2010 IEEE International Conference on Robotics and Automation

Anchorage Convention District

May 3-8, 2010, Anchorage, Alaska, USA

A Hovering Control Strategy for a Tail-Sitter VTOL UAV that

Increases Stability Against Large Disturbance
Takaaki Matsumoto ∗1 , Koichi Kita∗2 , Ren Suzuki∗1 , Atsushi Oosedo ∗1 ,
Kenta Go∗1 , Yuta Hoshino∗1 , Atsushi Konno∗1 and Masaru Uchiyama∗1
∗1 Department of Aerospace Engineering, Tohoku University.

Aoba-yama 6-6-01, Sendai, Miyagi, 980-8579, Japan

∗2 Fuji Heavy Industries Ltd., 1-1-11

Yonan, Utsunomiya, Tochigi, 320-8564, Japan

{takaaki, konno, uchiyama}@space.mech.tohoku.ac.jp, [email protected]

Hovering
Abstract— The application range of UAVs (unmanned aerial
vehicles) is expanding along with performance upgrades. Ver-
tical take-off and landing (VTOL) aircraft has the merits of
both fixed-wing and rotary-wing aircraft. Tail-sitting is the
simplest way for the VTOL maneuver since it does not need Transition from Takeoff Transition to Landing
extra actuators. However, conventional hovering control for
a tail-sitter UAV is not robust enough against large distur-
bance such as a blast of wind, a bird strike, and so on. It
is experimentally observed that the conventional quaternion
feedback hovering control often fails to keep stability when the
control compensates large attitude errors. This paper proposes Fig. 1. Takeoff and landing of the tail-sitter VTOL aircraft.
a novel hovering control strategy for a tail-sitter VTOL UAV
that increases stability against large disturbance. In order to wingtip rotors which generate a rotational force countering
verify the proposed hovering control strategy, simulations and the motor torque to their MAV [7].
experiments on hovering of the UAV are performed giving large However, those tail-sitter UAVs have some complex equip-
attitude errors. The results show that the proposed control
strategy successfully compensates initial large attitude errors ments such as a coaxial contra-rotating propeller [1], a ducted
keeping stability, while the conventional quaternion feedback fan and fins [2], side-by-side rotors [4],[5], and wingtip rotors
controller fails. [7] for the tail-sitting VTOL maneuver.
Only few attempts have been made to develop tail-sitter
I. INTRODUCTION UAVs without any extra equipment so far. However, since
VTOL UAVs make missions possible which are normally these simple tail-sitter UAVs have no extra equipment for
impossible to accomplish using either fixed-wing or rotary- countering the motor torque, robust stationary hovering is
wing UAVs alone; for example, search and rescue operations more difficult than other robots with complex equipments.
covering a broad area located at the rooftop of a building. Large disturbances in hovering such as strong wind or bird
There are several ways to perform VTOL maneuvers such as impact are major problems to overcome.
tilting-rotor, tilting-wing, thrust-vectoring and tail-sitting etc. Frank et al. have succeeded in the indoor flight experiment
The simplest way is tail-sitting since it does not need extra using a commercially available R/C acrobatic airplane and
actuators for the VTOL maneuver. A simple mechanism is the motion capture system [8]. However, since the flight
preferable for UAVs, because weight saving is crucial for the experiments were performed in a room, there was no dis-
VTOL maneuver and has the advantage of cost saving. Tail- turbance such as strong wind which is a major problem to
sitter VTOL aircraft switches between level flight mode and overcome [8]. Knoebel et al. proposed a new airframe design
hover mode by changing the pitch attitude of the fuselage [9]. They are working on flight tests using a commercially
by 90 ° as shown in Fig. 1. available single propeller R/C model which represents the
US Air Force Research Lab and AeroVironment Inc. “XFY1”. However, hovering performance with large distur-
have developed “SkyTote” which is equipped with a coaxial bance were not reported in [9]. Johnson et al. have developed
contra-rotating propeller [1]. The Defense Advanced Re- “GTEdge” which is a large scale R/C airplane weighting
search Projects Agency (DARPA) and Aurora Flight Sciences about 15 kg and studied the tail-sitter maneuver [10],[11].
have developed “GoldenEye” which is equipped with a They have succeeded transition flight and hovering; however,
ducted fan. It uses fins outside the duct during level flight the robustness of hovering control was not discussed.
and fins in the duct during hovering [2]. Stone developed “T- This paper is intended to propose a novel control strategy
Wing” which has a canard wing and tandem rotors [3],[4]. for robust hovering when large attitude errors are generated
Kubo and Suzuki proposed a twin-fuselage plane [5]. Green by some disturbances. Simulations and experiments on hov-
and Oh developed a micro air vehicle [6], and added two ering control of the UAV are performed giving large attitude

978-1-4244-5040-4/10/$26.00 ©2010 IEEE 54

 R/C Transmitter
Ground
Aerial Robot

Aileron Servo
R/C Receiver
Attitude Sensor
Module
Elevator Servo
Fig. 2. Tail-Sitter UAV.
Ultrasonic Sensor

errors to verify the robustness against the errors. The results Rudder Servo
show that the proposed strategy successfully compensates the Pressure Altitude Microcomputer
large errors, while the conventional control strategy failed to Sensor

stabilize the UAV. The UAV used in this paper is equipped

with all necessary sensors and computers on the fuselage
[12]. GPS Module Micro-SD Card Module Motor Controller Thrust Motor

II. SYSTEM CONFIGURATION Fig. 3. On-board electronics system.

An overview of the tail-sitter VTOL UAV is shown in

Aileron
Fig. 2. The main wingspan is 1.0 m, and the weight is
0.75 kg. The main and tail wings are parts of commercially X
available R/C airplane (Hyperion Co., Sniper 3D), and other Elevator
parts such as the body are newly developed. The motor and
propeller, of which the static thrust amounts to 120 % of the
fuselage weight at a continuous maximum motor load are
selected. Y
The UAV is equipped with the following processors and Rudder
sensors. Z
• A microcomputer board (Alpha Project Co., STK-7125) Fig. 4. Aircraft body coordinates and control surfaces.
that has an SH2 microcomputer made by Renesass
Technology Co. The microcomputer calculates control
input based on each sensor data, and sends pulse-width III. HOVERING CONTROL
modulated (PWM) signals to servo motors to control
surfaces (aileron, elevator, and rudder) and the thrust A. Quaternion Feedback Control
motor.
• An attitude sensor module (Microstrain Co., 3DM-
The earth fixed coordinate system defines X axis as true
GX1). This module provides the attitude, azimuth, north, Y axis as east, and Z axis as perpendicular downward.
three-axis angular velocity and acceleration. The sen- The fuselage fixed coordinate system is defined as shown
sor’s datasheet gives its attitude angle accuracy as ±2 °. in Fig. 4 as a principal axis of inertia. The attitude of
• An ultrasonic sensor to detect altitude when the air-
the fuselage is expressed with respect to the earth fixed
craft’s distance from the ground is less than 6 m. coordinate system.
• An atmospheric pressure sensor to detect altitude when Because the tail-sitter maneuver covers a wide range of
aircraft distance from the ground is more than 6 m. attitudes, quaternion expression which theoretically has no
• A global positioning system (GPS) receiver module singularity is used as a method of describing the attitude.
(Garmin Co., GPS 18-5Hz). The GPS module obtains Quaternion expresses the attitude by a three dimensional unit
absolute position on the earth and absolute velocity of vector r and its rotation angle ζ , as follows:
the three axes.  
• A micro-SD card module to record flight data and other cos(ζ /2)
q= = [q0 q1 q2 q3 ]T . (1)
information for postexperiment analysis. r sin(ζ /2)
• A R/C receiver to control aircraft by a human in emer-
gency. The main computer receives commands from Quaternion feedback is generally used for UAVs. In con-
an R/C transmitter, but these are not used in control trolling the attitude of the UAV, following three quaternions
calculation. are defined: q r that shows the desired reference attitude, q c
A configuration diagram of the electronic system is shown that shows the current attitude, and q e that shows the error
in Fig. 3. or deviation between q r and qc . The qe is shown as follows

55
 T mg XN XN XN
XC
θ θtwist
Propeller mg tanθ
≈ mgθ
Center of Gravity θtilt
YN
v

Downwash mg
(a) Step 1. Tilt (b) Step 2. Twist (c) Reference Attitude
(Current Attitude)
Control Surface
Fig. 6. Concept of the resolved tilt-twist angle control.
Fig. 5. Operating principle for hovering.
Step 1 Derive pitch and yaw errors based on an analogy
of inverted pendulum
by using q c and qr [13]: The first step derives the pitch and yaw errors.
Current attitude CO R and reference attitude O
 
qr0 qr1 qr2 qr3 N R of the
−qr1 qr0 qr3 −qr2  UAV are defined as follows:
qe =  q , (2)
−qr2 −qr3 qr0 qr1  c O
CR ≡ [exC eyC ezC ], (4)
−qr3 qr2 −qr1 qr0
O
NR ≡ [exN eyN ezN ], (5)
where qr = [qr0 qr1 qr2 qr3 ]T . The vector part of q e
(qe1 , qe2 , qe3 ) calculated by (2) shows amount of error about where e jC and e jN ( j = x, y, z) are the unit vectors
each axis in the body coordinates. along j axis of the body coordinate frame with
Each three axes are controlled by a PID controller. The respect to the world coordinate frame at current
control command is sent to control surfaces corresponding attitude and reference attitude, respectively.
to each axis as follows: Considering the UAV as an inverted pendulum, its er-

ror angles can be calculated. The attitude of inverted

δi = −2(KP qei + KI qei dt + KD q̇ei ), (3) pendulum is defined as follows,
 
where δ1 , δ2 and δ3 are the aileron angle, elevator angle r11E r12E r13E
and rudder angle, respectively. The PID gains are provided RE =O TO
N R C R = r21E
 r22E r23E  . (6)
by the ultimate sensitivity method, and tuned by trial and r31E r32E r33E
error. The attitude is operated by blowing a slip stream of The X axis elements of R E gives pitch and yaw errors
the propeller to each control surface as shown in Fig. 5. as follows,
B. Resolved Tilt-Twist Angle Feedback Control θY = atan 2(r31E , r11E ), (7)
Quaternion feedback works well when attitude errors are
θZ = atan2(r21E , r11E ), (8)
not very large. However, when the rolling error is large, the
quaternion feedback control presented in the previous section where atan 2(y, x) is a function that calculates
may fail to stabilize the UAV. tan−1 (y/x). θY and θZ define the tilt angle of inverted
For example, let (α β γ ) = (0 90 0) ° in a reference attitude pendulum θtilt as follows:
and let (α β γ ) = (180 80 0) ° be the current attitude, where 
α , β , and γ are ZYX Euler angles (yaw, pitch, and roll, θtilt = θY2 + θZ2 . (9)
respectively). In this case, the error quaternion is calculated
Step 2 Derive roll error
as [0 − 0.57 0 − 0.34] T . This error quaternion derives no
The second step derives the roll error. The rotation
error around the Y axis of the aircraft body coordinates.
of θtilt is given by Rodrigues’ rotation formula as
Therefore, the pitch error (error around Y axis) is not
follows,
compensated in the beginning of the quaternion feedback 
2
control.  E + v̂sin θtilt + v̂ (1 − cos θtilt ), for RE = E

We propose a novel hovering control strategy based on an Rv = (10a)
analogy of inverted pendulum to achieve robustness against

for RE = E (10b)

E,
large attitude errors. The proposed hovering control strategy
is named Resolved tilt-twist angle control. In this control, where E is a 3 × 3 identity matrix, v is the rotation
attitude error is resolved into the tilt and twist angles. The tilt axis vector given by the normalized cross product of
angle is composed of two angles of orthogonal axes. Fig. 6 exC and exN as follows,
shows the concept of the control. The resolved tilt-twist angle exC × exN
v= ≡ [vx vy vz ]T . (11)
control is composed of the following four steps. |exC × exN |

56
 The hat operator transforms a vector v into a skew- C. Altitude Control
symmetric matrix as follows, The altitude controller is independently designed. The
  desired propeller reference rotation speed is calculated from
0 −vz vy the reference and current altitudes. Altitude control is gener-
v̂ =  vz 0 −vx  . (12) ally possible without propeller rotation speed feedback, but
−vy vx 0 control performance is deteriorated by changes in battery
conditions and motor load due to disturbance. Therefore, a
The UAV attitude after compensating θ tilt (see the
feedback control of propeller rotation speed is introduced in
Fig. 6(b)), is given using R v as follows:
altitude control system to enhance robustness against these
changes. Control gains of the altitude control system were
RP = Rv CO R ≡ [exP eyP ezP ], (13) determined through simulation.
where e jP ( j = x, y, z) are the unit vectors along j IV. SIMULATION
axis of the body coordinate frame after compensating A. Mathematical Model
θtilt with respect to the world coordinate frame. The To evaluate the hovering algorithms, a two-dimensional
absolute roll error is defined as follows, tail-sitter UAV simulator was developed. The translational
  mathematical model of the UAV in the aircraft body coordi-
−1 e zP · e zN nates is represented as follows,
θtwist = cos . (14)
|ezP ||ezN |  
m U̇ + QW = L sin α − D cos α − mgsin θ + T − DP , (19)
Since aircraft roll angle range is −180 ° ∼ 180 °, the  
m Ẇ − QU = −L cos α − D sin α + mg cos θ , (20)
sign of the roll error must be identified. In order to
identify the sign of the roll error θ X , θsign is defined where U and W are velocities along the X and Z axes in the
as follows: aircraft body coordinates, L and D are lift and drag forces,
 
α is the attack angle, θ is the pitch angle, m is the fuselage
−1 eyP · ezN mass, g is the gravitational acceleration, T is the thrust force,
θsign = cos . (15) DP is the propeller drag force, and Q is the angular velocity
|eyP ||ezN |
of the Y axis around the aircraft body coordinates.
Using θsign , the roll error θ X of the UAV is identified The rotational mathematical model of the UAV is repre-
as follows: sented as follows,

θtwist , for θsign ≤ π2 (16a) Ixx Ṗ + C p P = Ma + M p, (21)
θX = π Iyy Q̇ + Cq Q = Mt + Me , (22)
−θtwist . for θsign > 2 (16b)
where P and Q are angular velocities around the X and Y
Step 3 Projection of pitch and yaw errors onto the rolling axes of the aircraft body coordinates, I xx and Iyy are inertia
body coordinate frame around the X and Y axes of the aircraft body coordinates, C p
In order to simultaneously compensate pitch, yaw, and Cq are viscous resistance coefficients, Ma and M p are the
and roll errors, the pitch and yaw errors must be aileron and propeller rolling momentum around the X axis
projected onto the body coordinate frame which is of the aircraft body coordinates, M t and Me are fuselage
rolling with respect to the world coordinate frame. and elevator pitching momentums around the Y axis of the
Errors around each axis in the aircraft body coordi- aircraft body coordinates.
nates are given as follows: To identify aerodynamic forces (L, D, D p , Ma , M p , Mt , Me ),
     experiments including wind tunnel test are performed with
d1 1 0 0 θX scale model of the UAV. Coefficients of main wing aerody-
d2  = 0 cos θX − sin θX   θY  . (17) namic forces (CL ,CD ,CMt ) are measured in all attack angle
d3 0 sin θX cos θX θZ range (−180 ° ∼ 180 °). Inherent parameters of the propeller
are measured through wind tunnel test. The momentum the-
Step 4 Feedback control for each control surface ory is used for its aerodynamic force calculation. Electrical
Control command is sent to control surfaces based and mechanical time constants of the DC motor are identified
on individual axes as follows: by experiment.

B. Simulation Results
δi = −(KP di + KI di dt + KD d˙i ), (18)
A typical hovering simulation result of quaternion feed-
back is shown in Fig. 7. The initial attitude is (α β γ ) =
where δ1 , δ2 and δ3 are the aileron angle, elevator (0 0 90) ° and the reference attitude is (α β γ ) = (170 0
angle and rudder angle, respectively. d 1 ∼ d3 are cal- 80) ° , where α , β , and γ are ZXY Euler angles. The error
culated by (17). PID gains are same as the quaternion angle around Z axis decreased rapidly. However, note that
PID feedback gains. the error angle around Y axis increased in the early stage

57
 150 150
120 120
90 90

ZXY Euler angle [°]

ZXY Euler angle [°]

60 60
30 30
0 0
−30 −30 Z result
−60 −60 Z reference
−90 Z result −90 Y result
−120 Z reference −120 Y reference
Y result X result
−150 −150
Y reference X reference
−180 −180
0 1 2 3 4 0 1 2 3 4
Time [s] Time [s]
Fig. 7. Quaternion feedback control simulation. Fig. 12. Quaternion feedback control experiment.
150 150
120 120
90 90

ZXY Euler angle [°]

ZXY Euler angle [°]

60 60
30 30
0 0
−30 −30 Z result
−60 −60 Z reference
−90 Z result −90 Y result
Z reference Y reference
−120 −120
Y result X result
−150 −150 X reference
Y reference
−180 −180
0 1 2 3 4 0 1 2 3 4
Time [s] Time [s]
Fig. 8. Resolved tilt-twist angle control simulation. Fig. 13. Resolved tilt-twist angle control experiment.
8
Resolved tilt-twist angle feedback control
7 Quaternion feedback control V. EXPERIMENTAL RESULTS
6
A. Hovering with Quaternion Feedback Control
5
Time [s]

4
Fig. 10 shows snapshots of one of the hovering experi-
ments with quaternion feedback control. In this experiment,
3
the reference and initial attitudes are about the same as
2
simulation. The result of the experiment is shown in Fig. 12.
1
In the beginning of experiment, the error angle around Y
0
0 20 40 60 80 100 120 140 160 180 axis increased and the UAV lost stability. This result is the
Yaw [°] same as the computer simulation. Moreover, the error angle
Fig. 9. Simulated recovery time comparison resolved tilt-twist angle control
with quaternion feedback control.
around Y axis increase and the error angle around Z axis
decrease caused the error angle around X axis. As a result,
the UAV couldn’t continue hovering.
of simulation. This error increase causes a long horizontal It is notable that quaternion feedback control works well
movement. when errors are not very large. However, in some cases
Fig. 8 is the result of simulation on a resolved tilt-twist as shown in Fig. 10, quaternion feedback control causes
angle control hovering. Same conditions are given in both problem.
simulations. The deceleration in error angle around Z axis is
slightly slower than quaternion feedback. Nevertheless, the B. Hovering with Resolved Tilt-Twist Angle Feedback Control
error angle around Y axis deceleration is very fast. As a Fig. 11 shows snapshots of one of the hovering experi-
result, with short horizontal movement, stable hovering is ments with resolved tilt-twist angle control. The experiment
realized. conditions are largely similar to the quaternion feedback
Fig. 9 shows a comparison of recovery times of both the expriment. The result of the experiment is shown in Fig. 13.
strategies. In quaternion feedback control, when the error The angles around Y and X kept reference values, respec-
angle around Z axis surpasses approx 70 °, the recovering tively. The error angle around Z axis decreases smoothly.
time increased exponentially. On the other hand, in resolved This arises from independent calculation steps for tilt and
tilt-twist angle control, the rate of increase of recovering twist angles in resolved tilt-twist angle control.
time is linear. Therefore, resolved tilt-twist angle control has Additionally, even when a human inflicted large distur-
superior stability against the large error angle around Z axis. bance during hovering the UAV continued stable flight and
Furthermore, the error angle around Y axis was converged errors were converged (Figs. 14 and 15). This robustness will
very quickly in resolved tilt-twist angle feedback in all be effective in order to overcome dynamic disturbance like
error angle ranges around Z axis. However, in quaternion a bird strike during hovering. These flights are experimented
feedback, the larger the error angle around Z axis exists, the indoors, but the strategies brought out same performance in
longer the error angle around Y axis converge time is needed. the open air.

58
 Fig. 10. Quaternion feedback control experiment. The UAV couldn’t continue hovering.

Fig. 11. Resolved tilt-twist angle control experiment. Since large aileron angle caused drag force, the UAV lost altitude slightly. However, the UAV could
continue hovering stably.

R EFERENCES
[1] T. Cord, Air Force Research Laboratory, Wright-Patterson AFB:
“SkyTote Advanced Cargo Delivery System,” AIAA International Air
and Space Symposium and Exposition: The Next 100 Years, AIAA-
2003-2753, 2003.
[2] C. Schaefer, L. Baskett: “GOLDENEYE: The Clandestine UAV,” 2nd
AIAA Unmanned Unlimited Systems, Technologies, and Operations,
Fig. 14. Human inflicted rotational disturbance while hovering with AIAA-2003-6634, 2003.
resolved tilt-twist angle control, but the UAV continued hovering stably. [3] H. Stone and G. Clarke: “Optimization of Transition Maneuvers for
a Tail-Sitter Unmanned Air Vehicle (UAV),” Australian International
Aerospace Congress, pp. 105, 2001.
[4] H. Stone: “Control Architecture for a Tail-Sitter Unmanned Air
Vehicle,” Proceedings of the 5th Asian Control Conference, vol. 2,
pp. 736-744, 2004.
[5] D. Kubo and S. Suzuki: “Tail-Sitter Vertical Takeoff and Landing
Unmanned Aerial Vehicle: Transitional Flight Analysis,” Journal of
Aircraft, vol. 45, no. 1, pp. 292-297, 2008.
[6] W. E. Green and P. Y. Oh: “Autonomous Hovering of a Fixed-Wing
Fig. 15. Human inflicted translational disturbance while hovering with Micro Air Vehicle,” IEEE International Conference of Robotics and
resolved tilt-twist angle control, but the UAV continued hovering stably. Automation, Orlando, FL, pp. 2164-2169, May 2006.
[7] W. E. Green and P. Y. Oh: “Optic-Flow-Based Collision Avoidance -
Applications Using a Hybrid MAV,” IEEE Robotics and Automation
Magazine, vol. 15, no. 1, pp. 96-103, 2008.
VI. CONCLUSIONS
[8] A. Frank, J. S. McGrewy, M. Valentiz, D. Levinex and J. P.
In this paper, we presented a novel hovering control How:“Hover, Transition, and Level Flight Control Design for a Single-
Propeller Indoor Airplane,” AIAA Guidance, Navigation and Control
strategy and applied it to PID controller to realize robust Conference and Exhibit, AIAA-2007-6318, 2007.
UAV hovering. The hovering control strategy is based on [9] N. B. Knoebel, S. R. Osborne, D. O. Snyder, T. W. McLain, R.
the analogy of an inverted pendulum model and composed W. Beard and A. M. Eldredge: “Preliminary Modeling, Control, and
Trajectory Design for Miniature Autonomous Tailsitters,” Proceedings
of four steps. The two-dimensional UAV simulator was de- of the AIAA Guidance, Navigation, and Control Conference and
veloped to evaluate the strategy. The resolved tilt-twist angle Exhibit, 2006.
control achieves superior stability to quaternion feedback [10] E. N. Johnson, M. A. Turbe, A. D. Wu, S. K. Kannan and J. C.
control when aircraft has large error angle around Z axis Neidhoefer: “Flight Test Results of Autonomous Fixed-Wing UAV
Transitions to and from Stationary Hover,” Proceedings of the AIAA
through simulation and experiment. Guidance, Navigation, and Control Conference Exhibit, 2006.
The application of the resolved tilt-twist angle feedback [11] E. N. Johnson, A. Wu, J. C. Neidhoefer, S. K. Kannan and M.
control for UAVs is not limited in hovering motion. It doesn’t A. Turbe: “Flight-Test Results of Autonomous Airplane Transitions
Between Steady-Level and Hovering Flight,” Journal of Guidance,
depend on any aircraft current and reference attitude. We be- Control, and Dynamics, vol. 31, no. 2, pp. 358-370, 2008.
lieve it can be extended for many kind of aircraft maneuvers [12] K. Kita, A. Konno and M. Uchiyama:“Hovering Control of a Tail-
which dynamically shift attitude with stall condition, not just Sitter VTOL Aerial Robot,” Journal of Robotics and Mechatronics,
normal motion such as level flight. Vol.21, No.2 pp. 277-283, 2009.
[13] B. Wie, H. Weiss and A. Arapostathis: “Quaternion Feedback Regula-
tor for Spacecraft Eigenaxis Rotations,” Journal of Guidance, Control
VII. ACKNOWLEDGMENTS and Dynamics, vol. 12, no. 3, pp. 375-380, 1989.
This work was supported by Grant-in-Aid for Exploratory
Research (No. 21656219), and Grant-in-Aid for JSPS Fel-
lows (21-6015).

59