Design and control of an unmanned aerialvehicle for autonomous parcel delivery with

transition from vertical take-off to forward flight-

VertiKUL, a quadcopter tailsitterMenno Hochstenbach, Cyriel Notteboom

Abstract—This paper presents the design and control of VertiKUL, a Vertical Take-Off and Landing (VTOL) transitioningtailsitter Unmanned Aerial Vehicle (UAV), capable of hover flight and forward flight for the application of parcel delivery.Autonomous parcel delivery by air is proven technically feasible, reduces traffic congestion, delivery time and deliverycost. In contrast to existing transitioning UAVs, VertiKUL is not controlled by control surfaces, but exclusively by fourpropellers using differential thrust during hover flight, transition and forward flight. A numerical design method optimisingrange and payload is developed for initial sizing. A dynamic model is implemented in Simulink to evaluate different controlstrategies before conducting test flights. A unique mid-level control strategy enabling intuitive control of VertiKUL whichrequires no pilot skills is developed. Fluent transition from hover to forward flight is achieved through an autonomouscontrol strategy. Attitude control based on quaternions instead of Euler-angles is implemented to avoid singularities.A high-level control system exclusively using GPS waypoints as input is developed enabling fully autonomous parceldelivery. The resulting design, VertiKUL, is built and tested.

Index Terms—VTOL, hybrid UAV, transition, parcel delivery, tailsitter, numerical design optimisation, autonomousnavigation

F

1 INTRODUCTION

The past years UAVs have known anincrease in interest and availability. Theyare widely used in civil applications andareas such as agricultural observation andaerial photography. Another interestingapplication is parcel delivery by air. For thispurpose VertiKUL is designed. Figure 1 showsVertiKUL in hover flight. Its main designfeature is the capability of transitioning fromhover flight to forward flight. Manoeuvrabilityof a quadcopter and efficient forward flightof a conventional airplane are combined intoone hybrid UAV. This is nescessary for parceldelivery applications, where long distances arecovered and tight landing spots are present.A high-level control system is designed to

• M. Hochstenbach and C. Notteboom are M.Sc students at theDepartment of Mechanical Engineering, KU Leuven, Belgium.

accommodate automated parcel delivery. Forintuitive manual control of VertiKUL, a uniquecontrol strategy requiring no flying skills fromthe user is developed. In this thesis onlytechnical aspects were considered, howevereconomical and political considerations shouldstill be investigated.

Figure 1: VertiKUL UAV during hover flight.

There are several ways to perform transitioningmanoeuvres: thrust-vectoring, tilting-rotor,tilting-wing, etc. The most famous transitioningmanned vehicles using thrust-vectoring andtilting-rotor are Harrier and V-22 Ospreyrespectively. Examples of unmanned tilt-rotor vehicles are Wingcopter and FireFLY6.These concepts involve significant extramechanical complexity which increases cost,maintenance and risk of failure. The explicitchoice is made to avoid the use of such tiltingmechanisms. This means VertiKUL purelyrelies on differential thrust created by thefour propellers to perform a transition and tocontrol the UAV in both hover and forwardflight. This concept can be regarded as tilting-body. Other examples of unmanned vehiclesthat transition using differential thrust onlyare Quadshot and ATMOS UAV (prototype).However, these UAVs still have tilting rotorsor actuated control surfaces to control thevehicle in level flight. With respect to theabsence of additional actuators few similarprojects were found in literature.

The design of VertiKUL consist of a conceptualphase, in which different configurations arecompared, and a numerical optimizationdesign strategy for initial sizing. The finalsizing and performance estimates are given insection 5. In section 3 a dynamical modelof VertiKUL is derived. This model isimplemented in Simulink. Using Simulink, acontrol strategy is developed and tested insimulation flights. The control of VertiKULconsists of a low-level, a mid-level and high-level controller which provide respectivelyacrobatic, intuitive and autonomous flightcapabilities. The control strategy is explainedin section 4. Section 5 presents the realizationof VertiKUL. Based on test flights with differentprototypes the design and control strategy areevaluated.

2 DESIGN OF VERTIKUL VTOL UAV

VertiKUL has a unique fixed wing quadcopterhybrid design, capable of both VTOL and ef-ficient forward flight. The design has the ma-jor innovation of having no additional controlsurfaces for attitude control and stability. Thisgreatly reduces cost, maintenance and risk offailure. VertiKUL is capable of transportinga payload up to 1 kg and is optimised forrange. The overall design process of VertiKULis shown in figure 2. First, a configuration bestmeeting the requirements is designed. Next,initial sizing is done using a numerical ap-proach optimising the range. After analysing,the proposed solution is designed in detail.Finally, tests with several prototypes are doneto confirm the analyses and design.

2.1 Configuration

VertiKUL has a fixed low wing configuration.Four fixed-pitch propellers provide lift duringhover flight and thrust during level flight.They also provide attitude control by exertingmoments on the frame using differentialthrust. The payload is placed at the centre ofgravity of VertiKUL to ensure the same flightcharacteristics for all payload masses.

Market analysis

Designrequirements

Configurationdesign

Detaildesign

PrototypingFlighttests

Componentmodels

NumericalDesign

Optimization

Figure 2: VertiKUL design process.

2

FixedhlowhwinghdesignHighhcapacityLiPohbattery

Reliablepropulsionhsystem

Parcelhuphtoh1hkg100x150x200hmm

Inclinedhwinglets

Low-complexitypassivehwings

Figure 3: VertiKUL design features.

Figure 3 shows the following design features:• Fixed low wing tailsitter design• Passive wing without control surfaces re-

ducing complexity• Four fixed-pitch propellers• High capacity rechargeable lithium poly-

mer batteries for long range• Rear-end opening for payload insertion• Winglets for directional stability and re-

duced drag during forward flight

2.2 Numerical design optimisationIn the numerical design optimisation, range ismaximized with given constraints. The rangeof an electric propeller-driven aircraft can beexpressed as:

R =kbatg

mfbat︸ ︷︷ ︸battery

· CLCD︸︷︷︸

aerodynamics

· ηpropulsion︸ ︷︷ ︸technology level

(1)

kbat: gravimetric energy density battery [J/kg]mfbat: mass fraction battery [-]

CL/CD : aerodynamic efficiency [-]ηpropulsion: propulsion group efficiency

(ESC, motor, propeller) [-]

Solutionset

Best?solution

Designspecifications

Componentmodels

FilterDesign

constraints

Calculaterange

Iterativealgorithm

Combine

Propellerdatabase

Wing?profile?database

Payload?

Cruise?speed

OK?

Primarydesign?parameters

add

Figure 4: Design optimisation approach.

Range is maximized when flying at maximumlift to drag ratio. This occurs at minimum dragspeed.

VertiKUL was designed following a numericalmethod written in Matlab. The proposedmethod combines models and databases ofcomponents and selects the best solutiontaking all constraints into account. Figure4 shows the general design approach. Thefiltered solution set is shown in figure 5.The best solution for 1 kg payload formsthe starting point for the detailed design ofVertiKUL which is described in section 5.

0 10 20 30 400

1

2

3

VertiKUL

R [km]

mpl

d[k

g]

Figure 5: Filtered solution set showing the bestsolutions for different payloads.

3

Fquad

Eqyφq

Eqz

Eqx

ψq

θq

Fplane

Epy

φp

Epz

Epx

ψp

θp

Fworld

Ewy

Ewx

Ewz

Figure 6: Quadcopter body frame Fquad, plane body frame Fplane and world frame Fworld.

3 MODELLING AND SIMULATION

Figure 6 shows two reference frames to rep-resent the attitude of VertiKUL. A first frameFquad(E

qx, E

qy , E

qz) is used to represent the UAV

in quadcopter mode, when it’s in the hoveringflight regime. A second frame Fplane(E

px, E

py , E

pz )

is used to represent the UAV in plane mode,when it’s in the forward flight regime. Thisway it is possible to think intuitively aboutVertiKUL’s attitude, i.e. roll, pitch and yaw,while in quadcopter mode as well as in planemode. The origin of both frames coincides withthe centre of gravity of VertiKUL. The attitudeexpressed in Fquad is represented by the Eulerangles roll φq, pitch θq and yaw ψq. Expressedin Fplane the attitude is represented by φp, θpand ψp.

3.1 General Equations of MotionThe general equations of motion expressed inthe quadcopter body frame Fquad are obtainedbased on Newton Euler formalism. They arepresented mathematically in equation 2.[m13 00 qI

] [q~v

q~ω

]+

[q~ω ×m q~v

q~ω × qI q~ω

]=

[∑q~F∑

q~M

](2)

where q~v and q~ω are the linear and angularvelocities respectively. The external forces andmoments are represented by q

~F and q~M . Index

‘q’ is used to indicate vectors are expressed inquadcopter body coordinates. External forcesconsist of gravity, aerodynamic and thrustforces. Similarly, external moments consist ofaerodynamic and differential thrust moments.

Matrix equation 2 includes the translationaland rotational equations of motion which arepresented by equation 3 and 4 respectively.

mtot qv = −q~ω ×m q~v + q~W + q

~Lwing

+q~Dwing + q

~Dbody + q~Ttotal

(3)

qI q~ω = −q~ω × qI q~ω + q~Mprop,gyro + q

~Mdrag

+q~Mwing + q

~Mwinglet + q~Mψ

+q~Mφ + q

~Mθ + q~Mψ

(4)

The total mass of VertiKUL mtot includes 1.6kg battery mass and 1 kg payload. Gravityand collective thrust are defined using q

~Wand q

~Ttotal. Aerodynamic forces included in thismodel are lift and drag. Lift is represented byq~Lwing and drag consists of wing drag q

~Dwing

and body drag q~Dbody. The lift and drag coef-

ficients of the wing are estimated using Xfoil.The body drag coefficient is determined exper-imentally. qI is the inertia matrix of VertiKULwith respect to its centre of gravity expressed inFquad. It is assumed the axes of Fquad are princi-pal axes such that qI is a diagonal matrix withelements Ixx, Iyy and Izz. Angular body drag isrepresented by q

~Mdrag and q~Mprop,gyro denotes

the propeller gyroscopic effect. Other angularaerodynamic effects included in this modelare wing moment q

~Mwing, stabilizing momentq~Mwinglet due to the winglets and rolling mo-

ment q~Mψ due to small differences in lift and

drag between the left and right wing. q ~Mφ, q ~Mθ

and q~Mψ denote roll moment, pitch moment

and yaw moment created by differential thrustof the propellers.

4

3.2 Propeller dynamics

In the simulation program a lag in collectiveand differential thrust is added to incorpo-rate the effect of propeller inertia. This lag isrepresented by a first-order system with timeconstant τ .

3.3 Quaternion representation

Although Euler angles provide an intuitiveway to represent VertiKUL’s attitude it suffersfrom singularities at a pitch angle of 90◦ or -90◦. Because in forward flight attitude controlis done close to such a singularity (θq = −90◦)the quaternion attitude representation is intro-duced. A quaternion is a set of four parame-ters and offers a singularity free mathematicalrepresentation of the attitude.

3.4 Simulation

The dynamical model is implemented in aSimulink simulation program to be able to testcontrol strategies with a virtual UAV in simu-lation flights. Figure 7 shows the main compo-nents of the simulation program. The first twoblocks represent the control which consists of areference builder and a controller. The referencebuilder generates reference signals out of theuser’s joystick input. Depending on the controlmode the controller regulates the angular ratesin low-level control, heading and altitude inmid-level control and finally position in high-level control. The third block represents thedynamic model and simulates the position andattitude at every time step. The forth blockvisualises the virtual UAV in a 3D environmentusing ‘Simulink 3D Animation’.

Control

Referencegenerator

Feedback loop

Controller Dynamics Visualisation

Figure 7: Structure of Simulink simulation pro-gram.

4 CONTROL

VertiKUL can be controlled in a low-level, mid-level or high level mode which provide ac-robatic, intuitive and autonomous flight capa-bilities respectively. As can be seen in figure7, each mode has its own reference genera-tor which translates user input into referencesignals and controller. The reference generatorand controller are designed using the Simulinkprogram and are implemented on a Pixhawkautopilot to control VertiKUL. The developedPixhawk’s firmware is based on ArduCoptersource code and is called ArduVTOL. Figure 8presents a diagram of the complete high-levelcontrol mode and shows how the high-levelcontroller is built from the mid-level and low-level controller.

4.1 Low-level Control ModeIn low-level control mode the transmitter’schannels correspond to desired collective thrustT , roll rate φd, pitch rate θd and yaw rate ψd. Thelow-level controller consists of three separatePID-controllers that translates desired angularrates into actuator moments that need to beinvoked by the propellers through differentialthrust. Only experienced UAV pilots can fly inthis low-level control mode.

4.2 Mid-level Control ModeBecause piloting VertiKUL in low-level moderequires a lot of practice a unique mid-levelcontrol mode is developed such that anyuser without experience can pilot VertiKULsafely and intuitively. A distinction is madebetween two flight modes. Quadcopter modecorresponds to VertiKUL hovering like aconventional quadcopter and plane modecorresponds to the forward flight regime. Inmid-level control mode a smooth transitionfrom hover to forward flight or back iscommanded by a switch. During a transitionto forward flight the control system ignorespilot input and decreases pitch gradually untilthe stall speed has been exceeded and optimalangle of attack has been reached. Transitionback to hover can be performed faster becauseno forward speed needs to be built up.

5

High-Level=control

Reference=generator

Mid-level=controller

Feedback=loop

Altitude=controller

Low-level=controllerAttitude=controller

Navigation=controller

High-level=controller

Scheduler

mission==={===take-off;===waypoint_1;===waypoint_2;===...===land;}

Figure 8: High-level control mode diagram with high-level reference generator and controller.

In quadcopter mode the transmitter’s channelscorrespond to desired climb rate hd, roll angleφd, pitch angle θd and yaw rate ψd. Desiredaltitude hd and yaw angle ψd are obtainedby integrating the desired climb rate andyaw rate input respectively. This way drift onaltitude can be compensated and heading canbe maintained, i.e. heading-lock. The altitudecontroller converts an altitude error into asupplementary climb rate which is added tothe user’s commanded climb rate hd. Thiscombined reference climb rate is convertedinto a reference climb acceleration which isregulated by a PI-controller that outputs adesired collective thrust. A feedforward termequal to VertiKUL’s weight is added.

In plane mode only two transmitter channelsare used to command climb and turning rate.Desired yaw angle ψd and altitude hd are againobtained by integrating the their rates. Withboth transmitter sticks centred VertiKUL willmaintain altitude and heading autonomously.Forward speed is not regulated. Two altitudecontrol approaches are presented.

4.2.1 Thrust altitude controlA first approach uses the same controller as inquadcopter mode but with PI-gains optimizedfor forward flight. Also a feedforward termequal to steady-state thrust in forward flightor climb is added. This term is estimated fromstatic equilibrium. Reference pitch, expressed

in Fplane, equals optimal angle of attack minusrigging angle and plus climb angle. The advan-tage of this strategy is that optimal angle ofattack will always be maintained. This meansthat even with a different mass VertiKUL willfly most efficiently. A disadvantage are the pos-sible variations in collective thrust. Also, dur-ing descend the thrust can drop significantlywhich reduces the angular controllability bydifferential thrust. Furthermore, when angle ofattack becomes negative, thrust will amplify analtitude error instead of reducing it.

4.2.2 Pitch altitude controlIn a second approach a PI-controller regulatesthe combined reference climb rate and outputsa reference pitch. This pitch is supplementedwith the same feedforward as in the firstapproach. Thrust is now solely determined bythe feedforward added in the first approach.This strategy ensures a smooth collectivethrust output. Another advantage is that whenlosing altitude the pitch will be commandedto rise and VertiKUL approaches a safe hoverregime where collective thrust will be able tocompensate gravity. One disadvantage is thatoptimal angle of attack is not always ensured,especially not with changing weight. However,weight can be estimated during hover flight tooptimize the feedforward terms.

Whether in quadcopter mode or in planemode the desired roll, pitch and yaw angles

6

φd, θd and ψd are regulated by a quaternion-based attitude controller which outputsreference angular rates expressed in Fplane.Experimental evaluation of the mid-levelcontrol mode, using thrust altitude control, ispresented in section 5.3.

4.3 High-level ControlA high-level control mode is designed todeliver packages fully autonomously. Insteadof using a transmitter to control VertiKULdirectly, waypoints are programmed into amission plan. As can be seen in figure 8, amission plan consists of a ‘take-off’ command,waypoints and a ‘land’ command. A missionscheduler provides the waypoint coordinatesto the high-level controller at appropriatetimes. The navigation controller translatestarget coordinates (longitude, latitude andaltitude) into reference signals sent to themid-level controller and replaces the user’sinput in the mid-level control mode. Basedon the distance between current locationand commanded waypoint the navigationcontroller uses three control strategies to guideVertiKUL to the waypoint.

4.3.1 Forward flight navigationWhen the distance to the waypoint is greaterthan 100 m a transition to forward flight iscommanded. Altitude is regulated to match thedesired altitude. Desired heading is calculatedfrom the difference in longitude and latitudecoordinates of desired and current position.

4.3.2 Hover flight navigationWhen the distance to the waypoint drops be-low 100 m, transition back to hover flightis commanded. A desired forward speed isproportional to the distance to the target andregulated using a PD-controller which outputsa desired pitch angle. Desired roll angle is keptzero. Desired yaw is calculated the same wayas during forward flight navigation.

4.3.3 Precision navigationWhen the distance to the waypoint is smallerthan 20 m desired yaw is kept constant.

Roll and pitch are commanded separately tominimize the distance to the waypoint alongthe transverse- and longitudinal-axis of Fquadrespectively.

Only discrete information on target position isused to navigate to the waypoints. The resultof a simulation test flight in high-level controlmode is shown in figure 9. Here the virtualUAV takes off to an altitude of 10 m. Thenavigation controller then receives a targetwaypoint of 500 m longitude and 1000 mlatitude. A transition is initiated and after thetransition the heading is changed. The targetaltitude is changed during flight. Changesto hover flight navigation and precisionnavigation are also indicated. Finally thevirtual UAV lands.

Take-off

10m

Transition Climb

Descent

Precision Nav.

Land10m

50m50m

30m

20m

10m

Transition

Command

Altitude

x

y

Figure 9: Simulation of high-level control.

5 REALISATION AND TESTING

The design of a non-conventional aircraft re-quires a trail-and-error approach where func-tionality increases with every iteration. It isafter all not possible to base the design on al-ready existing aircraft. Therefore, six fully func-tional prototypes are built and tested (figure10). Based on the result of the numerical designoptimisation approach described in section 2the final VertiKUL prototype is built. Table 1shows its design parameters and performance.

5.1 StabilityFor VertiKUL to be longitudinally stable, thecentre of gravity lies with a distance of 5%of the mean aerodynamic chord in front ofthe aerodynamic centre. Figure 11 shows theinternal component arrangement.

7

VertiKUL

X2000

X3000X1000

MC2000VertiKUL mini

Figure 10: The different prototypes built during the thesis.

Winglets and wing sweep are added fordirectional stability. For structural reasons,the wing has no dihedral. However, wingletdihedral is used to accommodate for rollstability. As can be seen in figure 10, VertiKULhas inclined propellers. Thus sacrificing thrustbut gaining sideways force for better yawcontrol.

5.2 General structureMainly carbon rods are used in the wing andfuselage for strength. A laser-cut multiplexstructure is built around the carbon frame forpayload and avionics support. The hull is madeout of kraft paper reinforced polystyrene, cutwith hot wire. A strong yet light sandwichstructure from polystyrene and balsa wood isused for the wing and stabilisers.

payload

battery

avionics

motor

motor

c.o.g. a.c.

Figure 11: VertiKUL internal component ar-rangement.

5.3 Test flights

To validate the mid-level controller, ArduVTOLsource code is tested and evaluated with pro-totypes X2000 and X3000.

5.3.1 Transition flight

Figure 13 shows commanded and measuredpitch angle, GPS ground speed and com-manded and measured altitude during the testflight. During transition, angle of attack de-creases gradually until a pitch set point isreached and forward speed increases whilekeeping altitude constant. During forward,climbing and descending flight altitude is reg-ulated with thrust. While climbing with a con-stant climb rate, pitch angle increases. Transi-tion to hover occurs fast and with little over-shoot in pitch angle. Figure 12 shows VertiKULduring transition.

Figure 12: VertiKUL during transition.

8

− 60

− 40

− 20

0

20

0

2

4

6

8

0 5 10 15 20 25 30 358

10

12

14

16

18

1GCHoverCflight5GCDescendingCflight

2GCTransition6GCForwardCflight

3GCForwardCflight7GCTransitionCandChover

4GCClimbingCflight

1 2 3 4 5 6 7

Pitc

hCan

gleC

[º]

Alti

tude

C[m]

TimeC[s]

Gro

undC

spe

edC[m

/s]

Altitude:CommandedMeasured

PitchCangle:CommandedMeasured

GroundCspeed

Figure 13: Altitude, GPS ground speed andpitch angle during transition test flight.

5.3.2 Turning flightFigure 14 shows the 3D circular flight path.A clockwise 360° turn is done in mid-levelcontrol mode. Altitude and turning rate stayconstant through differential thrust regulationas can be seen in figure 15. Because no GPSsignal was used for control, wind causeddrifting. The end point of the circle thereforeshifted relative to the starting point in thedirection of the wind.

Start

Stop Land

Wind

Figure 14: 3D circular flight path of the turningflight test.

0

200

400

0 10 20 30 40 50− 10

0

10

Alti

tude

F[m]

Yaw

Fang

leF[º

]

TimeF[s]

YawFangle:CommandedMeasured

1)FForwardFflight 2)FLevelFturningFflight 3)FDescendingFturningFflight

4)FTransitionFandFhoverFflight

1 2 3 4

Altitude:CommandedMeasured

Figure 15: Reference- and measured yaw angleand altitude during the turning test flight.

battery1608 g

payload1000 g

body776 g

wing300 g

propulsion

1064 g

avionics

111 g

Mass distributionVertiKUL

Parameter Valuemtot MTOW 4.86 kgmpld Payload 1 kgR Range 26 kmttot Endurance 29 minvc Cruise speed 15.5 m/svstall Stall speed 12.9 m/s- Wing airfoil NACA23012b Wing span 1.60 mS Wing area 0.36 m2

AR Aspect ratio 7.11ε Cruise lift to drag ratio 7.5CL Cruise lift coefficient 0.9- Thrust to weight ratio 2.17- Power to weight ratio 610 W/kg- Wing loading 13.5 kg/m2

- Propellers APC TE 9x5”- Battery 6S2P 10,000 mA hPmax Max. single motor power 740 W- Motor kv-value 970 min−1 V−1

Table 1: Design parameters and performance ofVertiKUL.

9

6 CONCLUSIONS AND FUTURE WORK

In this paper the design and control ofVertiKUL was described. For initial sizingof VertiKUL, a numerical design approachoptimising range was developed. In thedynamic model only the relevant aerodynamiceffects were considered. However, estimatingrealistic aerodynamic coefficients proved tobe difficult. Two control strategies for altituderegulation in mid-level control mode wereproposed. By regulating the altitude withpitch angle, a smooth thrust output is ensured.Also, when losing altitude the pitch will becommanded to rise and VertiKUL approachesa safer hover regime. However, only test flightsusing the thrust strategy were conducted up tillnow. Sufficient duration of transition appearedto be necessary to ensure stall speed isexceeded. Six fully functional prototypes werebuilt and successful transitional flight wasachieved. It was possible to perform transitionand forward flight without using additionalactuators and control surfaces. Intuitivemid-level control was demonstrated withautonomous transitional, turning, climbingand descending flight. In test flights, windgusts turned out to have a big influence onhover stability. To overcome this problem,heading lock in quadcopter mode was relaxedand propellers were inclined for better yawcontrol.

This paper proved the technical feasibilityof parcel delivery by air. However, otherfields still need to be investigated. Mostimportant are obstacle avoidance, publicsafety regulations, political and economicalaspects. Also technical improvements can bemade. The numerical design program canbe expanded for more general configurationsand with updated component models. Also, itwould be interesting to replace the simulationmodel in Simulink by a more advanced flightsimulator, for example X-Plane. CFD analysisand wind tunnel tests can be done for amore precise estimation of the aerodynamiccoefficients. To evaluate control strategies, testflights were performed at high angle of attack.However, to evaluate forward flight efficiency,

tests at optimal angle of attack should still bedone. Also, pitch-regulated altitude controlstill needs to be evaluated in test flights. Thestability advantage of a reflex profile shouldbe re-evaluated because they suffer fromlow CL,max values. Lower stall speeds canbe obtained with non-reflex profiles, makingtransition possible at lower speeds. Also,the use of multi-blade propellers should beinvestigated because propeller diameter isrestricted by VertiKUL’s configuration type.To reduce wind disturbances in hover flight,wing area can be reduced when using partof the propeller thrust for lift during forwardflight.

10