See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/239414531

Flight Testing of the T-Wing Tail-Sitter Unmanned Air Vehicle

Article in Journal of Aircraft · March 2008

DOI: 10.2514/1.32750

CITATIONS READS

111 1,894

6 authors, including:

Peter W. Gibbens
The University of Sydney
33 PUBLICATIONS 667 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

A Virtual Reality Learning Laboratory for Diverse Environments View project

All content following this page was uploaded by Peter W. Gibbens on 19 January 2015.

The user has requested enhancement of the downloaded file.

 JOURNAL OF AIRCRAFT
Vol. 45, No. 2, March–April 2008

Flight Testing of the T-Wing Tail-Sitter Unmanned Air Vehicle

R. Hugh Stone,∗ Peter Anderson,† Colin Hutchison,‡ Allen Tsai,† Peter Gibbens,§ and K. C. Wong§
University of Sydney, Sydney, New South Wales 2006, Australia
DOI: 10.2514/1.32750
Since October 2005, the T-wing tail-sitter unmanned air vehicle has undergone an extensive program of flight tests,
resulting in a total of more than 50 flights, many under autonomous control from takeoff to landing. Starting in
August 2006, free flights with conversion between vertical and horizontal flight modes have also been undertaken.
Although the latter flights have required some guidance-level ground-pilot input, significant portions of them were
performed in autonomous mode, including the transitions between horizontal and vertical flight. This paper
considers the overall control architecture of the vehicle, including the different control modes that the vehicle was
flown under during the recent series of tests. Although the individual controllers for each flight mode are
unremarkable in themselves, it is notable that the aggregate system allows the vehicle to fly throughout its entire flight
envelope, which is considerably broader than that of conventional fixed- or rotary-wing vehicles. The performance of
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

the controllers for the different flight modes will also be considered, with a particular focus on hover dispersion
results, in differing wind conditions. The majority of these flights were performed on a tether test rig during
autonomous control development, to ensure vehicle safety with minimal impact on vehicle dynamics. The
demonstration of autonomous flight under the constraints imposed by the tether system in winds up to 18 kt is a
significant achievement. Results from the more recent horizontal flight tests with conversions between vertical and
horizontal flight are also presented. Most important, these results confirm the basic feasibility of tail-sitter vehicles
that use control surfaces submerged in propeller wash for vertical flight control.

Nomenclature Subscripts
A, B = state-space system matrices
comm = commanded
Fx , Fy , Fz = body-axis force components
0 = value taken at reentry to controller
g = acceleration due to gravity
K = gain vector
KQ , KW , = control system gains I. Introduction
K
V ,KIW
L, M, N
nx , ny , nz
=
=
moments about body axes
unit vector components
T HE T-wing is a tail-sitter unmanned air vehicle (UAV) that aims
to blend the vertical takeoff and landing (VTOL) characteristics
and hence operational flexibility of the helicopter with the forward-
P, Q, R = body-axis rates flight efficiencies of a conventional aircraft in the simplest form
pn , pe , h = north, east and height positions possible. Consequently, it uses propeller wash over its wing and fin
Q, R = linear quadratic regular state and control control surfaces to effect control during vertical flight, rather than
weighting matrices cyclic blade pitch variation, as in a helicopter. Furthermore, instead
q0 , q1 , q2 , q3 = quaternion components of rotating engines or wings to transition between horizontal and
U, V, W = body-axis velocity components vertical flight, it instead rotates the complete airframe. This makes
x, u = state and control vectors the vehicle mechanically simpler than other convertiplane designs
x, y, z = standard body-axis coordinates, centered at the such as the tilt wing or tilt rotor, while still retaining the ability to fly
c.g. both vertically and horizontally. A picture of the basic T-wing
e = elevator angle (average angle of wing elevons) configuration is shown in Fig. 1, and a typical flight path is given in
eW = trim elevator angle per unit of W-translational Fig. 2. The concept-demonstrator version of the T-wing is a 65-lb
speed vehicle, with a wingspan of 7.14 ft, powered by twin 100-cm3 twin-

VW = trim vertical tilt angle per unit of W- cylinder two-stroke engines, driving 25-in. counter-rotating fixed-
translational speed pitch propellers. Other pertinent vehicle parameters are given in
,
, = standard Euler angles Table 1.
V ,
V , V = vertical Euler angles The concept-demonstrator version of the T-wing was designed

= total rotation angle specifically as a research platform to prove the feasibility of
controlling such a UAV through its complete flight envelope.
Received 11 June 2007; revision received 24 September 2007; accepted for Consequently, it only carries 0.528 gal (2 liters) of fuel and its
publication 24 September 2007. Copyright © 2007 by University of Sydney. avionics system [inertial measurement unit (IMU), GPS, flight
Published by the American Institute of Aeronautics and Astronautics, Inc., computer, and batteries] accounts for 20% of its maximum takeoff
with permission. Copies of this paper may be made for personal or internal weight (MTOW). These limitations make the current vehicle
use, on condition that the copier pay the $10.00 per-copy fee to the Copyright unsuitable for real-world applications. Notwithstanding this, other
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923; include work has been done by Stone [1–3] on developing multidisciplinary
the code 0021-8669/08 $10.00 in correspondence with the CCC. configuration optimization software for the preliminary sizing of T-
∗
Lecturer, School of Aerospace, Mechanical and Mechatronic Engineer- wing-type vehicles up to 1000-lb MTOW, given mission
ing, Building J07. Senior Member AIAA.
† requirements in terms of payload, range, endurance, and hover
Graduate Student, School of Aerospace, Mechanical and Mechatronic
Engineering, Building J07.
control performance. Thus, although the current vehicle is only
‡
Research Assistant, School of Aerospace, Mechanical and Mechatronic useful as a research platform, [1–3] suggest that the concept has
Engineering, Building J07. practical merit.
§
Senior Lecturer, School of Aerospace, Mechanical and Mechatronic In configuration, the T-wing is most similar to the Boeing
Engineering, Building J07. Heliwing UAV of the early 1990s [4,5], however, it differs
673
 674 STONE ET AL.

Table 1 Specifications for T-wing concept demonstrator

Property Value (imperial) Value (SI)
Wing span 7.142 ft 2.177 m
Wing chord 1.103 ft 0.336 m
Wing leading edge station 3.040 ft 0.927 m
(from nose)
Canard span 2.458 ft 0.749 m
Canard chord 0.472 ft 0.144 m
Propeller y position 1.993 ft 0.607 m
Propeller diameter 2.083 ft 0.635 m
Height (standing on ground) 5.000 ft 1.524 m
Engine power (per engine) 7.64 hp 5.70 kW
Fig. 1 Generic T-wing configuration: vehicle sits vertically on its fins Takeoff weight (nominal) 65.00 lb 29.48 kg
when taking off and landing; concept-demonstrator fins are not canted. Total installed thrust (SSL) 93 lb 42 kg
Specified maximum hover position 2.00 ft 0.61 m
excursion (10-kt sharp-edged gust)

wing, though with one central propeller rather than two. Based on
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

some simple aerodynamic modeling, they describe a feedback-

linearization control structure that allows their vehicle to fly
simulated transition maneuvers as well as hover-flight translations.
The most appealing feature of their approach is that attitude control is
performed directly in terms of quaternions, which obviates the need
for different attitude representations for vertical and horizontal flight,
as is done in the present work. Although quaternions and related
representations have been used in attitude control for a variety of
other aerospace vehicles, including helicopters [13], ducted-fan
UAVs [11], spacecraft [14–16], and even the space shuttle [17] back
in 1976, their use is particularly appropriate for a tail-sitter UAV that
must transition between horizontal and vertical flight. Kubo and
Suzuki [18] also present simulation work on developing robust gain-
scheduled linear quadratic regulator (LQR) flight controllers for a
Fig. 2 Composite picture of T-wing UAV; multiple vehicle pictures in tail-sitter vehicle. Although the control system design of the T-wing
some frames are from successive images from a still camera. is less sophisticated than the preceding approaches mentioned, flight
tests and simulation have shown it to work satisfactorily.
The design of autonomous flight controllers for the T-wing vehicle
fundamentally from that design in its mode of hover control.¶ is complicated by a number of factors. Foremost of these is that the
Whereas the Heliwing used conventional cyclic rotor controls, the T- vehicle ranges over a greater variety of flight conditions than a
wing uses control surfaces submerged in the slipstream of its conventional aircraft: from high-speed forward flight through to
propellers, similar to the systems used on the manned tail-sitters of vertical attitude descent. This means that the aerodynamics
the 1950s: the Convair XF-Y1 [6] and Lockheed XF-V1 [7]. These associated with this vehicle are both different from and more severely
vehicles suffered from hover-mode attitude instability and were nonlinear than those associated with conventional vehicles. This has
difficult to pilot, particularly when landing. By transitioning this necessitated the decomposition of the autonomous flight control
original tail-sitter concept to that of an unmanned vehicle and taking problem into four major modes dealing with vertical flight,
advantage of modern electronics to automate the control system, the horizontal flight, and the two transition flight regimes: vertical to
T-wing aims to obviate these problems and make this original horizontal (V2H) and horizontal to vertical (H2V). Further flight
concept a success. control modes are provided to allow different levels of manual
The 1950s and 1960s saw considerable interest in convertiplane control in the event of problems with the autonomous system, as well
vehicles such as tail-sitters. At that time, much of the research was as to allow for different reversionary modes in case of sensor
focused on the aerodynamics and basic feasibility of concepts such as malfunction. The most challenging flight mode from a control point
the tail-sitter, tilt-rotor, tilt-wing, and ducted-fan types of vehicles. of view is the vertical flight mode. This is when the vehicle is most
Details of (and references to) much of this work can be found in unstable and when the aerodynamics and dynamics are least similar
McCormick [8]. With growing interest in UAVs, many of these to a conventional aircraft. Attempts to fly the vehicle manually in this
VTOL concepts have been revived in unmanned form. Apart from mode have not been successful. Furthermore, because this is the
standard helicopter UAVs, the most common VTOL concept appears takeoff and landing mode of the vehicle, it is the most fundamental.
to be the ducted fan, starting with the Sikorsky Cypher UAV [9] of Because of these considerations, most testing of the T-wing to date
the early 1990s. Much of the modern research interest in these has been done in vertical flight using a tether test rig to protect the
vehicles has centered on flight control. Avanzini et al. [10] describe a vehicle during control system development. The use of this rig has
robust control structure for a shrouded-fan vehicle, and Johnson and been vital in allowing the T-wing flight test program to proceed
Turbe [11] describe a dynamic-inversion controller with neural- without damaging the airframe. Furthermore, because the vehicle is
network adaptation for a smaller ducted-fan vehicle. The latter flown with two tethers and is thus always attached to the ground by
reference also describes flight tests of the vehicle, including an air (slack) ropes,∗∗ flight tests can be carried out at smaller, less remote,
deployment by another UAV, an R-MAX helicopter. test sites than are required for free flights. During tethered testing of
Less work appears to have been done on flight control or flight the vehicle, the focus has been on precise hover control with and
testing of tail-sitter vehicles such as the T-wing. Knoebel et al. [12] without wind, because this is crucial for accurate landing of the
describe the modeling of a vehicle substantially similar to the T- vehicle. So far, the vehicle has flown in crosswinds gusting up to
18 kt (30 ft=s).
¶
The T-wing also differs from the Heliwing in the use of a canard. This
∗∗
surface is currently fixed, though the potential exists to use it for forward- The tether ropes weigh 0:108 oz=ft (10 g=m) and are 0.158 in. (4 mm) in
flight control or trim. diameter.
 STONE ET AL. 675

In addition to the tethered testing of the vehicle, the vehicle has

also been flown freely, during which it has performed autonomous
takeoffs, V2H and H2V transitions, and horizontal flight. The V2H
transitions were accomplished with no significant problems,
although some control improvements are still required on the
reverse-H2V maneuvers, during which the vehicle experiences
angles of attack in excess of 60 deg. Although the vehicle is yet to
complete a transition flight with a full autonomous landing, the
results to date indicate that this is achievable.
The contribution of the present work is fourfold. First, the flight
test results demonstrate the basic feasibility of this type of tail-sitter
vehicle in vertical, horizontal, and transition flight modes. Second,
these tests and simulation results indicate that a relatively simple
control approach can be made to work effectively. Third, the results
demonstrate that the aerodynamic modeling of the vehicle [19] is at
least sufficiently accurate for control design purposes, without the
need for adaptive elements (though these would undoubtedly
improve the vehicle’s performance). Finally, some readers may be
interested in particular details of the tests carried out and the way in Fig. 3 Z body-axis force plotted versus velocity (fps) and angle of attack
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

which particular problems were solved: for instance, the balancing of (deg) at throttle settings of 0.4, 0.6, 0.8, and 1.0.
engine thrust from side to side, the handling of switching between
different control modes, and the like.
This paper is set out as follows. Section II discusses the basic forces dominate) to high-speed horizontal flight (freestream lift and
equations of motion of the vehicle, axis systems, and vehicle drag forces dominate).
aerodynamics. Section III outlines the structure of the complete T-
wing flight control system, and Sec. IV discusses the physical flight
control sensors and hardware. Section V presents simulation results
for later comparison with flight test data. Vertical flight test results A. Attitude Representations
are presented in Sec. VI, followed by preliminary horizontal and Three distinct attitude representations are used for simulation and
transition test results in Sec. VII. Finally, Sec. VIII concludes the flight control of the T-wing vehicle. A quaternion representation is
paper with discussion of results to date and an outline of future work. used exclusively in the flight simulation of the vehicle, as well as in
the guidance controller for the transition maneuvers between
horizontal and vertical flight. The quaternion representation has the
II. T-Wing Model and Aerodynamics advantage of being unique (to within a choice of sign) and of having
The T-wing is modeled using the standard 6-DOF nonlinear rigid- no areas of degeneracy of its solution. For horizontal flight, the
body equations of motion as outlined by Stevens and Lewis [19], standard Euler-angle rotations of yaw , pitch
, and roll  are used
whose notation will be adopted here. Because of the fact that the to describe the vehicle’s attitude with respect to the north-east-down
propellers are counter-rotating, gyroscopic terms are omitted from (NED) frame. The Euler-angle representation is degenerate at
the moment equations. Although the use of differential thrust (via
 90 deg, which is problematic for the T-wing because it spends
differential engine rpm) may introduce minor gyroscopic effects, significant amounts of time in a near-vertical attitude.
these will be small and are ignored. The aerodynamic forces and Because of the degeneracy of the standard Euler angles for vertical
moments required for these equations are supplied through a flight, it is useful to define a second set of nonstandard Euler angles
database that covers more than 4000 flight conditions, ranging over for vertical mode flight as follows:
different flow angles  and , speeds V, and throttle settings T . The 1) Start from a vertical attitude with the vehicle belly (z axis)
database is constructed using an integrated propeller-blade element facing north.
and panel-method model. This model uses corrections based on 2) The vehicle is “rolled” about its longitudinal x axis by a vertical
measured 2-D viscous data for those regions of the vehicle flowfield roll angle v (opposite sense to ).
well outside the bounds of normal linear aerodynamic analysis 3) The vehicle is then pitched about its once-moved lateral y axis
[1,20,21]. This allows the vehicle to be modeled for angles of attack by a vertical pitch angle
v .
and sideslip between 90 deg and for velocities ranging from low- 4) The vehicle is now yawed about its twice-moved belly z axis by
speed vertical descent to high-speed forward flight.†† a vertical yaw angle v to complete the attitude of the vehicle.
The nonlinearities associated with the aerodynamic forces and Unlike the standard Euler angles, this set is not degenerate when
moments are more extensive than those typically associated with the vehicle is at or near vertical, when the standard pitch angle is close
conventional aircraft. This is because the range of flow speeds and to 90 deg. Furthermore, the sense of all the angles remains consistent
angles is greater. A partial appreciation of the force and moment between horizontal and vertical representations: roll rate is always
nonlinearities can be gained from a plot of the Z body-axis force, as about the body x axis, and yaw rate always about the body z axis. If a
shown in Fig. 3, taken from the aerodynamic database for the vehicle helicopter attitude convention had been used for vertical mode flight,
[21]. In this graph, the four stacked surfaces represent Z force values then the sense of roll and yaw would have varied between the
at different throttle settings plotted against angle of attack and different flight modes. It should be noted that although useful for
velocity. Stall is clearly seen occurring at   20 deg, after which vertical flight control, this vertical system becomes degenerate when
the four different throttle surfaces separate, due to nonlinear effects. the vertical pitch angle reaches 90 deg (horizontal).
Unlike conventional aircraft in which the nonlinearity in the force Transformations between these three representations can be easily
and moment data are largely confined to stall and the parabolic calculated [1]. Similarly, it is also possible to express the kinematic
variation of the force terms with speed, the nonlinearities for the T- equations of motion in terms of either the vertical or horizontal Euler
wing are more complicated, due to the changing relative importance angles. This is often appropriate when performing control design,
of the propeller-generated forces in comparison with those due to the because it is more difficult to relate the quaternion representation to
freestream dynamic pressure. This change occurs as the vehicle goes physically meaningful control objectives. An alternative method to
from low-speed vertical flight (propeller and propeller slipstream using two distinct attitude systems for a tail-sitter vehicle has been
proposed by Knoebel et al. [12]. They develop guidance and control
††
The use of negative velocities obviates the need for considering flow strategies that cover all flight regimes based solely on a quaternion
angles outside the range of 90 to +90 deg for descending flight conditions. attitude description, which unifies the control problem. Similar
 676 STONE ET AL.

vertical flight, the vehicle is clearly flyable in this condition. The

shading in Fig. 4 represents the region of greatest uncertainty in the
vehicle trim properties. This occurs in the transition region between
40 and 90 ft=s, in which the slipstream regions may be significantly
affected by stall and in which the poststall loads from nonslipstream
regions become significant. Consequently, there is no intention to try
to maintain trimmed flight at these speeds. As mentioned previously,
the T-wing is significantly unstable in vertical flight and marginally
stable (13% static margin) in forward flight, with a midrange c.g.
position.

Fig. 4 Predicted trim pitch angle, elevator setting and throttle versus
III. T-Wing Flight Control System
speed for nonclimbing flight; results given for three c.g. locations The flight control system for the T-wing consists of a relatively
(measured from the nose) with 40% thrust margin and for mid-c.g. at complex amalgam of simple control systems that are used for
20% thrust margin. different autonomous and semi-autonomous flight modes. The fully
autonomous modes span the basic operating conditions of the
vehicle: vertical, horizontal, and the two transition modes. Further
approaches for the T-wing have not been used to date (except for autonomous modes are used to provide some degree of control
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

transition-mode guidance). system fail-safe behavior in the event of particular types of sensor
As mentioned in the Introduction, quaternions and other attitude malfunction (such as loss of GPS signal). Another set of semi-
representations have been used for a variety of other aerospace autonomous/semimanual modes is provided to allow different levels
vehicles, including helicopters [13], generic spacecraft [14–16], and of ground-pilot intervention during initial flight testing. The major
the space shuttle [17]. Schuster [22] provides a comprehensive flight control modes and their principal characteristics are outlined in
summary of different representations in use across a variety of fields, Table 2. The different modes are either selectable from the pilot-
and Phillips et al. [23] summarize representations in use for aircraft. control station or are selected automatically in the case of
Although the vertical Euler-angle representation used here for hover- autonomous vehicle transition or failure modes. Because most of
mode flight control lacks the appeal of the unified representation these modes use relatively simple gain-scheduled classical
afforded by quaternions, it has been found useful in designing the controllers designed using frequency-based techniques [19,25,26],
vertical flight control system for the vehicle and also provides a no further details will be given as to their structure. In the case of
convenient and meaningful way to describe the attitude of the vehicle vertical hover flight, however, a brief description of the control
in vertical flight. structure will be provided. Although these controllers use standard
LQR designs [19] made with subspace models of the vehicle
dynamics, their description is warranted, due to the relative novelty
B. Qualitative Description of Vehicle Characteristics
of the vehicle and the fact that this structure is important when
As indicated in Table 1, the nominal thrust margin for the T-wing considering reentering this mode after performing a H2V transition.
vehicle is 40% at standard sea-level (SSL) conditions. Because of
variations in density altitude and takeoff weight, some flight tests
have taken place with thrust margins less than 20%. Figure 4 shows A. Autonomous Vertical Velocity Mode
longitudinal trim characteristics for the vehicle, as predicted using During autonomous vertical flight the T-wing uses a set of gain-
the aerodynamic database [21] for three separate c.g. positions scheduled LQR translational velocity controllers to control velocities
(nominal and one on either side). It also shows the same quantities for in the body-axis z and y directions using the elevators and rudders,
a reduced thrust margin of 20%. (Assuming thrust is proportional to respectively. These are combined with a classical proportional plus
density for a normally aspirated piston engine [24], a 20% thrust integral (PI) vertical roll-angle controller (for heading control or, in
margin represents the nominal vehicle operating at a density altitude other words, in which direction the belly is pointing) and a similar PI
of 5000 ft.) The kink in the trim elevator angle observed after 12 ft/ vertical velocity throttle controller. The vertical flight mode is when
s is associated with stall onset in one lobe of the propeller slipstream the vehicle is most unstable and also the mode of operation that is
velocity distribution while in (near) vertical flight. Figure 4 shows most novel in comparison with that of a conventional aircraft.
that it is increasingly difficult for the vehicle to be trimmed vertically The following section will briefly consider the design of a vertical
in strong winds as the c.g. is moved forward. The results at reduced attitude translational W-velocity controller for the vehicle. The
thrust margin show that despite the higher throttle setting required for design of a sideways V-velocity controller is similar. In doing so use

Table 2 Major vehicle control modes

Control mode Guidance Low-level control
Autonomous vertical velocity mode PID waypoint navigation emitting 1) LQR translational velocity controllers
1) Translational velocity commands 2) PI pointing angle and vertical
2) Belly pointing angle velocity controllers
3) Climb rate
Autonomous V2H and H2V transition modes Quaternion-based attitude guidance emitting 1) Classical P, Q, and R angular-rate
P, Q, and R pitch-rate commands controllers
2) Throttle scheduled for transitions
Autonomous horizontal flight mode Bank-to-turn proportional navigation in two stages: 1) Classical P, Q, and R angular-rate
1) First stage emits climb-rate, bank angle, controllers
speed commands; 2) Classical speed control via throttle
2) Second stage converts these to P, Q, and R
and speed commands
Autonomous vertical angle mode (for autoland 1) Tilt-angle commands either provided by ground 1) Classical PI tilt-angle controllers
when there is a sustained GPS failure and pilot or set to zero (for autoland) 2) PI pointing angle and vertical velocity
horizontal velocities are unavailable) 2) Vertical speed set by ground pilot or set to 4 ft=s controllers
Full manual mode (for ground checks only) Surfaces and throttles directly controlled by the pilot None
 STONE ET AL. 677

V
Q
_ -KQ
Fx θ vW + Kθ v θv

x, U δe ++ δe
Fz W + Vehicle
++
M,Q
W
z, W 1 +_ KW
1/4 Chord Wcomm
1
_ KI W
s
mg
Fig. 6 T-wing W-velocity control system for vertical flight translation.
e

Fig. 5 Vertical flight free-body diagram (all quantities shown positive;

belly on the right). K  2:67  19:0  46:2  KW KQ K
V (5)

Using these gains gives a closed-loop system with a gain margin of

will be made of Fig. 5, which gives a free-body diagram of the vehicle 8:75 dB, a phase margin of 41 deg, and poles of 1:88 and 1:64,
without lateral states. 2:00i. The regulator is converted into a controller by simply
Using a body-axis system and a vertical pitch angle, as shown in regulating the error between a commanded and actual velocity. Such
Fig. 5, for the case of nonclimbing vertical mode flight, the equations a controller will typically not have zero steady-state error. To fix this,
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

of motion can be linearized as shown in Eq. (1) [27]: the trim values for the tilt angle
v and the elevator deflection e are
2 3 2 32 3 2 3 used in a modified control law, which ensures that the steady-state
W_ F zW F zQ g W F ze value of the elevator deflection is exactly what is required for flight at
4 Q_ 5  4 M W M  Q 0 54 Q 5  4 M  e 5e (1) the given velocity. The addition of an integral W state to the
_

V 0 1 0
v 0 controller then provides robustness against modeling errors and
disturbances:
In the preceding equation, the force and moment terms are divided
by mass and pitch inertia, respectively (represented by overscores), "W  Wcomm  W
and standard mechanics of flight notation are used to represent the Z
significant force and moment derivatives. No attempt is made to use e  KW "W  KIW "W dt  KQ Q (6)
nondimensional units, due to the problems associated with the choice
of an appropriate velocity to use in the nondimensionalizing process.  K
V

VW Wcomm 
V   eW Wcomm
From this equation, the trim elevator deflection and the trim vertical
pitch angle per unit of translational velocity W can be easily In actual flight tests, controllers with and without the integral term
determined [2,28], as shown in Eq. (2). (Note that in these equations, were used. A block diagram of this controller is given in Fig. 6. The
we use the full force and moment derivatives. This is why “mass” integral term in Eq. (6) may cause problems because constant (or
appears in the denominator of the
VW equation.) average) wind disturbances occur in the fixed NED frame, whereas

 the integral error (as written) is stored in the body-axis frame. Thus,
Mw FZ FZW Mw during a pirouette, an average wind-disturbance error related to a
eW 
VW  e  (2)
Me mg FZe Me particular orientation is carried into another orientation. This may
cause a significant transient while the integral state adapts to the
This reduced set of equations can also be used to develop LQR disturbances in the new orientation.
controllers for the vehicle. The equations given in Eq. (1) are in the The gains used in this controller are scheduled with vertical flight
standard state-space form. speed and throttle setting. A better scheduling parameter would be
the prop-wash dynamic pressure. Unfortunately, this is hard to
x_  Ax  Bu (3) measure explicitly and must be inferred from vertical speed, throttle
setting, and (if available) local relative wind. The design of the
During flight, the W-velocity component (ft/s), the pitch rate Q sideways translational V-velocity controller is exactly analogous,
(rad/s), and the vertical pitch angle
v (rad) are all available for except that we include differential thrust using a proportional gain
feedback control of the elevator (deg) via the onboard inertial and tied to rudder deflection. This gain is currently set at 0:006= deg.
GPS sensors. Thus, to develop a linear quadratic regulator, all that we Thus, at full rudder deflection (38 deg) this gain leads to a 24%
require is a state-error weighting matrix Q and the control effort throttle-setting imbalance from side to side. This particular feature of
weighting R. For the current demonstrator vehicle in hover flight in the control system is hard to design accurately, because it requires
zero wind, the plant matrices are as given next (with imperial units, as knowledge of the engine time constants. For this reason, the
indicated before). Typical Q and R weighting matrices have also proportional interconnect between the rudder and differential
been indicated. This system has an unstable pole located at 1.87 rad/s throttles was chosen based on achieving acceptable simulation
and a stable pair at 1:64  1:99i rad=s. results over a range of conditions, with values confirmed via flight
2 3 2 3 test.
0:8830:194  32:2 0:241 Fully autonomous vertical flight involves the UAV navigating
A 4 0:385  0:5250 5; B  4 0:231 5 between a set of points in space that are defined before takeoff. The
01:0000 0 waypoint definition involves specifying the Cartesian coordinates of
2 3 (4) each point as well as the belly-pointing angle of the UAV. By altering
1 0 0 the pointing angles at different waypoints, the UAV can be instructed
Q  4 0 0 0 5; R  10 to translate between points in a flatwise (W velocity) sense, sideways
0 0 0 (V velocity), or in any skewed combination of these. Furthermore, by
specifying waypoints with different pointing angles but the same
This gives rise to the following control gains on the W-velocity, spatial coordinates, the vehicle can be made to perform pirouettes.
pitch-rate, and pitch-angle states‡‡: Guidance in autonomous vertical flight is provided by proportional
plus integral plus derivative (PID) guidance controllers in the local-
‡‡ vertical, local-horizontal frame, to keep the UAV on a track between
The units of these three gains are deg/(ft/s), deg/(ft/s), and deg/rad for the
W, Q, and
v states, respectively. the previous waypoint and the next one. The outputs of these
 678 STONE ET AL.

pN, pE, ϕ
.
.
h,h
h,h
Qcomm
WAYPOINTS hcomm
FLIGHT STATE
Horizontal Flight +
Height Control +

(pN , pE )comm Qcomm Q

pN, pE, ϕ (P, R) comm
Horizontal Flight Q
Proportional Navigation Qcomm _ s 2 +3.3s +2.7 δe .
+ KQ δe h,h
pN, pE, ϕ
attitude Qcomm s2
attitude
(P, R) comm Q_Controller
Vehicle
HORZ target attitude

V2H Guidance
Multiport VERT => (ϕ v = -ψ0 , θv = +10°, ψ v =0)
Switch
attitude Qcomm HORZ => (ϕ = 0, θ = 0, ψ = -ϕ v0 )
(P, R) comm
VERT target attitude

H2V Guidance (q0 ,q1 ,q2 ,q3)

Fig. 7 Pitch-rate control for horizontal flight and transition maneuvers.
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

guidance controllers are limited to help avoid control surface the horizontal and transition flight modes and no need to reset the
saturations in the low-level controllers. integrator states within the angular-rate controllers. The switch from
V2H to horizontal guidance occurs automatically when the pitch
B. Autonomous Transition Flight Modes
angle passes through 15 deg. The switch from horizontal to H2V
guidance occurs automatically when the vehicle passes through a
The V2H and H2V transition flight modes use low-level classical horizontal waypoint and detects that the next waypoint is vertical
angular-rate controllers combined with attitude guidance based on (indicated by a flag value attached to the waypoint).
quaternions. The controller for pitch rate is shown in Fig. 7. The KQ
gain in this controller is scheduled with speed and throttle setting, and
antiwindup limits are placed on the internal controller integrator D. Autonomous Mode Switching
states. This figure also shows the guidance blocks that generate the At the start of the V2H transition and the end of the H2V transition
pitch-rate commands for horizontal flight and the transition it is necessary for the vehicle to change its low-level controllers. Of
maneuvers. On entering a transition maneuver, the error between the these two cases, simulation and experience have shown that the most
current attitude and the desired attitude is expressed as a quaternion. serious mode switch arises at the top of the H2V transition, in which
The normalized vector part of the error quaternion, nx ; ny ; nz T , the vehicle reenters direct translational velocity control. This mode
gives the unit rotation vector required to achieve the attitude change. change is triggered 2 s after the vehicle passes through the vertical on
Using this vector (expressed in body axes) multiplied by the product its way to its target oververtical attitude
v  10 deg. This mode
of the total rotation angle
(also obtained from the quaternion) and a change gives the greatest problems, because at the top of the
guidance gain then gives the appropriate angular-rate commands transition, the vehicle typically still has significant translational
required for the attitude change. This is shown in Eq. (7) and is velocity, and the control settings and states will be a long way from
exactly the procedure outlined by Knoebel [12]. trim values desired by the guidance system. To overcome extreme
2 3 2 control transients at this point, the following procedure is used:
q0 3 2 3 2 3
cos

=2 1) At entry to vertical velocity mode, the current value of the
6 q1 7 6 nx sin

=2 7 Pcomm nx
control surface setting and the relevant states of the vehicle are
q err  6
4 q2 5
74
ny sin

=2 5 4 Qcomm 5  K

4 ny 5
recovered. If present, the integral velocity error state is set to zero.
nz sin

=2 Rcomm nz 2) Eq. (6) is then used to backcalculate the W command that
q3
corresponds to the current states and control surface setting at entry to
(7) the vertical flight mode, given the known gains of the controller, as
shown in Eq. (8):
An alternative simpler method for transition maneuver guidance is
to command a pure pullup (H2V) or pure pushover (V2H) at a fixed KW W0  KQ Q0  K
V
V0  e0
(or scheduled) pitch rate until a target attitude is attained. Although Wcomm0  (8)
this only works if the vehicle enters the maneuvers with its body x–z KW  eW  K
v
vW
plane approximately vertical, this is not a major restriction. This was
the procedure adopted in the horizontal flight test detailed later. For 3) This W command is then ramped down to zero over 5 s to bring
the V2H transition, the throttle is set to 100%, whereas for the reverse the vehicle to a stable hover.
H2V transition, the throttle is scheduled with forward speed to 4) A similar procedure is used for the lateral translational V-
balance conflicting requirements of maintaining adequate airflow velocity command.
over all control surfaces versus limiting the total height gained in the 5) After this, vertical waypoint guidance proceeds as normal.
maneuver. This procedure appears to work well in simulation and also in
flight test. An alternative procedure to achieve bumpless transitions
between flight modes is to initialize a controller integrator state to a
C. Autonomous Horizontal Flight Mode value that corresponds to both the current control surface setting and
The autonomous horizontal flight mode uses the same angular-rate the current states and commands. An example of the use of this
controllers as for the transition maneuvers at the lowest level and a technique for controllers in a highly maneuverable helicopter is
proportional navigation bank-to-turn guidance algorithm [25]. The given by Gavrilets et al. [29]. Such a technique was not applicable
guidance algorithm emits bank-angle, climb-rate, and velocity here, because the vehicle was not always flown with integral states in
commands. The bank-angle and climb-rate commands are further the W and V controllers.
converted to angular-rate commands, which are then sent to the low-
level classical P, Q, and R controllers. This structure is used because
it means that the low-level controllers are common between the IV. Flight Control System and Hardware
horizontal and the V2H and H2V transition modes. For this reason, The vehicle has a simple but accurate suite of onboard sensors to
there are no significant control transients when switching between enable it to estimate its current state. A schematic of this system is
 STONE ET AL. 679
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

Fig. 9 Vehicle in hover-mode flight in wind with ultrasonic wind sensor

on nose.

flight test data. Results will be given for one hover and one forward-
flight condition. To allow useful comparisons, wind measurements
from the flight tests given later were imported into the simulations.
These measurements were obtained at the ground station, 60 ft
from the takeoff and landing point. Their use ensures that the
simulated conditions are comparable with (if not exactly the same as)
Fig. 8 T-wing flight control system. those encountered by the vehicle during test.

shown in Fig. 8. The flight control system consists of a 400-MHz

Celeron flight computer in a PC-104 form factor; a Honeywell A. Vertical Flight Simulation
HG1700AG17 Ring-Laser Gyro IMU; a Novatel ProPak-G2 Plus 5- For all hover tests a standard  flight pattern was used, each leg
Hz GPS receiver, which includes a dedicated Kalman filter to being 8 ft, for a total width of 16 ft. This pattern involves a total of 27
interface with the Honeywell IMU to give filtered position, velocity, waypoints that the vehicle must pass through and exercises as much
and attitude (PVA) estimates; and an analog-to-digital converter to of the vertical flight functionality as possible within the confines of
monitor battery voltages and the static pressure sensor. A 900-MHz the tether test rig (described later). This functionality includes
radio modem is used to communicate with the ground station. All translations, skewed translations, pirouettes, climbs, and descents.
significant state data are transmitted and recorded at 80 Hz. As an The capture tolerance for waypoints is 6 ft in position and 10 deg in
option, the vehicle can be flown with a Gill Instruments Windsonic pointing angle. Although the position tolerance seems large in
ultrasonic wind sensor mounted on the nose of the vehicle. This comparison with the  dimensions, the vehicle dwells at each
measures V and W components of relative wind at the nose at a rate of waypoint for 6 s after capture, during which time it still tracks toward
4 Hz and a resolution of 0.1 m/s. that waypoint. To force the vehicle to land, the last waypoint is
The current flight control system with RTK GPS provides position simply placed 1 ft below ground. The standard test pattern is as
estimates with 2-cm errors and velocities within 2 cm/s. Although the follows:
system occasionally reverts to plain differential mode (with position 1) Climb to 10 ft (3.05 m), with the vehicle belly (z body axis)
accuracies of 1 m), the relatively high-quality IMU means that short- pointing north.
length dropouts can be tolerated before errors grow too large and that 2) Maintain this altitude and orientation and fly a  pattern with 8-
long-term dropouts only lead to slow drifts between true and ft (2.44-m) legs in each direction.
measured positions. The IMU also provides precise angular 3) Return to center and reorient the vehicle with the belly pointing
measurements, with quoted accuracies of 0.05 deg. Although this north–east.
level of accuracy is not required for a general tail-sitter vehicle 4) Maintain this orientation and fly the same  pattern as in step 2
operating freely, it is required for fully autonomous flight within the (but now all translations involve a combination of W and V
limited confines of the tether test rig, because significant state errors commands).
can lead to position biases, which may place the vehicle beyond the 5) Reorient the vehicle with the belly pointing north and perform a
bounds of the test rig. To date, the vehicle has not been equipped with clockwise pirouette (N ! E ! S ! W ! N) about the center
an air data system other than a simple static pressure sensor. Some point, with stops at each compass leg, followed by a similar
hover-mode flights were conducted with an ultrasonic wind sensor anticlockwise pirouette.
mounted on the nose (see Fig. 9); however, the shape of this sensor 6) Climb to 12 ft (3.66 m); descend to 4 ft (1.22 m); then land.
precludes its use for forward flight. Results from performing a simulation using the wind data from the
flight of 18 July 2006 are given in Figs. 10 and 11. The wind inputs
are shown in Fig. 12 and reach speeds of 18 kt (30 ft=s). The
V. Simulation Results simulation position results (Fig. 10) show that in these winds the
Although this paper is focused on the flight testing of the T-wing, a hover precision has a 2 bound of 5.32 ft. Significant position
few simulation results will be presented for comparison with the deviations are observed during the pirouettes. The vertical pitch- and
 680 STONE ET AL.

20
pN, (ft)

500
0

−20
20 0
p , (ft)

0
E

−20
−500

South − North (ft)

20
h, (ft)

10

0 −1000

0
φv, ( °)

−200
−400 −1500
0 50 100 150 200 250 300
Time (s)
Fig. 10 Simulation position states and belly-pointing angle
v using the −2000
same wind as for the 18 July 2006 flight; 
2:66 ft.

−2000 −1500 −1000 −500 0 500 1000

West − East (ft)
30
20 Fig. 13 Simulated horizontal flight pattern with wind as measured
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

during the flight of 30 August 2006 (waypoints are indicated with circles).
θ V (° )

10
0
−10
100
−20
50

θ, ( ° )
40 0
−50
20 Velocity, (ft/s)
ψ V (° )

150
0
100
−20
50
−40 0
0 50 100 150 200 250 300
Time (s)
Altitude, (ft)

600
Fig. 11 Simulation vertical Euler pitch and yaw angles using the same
wind as for the 18 July 2006 flight. 400
200
0
0 50 100 150 200 250 300
Time (s)
30
Wind Speed (ft/s)

Fig. 14 Pitch-angle, velocity, and altitude profiles for simulated

25 horizontal flight pattern.
20

15
100
10
θ, (°)

50
250
0
Wind Dn (°)

200
Velocity, (ft/s)

150

150 100
50
100
950 1000 1050 1100 1150 1200 1250 1300 0
Time (s)
350
Altitude, (ft)

Fig. 12 Wind speed and direction for 18 July 2006; wind between 15
300
and 30 ft/s (9–18 kts), mainly from the south; time base reflects the test
data file: for simulation, the time base was shifted by 950 s. 250
200
133 134 135 136 137 138 139 140
Time (s)
yaw-angle results given in Fig. 11 show the vertical pitch angle Fig. 15 Pitch-angle, velocity, and altitude during transition for
varying in the range of 15 to 25 deg. simulated horizontal flight pattern.

B. Horizontal Flight Simulation transitions can be completed in 300-ft altitude gain without
Simulation results for a complete flight involving both transitions breaking a 12-deg angle-of-attack boundary until a specified exit
between horizontal and vertical flight are presented in Figs. 13–15. translational speed is reached. This more conservative transition was
Wind data were taken from the actual transition flight of not used in the present tests, because it would add 75 s to the
30 August 2006 and varied between 13 and 20 ft=s (7.8 to 11.6 kt). descent time.
Figure 13 shows a plan view of the simulated flight path and Figs. 14
and 15 show details of the pitch attitude, velocity, and height. The
vehicle is seen to climb slightly during the V2H transition (Fig. 14) VI. Vertical Flight Tests
and gains approximately 120 ft during the reverse-H2V transition The analysis of results from the recent T-wing vertical flight test
(Fig. 15). The extra 80-ft climb at the end of the H2V transition is to campaign (from October 2005 to August 2006) will focus on the
enable the vehicle to pass through a waypoint at 400 ft and is not part hover precision of the vehicle in both windy and calm conditions.
of the transition maneuver per se. Other work by Stone [30] on Before describing test results, some aspects of the test procedure will
transition maneuver optimization has shown that unstalled H2V be detailed.
 STONE ET AL. 681

A. Tether Test Rig

The tether test rig used for the vertical flight tests consists of a 66-ft
(20-m) aluminium pole erected at a 45-deg angle to give a working
height for flight tests of approximately 46 ft (14 m). Figure 16 shows
the basic rig, and Fig. 17 presents a picture from an actual flight test.
The usable envelope for tether tests is significantly less than 46 ft
to ensure that no part of the vehicle can contact the tether pole and to
allow attachment of a weight on the tether rope above the vehicle.
This ensures that when descending, the vehicle does not directly pull Fig. 18 Tilt calibration of engines (fins are not canted on the concept
rope through the top pulley. A bottom tether rope is also used demonstrator).
(attached to the ground) to limit the maximum height that the vehicle
can fly. This is given a length of 16 ft (5 m). Flying at 10 ft (3 m)
above the ground gives the vehicle a 13-ft (4-m) sideways envelope observed between them. This is most likely due to slight differences
to move in. in the torsional stiffness characteristics of the wood, resulting in
differing amounts of twist under load.]
4) Wind variations can lead to takeoff throttle differences in excess
B. Tilt Calibration of Engines of 10% from day to day.
To overcome these problems, the T-wing currently performs an
The T-wing currently uses two Desert Aircraft DA-100 engines
autonomous throttle tilt calibration before all flights. This procedure
driving 25 10 in: Garvin wooden fixed-pitch propellers. During
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

is invoked after engine start. First, the port engine autonomously

initial autonomous flight control tests, one problem that soon became
ramps up at a rate of 3%/s while the starboard engine is held at idle
apparent was engine thrust variability. Variations were observed
(20%). When the IMU detects a tilt angle of 3 deg, the throttle setting
both in total installed thrust and between the port and starboard
is recorded and then the engine is reduced to idle. A similar procedure
engines on a day-to-day basis. Some of the reasons for these
is then performed on the starboard engine. Because the landing gear
variations are as follows:
are directly below the engines, the throttle settings so recorded are the
1) There were different atmospheric conditions between test days.
hover thrust settings required for flight on that day. Although this
2) There were slight differences in the tuning of the two-stroke
procedure ignores ground-effect contributions, these are known to be
engines.
small. A picture of the procedure is given in Fig. 18.
3) There were variations in the propellers. [Although the contra-
A further use for the tilt-calibration procedure is the determination
rotating propellers are manufactured by a numerically controlled
of the total excess thrust available to the vehicle. This can be done by
(NC) milling machine, significant performance variations are still
simply repeating the tilt-thrust calibration while steadily adding
weights to each nacelle. When a given engine is no longer able to lift
its side with the throttle at 100%, the maximum thrust point for that
side has been exceeded.
The tilt-calibration procedure was of significant use in
determining feedforward-trim hover throttle settings for balanced
flight. Errors in the calibration are covered by the rudder-to-
differential throttle interconnect described earlier (which also
provides extra yaw-control authority). One deficiency of the tilt-
calibration procedure occurs when operating on a windy day with the
wind blowing along the axis of the wings (y body axis). In this case,
the windward engine may “see” a higher wind than the downwind
engine, which may be partially blanked by the fuselage. Because any
crosswind helps increase the engine thrust at a given throttle setting,
the tilt-calibration may be biased toward the windward engine. Once
the vehicle takes off and alters its orientation, this bias in
Fig. 16 Tether test-rig setup. feedforward-trim differential throttle must be overcome by the
control system.

C. Hover Tests: Dispersion in Moderate Winds

Figures 19–21 show the hover performance of the vehicle in
performing the standard  pattern maneuver. The plan-view plot
(Fig. 19) shows the vehicle track over a full autonomous flight
conducted on 13 July 2006, including pirouettes, and shows no sign
of biases with a 2 hover precision of less than 2.5 ft. This flight was
conducted in winds of 6–8 kt. Figure 20 shows the actual position-
state traces for this flight. The position doublets are clear and show no
significant difference between skewed and nonskewed translations.
The winds during this test as measured at the ground station are
given in Fig. 21. The only slight problem with this particular test is
that the vehicle became stuck on one of the reverse pirouette
waypoints for approximately 170 s. This was due to the integral state
in the pointing-angle controller being inactive for this particular
flight. Simulation results (not shown here) using the wind data from
this flight show comparable position errors, with a slightly higher 2
bound of 2.88 ft and more wander during the pirouettes.

D. Hover Tests: Dispersion in Strong Winds

Fig. 17 Tether test rig with T-wing flying below; both tethers are slack; The vehicle was also flown using the standard  pattern in winds
first author is to the left, managing the tether rope. up to 18 kt (see Fig. 12). Plots for this are shown in Figs. 22–24.
 682 STONE ET AL.

10
15

5 10

5
South − North (ft)

South − North (ft)

0

−5
−5

−10
−10

−15
−15 −10 −5 0 5 10 −20 −15 −10 −5 0 5 10 15
West − East (ft) West − East (ft)
Fig. 19 Plan-view position plot for normal hover flight on 13 July 2006; Fig. 22 Plus pattern plan view (ft) in strong winds; plot covers full
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

winds at 6–8 kt; plot covers full flight including both  patterns and flight, including both  patterns and both pirouettes; 
2:79 ft (thick
both pirouettes (bounding cross indicates 2   error variation for flight lines indicate 2 error bounds).
path); 
1:21 ft.

20 20

pN, (ft)
pN, (ft)

0 0

−20 −20
20 20
pE, (ft)

pE, (ft)

0 0

−20 −20
20 20
h, (ft)

h, (ft)

10 10
0 0

0
φ , ( °)

0
φv, ( °)

−200
v

−200
−400
200 300 400 500 600 700 800 −400
Time (s) 950 1000 1050 1100 1150 1200 1250 1300
Time (s)
Fig. 20 Position states during hover tests (commanded positions Fig. 23 Position states (ft) for plus pattern in strong winds
dashed); north and east doublets are clear. In the first doublet, the vehicle (commanded position doublets dashed); hover precision is clearly less
belly is facing north and the vehicle moves than for the moderate-wind case.
N ! S ! N ! E ! W ! E; in the second doublet, the vehicle is
skewed facing NE, but the vehicle performs the same N ! S ! N !
E ! W ! E doublet cleanly.
pitch
V tilt angles (up to about 40 deg) that are required to hold
position in this wind are clear.
20
Comparing these results with the simulation (Figs. 10 and 11), the
hover precision is marginally worse, and the responses (particularly,
Wind Speed (ft/s)

the vertical pitch angle) exhibit less damping and greater excursions
10 than in the simulation. These discrepancies point to possible errors in
the aerodynamic database, though the presence of experimental
0 imprecision (such as significant engine vibrations causing filter
errors) may also be important. The vertical pitch angle also shows a
Wind Direction ( ° )

250

200 60
40
150
20
θV (°)

100 0
200 300 400 500 600 700 800
−20
Fig. 21 Wind speed and direction between 8 and 14 ft/s (6–8 kt) from −40
the south (measured at the ground station 60 ft from the takeoff point).
40

20
Although the precision is clearly less than before (this time the 2
ψV (°)

0
error bound is 5.58 ft), it is reasonable considering the strength of the
wind. The error is also commensurate with the original design −20

specification for the vehicle, which required a maximum 2.0-ft −40

950 1000 1050 1100 1150 1200 1250 1300
position deviation during hover in the presence of a 10-kt (17 ft=s) Time (s)
sharp-edged gust. Figure 23 also shows the obvious pirouettes V Fig. 24 Vertical Euler angles in strong winds; the v tilt angle (angle
that occur between 1130 and 1240 s. Finally, Fig. 24 shows the away from vertical) reaches 40 deg at some points when the belly is
vertical Euler tilt angles during this flight. From this plot, the large opposite the wind (belly facing north and wind from the south).
 STONE ET AL. 683

different trim bias in the test, which is probably due to a z-wise c.g.
offset toward the belly of the vehicle, causing an increase in the trim
vertical pitch angle.

VII. Horizontal and Transition Flight Test Results

During August 2006, the first three transition and horizontal flight
tests of the T-wing were conducted. A further horizontal flight was
also conducted in February 2007. To date, the T-wing has completed
autonomous takeoffs, V2H transitions, horizontal transitions, and
H2V transitions. Some problems were encountered at the top of the
H2V transition, due to issues with rudder gain scheduling and the
extreme angles of attack encountered. Landings have thus been
performed with pilot-assisted guidance-level control. The data
presented here show various qualitative aspects of the T-wing
horizontal and transition flight performance.
Figure 25 shows the flight path of the vehicle for the flight on
30 August 2006,§§ overlaid on an aerial photograph of the University Fig. 25 Flight path on 30 August 2006 in Marulan, New South Wales,
of Sydney Marulan test site. Graphs of pitch-angle, velocity, and Australia (circles represent GPS readings).
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

altitude for the complete flight are provided in Fig. 26. This flight
consisted of the following phases: 1) autonomous takeoff and
tracking through two vertical waypoints: one at 100 ft and the other at
300 ft, 2) autonomous transition to horizontal flight, 3) autonomous 100
tracking through the first horizontal flight waypoint and turn to track

θ, (°)
50 st
Final H2V Transition
1 V2H Transition
the next waypoint, 4) pilot guidance-level intervention at the end of
0
this turn to correct altitude loss (due to a complicated interaction
between underpredicted drag and gain scheduling of the pitch-rate 200
controllers), 5) performance of first H2V transition semi- 1st H2V Transition,
V, (ft/s)

2nd V2H Transition

autonomously followed by a commanded V2H transition and a 100

further pilot-guided circuit, and 6) performance of second H2V 0

transition semi-autonomously followed by descent to landing under
pilot guidance-level control. 400
h, (ft)

In the preceding description, the H2V transitions are described as 200

semi-autonomous because the pilot stick input (which maps to
0
vehicle pitch-angle commands during pilot-assisted horizontal 250 300 350 400 450 500 550
Time (s)
flight) is set so that full backstick is treated as a logic signal, forcing
the vehicle to perform a transition maneuver. Provided backstick is Fig. 26 Pitch angle, velocity, and altitude for flight on 30 August 2006.
maintained through the maneuver, the maneuver guidance is
autonomous until a vertical attitude is reached.
This flight demonstrates a variety of important characteristics of
80
the concept-demonstrator version of the T-wing.
θ, (°)

60
40
A. Altitude Loss in the V2H transition 20

The total altitude loss in the first V2H transition was 10 ft.
100
Furthermore, because the vertical flight guidance of the vehicle
V, (ft/s)

requires that it dwells for 3 s¶¶ at each vertical mode waypoint, the 50
vehicle performed this transition virtually from a standing start. (The
total velocity at entry was 3:61 ft=s). This result shows a slight 150
difference with the simulation data, in which the vehicle climbed
h, (ft)

slightly through the transition. This suggests that the simulation 100
overpredicts the total engine thrust, particularly at the top end of
415 415.5 416 416.5 417 417.5 418 418.5 419 419.5 420
the throttle. Time (s)
Fig. 27 Details of pitch angle, velocity, and altitude for flight on
B. H2V Transition Details and Mechanism 30 August 2006 at the point of the second H2V transition.
In the final H2V transition, the total height gain was 80 ft. A
detailed graph of the pitch-angle, velocity, and altitude states is and W velocity components are 110.3, 13:3, and 26.2 ft/s; at the end
provided for the 5-s period around this transition in Fig. 27. A further of the sharp pitch change, they are 31.78, 17:85, and 82.37.
graph of the vertical pitch and yaw angles for this period is provided Ignoring wind,∗∗∗ this final value corresponds to an angle of attack of
in Fig. 28. Figure 27 highlights the very abrupt nature of this 68.7 deg. Over the abrupt pitch change part of the maneuver, the total
particular transition. The predominant pitch change occurs between altitude gain is 17.0 ft. The reason for the extreme abruptness of the
t  415:74 and 416.25 s, when the pitch- angle changes from 26.2 to pitch-up is believed to be associated with a dynamic stall of the rear
82.40 deg. During this period, the vertical pitch angle increases from (high aspect ratio) main wing under low-throttle conditions, which
63:4 to 1.0 deg, after which it reaches a maximum value of then conveniently allows the back of the vehicle to “fall under” the
13.3 deg, providing dynamic braking to allow the vehicle to wash off nose, leaving the vehicle in a vertical attitude. The abrupt pitch
excess translational velocity. At the start of this maneuver, the U, V, change and dynamic braking parts of the maneuver are clear in the
reconstructed vehicle plots provided in Fig. 29, which shows the
§§
Videos of this and other T-wing flights can be found at http:// vehicle attitude and velocity vector at 0.5-s time increments between
www.aeromech.usyd.edu.au/uav/twing [retrieved 24 Sept. 2007].
¶¶
The dwell time at vertical waypoints was reduced for this flight from the
∗∗∗
6 s used previously. The vehicle was not equipped with an air data probe.
 684 STONE ET AL.

20

0
conditions). This is clearly inappropriate for such a dynamic
maneuver as seen here; in this case, the vehicle was using scheduling
−20
for standard vertical hover flight, because the pitch angle was
θV (°)

−40
approximately vertical, even though it had a translational velocity of
−60
90:1 ft=s. Even though the rudder aerodynamic contributions may be
−80
unpredictable under such conditions, the engine throttle-balancing
mechanism tied to the rudder deflection is still very effective at these
10
angles of attack and severely underpredicted using its hover values.
0
No oscillations are seen in the slower simulated transition. The fact
−10
ψV (°)

that no pitch transients are observed at the top of the transition when
−20
the low-level translational velocity controllers are engaged suggests
−30
that the mechanism outlined in Eq. (8) to perform this mode
−40
415 415.5 416 416.5 417 417.5 418 418.5 419 419.5 420 transition is effective.
Time (s)

Fig. 28 Vertical Euler tilt angles during transition; lateral oscillations

are clear (flat spots in the graphs indicate missed data points.)
VIII. Conclusions
The primary result from flight tests over the last two years is the
affirmation of the feasibility of the T-wing tail-sitter concept. The
vehicle was flown autonomously from takeoff to landing in vertical
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

flight in the presence of considerable winds with success. It has also

completed all portions of the complete flight envelope, including
transition maneuvers, though with the necessity of pilot guidance-
level input during the H2V transition and landing phases. The current
control structure, which uses an amalgam of simple controllers for
the different primary flight phases, has proved adequate to tackle the
control problem. In addition to some specific fixes for known control
issues, the main problem observed with this particular structure is
that it requires the presence of significant mode-control logic and
special consideration to be given to controller switching. The flight
tests also demonstrate that the aerodynamic modeling used to build
the simulation is at least accurate enough for control design purposes,
though certain particular discrepancies were observed between test
and simulation results.
Fig. 29 Transition attitude and velocity vector at 0.5-s intervals Future work will concentrate on achieving autonomous flight
(reconstructed from flight test data); the length of the velocity vector is through the complete flight envelope when time and funding permit.
scaled with respect to the vehicle size to represent the distance traveled in The most obvious area requiring improvement is the yaw oscillation
1=10 of a second. at the top of the H2V transition. This will be done by improving the
gain scheduling of rudder controllers to account for the effect that
significant W velocities have on rudder effectiveness, due to the
t  416:75 and 418.25 s. Note that the dynamic braking is an presence of the rudder-to-differential throttle interconnect. Because
intentional part of the maneuver. The last frame in this sequence also the current higher-quality inertial system is able to withstand much
shows the yaw instability at the top of the transition (discussed later). longer GPS outages, the full velocity-state information can be used.
In comparison with the simulation results (Fig. 15), the pitch The addition of a dedicated air data system will also facilitate the
change is more rapid and the altitude gain is significantly less. prediction of velocity-state information, though it is unlikely that a
Although the results are not directly comparable,††† the differences low-cost sensor would handle the extreme angles of attack observed
suggest that there may be some important discrepancies in the during the H2V transition reliably. The maneuver itself may be
simulation database that affect this transition. For instance, the refined to limit the angles of attack reached, though with the penalty
simulation shows a small pitch oscillation as the vehicle passes of greater height gained. This would be done by increasing entry
through 70 deg during the transition, which is not seen in the test. speed and throttle profile during the maneuver. With a demonstrated
Such discrepancies are not altogether unexpected, because it was steady-state descent velocity of 4 ft=s, an extra 300 ft of altitude gain
always known that the database would have problems at high-speed comes at the cost of an extra 75 s of descending flight.
poststall angles (approximately the shaded region in the trim graph in For hover-mode flight, improvements to the guidance and control
Fig. 4). strategies through the use of a quaternion formulation of attitude will
be considered, as is done in the transition mode. The use of a model-
C. Oscillations at the End of the H2V Transition
predictive control strategy will also be assessed in an effort to
eliminate the multiplicity of different controllers in use. (Initial
In the graphs of pitch angle, some oscillations are observed after vertical mode flights have been conducted.) Finally, parameter
the completion of the H2V maneuver. These are, in fact, oscillations identification work, particularly in hover flight in strong-winds, is
in the vertical yaw direction, rather than vertical pitch direction, as needed to establish the quality of the aerodynamic modeling. This
confirmed by Fig. 28. They are associated with incorrect rudder gain would allow deficiencies to be identified and controllers to be
scheduling, which caused the rudder to saturate in both directions at improved.
the top of the maneuver. This is because the vehicle was originally
equipped with much-lower-quality inertial sensors, for which the
velocity estimates would diverge rapidly in the event of a short GPS
outage. For this reason and the lack of an air data probe, the original Acknowledgments
gain scheduling of controllers was based on pitch angles and throttle This work would not have been possible without a grant from the
settings (and the predicted correlation of these back to trim-velocity U.S. Air Force Office of Scientific Research (AFOSR), contract
number FA5209-04-P-0563, which was instrumental in purchasing
†††
The simulation was done fully autonomously, whereas the flight test was the much-higher-precision real-time kinematic global positioning
performed with pilot guidance-level input up to the entry point of the system sensor with inertial measurement unit interface that
maneuver. has enabled fully autonomous flights on the tether test rig. The
 STONE ET AL. 685

continued support of Sonacom Pty, Ltd., in this project is also Control, and Dynamics, Vol. 9, No. 1, 1986, pp. 99–107.
acknowledged. [15] Junkins, J. L., Akella, M. R., and Robinett, R. D., “Nonlinear Adaptive
Control of Spacecraft Maneuvers,” Journal of Guidance, Control, and
Dynamics, Vol. 20, No. 6, 1997, pp. 1104–1110.
References [16] Vidali, S. R., Kraige, L. G., and Junkins, J. L., “New Results on the
[1] Stone, R. H., “Configuration Design of a Canard Configured Tail-Sitter Optimal Spacecraft Attitude Maneuver Problem,” Journal of Guidance,
Unmanned Air Vehicle Using Multidisciplinary Optimisation,” Ph.D. Control, and Dynamics, Vol. 7, No. 3, 1984, pp. 378–380.
Dissertation, Dept. of Aeronautical Engineering, Univ. of Sydney, [17] Klumpp, A. R., “Singularity-Free Extraction of a Quaternion from a
Sydney, Australia, 1999. Direction-Cosine Matrix,” Journal of Spacecraft and Rockets, Vol. 13,
[2] Stone, R. H., “The T-Wing Tail-Sitter Research UAV,” 2002 Biennial No. 12, 1976, pp. 754–755.
International Powered Lift Conference and Exhibit, AIAA Paper 2002- [18] Kubo, D., and Suzuki, S., “Transitional Flight Control of Tail-Sitter
5970, 2002. Vertical Takeoff and Landing Mini Unmanned Aerial Vehicle,” AIAA
[3] Stone, R. H., and Wong, K. C., “Preliminary Design of a Tandem-Wing Infotech@Aerospace 2007 Conference and Exhibit, Rhnert Park, CA,
Tail-Sitter UAV Using Multi-Disciplinary Optimisation,” AUVSI ’96 AIAA Paper 2007-2752, 2007.
Proceedings, Vol. 1, Association of Unmanned Vehicle Systems [19] Stevens, B. L., and Lewis, F. L., Aircraft Control and Simulation,
International, Arlington, VA, 1996, pp. 163–178. 1st ed., Wiley, New York, 1992.
[4] Albion, N., “Vertically Launchable and Recoverable Winged Aircraft,” [20] Stone, R. H., “Aerodynamic Modelling of a Wing-in-Slipstream Tail-
The Boeing Company, Chicago, IL, U.S. Patent No. 5765783, 1998. Sitter UAV, 2002 Biennial International Powered Lift Conference and
[5] Munson, K., Jane’s Unmanned Aerial Vehicles and Targets, Jane’s Exhibit, AIAA Paper 2002-5951, 2002.
Information Group, Surrey, England, U.K., 1998. [21] Stone, R. H., “Aerodynamic Modelling of the Wing-Propeller
[6] Bridgman, L., Jane’s All The World’s Aircraft 1954–55, Jane’s All The Interaction for a Tail-Sitter UAV,” Journal of Aircraft (to be published).
[22] Shuster, M. D., “A Survey of Attitude Representations,” Journal of the
Downloaded by UNIVERSITY OF SYDNEY on August 20, 2014 | http://arc.aiaa.org | DOI: 10.2514/1.32750

World’s Aircraft Publishing, London, 1955, p. 227.

[7] Bridgman, L., Jane’s All The World’s Aircraft 1955–56, Jane’s All The Astronautical Sciences, Vol. 41, No. 4, 1993, pp. 439–517.
World’s Aircraft Publishing, London, 1956. [23] Phillips, W. F., Hailey, C. E., and Gebert, G. A., “Review of Attitude
[8] McCormick, B. W., Aerodynamics of V/STOL Flight, Dover, Mineola, Representations Used for Aircraft Kinematics,” Journal of Aircraft,
NY, 1999. Vol. 38, No. 4, 2001, pp. 718–737.
[9] Cycon, J. P., Rosen, K. M., and Whyte, A. C., “Unmanned Flight [24] Hale, F. J., Introduction to Aircraft Performance, Selection and Design,
Vehicle Including Counter Rotating Rotors Positioned Within a 1st ed., Wiley, New York, 1984.
Toroidal Shroud and Operable to Provide all Required Vehicle Flight [25] Blakelock, J. H., Automatic Control of Aircraft and Missiles, 2nd ed.,
Controls,” United Technologies Corp., Hartford, CT, U.S. Patent Wiley, New York, 1991.
No. 5152478, 1992. [26] Etkin, B., Dynamics of Atmospheric Flight, 1st ed., Wiley, New York,
[10] Avanzini, G., Ciniglio, U., and de Matteis, G., “Full-Envelope Robust 1972.
Control of a Shrouded-Fan Unmanned Vehicle,” Journal of Guidance, [27] Stone, R. H., “Control Architecture for a Tail-Sitter Unmanned Air
Control, and Dynamics, Vol. 29, No. 2, 2006, pp. 435–443. Vehicle,” 5th Asian Control Conference, Vol. 2, Inst. of Electrical and
[11] Johnson, E. N., and Turbe, M. A., “Modeling, Control, and Flight Electronics Engineers, Piscataway, NJ, 2004, pp. 736–744.
Testing of a Small Ducted-Fan Aircraft,” Journal of Guidance, Control, [28] Stone, R. H., “Design Considerations for a Wing-in-Slipstream Tail-
and Dynamics, Vol. 29, No. 4, 2006, pp. 769–779. Sitter Unmanned Air Vehicle (UAV),” 10th Australian International
[12] Knoebel, N. B., Osborne, S. R., Snyder, D. O., McLain, T. W., Beard, R. Aerospace Congress, Royal Aeronautical Society, Australian Div.,
W., and Eldredge, A. M., “Preliminary Modeling, Control, and 2003, Paper aiac2003-068.
Trajectory Design for Miniature Autonomous Tailsitters,” AIAA [29] Gavrilets, V., Mettler, B., and Feron, E., “Human-Inspired Control
Guidance, Navigation, and Control Conference, Keystone, CO, AIAA Logic for Automated Maneuvering of Miniature Helicopter,” Journal
Paper 2006-6713, 2006. of Guidance, Control, and Dynamics, Vol. 27, No. 5, 2004, pp. 752–
[13] Johnson, E. N., and Kannan, S. K., “Adaptive Trajectory Control for 759.
Autonomous Helicopters,” Journal of Guidance, Control, and [30] Stone, R. H., “Transition Maneuver Optimization for the T-Wing Tail-
Dynamics, Vol. 28, No. 3, 2005, pp. 524–538. Sitter UAV,” 10th Australian International Aerospace Congress,
[14] Carrington, C. K., and Junkins, J. L., “Optimal Nonlinear Feedback Royal Aeronautical Society, Australian Div., 2003, Paper aiac2003-
Control for Spacecraft Attitude Maneuvers,” Journal of Guidance, 069.

View publication stats