Proceedings

July 10th, 2009

2nd International Robotic

Sailing Conference

Autonomous Robotic Boat of ENSIETA 1Jan Sliwka, Pierre-Henri Reilhac, Richard Leloup, Pierre Crepier, Henry De Malet, Patrick Sittaramane, Fabrice Le Bars, Kostia Roncin, Bruno Aizier, Luc Jaulin

Development of the USNA SailBots (ASV) 9Paul Miller, Owen Brooks, Matthew Hamlet

Design and Construction of the AutonomousSailing Vessel AVALON 17

Lian Giger, Stefan Wismer, Sebastian Boehl, Gion-Andri Büsser, Hendrik Erckens, Jürg Weber, Patrick Moser, Patrick Schwizer,Dr. Cédric Pradalier, Prof Dr. Roland Y. Siegwart

Technologies for Autonomous Sailing: Wings and Wind Sensors 23

Mark Neal, Colin Sauzé, Barry Thomas, José C. Alves

Communication Architecture for Autonomous Sailboats 31Roland Stelzer, Karim Jafarmadar

Model Sailboats as a Testbed for Artificial Intelligence Methods 37

Ralf Bruder, Birgit Stender, Alexander Schlaefer

AAS Endurance: An Autonomous Acoustic Sailboat forMarine Mammal Research 43

Holger Klinck, Roland Stelzer, Karim Jafarmadar, David K. Mellinger

Contents

1

Autonomous Robotic Boat of ENSIETAJan SLIWKA, Pierre-Henri REILHAC, Richard LELOUP, Pierre CREPIER, Henry DE

MALET, Patrick SITTARAMANE, Fabrice LE BARS, Kostia RONCIN, Bruno AIZIER, LucJAULIN et al.

E-mail: "name's �rst 6 letters + surname's �rst 2 letters"@ensieta.frexample: sliwkaja@ensieta.fr

(WRSC/IRSC-2009 Paper)

Abstract�This year, we launched MicroTransatProject in our school to prepare the next year transat-lantic race. In this article we will talk about thesolutions in mechanics, electronics, sailing strategiesand simulation that we developed for our autonomousrobotic boat. As for the mechanics, the hull is home-made using mainly glass �ber mat bound togetherwith a resin binder. As for the electronics, we triedto use off-the-shelf components as much as possibleto ensure the maintainability of the system. In orderto test the sailing algorithm we are using a simulatormade with SCILAB.

Index Terms�Autonomous, Sailing boat,WRSC/IRSC 2009.

Fig. 1. Our robotic boat

I. ABOUT OUR PARTICIPATION IN THE WRSCThis year, our school ENSIETA (French Graduate

Engineering School) has launched the project ofparticipating in the MicroTransat challenge. Manyfactors helped in the project creation. First, ENSI-ETA is located in Brest, a city on the North-Westof France on the Atlantic ocean shore. Moreover,our school is multidisciplinary since there are de-partments of informatics, electronics and mechan-ics and particularly the sub-department of navalarchitecture. Because ENSIETA take part of thechallenge for the �rst time, we will not participatethis year in the real challenge. However, we willbe participating next year for sure (if nobody hadcrossed the Atlantic of course).

Fig. 2. Brest harbour

In order to gain some experience and �nd spon-sors for the real challenge, we decided to makea smaller intermediate private challenge of au-tonomously crossing the Brest harbour (see �gure

IRSC'09 - page 1

2

2). For that purpose we are building a small boat1m20 long that will only be able to do shortdistances. As for an of�cial challenge, instead ofMicroTransat , the WRSC will be the �rst of�cialchallenge for our small boat.

II. OUR FIRST ROBOTA. Mechanical design and construction

Fig. 3. IMOCA Class

Our sailing robot design is based on the IMOCAclass design (race boats see �gure 3). We only madethe mast smaller in order to enhance its stability.The next sub-parts will talk about the mechanical

architecture.

Fig. 4. Created using Delftship (free version)

1) Hull construction: In order to build the hullwe used the following steps.

� Create a mould with a form corresponding tothe hull using plaster.

� Cover the inside of the mould with a specialresin in order to be able to remove the hullfrom it

� Put some Gel Coat and obtain the result in�gure 5-(a).

� Put 2 layers of glass �ber mat bound togetherwith a resin binder (�gure 5-(b)).

� Extract the hull (�gure 5-(c)) and add theinternal structure elements using wood andglass �ber mat bound together with a resinbinder (�gure 5-(d)).

Fig. 5. Hull construction

2) Waterproofness: The deck normally seals wellenough the inside of the boat. However, we wouldlike to be able to remove the electronics fromthe inside of the boat for maintenance and debugpurposes. This creates waterproofness problems. Wesolved this problem using a waterproof box andSwitchcraft waterproof connectors (see the �gure6). In fact, we cut a rectangular hole in the deckin which we put the box. Then we cut the box'sbottom and glued the borders of the box to the deckwith special glue. The communication between theexternal sensors and the inside electronics is made

IRSC'09 - page 2

3

through waterproof connectors.

Fig. 6. Waterproofness

3) Mast: As for the mast, we used an aluminummast ordered in Germany. It is a pro�led mast witha groove for the sail.4) Sail: The sail is not triangular but has the

form as in the �gure 7 in order to lower the windforce application point keeping the same ef�ciencybut at the same time rising the stability of the boat.

Fig. 7. Sail

5) Rudder blade: Like in the IMOCA classboats, there will be two rudder blades to allowsteering even when the boat is leaning sideways.

6) Servos: Because the boat is designed for shortdistances, we do not need resistant servomotors. Weused rc-model servos.7) Keel: The keel's lead bulb was cast using a

mould as you see the �gure 8.

Fig. 8. Bulb mould

B. Autonomous boat electronics1) Solutions:

Fig. 9. Electronic architecture

While designing our electronics (�gure 9), wehave �rst thought of the solution shown in the �gure10.Even if we didn't chose this approach, it is an

interesting one. In fact, the solution is based on twoprinciples:

IRSC'09 - page 3

4

� Use of COTS (Commercial off-the-shelf) ele-ments

� Use most integrated elements (that do most ofthe tasks as the Neo and the Labjack)

In fact, the Neo Freerunner is a mobile phonewith an LCD touch screen. Besides a GSM commu-nication module, the Neo has WIFI and Bluetooth,useful for debugging and con�guration, a GPS, 2Accelerometers and an embedded Linux Debianallowing an easy programming (In C/C++ languageusing standard libraries).

Fig. 10. Old solution

The other "universal" component is the Labjackthat makes the interface between the Neo andthe Sensors/Actuators. This component has severaldigital I/O, ADC (Analog to Digital Converter) andDAC (Digital to Analog Converter). It can alsoconnect to an I2C bus.We were able to make this solution work properly

with a robotic car. But the problem is the complex-ity of the Neo. In fact, because there is an embeddedLinux, the system has to boot when powered. Thiscauses problems since sometimes the system werenot working properly after rebooting. However, thisapproach is promising since it provides with high

quality integrated systems at a good price. Besides,the development of electronics becomes very simplesince we can add a new sensor just by connectingit to the Labjack and writing the program thatmanages the data acquisition.As for our current architecture, we used more

components that have exactly the same behaviorwhen rebooted and have lesser power consumption.In fact, we are preparing the electronics for thetransatlantic race where rebooting might be frequentdue to frequent power shortage.

Fig. 11. New solution

2) GPS: As for the GPS we used the EB-85A(FV-M8). In fact, this GPS is small and has goodprecision.3) Compass: We used the HMC6352 Compass

module since it has one degree resolution.4) Intelligence: We used a PIC18F2550 from

Microchip mounted on the 28-pin Pickit demoboard [8]. We have chosen the PIC since it is morerobust to power shortage (no booting) and use lessenergy than the Neo Freerunner.5) Energy source: As for now, the energy source

are lithium polymer batteries (�gure 12) from RC-Systems that will be suf�cient for crossing the Brest

IRSC'09 - page 4

5

harbour and might even be suf�cient to do the 48hWRSC race. Regarding the progress of the project,we will start the development of the solar energymodule.

Fig. 12. Lithium Polymere battery

6) Communication: Even if the boat is supposedto be autonomous, we need a way to communicatewith it for debug and con�guration purposes. Weuse two Adeunis ARF53 HF modems (one oneach side). This modem is long-range and can becontrolled by the PIC and the base PC throughRS232. At �rst, we wanted to use the GSM/GPRS(�gure 13-(b)) but we would have to pay for thesubscription fee for the GPRS data line both forthe boat phone and the PC modem so we gave upon this solution.As for the MicroTransat challenge we might

use IRIDIUM modem (�gure 13-(c)) or the SPOTmessenger(�gure 13-(a), found at the MicroTransatmailing list!). In fact, the SPOT messenger is alot cheaper and gives the possibility to track theposition of the robotic boat using E-mail or GSMphone.

Fig. 13. Communication modules

7) Servo control: At �rst, we wanted to controlthe servos directly from the PIC but it used toomuch of its resources so we decided to add a

separate servo controller. As for now, we will use ahomemade servo controller. In order to increase thereliability, we might use COTS IC as the POLOLUor PARALLAX (�gure 14-(a) and 14-(b) resp.).

Fig. 14. Servocontrollers

8) Anemometer: We will use the CV7 ultrasonicanemometer from LCJ Capteurs. We avoid usingmechanical anemometers since there are more likelyto break.9) Reliability: We are trying to make the boat

able to move forward using only the GPS. The othersensors would only increase the accuracy of theactions. As an example, if the anemometer stopsworking, we will have to infer the wind orientationfrom the GPS position/speed and the COMPASS.The trick is to stop the boat. A immobile boat willautomatically take a speci�c orientation from whichwe can infer the wind orientation.We are still working on that kind of aspects so

we do not have any results yet.

C. Sailing algorithmIn order to test our algorithms, we are using a

simulator written in SCILAB language. The sailingboat represented is described by the following stateequations8>>>>>>>>>>>>><>>>>>>>>>>>>>:

_x = v cos �; (i)_y = v sin � � �V; (ii)_� = !; (iii)_�s = u1; (iv)_�r = u2; (v)_v =

fs sin �s�fr sin �r��fvm ; (vi)

_! = (`�rs cos �s)fs�rr cos �rfr���!J ; (vii)

fs = �s (V cos (� + �s)� v sin �s) ; (viii)fr = �rv sin �r: (ix)

IRSC'09 - page 5

6

The inputs u1 and u2 of the systems are thederivatives of the angles �s and �r. The state vectorx = (x; y; �; �s; �r; v; !)

T 2 R7 is composed with� the coordinates x; y of the inertial center G ofthe boat

� the orientation �,� the sail angle �s� the rudder angle �r� the tangential speed of G� the angular velocity ! of the boat around G.The intermediate variables are� the thrust force fs of the wind on the sail,� the force fr of the water on the rudder.The parameters (that are assumed to be known)

are� the speed V of the wind,� the distance rr between the rudder and G,� the distance rs between the mast and G,� the rudder lift �r,� the sail lift �s,� the tangential friction �f of the boat withrespect to the water,

� the angular friction �� of the boat with respectto the water,

� the angular inertia J of the boat,� the distance ` between the mast and the thrustcenter of the sail,

� and the drift coef�cient �.These parameters will be chosen as

� = 0:05; rs = 1; rr = 2; V = 10;

m = 1000; J = 2000; �f 2 60;�� 2 500; �s = 500; �r = 300:

We simulate the behaviour of the boat using Eulerapproximation.

x(t+ dt) = dt � f(x) + x(t)

f represents the state equations.In order to be able to develop strategies, we �rst

need a regulator that regulates the boat's sail angle�s and orientation � to a speci�c target (�s; �). (seearticle [1] for more details)

u = r (x;w) = r(x;�s; �)

As for the strategies, in order to sail to a speci�cwaypoint, we can use a hybrid second stage reg-ulator. In �gure 15, we can see the four differentdirections �i to follow that will be chosen by theregulator with regards to the position of the boatwith regards to the target. Denote by �hr the currentdirection to be followed by the hybrid regulator. Forexample, if the boat is in zone q = 3 then �hr = �3.The target orientation � will the �ltered response of�hr with a �rst order �lter in order to avoid brutaltransitions of �. The sail angle depends directly ofthe orientation �s = h(�).

Fig. 15. Four possible target orientations

Some results of the SCILAB simulatorare displayed on �gure 16. The scilab �lecan be found on the following addresshttp://www.ensieta.fr/sliwka inthe MicroTransat section.

D. Real tests and debuggingIn order to be able to develop the electronics

at the same time as boat construction, we decidedto test the electronics and the different algorithmsusing a robotic car as shown in �gure 17.

IRSC'09 - page 6

7

Fig. 16. Simulation

Fig. 17. Robot for electronics debuging

III. CONCLUSIONThe boat is already working but we will still have

to do many tests and improvements and re�ne thewinning strategy ;).

References

[1]L. Jaulin (2004) Modélisation et commanded'un bateau à voile, CIFA2004 (Conférence In-ternationale Francophone d'Automatique), Douz(Tunisie)

IRSC'09 - page 7

IRSC'09 - page 8

Development of the USNA SailBots (ASV) Paul Miller1, Owen Brooks2, Matthew Hamlet3

1-2Naval Architecture and Ocean Engineering Department 3Systems Engineering Department

United States Naval Academy Annapolis, Maryland, USA

1phmiller@usna.edu 2oebrooksiii@gmail.com

3matthew.r.hamlet@gmail.com

USNA SailBot #1

Abstract— The development of two 2-meter long autonomous, sail-powered, surface vessels at the United States Naval Academy are described. Key design features and characteristics are presented along with supporting research and relevant background information. Efforts in naval architecture research focussed on velocity prediction program trade-off studies on beam, displacement and stability versus sail area. Systems development included gps-based navigation and vessel control operated through a Rabbit 3000 microprocessor. Keywords— autonomous surface vessel, SailBot, velocity prediction program (VPP)

I. INTRODUCTION Autonomous surface vessels (ASV) provide opportunities

in surveillance, monitoring and oceanographic research. A requirement of these vessels is the need for power for propulsion as well as control and communications. For long-term endurance on-board storage for traditional fuel sources is problematic, so the energy must be harvested while at sea. While many options are available this paper describes the development of traditional, small, sail-powered ASVs. The mission statement for the vessels described in this paper is only toward competition rather than any specific scientific, military or commercial task.

In 2004 Erik Berzins, an engineering student at the University of British Columbia began developing a small sail-powered ASV for a class assignment. A requirement was that the project could be used in a student competition. After contacting other Canadian universities a set of rules for a “SailBot” competition were developed and the first event was held at Queen’s University in Canada in 2006. The rules limit boats in the SailBot Class to two meters in length, three meters in beam (allowing for multihulls), 1.5 meters in draft and 5 meters in height from the bottom of the keel to the top of the fixed mast (not including wind instruments)[1]. The relatively small size allows for easy transportation and handling on shore while also keeping the construction and shipping costs down. Competition is intended for undergraduate students and the contests include a design presentation along with on-the-water events that test navigation, station keeping, performance and endurance[2].

The United States Naval Academy (USNA) started a team in January 2007 through the efforts of Jake Gerlach, a junior majoring in naval architecture and Associate Professor Paul Miller of the naval architecture major. With the assistance of Associate Professor Brad Bishop of the systems engineering major a team was created and funding secured for the following academic year. The USNA team comprising students majoring in naval architecture and systems engineering designed and built a boat the following year for the 2008 SailBot competition. Based on the lessons learned from that event and further research, the team designed and built a second boat for the 2009 competition.

IRSC'09 - page 9

While the mission statement for both boats was to win the SailBot competition, the team has a secondary goal to develop a small, sail-powered ASV for long distance passages. With the knowledge that the current holder of the “smallest vessel to sail across the Atlantic” is a mere 5’4”[3], the team is committed to an endurance vessel that also meets the SailBot Class rules. To date the team has spent approximately 900 man-hours and US$16K on developing the two boats.

II. VESSEL DESCRIPTION – NAVAL ARCHITECTURE While the two USNA boats have names (First Time and

Luce Canon respectively), for simplicity in this paper the hulls will be referred to as Boat 1 and Boat 2. Similarly, the keels, rigs and sails will be named according to their chronological design and construction.

Table 1 shows the two boats’ principal characteristics with their largest rigs. The influence of the SailBot Class rules is clearly seen in that both boats’ designs reflect the performance enhancing characteristics of maximum length and stability (via maximum draft).

TABLE I PRINCIPAL CHARACTERISTICS

Boat 1 Boat 2LOA m 2 2LWL m 2 2Beam m 0.36 0.28Draft m 1.5 1.5Sail Area m2 3.1 3.1Disp kg 26.7 24Cp 0.57 0.54LCB -53% -55%LCF -55% -57%"SA/Disp" 35.7 37.9"L/Disp" 6.7 7.0

Figure 1 shows the two boats’ hull lines with Boat 1 on the

top. The primary differences between the two boats are the trend in the newer boat to a narrower beam and reduced hull weight through lower freeboard amidships. The two boats have approximately the same ballast weight concentrated in a bulb located at the bottom of the keel. Boat 2 also has more V-shaped forward sections.

The two hulls and decks are constructed similarly based on a trade-off between available time, weight, the students’ building skills and the tools available to them. The hulls are skinned with one layer of #282 (T300) carbon cloth on both the outside and inside of a foam core. No moulds were used in the hull construction; rather the hull core was milled to shape on a ShopBot 3-axis mill from a solid block of closed cell modelling foam. Boat 1 used a 288 kg/m3 density foam while Boat 2 used 160. Bulkheads were integrally cut for Boat 2 to reduce flexing during construction and to save time from the secondarily bonded bulkheads on Boat 1. Boat 1 used ProSet 125/226 Epoxy as the adhesive system while Boat 2 used WEST 105/205 due to its greater viscosity and user-friendliness. Figure 2 shows the compartmentalization for

Boat 2. The core thickness for Boat 1 was a uniform 12.8 mm and Boat 2’s is 10 mm.

Fig. 1 USNA Boats 1(top) and 2, showing the trend toward narrower beam

Fig. 2 Integral bulkhead placement allowing for watertight compartmentalization and structural support

Chainplates were fabricated using two plies of #282 over a carbon tube, with the plies extending 40 mm each side on to the hull shell. Keelboxes were built using six plies of 150 g/m2 E-glass cloth and used the actual keel head as the plug. The rudder for Boat 1 is a NACA0012 section cored with 400 kg/m3 Renwood closed-cell foam and skinned with one ply of 150 g/m2 E-glass cloth. Rudder 2 is similar but uses a S8035 12% section[4]. Rudder shafts are 11 mm diameter silicon bronze. A plain bearing is used in the hull and the shaft is keyed for a double-sided tiller arm. To achieve a balanced rudder the shafts were located with 15% area in front of the shaft centreline. Rudder 1 is a simple trapezoid with a 0.4 taper ratio and Rudder 2 is elliptical. Both have a projected area approximately 1.2% of sail area. Experience has shown that a larger area may be beneficial in manoeuvring.

The mast step on Boat 1 was fabricated of Delrin with partial depth holes at two mast diameter spacings to allow for differing mast placements. Boat 2’s mast step was redesigned to allow for a wider range of mast step placements and features a fixed aluminium strip with an adjustable aluminium plate acting as a mast-step base plate. The long strip doubles as an additional support for the two keelboxes. Figure 3 shows the deck layout for Boat 2.

IRSC'09 - page 10

Fig. 3 Boat 2 showing mast step and rigging details.

Three keels were built for the two boats. The keel material

is 17-4 precipitation hardened stainless steel in an H1150 heat treat. Keel 1 has a 1.64 meter long rectangular (38 x 12 mm) stainless strip covered with a foam fairing to produce an airfoil section. The constant NACA0016 section shape has a 140 mm chord length and is covered with one layer of 150 g/m2 E-glass cloth. Keel 2 is a machined section of stainless with a 110 mm root chord and 38 mm tip chord. The section varies from 14% at the root to 16% at the tip. Keel 3 is a smaller version of Keel 2 with a 98 mm root chord and a 30 mm tip. The S8035 section is 13% at the root and 16% at the tip. Keels 1 and 2 have a 15 degree aft sweep while Keel 3’s is 5 degrees.

All three bulbs were built by cold casting lead shot in a two-part female mold. The matrix for Bulb 1 was Type 2 Portland cement while epoxy was used for Bulbs 2 and 3. Although the cement has a higher density, it proved too brittle and the curing time was longer than desired. Bulbs 1 and 2 used a NACA0016 section with a 3:2 squash ratio and beaver tail. Bulb 3 maintained the beaver tail and squash ratio but used a 10.5% S8035 section. Keels 1 and 2 are interchangeable. In case of problems with Keel 3, either earlier keel can be mounted in Boat 2 as it has keelboxes installed for both keel designs. Figure 4 shows the installation on Boat 2.

Four rigs were designed and built for the two boats and have seen the most post-launch development. Rig 1 was designed to the Nordic Boat Standards [5]as a strength minimum. While all the rigs have turned out to be durable, the major effort has focussed on finding a rig that works across the wind range. The Rig 1 goal was to create a lightweight rig that would automatically depower through bending. This was accomplished by using a 70% fractional rig with a long unsupported top mast. This rig worked well in winds from 4-24 knots, but the large mast bend in the upper wind ranges led to unreliable wind readings from the anemometer. The solution was to stiffen the top mast with jumpers. While this solved the anemometer issues it decreased the depowering ability.

Fig. 4 Multiple keelboxes on Boat 2 for backward compatibility with

Keels 1 and 2. Rigs 2 and 3 were designed as conventional 85% fractional

double-spreader rigs with 20 degree swept-back spreaders. The tubes are off-the-shelf braided carbon with a 16 mm diameter and a 1 mm wall thickness. To reduce deck penetrations they are deck stepped. The topmast uses a tapered section. The frontispiece shows Rig 2. The spreaders are 316-stainless tubes that slide over solid stainless rods that are glued through the mast tube. The booms are also carbon tubes and the battens are carbon strips. Rig 4 is a freestanding single sail rig (similar to a Laser) that has not yet undergone evaluation.

Sails were designed and built by the students using SMSW6[6]. Sail cloth is a lightweight scrim mylar. Approximately 85% of the sail area is in the main and each boat has a light air and heavy air main. The mains are attached to the mast with “zip ties” through grommets to allow quick changes. All rigging wire is 1.6 mm 7 x 7 316 stainless.

III. VESSEL DESCRIPTION - SYSTEMS While a fast boat is important in winning the SailBot

contest or making progress against ocean currents, reliable control systems are equally important. Boat 1 clearly demonstrated this concept as it was the fastest boat at SailBot 2008 but had unreliable controls and finished second. The current control systems are identical in the two boats. The primary controller is a Rabbit 3000 Microprocessor.

To fit within Boat 2’s more limited design space the systems constraints included: weight less than 3 kg in the hull and 1 kg at the masthead, able to fit through a 180 mm hatch opening, and minimum 24-hour endurance. Functional requirements include three modes; autonomous control of navigation and sail control, autonomous sail control with manual rudder control, and full manual control. For ease in transition between student year-groups and spares integration, a design driver was the desire to use in-house or off-the-shelf components as much as possible. Figure 5 shows the basic systems assembly.

IRSC'09 - page 11

Fig. 5 Systems Assembly for Boats 1 and 2.

The standard USNA (TSD) navigation board includes a

Rabbit 3000 Microprocessor, MicroMag 3-axis compass, Trimble IQ GPS, accelerometer, PWM outputs, Zigbee modem, 10 channels of 12-bit analog-to-digital conversion, 4 serial ports, external interrupt, general purpose I/O port, and statues LED. To provide additional watertight integrity for the main electronic components the navigation board was but into a plastic container. Holes were drilled into the side at mid level and a pipe inserted to run wires through. Within this pipe silicon rubber is applied to further reduce the flooding risk.

Accurate positioning is critical in this project as the finish line in one SailBot contest is only 3 meters wide. The standard Trimble GPS accuracy on the in-house navigation board is approximately 7 meters. To supplement the standard GPS a Magellan AC-12 DGPS was added. The AC-12 is a low cost, small, DGPS with an accuracy of 0.8 meter and a power consumption of approximately 200 mA at 3 volts. To improve reception when heeled, the AC-12 uses an on-deck Garmin 29 antenna.

Wind direction is sensed with a Davis anemometer. While heavier than desired and having a 20-degree deadband, it was off-the-shelf and is water-resistant. The rudder servo is a standard servo for remote control sailing yachts although it is upgraded with metal gears. To control the sails a single RMG 380HD Smartwinch is used with a traveller line on deck.

A Futuba transmitter and receiver is used to manually control the SailBot. In order to switch between manual and autonomous modes, a PIC microprocessor is used to read a toggle channel from the Futuba TM/RC, which triggered relays to switch the rudder and sail winch between the two modes.

This system requires constant communication with the Futuba Transmitter. It is expected in a long distance autonomous race that the SailBot will sail beyond the range of the controller. A commercial available Duratrax Failsafe Unit was purchased to address this situation. This devise senses losing the loss transmitter signal, in which case it provides a preset value. We set this value to default to the autonomous mode.

Another SailBot competition requirement is to be able to steer manually but have autonomous sail control. An override switch was added which forces the rudder to receive manual

control inputs regardless of the override status. This allows us to always drive the boat while switching in and out of automatic sail control.

Power to the three systems (navigation, winch and rudder) is independent to reduce feedback and for redundancy. The winch is powered by four rechargeable C-cells while the navigation board and rudder servo are powered by 6 volt, 1100 ma NiMH batteries. Figure 6 is the functional block diagram for the two boats.

Programming is accomplished through two methods. To avoid compromising watertight integrity a waterproof serial connector is used on deck. While this is quick and reliable it is not convenient in rough water or at a distance. In those cases the slower Zigbee modem is used to reprogram the microprocessor. A Zigbee modem is also used to serial communicate boat performance data back to the observer.

The code is in multiple parts, including taking in sensor information, navigation, rudder control and sail control. The navigation co-statement begins with assigning values to variables including magnetic wind direction, port and starboard close hauled course, velocity made good (VMG) on the calculated close hauled course angles, bearing to the waypoint, and danger bearings. Magnetic wind direction is then processed through a digital low pass filter. With these values calculated it then decides if the next waypoint is upwind (in the no go zone) or downwind. If it is downwind, it will drive straight there. If the boat has to tack to go upwind, the program then calculates which course (starboard or port tack) has the VMG towards the next waypoint. The program will then head the boat in that direction. If the boat completes a tack then the program with hold the given course for a pre determined about of time to allow it to get stabilized and up to speed. It will then start the above loop again, calculating which tack would have the better VMG and then steering accordingly.

Primarily the sails are trimmed to the current wind direction. The rudder is controlled by comparing the difference between the desired heading calculated in the navigation portion of the control and the current heading. A very simple proportional controller is used. The “gains” were calculated based on experience with sailing the two boats. We have found that this heuristic approach to rudder control to be reasonably reliable, however the sail trim significantly affects the ability to turn the boat. Most prevalently, in higher breezes it is almost impossible to turn the boat from close hauled to a beam reach with out easing the main. Similarly, the main tends to over power the rudder downwind and sometimes will prevent the boat from jibing.

IV. RESEARCH AND DEVELOPMENT – NAVAL ARCHITECTURE The primary tool used in developing the two boats from the

naval architecture perspective was a velocity prediction program (VPP) called PCSail[7]. A VPP solves for equilibrium of four of the six degrees of freedom (yaw and pitch are ignored). This Excel-based program is typical of simple VPPs in that it uses a relatively simple user interface

IRSC'09 - page 12

with inputs for the key vessel geometry and then solves through iteration using the built-in solver module. Key outputs include speed, heel, and the optimal reefing amount.

Fig. 6 Functional block diagram for Boats 1 and 2.

To determine if a particular change is worthwhile the results were displayed in delta seconds per mile for each heading or in VMG. For the SailBot competition these were then applied to each leg of a known course, or were applied as summed weighted averages for unknown conditions. An added level of analysis included comparing the resulting proposed boat against the competitors to determine potential win/loss records. Figure 7 is an example of a predicted match race in six knots showing the potential speed per leg for three boats. Boat 2 (Luce Canon) is shown to have a speed advantage on each leg.

Like most general purpose VPPs, PCSail estimates the hull resistance using parametric analysis of tank test results. The Delft series focused on full-size yachts and the parameters were for relatively beamier and much larger vessels than Boat

1 and 2. To validate the VPP, Boat 1 was tank tested and the results showed acceptable correlation.

While the VPP runs are quick, realistic trade-off studies are time intensive as a change in one hull variable necessitates

changes in many others. For instance, an increase in beam requires a reduction in canoe body draft for a constant displacement. This means that in hull studies a new hull must be generated for each data point. A complicating factor is the wide wind range which requires a design that is good across the full range.

First year studies included the key vessel parameters of length, beam, displacement and stability. Length was the easiest to solve as the VPP gave a strong trend toward maximum waterline length and as the freeboard is so small that also meant maximum deck length. The result was a plumb bowed and sterned vessel at the SailBot maximum length of 2 meters. During the second year the VPP used Boat 1 as a baseline.

IRSC'09 - page 13

Match Race ~ 6 knotsTriangle Course at 6 Knots

1.00

1.50

2.00

2.50

3.00

3.50

1 2 3 4 5

Course Leg

VMG

for

Leg

(kts

)

1st TimeLuce CanonNorth Star

1,4

2

3

5

Fig. 7 Predicted speeds for a match race in 6 knots. Beam was the second variable studied. The decision for the

first year was a trade-off in narrow beam versus volume and access. With the systems not yet defined by the time construction started, the decision was made to make the boat just slightly wider than the larger of the two available Holt Allen inspection ports. A more detailed study was done the second year and indicated that narrower beam would significantly increase performance. The beam variation study kept displacement and length constant while beam and canoe body draft varied. Beams from 390 mm down to 228 were compared and the results showed an increase in performance for the narrower boat in all wind speeds and on all courses. The trend appeared to reach a limit near 228 in reaching and running conditions however. The 280 mm final beam was selected based on mast stability considerations and the size of the smallest practical hatches.

It was relatively easy to study some characteristics, such as the prismatic, in isolation. Sail area, stability and displacement however, required a combined study. This was because it quickly became apparent that as sail area increased the boat gained performance if the stability increased, but that was only possible if displacement also increased. In essence, as the rig, hull, deck, and systems became a fixed weight, the variable weights were the keel and primarily the bulb.

It quickly became apparent that the most effective righting moment was achieved with a rule maximum draft. After that the keel weight became fixed and the variable was the bulb weight. The trade-off study thus focussed on variations in bulb weight which effectively was a trade-off in VCG versus displacement. The results showed some sensitivity to wind strength in that in light air a light bulb and light displacement were superior. Once the wind reached approximately 12 knots however the results converged to a 11.6 kg bulb weight of and 24 kg displacement. It is important to point out that VCG was critical and efforts were made to reduce weight in the hull and rig.

With the displacement and stability determined a trade-off study in sail area was performed. The results were somewhat disconcerting in that light air performance was strongly influenced by sail area. Essentially in winds less than six

knots it was critical to put as much sail on the boat as possible! The concern however was that in winds above 10 knots a large sail area would cause significant control issues. The solution was to have different main sails for light and heavy winds and have a relatively quick means for switching sails. Clearly this approach is not acceptable for long distance events. Current development focuses on sails and rigs that depower automatically but are robust.

As mentioned above, the need for stability forced the keels to the maximum 1.5 meters draft. At the same time a deep draft keel using normal proportions would create a large wetted surface area, decreasing performance. The solution was to aim for a very high aspect ratio keel. This created a design challenge in that a thin, high aspect ratio keel would bend, decreasing righting moment. Material selection for the keel focussed on a high stiffness material with good strength, durability and relatively low cost. Weight was not a significant concern due to the keel’s secondary function as ballast and the relatively small volume.

Keel 1 had a short lead time and was designed for easy fabrication. While this design was stiff, strong and easy to construct, it had more surface area than desired.

Keel 2 addressed this by cutting an 80 x 19 mm strip to an airfoil on a 5-axis mill, with a 0.4 chord/root taper ratio. The structural foil comprised the airfoil section from 15-75%. The leading edge used foam and a 4 mm rod to create the airfoil shape, while the trailing edge was constructed using filled epoxy. While this keel was both lighter and had less area, it was significantly more difficult to produce due to the added work on the leading and trailing edges.

Keel 3 was designed for Boat 2 and applied the lessons learned from the first two keels’ construction. While similar in construction to Keel 2, the structural portion extends from a constant 3mm aft of the leading edge all the way to the trailing edge. A 3mm diameter rod is glued on the leading edge to form the correct leading edge radius on the S8035 section.

To balance stiffness and strength a series of finite element analyses were performed on Keels 2 and 3. The limit state to yield was 90-degrees heel with the keel in air and a dynamic amplification factor of 2. The maximum permitted deflection at that condition was 125 mm with a 2 degree rotation. Figure 8 shows a typical plot. In practice, Keel 3’s torsional stiffness is a little less than desired.

Keels 1 and 2 were designed with 15 degrees of sweep in an effort to improve weed shedding and increase the second moment of area in yaw to aid directional stability. Keel 3 had the sweep reduced to 5 degrees to reduce induced drag and torsional stress. An unintended benefit of the relatively large keel deflections is that the boats have positive righting moment throughout the stability curve which means they will always self-right in a capsize.

IRSC'09 - page 14

Fig. 8 VonMises stress plot of Keel 3. The keel was designed so that the

highest stress was away from the connections. Other studies looked at hull shape and seakeeping. Boat 1

has relatively U-shaped sections and has a tendency to pound which shakes the rig. In response to this, Boat 2 has more V-shaped sections forward. The trade-off in this approach became apparent in early trials when the shaking was clearly less, but the boat showed a tendency to track divergently. The forward sections on Boat 2 were reshaped after the SailBot competition.

Future research areas include a more detailed look at added resistance in waves, automatically depowering rigs and low-power steering and sail trim systems.

V. RESEARCH AND DEVELOPMENT - SYSTEMS Choosing an appropriate wind sensor was difficult as no

sensor is made in a small lightweight and watertight size. The Davis anemometer was picked because of its relatively easy interfacing, minimal price and perceived accuracy. It is heavier than desired, with a two pound weight. The intent was to reduce the weight by rewiring it and rebuilding its frame. As the wind speed was not needed; additional weight was saved by taking these parts off. Wind direction is sensed with a potentiometer which returns an analog voltage back to the navigation board.

Initial testing showed the anemometer was not adequate. At heel angles greater than 30 degrees or light winds the instrument did not register properly. The Davis instrument seemed to have a high threshold velocity while our specifications required a starting threshold of less than 1 knot. These failures prompted us to make our own sensor using a potentiometer which would offer a lower threshold and create a new tail wing design.

The prototype used a Bourns 6534S-1-502 potentiometer with a carbon fibre rod for the structural support. We added a large plastic tail feather and counter-weighted it with a bolt and nut. Figure 9 shows the completed wind sensor.

Fig. 9 In-house lightweight wind direction sensor.

To test the anemometer we placed an aluminium ladder in

front of a ventilation fan. We then took a handheld wind speed sensor and mounted it to the ladder in order to measure wind speed. Comparing the Davis to our design we found they had the same 20-degree deadband, but our design had a lower threshold. The Davis wind direction sensor was water resistant however! Given the lower threshold from our larger tail fin we modified the Davis to include a larger tail fin.

While the RMG 380HD SmartWinch is strong enough to trim the sails in all normal conditions, it has no ability to prevent back turning the gears when the motor is turned off. This unfortunately means it requires significant amounts of current (1.5+amps) when holding the sail. This will quickly run the batteries down, and is unacceptable for long duration sailing. One solution we explored is a winch system that uses a worm gear to prevent the back turning. Using a Maxim motor and a worm gear we designed our own sail winch. Figure 10 shows the new winch.

The winch is controlled with the navigation board through a LM298 Dual Full Bridge Driver. A 10 turn potentiometer is attached to the output shaft in order to provide feedback on the winch position. The Maxim motor only draws 200mA, which is great for long distance racing. Unfortunately, to get the proper torque we had to run the motor at 22 volts and the response rate was too slow for round-the-buoys racing.

Fig. 10 In-house worm gear winch.

The Maxim motor was a convenient choice as it was in

stock, and was used for a proof-of-concept. Better selection would reduce the voltage needed with the same low current draw.

Future research includes power consumption and generation. Solving the winch’s power consumption problem, coupled with the low current draw of the navigation board and rudder (approximately 6 volts, 30mA each) application of

IRSC'09 - page 15

solar cell technologies can be used to allow self sustainability for a transatlantic crossing.

Geographic positioning for a transatlantic crossing is important but not as important for buoy racing. A high degree of accuracy on the race course is needed in order to round marks. DGPS is one solution, but since buoys tend to have swing circles, the exact location of the buoy may change. One solution is to use GPS to get close to the mark then use cameras and image tracking for close in navigation. We experimented with using two CMUcam2 giving us a ninety degree field of vision in front of the boat. These cameras would be set to track the highly visible orange color of the buoys. The microprocessor on these cameras would then send a serial signal to the navigation board with information including the size of the image and location of its centroid. This data would allow the boat to navigate around the mark. Limitations of this camera include glare off the water in some satiations, this could be fixed with the application of an IR filter.

VI. CONCLUSION While small, sail-powered autonomous vessels offer

promise in numerous applications, the field is still young and numerous opportunities exist for development. This paper highlighted the initial development of two “SailBots” by the students at the United States Naval Academy. This information may help others in developing their vessels. In addition to the educational objectives reached through the

boats’ design and construction, competition encouraged more rapid development of the vessels and will encourage others to participate.

ACKNOWLEDGMENT Funding for the USNA SailBot team was provided by the

Naval Sea Systems Command and Project Engineering Office – Ships, US Navy. Significant help on the project was provided by Professor Brad Bishop, Mr. Joe Bradsaw, the Technical Services Department, the Sail Loft of the United States Naval Academy and the other members of the USNA SailBot Team.

REFERENCES [1] SailBot Class Rules – as revised 2006, 28 Nov. 2008,

www.usna.edu/Users/naome/phmiller/SailBot/SAILBOT%20Class%20Rules%2021Nov08.doc

[2] P. Miller, “SailBot: The North American Autonomous Sailboat Competition,” The Journal of Ocean Technology, Volume IV, Jan. 2009, pp. 1-5.

[3] H. Vihlen, J. Kimberlin, The Stormy Voyage of Father's Day, Marlor Press, ISBN-10 0943400910, Jan. 1997.

[4] http://www.ae.uiuc.edu/m-selig/ads/coord_database.html. [5] L. Larsson, R. Eliasson, Principles of Yacht Design, Adlard Coles

Nautical: London, 1994. [6] SMSW6 User’s Manual, Autometrix, www.autometrix.com, 2009 [7] D. Martin, R. Beck, “PCSail, A Velocity Prediction Program for a

Home Computer,” in the Proceedings of the 15th Chesapeake Sailing Yacht Symposium, Annapolis, MD, USA, January 2001

IRSC'09 - page 16

Design and Construction ofthe Autonomous Sailing Vessel AVALON

Lian Giger∗, Stefan Wismer, Sebastian Boehl, Gion-Andri BusserHendrik Erckens, Jurg Weber, Patrick Moser, Patrick Schwizer

Dr. Cedric Pradalier, Prof Dr. Roland Y. SiegwartAutonomous Systems Lab, Swiss Federal Institute of Technology (ETH) Zurich

Leonhardstrasse 25, 8092 Zurich, Switzerland∗ To whom correspondence should be adressed.

E-mail: ligiger@student.ethz.ch

Abstract—This paper presents relevant mechanical and tech-nical aspects about the autonomous sailing vessel AVALON.AVALON’s purpose is to cross the Atlantic Ocean completely au-tonomously without any direct influence by humans. The system’srobustness and stability are as important as an acceptable sailingperformance. The entire mechanical and electronic system wasdeveloped in order to fulfill that single purpose and take up thechallenge.

I. INTRODUCTION

AVALON is an autonomous sailing vessel designed forthe MICROTRANSAT CHALLENGE, a competition in whichseveral international sailing robots are trying to cross theAtlantic Ocean autonomously. During its cruise from southernIreland to the Caribbean, AVALON has to withstand severalunfavorable weather conditions . Winds within a range of 10up to 50 knots and steep waves with a height of about 9meters may occur especially around the Irish coast. In spiteof powerful batteries the energy of the boat is limited dueto the requested autonomy of about 4500 hours. As a result,efficient and energy-saving mechanic and electronic systemsare required.

A. State of the Art

1) Harbor Wing Technologies: The company HARBORWING TECHNOLOGIES developed an autonomous, unmannedsurface vehicle that was originally intended to be used withthe US Navy. The vehicle is characterized by the followinginnovative components:

• A 60 foot Wingsail capable of turning 360 degrees• A horizontal winglet on the Sail that controls the driving

force produced by the wing• Hydrofoils on the rudder and fin that increase the effi-

ciency and speed of the boat2) Microtransat Competitors:ROBOAT was developed by the Austrian Society of Inno-

vative Computersciences (InnoC) and is based ona sailing dinghy LAERLING that was technicallymodified and supplemented by a self-developedrobotic unit.

FAST is like AVALON, a self-construction made of glassfiber. Contrary to AVALON, the Portuguese hull

Fig. 1. Prototype of Harbor Wing Technologies [1]

was built on a positive mould made of wood. Therobotic unit is completely self-developed.

Details about several other competitors are listed on theMICROTRANSAT 09-Homepage[4].

II. PROJECT ORGANISATION

A. Modules

The Team Students Sail Autonomously (SSA) consists ofeight mechanical engineering students of the Federal Instituteof Technology, Zurich. The Project is structured into threemodules: One contains all the management and organizationprocesses and the other two the boat structure respectively theentire control system.

B. Timetable

The Project AVALON started in August 2008 and willformally end in September 2009. The complete boat and itscontrol system has to be developed, built and tested withinthis period. The timetable in figure 2 illustrates the differentphases of the project, starting with the design and constructionphases to the point of testing and optimization.

IRSC'09 - page 17

Concept

Design

Construction

Testing

Sept Jan May

Fig. 2. Official timetable

C. Finances

The costs of the project currently amount to 209 000.-sFr. Split into Control System, Mechanical Structure, EnergySupply and Logistics (see fig. 3), it is obvious that the partfor the mechanical structure constitutes the largest amount tothe budget, mainly because we use high-quality materials likecarbon fiber and CNC-milling of all the molds. The entire

Fig. 3. Costs of different modules

project is financed by sponsors from the national industryacquired by students of the team.

III. MECHANICAL STRUCTURE

A. Hull

The hull design is based on form parameters, e.g. length,draft or beam. Parameters are used as input data for B-splinecurves and surfaces. Instead of modifying those curves andsurfaces, we use constrained optimization algorithms in orderto vary parameters and generate geometry. The algorithmminimizes a target function, which is defined so that a certaincombination of parameters results in a desired hull shape.The hull design of AVALON is based mainlyon four fixed parameters and seven variableparameters. As a result, four B-spline curves can bederived and the hull surface generated (Figure 4).The basic data of the hull AVALON can be given inTable III-A.

The final hull was laminated inside a female mold usingglass fiber sandwich. Epoxide resin was sucked into dryglass fiber material by vacuum using a so called INFUSION

TABLE INOMENCLATURE

Variable Description Value

General VariablesBmax Half maximal beam 0.7 mT Draft without keel 0.25 m∇ Displacement 0.44 m3

Fixed Design ParametersLOA Length over all 3.95 mAngle Central angle (DeckContour) 65°Radius Radius of DeckContour at bow 0.05 mDeckHight Z-Position off Deck from water-

line0.4 m

Variable Design ParametersBeamAft Half beam at aftBeamMaxX X-Position of maximal beamBeamMaxY Value of maximal half beamDraftBow Value of draft at bowDraftMax Maximal value of draftDraftMaxX X-Position of maximal DraftTransomEndZ Z-Position of Transom at stern

Fig. 4. Hull

PROCESS. Compared to conventional laminating, this methodis much cleaner and more convenient. It is the state of the artof laminating processes.

B. Rig

The rigging system is one of the most important parts of thewhole assembly. A defect in its structure will inevitably causea shipwreck of the project. Following aspects were consideredfor the design:

• High loads and forces on the mechanical structures dueto strong winds and heavy weather conditions.

• Highest demands on reliability. There is no chance torepair anything througout the journey.

• Preferably efficient mechanical transmission in order tosave as much energy as possible

During the design phase it turned out that a balancedrig (figure 5), comparable to RC-models, prevails over othersolutions like a conventional rig. The criteria for the decisionfor a balanced rig were:

• Costs• Mechanical energy efficiency

IRSC'09 - page 18

• Sailing performance• Reliability

Fig. 5. Design studies of the rig

The balanced rig is much more efficient relating to energyconsumption than any other conventional rig. The load ontrimming ropes is reduced by over 50% due to the balanceddistribution of the sail load. Furthermore, AVALON’s rigdoesn’t use any ropes that may generate knots and jams. Therig is pivoted on a central bearing without shrouds and stays.The result is a technically simple and reliable construction.

C. Keel and Rudders

1) Keel: In order to achieve a sufficient righting momentand stability during heavy weather situations, AVALON’s keelwith a draft of two meters consists of a slight fin with a 160 kgballast bulb. This configuration allows the boat to sail in windsof up to 50 knots without capsizing.

The fin and parts of the bulb were made by high-moduluscarbon fiber in precisely milled polyurethane molds. Afterhardening and tempering, the bulb was filled with lead andthe two halves were glued together.

Fig. 6. Keel with fin and ballast bulb

2) Rudder: A twin-rudder system was selected for AVALONto make sure to have sufficient steering effect in every sailingsituation. Angular mounted twin-rudders deliver a better gripin the water at a certain heeling compared to single rudders.

Assembled inside the hull, the rudder actuators are wellsealed and protected against water and humidity.

Fig. 7. Twin rudder system

IV. ELECTRONIC HARDWARE

A. Main Computer

The brain of the system consists of different hardware com-ponents located in the center of the hull. The main computeras well as different peripheral devices such as a serial-to-USBadapter, satellite modem and a wireless LAN hub are in a dou-ble sealed fiber glass box. The main computer MPC21 from

Fig. 8. Main PC and brain of AVALON [2]

DIGIALLOGIC is a 500 MHz device with 1024 MB RAM and acompact flash harddrive. With an average power consumptionof about 8 watts, the protection of a metal housing and a totalweight of less than 1 kg, this device is designed for operatingin a rough environment.

B. Sensors

All desired information such as position, heading or speedare measured by several sensors located all over the boat.To get a fully controllable system, data is collected from thefollowing sensors:

GPS Provides position, heading and speed. The sensor’santenna is mounted on top of the mast. The elec-tronic unit is located in the hull.

IMU The Inertial Measurement Unit provides all infor-mation about velocity and acceleration in all 6degrees of freedom. Combined with an additionalGPS-antenna, the accuracy of the boat’s positioncan be increased to ±10 cm.

Wind Mounted on top of the mast, the windsensor pro-vides wind speed and direction. Unlike other sailingboats, AVALON has an ultrasonic windsensor thatpromises less failure than a conventional sensorwith a turning wheel.

AIS This Sensor receives data such as position andheading from other boats via VHF-antenna. TheAIS system is an additional means of perceptionthat ensures that collisions with large commercialships are avoided.

IRSC'09 - page 19

C. Satellite Connection

In order to guarantee a safe connection to the boat during thetransatlantic journey, a very reliable communication system isinevitable for such a project. On AVALON, the communicationsystem will not only be used to provide periodic positionand status reports, but also to have the possibility to interactwith the boat in case of an emergency and to downloadweather updates. All these tasks can only be accomplishedby using a satellite network. With our main goal to cross theAtlantic Ocean, the respective satellite network has to providecoverage in this specific region. This is only the case with thenetworks IRIDIUM and INMARSAT. The choice between thesetwo options has been based on the available hardware for eachof these networks. For the INMARSAT network, the favoriteoption is the Mini-C-System. It has an integrated GPS-receiverwith position reporting capability. The IRIDIUM device whichwas finally chosen, the 9522-B modem, has a more flexibleinterface, higher data rates and lower airtime fees. Weatherupdates can be downloaded by the boat autonomously. AsIridium is also acting as an Internet Service Provider (ISP),there is the possibility to establish a Secure Shell (SSH)connection to the boat in case of a very severe malfunction ofthe software.

D. Power Supply

The main power supply is realized through four solarpanels of 90 Wp each and a total area of two square meters.The collected power gets stored in four lithium-manganesebatteries. Each battery consists of 70 single cells and has acapacity of 600 Wh at a nominal voltage of 25.2 volts. Lithium-manganese were chosen mainly because of their weight butalso because they are fairly safe to use.

For back-up power, the boat has a direct-methanol fuel cellon board. This fuel cell is automatically activated when thevoltage drops under a certain value, the switch on voltage.It then charges the batteries until the switch off voltage isreached. In theory, the solar cells provide enough power for theboat’s systems. The fuel cell only serves in case of enduringbad weather or other unforeseeable circumstances.

E. Actuators

1) Sail Motor: The Sail is driven by a 200 watt motorby MAXONMOTORS. Motor and gear are encapsulated witha glass fibre box in order to be protected against humidity.The actuation moment is increased by a factor of three andtransmitted over angular wheels to the shaft of the rig bearing.With a nominal torque of 600 Nm, the sail actuator is capableof handling most of the heavy weather situations. If the torqueexceeds the nominal value, a friction clutch protects the gearsand the motor from demolition.

2) Rudder Motors: Each rudder is separately driven by a150 watt motor by MAXONMOTORS. Protected by a fiber glassbox, the moment is aswell transmitted over angular wheelsto the rudder shaft. In case of maximal load, the rudder canactuate with a torque of 30 Nm which is far beyond common

load cases on rudders of comparable size. Therefore it wasneedless to implement any kind of clutch on the rudders.

Fig. 9. Rudder motor with gear

V. SOFTWARE

A. Software Organisation

All software is LINUX-based and mostly self developed. Thebase of the entire software structure is the so called middlewareDDX. This is a software that manages any storing activitiesand represents the connection between a shared memory andseveral independent subprograms (see figure 10). Thanks toDDX, the control system is structured in closed program partswhich are connected to the store by reading and writing globalvariables. For instance there are sensor drivers which read outactual sensor data from the hardware and write it directly tothe store. Another program requiring sensor data will readneeded data from the store. A distinct advantage of DDX is themodular software structure. If ever a program part crashes, itcan be restarted separately by certain watchdogs. The differentsubprograms are:

Fig. 10. Software organisation

Drivers Read out any sensor information, trans-mit them to the store and actuate rudderand sail.

Navigator Calculates the optimal route based onsensor data and transmits a desired head-ing to the store.

IRSC'09 - page 20

Helmsman Controls and steers the boat. Requiresinformation from store respectively sen-sors.

Communication Establishes and ensures the satellite con-nection between boat and headquarters.Required to transmit status messages andlogfiles.

Structure and function principles of Helmsman and Nav-igator are specified in the following sections. Basically thesoftware structure is comparable to a common hierarchy ona sailing yacht. On one hand there is the Navigator decidingabout the strategy and the heading. On the other hand, theHelmsman is responsible for the rudder and the trim of thesails. The only difference to a real sailing team is the fact thatall positions are occupied by computers, sensors and motors.

B. Navigator

This program part collects information about weather andgeography and calculates the optimal route to a given waypointbased on a grid-based A* path planning algorithm. Theprogram is split into a global and a local planner. The globalplanner respects mainly weather forecasts within the entireroute on the Atlantic, whereas the local planner sets localwaypoints and headings and is capable of recognizing andavoiding other large boats in the area. Weather informationis transmitted via a colored wind chart over the satellitecommunication system and will be parsed and convertedto useful wind information. The calculated heading will betransmitted to the Helmsman.

Fig. 11. Route planning at different wind directions

C. Helmsman

The Helmsman reads out the actual heading from the storeand keeps the desired heading of the boat. The sail’s positionis directly related to the wind angle. The Helmsman alsoperforms maneuvers such as gybing and tacking or luffingand bearing out. The program is built as a state machine withdifferent modes:

• Upwind Mode, during tacking against the wind• Downwind Mode, during gybing downwinds

• Normal Sailing, if a constant heading is required• Emergency Mode, rudders crossed and sail straight in

wind

Fig. 12. Sailing modes on different headings

Depending on the required heading, the Helmsman switchesto the corresponding mode. Control parameters have beenoptimized for every different mode in order to achieve thebest sailing performance.

-60

-40

-20

0

20

2600 2650 2700 2750 2800 2850

[deg

rees

]

time [s]

current headingdesired heading

wind direction

Fig. 13. Wind direction and current heading plotted over initially desiredheading

VI. CONCLUSION

The result of all previous information is an energy-efficientand solid autonomous sailing vessel named AVALON. Con-struction was completed on time. Several mechanical andsoftware tests have already taken place on the lakes of Zurichand Lucerne. The boat showed stable sailing behavior indifferent wind conditions. The robustness of the mechanicalstructure has been proofed up to 6 beaufort and the controlsystems already allow to navigate to a given waypoint. The

IRSC'09 - page 21

team is convinced to be ready and fully equipped to start thelong race over the Atlantic.

Fig. 14. Avalon ready to sail

ACKNOWLEDGMENT

The author would like to thank the ETH Federal Instituteof Technology Zurich and our supporters of the AutonomousSystems Lab (ASL), especially Prof. Dr. Roland Y. Siegwartand Dr. Cedric Pradalier for their competent support.

REFERENCES

[1] 2009/04 - http://www.harborwingtech.com[2] 2009/04 - http://www.digitallogic.com[3] Team SSA, ETH Zurich, Development of the Autonomous Sailing Vessel

AVALON (internal report), Zurich Switzerland: 2009.[4] Official Microtransat Challenge, www.microtransat.org[5] HARRIES, S., ABT, C.: Parametric Design and Optimization of Sailing

Yachts., In 14th Chesapeake Sailing Yacht Symposium, Annapolis, Mary-land, USA, 1999.

IRSC'09 - page 22

Technologies for Autonomous Sailing: Wings andWind Sensors

Mark Neal, Colin Sauze and Barry ThomasDepartment of Computer Science, Aberystwyth University,

Aberystwyth, Ceredigion, Wales, United Kingdom, SY23 3DBEmail: {mjn,cjs06,bmt7}@aber.ac.uk

Jose C. AlvesUniversity of Porto, Faculty of Engineering

Porto, PortugalEmail: jca@fe.up.pt

Abstract—The current generation of sailing robots requirea small number of essential components in order to functionsuccessfully. These include some kind of sail and a device for de-tecting the direction of the wind in order to ensure that the angleof attack of the sail is suitable for the course to be sailed. Thesetwo devices present some of the most difficult engineering andcontrol system challenges in building sailing robots. This papersummarises a number of experimental designs and approachesto the construction of these components. In particular a numberof wingsail construction and control techniques are presentedas well as designs for mechanical and ultrasonic wind directionsensors. All of the devices presented have been built and testedby the authors. Commentary on the performance and interactionof the devices is also presented.

I. INTRODUCTION

Sailing robots require some kind of sail to propel themthrough the water, to date two key designs have emergedwing sails and traditional fabric sails controlled through sheets(ropes). Wing sails offer far fewer points of failure but sufferfrom poor downwind performance and currently lack a reliablemethod of reefing. In order to set the sail position correctly theboat’s control system must know the current direction of thewind and therefore some kind of wind direction (and possiblywind speed) sensor is required. There are two main approachesto sensing wind direction, mechanical sensors using a windvane and ultrasonic sensors which sense the movement of airbetween an ultrasonic transmitter and receiver.

II. SAILS AND WINGSAILS

Sailing vessels have evolved over many thousands of yearsthrough a huge range of shapes, sizes and technologies. Allof these vessels until the last few years have been sailed byhumans with varying amounts of mechanical assistance rang-ing from simple rope purchases, through manually operatedand steam-powered capstans to modern electric and hydraulicwinches on large modern yachts. The role of conservatism andtradition in this evolution should not be underestimated, andis often reinforced in the current era by the nature of racingclasses and regulations. Despite this inherent conservatism awide range of innovative designs have been experimented withover the years and some of these designs have shown greatpromise. Modern junk rigs, wing sails and kites are goodexamples of these technologies and clearly demonstrate that

there is nothing particularly special about the conventionalflexible fabric sail.

Flexible fabric sails have a number of useful properties onmanned vessels under conventional conditions:

• They can be conveniently lowered and stowed when inharbour.

• They can be reduced in area relatively easily by eitherconventional “reefing” or by exchanging sails.

• They can be relatively easily repaired and modified.• Their shape and camber can be altered by tensioning and

releasing control lines.They also have a number of problems:

• They are prone to wearing and tearing when incorrectlyset.

• They lose their shape when not kept with a sufficientangle of attack leading to “luffing” which reduces sailingefficiency when close-hauled and eventually leads to“flogging” and potentially catastrophic failure.

• They require rigid structural spars and (often) wire rig-ging to maintain their shape: these introduce aerodynamicdrag weight high above the waterline.

• They tend to twist which leads to different angles ofattack at different points on the sail, this reduces sailingefficiency.

From the perspective of designing a sailing robot there aresome very good reasons for considering the use of alterna-tive sail types and in particular we have experimented withwingsails for various reasons:

• They can easily be designed such that they do not sufferfrom problems with chafing.

• They will not “flog” even when the control system failsto maintain the correct angle of attack.

• They maintain efficiency even when sailing very close tothe wind.

• They do not necessarily require any additional structuralelements to support them.

There are however significant disadvantages which shouldnot be ignored. These include the fact that it is extremelydifficult to design a wingsail which can be reefed reliably andthat it is relatively difficult to construct strong, lightweightrotatable wingsails at reasonable cost. We however maintainthat the potential gains in reliability and efficiency outweigh

IRSC'09 - page 23

these problems and we have successfully constructed andtested a number of sailing robots equipped with wingsails ofvarious designs. We have also experimented with a number ofactuator technologies appropriate for their control. We havefocused on designing for longevity, low power consumptionand simplicity and constructing for reliability and robustness.We do not aim for or claim that our systems are in any senseoptimal in terms of sailing efficiency, but the later systemsare sufficiently efficient, robust, controllable and low in powerconsumption to allow long term autonomy to be possible.

A. AROO’s wingsail

AROO[1][2] (Autonomous Robot for Ocean Observation)was the first sailing robot constructed by Aberystwyth Univer-sity during the autumn of 2004. It was decided that a wing sailwas best suited due to their robustness. The wing constructedfrom a folded sheet of scrap aluminium (which was originallypart of a London Bus!) and was 125 cm tall and 18 cm long(225 cm2 area). It was controlled by a DC electric motor withposition detected by a potentiometer as shown in figure 2.A mechanical wind vane and potentiometer were placed ontop of the sail to sense wind direction, as shown in figure 1.One unfortunate problem with this design was that the sailcould be continually rotated and the cable linking the windsensor to the rest of the boat could easily become tangledaround the mast. Despite these problems the wing performedexceptionally well in winds up to 30 knots. The sail was ifanything too large for the boat (a 1.5 m long racing monohull)and caused some stability problems and difficulties for thesteering system. Given the inability to reef a wing sail thisdesign would not have been appropriate for a sea going boatwhich is likely to encounter winds far in excess of 35 knots.

Fig. 1. AROO’s Aluminium Wing sail with a rotary wind sensor on top.

Fig. 2. AROO’s sail drive mechanism.

B. ARC’s schooner wingsails

In designing the second boat at Aberystwyth, ARC [3][2](Autonomous Robotic sailing Craft) we opted for dual wingsails in a schooner configuration.This was intended to counterthe instability which had been observed with AROOs sail.These were constructed of lightweight acrylic wrapped aroundseveral wooden blocks to retain shape, making them signifi-cantly lighter and easier to handle than AROOs sail. Theywere relatively easy to construct, needing only to cut thewood blocks, fold the acrylic and then place securing boltsalong the narrow edge of the sail to hold the two sides ofthe fold together. Each wing is 107 cm tall and 20 cm long (214 cm2 area), a photo of these can be seen in figure 3. Thisdesign created a very balanced sailing configuration and gavethe potential to use the sails to trim steering or to replace thesteering should the rudder fail. We conducted several tests ofthis boat without any control system running and found thatit was able to hold a course providing the sails had been setcorrectly. It was able to “goose swing” when sailing downwind where the sails are set to opposite tacks. This greatlyenhanced downwind stability compared to AROO’s single sailconfiguration. We also tested “heaving too” (where the sailsand rudder are configured to counteract each other and keepthe boat in one place) as a method of station holding butthe boat was dragged sideways by wind and currents, in partdue to its small shallow keel. The inherent stability of thisconfiguration offers great hope for one of the key requirementsof a sailing robot, a boat which requires virtually no actuatoruse to maintain itself on a present course, thus keeping powerconsumption to an absolute minimum. As with AROO wefound that the sails were actually too big and although theysailed fine in 30 knots of wind, anything more and the boatwould have healed excessively. To remove the problem ofcables running through the mast, the wind sensor was movedfrom the sail to its own mast near the stern where it was

IRSC'09 - page 24

less likely to experience any turbulence caused by the sails.The wing sails were controlled by two stepper motors takenfrom an old printer, these worked acceptably well in lightwinds and laboratory tests but in stronger winds the gearsdriving the sails would slip and the sails would drift from theiroriginal position. Our original control algorithm kept trackof sail position by keeping a record of the distance movedsince the sail was last calibrated, however when the sail beganslipping this strategy failed. We later added a potentiometer tokeep track of sail position to counter this problem.

Fig. 3. Arc’s Dual Wing Sails.

C. BeagleB’s wingsail

BeagleB [2] was developed commercially by Robosoft 1 forAberystwyth University and took on much of the knowledgegained in the previous two boats. BeagleB is 3.5 m long andits hull is based on sailing dinghy intended for disabled sailors,this design is particularly stable and designed to self right veryquickly should it capsize. BeagleB is propelled by a 2.55 m2

(3 m x 85 cm) carbon fibre wing sail, this is only 60 percentof the sail usually used on this hull, a photograph of the sailcan seen in figure 5. However this is probably still too largefor sailing under extreme conditions. Construction of the wingtook significant effort and several weeks, the sail representedover a quarter of the cost of entire boat. Although experiencewith ARC had shown that a dual sail configuration was highly

1www.robosoft.fr

stable the hull design was not suitable for two wings and thedesign actually proved to be sufficiently stable. The sail islimited to only 130 degrees of rotation by stays on either side,to the best of our knowledge it is the only example of a sailingrobot with a stayed wing sail. Wind sensing is provided by acommercial ultrasonic wind sensor (a Furuno Rowind) on topof the sail. This is mounted on an aluminium tube which runsdown the centre of the sail and does not rotate, this removesthe need to take sail position into account when determiningthe wind direction. As shown in figure 4 the sail is moved byan LA12 linear actuator mounted on the deck below the sail,the end of the actuator arm consists of a toothed plastic rack.The base of the sail contains a circular pinion which is drivenby the actuator. Beagle’s wing sail has proven to be highlystable, capable of sailing in winds as light as 1 knot howeverit has only been tested in winds of approximately 20 knots,mainly due to the danger to humans in deploying such a largeboat under strong winds.

Fig. 4. Beagle B’s Rack and Pinion connector for the wing sail.

D. MOOP’s wingsail

The latest boats produced by Aberystwyth University areknown as the MOOPs (Miniature Ocean Observation Plat-form). They are an attempt to build a set of small, cheap,simple, mass produceable and lightweight but highly robustrobots capable of crossing the Atlantic but also intended forshorter term missions to research control system strategies, aphotograph of the first prototype can see in figure 6. Theirhulls are only 72 cm long and the total weight is only around4kg. Such a small hull has been selected to reduce cost andthe difficulties in handling the boat, especially when launchingand recovering. We had found that with Beagle-B at least twopeople were required to rig and launch the boat and that inbusy waters a sufficiently fast chase boat was always required.With the MOOPs we wanted to develop a boat which couldeasily be handled by one person and that could be transportedin a normal car or checked in as baggage on a flight. Thesmall size also reduces the probability of causing damage toanother boat in the event of a collision. The low cost andrelatively simple construction process now allow us to producea new boat in under three weeks and we hope to deploy asmall fleet of them during summer 2009. Each boat features a

IRSC'09 - page 25

Fig. 5. Beagle B and its wing sail.

single wing sail 52.5 cm x 13 cm (68 cm2) constructed froma polystyrene and glass fibre composite, these are intended tobe small enough to remain sailing in strong winds. A carbonfibre rod run’s through the centre of the sail to reinforce it. Thesails also float (adding extra stability in the event of capsize)and are built to keep all the internal wind sensor electronicsdry. In addition to being small, light and floating these sailswere incredibly cheap compared with Beagle-B’s carbon fibresail, however it is unlikely that we could scale this design upto a 2.5 m2. The sails were relatively simple to construct, thepolystyrene is cut into wing sail shape from a larger block ofpolystyrene using an electrical cutter (which is simply a wirewhich heats up from the electricity running through it) andusing a cardboard outline as a guide. The centre of this mustthen be hollowed out and the carbon fibre rod inserted, thewind sensor must be mounted on the top. The wing is thenwrapped in fibre glass cloth which is attached with epoxy. Theentire process takes at least one day to complete (not includingconstruction of the wind sensor).

Although previous experience with ARC demonstrated thatdual sail configurations are preferable, the small size of thehull makes it difficult to place two sails. We have consideredattempting to place two sails upon the hull and possibly a slightincrease in hull length in future boats. Variants of the winghave been developed with both rotary and ultrasonic windsensors. The sail is positioned by a heavy duty servo and canrotate a maximum of 210 degrees. Although over the long terma servo is likely to wear out and if used incorrectly can easilyburn out they are exceptionally simple to program, cheap, fastand repeatable within a few degrees. So far (during relatively

short and “gentle” tests) they have performed exceptionallywell. From casual observations this wing sail appears to beable to sail at least 45 degrees to the wind and is very stableclose hauled or reaching. However their stability down wind,especially under gusty conditions is poor and frequent jibes areexperienced. This is suffered by all wing sails (and arguablymany other sail designs) and is not a problem unique to theMOOPs but because of their small size only minimal force isrequired to induce a jibe.

Fig. 6. The first MOOP sailing off Aberystwyth.

E. Conclusions on Wing Sail Design

Although each boat presented here differs in size, shape andits intended design goal we have observed a number of usefulattributes for a wing sail. The sails should be waterproof andbuoyant so that they can survive being submerged and so thatthey will aid in righting the boat in the event of capsize. Theyshould also be lightweight to simplify storage, transportationand rigging. If possible cabling and electronics should not berun through the sail, if they are then either the rotation of thesail must be limited or they should run through a tube whichdoes not rotate. Finally the size of the wing sail needs to bekept small, although when designing boats for racing there isthe temptation to increase sail size to increase speed this isoften counterproductive when sailing in winds over 30 knotswhen such boats find themselves leaning beyond 45 degreesmost of the time. Given that there is currently an absence ofsuitably reliable and simple reefing mechanisms for wing sails,any boat which wishes to continue sailing in strong winds mustbe equipped with a very small sail.

III. WIND SENSING

Wind direction sensing is a key requirement for a sailingrobot in order to allow it to set its sail position and course

IRSC'09 - page 26

correctly. Wind speed information is less important but maystill be useful to know whether it is futile to attempt to sailthe boat either due to there being too much or too little wind.

In order to deploy sailing robots unassisted for long periodsof time the ideal wind sensor must be robust enough towithstand strong winds, salt corrosion and the buildup ofsalt deposits and waterproof enough to sustain occasionalsubmergence and prolonged periods of rain and spray.

There are essentially three classes of wind sensor, pure me-chanical sensors which use a potentiometer to measure winddirection, contactless mechanical sensors which use magnetsand hall effect sensors to sense direction inside a waterproofenclosure or ultrasonic sensors which detect the movement ofair inside them. Mechanical sensors can suffer from wear andtear and can be difficult to waterproof. Ultrasonic sensors offerthe obvious benefit of being totally free of moving parts but aretypically more expensive and can experience problems whenwater droplets collect on the sensor. Contactless mechanicalsystems in many ways offer the best of both worlds as they arerelatively simple to manufacturer and operate but far easier towaterproof than traditional mechanical systems, however theyare likely to suffer from some level of mechanical wear overprolonged periods.

A. Mechanical

The traditional approach to wind sensing has been to simplyattach a wind vane to a continuous rotation potentiometer.Such an approach is taken by many off the shelf commercialproducts and offers a cheap and simple method for sensingwind direction. Using a simple analogue potentiometer allowsfor a typical resolution of 8 or 10 bits depending on theanalogue to digital converter being used. Typical accuracy iswithin a few degrees, dependant upon the exact design of thevane. As the only component is essentially a resistor of afew kohms, power consumption is very low. However theysuffer from a number of major drawbacks:

• They can suffer from mechanical wear and tear reducingtheir effectiveness over time.

• They are difficult to waterproof as the shaft of thepotentiometer must somehow connect with the vane.

• If the boat lists to one side and the vane may be affectedby gravity.

• They perform poorly in light winds when the wind is tooweak to keep them in position.

• Most potentiometers have small dead band where theycannot accurately measure the position.

Simple mechanical sensors were used on the first two boatsbuilt at Aberystwyth - AROO [1] and ARC [3] and havesince been employed on some variants of the MOOP boats, aphotograph of one of these sensors on a MOOP wing sail canbe seen in figure 7.

B. Contactless Mechanical

A contactless mechanical system has been developed bythe University of Porto for the FASt project [4]. This solutionis based upon contactless magnetic rotary encoders using the

Fig. 7. The Mechanical Wind Sensor used on AROO, ARC and MOOP.

family of integrated circuits available from AustriaMicroSys-tems (AS50xx)2. These small devices integrate in a single chipa set of Hall effect sensors with analogue and digital inter-facing circuitry and provide an absolute angle measurementby computing the absolute orientation of the magnetic fieldcreated by an appropriate magnet placed at a close distanceof the device case. Main features of these devices includeresolution from 8 to 12-bit, a maximum integral non linearityof ±1.4, digital output through a serial interface, sampling rateabove 2.5 kHz and power consumption below 20 mA (somechips even provide low power modes with consumption below2 mA).

Reading the absolute position of a mechanical wind vanecan be done by simply attaching a small magnet (6 mmdiameter, 0.55 grams) in the axis of the wind vane, andplace one of these rotary encoders close to the magnet andconveniently aligned with the rotational axis. To protect thedevice from moisture and water the whole electronics canbe embedded in some isolating material like Epoxy or liquidrubber. Figure 8 shows a possible arrangement for this device.

Fig. 8. A schematic of a mechanical wind vane read by a contactless magneticrotary encoder.

The arrangement shown in figure 8 was assembled in a

2http://www.austriamicrosystems.com

IRSC'09 - page 27

Fig. 9. The magnetic sensor embedded in Epoxy resin (a), the magnetattached to the vane axis (b), the assembly of the two parts (c) and the finalwind sensor (d).

prototype wind vane for a robotic sailboat. The sensor usesthe integrated circuit AS5040 (10 bit resolution), providingan accuracy of ±1 degree. The chip was mounted in a smallprinted circuit board and the electronic board was mouldedwith Epoxy resin (figure 9-a). The small magnet was attachedto the shaft with an aluminium case and a small ball bearingholds the rotating part of the assembly (figure 9-b). Bothparts are fixed together with screws that allow the mechanicalalignment of the sensor with respect to the magnet axis(figure 9-b). The whole set was then housed in a carbonfibre tube, that also supports the anemometer at the other end(figure 9-d). The anemometer is a conventional 3 cup rotorwith another magnet that activates a Hall effect switch in eachrevolution.

Fig. 10. The wind direction average calculator.

1) Reading the wind sensor: The interface with the AS5040is done with a synchronous serial interface that reads the 10-bitdigital word representing the absolute position of the magnet,

plus a few additional status bits used to validate the data read.This interface is implemented as a custom digital controlleron a FPGA device that hosts the whole digital system ofthe autonomous sailboat, sampling the sensor at a 50 Hzrate. To filter the readings, a low-pass mean filter was alsoimplemented in digital logic, using a 64-tab sliding window.Because of the discontinuity from 359 to 0, the computationof the angle average cannot be done as a simple arithmeticmean. This module implements an averaging process that, foreach new sample, computes the arithmetic mean of all thedeviations from the current averaged wind direction and thenadds it to the previous average. The output averaged valueis made available to the central processor as a 9 bit two’scomplement integer in the range [−180, +180]. Figure 10shows a block diagram of the module that performs thisoperation.

C. Ultrasonic Sensors

Ultrasonic sensors offer the promise of a sensor which isfree of moving parts allowing them to operate over prolongedperiods without suffering from mechanical wear. The theoryof operation is relatively simple, a transmitter transmits a burstof ultrasonic sound and this is picked up by a receiver. Thestrength and direction of the wind will affect the amount oftime the signal takes to reach the receiver. A measurement ofthis is taken by measuring the time of flight or phase differencebetween the transmitter and receiver. By using two receivers(although depending on configuration only one transmittermaybe required) placed 90 degrees apart two axis informationcan be derived. By taking the arc tangent of the two phasedifferences from each receiver the angle of the wind can bedetermined. Wind speed can be determined by taking the sumof the axes.

Although several off the shelf sensors are available they costseveral hundred pounds/dollars/euros and are not particularlysmall as they were designed for yachts not sub 4 metre sailingrobots! We chose to try and build a simple and low cost sensorfor the MOOPs using a water proof ultrasonic transmitter andtwo receivers as well as two NAND gates, comparators and ca-pacitors. Instead of measuring the time of flight directly (whichis highly processor intensive) we measure the phase differencebetween the signal we transmit and the one we receive. Therate of movement of air through the sensor is measured by tim-ing the flight of a 40kHz ultrasound signal reflected off the topplate from the transmitter to the receiver. The received signalis amplified by a comparator and NAND’ed with the original40khz signal (which was generated by the PWM channel ona PIC microcontroller) and used to charge a capacitor. Thevoltage at which the capacitor stabilises depends on the degreeof overlap between transmitted and received pulses. Changesin air speed through the sensor cause changes in the degree ofoverlap. A block diagram of this process is shown in figure 12and a photograph of the transmitter and receiver arrangementis shown in figure 11. Whilst this scheme is very simple, theresults vary depending (as are all devices based on the speedof sound in air) on the density of the air in the sensor. The

IRSC'09 - page 28

chief cause for variation in density is variation in the ambientair temperature which requires compensation. At the time ofwriting our prototypes do not have this compensation, thusthey require recalibration before each use to compensate forchanges in the ambient temperature. Calibration is performedautomatically on start up, and is thus not too onerous, butwork is on-going to incorporate temperature compensationinto the sensor and processing software. We have been able toreduce the level of noise in the sensor by moving some of theelectronics from inside the hull to the top of the mast. Thisreduces the level of attenuation in the signal by cutting thelength of the cable between the receiver and the comparator.Development of the electronics and software for the sensor willrequire significant further effort to achieve a standard suitablefor long term missions, but the advantages offered seem to beworthwhile: no moving parts, a total weight of around 50g,a total cost of less than £30 (sterling) and very low powerconsumption.

Fig. 11. MOOP’s Ultrasonic Wind Sensor.

We conducted a series of laboratory tests using a small deskfan. Figure 13 shows the raw wind direction as recorded whilethe sensor was spun 360 degrees. An averaging algorithm (de-scribed in section III-D) was used to generate a line of best forthe data. Prolonged testing suggests the accuracy is relativelypoor at somewhere between 10 and 20 degrees, however wehave found this is sufficient to position sail appropriately aswe only use a total of 10 unique sail positions. Unfortunatelydespite this promising laboratory test real world operationwas less successful due to the problems with temperaturecompensation. We have also found that if large droplets ofwater appeared on the sensor that the properties of the signalwere severely distorted although it was possible to detect thisas the sensor would return an extremely large raw value.Smaller droplets or a thin film of water on the sensor didnot appear to be a major problem.

D. Algorithms for Long Term Averaging of Wind Direction

Even the best wind sensors are likely to experience somelevel of noise and wind typically varies by a few degrees even

Fig. 12. A block diagram of the ultrasonic wind sensor.

Fig. 13. Raw data from the ultrasonic sensor show against averaged data.This with the sensor initially pointing into the wind. The sensor was thengradually spun over 360 degrees.

under optimal conditions. For this reason it seems sensible toattempt to average the wind sensor readings over time. Onepossible approach is to store a large number of readings in asliding window and constantly take an average (as describedin section III-B1), however as we increase the amount of timewe wish to average over the size of this buffer increases.Depending on the target computer architecture memory maybeextremely limited. The sliding window approach also givesequal weighting to all readings, regardless of age. If readingsare taken over a few seconds or even minutes this is unlikelyto be a problem, but if readings are taken over hours thenwe would probably need to reduce the importance of olderreadings. If we do not then changes in wind direction risk

IRSC'09 - page 29

being preserved in the average reading and the values will notbe reliable enough.

Instead of the sliding window approach, we can take aconstant running average. In order to take an effective averagewe must introduce a significant amount of historical databut we also wish to introduce a bias towards newer dataas the wind may have genuinely shifted and we would likeour control system to be able respond. Any system dealingwith rotational data is likely to encounter a problem whenaveraging data which lies either side of the wrap around point,for example averaging 350 degrees and 5 degrees. Our solutionto this problem is to take an average of the sine and cosinecomponents of the angle and then recombine them by takingtheir arc tangent. The algorithm is defined as follows:

s = s + (sin(w)− s)/r

c = c + (cos(w)− c)/r

d = atan2(s, c)

Where s is the average sine of the wind, c is the averagecosine, w is the current wind direction and d is the calculatedwind direction, r is the rate of change. The larger the valueof r the slower the average will change.

In all our robot designs the wind is measured relative tothe boat direction of the boat. In order to produce a long termaverage we must use the true wind direction (the compassheading of the wind) so we must subtract the current compassheading from the wind direction. As it takes sometime for theaverage heading to update we bootstrap the averages with thefirst reading we take in the hope that this is approximatelycorrect and will result in a faster convergence upon the realwind direction.

This algorithm has been shown to work well, but in con-ditions where the wind is constantly shifting (such as onmountain lakes where the wind funnels around nearby terrain)it can be too slow to update. In cases such as these takingan instantaneous or near instantaneous reading seems moresensible. Therefore it might be more appropriate to use thesliding window method described in section III-B1 or to setthe value of r very low.

IV. CONCLUSIONS AND FUTURE WORK

We have demonstrated the feasibility of wing sails to drivesailing robots and their potential to be highly robust butalso discovered their limitations with regards to reefing anddownwind sailing. Further work needs to be undertaken tooptimise wing shapes and sizes and to test the full potential ofusing multiple wing sails to improve performance and stability.

We have shown that a waterproof ultrasonic wind sensorcan be constructed using simple electronics and at a low cost,but that temperature calibration and long term averaging areessential. We have also shown that a contactless mechanicalsensor is also a viable option for long term use. For sailing

robots operating for short periods of time in sheltered condi-tions a basic mechanical sensor provides a cheap and viableoption but is not suitable for long term usage.

REFERENCES

[1] M. Neal, “A hardware proof of concept of a sailing robot for oceanobservation,” IEEE Transactions on Oceanic Engineering, vol. 31, 2, pp.462– 469, 2006.

[2] C. Sauze and M. Neal, “Design considerations for sailing robots perform-ing long term autonomous oceanography,” Austrian Journal of ArticialIntelligence, vol. 2, pp. 4–10, 2008.

[3] C. sauze and M. Neal, “An autonomous sailing robot for ocean observa-tion,” in proceedings of TAROS 2006, 2006, pp. 190–197.

[4] J. Alves, T. Ramos, and N. Cruz, “A reconfigurable computing system foran autonomous sailboat,” in proceedings of International Robotic SailingConference 2008, 2008, pp. 13–20.

IRSC'09 - page 30

Communication Architecture for AutonomousSailboats

Roland Stelzer∗† and Karim Jafarmadar∗∗Austrian Society for Innovative Computer Science

Kampstraße 15/1, 1200 Vienna, AustriaEmail: {roland.stelzer,karim.jafarmadar}@innoc.at

†Center for Computational IntelligenceSchool of Computing at De Montfort University

The Gateway, Leicester LE1 9BH, United Kingdom

Abstract—Although an autonomous sailboat can operate with-out human intervention a data link between boat and shore isnecessary. During development a reliable connection for moni-toring, debugging, and remote control in case of emergency isessential. When used for long-term observation tasks the operatoron shore is keen to receive real-time measurement data.

A three-stage communication system for autonomous sailboatshas been designed, implemented and tested successfully. It com-bines wireless LAN, GPRS/UMTS and satellite communicationfor a reliable data link between shore and sailboat.

I. INTRODUCTION

The vision of machines which relieve more and moretasks of humans induced many scientific initiatives in mobilerobotics. Research in fully autonomous sailing boats wasrecently stimulated by the Microtransat idea of Yves Briere[1]. The Microtransat Challenge is a transatlantic race offully autonomous sailing boats. The race aims to stimulatethe development of autonomous sailing boats through friendlycompetition. The Transatlantic race is currently planned forlate 2009.

An increasing number of research teams around the worldtry to teach their boats the complex task of sailing [2], [3].The best routeing decision, perfect handling of ever changingwind conditions and perfect timing during tack and gybe aresome of the skills an autonomous sailing vessel has to master.

A robotic sailboat is able to autonomously navigate towardsany given target without human control or intervention. Theoptimal route is calculated dependent on strategic goals andweather parameters. Rudder and sails are autonomously con-trolled in order to keep course and to execute manoeuvreslike tack and gybe. As sailboats operate in a highly dynamic,environment an autonomous sailboat has to respond quickly toever-changing environmental conditions. Incoming data fromsensors (GPS, compass, anemometer, etc.) have to be analysedpermanently by intelligent control mechanisms.

During the research process of an autonomous sailing boatit is very important to have a permanent data link to thevessel. On the one hand to take over control manually forsafety reasons if the artificial sailor on board does not workas expected or an obstacle crosses the boat’s trajectory. On theother hand it is quite convenient to have an opportunity for

real-time monitoring from shore in order to identify room forimprovement in the control algorithms of the boat.

Not only for research, but also for many possible applica-tions of autonomous sailboats [4], [5], a reliable communica-tion system between shore and boat is essential. Some of theseapplications are the following:

• Intelligent Sensor Buoys: An autonomous sailing boatcan operate energetically autonomously and can thereforecollect unlimited measured data from world’s lakes andseas. In other words this application would make possible,that surveying and mapping as well as water ecologicalstudies or recording of fishing resources would be rathercost-effective.

• CO2-neutral in Transportation of the Goods: The fuelprice is expected to grow dramatically in the next decadesand complying with Kyoto standards is getting more dif-ficult. Therefore better alternatives for the transportationof the goods need to be sought. Traditional sailing boatsare environment-friendly but they require a rather highnumber of human resource. An autonomous sailing boatwith its attributes from calculating the optimal itineraryto independently executing the right manoeuvres can bea serious option.

• Reconnaissance and Surveillance: An autonomous sail-ing boat can be sent out to far reaches or dangerousregions. Due to its silent, pilotless and energy self-sufficient attributes it is a safe alternative for surveillanceof smuggler-boats.

• Supply Vessel: Secluded regions with lower number ofinhabitants for instance researching base camps on is-lands can be cost-effectively supplied by the autonomoussailing boats with equipment, medicine, food or corre-spondence.

At least geographic coordinates of the boat or of the nexttarget need to be transmitted for the applications mentionedabove.

The next chapters gives an overview of the communicationpartners involved and a detailed description of a multi-stagecommunication architecture. Finally, some experiments carriedout in the Irish Sea are analysed and conclusions are made.

IRSC'09 - page 31

II. COMMUNICATION PARTNERS

Three entities are involved in the communication process:sailboat, visualisation software (Fig. 1), and remote controller.

Fig. 1. Screenshot of visualisation software

During normal operation the sailboat transmits sensor valuesto the visualisation. The visualisation is a software programwhich runs on a computer on shore and represents thetransmitted data like position or information about the boat’sstrategy clearly. If required it is also possible to send importantinstructions or data, such as new target coordinates, obstacleinformation, or a new desired course from the visualisation tothe sailboat.

In case of emergency, especially during test runs, an er-gonomically designed remote control device can be used tooverrule the autonomous onboard control of the sailboat.Therefore the desired actuator values, including position ofrudder and sails are transmitted to the boat in real time.

III. MULTI-STAGE COMMUNICATION ARCHITECTURE

The System consists of three alternative communicationchannels between boat and shore. Each of them featuresspecific advantages and disadvantages. Therefore it has to beensured, that the system switches dynamically between theavailable communication channels in order to use the mostappropriate way at any time.

A. First Stage: Wireless LAN

A radio mast, equipped with a directional wireless LANantenna, is mounted on shore. On the mast top of the boat anomnidirectional antenna is mounted. The higher the antennasare mounted, the longer is the distance that can be coveredby this technology. Experiments have shown a reliable 10 Mbdata link between boat and shore for up to 3 km distance.Both antennas have been mounted at a level of about 5 m.

Advantages of this technology are:• No base fee and no connection fee• High bandwidth (up to 57.6 kbit/s for GPRS respectively

up to 384 kbit/s for UMTS which allows even real-timevideo transmission)

• TCP-based communication; software maintenance can becarried out during runtime

Disadvantages are:• Infrastructure (antenna, router) needs to be set up• Limited operating distance

B. Second Stage: Data Service of Mobile Phone Provider

The boat as well as a server on shore is equipped witha data modem of a commercial mobile phone provider. Thisallows internet-based communication between these two sta-tions. Common data modems provide UMTS (Universal Mo-bile Telecommunications System) and GPRS (General PacketRadio Service) and switch over automatically depending onthe availability of the services.

Advantages of this technology are:• Infrastructure is provided by the mobile phone service

provider• High bandwidth which allow even real-time video trans-

mission, at least for UMTS• TCP-based communication; software maintenance can be

carried out during runtimeDisadvantages are:• Base fee and connection fee, can be extremely high

abroad (roaming)• Operating distance is limited by the network coverage of

the service provider 1

C. Third Stage: Satellite Communication

The sailboat is equipped with an Iridium satellitetransceiver. The Iridium satellite constellation is a system of66 active communication satellites with six spares in orbit andon the ground. It allows worldwide voice and data communi-cations. The Iridium network is unique in that it covers thewhole earth, including poles, oceans and airways.

Various Iridium based services are available. The mostappropriate for communication to an autonomous sailing boatis the Iridium Short Burst Data (SBD) Service. Iridium SBDService is designed to serve a range of applications that needto send data messages that on average are typically less than300 B. The used Iridium 9601 SBD transceiver on the sailboatis controlled by AT commands over an RS232 serial port.An Iridium Gateway allows receiving data from respectivelytransmitting data to the sailboat via e-mail [8].

Advantages of this technology are:• Covers the whole surface of the earth• The Iridium SBD transceiver offers not just message

transmission, but delivers rough geographic position in-formation as well. This can be used as backup system ifthe GPS receiver on board fails.

Disadvantages are:

1According to GSM, which is the basis for GPRS communication, technicalspecification the maximum distance between mobile station and base stationis 35 km. [6]. However, a viable communication link can be expected for adistance of about 15 km as shown in [7].

IRSC'09 - page 32

• Low data volume (max. about 2 kB pro message [8],which can be sent approximately every 30 s [9])

• High transmission latency (between 7 s and 22 s modemprocessing time [10])

• Base fee and connection fee (USD 21.00 per month plusUSD 1.30 per 1,000 B at the provider used for thisexperiment)

IV. SELECTION OF THE APPROPRIATE COMMUNICATIONSTAGE

Not every communication technology is adequate for everycombination of communicating devices. Any of the threestages can be used between sailboat and visualisation, eventhough with different scope of operation. A reasonable com-munication between sailboat and remote control requires aline of sight to the boat and real time transmission of theinstructions. Therefore, for this case only Wireless LAN andUMTS/GPRS are suitable. (Fig. 2)

Fig. 2. Communication partners and available communication stages

The selection of the appropriate communication stage ismainly based on the availability of the data networks. Con-sidering advantages and disadvantages of each technology thefollowing strategy is implemented.

If a proper wireless LAN link to the boat is available, itwill be used for visualisation and remote controller. If thelink quality decreases below a certain threshold, the systemautomatically switches over to UMTS/GPRS if available. Forboth communication partners this happens transparently. Ifneither wireless LAN nor UMTS/GPRS is available, satellitecommunication via Iridium SBD will be activated. In this casedirect manual control of rudder and sail position is not possibleand not reasonable. This is because of the limited sight causedby long distance to the sailboat and the high latency of satellitecommunication.

The availability of all communication stages is verifiedperiodically. If indicated, a switch over will be initiated. Ahysteresis condition is implemented in order to avoid con-stantly switching between the communication stages.

When no communication stage is available, the boat keepson sailing fully autonomously. The sailing functionality ofthe boat is completely independent from the communicationsystem. A user on shore is not able to monitor the boat’s dataor to influence its strategy in this situation.

V. FEATURES OF THE INDIVIDUAL COMMUNICATIONSTAGES

Wireless LAN and UMTS/GPRS enables the user to requestall available data from the sailboat. There are two differentmodes:

• Pull mode: a request for a certain value is sent to thesailboat and immediately answered.

• Push mode: the client software can subscribe to a setof values. The sailboat delivers every update of thesubscribed values automatically.

These communication stages also allow transmission ofmultimedia data, such as images, videos, or sound files.Furthermore it is possible to remotely log into the boat’s maincomputer. This can be used to check log files, update software,or to restart system services during runtime.

The third communication stage, satellite communication,has restrictions concerning bandwidth and latency. Therefore itfocuses on transmission of short and concise data packages. Bydefault only the location data (GPS coordinates) of the sailboatare transmitted regularly. Further values (wind speed, batteryvoltage, etc.) are transmitted if certain thresholds are reached.As Iridium SBD provides communication in both directionsany other value can be requested on demand and high levelcommands such as new waypoints can be submitted.

VI. EXPERIMENTS DURING MICROTRANSAT 2007

A. Microtransat 2007

Prior to the ultimate goal of the Microtransat Challenge,the fully autonomous crossing of the Atlantic Ocean, severalsmaller competitions have already taken place. These compe-titions allowed researchers to exchange ideas and test theirboats in less harsh environments.

The second of these pre-races took place on the Irish Seaoff the coast of Aberystwyth, Wales, UK in September 2007.It was the first competition on sea. Aside from ASV roboatfrom the Austrian Society for Innovative Computer Science(InnoC), which is the basis for the current work, three otherteams (Aberystwyth University, ENSICA/IUT de Nantes, andQueens University) took part.

The InnoC team was announced the winners by the judges,who released the following statement:

”We are very pleased to announce that the overallwinners of this event were the Roboat team fromAustria. This team demonstrated true Autonomoussailing for the full 24 h period of the challenge anda very robust control system. ...” [11]

B. ASV Roboat

The ASV Roboat is an autonomous robotic sailboat, whichis able to autonomously navigate towards any given targetwithout human control or intervention. The optimal routeis calculated by weighting drift coordinates against weatherparameters. The rudder and sails as well as the tacks and gybesare autonomously controlled by incoming data from sensors(GPS, compass, anemometer, etc.) [12]–[14]. The Roboat won

IRSC'09 - page 33

the international Microtransat event in Toulouse. In September2007 the InnoC team won the Microtransat again on the IrishSea in Aberystwyth, Wales. And in May 2008 the ASV roboatbecame first World Champion in robotic sailing. The WorldRobotic Sailing Championship was held on Lake Neusiedl inAustria.

The basis for the autonomous sailboat ASV roboat is a com-mercial sailboat designed by Jan Herman Linge, the boat typeLaerling. The Laerling is a product from the same drawingboard as the well-known Olympic class Soling and Yngling.The boat was originally created for kids to learn sailing, andtherefore safety and stability are the major characteristics ofthe boat. It comprises a 60 kg keel-ballast, which will bringthe boat upright even from the most severe heeling. A briefoverview of the adaptations and the technical equipment onboard can be found in Figure 3.

ASV Roboat at a glance

Boat characteristicslength: 3.75 m• total displacement: 300 kg• main and foresail: 4.5 m²• self-righting hull design•

ComputerMini-ITX with Linux• 800MHz / 512MB RAM• 4GB CF Card• Control software in Java und C++•

Actuatorssheet chain drive • balanced rudder • automatic bilge pump• fog horn, lights•

Sensors (NMEA2000/CAN-Bus)position and speed over ground (GPS)• speed through water• battery voltage and power consumption• ultrasonic wind sensor • tilt-compensated compass• humidity, air and water temperature, depth•

CommunicationWireless LAN• UMTS/GPRS• Iridium satellite modem•

Energy balance50W avg. power consumption• 1.5m² solar panels (285 Wp)• 2 x 230 Ah lead batteries• direct methanol fuel cell (65 W) •

Fig. 3. ASV Roboat at a glance

C. Communication Setup

The experimental setup provides all three communicationstages. Figures 4 to 6 illustrate the interaction of the involvedcommunication partners for each stage separately.

First stage - Wireless LAN communication: A laptop forvisualisation and remote control is connected to a directionalwireless LAN antenna, which is mounted on a 5 m mast onshore. This is used to establish a connection to an omnidirec-tional wireless LAN antenna on mast top of the boat. (Figure4)

Second stage - UMTS/GPRS communication: Both com-munication partners are connected to the internet via aUMTS/GPRS modem. The communication partners are con-nected via a virtual private network (VPN). (Figure 5)

Third stage - satellite communication: The sailboat isequipped with an Iridium SBD modem, which transmits datavia the Iridium satellite network. Data packages are forwardedto a mail server by the Iridium provider. The visualisationsoftware is connected to the Internet via a UMTS/GPRS datalink in order to fetch transmitted data from the mail server.(Figure 6)

Fig. 4. First stage - Wireless LAN communication

Fig. 5. Second stage - UMTS/GPRS communication

D. Experimental Results

Figure 7 compares the network coverage of all three com-munication stages in Aberystwyth, where Microtransat 2007took place. For runs near shore, wireless LAN coveragewas sufficient and UMTS/GPRS served as backup system.Whenever the sailboat left the area covered by wireless LANthe system switched to UMTS/GPRS automatically withoutuser intervention and vice versa.

Fig. 6. Third stage - satellite communication

IRSC'09 - page 34

Fig. 7. Network coverage in Aberystwyth (UK), Microtransat 2007

As there was no race outside UMTS/GPRS coverage, theUMTS/GPRS modem was deactivated manually to test theswitchover to satellite communication. As soon as the boatsailed out of the wireless LAN covered area it switched overdirectly to satellite communication, because UMTS/GPRS wasnot available. In this stage the system transmitted the boat’sGPS position every 5 min as configured. Remote control of theboat was not possible until the boat sailed back autonomouslyand received a wireless LAN signal again.

Power consumption is a very important feature for au-tonomous sailboats especially in long term missions. Thereforepower consumption of the communication equipment of theindividual stages is compared in Table I. Since stages I andII are primarily used for test runs and short missions closeto shore, power consumption is not of particular interest.Therefore for long term missions only stage III is integratedin the energy balance.

Communication Device Power ConsumptionWireless LAN (Buffalo WHR-HP-54G) 3.4 W [15]

UMTS (HUAWEI E220) ≤ 2.5 W [16]Satellite (Iridium SBD 9601) 1.75 W

(0.33 W stand-by)2 [17]

TABLE ICOMPARISON OF POWER CONSUMPTION OF USED COMMUNICATION

DEVICES

VII. CONCLUSIONS

Several test runs have shown that real-time communicationfor monitoring and remote control is of importance for safetyespecially in the field of vision close to shore. In case ofemergency immediate manual control has to be possible.

On the other hand, for long-term missions over huge dis-tances the main focus is on global network coverage andreliable transmission of a few essential values. Higher trans-mission latency can be accepted.

21.75 W is the average power consumption when a SBD message transfer isin process. The actual power consumption depends on the view of the satelliteconstellation from the antenna.

A three-stage communication system for autonomous sail-boats has been designed, implemented and tested successfully.Advantages and disadvantages of the individual stages havebeen illuminated with regard to availability, costs, bandwidth,and real-time abilities. Requirements during development ofautonomous sailboats as well as its applications have beenconsidered separately.

Based on the experiments carried out, the presented conceptturned out as appropriate. Long-term test runs have to be donein order to get more valuable results on longevity, maintenanceeffort, and robustness, especially in harsh environments.

REFERENCES

[1] Y. Briere, F. Bastianelli, M. Gagneul, and P. Cormerais, “Challengemicrotransat,” in CETSIS’2005, Nancy, France, 2005, vol. 5, Issue 2,DOI:10.1051/j3ea:2006031.

[2] C. Sauze and M. Neal, “Design considerations for sailing robots per-forming long term autonomous oceanography,” in International RoboticSailing Conference (IRSC). Breitenbrunn, Austria: Austrian Society forInnovative Computer Science, May 2008, pp. 21–29.

[3] J. C. Alves, T. M. Ramos, and N. A. Cruz, “A reconfigurable com-puting system for an autonomous sailboat,” in International RoboticSailing Conference (IRSC). Breitenbrunn, Austria: Austrian Societyfor Innovative Computer Science, May 2008, pp. 13–20.

[4] N. A. Cruz and J. C. Alves, “Ocean sampling and surveillance usingautonomous sailboats,” in International Robotic Sailing Conference(IRSC). Breitenbrunn, Austria: Austrian Society for Innovative Com-puter Science, May 2008, pp. 30–36.

[5] C. Sauze and M. Neal, “An autonomous sailing robot for ocean obser-vation,” in Towards Autonomous Robotic Systems (TAROS), Surrey, UK,2006, pp. 190–197.

[6] M. Medved and A. Tekovic, “Extended range functionality for gsmnetworks,” in 46th International Symposium Electronics in Marine, June2004, pp. 211–216.

[7] A. V. Alejos, I. Culnias, M. G. Sanchez, and J. A. G. Fernandez,“Distress beacons for maritimal accidents: Measurements of coastalsea coverage,” in EuCAP 2007. The Second European Conference onAntennas and Propagation, November 2007, pp. 1–6.

[8] (2003) Iridium satellite data services white paper, version 1.0.[9] A. Sybrandy. (2007) Remote data acquisition telemetry options.

[Online]. Available: Scripps Technical Forum, University San Diego(accessed on 20 May 2009) http://stf.ucsd.edu/presentations/2007-02%20STF%20-%20Sybrandy%20-%20Telemetry.pdf

[10] M. M. McMahon and R. Rathburn, “Measuring latency in iridiumsatellite constellation data services,” in International Command andControl Research and Technology Symposium (ICCRTS), 2005.

[11] (2009) Official microtransat website. [Online]. Available:Aberystwyth University, Wales (accessed on 18 April 2009)http://www.microtransat.org/2007.php?lang=en

[12] R. Stelzer and K. Jafarmadar, “A layered system architecture to controlan autonomous sailboat,” in Towards Autonomous Robotic Systems(TAROS 2007), Aberystwyth, UK, September, September 2007, pp. 153–159.

[13] R. Stelzer, T. Proell, and R. I. John, “Fuzzy logic control system forautonomous sailboats,” in FUZZ-IEEE, London, UK, July 2007, pp. 97–102.

[14] R. Stelzer and T. Proell, “Autonomous sailboat navigation for shortcourse racing,” Robotics and Auotnomous System, vol. 56, no. 7, pp.604–614, July 2008.

[15] Buffalo airstation high power turbo g high power wirelesscable/dsl smart router whr-hp-g54. [Online]. Available:Buffalo Technology Inc. (accessed on 20 May 2009)http://www.buffalotech.com/files/downloads/WHR-HP-G54 DS.pdf

[16] System description of huawei e220 usb modem. [Online].Available: Huawei Technologies Co., Ltd. (accessed on 20 May 2009)http://corporate.proximus.be/download/VMC Manual Huawei E220.pdf

[17] Iridium 9601 sbd transceiver datasheet. [Online]. Available:Iridium Satellite LLC (accessed on 20 May 2009)http://www.iridium.com/about/press/pdf/Iridium 9601 r2.pdf

IRSC'09 - page 35

IRSC'09 - page 36

Model Sailboats as a Testbed for ArtificialIntelligence Methods

Ralf Bruder, Birgit Stender, Alexander SchlaeferInstitute for Robotics and Cognitive Systems

University of LuebeckRatzeburger Allee 160

D-23538 Luebeck, GermanyEmail: {bruder,stender,schlaefer}@rob.uni-luebeck.de

Abstract—The design of autonomous robots is a challengingresearch topic involving a wide range of algorithms and methodsfrom artificial intelligence. A number of ambitious projects likethe DARPA challenge, RoboCup, or the Microtransat present welldefined scenarios for the development of specialized autonomousrobots. We propose to add robot sailing using model sailboatsas another scenario. Similarly to the RoboCup soccer robots,model sailboats are relatively cheap and easy to transport. Onthe one hand, this allows to study methods needed to build andimprove ocean going autonomous vessels. Moreover, it wouldbe fairly straightforward to have one design races of robotsailboats against humans. We present results on building twosmall and lightweight robot sailboats which we plan to use as atest environment in a student project on artificial intelligence.

I. INTRODUCTION

Autonomous robots have long been a fascinating researcharea with applications ranging from fully autonomous vehicles[1] to robotic soccer [2], [3], [4]. The success of RoboCupillustrates how a well defined experimental environment canfoster the understanding of concepts and methods to imple-ment autonomous behavior.

Only recently, robotic and autonomous sailing has beenstudied [5], [6], [7], [8], [9], [10]. While auto-pilots havebeen in use for long time, implementing optimal steering ofa sail boat is a rather intricate task [6]. Moreover, completelyautonomous sailing involves planning and decision makingin an unknown and constantly changing environment. Hence,apart from having useful applications [7], [9], [10], it presentsan interesting testbed for methods from artificial intelligence.

As is obvious to every avid sailor, two boats make a race.While this holds for both, long distance competitions andaround-the-buoys regattas, the latter have the advantage ofbeing easier and cheaper to organize. We propose to use small,lightweight model sailing boats to study various aspects ofautonomous sailing. To this end we plan to use two boats as atest environment for a student project in artificial intelligence.In this paper we present the current design of the boats, andillustrate the problems that can be analyzed within our testenvironment.

II. TECHNICAL IMPLEMENTATION

Model sailboats are popular and a number of classes exist,for example the International One Metre (IOM) class, or the

TABLE IDIMENSIONS OF THE GRAUPNER UNIVERSITY CLUB

Dimension Value

Length 515 mm

Beam 120 mm

Height 857 mm

Weight 930 g

RC Laser class. Typically, the boats are remote controlled withtwo degrees of freedom, the angle of rudder and the sails.While the boats vary in size, serious racing designs often costmore than 1000 EURO, without being autonomous, of course.Although it would be preferable to study boats that are alreadywidely popular, we put the focus on a quick and affordablesolution.

A. The boat

Serious model racing boats are rather costly and often builtby specialized boat builders. After some research on ready-to-sail models we opted for a University Club (Graupner,Germany), mainly because the model is small and easy totransport and offers a lot of space inside the hull for its size.The overall dimensions are given in Table I.

The boat can hold approximately 300 g additional weight,and some of the original parts, e.g., servos, radio controland battery package were removed or replaced by lightercomponents. Particularly, the original design used sheets toposition the main sail and jib. Both sheets were connected toa single servo, leading to a number of disadvantages. First,the actual position of the sails could not be determined, e.g.,as loose sheets would allow the boom to take any positionfrom maximum starboard to maximum port. Second, the twosails could not be controlled independently, e.g., to backwindthe jib or to run wing-on-wing. Therefore, a small servo wasmounted to each boom, allowing a boom position of ±90◦

from centerline. Note that model boats often use a boom forthe jib, too. A third servo of the same type was used to controlthe rudder, where the actual angle is approximately ±45◦.

IRSC'09 - page 37

Fig. 1. The controller board including an Atmel processor for serialcommunication and to control servos, inclinometer and wind vane.

Fig. 2. Communication between ATMEGA8 processor, sensors and actuators.

B. Sensors and servo control

Clearly, a good helmsman will sense the boat and thewind in many ways. While large keel boats often includeinstruments summarizing the speed of wind and boat indifferent ways, every dinghy sailor first learns to pay attentionto the wind direction and the boat heel. Hence, we includean inclinometer and a small wind vane as essential sensors inour design. Additionally, an autonomous vessel would need todetermine its position, speed, and direction with respect to theworld. In the current design we rely on a small bluetooth GPSmodule to obtain position, speed over ground and course overground. To interface with sensors and servos, a small controllerwas developed as a printed circuit board (PCB) with an AtmelATMEGA8 processor. Figure 1 shows the controller board.

The processor runs a simple program which receives andprocesses messages from a serial RS232 interface. Thus,output parameters can be set while measured sensor data arecollected and periodically transmitted in the opposite directionover the same serial link.

TABLE IIFEATURES OF THE INCLINOMETER

Cost-efficiency approx. 10 EURO

Full scale acceleration ± 2 g

Bandwith 640 Hz

Resolution 12 bit

Interface I2C or SPI

Low power consumption max. 0.8 mA

The ATMEGA8 is equipped with one 16-bit and two 8-bit timers with additional output logic which can be usedto generate up to four asynchronous pulse width modulatedsignals. We use three of those outputs to generate phase andfrequency correct servo control signals. In this way the threeRC servos for the rudder, main and foresail are controlledwith highly accurate timings and without interrupting the mainprogram.

In addition to the pulse width modulation (PWM) unit theprocessor features an asynchronous analog to digital converterwith eight selectable inputs and up to 10 bit resolution. Thisunit is used to constantly monitor the voltage of the mainbattery. Furthermore, the analog output of the wind sensor isdigitalized and converted to an angular value representing thewind direction.

Moreover the results of a three-axes acceleration sensor areperiodically polled via I2C bus using a chip-specific protocol.Figure 2 shows a schematic of the different input and outputsignals to the Atmel processor.

1) Inclinometer: A simple and effective way to estimatethe motion of the boat is to measure the rotation about itsaxis, i.e., roll, pitch, and yaw. The roll value is of particularinterest, as it represents the heeling of the boat and presentsa surrogate for the wind speed. Typically, boats sail faster ifthey are upright, and hence the roll, together with the winddirection, will be an important input to algorithms controllingthe boat. Usually, two or more sensors are mounted in a boatto measure the acceleration about different axes, with the roll,or heeling, computed from those measurements. For our smallboat we included the LIS3LV02DQ inclinometer because itintegrates acceleration sensors for all three axes in one IC andoffers both I2C and SPI interface for easy connection to microcontrollers.

The chip is placed inside the boat so that the x-axis foracceleration measurement is aligned to the forward axis ofthe boat and the z-axis of the sensor is pointing downwards.In this way the acceleration readings in x- and y-axis can bedirectly converted to roll and pitch values using trigonometricfunctions. The chip’s main features are summarized in TableII.

2) Wind sensor design: The only possible position on boardfor undisturbed measuring of the wind direction is at the masttop. One major goal of the design was to achieve a mass below20 g in order to reduce the effect on the boat dynamics dueto the sensor position. Other design goals are listed in TableIII below.

IRSC'09 - page 38

Fig. 3. Schematic diagram of the wind sensor.

Fig. 4. The wind vane.

We decided to use the contactless magnetic encoder AS5030which measures the angular position of a diametrically mag-netized 2-pole permanent magnet with a resolution of 8 bit.A hall array inside the sensor chip determines the orientationof the magnetic field strength. The PWM sensor output signalis low pass filtered and divided due to different voltage levelsof the sensor chip and the controller. In principle, the sensordesign would be suitable for measuring wind direction or windspeed. The latter would require a cup-type anemometer, andcurrently the boats are only equipped with a wind vane tosense the direction. The printed circuit board is placed insidea 2x2 Lego brick. The ball bearing rotary axis (sewing needle)with the permanent magnet is located outside the Lego brickto ensure watertightness. To compensate the orientation of the

TABLE IIIFEATURES OF THE WIND SENSOR

Small total error below ±5◦

Cost-efficiency approx. 30 EURO

Low weight 5 g

Compensation of orientation balancing out wind vane

Simple interface analog output 0 - 3.3 V

Watertightness sensor and rotary axis separated

Low power consumption approx. 20 mA

Fig. 5. Power distribution to the different components.

boat the wind vane needs to be carefully balanced out acrossthe rotary axis. A schematic illustration of the design is givenin Figure 3 and the actual wind vane is shown in Figure 4.

3) Power distribution: As main battery we have integrateda lithium polymer accumulator with 7.4V and 2.1Ah. Thisbattery has sufficient capacity to drive the boat for at least threehours. For the controller board the input voltage is regulatedto 3.3 Volt. Another regulator provides 5 Volt to drive theservos. While PDA and GPS come with batteries yieldingapproximately 3 hours of operation, the main battery wasadditionally connected to charge both modules. The internalbatteries are therefore no longer necessary for operation andmay be removed later to further reduce the weight.

C. Control and communication

While the sailboat will be able to operate completelyautonomous, there are various reasons to keep a communi-cation link. First, it is always advisable to have a fallback tomanual control, either in emergencies or to avoid collisionswith objects not known to the boat. Second, the boat has alimited set of sensors and given its size it will be virtuallyimpossible to add, e.g., cameras or laser range finders. Yet,it is easy to maintain a world model on a central server, andto simulate objects within this world model. This informationcan be transmitted to the vessel and handled like sensor data,therefore extending the experimental setup substantially. Third,it would be interesting to keep track of the onboard sensorreading while sailing. Forth, wireless communication withonboard sensors like the GPS is easy to implement, e.g., usingBluetooth.

Further requirements for a main control unit are lightweight and easy programmability. After considering the costand time effort to build a board ourselves we decided tobuy a used PDA instead. At a weight of 139 g the PDAincludes WLAN, Bluetooth, an SD slot and a battery. For ourcurrent experiments the installed Windows Mobile 2003SE isa suitable operating system, particularly because it is easy toprogram.

The PDA runs a basic program for sensor/actuator handlingand communication. Two serial handlers process messages

IRSC'09 - page 39

Fig. 6. Thread-safe access to sensor and actuator values.

TABLE IVCOSTS FOR ONE SAILBOAT

Component Description Price (EURO)

Sailboat Graupner University Club 99.00

Servos (Jib, Main, Rudder) Hitech 140BB 33.00

LiPo Kokam (2100mA/7.4V) 31.00

PDA Dell Axim X30 (used) 70.00

GPS Royaltek BT2100 30.00

Controller Atmel + Incliniometer 20.00

Wind vane AS5030 chip, bearings, etc. 30.00

Board, components 30.00

from the Bluetooth GPS module and the controller board. Allsensor readings are stored in one central, thread-safe object.There are further interfaces to this object for a network TCPserver which can be used for remote control or distributionof external data and for a user-defined program used forautonomous control. An overview about this structure is givenin Figure 6. Table IV summarizes the main components andtheir approximate cost. The boat and its components is shownin Figure 7.

III. SAILING AS A TESTBED FOR ARTIFICIALINTELLIGENCE

It is planned to use the two sailboats as a testbed for variousmethod from artificial intelligence. Particularly, we plan to usethe boats for a student project, where algorithms can be studiedin practice. There is a wide range of practical problems thatcan be studied, as detailed below.

A. Test scenarios

The basic requirement for sailing is boat control. Hence,a first test would involve the optimal and robust implemen-tation of maneuvers like tack and jibe, sailing a course, andsteering relative to the wind. For example, machine learningapproaches can be used to find the way to detect and reactto changes in wind direction and wind speed. While the wind

direction can be read from the wind vane, the wind speedresults in an indirect reading: the boat starts to heel as thewind increases, which can be read from the inclinometer. Thealgorithms need to be trained to react such that the boat keepsits speed, i.e., by slacking the sails slightly or luffing early.

As a second test, we plan to run small round-the-buoy raceswith the two practically identical boats. Hence, there is a clearmetric to evaluate the implemented methods: the first boat tofinish wins. Since the sailing area can be covered by WLAN,it is also straightforward to implement a more complex racingscenario with realistic rules of racing. For example, both boatsare sending their respective position, speed and direction to acentral server, which in turn broadcasts the data. Then, theboats can implement strategical and tactical behavior reactingto the other boats performance. While boat-to-boat communi-cation would be possible, there is another advantage to thecentral server approach. Much like data on the competingboats, data on other real or virtual objects could be sentto the boats. For example, a large vessel following a trafficseparation scheme could be simulated, and the boat’s reactionwith respect to the Convention on the International Regulationsfor Preventing Collisions at Sea (COLREG) can be evaluated.This all is possible without adding further sensors to the boats.

A third scenario would be autonomous long distance trips.Clearly, long distance will be in the range of a few nauticalmiles and far from crossing the atlantic ocean. However, itis still possible to test the boats performance when runningcompletely autonomous and adapting to different wind andweather conditions on the way.

B. Problems to be studied

Sailing is a fairly complex matter, and the problems thatcan be studied when developing an autonomous sailboat havemany different applications. Clearly, robust localization is akey requirement for autonomous robots. Likewise, path plan-ning and collision avoidance methods can be studied. Methodsand filters to read and fuse sensor data have applications fromaviation to medical devices. The same holds for predictionmethods, which in sailing could be applied to forecast changesin the wind direction or in the position of other vessels.

IV. RESULTS AND LESSONS LEARNED

Both boats have been modified and first test runs undermanual control have been successfully completed. From thesefirst results, all components worked robust and the boatssailed surprisingly stable in approximately 2-3 Bft winds. Theoverall effort in time and money was relatively small, withapproximate costs of less than 700 EURO for both boats,and about one person month time involved. Yet, while thecurrent boats cost a fraction of an IOM class boat, it wouldbe preferable to have a standard platform to compare resultsagainst both, autonomous and human competitors. Moreover,the current boat is too small to carry additional sensors.Particularly, we would like to include sensors to scan thedirect proximity of the boat for fully autonomous collisionavoidance.

IRSC'09 - page 40

Fig. 7. The autonomous model sailboat with all components: A) boat and host computer, B) PDA and control software, C) rudder servo, D) PDA, E) mainbattery, F) main sail servo, G) controller board, H) jib sail servo, I) GPS module.

For our teaching purposes, complete autonomy may not be akey requirement, though. In fact, from our current experiencewith a group of seventeen students we would argue that anapproach where the boats can be controlled from a computeris preferable. The students quickly implemented basic control,filter, and search algorithms, and learned from studying datain realtime. Such a setup could also be used for robotic sailingcompetitions.

A number of approaches to autonomous sailing have usedlarger model boats to start their research [5], [7], [8]. We haveillustrated that it is possible to build very small, lightweight,and cost-efficient boats which could attract new participantsto robotic sailboat racing.

V. CONCLUSION

We have presented a quick and cheap approach to establisha testbed using artificial autonomous model sailboats. Theboats have been successfully tested and it is planned toimplement intelligent and autonomous behavior in the courseof a student project. While we agree that the design of au-tonomous sailboats for long distance applications, e.g., withinthe Microtransat is an important research goal, development,maintenance and test of the boats can be relatively costly.Many of the underlying problems may also be studied in modelsailboats, having the advantage of being small, transportable,

and relatively cheap. Hence, we propose to consider a sailingleague comparable to RoboCup, where teams compete withinone design or construction classes. While RoboCup has de-clared the goal to ”create robots capable of beating the worldcup in 2050” [4], winning the world championship in a onedesign model sailboat class would be an challenging goal, too.

REFERENCES

[1] S. Thrun, M. Montemerlo, H. Dahlkamp, D. Stavens, A. Aron, J. Diebel,P. Fong, J. Gale, M. Halpenny, G. Hoffmann, K. Lau, C. Oakley,M. Palatucci, V. Pratt, P. Stang, S. Strohband, C. Dupont, L.-E. Jen-drossek, C. Koelen, C. Markey, C. Rummel, J. van Niekerk, E. Jensen,P. Alessandrini, G. Bradski, B. Davies, S. Ettinger, A. Kaehler, A. Ne-fian, and P. Mahoney, “Stanley: The robot that won the DARPA grandchallenge,” Journal of Field Robotics, 2006, accepted for publication.

[2] H. Kitano, M. Asada, Y. Kuniyoshi, I. Noda, E. Osawa, and H. Matsub-ara, “Robocup: A challenge problem for ai,” AI Magazine, vol. 18, pp.74–85, 1997.

[3] H. Kitano, M. Asada, Y. Kuniyoshi, I. Noda, and E. Osawa, “Robocup:The robot world cup initiative,” in AGENTS ’97: Proceedings of the firstinternational conference on Autonomous agents. New York, NY, USA:ACM, 1997, pp. 340–347.

[4] H.-D. Burkhard, M. Asada, A. Bonarini, A. Jacoff, D. Nardi, M. Ried-miller, C. Sammut, E. Sklar, and M. Veloso, “RoboCup: Yester-day, Today, and Tomorrow Workshop of the Executive Committee inBlaubeuren, October 2003,” ser. LNAI 3020, D. P. et al., Ed., 2004, p.1534.

[5] J. Abril, J. Salom, and O. Calvo, “Fuzzy control of a sailboat,”International Journal of Approximate Reasoning, no. 16, pp. 359–375,1997.

IRSC'09 - page 41

[6] M. L. van Aartrijk, C. P. Tagliola, and P. W. Adriaans, “Ai on the ocean:the robosail project,” in ECAI 2002, F. van Harmelen, Ed. IOS Press,2002, pp. 653–657.

[7] C. Sauze and M. Neal, “An autonomous sailing robot for ocean obser-vation,” in Proceedings of TAROS 2006, 2006, pp. 190–197.

[8] R. Stelzer, T. Proll, and R. John, “Fuzzy logic control system forautonomous sailboats,” in IEEE International Conference on FuzzySystems, 2007.

[9] C. Sauze and M. Neal, “Design considerations for sailing robots per-forming long term autonomous oceanography,” in Proceedings of IRSC2008, 2008, pp. 21–29.

[10] N. A. Cruz and J. C. Alves, “Ocean sampling and surveillance usingautonomous sailboats,” in Proceedings of IRSC 2008, 2008, pp. 30–36.

IRSC'09 - page 42

AAS Endurance:An autonomous acoustic sailboat for

marine mammal researchHolger Klinck∗, Roland Stelzer†‡, Karim Jafarmadar† and David K. Mellinger∗

∗Cooperative Institute for Marine Resources StudiesOregon State University, Hatfield Marine Science Center

2030 SE Marine Science Drive, Newport, OR 97365, USAEmail: {holger.klinck,david.mellinger}@oregonstate.edu

†Austrian Society for Innovative Computer ScienceKampstraße 15/1, 1200 Vienna, Austria

Email: {roland.stelzer,karim.jafarmadar}@innoc.at‡Center for Computational Intelligence

School of Computing at De Montfort UniversityThe Gateway, Leicester LE1 9BH, United Kingdom

Abstract—This paper presents a joint research project ofthe Austrian Society for Innovative Computer Science, Austriaand Oregon State University, USA which is intended to berealised within the next three years. The aim of the project is todevelop an autonomous sailboat for passive acoustic monitoringof marine mammals and mitigation of human impacts on them.Performance tests of the autonomous acoustic sailboat - AASEndurance - will include an open sea transect of at least onemonth duration. The work presented here discusses shortcomingsof current ways of acoustic marine mammal monitoring andoutlines advantages of a robotic sailboat for this task, as well asproblems to be solved with this new technology.

Index Terms—autonomous sailboat, robotics, marine mam-mals, bioacoustics, passive acoustic survey, underwater acoustics,line transect

I. INTRODUCTION

Passive acoustic monitoring (PAM) is a widely used tech-nique to estimate the abundance and distribution of marinemammals. A principal problem of PAM is the limitation inspatial and temporal coverage of the observations (see Fig. 1).Measurement can either be done with a moving platform(e.g., research vessel) or stationary recording devices (e.g.,anchored autonomous recorders). Moving platforms offer thepossibility of sampling a large area in a short period of time.However, because of the high costs of ship time, such passiveacoustic line transects can be conducted only occasionally,and temporal coverage is very limited. In contrast, stationaryrecording devices [1] allow continuous sampling of an area.Their disadvantage lies in the limited spatial coverage of thedevices.

Autonomous and remotely navigable passive acoustic plat-forms offer the possibility of sampling an area of interestwith high temporal and spatial resolution at low cost. In thispaper we introduce such a technology based on an autonomousacoustic sailboat (AAS). The extended payload and availability

Fig. 1. Comparison of the spatial and temporal coverage of ship transects(dotted line) and stationary recorders (dashed circles)

of energy on the proposed research platform allows operationof additional sensors such as measurement of chlorophylland zooplankton density. The multi-sensor platform is there-fore well-suited for investigating broader oceanographic andecological questions, including predator-prey dynamics, patchscales, prey densities, and trophic energy flow.

II. AUTONOMOUS AND REMOTELY NAVIGABLE PASSIVEACOUSTIC PLATFORMS FOR MARINE MAMMAL RESEARCH

To date, two autonomous and remotely navigable passiveacoustic platforms are available for marine mammal research:wave-powered vessels (e.g., the Wave GliderTM [2]) andocean gliders (e.g., the SeagliderTM [3]).

The Wave Glider provides a submerged (swimmer) anda surface (float) unit. Both units are connected via a tetherand allow the swimmer to move up and down as a result

IRSC'09 - page 43

of wave motion. The swimmer includes several fins whichinteract with the water as the swimmer moves up and down,and generate forces which propel the vehicle forward. TheWave Glider, developed by Liquid Robotics, Inc., has provenlong-term capabilities in a five-month test trial, and the deviceseems well-suited for long-term passive acoustic monitoring ofmarine mammals. However, as the Wave Glider is a relativelynew device, and to date there have been no reports of long-term acoustic recording capability, the following discussionwill focus on the comparison of gliders with the proposedautonomous acoustic sailboat.

Gliders are commercially available from several manufac-turers (e.g., [4]), and all types are based on the same principle.Changes in buoyancy cause the glider to move down and upin the water, and as with a airplane gliders, wings transformthis vertical motion into forward motion. A stable, low-drag,hydrodynamic shape allows the glider to fly efficiently throughthe oceans. These devices are optimized for extremely lowenergy requirements and designed to operate at depths up to1000 m. Gliders are capable of long-term operation and havebeen used extensively for oceanographic research for a numberof years.

In the last years several research groups in the United Statesand Canada have started using gliders to investigate cetaceans[5], especially the deep-diving species in the beaked whalefamily (Ziphiidae; [6]). Research on beaked whales cameto the fore because little is known about these animals andbecause of atypical stranding events which are suspected tobe related to military sonar activities [7]–[9].

Because of their long dives (up to 1.5 h) and brief surfacingperiods these animals are difficult to detect visually. Beakedwhales vocalize extensively underwater to navigate and detectprey [10]. PAM is therefore the preferred method to deter-mine presence/absence of beaked whales. However, as beakedwhales appear to start echolocating at depths greater than400 m, and because their emission beam pattern is narrow [11],the detection probability increases with depth [12] and soundreaches the surface only occasionally. Accordingly gliders arebetter suited for investigation of these animals than surfacevessels.

Gliders are also used to investigate baleen whales [5].Because of the glider’s low speed (0.25-0.5 m/s, or 0.5-1 kt), flow noise is relatively low, which is advantageous forrecording low-frequency baleen whale vocalizations. However,the internal electronics and mechanics of gliders periodicallyproduce self-noise, and during such periods passive acousticobservations are not possible. An advantage of submergedoperated vehicles is the limited surface time, which minimizesthe risk of a collision with other obstacles, reduces damagefrom high-energy surface phenomena (wind and waves), andreduces the possibility of potentially harmful human action.Furthermore gliders can be deployed in polar regions, whereice coverage prohibits the usage of surface vehicles, and inareas with high wind and waves where the traditional visualmeans of marine mammal observation are ineffective.

III. LIMITATIONS OF PASSIVE ACOUSTIC GLIDERS

Submerged operated platforms such as gliders also sufferfrom some drawbacks:

• Speed: The typical horizontal cruise speed of mostgliders is approximately 0.25 m/s (0.5 kt). This lowspeed does not allow surveying a large area for a targetspecies in a reasonably short time period. To be able toconduct a survey in a shorter amount of time, a largernumber of gliders (number depending on the size of thearea of interest) must be deployed. A larger number ofdevices significantly increases the complexity and cost ofa survey.

• Payload: Most gliders are relatively small instrumentsand provide relatively limited payload capacity. Largerpayloads allow for more batteries and sensors, so thesmall capacity of gliders limits both their deploymentduration and their capability for measuring a wider suiteof oceanographic parameters. An additional constraint ingliders is that the payload must be horizontally balanced.

• Continuous real-time access: As gliders stay submergedmost of time, these platforms do not provide continuousreal-time access. For real-time monitoring, such to warnof the presence of an endangered species, the minimumresponse time of a glider is the time it takes to rise to thesurface - potentially several hours - plus a small amountof data transmission time.

• Sensors: The operating power for gliders comes frombatteries. Because of constraints in payload mass, theamount of energy available for operating power-intensiveelectronics such as optical sensors is small.

• Computational power: Because of the energetic limita-tions, sophisticated and thus energy-intensive computa-tions cannot be run continuously onboard a glider.

• Reliability: A malfunction at depth can cause the loss ofa glider.

• Duration: Because of the limited energy capacity, acous-tic glider deployments for marine mammal studies arelimited to a duration of several weeks.

IV. AUTONOMOUS SAILING VESSELS

An autonomous sailing vessel (ASV) is a sailboat equippedwith sensors for wind speed and direction and motor-drivenactuators for controlling sails, rudder, trim, etc. Using itsintelligent control system [13]–[16], it can automatically steerthe vessel to a desired point, maintain station at a locationwhen desired, or follow any other long-term directions a shore-based pilot provides it. Autonomous sailboats are aimed to beused for several tasks on sea, especially for ocean samplingand observation [17]–[21].

The Roboat (see Fig. 2) is a type of ASV in developmentand use since 2007 [14], [15], [22], [23]. The basis for theRoboat is a commercial sailboat designed by Jan HermanLinge, the boat type Laerling. The boat was originally createdfor kids to learn sailing, and therefore safety and stability arethe major characteristics of the boat. It has a length of 3.75 m

IRSC'09 - page 44

Fig. 2. The Roboat autonomous sailing vessel (ASV)

and comprises a 60 kg keel-ballast, which will bring the boatupright even from the most severe heeling. The boat can carrylarge payloads such as a battery bank and multiple sensors.Including batteries the overall weight of the boat is 300 kg.Additional payload of up to 50 kg is possible without impacton the sailing behaviour. The sail area of mainsail and foresailtogether is 4.5 m2. It is equipped with solar panels providingup to 285 W of power during conditions of full sun and a directmethanol fuel cell delivering 65 W as a backup energy source.The Roboat features a three-stage communication system,combining WLAN, UMTS/GPRS and an IRIDIUM satellitecommunication system, allowing continuous real-time accessfrom shore [23]. This can be used, for example, to trackand navigate the ship, or to transmit information on acousticdetections, to a shore-based command center. The rudder andsails as well as the tacks and jibes are autonomously con-trolled by incoming data from various sensors (GPS, compass,anemometer, etc.) on an NMEA200-bus, which are analysedon an onboard PC running Linux. It has been successfullytested on Austrian Lakes, the Adriatic Sea in Croatia, and theIrish Sea in Wales. The Roboat is virtually unsinkable, so the

danger of losing the device is small, and any detected systemmalfunctions can be immediately reported to the commandcenter.

V. THE AAS ENDURANCE

The AAS Endurance will be a specially-equipped Roboat.Unique features of the AAS Endurance include the following.

A. Acoustic System

An acoustic streamer (towed array) will contain threehydrophones, a depth sensor, and a compass module fordetermining the orientation of the streamer. The capturedsound will be sent to a BARIX Instreamer, which will digitizethe analog signals with sampling rates up to 48 kHz. Datawill be streamed continuously via the boat’s WLAN interfaceto a base onshore, or to a manned vessel if within reach.This arrangement was successfully implemented and is beingused in an autonomous listening station in Antarctica [24].In parallel, the analog hydrophone signals will be sent to anonboard high-quality recording system with sampling rates upto 192 kHz and resolution of 24 b running on a low-power PC.Signals will also be sent to automated call-detection softwarerunning on DMON hardware developed by Mark Johnson ofWoods Hole Oceanographic Institution. Such software willlisten for calls of target species of marine mammals; suchalgorithms have been developed for many species of cetaceans(whales, dolphins, porpoises) and pinnipeds (seals, sea lions,walrus) (e.g., [25]–[28]). Most cetaceans and pinnipeds arereliably detectable from the surface, and data recorded fromsurface vessel towed arrays make clear that even beakedwhales can be detected [29], although the detection probabilityis lower.

This acoustic data-capture and processing system will allowonboard real-time detection of marine mammal calls and stor-age of high-quality data for further laboratory analysis. If thesailboat’s WLAN is within reach of shore, acoustic data canbe streamed to the command center in real time. In addition,the spatially separated hydrophones provide information forestimating the direction to any sound sources encounteredusing time-of-arrival delay methods [30].

B. Optical System

An optical camera mounted at top of the mast can beaimed in any desired direction. The acoustic system willuse its multiple hydrophones to estimate the bearing to amarine mammal sound source and provide this bearing to theoptical system. The optical system can then be aimed in thedesired direction to potentially allow visual identification ofany vocalizing marine mammals when they surface.

C. Energy System

To produce energy independently of weather conditions, amethanol fuel cell is integrated as a backup system, allowingcontinuous provision of 65 W over a period of four weeks.The advanced energy system allows the Roboat to run sophis-ticated algorithms, such as for detection and classification of

IRSC'09 - page 45

marine mammal calls, continuously over extended periods oftime. This energy system is not available on other types ofautonomous acoustic platforms.

D. Speed

AAS Endurance will have a maximum speed of approx-imately 2.3 m/s (4.5 kt). This allows sampling an area ofinterest with high temporal and spatial resolution at low cost.

VI. CHALLENGES

A. Obstacle Detection and Avoidance

An important problem to be solved for long-term unmannedand autonomous missions on sea is reliable obstacle detectionand avoidance. Static obstacles such as landmasses can bepredefined on the sea map which is the basis for the routingsystem. A combination of multiple techniques, such as thermalimaging, radar, camera, and automatic identification system(AIS) will be used to detect dynamic obstacles. Research inthis field has been carried out for autonomous underwatervehicles [31] and motorised autonomous surface vehicles [32]–[35]. The obstacle avoidance task is different for sailingvessels, as they can not navigate in any direction directly,depending on wind conditions. Therefore a novel approachto autonomous obstacle avoidance will be an essential part ofthis research project.

B. Energy Balance

The currently used ASV Roboat can operate energeticallyautonomously with an average power consumption of 30 W.The solar system generates enough energy to sail continously,but doesn’t provide any additional energy for the acousticmonitoring facilites. In order to compensate this lack ofenergy, there are basically two possible approaches: generatingmore power or increasing efficiency. The first approach withinthe research project will be to save power by the use ofmore efficient components (computer, sensors, drives) and byoptimising the control algorithms. Furthermore, a balanced rigdesign (also known as Balestron rig, AerorigTM , swing rig,and EasyRigTM ) provides great potential to save power [36],[37]. A balanced rig consists of an unstayed mast carrying amain and jib (see Fig. 3). The main boom extends forward ofthe mast (the mast passes through the boom) to the tack ofthe jib. The main and jib are sized so that the force from themainsail is slightly higher than that from the jib. That is, thecombined center of effort is just behind the mast. Thereforethe force needed to control the sheets is much lower than fora conventional sloop rig. The new rig will be equipped withmotors for autmatic reefing in order to avoid damage duringstorms.

VII. PROJECT TIMELINE

To date (April 2009) the planning phase of the project iscompleted and funding has been requested. We plan to buildand test AAS Endurance over the next three years.

In the first year of development the sailboat will be equippedwith the control and energy system in Vienna, Austria. A

Fig. 3. Balanced rig example (source: [37])

first system test will be conducted on Lake Neusiedl, Austria.In a second step the acoustic system will be integrated. Amore comprehensive test will be performed on the coast ofthe Baltic Sea in northern Germany. Goals of this test are(1) to verify that the control (including obstacle avoidance)and energy systems are working properly, (2) to evaluate theimpact of the acoustic streamer on vessel speed and behavior,(3) to test mechanisms to optimize the depth and alignmentof the acoustic streamer, and (4) to test the optical system forthe potential verification of recorded sounds. A final tuningbased on the result of the Baltic Sea test will be conducted inVienna, Austria.

In the second year, AAS Endurance will undergo its firstdeep-water tests over 3-5 days off the coast of Newport,Oregon, USA. The goals of this test are optimization of theacoustic systems, especially noise reduction; assessment ofvessel self-noise in various sea states; and testing of marinemammal detection capability. Some acoustic data will betransmitted in real time to shore, allowing analysis of acousticsystem performance and wave and flow noise levels in variousmodes of sailing. Real-time marine mammal call detectionalgorithms will be implemented in the on-board acousticsystem, allowing sending of encounter information nearlyinstantaneously via IRIDIUM communication link while ontransect.

After successful completion of these tests, AAS Endurancewill be transported to Hawaii, USA. After a final test offKailua, Hawaii, USA, AAS Endurance will be sent on atransect from Kailua, Hawaii, USA to Newport, Oregon, USA,a direct distance of approximately 4100 km. The estimated

IRSC'09 - page 46

transect time is approximately 4 weeks. A comprehensive dataanalysis to characterize the system’s performance at detectingmarine mammal vocalizations will be conducted afterwards inthe lab.

After the two-year development period, AAS Endurance willreach operational capability. A first scientific survey of marinemammals will be conducted in the third year.

VIII. DISCUSSION AND FUTURE WORK

The autonomous acoustic sailboat offers major advantagescompared to submerged operated vehicles, including payload,speed, continuous real-time access, energy, and onboard com-putational power. However there are also challenges suchas reliable obstacle avoidance linked to this new technologywhich must be addressed.

Gliders remain an important and powerful platform toinvestigate deep diving animals such as beaked whales orsurveying polar regions where ice coverage prohibits the usageof surface vehicles. Both platforms are useful tools to gainknowledge of marine ecosystems, especially - as here proposed- of marine mammals.

AAS Endurance offers the operation of a multi-sensor plat-form and is therefore suitable to investigate broader ecologicalquestions. The autonomous acoustic sailboat could, for ex-ample, be navigated to follow tagged animals using positioninformation transmitted by the tag. Such a mission wouldhelp gain information on species-specific seasonal and diurnalvocalization in behavior. This baseline information is veryimportant for projects utilizing passive acoustic recordings toestimate the distribution and abundance of marine mammals.Additional sensors for oceanographic variables such as chloro-phyll and zooplankton density could help to understand theecology of many marine mammal species.

REFERENCES

[1] D. K. Mellinger, K. M. Stafford, S. E. Moore, R. P. Dziak, andH. Matsumoto, “An overview of fixed passive acoustic observationmethods for cetaceans,” Oceanography, vol. 20, no. 4, pp. 36–45, 2007.

[2] (2009). [Online]. Available: accessed on 9 April 2009,http://www.liquidr.com

[3] C. C. Eriksen, T. J. Osse, R. D. Light, T. Wen, T. W. Lehman,P. L. Sabin, J. W. Ballard, and A. M. Chiodi, “Seaglider: A long-range autonomous underwater vehicle for oceanographic research,” IEEEJournal of Oceanic Engineering, vol. 26, no. 4, pp. 424–436, 2001.

[4] (2009). [Online]. Available: accessed on 9 April 2009,http://www.seaglider.washington.edu

[5] M. F. Baumgartner and D. M. Fratantoni, “Diel periodicity in both seiwhale vocalization rates and the vertical migration of their copepod preyobserved from ocean gliders,” Limnol. Oceanogr., vol. 53, no. 5, part 2,pp. 2197–2209, 2008.

[6] J. A. Theriault, D. Mosher, J. Hood, D. Flogeras, and T. Murphy,“Detection of beaked whales using autonomous underwater vehicle(glider),” Presentation with abstract, 3rd Intl. Workshop on Detectionand Classification of Marine Mammals using Passive Acoustics, Boston,p. 24, 2007.

[7] A. Fernandez, J. . F. Edwards, F. Rodriguez, A. E. de los Monteros,P. Herraez, P. Castro, J. R. Jaber, V. Martin, and M. Arebelo, “Gasand fat embolic syndrome involving a mass stranding of beaked whales(family ziphiidae) exposed to anthropogenic sonar signals,” VeterinaryPathology, vol. 42, pp. 446–457, 2005.

[8] T. M. Cox, T. J. Ragen, A. J. R. amd E. Vos, R. W. Baird, K. Balcomb,J. Barlow, J. Caldwell, T. Cranford, L. Crum, A. D’Amico, G. D’Spain,A. Fernandez, J. Finneran, R. Gentry, W. Gerth, F. Gulland, J. Hilde-brand, D. Houser, T. Hullar, P. D. Jepson, D. Ketten, C. D. Macleod,P. Miller, S. Moore, D. C. Mountain, D. Palka, P. Ponganis, S. Rommel,T. Rowles, B. Taylor, P. Tyack, D. Wartzok, R. Gisiner, J. Mead, andL. Benner., “Understanding the impacts of anthropogenic sound onbeaked whales,” J. Cetacean Res. Manage, vol. 7, pp. 177–187, 2006.

[9] J. A. Hildebrand, “Impacts of anthropogenic sound,” in Marine MammalResearch: Conservation beyond Crisis, J. E. R. III, W. F. Perrin, R. R.Reeves, S. Montgomery, and T. Ragen, Eds. Baltimore: Johns HopkinsUniv. Press, 2005, pp. 101–124.

[10] W. M. X. Zimmer, M. P. Johnson, P. T. Madsen, and P. L. Tyack,“Echolocation clicks of free-ranging cuvier’s beaked whales (ziphiuscavirostris),” Journal of the Acoustical Society of America, vol. 117, pp.3919–3927, 2005.

[11] J. Ward, R. Morrissey, D. Moretti, N. DiMarzio, S. Jarvis, M. Johnson,P. Tyack, and C. White, “Passive acoustic detection and localization ofmesoplodon densirostris (blainville’s beaked whale) vocalizations usingdistributed bottom-mounted hydrophones in conjunction with a digitaltag (dtag) recording,” Can. Acoust., vol. 36, pp. 60–66, 2008.

[12] W. M. X. Zimmer, J. Harwood, P. L. Tyack, M. P. Johnson, and P. T.Madsen, “Passive acoustic detection of deep-diving beaked whales,”Journal of the Acoustical Society of America, vol. 124, pp. 2823–2832,2008.

[13] J. C. Alves, T. M. Ramos, and N. A. Cruz, “A reconfigurable com-puting system for an autonomous sailboat,” in International RoboticSailing Conference (IRSC). Breitenbrunn, Austria: Austrian Societyfor Innovative Computer Science, May 2008, pp. 13–20.

[14] R. Stelzer, T. Proell, and R. I. John, “Fuzzy logic control system forautonomous sailboats,” in FUZZ-IEEE, London, UK, July 2007, pp. 97–102.

[15] R. Stelzer and T. Proell, “Autonomous sailboat navigation for shortcourse racing,” Robotics and Auotnomous System, vol. 56, no. 7, pp.604–614, July 2008.

[16] M. L. V. Aartrijk, C. P. Taglioloa, and P. W. Adriaans, “Ai on the ocean:the robosail project,” in European Conference on Artificial Intelligence,2002, pp. 653–657).

[17] C. Sauze and M. Neal, “Design considerations for sailing robots per-forming long term autonomous oceanography,” in International RoboticSailing Conference (IRSC). Breitenbrunn, Austria: Austrian Society forInnovative Computer Science, May 2008, pp. 21–29.

[18] N. A. Cruz and J. C. Alves, “Ocean sampling and surveillance usingautonomous sailboats,” in International Robotic Sailing Conference(IRSC). Breitenbrunn, Austria: Austrian Society for Innovative Com-puter Science, May 2008, pp. 30–36.

[19] ——, “Autonomous sailboats: an emerging technology for ocean sam-pling and surveillance,” in MTS-IEEE Conference Oceans’2008, Quebec,Canada, September 2008.

[20] C. Sauze and M. Neal, “An autonomous sailing robot for ocean obser-vation,” in Towards Autonomous Robotic Systems (TAROS), Surrey, UK,2006.

[21] M. Neal, “A hardware proof of concept of a sailing robot for oceanobservation,” IEEE Journal of Ocean Engineering, vol. 31, pp. 462–469, 2006.

[22] R. Stelzer and K. Jafarmadar, “A layered system architecture to controlan autonomous sailboat,” in Towards Autonomous Robotic Systems(TAROS 2007), Aberystwyth, UK, September, September 2007, pp. 153–159.

[23] ——, “Communication architecture for autonomous sailboats,” submit-ted to International Robotic Sailing Conference (IRSC), Porto, Portugal,July 2009.

[24] H. Klinck, “Automated passive acoustic detection, localization andidentification of leopard seals: from hydro-acoustic technology to leop-ard seal ecology,” Reports on Polar and Marine Research, vol. 582,145 pp., 2008.

[25] D. K. Mellinger and C. W. Clark, “Recognizing transient low-frequencywhale sounds by spectrogram correlation,” J. Acoust. Soc. Am., vol. 107,pp. 3518–3529, 2000.

[26] D. Gillespie, “Detection and classification of right whale calls usingan edge detector operating on a smoothed spectrogram,” Can. Acoust.,vol. 32, pp. 39–47, 2004.

[27] D. K. Mellinger, K. M. Stafford, and C. G. Fox, “Seasonal occurrence

IRSC'09 - page 47

of sperm whale (Physeter macrocephalus) sounds in the gulf of alaska,1999-2001,” Mar. Mamm. Sci., vol. 20, no. 1, pp. 48–62, 2004.

[28] M. A. Roch, M. S. Soldevilla, R. Hoenigman, S. M. Wiggins, andJ. A. Hildebrand, “Comparison of machine learning techniques for theclassification of echolocation clicks from three species of odontocetes,”Can. Acoust., vol. 36, pp. 41–47, 2008.

[29] G. Pavan, C. Fossati, M. Priano, and M. Manghi, “Recording cuvier’sbeaked whales (Ziphius cavirostris) with a wideband towed array,”Document SC/58/E18 submitted to the 58th International WhalingCommission Scientific Committee, 2008.

[30] W. A. Watkins and W. E. Schevill, “Spatial distribution of Physetercatodon (sperm whales) underwater,” Deep-Sea Res., vol. 24, pp. 693–699, 1977.

[31] S. Showalter, “The legal status of autonomous underwater vehicles,”Mar. Techn. Soc. J., vol. 38, no. 1, pp. 80–83, 2004.

[32] J. Larson, M. Bruch, R. Halterman, Rogers, and J. R. Webster, “Ad-vances in autonomous obstacle avoidance for unmanned surface vehi-cles,” in Proceedings of AUVSI Unmanned Systems North America 2007,vol. 582, Washington, DC, USA, 2007, pp. 154–168.

[33] M. R. Benjamin, J. J. Leonard, J. A. Curcio, and P. M. Newman,“A method for protocol-based collision avoidance between autonomousmarine surface craft,” J. Field Robot., vol. 23, no. 5, pp. 333–346, 2006.

[34] T. Statheros, G. Howells, and K. McDonald-Maier, “Autonomous shipcollision avoidance navigation concepts, technologies and techniques,”J. Navigation, vol. 61, no. 1, pp. 129–142, 2008.

[35] R. Smierzchalski, “Evolutionary-fuzzy system of safe ship steering ina collision situation at sea,” in International Conference on IntelligentAgents, Web Technologies and Internet Commerce, vol. 1, 2005, pp.893–898.

[36] (2009). [Online]. Available: accessed on 16 April 2009,http://www.multirig.com/the balestron rig.htm

[37] (2009). [Online]. Available: accessed on 16 April 2009,http://balancedrig.com/description.html

IRSC'09 - page 48