yaw commands. The severity
46 perform aggressive maneuvers, especially in pitch and yaw. of external disturbances therefore determines whether roll
ce

47 stability is possible. Real flight tests are presented that show

48 XI. C ONCLUSIONS that stable hover flight is realised in combination with a tail
49 and that this system is also able to perform fast turns in this
In this study a control mechanism is presented for a
50 condition. Flight test with a tailless configuration are also
flapping wing MAV having two pairs of wings in a bi-
51 presented, demonstrating hovering and slow forward flight,
plane configuration. The mechanism serves as a light weight
52 and also sideways flying. Other experiments indicate that
control solution that allows for fast and large wing root
Ac

53 deflections. Both symmetric and asymmetric wing root de- aggressive pitch and yaw manoeuvres can be combined in a
54 flections are possible by which pitch and yaw moments can configuration with a tail.
55 be created. Furthermore the proposed mechanism allows for
56 lateral translations of the wing roots by which the tension of
57 the wings can be adapted, enabling roll moments. Static tests
58
59
60
Page 13 of 13 AUTHOR SUBMITTED MANUSCRIPT - BB-101277.R1

1
2 R EFERENCES [21] G. C. H. E. de Croon, M. Perin, B. D. W. Remes, R. Ruijsink,
and C. de Wagter, The DelFly: Design, Aerodynamics, and Artificial
3 [1] C. De Wagter, D. Dokter, G. C. H. E. De Croon, and B. D. W. Intelligence of a Flapping Wing Robot. Springer Netherlands, 2016.
Remes, “Multi-lifting-device uav autonomous flight at any transition
4 [22] H. Tokutake, S. Sunada, and Y. Ohtsuka, “Active control of flapping

percentage,” Proceeding of: EuroGNC, vol. 2013, 2013.
5 [2] A. F. Şenkul and E. Altuğ, “System design of a novel tilt-roll rotor
6 quadrotor uav,” Journal of Intelligent & Robotic Systems, vol. 84, no.
1-4, pp. 575–599, 2016.
7 [3] K. Wang, Y. Ke, and B. M. Chen, “Autonomous reconfigurable hybrid
8 tail-sitter uav u-lion,” Science China Information Sciences, vol. 60,
9 no. 3, p. 033201, 2017.
pt
wings using wing deformation,” Transactions of the Japan Society for
Aeronautical and Space Sciences, vol. 52, no. 176, pp. 98–103, 2009.
[23] H. V. Phan and H. C. Park, “Generation of control moments in an
insect-like tailless flapping-wing micro air vehicle by changing the
stroke-plane angle,” Journal of Bionic Engineering, vol. 13, no. 3, pp.
449–457, 2016.
[24] C. De Wagter, M. Karásek, and G. de Croon, “Quad-thopter: Tailless

[4] C. De Wagter, R. Ruijsink, E. Smeur, K. van Hecke, F. van Tienen,
10 E. van der Horst, and B. Remes, “Design, control and visual navigation
11 of the delftacopter,” arXiv preprint arXiv:1701.00860, 2017.
12 [5] W. Send, M. Fischer, K. Jebens, R. Mugrauer, A. Nagarathinam, and
F. Scharstein, “Artificial hinged-wing bird with active torsion and
13 partially linear kinematics,” in 28th Congress of the International
14 Council of the Aeronautical Sciences, 2012, pp. 23–28.
15 [6] L. Roberts, H. A. Bruck, and S. K. Gupta, “Autonomus loitering
control for a flapping wing miniature aerial vehicle with indepen-
16
cri
flapping wing robot with 4 pairs of wings,” 9th international micro
air vehicles, 2017.
[25] S. B. Fuller, E. F. Helbling, P. Chirarattananon, and R. J. Wood, “Using
a mems gyroscope to stabilize the attitude of a fly-sized hovering
robot,” in IMAV 2014: International Micro Air Vehicle Conference
and Competition 2014, Delft, The Netherlands, August 12-15, 2014.
Delft University of Technology, 2014.
[26] E. F. Helbling, S. B. Fuller, and R. J. Wood, “Pitch and yaw control
of a robotic insect using an onboard magnetometer,” in Robotics and

us
dent wing control,” in ASME 2014 International Design Engineering Automation (ICRA), 2014 IEEE International Conference on. IEEE,
17 Technical Conferences and Computers and Information in Engineering 2014, pp. 5516–5522.
18 Conference. American Society of Mechanical Engineers, 2014, pp. [27] S. Tijmons, G. C. H. E. de Croon, B. D. W. Remes, C. De Wagter, and
V05AT08A013–V05AT08A013.
19 [7] N. Gaissert, R. Mugrauer, G. Mugrauer, A. Jebens, K. Jebens, and
M. Mulder, “Obstacle avoidance strategy using onboard stereo vision
on a flapping wing mav,” IEEE Transactions on Robotics, 2017.
20 E. M. Knubben, “Inventing a micro aerial vehicle inspired by the me- [28] S. S. Baek, F. G. Bermudez, and R. Fearing, “Flight control for target
21 chanics of dragonfly flight,” in Towards Autonomous Robotic Systems. seeking by 13 gram ornithopter.” in IEEE/RSJ Int Conf on Intelligent
Springer, 2013, pp. 90–100.
22 Robots and Systems., 2011.
23
24
25
26
27
28
[8] M. Keennon, K. Klingebiel, H. Won, and A. Andriukov, “Development

in 50th AIAA Aerospace Science Meeting, 2012, pp. 6–12.

an
of the nano hummingbird: A tailless flapping wing micro air vehicle.”

[9] K. Y. Ma, P. Chirarattananon, S. B. Fuller, and R. J. Wood, “Controlled

flight of a biologically inspired, insect-scale robot.” Science, vol. 340,
no. 6132, pp. 603–607, 2013.
[10] M. Karásek, A. Hua, Y. Nan, M. Lalami, and A. Preumont, “Pitch
and roll control mechanism for a hovering flapping wing mav,”
[29] G. C. H. E. de Croon, M. A. Groen, C. de Wagter, B. D. W. Remes,
R. Ruijsink, and B. W. van Oudheusden, “Design, aerodynamics, and
autonomy of the delfly,” Bioinspiration and Biomimetics, vol. 7, no. 2,
2012.
[30] S. Sane, “Review: The aerodynamics of insect flight,” The journal of
Experimental Biology, vol. 206, no. 23, pp. 4191–4208, 2003.
[31] J. E. Volder, “The cordic trigonometric computing technique,” Elec-
tronic Computers, IRE Transactions on, no. 3, pp. 330–334, 1959.
dM
International Journal of Micro Air Vehicles, vol. 6, no. 4, pp. 253–264,
29 2014.
30 [11] D. Coleman, M. Benedict, and I. Chopra, “Design, development and
flight testing of robotic hummingbird,” in 71st Annual Forum of the
31 American Helicopter Society, 2015.
32 [12] H. V. Phan, T. Kang, and H. C. Park, “Design and stable flight of a
33 21 g insect-like tailless flapping wing micro air vehicle with angular
rates feedback control,” Bioinspiration & Biomimetics, vol. 12, no. 3,
34 p. 036006, 2017.
35 [13] D. B. Doman, M. W. Oppenheimer, and D. O. Sigthorsson, “Dynamics
36 and control of flapping wing mavs,” in Handbook of Unmanned Aerial
Vehicles. Springer, 2015, pp. 329–346.
37 [14] P. Zdunich, D. Bilyk, M. MacMaster, D. Loewen, J. DeLaurier,
38 R. Kornbluh, T. Low, S. Stanford, and D. Holeman, “Development
pte

39 and testing of the mentor flapping-wing micro air vehicle,” Journal of

Aircraft, vol. 44, no. 5, pp. 1701–1711, 2007.
40 [15] C. Richter and H. Lipson, “Untethered hovering flapping flight of a
41 3d-printed mechanical insect,” Artificial life, vol. 17, no. 2, pp. 73–86,
42 2011.
[16] Q. V. Nguyen, W. L. Chan, and M. Debiasi, “Performance test of a
43 hovering flapping wing micro air vehicle with double wing clap-and-
44 fling mechanism,” in International Micro Air Vehicle Conference and
45 Competitions (IMAV), 2015.
[17] Q.-V. Nguyen, W. L. Chan, and M. Debiasi, “Hybrid design and
46 performance tests of a hovering insect-inspired flapping-wing micro
ce

47 aerial vehicle,” Journal of Bionic Engineering, vol. 13, no. 2, pp. 235–
48 248, 2016.
[18] C. De Wagter, S. Tijmons, B. D. W. Remes, and G. C. H. E. de Croon,
49 “Autonomous flight of a 20-gram flapping wing mav with a 4-gram
50 onboard stereo vision system,” in Robotics and Automation (ICRA),
51 2014 IEEE International Conference on. IEEE, 2014, pp. 4982–4987.
[19] G. C. H. E. De Croon, K. M. E. De Clercq, R. Ruijsink, B. Remes,
52 and C. De Wagter, “Design, aerodynamics, and vision-based control of
Ac

53 the delfly,” International Journal of Micro Air Vehicles, vol. 1, no. 2,

54 pp. 71–97, 2009.
[20] J. L. Verboom, S. Tijmons, C. De Wagter, B. D. W. Remes,
55 R. Babuska, and G. C. H. E. de Croon, “Attitude and altitude
56 estimation and control on board a flapping wing micro air vehicle,” in
57 Robotics and Automation (ICRA), 2015 IEEE International Conference
on. IEEE, 2015, pp. 5846–5851.
58
59
60