DOCKING CONTROL SYSTEM FOR A 21” DIAMETER AUV

Robert S. McEwen, Brett W. Hobson, and James G. BellinghamMonterey Bay Aquarium Research Institute

7700 Sandholdt Road, Moss Landing, CA 95039-9644rob@mbari.org, hobson@mbari.org, jgb@mbari.org

Abstract— The Monterey Bay Aquarium Re-search Institute (MBARI) has developed a 21-inch(54 cm) diameter docking AUV and companiondocking station. This program resulted in four con-secutive successful autonomous homing and dockingevents in the open ocean, which included download-ing data, uploading a new mission plan, rechargingthe battery, and complete power cycling of the AUV.We describe the design, simulation, and at-sea testof the homing and docking control system.

I. I NTRODUCTION

Autonomous Underwater Vehicles (AUVs) areplaying an growing role in the ocean sciences.At MBARI, a 6 km depth-rated mapping AUVis routinely generating bathymetry maps with un-precedented precision, resolution and speed, and asecond AUV routinely performs physical, chem-ical, and biological measurements in MontereyBay. Both of these AUVs are characteristic of aclass of vehicles carrying sophisticated, but power-intensive instrumentation. The high power drawof the instrumentation limits today’s AUVs toendurances of roughly one day. This means thatfor all but near-shore operations, the vehicles mustbe attended by a ship. As a consequence, thefundamental utility of the AUV is limited by shipsupport requirements. In some cases, the issue iscost. In other cases, the need for ship supportmakes activities impossible, for example in highsea states, or when events cannot be predicted inadvance for scheduling ship time.

Our docking station is designed to link tothe power and communication infrastructure of aseafloor observatory, enabling the AUV to oper-ate independent of a surface vessel for extendedperiods. In effect, the docking station allows thevehicle to connect to a mooring or a cabledinstrumentation node. Once attached, the AUV canrecharge its batteries, download data, upload new

Fig. 1. AUV in the Dock in a Seawater Test Tank

Fig. 2. AUV Entering the Dock in Monterey Bay

instructions, and safely park until the next mission.Once its batteries are recharged, the AUV can bedeployed nearly instantaneously. Instead of beinglimited to studying static or repetitive subjects, theAUV can be quickly dispatched to record episodicevents such as canyon turbidity flows or planktonblooms, greatly increasing the scientific utility ofan AUV.

The move toward seafloor observatories is in-

ternational in scope, and creates the possibilitythat power and communication infrastructure willbe available on the seafloor in a variety of in-teresting regions. At MBARI, the focus has beenthe development of the Monterey Accelerated Re-search System (MARS) cabled observatory andthe Monterey Ocean Observing System (MOOS)mooring-based observatory. MARS is a testbed forsubsequent cabled observatories, and provides anearly opportunity to develop and demonstrate theAUV docking capability in a logistically benignenvironment. The compatibility of MARS and thedocking system should ensure that the system iswell suited to being deployed on future Regional-scale Cabled Observatories (RCOs). The MOOSmooring system is also capable of supportingdocking, and while MOOS is unable to supply asmuch power to a docking station as a shore-cabledobservatory, it has the advantage of being muchmore easily relocated and deployed in regions ofinterest.

In the tests described below, the AUV was ableto successfully dock and undock repeatedly in theopen ocean, without assistance from the surface.Shore operators downloaded mission data and up-loaded the next mission through an underwaterEthernet link, charged the battery, and completelypowered the vehicle down and back up. Figure1shows the dock and vehicle in a test tank.

Figure2 shows the vehicle entering the dock inthe Monterey Bay where the AUV was tested ina near-shore, disturbance-intensive environment.The dock worked well despite the fact that itwas designed for a permanent deployment in adeeper more benign environment at the MARScabled observatory at 900 m depth in MontereyBay. This deeper MARS site is free from surfacewave motion and the intensive bio-fouling evidentin Figure2.

II. D OCKING SYSTEM OPTIONS

In past years, various research groups havedeveloped different and innovative dock designs.Some of these require a hovering vehicle, suchthe intervention-style grappling system describedin [1]. Another hovering vehicle has a tail hookand wire arrangement, where the vertical thrusterslower the vehicle into a cradle [2]. The designs of

docks for flying vehicles, which rely on sustainedforward motion for control, generally rely on ei-ther a pole or funnel. In the pole design [3], thevehicle carries a latching mechanism that grabson to a vertically moored pole or cable. In thefunnel or cone design, the vehicle enters a circularfunnel and is channeled into a tube where it iscaptured. There have been several designs of thistype for smaller vehicles; the Remus/LEO15 sys-tem [4], [5], the Flying Plug [6], and the KRISO-KORDI [7]. To the best of our knowledge, therehas never been a cone dock for a 21” (54 cm)vehicle that worked in the ocean.

We considered three classes of dock forMBARI’s torpedo-like flying vehicle: the pole, theweathervaning cone, and the fixed cone. The poleis omni-directional, and eases the constraint ondepth control, but requires the vehicle to carryan external, complex, motorized latching device.Establishing the power and data connection isdifficult since the docked vehicle is not rigidlyattached to the pole. A weathervaning cone is alsoomni-directional, but securely captures the vehicle,making power and data connections easier. How-ever the dock must communicate its heading to thevehicle prior to docking, requiring a communica-tion system. This dock must also transfer powerand data across a moving interface, most likelyusing a slip-ring.

III. T HE AUV

The Docking AUV is a Dorado/Bluefin 21”(54 cm) diameter type, which has a Gertler bodyshape, Series 58 Model 4154, with a cylindricalmidsection inserted at the point of maximum di-ameter, and a ring-wing tail [9], [10]. The vehicleis 358 cm long, and weighs 640 kg including en-trained water. The vehicle has the capacity to carrythree batteries, although during the short durationdocking sea-trials, it typically carried only one.Each battery provides 2 kWh of energy, giving amaximum range of approximately 70 km at 1.5m/s with all three batteries present. The vehiclemust maintain a minimum speed of approximately0.8 m/s for controllability and to counteract thestatic buoyancy.

The ring-wing propulsion unit, shown in Fig-ure 3, generates 52 N of thrust at 300 rpm. The

outer ring is attached to the body by 8 equally-spaced struts, which are small foils with a twist orpre-swirl. See TablesIII andIV for numerical val-ues, and [10] for a photograph and a mathematicalmodel. Both the ring and the struts are based ona NACA 0015 foil profile.

The propulsion unit has a fractional horsepowerDC brushless motor with a 10:1 reduction gearbox. A 2 degree-of-freedom gimbal provides con-trol in both vertical and horizontal directions.

The control system consists of outer and in-ner proportional-integral-derivative (PID) loops forboth heading and depth. The nominal horizontal-plane cross-track control loop is similar to thedocking control presented in Figure8. A fullexplanation of the control system in both verticaland horizontal planes is in [10].

Fig. 3. Solid model of the docking AUV showing USBLhoming system as magenta cylinder in the bow.

The vehicle navigation system is essentiallyDoppler velocity log (DVL)-aided dead reckoning,with periodic global positioning system (GPS)updates, which require surfacing. The expectedaccuracy is 4% of distance traveled (DT), circularerror probability (CEP) with the navigation andcontrol sampling rate of 5 Hz.

The AUV homes to the dock using an USBLsonar transceiver mounted in the vehicle nose, asshown in Figure3. MBARI selected an SonardyneScout inverted Ultra-Short Base Line (USBL)homing system, although other types of homingsystems have been proven, such as an electromag-netic system [8], and optical systems [6] and [7].The USBL has the advantage of much greaterrange, and is also reliable, mature, and commer-

cially available. The only drawback is that theSonardyne Scout USBL selected is relatively largeand takes up valuable space in the nose of theAUV. The USBL provides a bearing to its beaconin both the horizontal and vertical planes with aspecified accuracy of .28 degrees root mean square(RMS). The maximum sampling frequency of theUSBL is 1 Hz due to mostly to the beacon turn-around time and subsequent beacon lockout, whichare 300 and 600 ms respectively. A faster beaconfirmware design is under consideration, whichassumes a single, nearby vehicle. This samplingrate may go as high as 3 Hz.

MBARI has already placed and surveyed over100 USBL beacons in the Monterey Bay and theUSBL can home to any of the appropriate beacons.The USBL is useful as a navigation instrument,since it provides position fixes much like GPS,but without surfacing or stopping.

IV. T HE DOCK

The MBARI design approach keeps the dockas simple as possible and puts the complexityon the vehicle, since it is easier to retrieve thevehicle for maintenance than the dock. The dockis a benthic, fixed-heading cone that minimizesmoving parts and electronics on the dock, and doesnot require the vehicle to carry an external latchingmechanism. It also provides secure parking anda rigid data/power connection. The drawback isthat the vehicle must fly a near-bottom precisionapproach along a fixed flight path, possibly ina cross current, and must therefore contain thenecessary electronics and algorithms. The body ofthe dock is a fiberglass tube of diameter 57 cm,3 cm wider than the vehicle’s diameter.

The dock’s entry is conically shaped, built offiberglass staves held in place by two tubularstainless-steel rings with an entrance diameter oftwo meters. This entrance diameter was selectedbased on:

1) Observations that the vehicle was passingwell within two meters of the transponderduring initial homing tests [11].

2) The predicted position deviation of the vehi-cle described in SectionVIII using the datafrom SectionX.

3) Personal communications from local diversfamiliar with bottom currents in that area.

The dock body, consisting of the cone andtube, is mounted on a 3 m tall aluminum tripod(Figure 1). It contains a pressure housing withthe electronics necessary to support power anddata transfer between the vehicle and the cable,and a sonar beacon for AUV homing. Althoughthe dock body is fixed in heading, it gimbals inpitch and roll, due to a self-leveling pendulumdesigned for deployment at MARS. The forcesexerted by the short-period wave motion causedthe dock to continually pitch up and down withan amplitude of approximately 10 degrees. Thismotion proved to be helpful in preventing the AUVfrom binding during entry. A next-generation dockdesign will likely take more deliberate advantageof this pitching motion to prevent binding, aswell as employing other aids such as mechanicalrollers.

Power transfer and mechanical latching wereboth accomplished by having a peg inserted intothe AUV from the dock. The peg has an inductorembedded in the tip, as does the receptacle. Thesystem then transfers power inductively throughthe sea water without a metal-to-metal connection.The measured power transfer rate was 416 Wof power shore-side, with an efficiency of 48%.The peg and receptacle mechanism and electronicswere a slightly-modified version of a previousdesign [12].

We expect that a next-generation design willreverse the peg and receptacle, placing the movingpeg in the vehicle, in keeping with our designphilosophy of an uncomplicated dock. Even withthe peg located in the dock, the entire dockingsystem has only a single moving part with anassociated drive motor.

The AUV transfers data to the dock usingmodified off-the-shelf wireless Ethernet antennas.One is embedded in the top of the vehicle, withits mate correspondingly located in the top ofthe docking tube. The static vehicle buoyancyholds the antennas nearly in contact, with thepeg aligning the vehicle fore and aft. Seawatertank tests conducted at MBARI achieved a transferrate of 10 Mb/second in seawater with a 2.5 cmantenna separation. The system also worked verywell in the ocean trials.

V. HOMING AND DOCKING SEQUENCE

The general homing and docking sequence con-sists of the following steps:

1) Locate and home to the dock. The vehicleuses its on-board navigation to transit withinUSBL range of the beacon, about 2 km.The USBL is directional with a±85 de-gree field-of-view. The vehicle then homesto the dock using pure pursuit guidance,where the heading control system simplykeeps the vehicle pointed at the beacon. Thisguidance strategy does not compensate forocean current and thus the vehicle can beblown downwind while approaching. Pursuitguidance has the advantage of keeping theUSBL pointed right at the beacon for max-imum signal strength.

2) Compute a position fix. When the USBLattains good signal strength, the vehicle usesits compass heading and the USBL bearingand range to compute a position fix. Basedon previous testing this occurs between 1000and 200 meters away from the dock. Sincethe homing control does not use compassheading or propagated position, a jump inthe propagated position caused by the fixdoes not affect the vehicle control.

3) Fly to the start of the final approach path.The approach path is along the cone cen-terline,about 300 meters out. The approachmay require the vehicle to turn away fromthe dock and temporarily lose USBL con-tact.

4) Execute final approach. The vehicle ap-proaches along the cone centerline using across-track controller instead of pure pursuit.This means that the vehicle will track theapproach path to the cone, and acquire adrift correction angle (crab angle) if theocean current has a lateral component. Thistype of control uses both range and bearingfrom the USBL at 1 Hz, and uses the DVLand compass to dead-reckon between USBLupdates. The vehicle slows to 1.0 m/s about200 m from the dock. This allows time forthe control loop to zero the cross-track error,and also prevents the vehicle from hitting thedock with too much force.

Network present?

Pin Extended?

At Outer Marker?

(Vehicle not in posn andTimeout) or(Pin not extended and Timeout) orNew Mission Running?

Vehicle Moving Freely andNew Mission Running?

Vehicle Moving Freely andSame Mission Running?

Passed Dock?Docking

In Dock

MissedApproach

Undocking

No network andNo motion andTime out?

Pin Retracted?

3 Misses?

No motion andTimed out?

Abort(Surface)

Abort(In Dock)

Mission EndsOperator starts dropweight serverOperator commands Start/Stop chargeOperator loads and initiates new mission

Keep thrust onShut off Usbl pingingSense resolverCommand probe extensionWaitWiggle tailTime out

ExecutingMission

Homing

Retract pinFixed Duration Reverse ThrustWiggle tailFloat up until 3 mTime out

Docking ControlState Transition Diagram

Finite State Machine,Mealy Model

Latched

Fig. 4. State diagram of the mission, homing and docking sequence

5) Latch. The vehicle uses the inductive posi-tion sensor and Ethernet contact to deter-mine if it has fully entered. The code thenraises the peg and latches the vehicle. Thiscode was not fully debugged, but workedabout half of the time in test tank tests. Inthe open ocean tests, shore operators raisedthe pin remotely while watching the vehicleenter the dock on camera.

Figure 4 shows the various possibilities duringthe homing and docking sequence in state-diagramform. Items shown in green text were planned

but not implemented at the time of these tests.The ability of the vehicle to sense a decelerationwithout network presence, that is, to sense that itis stuck either partially in the dock, or somewhereelse, is important and planned for later develop-ment. In the meantime, if the AUV becomes stuck,we plan to wiggle the rudder/elevator, and pulsethe thruster to free the vehicle.

If the vehicle enters the dock, but cannot achieveacceptable fore/aft alignment, it executes an un-docking maneuver, returns to the waypoint, andretries. After three unsuccessful tries, the AUV

aborts the mission and surfaces.The homing and docking steps were tested sep-

arately. Homing was tested in the fall of 2005 incentral Monterey Bay off Moss Landing. Dockingwas tested during the summer and fall of 2006at the Monterey Inner Shelf Observatory (MISO),operated by Naval Postgraduate School (NPS) [16]and Associate Professor Tim Stanton, who gener-ously allowed MBARI full access to the facility.MISO is a cabled facility that provides powerand Ethernet 600 meters offshore, in 12 meters ofwater. The shallow depth makes this site subjectto 9-12 second short-period wave motion, as wellas 12 hour tidal currents. The observatory hasan upward-looking ADCP, and NPS maintains anarchive of current measurements stretching backfor many years.

The docking flight pattern was a rectangle of300 × 50 meters. The approach to the dock atMISO was a corridor about 150 meters wide,defined by kelp beds on either side, with a flightpattern through this corridor.

The position fix update in step2 was notimplemented for the sea-trials described here dueto time limitations. However, it is simple code toadd, and would eliminate surfacing the AUV fora GPS fix close to the dock, which increases risksat MISO due to the proximity of the shore, kelp,and recreational boating traffic.

VI. U NDOCKING

The vehicle was designed to undock by lower-ing the peg and reversing the thruster. The vehicleis unstable in reverse, and therefore thrusts with afixed elevator angle and duration. After the thrustershuts off, the vehicle floats up until reaching adepth of 3 meters, well above and clear of thedock. The vehicle then begins a normal missionwith ahead thrust, which consists of a U-turn anda transit down the corridor to the outer waypoint.

During the sea-trials the shore operators manu-ally lowered the peg and initiated the undockingmission while watching on the camera, becausethe peg control code was not yet reliable. TheEthernet path from the main vehicle computer tothe peg PIC passed through four Wireless AccessPoints (WAPS), and we believe the source of theproblems lay in the four routing tables.

A planned, but unimplemented feature, wasundocking behavior that would pulse the thrusterforward and backward to break loose if the vehiclehad difficulty backing out. Should the vehiclebecome permanently stuck, the positively buoyantdock would contain an acoustic release at the topof the base. A support ship could then trigger therelease and recover both the vehicle and the dock.

VII. H OMING CONTROL LOOP

This section explains the homing control loopthat operates in step1 above. SectionVIII explainsthe docking control loop of step4.

Figure5 shows angle and reference frame def-initions used in the following homing analysis.Angles are defined to be positive clockwise.

�������������� ������������������� �"!

#$

%'&

(�)

*

+

Dock

,

Cross-Track Error

Earth-Fixed FrameDocking Frame

Body Frame

-

. /

Fig. 5. Angle and reference frame definitions

The design and analysis of the homing anddocking control loops are fundamentally basedon the dynamics of the vehicle, described bycoupled nonlinear equations of motion (EOM)such as those in [13]. The vehicle travels mostfrequently in straight and level flight, with smallangular and translational deviations, in which casethe horizontal-plane and vertical-plane dynamicsdecouple. Equation1 is the horizontal-plane, orsway-yaw dynamics linearized about straight andlevel flight. It is fourth order and consists of∆ψ,heading deviation;y, cross-track displacement;v,lateral velocity component in the body frame;and finally r, which is heading rate. See [14],

m− Yv mxG − Yr 0 0

mxG −Nv Izz −Nr 0 00 0 1 00 0 0 1

vry

∆ψ

=

(Yv − YRF − YA)U (Yr −m− xRYRF − xAYA)U 0 0

(Nv − xRYRF − xAYA)U (Nr −mxG − x2RYRF − x2

AYA)U 0 01 0 0 U0 1 0 0

vry

∆ψ

+

YRU

2 + TP

(YRU2 + TP )xR

00

δR −

(Yv − YRF − YA)U(Nv − xRYRF − xAYA)U

10

vc (1)

p. 117, and [10] for derivations. TablesI-V in theappendix define the symbols and numerical valuesof the constants.

The depth control did not require any specialmodifications for docking, and so discussion ofthe vertical-plane dynamics and control [10] arenot repeated here.

The origin of body-fixed coordinate system as-sumed by Equation1, and the values in TablesIand II , is amidships, that is, along the centerlineat half the overall length. It isnot at the center-of-mass or the center-of-buoyancy.

In Equation1, YRF is the sway force coefficientdue to the rudder fixed (RF) in the neutral position,

YRF =12ρA(CLα + CDo)

+12ρAS(CSLα + CSDo). (2)

A is the area of the ring, defined as chord×diameter, that is,

A = cRdR. (3)

AS is the equivalent strut area, defined to be

AS = 2(1 +√

2)cSbS . (4)

CLα is the slope of the coefficient of lift of thering, andCDo is the coefficient of drag. Similarsymbols with subscriptsS andF refer to the strutsand RF/GPS fin, respectively.

Is is common to subsume the rudder-fixed por-tion of the force,YRF , into the stability derivatives.In this presentationYRF is added explicitly in

Equation1, and so the stability derivative valuesin Table II are for the fuselage only.YA is the sway force due to the RF/GPS antenna

fin,

YA =12ρAF (CFLα + CFDo). (5)

AF is the area of the RF/GPS antenna fin.YR is the sway force coefficient due to a de-

flected rudder,

YR =12ρACLα. (6)

The homing control loop implements pure pur-suit guidance, which requires that the vehicle’snose continuously point at the beacon during theapproach. Pure pursuit is a special case of the well-known proportional navigation technique used formissile guidance [15]. Pure pursuit is proportionalnavigation with a navigation gain of unity. Largergains will cause the pursuing vehicle to lead thetarget and intercept faster, but the algorithm ismore complicated. Proportional navigation is un-necessary in our case because the time-to-interceptis unimportant, and also because the vehicle doesnot come directly into the dock, but must mustfirst transit over to a fixed approach path after itgets to with several hundred meters. The dock hasa fixed heading and the vehicle must line up withit, much like an airplane approaching a runway.

When the homing loop is first initiated it usescross-track waypoint control to fly toward a pre-programmed dock location until the USBL re-ceives a response from the beacon. From that

�������

�� ��

LinearizedSway/Yaw

Eqn. (1)-1 ZOH

5 Hz

�����������

�

���

�

���

Gyro5 Hz

DVL5 Hz

Compass5 Hz

USBL1 Hz

�

5 Hz

"!�#%$'&)(+*-,/.1032�465�785:9<;.=0>&)53?3@�AB,/*-CD(

E�F

GH

I

JDK

L MNPO

Software Hardware

Q�RS

T

U:V

WXY[Z]\^

Fig. 6. Pure-pursuit homing control loop

point on, the algorithm uses the last good mea-sured position of the USBL beacon. The algorithmalso does this in the event of USBL dropoutsor failure. During cross-track waypoint control,navigation reverts to DVL-aided dead reckoning.Ocean currents are accounted for, in this case, butthe propagated vehicle position will slowly accrueodometry error due mostly to compass and DVLbias and misalignment.

Figure 6 shows the homing control loop. It ismulti-rate Proportional-Integral-Derivative (PID),with the option for feedback on the lateral speedstatev. The lateral speed feedback term was notused in the final design and consequently is shownin green in Figure6. The cross-track statey,and compass heading stateψ, are unnecessaryfor homing and are therefore uncontrolled andignored.

The reference input∆ψR represents the angularmotion of the dock relative to the vehicle. Ifthe AUV is being carried laterally by current,the body-referenced heading to the dock changes.Thus this control loop needs to be able to track atlow-frequencies. An integrator was included in thecontroller to give a total of two free integrators,

counting the one in the plant where∆ψ = r,as shown by Equation1. This closed loop systemwill track constant angular motion (a ramp) withno steady-state error. Additionally, the steady-statestep response to a lateral current has zero error.

A constant lateral current will cause the homingloop continuously to turn the vehicle so that asseen from above, the vehicle spirals in toward thedock. In an extreme case the vehicle will endup approaching from down wind. This has theadvantage of keeping the beacon centered in theUSBL field of view for maximum signal strengthwhile the vehicle gets close.

Figure7 shows the heading response to avc =25 cm/sec lateral current step, determined by simu-lation using Equation1. Assuming that the vehicleis at least two hundred meters from the dock, thisheading error is insignificant because the controlloop has ample time to recover.vc = 25 cm/secwas the maximum constant cross-current expectedat the test site in non-storm conditions, and isdiscussed further in SectionX.

In addition to displacing the vehicle, a lateralcurrent step exerts a yaw torque due to the leverarm between the lateral center-of-pressure and the

0 20 40 60 80 100 120−0.2

−0.1

0

0.1

0.2

0.3

0.4

0.5

Time, Seconds

Hea

ding

, Deg

rees

Closed Loop Cross−Current Response

Analytic Position Error Loop at 5 HzNumeric Position Error Loop at 0.33333 Hz

Fig. 7. Closed-loop heading Response to a 25 cm/sec. stepdisturbance

center-of-mass. Equation1 models this, as shownby the Nv term in the disturbance (vc) inputmatrix.

The red curve shows the response assuming theouter USBL loop has a sampling period of 3 sec-onds, which it does at the outer limit of it’s range.The inner gyro feedback loop still runs at 5 Hz.As the vehicle approaches, the USBL samplingperiod drops to one second. Figure7 shows thatthe slow outer loop does not significantly degradethe performance.

Between USBL sample periods, the 5 Hz head-ing loop maintains the last bearing determined bythe USBL and the compass.

The control gains were empirically chosen andresulted in a dominant pole pair with a dampingof .77 and a time constant of 17 seconds. Thismeans that the vehicle will travel about 75 metersbefore transients induced from step disturbancesdie out. Since we expect this loop to be in oper-ation from 1-2 km out, these time constants areacceptable. Gain values are in TableVI .

The eigenvalues were determined by substi-tuting analytically-determined stability-derivativevalues into Equation1, and combining the resultwith the control loop shown in Figure6. Numeri-cal values are shown in TablesI-V.

VIII. D OCKING CONTROL LOOP

Figure 8 shows the docking control loop. Itdiffers from the homing control loop because

now the cross-track error state is computed fromthe USBL and fed back. The vehicle will nowfly down a fixed approach path and will not beblown downwind by ocean currents. The integra-tor, which has been moved to cross-track error,will crab the vehicle into any current so as to keepthe cross-track error at zero.

In Figure8 there is a smoothing filter after thewaypoint control, which has a time constant of 2.5seconds. This filter was neglected in the analysisbelow since it is much faster than the slowest rootsof the closed-loop characteristic equation.

Since the vehicle position is now computedfrom the USBL, the navigation is free from theaccruing odometry error inherent in dead reckon-ing. The error is proportional to the range, and thusdecreases as the vehicle closes on the dock. Thisis fundamentally what makes the docking systemwork, and is the reason for having a homingsystem.

The heading of the dock,ψb, was measured afterdeployment, and is a known constant. It is usedinside the USBL feedback loop to convert headingdeviation to cross-track error.

As in the case of the Homing Controller, thiscontrol has a fall back in case the USBL dropsout. It latches the last dock position measuredby the USBL, and continues by DVL-aided deadreckoning.

Control gains were again empirically deter-mined, withKv kept at zero. Values are in Ta-ble VII . The cross-track error step response isdominated by a closed-loop pole pair with damp-ing .29 and a time constant of 55 seconds. Figure9shows simulated cross-track error response to a 25cm/sec current step.

Although the transients damp slowly, the error iswell below the one meter dock aperture radius af-ter 100 seconds, while the vehicle has 100 metersyet to travel before encountering the dock. Thisassumes that the docking control is initiated 200meters from the dock on the approach path, andthe vehicle has also slowed to 1.0 meter/second.The thrust is also reduced to 22 N. These twovalues were used in Equation1 along with thegains in TableVII to produce Figures9 and10.

In addition to crabbing into a constant tidalcross current, the control loop must perform in

�������

�� ��

LinearizedSway/Yaw

Eqn. (1)-1 ZOH

5 Hz

���

�

���

Gyro5 Hz

DVL5 Hz

LinearizedCompass

5 Hz

USBL1 Hz

���

5 Hz ��

��

�

��� �!#"$

Software Hardware

%

&(')+**-,/.

01+2 3

4�576 8:9<;=;:>@?A8CB<DFEGH8C8:9I8

JLK M

Navigation

N

OQP

R�S

T�UVXWAY[Z]\^\`_:acb<dFe�fhgibj_Ckmlon p

q�rcs

t�u:vcw

xzyy�{}|�~C�

SmoothingFilter

Fig. 8. Cross-track docking control loop

0 50 100 150 200 250−1.5

−1

−0.5

0

0.5

1

1.5

2

2.5

3

3.5

Time, Seconds

Cro

ss−T

rack

Err

or, M

eter

s

Closed Loop Cross−Current Step Response; vc = 0.25

Analytic Position Error Loop at 5 HzNumeric Position Error Loop at 1 Hz

Fig. 9. Cross-track error response to a 25 cm/sec step current

the presence of 8-12 second wave motion. Thegreen curves in Figure10 show the steady-statesinusoidal response of cross-track error to lateralcurrent. The frequency region bounded by thevertical magenta lines represents the wave energy.The black curve is the open-loop transfer function,which is coincident with the closed loop transferfunction at the wave energy, showing that thecontrol loop has rolled off there and has no effect.At the low end of the wave spectrum the closed-

loop gain is a factor of 2.5, and decreases to lessthan one at the high end. Typically the wave periodis close to ten seconds, where the loop gain is nearunity, implying that a 30 cm/sec wave motion willdeviate the vehicle less than 30 cm from the flightpath. ADCP data taken during the docking trials,plotted in Fig 15, showed that the wave periodwas 9.3 seconds with3σ amplitude of about 30cm/sec, and this is shown in Figure15.

The dotted lines show the multi-rate transferfunction with the outer USBL loop at 1 Hz, incomparison to the solid lines where 5 Hz wasassumed. The 1 Hz sampling doesn’t appreciablydegrade the performance.

IX. SEA-TRIAL RESULTS

The MBARI team conducted the first at-seahoming trials on 11 December 2005. The testarea had a flat bottom about 15 meters deep,without any nearby obstructions or kelp beds. Thetransponder was moored about 5 meters off thebottom. The vehicle was acquiring the transponderat distances exceeding 2 kilometers, and homingto within 50 cm. See [11] for details.

The USBL performed well during the homingtrials, with little noise and few dropouts,leadingus to conclude that little filtering was necessary.

10−4

10−3

10−2

10−1

100

101−150

−100

−50

0

50Bode Plot of Closed Loop Sway/Yaw System

Mag

nitu

de

y/yR

y/vc

y/vc 1 Hz

y/vc Open Loop

Waves

10−4

10−3

10−2

10−1

100

101−600

−400

−200

0

200

Frequency, Hz

Pha

se, D

egre

es

Fig. 10. Bode plot of cross-track docking control loop

However the noise environment at the MISO dock-ing site was much worse, due perhaps to theproximity of the shoreline, the surf, and nearbykelp beds.

The noisy USBL data at MISO was significantin that it was causing the vehicle to consistentlymiss the dock by 3-10 meters. This led to thedesign of a moving-window median filter whichproved to be very effective. The vehicle madethe first controlled dock after implementing thefilter, and successfully docked on every attemptthereafter. There were four consecutive dockings.

A median filter can be effective for sonar appli-cations because it rejects multipath, as long as thenumber of measurements corrupted by multipathare fewer than the uncorrupted ones.

Before median filtering, the USBL measure-ments were passed through simple outlier rejectiontests. The data had to pass all of the below to beconsidered valid:

1) There must be a preset number of goodconsecutive samples.

2) The vehicle must not be turning too fast.3) The measured bearing must not be too great.

These tests are not switched on until the vehicleis near enough to the transponder that the signalis strong.

The median filter then consists of taking the last11 valid USBL measurements, and computing themedian of each axis individually. The navigationfirst converts the measurements to UTM locations,

−200 −100 0 100 200

−250

−200

−150

−100

−50

0

50

100

Meters, East

Met

ers,

Nor

th

Initial UTM Coord: North = 4052097.4935 East = 600569.1043 Data Set: 2006.255.09

Navigated PositionGPSUsbl

Fig. 11. Vehicle track during a docking trial. The + pointsare USBL measurements of the dock location, color codedby distance.

so that each measurement appears as a fixedlocation on a map, such as those in Figures11and12. Median filtering the eastings and northingsseparately has the interesting effect of producinga location that is not the same as any of themeasurements, yet is contained within the largestcluster.

50 60 70 80 90

−285

−280

−275

−270

−265

−260

−255

−250

Meters, East

Met

ers,

Nor

th

Initial UTM Coord: North = 4052097.4935 East = 600569.1043 Data Set: 2006.255.09

Navigated PositionGPSUsbl

Fig. 12. Enlargement of the USBL dock hits. The +’s areraw data, and the x’s are the filtered result.

Figure11 shows the vehicle path, as seen fromabove, of the second docking mission. The axesare eastings and northings (UTM) in meters. Themagenta points at the start, near (0,0), are GPShits. The vehicle submerges, turns and flies north

to the first waypoint, then turns again to thesouth and passes through two more waypoints.The color-coded diamonds overlaying the pathshow the location of the vehicle at each USBLmeasurement. The color shows the distance fromthe transponder, and corresponds to the color ofthe measured location indicated by a+. Black×’sshow the location of filtered points.

Figure 12 shows a zoom of the dock locationin Figure 11. There is a distinct cluster of pointsabout 8 meters to the west of the dock, detectedby the USBL when the vehicle was around 200meters out. The error is presumed to be due toa bounce (multipath). This cluster contains manyvalid hits. As the vehicle closes, however, thiscluster is discarded by the moving window andthe valid hits then cluster at the dock, precisely asdesired.

The position error of the vehicle when it en-ters the dock, as viewed on the video, is nearlyundetectable, appearing less than 5 cm.

The jog after the third waypoint, at (0,-70), iscaused by the initiation of the docking controlloop. It turns the vehicle to the east to bring itonto the fixed flight-path heading ofψb = 171degrees. From that point in, the cross-track erroris computed directly from USBL measurements,and not from propagated position.

The top strip of Figure13 shows yaw com-mands during the approach to the dock. The greenline is commanded bearing to each of the threewaypoints, followed by a flight path approachof 171 degrees, beginning at 291 seconds. Theapproach bearing is labeledψb in Figure8. The redline in Figure13 is the filtered heading command,and the blue is the measured. They correspond toψR = ∆ψR +ψb, andψ = ∆ψ+ψb, respectively.The commands to turn the vehicle from eachwaypoint to the next are clearly visible at 128and 227 seconds, withψ tracking as expected.The maneuver at 291 seconds is interesting, whereψR increases, but the vehicle commanded headingappears to turn the opposite direction. This isbecause the docking controller has activated andmust now slew the vehicle not only onto thespecified bearing ofψb = 171 degrees, but mustdo so on the approach path. To do this, the dockingcontroller must first turn the vehicle in the opposite

direction to get onto the line, before turning backto the specified bearing.

The middle strip of Figure13 shows headingdeviation∆ψ, which corresponds directly with thesame symbol in Figure8. The green curve is thecrab angle, commanded by the integrator. It isapparent that there is little cross current during thisapproach, since the crab angle approaches zero asthe control reaches steady-state just before each ofthe two waypoint changes.

The character of the blue∆ψ curve changeswhen the waypoint control activates at 291 sec-onds. This is because the cross-track error was pre-viously computed from the propagated position,determined from DVL-aided dead reckoning. After291 seconds the cross-track error is computedfrom the filtered USBL measurements and thecompass. Several jumps in heading deviation areapparent between 320 and 450 seconds due toUSBL noise.

The bottom strip shows the cross-track error. Itis converging to under a meter after 400 seconds.Interestingly, the USBL apparently got some hitsthat took the vehicle almost a meter off-track at490 seconds, but then the measurements improveduntil the vehicle had zero measured cross-trackerror only about 4 seconds before contact. Themeasured error at the moment of contact, at 519seconds, was 10 cm, slightly larger than wasapparent in the video.

There were three additional docking events,each appearing much like this one, indicating thatthe docking control is repeatable, at least in thecross-current conditions shown in Figure15

Depth control is unchanged from the standardalgorithm as described in Ref. [10]. The vehicleflew at 12 meters depth during it’s final approach,until about 50 meters out, where it transitioned toa 3 meter altitude track. The dock is located on aslight slope and is 3 meters high in 12 meters ofwater.

X. DOCK TEST ENVIRONMENT

Figure 14 shows statistics of cross-shore andalong-shore currents at the MISO docking testsite during July and August 2005, which is arepresentative period for the docking trials. Themeasurements are at 1.8 meters altitude. Thealong-shore current is of primary concern because

0 100 200 300 400 500 6000

200

400Commanded and Smoothed Bearing; Data Set: 2006.255.09

Deg

rees

0 100 200 300 400 500 600−20

0

20

Deg

rees

Heading Deviation ∆ψ

0 100 200 300 400 500 600−2

0

2

Elapsed time (s)

y, m

eter

s

Cross−Track Error;

Fig. 13. Cross-track error, heading deviation, and headingduring the docking event

it is perpendicular to the approach path, and willtend to carry the vehicle off track. The vehicleapproach altitude was 3 meters, where the currentwas expected to be larger, and so the design hadto allow for currents greater than shown.

Blue points are the mean current over a six-hourperiod, which shows the tidal component withwave motion averaged out. The sampling periodwas between 1 and 2 seconds. Tidal currents actas a constant during a docking approach since anapproach takes only a few minutes.

The standard deviation of the tidal componentis 3.8 cm/sec, with a maximum surpassing 5cm/sec. The constant-current design specificationwas consequently taken to be 25 cm/sec, keepingin mind that the currents tend to be faster furtherabove the sea bed, and allowing a margin.

Figure 15 shows a 3 minute segment of thefull 1 second data set of along-shore current mea-surements. This data was taken on the day thatthe AUV docked. At 3 meters altitude, the tidal(average) current was 2.0 cm/sec to the north,with a 9.3 second wave component, with standardamplitude deviation ofσ = 9.4 cm/sec.

The figure shows that the short-period wavecurrent tends to increase with altitude, justifyinga design margin for current greater than what wasmeasured at 1.8 meters in Figure14.

180 182 184 186 188 190 192 194 196−5

0

5

10

15

Cm

/Sec

2005 Cross Shore Current

180 182 184 186 188 190 192 194 196−10

−5

0

5

10

Day Number

Cm

/Sec

2005 Along Shore Current

MeanStd. Deviation

Fig. 14. 6 Hour Current Statistics at 1.8 Meter Altitude

Seconds

Alti

tude

, Met

ers

Along−Shore (Sway) Velocity Component; t0 = 19.60 hours Z

0 50 100 150

4

6

8

10

12

14

−0.5

−0.4

−0.3

−0.2

−0.1

0

0.1

0.2

0.3

0.4

0.5

Fig. 15. Along-shore current as a function of altitude atMISO on the moment of the docking event in Figures11-13,12 September 2006 (day 255)

XI. CONCLUSIONS

A prototype AUV docking system has beendesigned, built and tested at sea. The dock wasattached to the MISO cabled observatory in south-ern Monterey Bay, and used to support dockingoperations with a Dorado 21” (54 cm) diameterAUV. All the of the capabilities required for adocking capability for seafloor observatories hasbeen demonstrated, including:

1) homing and capture of the AUV in thedocking station

2) establishing an Ethernet connection betweenthe vehicle and the dock using a short rangeradio frequency communication system

3) complete interactivity with the vehicle in-cluding recovering mission data to shore,downloading new missions to the vehicle,and even adding new code to the vehicle andcompiling in situ

4) inductive transmitting power to the vehicle5) charging vehicle batteries from observatory

power6) turning the vehicle off, and then waking it

up from the dock7) undocking the vehicle from the dock, and

commencing missions.

ACKNOWLEDGMENTS

This work would not have been possible withoutthe sustained effort of the entire MBARI Dock-ing team,including Jon Erickson, Lance McBride,Thomas Hoover and Farley Shane, as well asimportant contributions from Hans Thomas, RichHenthorn, Judith Connor (MBARI), Steve Rock(Stanford University), and others. We would par-ticularly like to thank Tim Stanton (Naval Post-graduate School) for allowing us full and unre-stricted access and use of the Monterey InnerShelf Observatory, which provided the underseapower and data cable, and infrastructure requiredto support the dock. We gratefully acknowledgethe support of the Packard Foundation. This workwas supported in part by the National ScienceFoundation under NSF-OPP 9910290. Any opin-ions, findings and conclusions or recommenda-tions expressed in this material are those of theauthor(s) and do not necessarily reflect those ofthe National Science Foundation.

REFERENCES

[1] J. Evans, P. Redmond, C. Plakas, K. Hamilton, andD. Lane, “Autonomous docking for intervention-styleauvs using sonar and video-based real-time 3d poseestimation,” in MTS/IEEE Oceans 2003 ConferenceProceedings, San Diego, CA, September 2003, pp.2201–2210.

[2] T. Fukasawa, T. Noguchi, T. Kawasaki, and M. Baino,“Marine bird - a new experimental auv with underwaterdocking and recharging system,” inMTS/IEEE Oceans2003 Conference Proceedings, San Diego, CA, Septem-ber 2003, pp. 2195–2200.

[3] H. Singh, J. G. Bellingham, F. Hover, S. Lerner, B. A.Moran, K. Heydt, and D. Yoerger, “Docking for an au-tonomous ocean sampling network,”Journal of OceanicEngineering, vol. 26(4), pp. 498–513, October 2001.

[4] B. Allen, T. Austin, N. Forrester, R. Goldsborough,A. Kukulya, G. Packard, M. Purcell, and R. Stokey,“Autonomous docking demonstrations with enhancedremus technology,” inMTS/IEEE Oceans 2006 Confer-ence Proceedings, Boston, MA, September 2006, pp.1–6.

[5] R. Stokey, B. Allen, T. Austin, R. Goldsborough, N. For-rester, M. Purcell, and C. V. Alt, “Enabling technologyfor remus docking: an integral component of an au-tonomous ocean sampling network,”Journal of OceanicEngineering, vol. 26(4), pp. 487–497, October 2001.

[6] S. Cowen, S. Briest, and J. Dombrowski, “Underwaterdocking of an autonomous undersea vehicle using opti-cal terminal guidance,” inMTS/IEEE Oceans 1997 Con-ference Proceedings, Halifax, Canada, October 1997,pp. 1143–1147.

[7] P. M. Lee, B. H. Jeon, and S. M. Kim, “Visual servoingfor underwater docking of an autonomous vehicle withone camera,” inMTS/IEEE Oceans 2003 ConferenceProceedings, San Diego, CA, September 2003, pp.2195–2200.

[8] M. D. Feezor, F. Y. Sorrel, P. R. Blankenship, andJ. G. Bellingham, “Autonomous underwater vehiclehoming/docking via electromagnetic guidance,”IEEEJournal of Oceanic Engineering, vol. 26, no. 4, pp. 515–521, 2001.

[9] M. Gertler, “Resistance experiments on a systematicseries of streamlined bodies of revolution – for appli-cation to the design of high-speed submarines,” NavyDepartment, The David taylor Model Basin, Report C-297, Apr 1950.

[10] R. S. McEwen and K. Streitlien, “Modeling and controlof a variable-length auv,” in12th International Sym-posium on Unmanned Untethered Submersible Tech-nology, University of New Hampshire, Durham, NH,August 2001.

[11] J. G. Bellingham, R. S. McEwen, and B. W. Hobson,“The development of a docking station for a regionalcabled observatory,” in4th International Workshop onScientific Use of Submarine Cables, Dublin, Ireland,February 2006.

[12] R. Coulson, J. Lambiotte, and E. An, “A modular dock-ing system for 12.75-inch class auvs,”Sea Technology,pp. 49–54, April 2005.

[13] M. Gertler and G. Hagen, “Standard equations of motionfor submarine simulation,” David W. Taylor Naval ShipResearch and Development Center, Report 2510, 1967.

[14] T. I. Fossen,Guidance and Control of Ocean Vehicles.New York: Wiley & Sons, 1994.

[15] M. Guelman, “A qualitative study of proportional navi-gation,” IEEE Transactions on Aerospace and ElectronicSystems, vol. AES-7, pp. 637–643, 1971.

[16] T. Stanton, “Monterey inner shelf observatory webpage,” http://www.oc.nps.navy.mil/∼stanton/miso.

APPENDIX

m 640.3 kg MassIzz 475.0 kg-m2 InertiaxG .119 m CG offsetxR -1.79 m Dist. to rudderxA -1.03 m Dist. to antennaU 1.54 m/s SpeedTp 52 N Thrustρ 1025 kg/m2 Density of seawater

TABLE I

GENERAL PARAMETERS AND MASS PROPERTIESIN

EQUATIONS 1-6 FOR THEHOMING CONTROL

Yv -595.0 kg Added massYr -57.91 kgNv -57.91 kg-m2 Added InertiaNr -381.4 kg-m2

Yv -106.0 kg/m Force deriv.Yr 37.90 kgNv -494.9 kg Munk momentNr -288.1 kg-m Torque deriv.

TABLE II

STABILITY DERIVATIVE VALUES IN EQNS. 1-6

dR .381 m DiametercR .127 m ChordCLα 4.80 radians−1 Coef. of lift slopeCDo .012 none Coef. of drag

TABLE III

HYDRODYNAMIC VALUES FOR RING WING IN EQNS. 1-6

bS .127 m SpancS .043 m ChordCSLα 6.28 radians−1 Coef. of lift slopeCSDo .012 none Coef. of drag

TABLE IV

HYDRODYNAMIC VALUES FOR RING WING STRUTS IN

EQNS. 1-6

bF .254 m SpancF .0762 m ChordCFLα 4.4 radians−1 Coef. of lift slopeCFDo .005 none Coef. of drag

TABLE V

HYDRODYNAMIC VALUES FOR GPS/RF ANTENNA FIN IN

EQNS. 1-6

Kp 0.4 none Position gainKr 2.0 seconds Rate gainKv 0.0 radians/meter/second Lateral speed gainKI .02 seconds−1 Integral gainT .2 seconds Sampling period

TABLE VI

HOMING CONTROL GAINS IN FIGURE 6

Kp 1.0 none Position gainKr 2.0 seconds Rate gainKv 0.0 radians/meter/second Lateral speed gainKwp .05 radians/meter Wpnt. posn. gainKIwp .003 radians/second/meter Wpnt. intgrl. gainT .2 seconds Sampling perioda 1/2.5 seconds−1 Filt. nat. freq.K 1− eaT none Filt. gain

TABLE VII

DOCKING CONTROL GAINS IN FIGURE 8