Field and Service Robotics - Corke P. and Sukkarieh S.(Eds) Part 12 doc

The control designfor the T-Wing is complicated by the large differences in vehicle dynamics betweenvertical and horizontal flight; the difficulty of accurately predicting the low-speedvehic

3 Navigation and Control System

In order to reach the goal of the project, the autopilot design phase followedthose principles:

• select components not only according to criteria of precision and resolution,but as well of weight and power consumption to be suitable for the targetedapplication

• use as much as possible digital output and calibrated sensor to reduce thedevelopment time and avoid additional need of A/D converter, interfacemicrocontroller, etc

• interface the sensors so that the central processor doesn’t have to wait onthem but can access directly and rapidly to the information on request.This applies for example to the GPS

3.1 Computer and Interfaces

Sky-Sailor will fly autonomously using an onboard autopilot, only high levelorders being given from the ground The system is mainly based on a singleboard computer, the X-board <861> which is a compact embedded PC designfor low power consumption

Fig 3 X-board <861> single board computer from Kontron

It includes a Geode SC1200 Processor, up to 128 Mbyte of DRAM and up

to 128 Flash storage media on board Despite the compact size of a businesscard, it offers a lot of interfaces: integrated Graphics, Ethernet, USB, RS 232,

I2C, audio The OS running on it is a reduced Linux distribution, based onDebian, that only contains the necessary features

3.2 Sensors

In Fig 4, one can see the power generator system and the autopilot, with allsensors and their interfaces to the X-board

Fig 4 Schematic view of the power and control parts of Sky-Sailor

Attitude

The attitude and angle rate of the airplane are given by the MT9-B IMU

7 at a frequency of up to 512 Hz Such a low-cost sensor is perfectly cient to perform inertial navigation compared to heavier one [6] It containsaccelerometers, magnetometers, gyroscopes and communicates through serialport (RS232) with the X-board on which data fusion is executed In the futureversion of this device, the sensor fusion will be done by a DSP chip, reducingthe computational cost on the central processor of the autopilot

suffi-VGA camera

One direction of the project is to achieve autonomous navigation based onvision, using SLAM techniques as shown in [1] [2] One or more lightweightVGA cameras will give 680 x 480 images of the landscape and allow localiza-tion and mapping of the terrain Efforts will be done in this direction in thefollowing month Cameras are connected to the central processor via USB.Absolute x-y position and altitude

The absolute position is given by an ultra low power GPS sensor with patchantenna from Nemerix This sensor consumes only 61 mW for a weight of 12.36

gr In terms of position accuracy, 95 % / 99.7 % of the time, the estimatedposition lies within 2.7933 m / 4.2028 m respectively of the actual position

7 Inertial Measurement Unit

Design of an Ultra-lightweight Autonomous Solar Airplane 445

A future version will accept WAAS/EGNOS correction for more precise surements The data are sent on a serial port at a fixed rate of 1 Hz to amicrocontroller that decodes the NMEA protocol, stores the value internallyand sends them on demand to the main processor via I2C.

mea-The same microcontroller interfaces the altitude pressure sensor MS5534.Pressure and temperature values, as well as four calibration factors allow thecomputation of the altitude with a resolution of 1 m The relation betweenpressure and altitude being variant with the atmospheric condition, the mi-crocontroller will achieve data fusion, using the GPS altitude as an absolutevalue to correct the drift of the MS5534

Airspeed

The airspeed sensor DSDX is a differential pressure sensor, with digital I2Creadout and temperature compensated It is connected to a Pitot tube fixed

at the attack border of the wing

3.3 Ground Control Station

The control of the airplane is executed onboard but there is a link to a groundcontrol station through a serial radio modem that allows a baudrate of 9600bps The goal is to:

• download and upload airplane and control parameters, but as well theflight plan, before the takeoff,

• get a visual feedback of the state of the airplane once airborne, modifyflight plans on-the-fly,

• retrieve and record the telemetry for flight analysis, system identification,etc

Fig 5 Ground control station and it’s graphical user interface

The GUI8was developed with QT graphical libraries under Linux (Fig.5)

It is composed of three main layers which ensures modularity:

8 Graphical User Interface

• the graphical interface, that allows a visual overview of the state of theairplane and its position on a 3D map of the terrain.

• a second layer which processes data and control the GUI

• a communication module that receives and sends the data in packets tothe airplane through the serial port connected to the radio modem.Control of the airplane from the ground

As shown in Fig 4, the commands given to the servos can come from the topilot or a human pilot on the ground using an RC transmitter The ”servoboard” decodes the PPM9 from the RC source and get the value given bythe autopilot through the I2C bus Based on one additional channel on the

au-RC remote, it switches from one source to the other It is also possible, forcontrol tuning purpose, to mix sources and, for example, allow the autopi-lot to command only the elevator while the other actuators are commandedmanually

3.4 Autopilot Design Results

The final design leads us to a navigation and control system with a total mass

of 140 g for a consumption of around 4 W One can see that 6/8 of the power

is used by the X-board and 1/8 for the transmission, the rest being used bythe sensors

Table 1 Autopilot power and mass distributionPart Weight [g] Power consumption [W]

Altitude sensor board 2 0.03

Airspeed sensor board 3 0.03

9 Pulse Period Modulation

Design of an Ultra-lightweight Autonomous Solar Airplane 447

4 Simulation of the Solar Flight

For the validation of a long endurance solar flight, a simulation was ized under Matlab Simulink Fig.6 represents the schematic of the model thatincludes first the irradiance model based on [12] and depending on the geo-graphic position, time and solar panels orientation We then take into accountthe surface of solar cells, their electrical efficiency and the efficiency of the con-nection configuration For the MPPT, the electrical and algorithm efficienciesare taken into account The power consumption is the addition of the autopi-lot power and the power needed for flight, which was measured in the case oflevelled flight and climbing phase Depending on the irradiance conditions andthe consumption, the battery is charged or discharged, taking into account theefficiency of the energy transfer

real-Fig 6 Schematic of the simulation model under Matlab Simulink

4.1 Study of Various Scenarios

The simulation environment allows to test different flight strategies in order

to accomplish a long endurance flight and analyze the benefit of a climbingphase or the influence of the other parameters on the feasibility of a multi-daysflight We will present here two scenarios

In the first simulation, Sky-Sailor starts a flight at EPFL location on the21th of June with an empty battery, keeping always the same altitude The twographs below show the evolution of the power distribution during 48 hours.With good sun conditions, the battery is fully charged at 13h30 At thismoment, the MPPT measures that the battery voltage reaches the maximum

Fig 7 Power distribution on Sky-Sailor during levelled flight

Fig 8 Battery charge/discharge current and energy during levelled flight

voltage of 33.7 [V] and adapts the maximum power point to avoid overcharge

In this phase, the total amount of energy that is not used but that could

be retrieved from the solar panels reaches 92.5 [Wh] During the night, thebattery supplies the all airplane but at 5h10 it is totally discharged

Another strategy is to better use the energy after the battery charge byincreasing altitude Fig 9, 10 and 11 show the same scenario presented beforebut with a climbing phase until 2000 [m]

Basically, Sky-Sailor uses the additional energy to gain altitude at 0.3[m/s] using an electrical power of 40 [W] Having reached 2000 [m], it stays atthis altitude until the energy is not sufficient anymore for levelled flight Atthis point, the motor is turned off and the descent starts Finally, at the mostcritical point at 6h13 in the next morning, the battery still has a capacity of4.7 [Wh] and the charging process starts again Globally, the unused energyduring the day is 61.5 [Wh]

Design of an Ultra-lightweight Autonomous Solar Airplane 449

Fig 9 Power distribution on Sky-Sailor during flight with climbing phase

Fig 10 Altitude during flight with climbing phase

Fig 11 Battery charge/discharge current and energy during flight with climbingphase

5 Status of the Project and Future Work

The mechanical structure of the airplane is actually ready, it has been fully tested and validated in terms of power and stability The solar generator,composed of the solar modules and the MPPT, is in the integration phase onthe wing

success-Concerning the autopilot, the different parts of the system are being sembled and all functionalities will be tested during the first half of this year

as-In the summer, we should have achieved many flights and experiments toclearly evaluate the capabilities of our UAV

6 Potential Applications

Small and high endurance UAVs find uses in a lot of varied fields, civilian ormilitary The civil applications, leaving side the military ones, could includecoast or border surveillance, atmospherical and weather research and predic-tion, environmental, forestry, agricultural, and oceanic monitoring, imagingfor the media and real-estate industries, and a lot of others The target mar-ket for the following years is extremely important [11]

The great advantages of Sky-Sailor compared to other solutions would bewithout any doubt its capability to remain airborne for a very long period, itslow cost and the simplicity with which it can be used and deployed, withoutany ground infrastructure for the lunch sequence

As an example, in the hypothetical case of forest fire risks during a warmperiod, a dozen Sky-Sailor, easily launched with the hand, could efficientlymonitor an extended surface, looking for fire starts A fast report would allow

a rapid intervention and thus reduce the cost of such disaster, in terms ofhuman and material losses

Sky-Sailor would be as well a very interesting platform for academic search, in aerodynamics or control

re-7 Conclusion

In this paper, the design of an ultra-lightweight UAV was presented, includingdetails about it’s mechanical structure, the solar generator and the autopilotsystem The approach adopted doesn’t aim only at building an efficient autopi-lot, but also keeps in mind it’s future application This is done by designingand selecting all the parts to obtain a lightweight and low-power airplane Weplan to perform the first experiments with the autonomous airplane duringthe first half of this year and a long endurance flight this summer

Design of an Ultra-lightweight Autonomous Solar Airplane 451

The authors would like to thank all the people who contributed to the nition study, Samir Bouabdallah for fruitful discussions and advices on flyingrobots, Walter Engel for the realization of the mechanical structure and allthe students who worked or are working on this project

defi-References

1 Davison A J (2003) Real-time simultaneous localization and mapping with asingle camera, IEEE Int Conf on Computer Vision, ICCV-2003, pp 1403-1410,Nice (France), October 2003

2 Lacroix S, Kung I K (2004) High resolution 3D terrain mapping with low tude imagery, 8th ESA Workshop on Advanced Space Technologies for Roboticsand Automation (ASTRA’2004), Noordwijk (Pays-Bas), 2-4 Novembre 2004

alti-3 Eisenbeiss H (2004) A mini unmanned aerial vehicle (UAV): system overviewand image acquisition, International Workshop on ”Processing and visualizationusing high-resolution imagery” 18-20 November 2004, Pitsanulok, Thailand

4 Kim J.-H, Sukkarieh S (2002) Flight Test Results of GPS/INS Navigation Loopfor an Autonomous Unmanned Aerial Vehicle (UAV), The 15th InternationalTechnical Meeting of the Satellite Division of the Institute of Navigation (ION)24-27 September, 2002, Potland, OR, USA

5 Kim J.-H, Wishart S, Sukkarieh S (2003) Real-time Navigation, Guidance andControl of a UAV using Low-cost Sensors In International Conference of Fieldand Service Robotics (FSR03), Japan, July 2003

6 Brown A K, Lu Y (2004) Performance Test Results of an IntegratedGPS/MEMS Inertial Navigation Package, Proceedings of ION GNSS 2004,Long Beach, CA, Sept 2004

7 Atkins E M et al (1998) Solus: An Autonomous Aircraft for Flight Controland Trajectory Planning Research, Proceedings of the American Control Con-ference, Pennsylvania, June 1998

8 Johnson E N et al (2004) UAV Flight Test Programs at Georgia Tech, ceedings of the AIAA Unmanned Unlimited Technical Conference, Workshop,and Exhibit, 2004

Pro-9 Granlund G (2000) Witas: An intelligent autonomous aircraft using active sion In Proceedings of the UAV 2000 International Technical Conference andExhibition, Paris, France, June 2000 Euro UVS

vi-10 DeGarmo M, Nelson G M (2004) Prospective Unmanned aerial vehicle tions in the future national airspace system, AIAA 4th Aviation Technology,Integration and Operations (ATIO) Forum, 20 - 23 Sept 2004, Chicago

opera-11 Wong K.C, Bil C (1998) UAVs over Australia - Market And Capabilities, Paper

No 4, Proceedings of the 13th Bristol International Conference on RPVs/UAVs,Bristol, UK, 1998

12 Duffie J A, Beckman W A (1991) Solar Engineering of Thermal Processes,Second Edition New York: Wiley-Interscience

Control and Guidance for a Tail-Sitter

Unmanned Air Vehicle

R Hugh Stone

School of Aerospace, Mechanical and Mechatronic Engineering,

Building, J07, University of Sydney, NSW, Australia 2006

hstone@aeromech.usyd.edu.au

Summary This paper details the control and guidance architecture for the T-Wingtail-sitter unmanned air vehicle, (UAV) The vehicle uses a mixture of classical andLQR controllers for its numerous low-level and guidance control loops Differentcontrollers are used for the vertical, horizontal and transition flight modes, gluedtogether with supervisory mode-switching logic This allows the vehicle to achieveautonomous waypoint navigation throughout its flight-envelope The control designfor the T-Wing is complicated by the large differences in vehicle dynamics betweenvertical and horizontal flight; the difficulty of accurately predicting the low-speedvehicle aerodynamics; and the basic instability of the vertical flight mode Thispaper considers the control design problem for the T-Wing in light of these factors

In particular it focuses on the integration of all the different types and levels ofcontrollers in a full flight-vehicle control system

Keywords: UAVs, Guidance, Control, VTOL, Tail-sitter

1 Introduction

The T-Wing is a VTOL UAV that aims to combine the greater efficiency ofwing-born flight with the operational flexibility offered by VTOL configura-tions such as the helicopter The T-Wing is a tail-sitter twin-engined vehicleand is controlled during vertical flight via propeller-wash over its wing andfin-mounted control surfaces In this respect it is similar to the early mannedtail-sitter vehicles of the 1950s, the Lockheed XF-V1 and the Convair XF-Y1[1, 2] This allows the T-Wing to be substantially less complicated than other

“convertiplane” configurations such as the tilt-wing, tilt-rotor or tilt-body Inoverall configuration the T-Wing is most similar to the Boeing Heliwing of theearly 1990s [3] This was also a twin-engined vehicle but unlike the T-Wingused helicopter cyclic and collective pitch controls during vertical flight Apicture of the T-Wing vehicle during fully autonomous vertical mode flight isshown in 1(a), while a diagram of its typical flight profile is given in 1(b)

P Corke and S Sukkarieh (Eds.): Field and Service Robotics, STAR 25, pp 453–464, 2006.

© Springer-Verlag Berlin Heidelberg 2006

The fact that the T-Wing operates across a much wider range of speedsand attitudes than a conventional aircraft complicates the design of the flightcontrol system for this vehicle It is required to operate in vertical flight whenthe vehicle aerodynamics are dominated by the propeller slipstream effects andthe dynamic modes are significantly unstable, as well as in forward flight whenthe vehicle behaves like a conventional aircraft Due to the widely differentbehaviour of the basic plant between these two fundamental flight modes(vertical and horizontal) it is necessary to have a range of controllers to coveroperation at these fundamental modes as well as the transitions between them.Although individually the controllers are relatively simple, the overall controlsystem is quite complex and involves considerable switching logic to determinewhich controllers to use at any given time as well as how to smoothly transitionbetween these controllers.

This paper will discuss the overall control system architecture of the Wing vehicle Section 2 will outline the basic mathematical model of thevehicle, while Section 3 will discuss the on-board sensors and filters Sections 4and 5 will introduce some aspects of the control and guidance of the vehicle forvertical and horizontal flight phases respectively, while Section 6 will deal withthe transition modes Section 7 considers the flight-state logic and controllertransition issues, before looking at the actual implementation of the controlsystem and some flight-test results in Section 8

T-(a) Autonomous hover flight (b) Flight profile

Fig 1 T-Wing Vehicle in full autonomous hover flight and Flight Profile Styrofoamballs attached to fins are for tip-over protection during initial vertical flight testing

2 Basic Vehicle Model

For the purposes of simulation and control design, The T-Wing vehicle is eled as a standard 6-DOF non-linear rigid-body aircraft model The modeling

mod-is done in standard body-axes centered at the aircraft center of gravity where

x points forward, through the aircraft nose; y is directed to the starboard,(right); and z is directed through the belly of the aircraft

Control and Guidance for a Tail-Sitter Unmanned Air Vehicle 455

The standard body-axis equations of motion are given below using thenotation of Stevens and Lewis [4] In order these equations are the force (1),moment (2), kinematic [quaternion form] (3), and navigation (4) equations

✷

✹ ˙U

˙V

˙W

✸

In the above equations (U, V, W ) are the body axis velocity states; (P, Q, R)are the body axis rates; (q0, q1, q2, q3) are the quaternion representation of thevehicle attitude; and (pN, pE, h) are the North, East and height positions.These variables form the basic vehicle state vector The aerodynamic forceand moment vectors are F = (F x, F y, F z) and M = (L, M, N) respectively

J is the standard inertia matrix for the vehicle, m is the mass, and B is thetransformation matrix that takes vectors from the North-East-Down (NED)frame to the body axis frame The gravitational acceleration in the body axisframe is (gx, gy, gz) The B-matrix can either be obtained directly from thequaternion parameters or equivalently from other Euler angle representations

(a) X-Body axis force (b) Z-Body axis force

Fig 2 X and Z Body forces plotted verses velocity (m/s) and angle of attack (◦)

In these equations non-linearities arise in both their basic structure aswell as in the force and moment vectors F and M This can be appreciated

by considering the X and Z body axis forces plotted verses velocity and angle

of attack as shown in Figure 2

In Figure 2(a) the four stacked surfaces represent X-force values at ent throttle settings This particular graph is taken from an estimated aero-dynamic database for the vehicle [5]

differ-Unlike conventional aircraft where the non-linearity in the force and ment data is largely confined to the parabolic variation of these terms withspeed, the non-linearities for the T-Wing are more complicated due to thechanging relative importance of the propeller generated forces in comparison

mo-to those due mo-to the free-stream dynamic pressure This change occurs as thevehicle goes from low-speed vertical flight (propeller and propeller slipstreamforces dominate) to high-speed horizontal flight (free-stream dynamic pressuredominate)

2.1 Attitude Representations

Three distinct attitude representations are used for the simulation and flightcontrol of the T-Wing vehicle These consist of a quaternion representationand two Euler angle representations

The quaternion representation is used exclusively in the flight simulation ofthe vehicle, as well as in the guidance controller for the transition maneuversbetween horizontal and vertical flight The quaternion representation has theadvantage of being unique (to within a choice of sign) and of having no areas

of degeneracy of its solution

For horizontal mode flight the standard ordered Euler angle rotations ofyaw (ψ), pitch (θ) and roll (φ) about the vehicle z, y and x body-axes respec-tively are used to describe the vehicle’s attitude

Due to the degeneracy of the standard Euler angles for vertical flight tudes, a second set of Euler angles has been defined for vertical mode flight.Starting from a vertical attitude with the vehicle belly (z-axis) facing North,these consist of a vertical roll (φv) [opposite sense to ψ] about the x axis; avertical pitch (θv) about the y-axis; and finally a vertical yaw (ψv) about the

atti-z axis

The advantage of coupling this new system in vertical flight with a dard aircraft system for horizontal flight is that the senses of pitch, roll andyaw are consistent between the different representations Transformations be-tween these three representations can be easily calculated [6]

stan-It is also possible to express the kinematic equations of motion in terms

of either the vertical or horizontal Euler angles This is useful for design and is in-line with standard aircraft control practice The rationalefor using Euler angles for control is threefold Firstly, their use allows thestate variables (including attitude states) and controls to be separated intodistinct longitudinal and lateral partitions Secondly, within these partitions,

control-the approximate assignment of control surfaces to attitude states is invariantwith Euler Angle attitudes Lastly, it is much easier to relate meaningfulcontrol objectives to Euler angles than to the quaternion parameters.

3 Sensors and Flight Control Hardware

The vehicle has a fairly simple suite of on-board sensors to enable it to estimateits current state A schematic of this system is shown in Figure 3 The primarycomponents of the flight control system are as follows:

• A 400MHz Celeron flight computer in a PC-104 form-factor Thisinterfaces with the IMU and GPS units via standard serial communication

• A Honeywell HG1700AG17 Ring-Laser Gyro inertial measurementunit (IMU) comprising 3 accelerometers and 3 rate gyros, with 10◦/hour drift

• A Novatel ProPak-G2 Plus 5Hz GPS receiver This receives standardRTCM differential corrections from the ground station and also includes adedicated Kalman filter to allow interfacing with the Honeywell IMU to givefiltered position, velocity, attitude (PVA) estimates

• Analog voltage signals are used for a static pressure sensor to enablemeasurement of pressure altitude and rate of climb as well as for simple voltagedividers to monitor battery voltages These are sampled via an AD card

• Control of the standard RC hobby servos is via PWM signals ated by a PC104 card

gener-• The data-link to the ground is via a spread spectrum radio-modem.All significant state data is transmitted and recorded at 80Hz

Throttle Servos (1x2) Elevon (2x2)

DGPS Corrections

Joystic for Manual Reversion Modes

RS232 RS232

Fig 3 Complete T-Wing UAV System

4 Vertical Flight Control and Guidance

During vertical flight the T-Wing uses a set of gain-scheduled LQR controllers

to control translational velocities in the body axis y and z directions using theelevons and rudders These are combined with a vertical roll-rate controller(for heading control – or in other words which direction the belly is pointingin) and a simple vertical velocity throttle controller The vertical flight mode

is where the vehicle is most unstable and also where the mode of operation ismost novel in comparison to that of a conventional aircraft

457Control and Guidance for a Tail-Sitter Unmanned Air Vehicle

(a) Vertical Flight

Free Body Diagram (b) Low-Level Vertical Flight Controllers indicat-ing state and command inputsFig 4 Vertical Flight Free Body Diagram and Overall Control Structure

In considering the design of a vertical attitude translational W -velocitycontroller use will be made of Figure 4(a), which gives a free-body diagram

of the vehicle, without lateral states For low-speed vertical-mode flight theperturbation system, linearised about the hover trim state (U 0) in standard

˙x = Ax + Bu state-space form is as follows:

In the above equation standard mechanics of flight notation has been used

to represent the significant force and moment derivatives, (eg MW = ∂M/∂Wetc.,) From this equation the trim elevon deflection and the trim vertical pitchangles can easily be determined [?]

This reduced set of equations can also be used to develop LQR controllersfor the vehicle During flight the W -velocity component, the pitch rate, Q, andthe vertical pitch angle, θv, are all available for feedback control of the elevatorvia the onboard inertial and GPS sensors The LQR design process gives rise

to a vector of control gains K = KQKθKW , which can be used directly in

a W -translational velocity controller (rather than regulator) by replacing zeroregulation on the W -state with zero regulation of a W -command error Thistype of body-axis velocity control, though slightly unusual for an air-vehicle,has been found to work well In practice, the gains are scheduled with verticalvelocity The design of a sideways V -velocity controller is similar

The suite of vertical flight controllers is completed by an aileron roll-ratecontroller (yaw rate when viewed as a helicopter) and a throttle vertical ve-locity controller These are simple SISO proportional controllers The overallstructure of the low-level vertical flight controllers is given in Figure 4(b).Guidance during vertical flight is implemented via simple proportionalguidance based on the current errors between the vehicle position and the next

waypoint Waypoint definitions consist of North, East, Height (NEH) positioncoordinates and a pointing angle, (which is the angle that the vehicle’s bellyfaces) The current errors in horizontal position are converted into errors inthe local vertical, local horizontal (LVLH) reference frame of the vehicle, andare then used (with judicious saturation limits) to generate appropriate Wand V velocity commands for the elevon and rudder controllers respectively.Height errors are similarly used to generate vertical velocity commands to thethrottle controller, while pointing errors are used to supply vertical roll-ratecommands to the aileron control-circuit.

5 Horizontal Flight Control and Guidance

For horizontal flight the vehicle uses pitch and yaw rate controllers for itselevator and rudder control surfaces as well as a roll-rate controller for theailerons Speed control is via throttle All these controllers are designed usingclassical SISO root-locus techniques The form of the pitch and yaw ratecompensators are as given in equation (5) for the pitch-rate controller

6 Transition Mode Guidance

For the transitions between vertical and horizontal flight the same low-levelrate controllers are used as for horizontal flight The main differences are to

be found in the guidance used For both transition maneuvers the guidancealgorithm is based on determining a “ quaternion velocity” that takes thevehicle from its current attitude to a target attitude as detailed below

459Control and Guidance for a Tail-Sitter Unmanned Air Vehicle

• A target attitude for the transition is selected.

• The target attitude is then converted to a quaternion and compared withthe current vehicle attitude expressed in similar form

• The quaternion difference between the current and desired attitude is theninterpreted as a quaternion velocity vector over a nominal time incrementand this is used to generate commanded rates for the vehicle controllers,(with suitable saturation limits applied) as given in (6)



PQR

is a nominal transition time interval that converts the quaternion differenceinto a quaternion velocity vector

During the transition manoeuvres, throttle can either be specified loop via a pitch-angle schedule or controlled with a velocity controller to match

open-a pre-specified velocity schedule with pitch-open-angle

7 Flight State Logic

The low-level and guidance controllers described so far allow the vehicle tooperate autonomously throughout its complete flight envelope under non fail-ure conditions given suitable logic to govern transitions between these differ-ent flight modes However, the situation is complicated because a significantnumber of extra controllers and guidance modes are required for a full vehiclecontrol system These other controllers implement functionality to allow forsensor and system failures as well as for various levels of manual intervention

in the vehicle control and guidance

The main system failures that the control system must address are loss ofGPS signal and loss of the communication link In the case of loss of GPS signalthe vehicle can only navigate for a relatively short period of time using its(low-accuracy) inertial sensors This means that any controllers that depend

on velocity information (such as the vertical flight velocity controllers) must bereplaced with other controllers and guidance action must be taken to attempt

a safe landing of the vehicle In the case of sustained communication failure,the vehicle needs to abort its mission and return home As well as handlingthese vehicle failure modes, it is also necessary to allow fully autonomousoperation to be overridden with different levels of manual guidance and controlfor testing purposes

The addition of manual and failure state modes of operation requires ful thought in terms of allowing or forcing mode transitions For instance a

care-transition back to autonomous operation must be forced if the vehicle is in

a manual mode at the time of a communication failure Failures of GPS sors or communications during any of the four normal flight modes (vertical,horizontal, and the two transition modes) must also trigger forced transitions

sen-to degraded modes of operation that involve the use of different sets of flightcontrollers or different guidance algorithms This requires complex supervisorylogic to sit above the guidance and control functionality

To handle this problem in a consistent and rigorous manner, use has beenmade of the Mathworks Stateflow Toolbox that integrates with the rest ofthe SIMULINK simulation and rapid-prototyping environment in which allthe flight control design and development is conducted This toolbox allowsgraphical representation of all the flight mode logic as a finite-state machine

An example of a typical state transition diagram is shown in Figure 5

twmodr143_novrml_modStateFlow/filt_guid_ctrl/flightStateLogic/Flight−State/Autonomous Autonomous

during: thetaDeg = R2D*state[11];

Fig 5 Stateflow diagram for standard autonomous mode

Figure 5 looks inside the autonomous mode block of the overall Stateflowdiagram at the transitions between the different autonomous flight conditionssuch as vertical and horizontal flight Further drilling down into the indi-vidual autonomous modes would reveal extra flight states for various failureconditions Going up a level would reveal transitions between autonomous andmanual modes of operation

The advantage of using Stateflow is that it formalizes the description of thestate transitions in a readily understandable graphical representation for thewhole control system This obviates the need for manual coding of complexdecision trees that is prone to coding errors and logical inconsistencies.7.1 Controller Transitions

The fact that the T-Wing does not have one set of consistent controllers thatoperate throughout its entire flight envelope requires attention be paid totransitions between different controllers during control mode changes The

461Control and Guidance for a Tail-Sitter Unmanned Air Vehicle

most critical transition occurs at the re-entry to vertical flight before landingwhen the transition rate controllers are replaced with vertical flight transla-tional velocity controllers This is because the vehicle may still have significantresidual translational velocity even after it has reached a vertical attitude andsuddenly switching to velocity based controllers may cause significant controltransients To handle this the following procedure is used.

• On re-entering the vertical mode the control surface deflections and variables are used to back-calculate translational velocity guidance com-mands corresponding to the current vehicle state based on the knownvelocity mode control algorithms

state-• These initial guidance commands are ramped down to zero over a set timeperiod to bring the vehicle to a stable hover

• The vehicle then starts to accept normal guidance commands to navigate

to its landing location

To see this consider the simplest form of the W -velocity controller

Control transition problems are not found in going from vertical flight tothe horizontal-to-vertical transition mode because the translational velocitycontrollers automatically keep angular rates close to zero Hence the entry

to the transition rate controllers is well-behaved As both sets of transitioncontrollers use the same low-level rate controllers as are used for horizontalflight, there are also no problems in changing at either end of the horizontalflight mode

8 Implementation Details and Results

The complete control system for the T-Wing vehicle is developed and lated within the SIMULINK simulation environment To convert the controlsystem to a hard real time flight controller, use is made of the MathworksRealtime Workshop and xPCTarget toolboxes These allow for the automaticgeneration of hard real-time code from the SIMULINK model as well as in-terfacing to external sensors and actuators through xPCTarget driver blocks

simu-The whole process of converting the simulated flight controllers into time controllers on-board the vehicle simply requires cutting and pasting theflight-control block from the simulation model into the flight control modeland performing a real-time build.

real-(a) Demonstrator undertaking tethered

hover tests March 2005 Both top and

bot-tom tethers are slack

650 700 750 800

−500 0 500 1000

θ V

650 700 750 800

−20 0 20

Teth-The T-Wing has flown numerous flights in different configurations ing:

includ-• Full manual hover flights (demonstrating the significant inherent ity of the vehicle in this condition);

instabil-• Tethered hover flights in which the pilot provides velocity guidance mands to the translational velocity controllers (Figure 6(a) and 6(b));

com-• Tethered 5-DOF autonomous flights in which operation is fully autonomousexcept for vertical velocity commands provided by the pilot;

• Velocity mode free flights (velocity guidance commands from pilot); and

• Fully autonomous vertical mode flight A picture of the vehicle in thismode is shown in Figure 1(a)

9 Conclusion and Future Work

This paper has considered the overall flight control system for the T-WingVTOL UAV, including low-level and mid-level guidance controllers for allthe non failure-state flight modes that the vehicle must operate in Althoughthe individual controllers are relatively simple the overall control structurerequires a complex layer of supervisory control logic to handle controller andflight-mode transitions This has been implemented using Stateflow So farthe T-Wing vehicle has been successfully flown in a variety of vertical flightmodes including fully autonomous vertical flight It is anticipated that flightthroughout the rest of the flight envelope will occur before the end of 2005

463Control and Guidance for a Tail-Sitter Unmanned Air Vehicle