IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 4, NO.

3, JULY 2019 2683

Attitude and Cruise Control of a VTOL Tiltwing UAV

David Rohr , Thomas Stastny , Sebastian Verling, and Roland Siegwart

Abstract—This letter presents the mathematical modeling, con-

troller design, and flight testing of an overactuated vertical take-off
and landing tiltwing unmanned aerial vehicle (UAV). Based on sim-
plified aerodynamics and first principles, a dynamical model of the
UAV is developed which captures key aerodynamic effects includ-
ing propeller slipstream on the wing and poststall characteristics
of the airfoils. The model-based steady-state flight envelope and
the corresponding trim-actuation is analyzed and the overactua-
tion of the UAV solved by optimizing for, e.g., power-optimal trims.
The developed control system is composed of two controllers. First,
a low-level attitude controller based on dynamic inversion and
a daisy-chaining approach to handle allocation of redundant ac-
tuators. Second, a higher level cruise controller to track a desired
vertical velocity. It is based on a linearization of the system and
lookup tables to determine the strong and nonlinear variation of the
trims throughout the flight envelope. We demonstrate the perfor- Fig. 1. The tiltwing UAV on which the presented control system is imple-
mance of the control system for all flight phases (hover, transition, mented, here shown in transition mode with partially tilted wing.
cruise) in extensive flight tests.
Index Terms—Aerial systems, mechanics and control, motion
control, hybrid UAV, overactuation.

I. INTRODUCTION
NMANNED aerial vehicles (UAVs) are extensively inves-
U tigated in the robotics community. Over the last decades,
various designs evolved to meet the requirements of specific mis-
sion profiles. Fixed-wing aircraft designs offer high endurance,
Fig. 2. Schematic of the tiltwing UAV in the different flight-phases hover
large range, and high speeds while rotary-wing platforms such (III), transition (I), and cruise (II). The body-fixed frame of reference is located
as the popular multirotor feature high maneuverability, hover- in the center of gravity (CG) and denoted with the superscript B, the inertial,
and Vertical Take-Off/Landing (VTOL) capabilities. There is in- earth-fixed frame is labeled with the superscript I.
creasing interest in the development of highly versatile, so-called
“hybrid” UAVs that can operate both as fixed- and rotary-wings
and thus combine the benefits of the respective designs [1]. Ex- see Fig. 1. Compared to the tailsitter design [2], [3], TWVs ben-
amples are the tiltwing presented in this work, the tiltrotor and efit from the fuselage remaining horizontal. Versus the tiltro-
the tailsitter. A tiltwing vehicle (TWV) features a wing that ro- tor, they feature improved stall characteristic and more effective
tates together with the propulsion system between a horizontal wing-born lift due to the continuous immersion of the wing in
(cruise-mode) and upright- (hover-mode) position (Fig. 2). At the well-aligned propeller slipstream.
intermediate wing-tilt angles, the lift-force resolves into contri- The large flight-envelope of a TWV imposes challenging re-
butions of both the propulsion system and the airfoils, leading to quirements on the flight-control system. The transition phase
a blend of fixed- and rotary wing operations (transition mode), is characterized by strong nonlinearities in the dynamics of the
aircraft. These result from the interaction of the wing and the
propeller slipstream, the wings wide range in angles of attack,
Manuscript received February 24, 2019; accepted April 14, 2019. Date of
publication May 1, 2019; date of current version May 10, 2019. This letter was and, generally, the large variation of trim-settings throughout
recommended for publication by Associate Editor V. Lippiello and Editor J. the flight envelope. For the TWV presented in this work, an ad-
Roberts upon evaluation of the reviewers’ comments. (Corresponding author: ditional control challenge is given by actuator redundancy that
David Rohr.)
The authors are with the Autonomous Systems Lab, ETH Zurich 8092, leads to an overactuation for both attitude- and cruise control.
Zurich, Switzerland (e-mail: [email protected]; thomas.stastny@mavt.
ethz.ch; [email protected]; [email protected]).
This letter has supplementary downloadable multimedia material available
A. Related Work
at http://ieeexplore.ieee.org provided by the authors. This includes a video file, Existing work on design, modeling and control of tiltwing
which shows a selection of conducted flight tests to assess the performance of
the control system. This material is 8.95 MB in size. UAV’s considers tandem-wing [4]–[6] and single-wing vehicles
Digital Object Identifier 10.1109/LRA.2019.2914340 [7]–[9]. Employed control systems are either unified [8], [9] or
2377-3766 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: Khalifa University. Downloaded on September 19,2023 at 07:19:32 UTC from IEEE Xplore. Restrictions apply.
2684 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 4, NO. 3, JULY 2019

switch between different controllers for hover, transition, and comes fully equipped with all required actuators including a
cruise [10]. For both attitude- and velocity control, decoupled tilt-mechanism for the wing. It features a wing-span of 0.94 m
PID and full state feedback LQR-architectures are reported [5], and its take-off mass amounts to 1.9 kg. In cruise configuration,
[9], [11]. They are typically combined with local linearizations flights of up to ∼ 20 min are possible while in hover, endurance
and gain-scheduling to address the strong non-linearities. Ex- is limited to ∼ 5 min (battery: 14.8 V, 3800 mAh).
amples of H∞ -based attitude- and cruise control are found as
well [4], [12]. A popular non-linear control technique involves
dynamic inversion (DI) [13] and is used both for tailsitters [2], A. Avionics
[3] and tandem TWVs [4]. It enables reference-model follow- In order to implement our own flight-control system, the UAV
ing but requires an accurate model to estimate state-dependent is refitted with the Pixhawk Autopilot [15] running the PX4 au-
moments and forces. High-fidelity models are required to ad- topilot software [16]. The Pixhawk provides a six-axis Inertial
dress the complex transition phase and typically consider the Measurement Unit (IMU) (3-axis accelerometer + 3-axis gy-
prominent propeller slipstream interaction with the wing [11]. roscope), a 3-axis magnetometer, and a barometer. Addition-
Instead of modeling, [9] describes a control system that is based ally, the system is complemented with a differential-pressure
exclusively on state- and control derivatives obtained from wind- airspeed-sensor and a GNSS-module. All sensor data is fused in
tunnel testing, [2] and [3] introduce lumped-parameter models the ready-to-use state-estimation available within PX4, resulting
to fit experimental data for a flying-wing tailsitter. in attitude, altitude and airspeed estimates that are subsequently
used in the feed-back flight-control system.
B. Contribution
In this work, we present a global, model-based control sys- B. Actuation Principle
tem that tracks the full desired attitude and vertical airspeed in Fig. 2 depicts the different available actuators. Their function
all flight phases. The attitude control system (ACS) is based on depends on the flight phase and the configuration of the UAV:
i) a high-fidelity model built from first principles, ii) dynamic In hover flight, roll and pitch are controlled by thrust-vectoring
inversion and iii) a daisy-chaining approach to handle overactua- of the main-propellers (pl, pr) and the tail-propeller (pt). Yaw is
tion. This combination is novel in its application to single-wing actuated by tilting the tail-propeller thrust vector around the body
TWVs. The good performance of the ACS in the flight-tests x-axis (tt). Redundantly, yaw-moment can also be generated by
indicates that the proposed model structure reasonably trades- differential deflection of the slipstream-immersed ailerons (al,
off between i) DI-required fidelity and ii) low-computational ar). Horizontal maneuvering is performed by tilting the UAV,
complexity to remain tangible for use on micro-controllers. The and hence the net thrust-vector, into the desired direction. Climb-
developed cruise control system (CCS) employs a linearized ing and sinking is achieved by collective throttling of all pro-
approach similar to that outlined in [9], i.e., it relies on look- pellers.
up trim-maps (TMs) to determine the nonlinear trim-actuation. In cruise flight, standard fixed-wing controls apply, i.e., roll,
However, contrary to [9] where wind-tunnel based TMs are used, pitch and yaw are controlled with the ailerons, elevator (e) and
we rely on model-based TMs obtained by offline full-state opti- rudder (r), respectively. Additionally, yaw- and negative pitch
mization to systematically handle non-uniqueness of the trims. moment can be generated by differential throttle on the main-
Additionally, the CCS presented includes feedback control to propellers and the tail-propeller thrust, respectively. Again, this
account for modeling errors and to attenuate disturbances. The provides redundancy and illustrates the overactuation for attitude
resulting vertical velocity tracking accuracy and -range improves control. Airspeed and climb-rate are controlled by coordinating
on the data presented in [9], thus rendering modeling with ve- main-propeller thrust and pitch-angle of the UAV.
locity feedback a valuable alternative to laborious wind-tunnel In the transition phase, the control strategies overlap, e.g.,
testing. rolling and yawing both require simultaneous thrust vectoring
and aileron deflection. Horizontal- and vertical velocity control
C. Outline includes combined wing-tilt actuation, pitch-angle- and throt-
The remainder of the letter is structured as follows: In tle selection. This combination is not necessarily unique, hence,
Section II, the system is introduced, followed by the modeling overactuation is again present: For example, hovering is possible
in Section III. Optimal trim-actuation is analyzed in Section IV with every wing-tilt angle (ζw ) and fuselage-pitch (θ) combina-
and forms the basis for the CCS. Sections V and VI introduce tion that leads to the main-thrust vector pointing upward, i.e.,
the attitude- and cruise-control architectures, respectively. The ζw + θ ≈ 90◦ .
performance of the controllers is demonstrated in flight experi-
ments presented in Section VII. Finally, an outlook is given in C. Nomenclature
Section VIII.
For system analysis, we introduce a body-fixed forward-right-
down frame of reference located in the UAV’s center of gravity
II. SYSTEM DESCRIPTION
(CG) (cf. Fig. 2). CG variation upon tilting of the wing is ne-
The employed tiltwing-UAV is a commercially available glected, it amounts to ±1 cm along the body z-axis w.r.t. the CG
radio-controlled (RC) aircraft [14]. It is a replica of the Canadair used for modeling. Vector-quantities are written in bold-face and
CL-84 manned tiltwing aircraft which flew in the 1960’s and denoted with lower pre-script B and I when expressed in body-
Authorized licensed use limited to: Khalifa University. Downloaded on September 19,2023 at 07:19:32 UTC from IEEE Xplore. Restrictions apply.
ROHR et al.: ATTITUDE AND CRUISE CONTROL OF A VTOL TILTWING UAV 2685

and inertial-frame, respectively (cf. Fig. 2). Normalized actua-

tor control inputs are denoted with δi , actual actuator positions
with ζi , and propeller speeds with ηi . The subscript defines the
actuator and will be introduced when used.

III. SYSTEM MODELING

The UAV is modeled as a single rigid body whose translational
and angular dynamics are driven by the net force and moment
Fig. 3. Schematic of an airfoil segment with a local frame of reference W =
acting on it. With the Newton-Euler equations and rigid-body (eW
x , ey , ez ) and the quantities involved to compute the aerodynamic forces
W W
kinematics, the position (x), translational velocity (v), attitude and -moments.
(RIB ) and angular rate (ω) of the UAV are defined by
I ẋ = Iv
CQ are approximated as affine function of J (cf. [17]). For sim-
mI v̇ = mI g + RIB · B F plicity, other components, such as the propeller rolling moment,
are neglected.
ṘIB = RIB [B ω]×
B IB ω̇ = B M − B ω × B IB ω (1) B. Airfoils

where m denotes the UAV’s total mass and B I the UAV’s mo- We divide the wing and stabilizers into multiple span-wise
ment of inertia expressed in B. The gravitational acceleration segments to account for the different inflow conditions which
is given by g, [B ω]× denotes the skew-symmetric cross-product depend on ω and, for wing segments located behind a propeller,
matrix of B ω. We express the attitude (RIB ) with a rotation- the propeller slipstream velocity w. Resulting forces and mo-
matrix mapping from body- to inertial frame. The net aerody- ments are calculated at the center of pressure rcp of each seg-
namic force F and moment M are formed by accumulating the ment, see Fig. 3. The local airspeed ua is given by
contributions of the different components, i.e., wing, stabilizers, ua (rcp ) = va + ω × rcp (+w) (5)
fuselage and propellers. Tiltwing-specific aerodynamic effects
to be modeled include the forces and moments generated by For simplicity, the spatial evolution of the propeller wake [17],
i) the airfoils subject to full ±180◦ free-stream angle of attack, [19] is neglected and the induced velocity w approximated by
ii) the propeller-slipstream effects on airfoils located down- the value at the corresponding propeller-hub. From disk actuator
stream of the propellers, and iii) the propellers facing different theory [17]:
  
inflow-conditions throughout the flight envelope. 1 2T
w = p −V,∞ + V,∞ +2 (6)
2 ρA
A. Propellers
With va the air-relative velocity (airspeed) of the UAV’s CG, with A = πD2 /4 the propeller-disk area. Segment-wise lift
the local airspeed ua at the propeller hub is given by (ΔFL ), drag (ΔFD ) and moment (ΔM) contributions are
finally obtained by
ua (rp ) = va + ω × rp = V,∞ p + V⊥,∞ p⊥ (2)
1
ΔFL = CL (α, ζcs ) · ρV 2 · c · Δy · eL
with rp the CG-relative position of the propeller hub. The local 2
airspeed resolves in an axial (V,∞ ) and radial (V⊥,∞ ≥ 0) free- 1
flow component, p and p⊥ are unit vectors pointing in propeller ΔFD = CD (α, ζcs ) · ρV 2 · c · Δy · eD
2
forward and radial direction, respectively. According to [17],
1
[18], the net force of propeller p is composed of the thrust T and ΔMm,c/4 = CM (α, ζcs ) · ρV 2 · c2 · Δy · eW (7)
normal force N : 2 y

where the definition of most quantities is illustrated in Fig. 3.

B Fp = ρη 2 D4 CT (J) B p − ημN V⊥,∞ B p⊥ (3) Lift- and drag direction are denoted by eL and eD , respectively,
     
T N and V = uldp  (cf. Fig. 3). The aerodynamic coefficients CL ,
with the thrust-constant CT depending on the propeller advance- CD , CM depend on the angle of attack α ∈ [−π, π] and, if con-
ratio J and a lumped-parameter constant μN > 0. Further, ρ, η, trol surfaces (CS) are present, the CS deflection ζcs . Simple
and D denote air density, propeller speed and propeller diameter, linear and quadratic relations are employed if the segment is
respectively. not stalled (αs− < α < αs+ ). In post stall (αs− α, α αs+ ),
The reactive propeller-moment due to the air-drag of the pro- the wing is assumed to behave like a flat plate (fp) and we
peller blades amounts to approximate the coefficients by:
CLf p = CL,π/4
fp
· sin(2α)
B Mp = −ρη 2 D5 CQ (J)εB p (4)
with CQ the torque-constant. The propeller turning direction is
fp
CD fp
= CD,min fp
+ (CD,π/2 fp
− CD,min ) · sin(α)2
determined by ε ∈ {−1, 1} where ε = 1 if the turning direction
is positive along p (right-handedness) and vice-versa. CT and
fp
CM fp
= −CM,max · sin(sgn(α) · α2 /π) (8)

Authorized licensed use limited to: Khalifa University. Downloaded on September 19,2023 at 07:19:32 UTC from IEEE Xplore. Restrictions apply.
2686 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 4, NO. 3, JULY 2019

to match reported experimental data, e.g., [20], [21]. Close to A. Problem Formulation and Optimization
the stall angles (αs− , αs+ ), interpolation between both models
The trim-map is calculated offline in a nonlinear, constrained
yields a smooth changeover. optimization which minimizes translational (v̇) and angular (θ̈)
unsteadiness and, simultaneously, seeks to reduce a user-defined
C. Fuselage cost-function q which is included to render the trim solution
unique:
Modeling of the fuselage follows a simplified approach known
as “quadratic aerodynamic form” (cf. [22], p. 20): min v̇T Qv v̇ + Qθ θ̈2 + q(ut , θt ) s.t.
ut ,θ t
ρ B f f B f
B Ff =− e C Vuu + eB
y CD,y Vvv + ez CD,z Vww 1
2 x D,x I v̇ = Ig + RIB · B F(ut , θt )
m
B Mf =0
B ω̇ = B I−1 B M(ut , θt )
Vuu = u|u|, Vvv = v|v|, Vww = w|w| (9)
−π/2 ≤ θt ≤ π/2, ut ∈ U (12)
f
with (u, v, w) = B va and
T
CD,i
the drag coefficients of the with Qv , Qθ positive definite weightings and U the set of ad-
fuselage subject to i-axis-aligned airflow. missible, non-saturated actuator inputs. The equality constraints
follow from the system dynamics (1), F and M further de-
D. Parameter Identification pend on (va , γ). In q ≥ 0 we include penalties on i) net power-
consumption, ii) control-surface saturation, iii) deviation from a
The introduced parameters are either identified (experimental desired pitch-angle θ∗ and iv) deviation from solutions of close
assessment of static thrust- and moment curves of the propulsion operating points to penalize discontinuous trim-maps and, thus,
systems and CAD-based approximation of inertial properties) or prevent discrete switching of the feed-forward trim values in the
estimated based on values obtained from literature (all airfoil- cruise controller.
related data). At every (va (i), γ(j)) contained in the trim-map, we per-
formed the optimization using the lsqnonlin solver of the Matlab
IV. TRIM ANALYSIS optimization toolbox [23]. Finally, the resulting steadiness (v̇, θ̈)
was thresholded to decide upon incorporation of (va (i), γ(j))
Cruise-control system development is preceded by assessing in the steady flight envelope. It is worth noting that the resulting
the steady-state flight envelope, i.e., the set of operating points steady flight envelope is generally a conservative estimate due
for which a dynamic equilibrium exists. We restrict the formu- to the risk of the solver getting trapped in a local optimum or
lation of the cruise-controller and the system analysis to the too much weight being put on minimizing the additional cost
2-dimensional, longitudinal dynamics. An operating point can q. To minimize the risk of locally optimal solutions, the solver
therefore be defined by the flight-path angle γ and the airspeed requires appropriate initial guesses (IG).
magnitude va . The point is declared steady feasible if for the
given pair (va , γ) there exists a trim-pitch θt and a set of trim- B. Initial Guess Generation
actuations ut
We devise an iterative procedure to generate IGs during build-

θ (va , γ), u (va , γ) =
t t
δw
t
δpl,r
t
δal,r
t
δet δpt
t
(10) up of the trim-map: At every operation point (va (i), γ(j)) in the
trim-map, the optimization is solved once with every available
such that solution of the neighboring operation points as IG (eight in to-
tal for a Cartesian grid). The steady-feasible solution (ut , θt )
ω = ω̇ = v̇ = 0. (11) which yields the lowest cost is adopted as preliminary trim at
(va (i), γ(j)). If, in a subsequent iteration, a neighboring point
The superscript t relates to the trim setting and the subscripts of manages to further lower its cost with a new solution, the trim
the δ’s specify the actuator (cf. Fig. 2). If no such actuation/pitch- at (va (i), γ(j)) is revisited with this solution as IG and adjusted
angle exists, the operating point cannot be stabilized by any con- upon improvement. This procedure is conducted at every point in
trol system. The mapping T : (va , γ) → (ut , θt ) is referred to the map for multiple iterations until the solutions do not change
as trim-map. It is calculated in a discrete form to serve as look- anymore. At this instant, mutually lowest costs are achieved
up table for feed-forward actuation of the wing-tilt, throttle, and among neighboring points in the map.
pitch for cruise control (cf. Section VI-A). With T not necessar- The procedure requires at least one (va (i), γ(j)) to be solved
ily unique (overactuation, see Section II-B), selection from a set in advance, its IG is provided manually. If the grid points
of feasible trims follows an optimization which regards the use (va (i), γ(j)) are spaced close enough and assuming sufficient
of the trims for feed-forward actuation in cruise control. This smoothness of the optimal trim-map, this procedure provides
differs from the approach in [9], where θt is imposed to render IGs which are already close to the actual solution. Further, it
the trims unique—this potentially constrains the solution space, fosters propagation of good solutions through the map: though
i.e., the extent of the assessable flight envelope. Furthermore, in not guaranteed, a globally optimal solution at (i, j) might
[9] T is entirely based on wind-tunnel data, i.e., no model or render locally optimal solutions in (i + m, j + n) globally
optimization is involved. optimal as well.
Authorized licensed use limited to: Khalifa University. Downloaded on September 19,2023 at 07:19:32 UTC from IEEE Xplore. Restrictions apply.
ROHR et al.: ATTITUDE AND CRUISE CONTROL OF A VTOL TILTWING UAV 2687

in the attitude to a desired body angular acceleration ω̇ des us-

ing a cascaded P-PID-structure. It is already implemented in the
PX4 autopilot software for multirotor control and thus adopted
for the present work, [16]. An in-depth outline and analysis of
the controller is given by the respective authors in [24]. Due
to the tail-propeller generating only upwards forces (negative
pitch-moments) and the CG being close to the midpoint between
the main-propellers, a non-symmetric pitch authority is present
around hover: to improve pitch-response, a pitch-error depen-
dent gain-scheduling is therefore included in this flight-phase,
i.e., selecting weaker pitch-gains when pitching down to reduce
overshoot.

B. Dynamic Inversion
Fig. 4. Trim-maps obtained from the steady-state analysis and optimized for
the criteria outlined in Section IV-A. Shown are the trim-actuation for wing-tilt Given the desired angular acceleration ω̇ des calculated in
ζwt (left), main-propeller throttle δ t
pl,r (middle) and the trim pitch θ (right) as
t
the attitude control block, the total moment Mdes required to
function of airspeed va and flight-path angle γ.
achieve ω̇ des is obtained by rearranging (inverting) the angular
dynamics (1) of the aircraft:

B Mdes = B IB ω̇ des + B ω × B IB ω (13)

with ω the current angular body rate. In general terminology,
Mdes is known as virtual input to the system and, by the above
choice, linearizes the angular system dynamics to ω = ω des
[13]. However, note that in practice, Mdes can be generated only
Fig. 5. Architecture of the proposed attitude control system to track roll-
(φdes ), pitch- (θdes ) and yawrate (ψ̇des ) references. Wing-tilt (δw ) and nom-
approximately due to modeling errors, state-estimation uncer-
n ) are commanded manually or by the CCS.
inal main-propeller throttle (δpl,r tainty, actuator saturation and external disturbances. The overall
Estimated quantities are denoted by a hat. controller must thus be robust enough to compensate for the
resulting errors in the moment.
C. Results
C. Control Allocation
Fig. 4 shows the trim-maps for the wing-tilt angle ζw t
, the
In the last step, the system actuation required to generate
main-throttle setting δpl = δpr =: δpl,r and the aircraft pitch θt
t t t
Mdes is determined. For this purpose, we define the aerody-
obtained by the above described optimization. The trim-throttle
t namic moment Mact to be actuated as
map (δpl,r ) clearly shows that the steady flight-envelope in climb ⎡ ⎤
(γ > 0) is limited by main-throttle saturation. Also note how the lact
maximum possible airspeed va reduces with increasing flight- ⎢ ⎥
B Mact = ⎣ mact ⎦ = B Mdes − B M̂(B ω̂, B v̂ a , u )
n
(14)
path angles. Around hover, the trim-pitch θt is close to zero,
nact
whereas, in cruise-flight, pitch is aligned with the flight-path
angle. This solution is close to the desired trim-pitch θ∗ imposed where M̂ denotes the current estimate of the total aerodynamic
in the optimization (Section IV-A). Overall, the basic operation moment acting on the vehicle with nominal actuation un , defined
of the UAV is well illustrated: for a forward-transition, the wing by δ{al,ar,e,r,tt,pt}
n
= 0, δpl,r
n
and δw as imposed by the pilot or
is gradually tilted down with increasing airspeed and, as soon as
the CCS, Fig. 5. M̂ is based on the aerodynamic model and the
wing-born lift dominates, throttle is mainly used to counteract
estimated state of the UAV. Recalling the overactuated attitude
air-drag in forward flight and, therefore, it can be reduced.
of the vehicle, an approach to distribute the total control effort
Mact among the actuators is required. Due to limited onboard
V. ATTITUDE CONTROL computational power, a full online optimization is not feasible.
In order to stabilize the UAV attitude in all flight phases at a Instead, we thus employ the lightweight heuristic known as daisy
desired setpoint, we develop a model-based attitude controller. chaining [25]. Given a defined order of priority among redundant
Its structure is shown in Fig. 5. The setpoint consists of roll- actuators, this method sequentially allocates the actuators until
(φdes ), pitch- (θdes ) and yawrate (ψ̇des ) references that are pro- the total control effort is achieved. Actuators “further back in
vided manually or by the cruise controller. In the following, we the chain” remain in their nominal state. Fig. 6 illustrates this
present the three main parts of the control system: procedure and shows the choice of priorities among the actuators
and actuator-groups. Use of control surfaces is prioritized since
they are considered more energy efficient than thrust vectoring.
A. Controller
This is found to work well for the available actuators on the
The basic attitude controller is a model-free, non-linear con- UAV: Pitching, e.g., is performed with the elevator in cruise and
trol law based on a quaternion-formulation which maps errors becomes gradually assisted by thrust vectoring at low speeds if
Authorized licensed use limited to: Khalifa University. Downloaded on September 19,2023 at 07:19:32 UTC from IEEE Xplore. Restrictions apply.
2688 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 4, NO. 3, JULY 2019

Fig. 7. The architecture of the proposed cruise control system consists of

two main components: 1) a feed-forward path to set the (approximate) trims
for wing-tilt δw
t , throttle δ t
pl,r , and pitch-angle θ and 2) a feedback loop to
t

stabilize the UAV at the desired velocity va,des by commanding throttle- δpl,rc

and pitch corrections θc . Additionally, a turn-coordination is included in cruise

Fig. 6. The distribution of the total attitude control effort B Mact = which mixes desired roll angle φdes and yaw-rate ψ̇ tc to maintain zero lateral
(lact , mact , nact )T among redundant actuators employs the heuristic daisy acceleration. Estimates are denoted by a hat.
chaining approach. The subscript ()r denotes residual moments left after alloca-
tion of preceding actuators in the chain, the superscript ()Δ indicates differential
actuation. Due to control couplings, differential ailerons/throttle (block 3) and (Section IV-C) and then fed forward to the attitude controller.
tail-throttle/-tilt (block 4) are allocated in groups.
Since the trim-maps describe the steady-state actuation, they are
primarily suited for feed-forward when the airspeed setpoint is
the elevator saturates due to reduced effectiveness. Note that the constant. If, however, a varying setpoint is to be tracked, the
group of actuators on the wing (ailerons and main-propellers) is look-up-velocity (va,LU , Fig. 7) needs to be set such that the
the only source for roll-moment generation. Hence, the roll-axis trims i) remain close enough to the steady-state solution to re-
is not overactuated and, accordingly, not daisy-chained. tain performance of the feedback control-loop but ii) still al-
With the desired control effort assigned to an actuator (or low to perform an unsteady maneuver. The former is required
actuator group), the actual control inputs are obtained by solv- since the feedback control-law (Section VI-B) is based on a lo-
ing the aerodynamic model for the control surface deflections cal linearization of the system. Its performance thus degrades
and propeller speed increments, respectively. The corresponding with ‘off-trim’ feed-forward actuation. At the same time, an
equations are linear and quadratic in the desired variables. The ‘off-trim’ feed forward of the wing-tilt is required1 to accelerate
result is constrained to satisfy actuator limits and then added to or decelerate during transitions. Given a desired airspeed vec-
the nominal actuation δin . In case of control saturation, non-zero tor va,des = (va,x , va,z )T , the look-up in the trim-map (va,LU )
residual control effort is passed on to the next actuator. ±
is thus constrained to the proximity (va,i > 0) of the actual air-
Ailerons and differential main-throttle (block 3) as well as tail- speed vector (va ). For the horizontal (i = x) and vertical (i = z)
tilt and tail-throttle (block 4) are allocated in groups to handle components we require:
the control couplings. Since each group provides two degrees − +
of freedom for moment generation, actuator saturation needs va,i − va,i < va,i,LU < va,i + va,i (15)
to be addressed explicitly: a constrained quadratic optimization This method leads to a trade-off when selecting the bounds
trades off roll- and yaw-moment generation on the wing (block + −
va,i , va,i : If they are set too tight, the aircraft might fail to ac-
3), whereas attaining the pitch-moment on the tail is strictly celerate/decelerate or does so only very slowly. On the other
prioritized over the yaw-moment (block 4). hand, a far ‘off-trim’ situation might arise with the mentioned
Actuation of the wing-tilt and nominal main-propeller throt- performance degradation of the stabilizing feedback controller.
tle is not part of the attitude controller. Instead it is commanded Acceptable values for those bounds are found in flight-testing by
by either the pilot or by the higher-level CCS (Section VI). Fur- gradual relaxation until transitions become possible or feedback
thermore, for the sake of simplicity, the current implementa- control performance degrades—the former is found to occur first
tion of the control system ignores actuator dynamics, i.e., we if the trim-maps are accurate enough.
assume that propeller throttling and control-surface deflections
are immediate. B. Controller
In order to correct for modeling errors corrupting the trim-
VI. CRUISE CONTROL
maps, to attenuate disturbances, and to increase tracking per-
With the cruise control system, operation of the UAV is fur- formance of the desired airspeed vector va,des by providing
ther simplified by allowing the pilot to command a desired the required ‘maneuvering’ forces, an additional, stabilizing
horizontal- and vertical airspeed, thus automating wing-tilt-, feedback control law is inevitable. We use a PID structure to
throttle- and pitch-angle selection. Fig. 7 outlines the basic ar- map velocity errors to desired accelerations and—based on the
chitecture of the cruise-control system and its interface to the mass of the UAV—to corrective forces, respectively. Allocation
attitude controller. of the corrective throttle δpl,r
c
and -pitch θc follows a regular-
ized, weighted least-squares approach based on local control
A. Trim-Map Feed Forward Terms
1 Currently, we control the wing-tilt purely by feed-forward and, since it con-
The strongly and non-linearly varying trims for throttle, stitutes a key actuator for horizontal acceleration in early transition, trim-map
wing-tilt angle and pitch angle are obtained from the trim-maps look up needs to yield an unsteady actuation for acceleration.
Authorized licensed use limited to: Khalifa University. Downloaded on September 19,2023 at 07:19:32 UTC from IEEE Xplore. Restrictions apply.
ROHR et al.: ATTITUDE AND CRUISE CONTROL OF A VTOL TILTWING UAV 2689

derivatives J:
min (Juc − Fc )T W (Juc − Fc ) + (uc )T Kuc
uc
 c  ∂f ∂fx

x
θ
u =
c
, J = ∂fz ∂θ ∂δpl,r
(16)
δpl,r
c
∂θ ∂δ
∂fz
pl,r

with Fc = (fx , fz )T the desired corrective force obtained from

the controller and W, K symmetric, positive definite weighting
and regularization matrices, respectively. The control derivatives
are numerically approximated using finite differences and the
aerodynamic model of the UAV. Contributions of stalled airfoil
segments to ∂fi /∂θ are ignored due to modeling uncertainties.2
The wing-tilt δw is not part of this control law since its dynamics
are very slow compared to the throttle and pitch-response, it thus
remains being fed-forward only (∼ 5 s and ∼ 10 s for fully tilting
up and down, respectively).
The weighted least-squares approach allows to trade off the Fig. 8. Attitude control and tracking of roll (φdes ) and pitch angle (θdes )
reference. Top: during hover, Middle: in transition with ζw ≈ 45◦ and airspeed
realization of horizontal (fx ) and vertical (fz ) corrective forces ∼ 8 m/s, integrators are included on the level of roll- and pitch axis rate control.
using the weighting matrix W. Strong coupling effects between Bottom: during cruise and with airspeed ∼ 20 m/s (wind amounted to ∼ 8 m/s,
the corresponding axes are present in transition where, e.g., pro- presumably turbulent), steady roll-offset is persistent as no integrators were used
in this flight.
peller thrust contributes to both fx and fz . If an actuator saturates
or constraints are set on maximum allowed corrective pitch θt ,
simultaneous control of both axes entails degraded axis-wise
performance in comparison to single-axis control. For overall
safety, we thus prioritize vertical- over horizontal velocity con-
trol and ignore fx during transitions by appropriate scheduling
of W: The fz -weight, wzz , remains constant and wxx is lin-
early ramped up from wxx << wzz to wxx ∼ wzz as airspeed
increases from 12 m/s to 15 m/s (wxz = wzx = 0).

VII. EXPERIMENTAL VERIFICATION

The presented control-system was extensively tested both in
simulation and on the real platform. We performed initial tun-
ing of the attitude-controller gains in hover configuration and
with the UAV suspended by a tether for experimental safety.
Subsequently, we assessed the stabilization and tracking per-
formance in outdoor experiments for all flight phases, includ- Fig. 9. Back-transition with cruise controller commanded to keep zero vertical
velocity vz . Total altitude difference for full transition < 2 m. Bottom plot shows
ing partial- and full transitions of various durations. Finally, the main actuator inputs (normalized) used to control the transition, notice the
the CCS was included and investigated on vertical airspeed combined actuation of the tail-prop and the elevator to generate a negative pitch-
control capabilities. moment (cf. daisy chaining, Section. V-C). The main-throttle is minimal in the
first phase of the back-transition to keep the climb-rate low. The range of nor-
malized throttle-inputs is stretched from [0,1] to [−1, 1] for ease of display (δi∗ ).
A. Attitude Control
Fig. 8 shows roll- and pitch angle tracking in hover, transition from cruise to hover, is depicted in Fig. 9. As seen, pitch is
and cruise flight. Overall, acceptable performance is observed in disturbed in this regime but the pitch error can be kept below
most phases. However, pitch-angle tracking was found to exhibit 5◦ most of the time. Note that roll-axis control is generally less
degradation at wing-tilt angles ζw ∼ 20◦ . There, the free-stream demanding due to the symmetry of the system and thus performs
immersed parts of the wing are just stalled and pitching is about better than pitch.
to require support by the tail-propeller. Strong non-linearities
due to complex aerodynamic effects and modeling errors can B. Cruise Control
explain the observed degradation.
Attitude control during transitions did consider in particular Fig. 9 demonstrates vertical velocity stabilization during a fast
stabilization of the pitch-angle: a fast back-transition, i.e., going back-transition which is generally characterized by highly non-
linear and fast varying dynamics of the UAV: a strong increase in
lift and positive pitch moment is followed by a sudden decrease
2 Close beyond stall-angle, lift- and drag-curves typically exhibit a hysteresis
of the same when reaching stall. On-time throttling is therefore
in reality. Our model ignores this fact which, in turn, is found to cause pitch-
angle instabilities when employed for control-derivative calculation of airfoils key for vertical velocity control. The presented controller is able
close to stall. to maintain altitude within a 2 m band while fully decelerating
Authorized licensed use limited to: Khalifa University. Downloaded on September 19,2023 at 07:19:32 UTC from IEEE Xplore. Restrictions apply.
2690 IEEE ROBOTICS AND AUTOMATION LETTERS, VOL. 4, NO. 3, JULY 2019

REFERENCES
[1] A. S. Saeed, A. B. Younes, S. Islam, J. Dias, L. Seneviratne, and G. Cai, “A
review on the platform design, dynamic modeling and control of hybrid
UAVs,” in Proc. Int. Conf. Unmanned Aircr. Syst., Jun. 2015, pp. 806–815.
[2] S. Verling, B. Weibel, M. Boosfeld, K. Alexis, M. Burri, and R. Siegwart,
“Full attitude control of a VTOL tailsitter UAV,” in Proc. IEEE Int. Conf.
Robot. Autom., May 2016, pp. 3006–3012.
[3] R. Ritz and R. D’Andrea, “A global controller for flying wing tailsitter
vehicles,” in Proc. IEEE Int. Conf. Robot. Autom., May 2017, pp. 2731–
2738.
[4] K. Masuda and K. Uchiyama, “Flight controller design using μ-
synthesis for quad tilt-wing UAV,” in Proc. AIAA Scitech Forum, Jan.
2019, pp. 1–25.
[5] R. G. Mcswain, L. J. Glaab, C. R. Theodore, R. D. Rhew, and D. D. North,
“Greased lightning (gl-10) performance flight research: Flight data report,”
NASA Langley Res. Center, Hampton, VA, USA, Tech. Rep. NASA-TM-
2017-219643, 2017.
[6] M. Sato and K. Muraoka, “Flight test verification of flight controller for
quad tilt wing unmanned aerial vehicle,” in Proc. AIAA Guid., Navigation
Control Conf., Aug. 2013, pp. 1–17.
[7] E. Small, E. Fresk, G. Andrikopoulos, and G. Nikolakopoulos, “Mod-
elling and control of a tilt-wing unmanned aerial vehicle,” in Proc. 24th
Mediterranean Conf. Control Autom., Jun. 2016, pp. 1254–1259.
[8] J. J. Dickeson, D. Miles, O. Cifdaloz, V. L. Wells, and A. A. Rodriguez,
“Robust lpv H∞ gain-scheduled hover-to-cruise conversion for a tilt-wing
rotorcraft in the presence of CG variations,” in Proc. Amer. Control Conf.,
Fig. 10. Vertical velocity control subject to multiple step inputs: A) Reference Jul. 2007, pp. 5266–5271.
and actual vz showing good tracking, error at t = 150 s during a turn with [9] P. Hartmann, C. Meyer, and D. Moormann, “Unified velocity control and
high roll-angle φ ≈ 40◦ which is currently not compensated for. B) Throttle flight state transition of unmanned tilt-wing aircraft,” J. Guid., Control
setting obtained from trim-map δpl,rt and commanded nominal throttle δpl,rn ,
Dyn., vol. 40, no. 6, pp. 1348–1359, Jun. 2017.
the corrective throttle is given as δpl,r
c = δpl,r
n − δpl,r
t , C) as in B but for the [10] T. Ostermann, J. Holsten, Y. Dobrev, and D. Moormann, “Con-
trol concept of a tiltwing UAV during low speed manoeu-
pitch-angle, D) Feed forward trim wing-tilt δw t obtained from the trim-map.
vring,” in Proc. 28th Int. Congr. Aeronautical Sci., 2012. [On-
line]. Available: http://www.icas.org/ICAS_ARCHIVE/ICAS2012/
ABSTRACTS/752.HTM
from 20 m/s in ∼ 6 s. Vertical velocity tracking is demonstrated [11] P. Hartmann, “Predictive flight path control for tilt-wing aircraft,” Ph.D.
in Fig. 10A for maximum vz step-inputs and gradually vary- dissertation, Rheinisch-Westfälische Technische Hochschule Aachen,
Aachen, Nordrhein-Westfalen, Germany, Nov. 2017. [Online]. Available:
ing horizontal velocity (vx ∼ 15 m/s − 3 m/s), showing accurate https://publications.rwth-aachen.de/record/710450/files/710450.pdf
tracking and short response times. Comparing feed-forward and [12] J. J. Dickeson et al., “H∞ hover-to-cruise conversion for a tilt-wing rotor-
corrective values for throttle and pitch reveals the merits of both craft,” in Proc. 44th IEEE Conf. Decis. Control, Dec. 2005, pp. 6486–6491.
[13] H. Khalil, Nonlinear Systems, 3rd ed. Upper Saddle River, NJ, USA:
the trim-map and the feedback controller: For, e.g., δpl,r n
, the Prentice-Hall, 2002, ch. 13.
trim-map contributes up to 80% of the control signal in steady- [14] RC Canadair CL-84 Dynavert by Flyzone. (2019). [Online]. Avail-
state phases while the feedback terms dominate during unsteady able: https://modellflugwelt.de/flugzeuge/modelle/2734/flyzone-cl-84-
dynavert
phases to achieve the fast responses upon setpoint changes. [15] Pixhawk Autopilot Research Project. (Jan. 2019). [Online]. Available:
https://pixhawk.org/
[16] L. Meier, D. Honegger, and M. Pollefeys, “PX4: A node-based multi-
VIII. FUTURE WORK threaded open source robotics framework for deeply embedded platforms,”
in Proc. IEEE Int. Conf. Robot. Autom., May 2015, pp. 6235–6240.
Further steps in the development of the control system con- [17] M. S. Selig, “Modeling propeller aerodynamics and slipstream effects on
small UAVs in realtime,” in Proc. AIAA Atmospheric Flight Mech. Conf.,
sider a proper system identification and/or wind-tunnel testing Aug. 2010, pp. 1–23.
to obtain those parameters of the aerodynamic model which are, [18] P. Martin and E. Salaun, “The true role of accelerometer feed-
for now, only based on typical values from literature. Given the back in quadrotor control,” 2010, unpublished. [Online]. Available:
https://hal.archives-ouvertes.fr/hal-00422423v1
strong dependence of the proposed attitude- and cruise controller [19] W. Khan and M. Nahon, “Improvement and validation of a propeller slip-
on the aerodynamic model, we expect controller performance stream model for small unmanned aerial vehicles,” in Proc. Int. Conf.
to benefit from more accurate parameter values. Furthermore, Unmanned Aircraft Syst., May 2014, pp. 808–814.
[20] X. Ortiz, A. Hemmatti, D. Rival, and D. Wood, “Instantaneous forces
modifications of the tilting mechanism of the wing towards a and moments on inclined flat plates,” in Proc. 7th Int. Colloq. Bluff Body
faster and more reliable actuation would allow to include the Aerodyn. Appl., Shanghai, China, Sep. 2012, pp. 1124–1131.
wing-tilt in the cruise-control feedback-loop. The added control [21] S. F. Hoerner, Fluid-Dynamic Drag. Bakersfield, CA, USA: Hoerner Fluid
Dynamics, 1965.
authority would simplify simultaneous horizontal- and vertical [22] R. K. Heffley and M. A. Mnich, “Minimum-complexity helicopter simu-
cruise control in the transition phase. lation math model,” NASA Ames Research Center, Mountain View, CA,
USA, Tech. Rep. NASA-CR-177476, 1988.
[23] MATLAB Optimization Toolbox, MathWorks, Natick, MA, USA, 2016.
[24] D. Brescianini, M. Hehn, and R. D’Andrea, “Nonlinear quadrocopter at-
ACKNOWLEDGMENT titude control,” Eidgenössische Technische Hochschule Zürich, Zürich,
Switzerland, Tech. Rep., 2013, doi: 10.3929/ethz-a-009970340.
The authors would like to thank Dufour Aerospace [25] T. A. Johansen and T. I. Fossen, “Control allocation–A survey,” Automat-
(https://dufour.aero) for initiating and supporting this project. ica, vol. 49, no. 5, pp. 1087–1103, May 2013.

Authorized licensed use limited to: Khalifa University. Downloaded on September 19,2023 at 07:19:32 UTC from IEEE Xplore. Restrictions apply.