GPS-BASED SYGPS-BASED SYSTEM IDENTIFICASTEM IDENTIFICATIONTIONFOR PRECISION CONTRFOR PRECISION CONTROL OF OL OF AA

WING-SAILED CAWING-SAILED CATTAMARANAMARAN

GGABRIELABRIEL H.H. EELKAIMLKAIM

PPHH.D.D.. OORALRAL DDEFENSEEFENSE

13.M13.MARCHARCH.2001.2001

GOALS OF ATLANTIS PROJECT

STANFORD UNIVERSITY

• Wing-sail propelled• GPS-based navigation• Autonomous• Self-sailing capability• Self-guiding to better

than 1 meter

• “Systems” thesis withefforts in structures,aerodynamics, andcontrols.

MOTIVATIONS

• Reduced fuel costs• Increased speeds• Replace “deep water” buoys with

station keeping• Remote weather and/or

environmental monitoring• Marine equivalent of UAV with

infinite loiter time

STANFORD UNIVERSITY

CONTRIBUTIONS (1.3)• Conceived, designed, built, and

experimentally demonstrated anautonomous sailboat capable of precisioncontrol to better than 0.3 meters.

• Developed methodology to identify robustplant models and controllers that areinvariant under velocity changes.

• Described optimization scheme forsymmetric wingsail section based onrequirements unique to sailing vehicles.

STANFORD UNIVERSITY

STANFORD UNIVERSITY

CONTRIBUTIONS (2.3)

• Developed and experimentallydemonstrated novel quaternion-basedattitude estimation algorithm from vectorobservations. [with Demoz Gebre]

• Developed and experimentallydemonstrated novel method for calibratingany 3-axis sensor that requires no externalreference. [with Demoz Gebre]

STANFORD UNIVERSITY

CONTRIBUTIONS (3.3)• Identified robust model for Tractor and

Implement System based on experimental data.

• Demonstrated method for robust identificationof multiple implements and velocityconfigurations for GPS-guided tractor.

• Experimentally demonstrated precise control ofGPS-guided tractor to 3 cm error withimplement (vs. 12 cm for excellent humandriver) based on identified models.

PRESENTATION MAP• IntrIntroductionoduction• Anatomy of Atlantis• Attitude System• Propulsion System• Identification and Control• Experimental Results• Conclusions• Future Work

STANFORD UNIVERSITY

STANFORD UNIVERSITY

GLOBAL POSITIONING SYSTEM

© T.Walters

WINGSAIL HISTORY (1.3)

STANFORD UNIVERSITY

1926

Baden Baden

1940

Flaunder

1951

1968

Carl

1969

PlaneSail

1972

SKAMP Miss Nylex

STANFORD UNIVERSITY

WINGSAIL HISTORY (2.3)

1976

Beaty

1979

BergesonReport

1981

Shin Aitoko Maru

1983

MINI Lace

1985

Fekete

Cousteau

STANFORD UNIVERSITY

WINGSAIL HISTORY (3.3)

1988

1989

Ross “C”-class

1990

Zephyr

1996

Blue Nova

1999

RAFT BoatekStars & Stripeswins America’s Cup

Pelmatic

2000

AUTONOMOUS SAILBOATS

STANFORD UNIVERSITY

SKAMP project- RCA Astro-Electronics division- Uses two curved NACA 0030 wings- Elastomeric hull- Positioning via Transit- Speed ~4 knots (2 m/s)- “Station Keeping” within 0.2 nm (370 m.)

RELATIONSHIP project- Technical University of Furtwangen.- Conventionally sailed trimaran.- Satellite link between boat and control.- Currently anchored in Azores.

Attitude Estimation- Wahba [1965] proposes two-vector problem.- Bar-Itzhack [1985] extends solution to filter form.- Creamer [1996] solves via two successive rotations.- Hayward, et al [1999] constrains yaw, and solves for- Murdin [2000] solves via eigenvalue decomposition.

Magnetometer Swinging

- Bowditch [1865] solves using known headings.- Hine [1968] demonstrates complete error analysis.- Psiaki [1990] uses orbital dynamics and gyros.- Murdin [2000] solves via Information filter driven

from an INS.

STANFORD UNIVERSITY

PRIOR ART

Tn b

[ ]→

STANFORD UNIVERSITY

PRESENTATION MAP• Introduction• AnatomAnatomy of y of AtlantisAtlantis• Attitude System• Propulsion System• Identification and Control• Experimental Results• Conclusions• Future Work

ATLANTIS’ ANATOMY (1.2)

STANFORD UNIVERSITY

Wing:free to rotate on bearings,propels the sailboat.

Wing:free to rotate on bearings,propels the sailboat.

Ballast:centers the wing’s masson the bearings.

Ballast:centers the wing’s masson the bearings.

Rudders:provide steering,actuatedvia computer.

Rudders:provide steering,actuatedvia computer.

Attitude:senses the orientation ofthe sailboat.

Attitude:senses the orientation ofthe sailboat.

Anemometer:senses the wind speedand direction.

Anemometer:senses the wind speedand direction.

Tail:keeps the wing adjustedto the wind.

Tail:keeps the wing adjustedto the wind.

STANFORD UNIVERSITY

ATLANTIS’ ANATOMY (2.2)

Attitude Sensor

GNC Computer

GPS Receiver

Hullspeed Sensor

Rudder Angle Sensor

Rudder Actuator

Anemometer

Wing MagnetometerLower Flap Actuator

Center Flap Actuator

Upper Flap Actuator

Tail ActuatorSlip RingCAN Network

ATTITUDE SYSTEM

STANFORD UNIVERSITY

• Attitude system is requiredto create “synthetic” sensorat position different thanGPS antenna.

• Uses a 3-axis magnetometer,and 2-axis accelerometer to

determine attitude.

• Magnetometer heading isvery sensitive to errors inpitch and roll.

ANEMOMETER

STANFORD UNIVERSITY

V KTwind

Controller

= 1

∆ µ

• Anemometer mounted on wingsail“pod,” measures angle of attack ofwingsail.

• Three cup anemometer measuresdynamic pressure via differential drag.

• Microcontroller measures the time difference inclock cycles, wind velocity is inverselyproportional to ∆T.

WINGSAIL / PROPULSION• 5.37 m. in height• 1.25 m. chord• 71 kg. mass• mass balanced about quarter chord• full-flying tail• free to rotate about vertical axis

on bearings• construction materials are:

- plywood ribs and skin- plywood shear webs- mylar covering- aluminum stub-mast

STANFORD UNIVERSITY

STANFORD UNIVERSITY

RUDDER ACTUATOR• Mechanical lead-screw

translates rotary motion torudder angle

• Fractional horsepowerbrushed-DC motor

• PWM from Zanthic/SeimensSAB-505CA µController

• 500 count-per-revolutionencoder

• Infineon 5-A H-bridgemotordrive chip

RUDDER ANGLE SENSOR

STANFORD UNIVERSITY

• The rudder angle sensor is alohet mounted in between twomagnets.

• The output is proportional tothe flux crossing the lohet plane:

• The rudder was calibrated by deflectingto +/-30 and +/- 45 and zero degrees asmeasured perpendicular to the hull end.

V Koutput = sin( )Θ

STANFORD UNIVERSITY

HULLSPEED SENSOR

• Magnetic “Paddle-Wheel” and hall-effectsensor buried on hull centerline.

• Motion based on paddle wheel beingsemi-submerged in moving water.

• Four pulses per revolution.• Systematically very similar to

anemometer.• Calibration based on matching GPS

velocity and Hullspeed for out and backcourse using electric trolling motor.

GNC BOX• Waterproof case for electronics.• Pentium class GNC computer.• DC/DC converter runs

main computer off ofsystem 12V bus.

• Trimble Ag122 GPS receivercommunicates to GNC onRS-232.

• ESD CAN dongle providesCAN interface to GNCcomputer.

STANFORD UNIVERSITY

FLAP/TAIL ACTUATORS

STANFORD UNIVERSITY

• Actuators are used to movetrailing edge flap and tail

• Actuators are brushed DCmotors, connected to leadscrews

• Push rods link the leadscrews and the controlhorns

• Motors are controlled via aµcontroller using PWM drive

STANFORD UNIVERSITY

PRESENTATION MAP• Introduction• Anatomy of Atlantis• Attitude SystemAttitude System• Propulsion System• Identification and Control• Experimental Results• Conclusions• Future Work

STANFORD UNIVERSITY

ATTITUDE SYSTEM• Attitude is computed based on

vector “matching.”

• A new quaternion attitudeestimation algorithm was developed to take

advantage of low-costsensors.

• Uses a 3-axismagnetometer, and 2-axis accelerometer todetermine attitude.

VECTOR MATCHING

STANFORD UNIVERSITY

• Vector matching consists offinding the rotation thatbrings the measured (body)vector into alignment withthe known (inertial) values.

• Ambiguity of rotation aboutthe vector requires at leasttwo vectors to solve.

• Implemented withquaternion-based algorithmwhere:

q q qtrue e= ⊗ ˆ

STANFORD UNIVERSITY

PERFORMANCE ON SIMULATED DATA

150 100 50 0 50 100 150

80

60

40

20

0

20

40

60

80

startnose

startwing

startdown

azimuth (deg)

elev

atio

n (d

eg)

Unwrapped surface of sphere

nosewingdown

The trace of unit normal vectors (nose, wing, anddown) are shown converging on the true atti-tude from an initial guess.The surface of thesphere is unwrapped via a Mercator projectioninto azimuth and elevation.

Startnose

Startdown

Start wing

Target

Target

Target

MAGNETOMETER CALIBRATION

• Honeywell HMC2300 3-axis magnetometer,measures earth’s magnetic field in body frame.

• Hard Iron (bias) and Soft Iron (scale factor)errors need to be removed by calibration.

• Two-step calibration method requires noadditional information or instrumentation.

• Biases and Scale Factors solved in Measurementdomain.

STANFORD UNIVERSITY

IDEAL 2-D SENSOR TRACE

STANFORD UNIVERSITY

• A vector is constant in theinertial frame, but is measuredin the body frame.

• As the bodyframe isrotated the componentsmeasured along eachaxis trace out a circle.

vvectorector

inertial

bodybodyvvectorector

inertialbodybody

tracetrace

STANFORD UNIVERSITY

2-D MAGNETOMETER SIMULATION

• Break the estimation intotwo parts: Least squaresestimation of non-intuitivestates, then algebraicmanipulation of states toextract relevant parameters.

• Can be solved with only a small part of the circle.

-4 -3 -2 -1 0 1 2 3 4 5

-4

-3

-2

-1

0

1

2

3

Simulation of Two-Step Non-Linear Estimator

truth scaled biased noise reconstructed

x x y y

x

k

k y

k x k y

i i i i[ ] = −[ ]− −

2 2

0

2

2 0

1 02

2 02

2 2

2 2 1

k a R

ka

b

=

=

12 2

2

2

2

x x

a

y y

bR

−( )+

−( )=0

2

20

2

22

STANFORD UNIVERSITY

EXPERIMENTAL SETUP (GROUND)

• Attitude box secured to endof a long wooden boom,with magnetometerrecording data at 100 Hz.

• Honeywell navigation gradeINS records attitude forlater comparison.

• Setup is pitched, rolled, andyawed to generate data forcalibration run.

• Experiment repeated several times for validationrun.

STANFORD UNIVERSITY

EXPERIMENTAL RESULTS (1.2)

• Before calibration, trace is poormatch to the surface of the sphere.

• After calibration, trace is anexcellent match to surface ofsphere.

0.5

0

0.5

0.5

0

0.50.5

0.4

0.3

0.2

0.1

0

0.1

0.2

0.3

0.4

0.5

Hx

Calibrated Magnetic Field

Hy

H z

0.5

0

0.5

0.5

0

0.50.5

0.4

0.3

0.2

0.1

0

0.1

0.2

0.3

0.4

0.5

Hx

Raw Magnetic Field

Hy

H z

STANFORD UNIVERSITY

EXPERIMENTAL RESULTS (2.2)

• After calibration,validation runshows a veryconstantmagnitude ofmagnetic fieldvector.

• Before calibration,the magnitude ofthe magnetic fieldvaries greatlywith pitch androll inputs.

0 200 400 600 800 1000 1200 1400 1600 1800 20000.3

0.35

0.4

0.45

0.5

0.55Magnitude of Magnetic Field Vector

Time (sec)

|| H

n || (

Gau

ss)

Corrected Hn

Uncorrected Hn

STANFORD UNIVERSITY

QUATERNION ATTITUDE ON REAL DATA

0 5 10 15 20 25 30 35 40 45 50200

100

0

100

200

Attitude (Deg). fs = 1 Hz.

Yaw

estimatedtrue

0 5 10 15 20 25 30 35 40 45 5040

20

0

20

40

Pitc

h

0 5 10 15 20 25 30 35 40 45 50100

50

0

50

Rol

l

Time (min)

• Real data basedon flight test ofQueenAir fromLivermore toSan Jose.

• Truth is fromshort baselineGPS attitude.

• Coordinatedturns throughdue East anddue West (Yawis +/- 90)violatedconditions ofhaving two non-collinearvectors.

STANFORD UNIVERSITY

PRESENTATION MAP• Introduction• Anatomy of Atlantis• Attitude System• PrPropulsion Systemopulsion System• Identification and Control• Experimental Results• Conclusions• Future Work

WHY USE A WINGSAIL?• Self-trimming WingSail can be controlled with

very small actuators.

• WingSail can be self-trimming within a largerange of wind directions.

• WingSail is more efficient:

- Aeroelastically stable- Higher lift/drag ratio- Greater

• Tacking and Jibing become very docilemaneuvers.

STANFORD UNIVERSITY

CLmax

HOW DOES THE WINGSAIL WORK?

STANFORD UNIVERSITY

Wind

• Tail deflection causes WingSail to rotate on bearings,generating an “angle of attack” to the wind, and thus lift.

• Lift on WingSail pulls boat forward through water.

Lift

STANFORD UNIVERSITY

HOW DOES SELF-TRIMMING WORK?

Lift

• A change in the wind direction causes a load on the tail whichrotates the WingSail on its bearings.

• The WingSail orientation remains constant relative to the wind!

Wind

Lift

Wind

• The WingSail is tacked orjibed by centering the tailas the wind passes thecenter-line of the boat.

• The tail is then turned tothe opposite side of theWingSail and the tack orjibe is complete.

• The only difference betweena tack and a jibe is theorientation of the WingSailwith respect to the boat.

• The WingSail remains at aconstant orientation to thewind.

STANFORD UNIVERSITY

Wind

HOW DO TACKING & JIBING WORK?

STANFORD UNIVERSITY

AIRFOIL DESIGN METHODOLOGY

• Low Reynolds number (<300,000) requires specialdesign considerations.

• Conventional symmetric airfoil perform very poorly.

• Large, blunt leading edge for good

• Trip boundary layer at maximum thickness.

• Flat “rooftop” pressure distribution until trip point.

• Long, slow pressure recovery required.

CLmax

STANFORD UNIVERSITY

FINAL AIRFOIL SECTIONWINGSAIL:

STANFORD UNIVERSITY

CONFIGURATION CONSIDERATIONS

• The WingSail needs to weather vane into therelative wing, and at the same time, trim a largemaximum lift coefficient ( ).

• Control power, or pointing ability

• Mass balance about pivot point

• Minimum swept-radius WingSail

• Mechanical complexity

• Stability requires: ∂Cm/∂α < 0

• Trim requires: Cm = 0

CLmax

STANFORD UNIVERSITY

WINGSAIL CONFIGURATIONS

Canard “Free-Floating” Canard

“Flying Wing”Conventional

STANFORD UNIVERSITY

dy

Physical Parameters:

b = 5.37 m.

c = 1.425 m.

S = 7.65 m2.

m = 324 kg.

AR = b/c = 3.765

=1.8

ρair= 1.225 kg/m3

g = 9.81 m/s2.

l V cCl F

Cl

= +( )1

22ρ α α δ

α δ δ1 244 344

V cC ll= = [ ]1

22ρ N / m

L ldy V cC dy V cC dy V cC b lb Lh

h b

lh

h b

l h

h b

l= = = = [ ] = =+ + +

∫ ∫ ∫1

2

1

2

1

22 2 2ρ ρ ρ [N]

BASIC STRUCTURAL LOADING

CLmax

STANFORD UNIVERSITY

WING SPAR STRUCTURAL LOADS

lift [N/m]

b [m]

R1

M1

R2

s

lift [N/m]

R1R1

M1

M1

R2 R2

+

-

-

+

EIw Mlx

R x

EIwlx

Rx

c

EIwlx

Rx

c x c

′′ = − = − −

′ = −−

+

= −−

+ +

2

2

3

2

2

1

4

2

3

1 2

2

6 2

24 6

γ

γ

γ

Evaluate B.C’s:w’(b)=0w(b)=0w(b-s)=0

Rl

sb bs s= − +[ ]2

2 2

86 4

Forces Shear MomentsV lx R x= − −2

0γ Mlx

R x= − + −2

22γ

R lb R= −1 2 Mlb

R s= − +1

2

22

Mb

R2s-

-

Rb

R2

+

-

STANFORD UNIVERSITY

STUB-MAST STRUCTURAL LOADS

T

Rb

Mb

R1

R2

M1

T

T

Rb

Mb

h

R1

s

R2

M1 ~

EIw M R x R T x s M x s

EIwR x

R

′′ = − = + −( ) − − −

′ = + −

2 1 10

22

12

~

~TT

x sM x s c

EIwR x

R Tx s

Mx s

c x c

( ) − − − +

= + −( ) − − − + +

2

1 1

23

1

3

1

2

1 2

2

6 6 2~

Evaluate B.C’s:w’(s+h)=0w(s+h)=0w(s)=δcable

Forces Shear MomentsV R R T x s= + −( ) −2 1

0~

lb T Rb− =~

M R x R T x s M x s= − − −( ) − + −2 1 10~

Mlb

lbh hTb = − − +2

2~

wEI

h lb lbh hT

TL

AE

x s cable

cable

=+ −[ ]

=

=

=

2 21 3 4 4

12

~~

~ ~

δ

δ

STANFORD UNIVERSITY

• Stub mast is a 4” O.D. aluminumT6061 pipe with 3/8” wall thickness

• Wing ribswererouted outof 3/8”marineplywood

• Ribs assembled on jig, with a sprucespar epoxied in place

WINGSAIL CONSTRUCTION (1.2)

STANFORD UNIVERSITY

• Skin bonded to ribs with epoxyforming the front of the “D” tube

• Shear websare madeof 3/8”marineplywood

• Wingsail is covered with “industrial”grade mylar covering

WINGSAIL CONSTRUCTION (2.2)

STANFORD UNIVERSITY

CONSTRUCTION: LOAD TEST

• Load test consisted on hanging a 72 kg. dummyload off the end of the wing, while having thecrossbeam of the catamaran secured to a column.

• Based on results of load test, bottom section shearwebs were reinforced.

STANFORD UNIVERSITY

PRESENTATION MAP• Introduction• Anatomy of Atlantis• Attitude System• Propulsion System• Identification and ContrIdentification and Controlol• Experimental Results• Conclusions• Future Work

BASIC KINEMATIC MODEL

y VV

L

y

u

x

xψδ

ψδ

=

+

•0 0

0 0

0 0 0

0

0

1

• Simple linearized model assumes that ruddercannot slip sideways in water.

• Boat assumed to be moving along X-axis atconstant velocity.

• State-space model is triple integrator.

STANFORD UNIVERSITY

ψδ

δ

along track

crosstrack

X

Y

VELOCITY INVARIANCE

STANFORD UNIVERSITY

~

~

~

~y

L

y

uψδ

ψδ

=

+

•

0 1 0

0 01

0 0 0

0

0

1

ψδ

δ

along track

crosstrack

X

Y• Crosstrack error, y, and azimuth error,ψ, are

integrating with distance, not time.• Reformulate model so that the states for y andψ are normalized by velocity.

• Model now invariant with velocity.

~

~

yy

V

V

x

x

ψ ψ

=

=

STANFORD UNIVERSITY

SYSTEM IDENTIFICATION

Inputs Outputs

• Replace the “black box” with a mathematical model.• Observer Kalman-filter Identification (OKID)

method requires only input-output data, no a prioriknowledge of plant structure is needed.

• Using an identified “high-fidelity” model, greatercontrol system authority can be used without thedanger of instability.

Disturbances

STANFORD UNIVERSITY

-100 -80 -60 -40 -20 0 20 40 60 80 100

-30

-20

-10

0

10

20

30

Actuator Mapping Data

PWM [%]

Rud

der

Slew

Rat

e [°

/sec

]

Deadband

NON-LINEAR ACTUATOR MAPPING

• Actuator Mappedfor various PWMsettings and slewrates

• Noisy, but meanline readily visible

• Cubicinterpolation ofpoints outside ofdeadband

• Actuator is non-symmetric, non-linear, and has adeadband

• This same actuator,is now mapped as alookup table

TYPICAL SYS ID PASS

STANFORD UNIVERSITY

0 100 200 300 400 500 600 700 80030

20

10

0

Y [

m]

SysID pass #3

0 100 200 300 400 500 600 700 80050

0

50

Ψ [

°]

0 100 200 300 400 500 600 700 80050

0

50

δ [°]

0 100 200 300 400 500 600 700 800

20

0

20

U [°/

sec]

AlongTrack [m]

• Rudder wasslewed byhuman pilotto excite allmodes ofsailboat.

• Both GPS-derivedbearing andazimuth fromattitudesystem areshown.

• Sensors weresampled at100 Hz.

OKID RESULTS

STANFORD UNIVERSITY

0 20 40 60 80

10 10

10 5

10 0

10 5

Number

SV M

agni

tude

Hankel Singular Values (HSV)

0 5 10 15 2010 6

10 4

10 2

10 0

Number

Mag

nitu

de

Modal Singular Values (MSV)

0 10 20 307.5

7

6.5

6

5.5

5

Y [

m.]

SysID Model Validation

0 10 20 30-0.3

-0.2

-0.1

0

0.1

0.2

Ψ [

rad]

0 10 20 30-1

-0.5

0

0.5

1

Time [sec]

δ [r

ad]

MeasuredModelled

• Large drop offin singularvalues after 4thorder.

• Modal singularvalues confirm4th ordermodel.

• Reconstructionof data from adifferent dataset showsexcellentagreement withmeasurements.

STANFORD UNIVERSITY

PRESENTATION MAP• Introduction• Anatomy of Atlantis• Attitude System• Propulsion System• Identification and Control• Experimental ResultsExperimental Results• Conclusions• Future Work

STANFORD UNIVERSITY

TROLLING TEST SETUP

• • TTested in Redwested in Redwood City Harborood City Harbor• Ballasted Catamaran with 75 Kg.• Ballasted Catamaran with 75 Kg.

of lead to simof lead to simulate wulate weight ofeight ofwing.wing.

• • TTrrolling motor poolling motor powwerered with 12,ed with 12,24,24, and 36 vand 36 volts folts for moror more speed.e speed.

STANFORD UNIVERSITY

BIRD’S EYE VIEW OF TROLLING DATA

1000 800 600 400 200 0 200 400400

200

0

200

400

600

800

Closed Loop Control vs. Human Sailor

Nor

th [

m]

East [m]

• Data from sailingand trolling inRedwood cityharbor, takenover a yearapart.

• Human sailingdata shown inyellow.

• Computer datashown in red.

BIRD’S EYE VIEW OF CLOSED-LOOP

200 100 0 100 200 300

200

150

100

50

0

50

100

150

200

250

start

end

start

end

East [m]

Nor

th [

m]

Controller Performance for: 02_12_00_24.dat

• Closed loop controlalong straight linesegments.

• Very precise controlwas achieved.

• Trolling motor wascanted off centerlineaxis to simulate wind.

STANFORD UNIVERSITY

0 100 200 300 400 5002

1

0

1

2

Y [

m]

Controller Performance for: 02_13_15_07.dat

Mean: 0.027199 σ: 0.098294

0 100 200 300 400 50020

10

0

10

20

Ψ [

°]

0 100 200 300 400 5000

2

4

6

V [

m/s

ec]

AlongTrack [m]

• Close-up view offirst pass.

• OKID 4th order,vel. invariantcontroller.

• Mean:0.03 m.,StandardDeviation:0.10 m.

• Note bias inAzimuth due tocurrent.

• Boat Speed 4 kts.

TROLLING MOTOR CONTROL (1.2)

STANFORD UNIVERSITY

STANFORD UNIVERSITY

TROLLING MOTOR CONTROL (2.2)

0 100 200 300 400 500 600 7002

1.5

1

0.5

0

0.5

1

1.5

2

Along Track [m]

Cro

ssTr

ack

erro

r [m

]

Close Loop Control with Trawling Motor

σ = 0.1194σ = 0.0761σ = 0.1180σ = 0.1010

• All passes havestandarddeviationsbelow 0.15meters.

• All means arebelow 0.15meters.

• “Troll” pathnever exceeds1 meter fromdesired path.

SAILBOAT SETUP• Sailed in Redw• Sailed in Redwood City Harborood City Harbor• Sailed all points of sail.• Sailed all points of sail.• 215 kg.• 215 kg. of of “liv“live” ballast.e” ballast.• GPS and sensor data logged to• GPS and sensor data logged to

computer at 5 Hz.computer at 5 Hz.

STANFORD UNIVERSITY

BIRD’S EYE VIEW OF ALL DATA

STANFORD UNIVERSITY

500 0 500

200

400

600

800

1000

1200

Closed Loop Control vs. Human Sailor

Nor

th [

m]

East [m]

• Data from normalsailing and wing-sailing inRedwood Cityharbor, takenover a yearapart.

• Human sailingdata shown inyellow.

• Computer datashown in red.

BIRD’S EYE VIEW OF CONTROL

400 300 200 100 0 100 200 300

300

200

100

0

100

200

start

endstart

end

start

end

start

endstartend

start

end

East [m]

Nor

th [

m]

Controller Performance for: 06_15_58_28.dat

• Several straight linesegments were sailed.

• Precise control onstraight lines wasachieved.

• Tacks and jibes werevery gentle.

STANFORD UNIVERSITY

STANFORD UNIVERSITY

0 50 100 150 200 250 3002

1

0

1

2

Y [

m]

Controller Performance for: 06_15_58_28.dat

Mean: 0.029249 σ: 0.19474

0 50 100 150 200 250 30020

10

0

10

20

Ψ [

°]

0 50 100 150 200 250 3000

2

4

6

V [

m/s

ec]

AlongTrack [m]

• Close-up viewof first pass.

• OKID 4thorder,velocityinvariantcontroller.

• Mean:0.03 m.,StandardDeviation:0.20m.

• Wind 8-12 kts.

• Boat speed 4-5knots.

CLOSE-UP OF SAILING CONTROL (1.2)

STANFORD UNIVERSITY

CLOSE-UP OF SAILING CONTROL (2.2)

0 50 100 150 200 250 300 3502

1.5

1

0.5

0

0.5

1

1.5

2Closed Loop Line Following Performance

Distance along track [m]

Cro

sstr

ack

erro

r [m

]

σ = 0.1934σ = 0.2839σ = 0.2553

• All passes havestandarddeviationsbelow 0.3meters.

• All means arebelow 0.2meters.

• “Sail” pathnever exceeds1 meter fromdesired path.

STANFORD UNIVERSITY

PRESENTATION MAP• Introduction• Anatomy of Atlantis• Attitude System• Propulsion System• Identification and Control• Experimental Results• ConclusionsConclusions• Future Work

CONCLUSIONS (1.2)• Experimentally demonstrated precise control of an

autonomous sailboat to better than 0.3 meters.

• Identified a robust plant model and controller forautonomous sailboat that is invariant under velocitychanges.

• Developed optimized symmetric wingsail section based onrequirements unique to sailing vehicles.

• Developed and experimentally demonstrated novelquaternion-based attitude estimation from vectorobservations.

• Developed and experimentally demonstrated novel methodfor calibrating strap-down 3-axis magnetometer thatrequires no external reference.

STANFORD UNIVERSITY

STANFORD UNIVERSITY

CONCLUSIONS (2.2)

IT IT WWORKED!!ORKED!!

STANFORD UNIVERSITY

PRESENTATION MAP• Introduction• Anatomy of Atlantis• Attitude System• Propulsion System• Identification and Control• Experimental Results• Conclusions• FuturFuture e WWorkork

FUTURE WORK (1.2)• Implement Auto-Tacking and -Jibing.

• Implement reverse controller and stationkeeping.

• Optimal trajectories to destination.

• On-the-fly current estimation.

• Map/Gradient based higher level navigation.

• Implement distributed control system.

STANFORD UNIVERSITY

STANFORD UNIVERSITY

FUTURE WORK (2.2)• Experimentally determine actual WingSail

performance.

• Improve WingSail structural robustness.

• Improve user interface, and control software.

• Implement Real-Time-OS.

• Solar panels or wind turbine for on-board powergeneration.

• Unmanned sail to Honolulu?

ACKNOWLEDGMENTS• Advisor: Brad Parkinson• Committee: Dave Powell, Per Enge, Juan Alonso,

Chris Gerdes• Family: Silvia, Meyer, Helen, Larry, Estreilla,Ariella,

Sasha• Financial support: OTL, Mary Watanabe• Wing builder: Cris Hawkins• Testing crews: Konstantine, Lee, Chad, Jaewoo,

J.Fey, Demoz, Sharon, Sherman, James, Eric, Lude,Guttorm, Rob.

• Staff: Manni, Chris, Pete Cruz• All the labs that unknowingly donated material

to the project in the middle of the night.STANFORD UNIVERSITY