الرئيسية Robotic Sailing 2014: Proceedings of the 7th International Robotic Sailing Conference

Robotic Sailing 2014: Proceedings of the 7th International Robotic Sailing Conference


An autonomous sailboat robot is a boat that only uses the wind on its sail as the propelling force, without remote control or human assistance to achieve its mission. Robotic sailing offers the potential of long range and long term autonomous wind propelled, solar or wave-powered carbon neutral devices. Robotic sailing devices could contribute to monitoring of environmental, ecological, meteorological, hydrographic and oceanographic data. These devices can also be used in traffic monitoring, border surveillance, security, assistance and rescue. The dependency on changing winds and sea conditions presents a considerable challenge for short and long term route and stability planning, collision avoidance and boat control. Building a robust and seaworthy sailing robot presents a truly complex and multi-disciplinary challenge for boat designers, naval architects, systems/electrical engineers and computer scientists.

Over the last decade, several events such as Sailbot, World Robotic Sailing Championship and the International Robotic Sailing Conference (WRSC/IRSC) and Microtransat have sparked an explosion in the number of groups working on autonomous sailing robots. Many of the challenges in building truly autonomous sailing robots still remain unsolved. These proceedings present the work of researchers on current and future challenges in autonomous sailboat development, presented at the WRSC/IRSC 2014 in Galway, Ireland, 8th – 12th September 2014.

السنة: 2015
الطبعة: 1
الناشر: Springer International Publishing
اللغة: english
الصفحات: 95 / 98
ISBN 13: 978-3-319-10076-0
File: PDF, 10.22 MB
تحميل (pdf, 10.22 MB)
اقرأ الكتاب مباشرة
You can write a book review and share your experiences. Other readers will always be interested in your opinion of the books you've read. Whether you've loved the book or not, if you give your honest and detailed thoughts then people will find new books that are right for them.
Fearghal Morgan
Dermot Tynan

Sailing 2014
of the 7th International
Robotic Sailing


Robotic Sailing 2014

Fearghal Morgan · Dermot Tynan

Robotic Sailing 2014
Proceedings of the 7th International
Robotic Sailing Conference


Fearghal Morgan
Bio-Inspired Electronics & Reconfigurable
Computing Research Group
Electrical & Electronic Engineering
National University of Ireland

ISBN 978-3-319-10075-3
DOI 10.1007/978-3-319-10076-0

Dermot Tynan
Hewlett Packard HP Cloud

ISBN 978-3-319-10076-0


Library of Congress Control Number: 2014945762
Springer Cham Heidelberg New York Dordrecht London
c Springer International Publishing Switzerland 2015

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’s location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for any
errors or omissions that may be made. The publisher makes no warranty, express or implied, with respect
to the material contained herein.

Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)


These proceedings contain the papers presented at the IRSC 2014 (International
Robotic Sailing Conference) that has taken place in the National University of Ireland, Galway, in conjunctions with the WRSC (World Robotic Sailing Championship) from the 8th until the 12th of September 2014. This is the 7th edition in
a series of IRSC proceedings.
Robotic sailing offers the potential of long range and long term autonomous wind
propelled, solar or wave-powered carbon neutral devices. Robotic sailing devices
could contribute to monitoring of environmental, ecological, meteorological, hydrographic and oceanographic data. These devices can also be used in traffic monitoring, security, assistance and rescue.
The dependency on changing winds and sea conditions presents a considerable
challenge for short and long term route and stability planning, collision avoidance
and boat control. Building a robust and seaworthy sailing robot presents a truly complex and multi-disciplinary challenge for boat designers, naval architects, electronic
and embedded systems engineers and computer scientists. Since 2004, events such
as Sailbot, Microtransat, World Robotic Sailing Championship and the International
Robotic Sailing Conference have sparked an explosion in the number of groups
working on autonomous sailing robots. Despite this interest the longest distance
sailed autonomously remains only a few hundred miles. Many of the challenges in
building truly autonomous sailing robots still remain unsolved.
The International Robotic Sailing Conference (IRSC) provides researchers with
the opportunity to present and exchange ideas on their work on a wide range of
topics related to autonomous surface marine robotics (especially sailing robots).
WRSC2014 (organised in conjunction with IRSC2014) includes a series of short
distance racing, navigation and autonomy challenges. The competition proposes
tasks such as station, speed in different conditions, accuracy, obstacle avoidance,
target tracking, endurance and cooperation. The competition, originally designed
for sailboats also includes a motorboats category, in order to bring together the scientific communities that work on different types of autonomous marine vehicles.
Previous IRSC/WRSC events have been hosted in France (2013), Wales (2012),
Germany (2011), Canada (2010), Portugal (09) and Austria (08).



The proceedings is divided into three sections as follows:
Part I: Sailboat Platforms and Applications
• Design, construction and test sailing of a high-performance SailBot Sea Quester
(and a lower cost variant) built using basic tools and inexpensive materials.
• The design and development of a versatile, small, low cost and efficient sailing
robot platform (MaxiMOOP).
• The use of a sailboat to tow large objects, with simulated demonstration and
practical suggestions.
Part II: Power Management and Mission Planning
• Power management and energy saving strategy for a robotic sailboat.
• An interactive tool (METASail) for assisting the planning, supervision and analysis of missions performed by the autonomous sailboat FASt.
Part III: Controllers and Sensors
• The concept of active course markers relaying environmental conditions to support autonomous sailboats.
• An open source, low cost, Arduino-compatible sailboat controller, incorporating
a small and agile real time operating system.
• A piezoelectric sail trim/luffing sensor
The Editors thank all of the authors, the Programme Committee members, all
sponsors of IRSC2014 and WRSC2014, and all who have made WRSC2014/
IRSC2014 possible in Galway, Ireland.
July 2014

Fearghal Morgan
Dermot Tynan


IRSC2014/WRSC2014 General Chair
Dermot Tynan
Fearghal Morgan

Hewlett Packard, Galway
National University of Ireland, Galway

Fearghal Morgan
Dermot Tynan

National University of Ireland, Galway
Hewlett Packard, Galway

IRSC Co-Organisers
Sean Roberts
Patricia Walsh
Mark Gantly
Thomas Ditzinger

IRSC Expo Organiser
National University of Ireland, Galway Conference
Hewlett Packard, Galway
Applied Sciences and Engineering, Springer DE

WRSC Co-Organisers
David Vinnell
Brendan Smith
Oliver Leahy

WRSC2014 Principal Race Officer, Galway Bay
Sailing Club
Schools Outreach Co-Ordinator
Bottle Boat competition



IRSC2014 Programme Committee
José Carlos Alves
Taylor Barton
Brad E. Bishop
Ole Blaurock
Benoit Clement
Vincent Creuze

Nuno A. Cruz
Kjell Dahl
Maeve Duffy
Fabrice Le Bars
Oren Gal
Luc Jaulin
Edward Jones
Erik Maehle
Benedita Malheiro
Paul H. Miller
Fearghal Morgan
Mark J. Neal
Cedric Pradalier
Olivier Reynet
Kostia Roncin
Colin Sauze
Alexander Schlaefer
Michael Schukat
Roland Stelzer
Daniel Toal
Dermot Tynan
Diedrich Wolter

Universidade do Porto, Portugal
Massachusetts Institute of Technology (MIT),
US Naval Academy (USNA), USA
Lübeck University of Applied Sciences,
École nationale supérieure de techniques
avancées (ENSTA) Bretagne, France
Laboratoire d’Informatique, de Robotique et de
Microélectronique de Montpellier
(LIRMM), France
Universidade do Porto, Portugal
Åland University of Applied Sciences, Finland
National University of Ireland, Galway, Ireland
École nationale supérieure de techniques
avancées (ENSTA) Bretagne, France
University of Haifi, Israel
École nationale supérieure de techniques
avancées (ENSTA) Bretagne, France
National University of Ireland, Galway, Ireland
University of Lübeck, Germany
Instituto Superior de Engenharia do Porto
(ISEP – IPP), Portugal
US Naval Academy (USNA), USA
National University of Ireland, Galway, Ireland
Aberystwyth University, Wales, UK
Georgia Tech-Lorraine, Metz, France
École nationale supérieure de techniques
avancées (ENSTA) Bretagne, France
École nationale supérieure de techniques
avancées (ENSTA) Bretagne, France
Aberystwyth University, Wales, UK
Hamburg University of Technology, Germany
National University of Ireland, Galway, Ireland
Austrian Society for Innovative Computer
Sciences (Innoc), Austria
University of Limerick, Ireland
Hewlett Packard, Galway, Ireland
University of Bamberg, Germany


Sponsoring Organisations
Hewlett Packard Galway
National University of Ireland, Galway/Ollscoil an hÉireann, Gaillimh
Port of Galway/Calafort na Gaillimhe
Galway Bay Sailing Club
Intel Ireland



Part I: Sailboat Platforms and Applications
An Easy-To-Build, Low-Cost, High-Performance SailBot . . . . . . . . . . . . .
Paul Miller, Andrew Beeler, Beatrice Cayaban, Matthew Dalton,
Cassandra Fach, Christian Link, Joel MacArthur, Jericho Urmenita,
Robert Yerkes Medina
MaxiMOOP: A Multi-Role, Low Cost and Small Sailing Robot
Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Paul Miller, Matthew Hamlet, Colin Sauzé, Mark Neal, David Capper,
Daniel Clark, Ashley Iles, Louis Taylor
Towing with Sailboat Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Luc Jaulin, Fabrice Le Bars




Part II: Power Management and Mission Planning
Power Management Strategies for an Autonomous Robotic Sailboat . . . .
Kjell Dahl, Anton Bengsén, Matias Waller


METASail – A Tool for Planning, Supervision and Analysis of Robotic
Sailboat Missions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
José Carlos Alves, Nuno Alexander Cruz


Part III: Controllers and Sensors
Towards Active Course Marks for Autonomous Sailing Competitions . . .
Paulo Ferreira, Benedita Malheiro, Pedro Guedes, Manuel Silva


A Real-Time Sailboat Controller Based on ChibiOS . . . . . . . . . . . . . . . . .
Jorge Cabrera-Gámez, Angel Ramos de Miguel,
Antonio C. Domínguez-Brito, Jose D. Hernández-Sosa,
Jose Isern-González, Leonhard Adler


Piezoelectric Vibrational Sensor for Sail Luffing Detetection on
Robotic Sailboats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Halie Murray-Davis, David Barrett


Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .


Part I

Sailboat Platforms and Applications

An Easy-To-Build, Low-Cost,
High-Performance SailBot
Paul Miller, Andrew Beeler, Beatrice Cayaban, Matthew Dalton, Cassandra Fach,
Christian Link, Joel MacArthur, Jericho Urmenita, and Robert Yerkes Medina

Abstract. Boats built to the SailBot Class rules are high-performance, computercontrolled vessels and to win requires both good systems development and a good
boat. The loose rules allow for exotic design and construction in order to improve
vessel performance, though with increasing cost of materials. This paper describes
the naval architecture design, construction and test sailing of a high-performance
SailBot Sea Quester, built using basic tools and inexpensive materials, and costing
€1200. The paper proposes alternative construction methods with an estimated
built cost of approximately $600, though with the loss of some operating
performance. Sea Quester was launched and sailed under remote control on 25
April 2014 and testing is in progress.



The two-metre long SailBot class of autonomous, sail-powered, surface vessels
originated in the 2004 SailBot regatta held in Canada. The class rules are
relatively simple as they restrict just a few key dimensions. In the decade since,
interest and participation in the competition has increased. Teams are using
increasingly higher-cost materials and high-tech fabrication methods in order to
improve performance. A cost of over US$15,000 to build boat (excluding the
electronics) has been highlighted.
The goal of this project is to use hand-tools and low-cost materials to build a
SailBot with a performance potential equal to that of high-cost vessels. The
resulting boat, Sea Quester (SQ), Figure 1, has been designed, built and tested by
midshipmen at the United States Naval Academy (USNA) during the 2013-2014
academic year as part of the Autonomous Surface Vessel courses.

Paul Miller · Andrew Beeler · Beatrice Cayaban · Matthew Dalton · Cassandra Fach ·
Christian Link · Joel MacArthur · Jericho Urmenita · Robert Yerkes Medina
United States Naval Academy, Annapolis, Maryland, USA
© Springer International Publishing Switzerland 2015
F. Morgan and D. Tynan (eds.), Robotic Sailing 2014,
DOI: 10.1007/978-3-319-10076-0_1


P. Miller et al.


Fig. 1 Sea Quester on trial sail, May 2014

Forty easy-to-build vessel designs have been developed using a naval
architecture solid modeling program and analyzed with a Velocity Prediction
Program. The best design has demonstrated a performance potential greater than
any of the previous six USNA SailBot boats, with a predicted top speed excess of
6.6 knots in a 15 knots breeze. This paper describes the investigation of the
vessel’s performance characteristics. The dxf and igs files and additional
information on Sea Quester's design (set of hull planking patterns) and a detailed
description of the construction process can be obtained by contacting the lead
author. The 25 kg vessel has been built using 1.5 mm plywood skinned with
lightweight fiberglass; off-the-shelf (OTS) carbon rods for the mast, and OTS
pultruded carbon strips fashioned to form a keel. While the bulb has been
purchased, an alternative method using a room temperature bulb manufacturing
process is described. The boat, sailing in remote control mode, can be built for
US$1200 in material costs, of which $900 are the carbon keel and lead bulb costs.
A boat which uses a lower-cost alternative keel and bulb is also described,
offering somewhat reduced performance.


Design Criteria and Development

SailBot development has evolved at USNA as follows: From 2008-2010 and
2013-2014 five performance-driven boats designs have been built expressly for

An Easy-To-Build, Low-Cost, High-Performance SailBot


the SailBot competition. From 2012-2013 and also in 2014 students have designed
and built boats for the Microtransat Challenge. The 2010 boat, Gill the Boat
(GTB), is considered a fast and controllable boat and has served as the parent hull
for the 2013 boat, A Mid Ships. It has also served as the baseline for the 2014
SailBot described in this paper. [1, 2] provide further information on GTB and
highlight primary problems as 9% over design weight and a tendency for the bow
to bury bow when sailing downwind in strong winds.
The mission statement for the Sea Quester boat is "Design a high-performance
SailBot that is easy-to-build using hand tools and is inexpensive." The six students
each semester have been tasked with designing, building and testing the boat over
nine months. None of the students had previous boatbuilding experience and only
one had significant sailing experience. Due to time and access constraints the
students were limited to working only about five hours per week.
Early decisions included designing a vessel that could be built of plywood
using a male station-frame method, a composite keel, an OTS bulb and OTS
carbon tubes for the mast. The design task started with a review of previous
SailBot designs, a preliminary weight study and the development of a Work
Breakdown Structure (WBS) that was used to track progress and define
deliverables. Trade-off studies using a Velocity Prediction Program (VPP) [3]
have been selected to identify performance drivers and from these results each
student has designed a hull. Results for each design have been compared in a
virtual regatta. The fastest design has been selected for further refinement and
detailed design.


Weight Study

The basic requirement of any vessel design is that the vessel's displacement is
balanced by the buoyancy supplied by the underwater volume (hull, keel, bulb and
rudder). Earlier VPP studies have shown that a lighter weight vessel is generally
faster if it has the same stability as a heavier vessel. Therefore the goal is to make
lighter everything that is not contributing to stability. In SailBots the majority of
the stability is provided by the bulb, so all other aspects of the design have been
investigated to reduce weight. Each of the major components from previous
designs have been weighed and brainstorming design sessions have been held to
derive lighter alternatives.
Compared to previous USNA SailBots, five main items have been selected for
weight reduction; hull, deck, keel, rudder stock, and rudder blade. In past boats the
construction methods of those items has been chosen based on the availability of
in-house capabilities. An NC-router has been used to mill the hull and keel and
allow rapidly assembly of a hull through carving the 10 mm thick hull from a solid
block of foam and the keel from blocks of stainless steel or titanium. The resulting
weight reduction study indicates that a 24.5 kg displacement is possible for use as
the target displacement. The final displacement has been increased to 25.5 kg due
to a heavier than expected bulb. Table 1 shows the weight table.

P. Miller et al.

Table 1 Weight Table for Sea Quester

With a weight target established the VPP studies have been designed based on
previous studies and the students' desires to explore different topics.


Performance Trade-Off Studies

Seven VPP studies have been used to explore different design feature trade-offs.
As none are direct repeats of the more than a dozen studies performed previously
by USNA students, most do not investigate primary design features and the results
show secondary performance effects. Two of the studies provide significantly
useful information and one indicates a clear need for future work. The three
studies completed are:
Beam at waterline - increased beam increases stability and performance, but
also increases wetted surface friction and wavemaking drag.
Canoe Body Draft - a deeper canoe body decreases intersection drag between
the keel and hull at the expense of greater dynamic sinkage
Transom Width - increasing the transom width has similar trade-offs to the
beam studies but is also a factor in nose-diving with a wider transom
increasing the tendency.

An Easy-To-Build, Low-Cost, High-Performance SailBot


The VPP studies all started with a baseline design using the 24.5 kg
displacement and a V-hull shape. This hull shape was chosen as the angle of the V
(deadrise) could be varied from 0 (flat bottom) to as much as 60 degrees, allowing
for the analysis of many hull shapes. It is an easy-to-build shape with the
convenience of showing a clear centerline for aligning the keel and rudder. The
well-known and quite fast sharpies class of boats use this hull shape. The V-bow
reduces pitch acceleration but can lead to tracking issues and the V-aft reduces the
likelihood of planing. The hulls have been evaluated in wind speeds of 5, 10 and
15 knots. Figure 2 shows the beam study results for a range of 30.5 - 46 cm (1-1.5
ft.). The minimum beam has been selected due to the hatch opening required for
the systems when a maximum of 5 degree flare is allowed. The VPP results are
reflected in the time required by the vessel to sail a one nautical mile course
consisting of an equal amount of beating, reaching and running. Although the
vessel gains stability with increasing beam, the overall effect is slower times. This
is not unexpected as the large majority of a typical SailBot's stability comes from
the low center of gravity provided by the keel versus the form stability more
common in shallow draft vessels.

Fig. 2 Beam study results showing increasing beam increases the time to sail a one mile

The intersection between the keel and canoe body can be a significant source of
drag with the added problem that if the vessel heels enough, the keel root may
come out of the water and the keel may ventilate. This is particularly likely when
the boat is travelling quickly when a wave trough forms near midships. The
PCSail VPP includes a factor for this affect. From the baseline hull the deadrise

P. Miller et al.


angle has been varied from a flat bottom to an angle of 60 degrees, resulting in the
canoe body draft ranging from 6.4 - 21.6 cm. (2.5 - 8.5 in.). In order to keep the
displacement the same the waterline has been shortened and the beam narrowed
slightly as the canoe body draft increased. Figure 3 shows that in light air the
moderate draft is fastest but as the wind and boat speed increase the lowest drag is
seen with shallowest canoe body draft. In a combined wind space the best
compromise has been a canoe body draft between 10 and 15 cm (4 - 6 in).

Time (sec/mi)

5 knot wind


10 knot wind


15 knot wind







Canoe Body Draft (in)

Fig. 3 Canoe body draft results show the compromise choice as a moderate draft

How the transom width affects performance is similar to beam; a wider transom
increases stability and wetted surface area. Additionally however, a wide transom
prevents squatting by increasing buoyancy aft and at high speed by producing
dynamic lift, which directly leads to the bow burying and a loss of directional
control. A double-ended (canoe-style) stern can also be built lighter and may stay
in level trim when heeled. These are both desirable characteristics. This study has
investigate varying the transom width from full-beam to double-ended. For the
study the displacement has been kept constant and the beam at the waterline has
been varied. Sinkage is calculated and the center of flotation, buoyancy and wetted
surface area adjusted as necessary. Initial results have shown that a 50% width
transom has a slight advantage overall, with a double-ender showing better
performance in light air and for a transom width up to 90% showing promise in
heavier winds. The differences are small however and the relationship with heel
cannot be tested in the VPP. Since this characteristic strongly influences the wake
and is impacted by transom immersion, the best way to test for this effect is in a
towing tank.

An Easy-To-Build, Low-Cost, High-Performance SailBot


With the background research completed the students were then tasked
with developing their own designs and the results were compared in a virtual
regatta that simulated light, medium and heavier air races using a triangle plus
windward/leeward course. The vessel that performed best had a moderate canoe
body draft, narrow beam, 90% transom width and 23 kg displacement. This
became the starting point for Sea Quester.


Sea Quester Design

Another week was spent running VPP analysis on the selected design. This
resulted in flare reduction, greater deadrise at the transom and an increase in bulb
weight. The final design was then compared to GTB. The new design is 3.6 kg
lighter and has a 1 kg heavier bulb. The prediction has indicated a 4% reduction in
the time to sail the one mile course. The SQ design presented here is a slightly
modified version of the boat that has been built. The small changes have been
made to ease constructability and no significant impact on performance should be
seen. A V-bottom has been maintained with a moderate canoe body draft. The
transom has been shaped so that at small heel angles consistent with light air she
acts as a double-ender, but when heeled and moving fast, the transom increases
beam to roughly 50% of maximum beam. Figures 4 -7 shows her perspective,
body plan, profile and waterline views, respectively. Her Principal Characteristics
are shown in Table 2.

Fig. 4 Sea Quester perspective view


Fig. 5 Sea Quester body plan

Fig. 6 Sea Quester profile lines

Fig. 7 Sea Quester waterlines
Table 2 Principal Characteristics of Sea Quester

P. Miller et al.

An Easy-To-Build, Low-Cost, High-Performance SailBot



Hull and Deck Construction

The longest lead time item has been the hull and deck. The basic construction
method uses a ladder-frame with station molds printed as templates from the body
plan. Nine stations have been set up with the forward station staying in the boat as
a collision bulkhead. The fourth station also stays in and serves as the mast
support bulkhead. The last station serves as the transom.
Low-cost construction materials have been selected whenever possible. As
plywood is the intended main building material for the hull and deck, an epoxy
resin system has been chosen as the adhesive. The skin, selected after constructing
six samples is 1.5 mm Okume plywood with 200 g/m2 fiberglass cloth on the
outside and 150 g/m2 cloth on the inside skin. Figure 8 shows the station molds on
the strongback.

Fig. 8 Frames mounted on the strongback

The edges of the frames not to be kept in the boat are then taped to prevent
adhesion and the topside plank templates has been made using the jig with heavy
construction paper. These have then been traced and the four hull planks cut. To
provide greater plank rigidity the inside skin has been laminated to the plywood
on a flat surface before the planks were installed. The topside planks are hung
first. As they do not have any twist they are quite easy to fit. The bottom planks
have a significant amount of twist, particularly aft, and require some effort to fit. It
helps to allow those planks to overhang while curing and then trim them back. The
hull has been faired and the outside skin applied and the hull removed from the
mold. Each interior joint has been filleted and laminated with a layer of 150 g/m2
fiberglass cloth. Figure 9 shows the hull with the outside skin laminated.
The deck is built in a similar fashion with seven 6 x 25 mm balsa deck beams
added in the way of the rudder servo, main sheet fairlead, winch, keelbox and jib
sheet fairlead and the two handles. The rudder shaft tube goes through the hull and
deck and is a 14 mm ID, 1.5 mm wall thickness braided carbon tube (also used
as the boom material). The keelbox is laminated over the keel using four plies of

P. Miller et al.


Fig. 9 Fiberglass application as the outside skin

200 g/m2 fiberglass cloth and also goes through the hull and deck. Both the
keelbox and the rudder shaft tube extend 8 mm above the deck and are flush
with the bottom of the hull. The mast step is a departure from previous boats in
that it is a tube (Laser sailboat style) that extends from the deck to the bottom of
the hull rather than a deck-stepped arrangement. The goal has been to improve
the mast stiffness and although the deck stepped arrangement gives more
flexibility in mast position the design team was confident of accurately placing the
step. The mast step tube is an 18 mm ID, 1.75 mm wall thickness braided tube.
The joined hull and deck without the hardware attached weighs 2.1 kg (shown
in Figure 10).

Fig. 10 Deck layout


Bulb, Keel and Rudder Construction

Until 2012 USNA has constructed its sailboat bulbs using a "cold" process to
avoid creating lead fumes. Two-part female foam molds are carved by NC and a
waxed layer of 200 g/m2 fiberglass cloth is placed in the mold and allowed to cure.

An Easy-To-Build, Low-Cost, High-Performance SailBot


After sanding, unfilled epoxy (or higher density cement) is poured into the mold.
Recycled lead shot is then poured into the mold, displacing the epoxy (or cement).
The bulb process includes weight monitoring. While a simple and easy approach,
the overall density is roughly equivalent to that of steel. Therefore the bulb has
unnecessary wetted surface area and induces drag. In 2012 a new bulb has been
designed using computational fluid dynamics, with outsourced mold and solid
lead bulb construction by MarsMetals in Canada.
Although the option was available to design a new bulb, the VPP indicates that
a bulb of similar weight to the 2012 bulb is acceptable and with high-temperature
molds still available the MarsMetals bulb cost is $627. This mold is available for
others who wish to prepare a bulb weighing approximately 11 kg.
In general the design driver for SailBot keels is deflection rather than strength.
The amount of surface area needs to be as small as possible so the primary design
driver is the material's stiffness. Prior USNA keels have used one of two
approaches. The first and easier approach features an OTS rectangular piece of 174ph stainless, covered with foam and hand faired to shape before application of a
final layer of fiberglass. The second approach features machining keels made of
stainless or titanium. Neither method has met the two criteria of saving weight and
enabling hand tools preparation of the sail boat.
The 92 mm chord length keel for SQ has been built using OTS ultra high
modulus pultruded carbon rods. After laminating a centerplane of 200 g/m2
fiberglass to use as a reference, thirteen 12 x 3 mm rods have been glued on each
side and then hand faired to a 15% SD8020 section shape before wrapping one
layer of 190 g/m2 carbon cloth around the circumference. The pultruded rods have
been purchased from The Composites Store at a cost of $256. Figure 11 shows the
keel and bulb. A single, horizontal M6 bolt through the keelbox above the deck
holds the keel in place.

Fig. 11 Sea Quester's keel and bulb

P. Miller et aal.


The 100 mm chord, 66
6 cm span rudder has been similarly built. On each side oof
the fiberglass centerplate a layer of 6 mm balsa is laminated and faired by hand tto
a NACA0012 section shaape. A 10 mm wide slit is then cut 50 cm long along thhe
15% chord length for th
he rudder shaft. The 9 mm OD x 2 mm wall thickness
carbon shaft is then glueed in place, faired, and one layer of 200 g/m2 wrappeed
around the circumferencee. A solid plug has been installed in the top 50 mm oof
the rudder shaft so that it would not be crushed by the tiller arm shaft clamp
Two Torlon bearing sleeeves have been slid over the shaft, making it ready foor
installation in the hull.


Sail and Rig Construction

The SailBot and WRSC competitions
require the boats to operate in varying winnd
conditions from nearly caalm to over 30 knots. Past USNA boats have used a singgle
mast. Sails have been sw
witched out when needed. A typical sail switch takes15520 minutes. During this time
the wind may have changed again. Additionally, iin
the strongest winds thee mast is significantly taller than needed, creatinng
unnecessary drag. Follow
wing the lead of UBC and model yachting, three rigs havve
been built for winds of 0--6, 6-13 and 14-30 knots. With the final design of SQ iin
the VPP the process has relatively
simple, designing the sails so that she does not
heel too much in the cond
ditions while arranging the planform so that each smaller
set of sails moves the center of effort 1% of the waterline length further forward tto
onal weather helm.
compensate for the additio
The sails have been designed using SailCad, a freeware program. Sail shapees
are generally similar, witth 13% max draft located 40-45% off the chord lengtth.
Paper templates have beeen first used to check the designs and then the sails havve
been cut out, taped and sewn together. The medium and heavy air sails are madde
from 225 g/m2 mylar scrrim while the light air sails are made of 90 g/m2 dynaac.
Figure 12 shows the saill plan. Note that the drawing does not show the maiin
roach, which is 80 mm on
n the light air sail, 50 mm on the medium air sail and 330
mm on the heavy air sail. Table 3 shows the sail areas.
Table 3 Sail areas for Sea Quester

An Easy-To-Build, Low-Cost, High-Performance SailBot


Fig. 12 Sailplan showing light, medium and heavy air sets of sails and masts

Fig. 13 Detail at the hounds, showing shroud and forestay yolk before the bolt was trimmed
off and showing the local fiberglass reinforcement around the bolt hole

The masts have been made from carbon tubes purchased from The Composites
Store at a cost of cost $110. All use 17mm OD tubes for the lower section and
tapered upper sections. Two sets of spreaders are used on each mast. The spreader
bars are 3 mm stainless rods that go through the mast and project 25 mm on each
side. The spreaders are 5 mm OD, 1 mm wall thickness stainless tubes with a hole

P. Miller et al.


drilled on the outboard end for the 1.2 mm diameter wire shrouds. The shrouds are
connected to the mast using a M5 bolt at the hounds with the forestay connected
via a wire yolk. Figure 13 shows the hounds before the M5 was cut to length.



Sea Quester (Figure 1) was launched and sailed under remote control on 25 April
2014 and in the best of naval traditions was behind schedule (two weeks) and over
budget ($200 above the target price of $1000). Using the alternative construction
methods presented in the paper, it is estimated that could be built at a cost of
approximately $600, though with the loss of some operating performance. The
initial Work Breakdown Structure has been quite accurate, underestimating the
160 man-hour workload for construction by just 5%. She has been built using only
hand tools, and by students with no prior related experience. Whether her
performance is enough to win has yet to be verified.

1. Miller, P., Hamlet, M., Rossman, J.: Continuous improvements to USNA sailbots for
inshore racing and offshore voyaging. In: Sauze, C., Finnis, J. (eds.) Robotic Sailing
2012, vol. 121, pp. 49–60. Springer, Heidelberg (2013)
2. Miller, P., Beal, B., Capron, C., Gawboy, R., Mallory, P., Ness, C., Petrosik, R., Pryne,
C., Murphy, T., Spears, H., Hamlet, M.: Increasing Performance and Added Capabilities
of USNA Sail-Powered Autonomous Surface Vessels (ASV). In: 3rd International
Robotic Sailing Conference, Kingston, Canada, June 7 (2010)
3. Martin, D.E., Beck, R.F.: Pcsail, a velocity prediction program for a home computer. In:
Chesapeake Sailing Yacht Symposium (2001)

MaxiMOOP: A Multi-Role, Low Cost and Small
Sailing Robot Platform
Paul Miller, Matthew Hamlet, Colin Sauzé, Mark Neal, David Capper,
Daniel Clark, Ashley Iles, and Louis Taylor

Abstract. This paper describes the development, testing and operational results
from a small, autonomous sailing vessel that was designed to be easily launched
and retrieved by one person while carrying a 7.5 kg payload and with enough speed
under sail to overcome reasonable current. The hull is 1.2 metres long and fits in the
boot of a typical car. This paper focuses on the design and testing of four prototypes,
two designed for short course racing and two others designed for long endurance all
weather missions. Initial tests have shown top speeds of around 3 knots with a larger
racing rig and 2.4 knots with a small all weather rig. One of the prototypes has attempted a transatlantic crossing. This was cut short when it was accidentally caught
by a fishing boat. Two different autonomous control systems have been developed,
one based around a pair of microcontrollers and intended for low power operation
averaging less than 1 W and the other based around a Raspberry Pi and ATMega328
combination to ease development and test more complex sailing algorithms.

1 Introduction
The MaxiMOOP design was developed in response to a need for a small, inexpensive, easy to build and transport, special-purpose autonomous surface vessel (ASV)
for use in oceanographic research and autonomous systems development. It was
inspired by the original MOOP (Miniature Ocean Observation Platform) [10], a
74 cm long hull design that demonstrated the feasibility of a small scale hull but
suffered from poor upwind performance. The MaxiMOOP is capable of multi-day
missions without the need for refuelling or recharging. To date, sail-powered ASVs
have mostly been adaptations of either small keelboats designed for one person or
Paul Miller · Matthew Hamlet
United States Naval Academy
Colin Sauzé · Mark Neal · David Capper · Daniel Clark · Ashley Iles · Louis Taylor
Department of Computer Science, Aberystwyth University, UK
e-mail: {cos,mjn,dmc2,dac46,ati1,lot15}@aber.ac.uk
c Springer International Publishing Switzerland 2015

F. Morgan and D. Tynan (eds.), Robotic Sailing 2014,
DOI: 10.1007/978-3-319-10076-0_2



P. Miller

Fig. 1 The MaxiMOOP ABoat Time beginning her transatlantic voyage in May 2014

modified remote control model yachts. The former, while able to carry large payloads, have the problem of their size causing logistical problems. The latter have
little payload capacity (typically less than two kilograms) and are not robust, but are
relatively quick. The MaxiMOOP was designed from scratch for the purpose of easy
logistics and reasonable payload. It can fit in the boot of a typical small car, costs
less than US$1000 in materials to construct (approximately $250 for materials and
$750 for electronics), can be carried and deployed by one person, has a payload of
7 kg and can maintain speeds of 1-2.5 knots in most wind conditions. It can serve as
a base platform for students developing their first ASV system or for carrying small
oceanographic research payloads. At least seven vessels have been built so far and
they have performed admirably in both Europe and North America
Note 1. references from sailbot. The hull shape has been developed to support a reasonable payload and exceptional sea-keeping ability while remaining durable and
easy to build. The hull is similar to many sail-powered inshore fishing craft of the
19th century with the addition of a proportionately much deeper keel that is integrated to the hull with slack garboards. The hull/keel joint was designed to ease
construction, reduce stress concentrations in the hull to keel joint and to provide
additional storage space. The leading edge of the keel is swept back to ease weed
shedding. The large lateral plane is needed in light air and waves while it provides
a keel sump with sufficient volume down low for ballast. The flat deck eases construction and the mounting of hardware. Table 1 lists the principal characteristics
and Figure 2 shows the hull lines of a MaxiMOOP.
A two-part (port and starboard) hull plug was machined from foam at the U.S.
Naval Academy (USNA) and molds were taken from the plug at Aberystwyth Uni-

MaxiMOOP: A Multi-Role, Low Cost and Small Sailing Robot Platform


Table 1 Principal characteristics of the MaxiMOOP hull
Length Overall
Waterline Length
Beam max overall
Depth overall
Sail Area

1.2 m
1.1 m
16-23/7 kg
0.41 m
0.35 m
0.6 m
9-10.5 kg
0.24-1.0 m2

Fig. 2 MaxiMOOP hull shape lines

versity and at USNA. Most hulls have been built from approximately 800g/m2 of
fibreglass cloth set in epoxy. Decks are typically 3 mm plywood covered on each
side by 200 g/m2 fibreglass cloth set in epoxy. Recycled lead shotgun shot is used for
ballast, this is poured in to uncured epoxy in two or three steps to reduce heat buildup. The top of the lead line is approximately 100 mm above the base of the keel,
providing a low centre of gravity and high stability. Due to the variety of missions
planned for the MaxiMOOPs, there are options of three different rudder configurations; attached to the keel, under hung spade and transom hung. The first boat,
Morwyn was built by Aberystwyth University for research in ASV systems in the
Irish Sea and has an attached rudder to reduce the likelihood of catching weed or
other floating debris. The second boat (Dewi) was built by Aberystwyth students
for competing in the SailBot competition and featured a transom hung rudder to
provide more manoeuvrability while being easily removable for transportation.
Figure 3 shows the two vessels.


P. Miller

Fig. 3 Morwyn (left) and Dewi (right) showing attached and transom-hung rudders

The sixth (Mid Life Crisis) and seventh (ABoat Time) vessels have been built
by the midshipmen at the United States Naval Academy and feature permanently
attached spade rudders for reduced drag. Figure 4 shows the spade rudder design
on MLC.
In addition to the ability to use different rudder designs, the MaxiMOOP has been
designed to accommodate different rig designs. Two rigs have proven to be the most
successful. The first is a relatively standard sloop with a 15/16 fractional foretriangle
as seen on MLC in Figure 4 and in the right drawing in Figure 5. The height of the
mast has been chosen so that the two-part mast, when disassembled, can be stored
on deck without overlapping the ends of the boat. The sloop rig is suitable when
using the boat at competitions, as it provides higher boat speeds and allows the boat
to point closer to the wind. The Aberystwyth University team used this rig to achieve
third place in the 2013 SailBot competition. Although this rig has been trouble-free,
the multiple parts inherent in a stayed sloop rig lead to lower reliability over a long
period of time. The second rig is a smaller, lower-aspect ratio fixed gaff rig shown
in figure 1 and the left drawing in Figure 5. The lower portion of the mast is offset
to reduce the sail’s yaw moment, which results in lower energy consumed to trim
the sail and reduced weather helm. The balanced rig was discussed in earlier IRSC
papers [8, 12].
To achieve higher rig reliability three features must be factored in; fewer moving
parts, fewer free edges and fewer point loads. To reduce the moving parts, a free-

MaxiMOOP: A Multi-Role, Low Cost and Small Sailing Robot Platform

Fig. 4 Mid Life Crisis showing spade rudder design and racing sloop rig

Fig. 5 Racing sloop rig (left) and free standing, rotating balanced gaff rig (right)



P. Miller

standing rig (without shrouds) is the best choice due to its few parts. A typical main
sail has three edges, the luff, which is attached to the mast, the foot, which is attached
to the boom, and the leech, which is a free edge. Most sail damage occurs along the
free edges due to the higher stress resulting from large movements during luffing.
Point loads on sails occur when the sails are attached via grommets or webbed
straps, rather than a boltrope or sleeve. The point loads can easily overload the local
material and cause failure.

2 Sail Tests
To determine the relative efficiency and reliability of different free-standing rigs for
the MaxiMOOP, a MaxiMOOP was built as a rig test platform. MLC (Figure 4) has
multiple mast step tubes to accommodate 14 different mast positions, corresponding to everything from a stayed sloop to very low aspect ratio freestanding sails.
Midshipman Padraig O’Brien compared results from the PCSail Velocity Prediction
Program (VPP) [7] to on-the-water results for four rigs. The VPP takes the boat’s
physical characteristics, such as length, beam, displacement and sail dimensions and
applies a wind velocity and relative direction as the input force and iteratively solves
for the boat speed, heel and leeway. Based on the earlier work on route planning for
the Microtransat [6], the rig tests focused on mostly sailing on a reach (close and
broad) due to the expected conditions in a transatlantic crossing from west to east
in winds of 8-30 knots. The four rigs all have 0.24 m2 of sail area and the height of
the boom above the deck was constant at 125 mm. The maximum draft is 13% and
the draft is located at 45%. The sail area selected has been chosen as a compromise
of minimal light air performance versus heavy air controllability. Figure 6 shows
the four rigs, which included an aspect ratio (AR) of three and six Marconi (three
sided) and gaff (four sided) rigs in the positions that give neutral helm balance. In
both rig types all the spars were joined together with fixed connections, increasing
reliability but potentially sacrificing some performance. Unlike the gaff rig shown in
Figure 6, these rigs are not self-balanced in order to reduce construction complexity
and time. The figure also shows the boom overlap with the solar panel, an important
consideration in solar charging.

Fig. 6 Four tested rigs: Marconi (AR=3), Marconi (AR=6), gaff (AR=3), gaff (AR=6)

MaxiMOOP: A Multi-Role, Low Cost and Small Sailing Robot Platform


Figure 7 shows the results from the VPP and multiple on the water tests normalized to the seconds required to sail one nautical mile with a wind speed of nine knots.
The course sailed in both prediction and on-the-water cases is a combined close and
broad reach. While the VPP favours high-aspect ratio sails over low-aspect ratio
sails due to their greater upwind efficiency, the on-the-water (OTW) tests demonstrate that the low-aspect ratio sails performed better. The reason for this is that
the very small size of the MaxiMOOP results in large amounts of rolling, causing
greater apparent wind shifts, which the low aspect-ratio sails are more forgiving of.
The VPP also favours Marconi over gaff rigs. This has not been seen on the water
possibly because the fixed gaff controls twist more effectively than the rotating gaff
normally used. The overall speeds seen on the water are also much higher than predicted by the VPP. This is not uncommon for VPPs, which are generally considered
more accurate for relative performance than absolute.

Fig. 7 VPP and on-the-water results for the four rigs

The final rig selected for the voyaging boat is the AR=3 gaff rig. It operates at
nearly the performance of the Marconi rig and has a shorter boom, which is desirable as it does not cover the aft solar panel as much. For the racing boats the final rig
selection is a 1.0 m2 Marconi rig with a low centre of gravity, consistent with participating in the SailBot competition, the VPP results of this are shown in Figure 8 and
show a top speed of three knots while running in 20 knots of breeze. In comparison,
with the smaller sail and higher centre of gravity and displacement consistent with
ocean voyaging, the performance of the voyaging gaff rig is noticeably less, as seen
in Figure 9. Convergence has not been reached with the VPP in winds of less than
11 knots for this case.

3 Energy Balance
A challenge with all vessels is power management at sea. In the case of sail-powered
ASVs, while the main propulsion system uses the wind, the computers, rudder and
sail winch need power. For small vessels such as the MaxiMOOP, the ability to be
self-sustaining offshore is a practical requirement. Initial work was carried out in


P. Miller

Fig. 8 Predicted MaxiMOOP performance versus wind angle in wind speeds of 6 to 20 knots
with the 1.0 m2 racing rig, lower centre of gravity and 16 kg displacement

Fig. 9 Predicted MaxiMOOP performance versus wind angle in wind speeds of 6 to 20 knots
with the 0.24 m2 voyaging sail and higher centre of gravity and 20.5 kg displacement

2012 [8], which showed that the solar panels mounted on the ASVs produced about
13-27% of their rated capacity. Two experiments were run by Midshipman Chris
Hein, testing whether the relative angle of the panel to the deck was significant and
whether a newer model of marine solar panel by Boulder produced higher output.
The first test varied the angle of the panel relative to the deck. As the course from
North America to Europe is a relatively constant course, with the wind mostly from
one direction, the question was raised whether some tilt angle other than flat was
optimum. For the mid-latitude of the course track and using a standard solar tilt
equation, the optimum tilt would be 24 degrees from horizontal, pointing south. As

MaxiMOOP: A Multi-Role, Low Cost and Small Sailing Robot Platform


the vessel was expected to heel an average of 15 degrees to port, the theoretical best
angle would be 39 degrees from the deck. Two experiments were conducted; the
first on shore and the second on MLC while sailing a relatively constant course. The
tilt angle varied from zero to sixty degrees. The total energy produced was measured
using a WattsUp watt-meter. The mean value was 63% of the panel’s rated output
with an 8% coefficient of variation. No discernible trend was seen however based on
the tilt angle. On the water tests also did not show a discernible trend in tilt angle,
but the amount produced was roughly 42% of the rated output, which was nearly
double the output of the panels tested in 2012. The doubled output of energy collection likely resulted from the combination of Boulder solar panel and the addition
of a Maximum PowerPoint Controller (MPPT) device put inline between the solar
panel and the battery. The MPPT better matches the battery draw conditions to the
solar panel given the current efficiency of the panel at any given time. The MPPT
used was purchased from Genasun LLC in Cambrige, Massachusetts, USA and was
specifically designed to work with our LiFePO4 4-cell battery.
After factoring size and weight of solar panels, it is clear lower voltages are
more efficient to produce energy with current solar technologies. In order to minimize power requirements, two independent power systems were installed on A Boat
Time. All electronic sensing and decision making devices were powered by a nominal voltage of 3.7 volts and a 6 volt rated solar charging system. In order to keep the
current consumption low, the actuator motors were run on a different power system
with a nominal voltage of 14.2 volts, which was paired with the charging system
Midshipman Hein tested.

4 Control System Design
A variety of control system architectures have been developed for sail-powered
ASVs, ranging from single low power microcontroller systems [5, 4] to FPGAs
[2] and single board computers [11, 10, 9, 3]. Each of these brings a different set
of trade-offs between power consumption, computing power, ease of use and reconfigurability. A single microcontroller system is most suited to longer term low
power operations but this comes at the expense of the ease of development and testing or the ability to execute more complex algorithms. In Dewi, which has been
built for short races lasting at most a few hours, a Raspberry Pi single board computer has been used together with an Arduino Uno microcontroller. The Raspberry
Pi is responsible for high level control decisions, logging and sending telemetry
data over WiFi or an XBee Pro radio modem. The Arduino is connected directly
to the servos, compass and wind sensor, with a series of simple commands issued
by the Raspberry Pi either requesting data or sending positions to these. This split
architecture sees the timing critical code such as reading the PWM wind sensor and
controlling the servos moved to the Arduino while the bulk of the code executes
on the Raspberry Pi using the Raspbian Linux distribution. The presence of a full
operating system greatly simplifies the development and testing of control system


P. Miller

code, makes performing “over the air” code updates easy, allows threading/locking
or concurrent processes and allows logfiles to be easily stored and accessed.
Similar to the control design and testing in Dewi, AT system development started
with a Raspberry Pi and Arduino combination. Uniquely, a combination of Arduino
boards were used for sensing, actuator control and decision making processes. Due
to the Pi’s relatively higher energy consumption it was used for remote interaction
with the Arduino for troubleshooting, with the intended purpose to remove the Pi
for the actual transatlantic launch. The primary controller on AT is a 3.3V Mega Pro
running at 8 MHz Atmel 8-bit microcontroller provided by Sparkfun Electronics in
Boulder, Colorado, USA. The system takes data in from a GPS device with a helical
antenna allowing for greater reception in moving seas, a tilt compensated compass
(running its own configurable microprocessor), and an IP67 industrial rated hall effect potentiometer for the wind sensor, located on a separate stern mounted mast.
In order to improve survivability of the wind sensor, an industrial grade sensor was
chosen. Sensitivity was sacrificed due to greater friction in the devises bearings.
This friction was overcome by increasing the moment of inertia of the wind sensor by increasing its length. An additional logging device was added that received
serial signals from the main microprocessor and writing it to a miniSD card. An
independent satellite tracking devise was installed to monitor the performance of A
Boat Time while in the ocean. The tracker was installed separate and independent
of any other system to increase its survivability probability. The combination of all
these components led to an average power consumption of approximately 50mA.
System survivability was a key addition for the A Boat Times system design. Two
GPS and Compass systems were installed and coded to act as fall back redundant
systems by Midshipmen Kevin Flaherty and Aaron Dougherty. Additionally, wind
sensor performance was tracked by the system and if a failure was identified, follow
on conditions for sail trim and steering were used to maximize the ability for A Boat
Time to reach its desired location.
Morwyn has been designed for long distance voyaging and power consumption
is a key concern in extending endurance. Morwyn’s target power budget is 1 W
average, although this has not yet been confirmed under real world conditions. If this
is achieved then working on a conservative figure of 10% average efficiency, a 10-15
Wpeak solar panel should provide sufficient to power. Morwyn uses two Olimexino
32U4 boards, these have been selected for their very low power consumption of
approximately 20 mA (at 3.3 V) when active and less than 1 mA in sleep mode.
One of these is responsible for controlling the sail and rudder actuators, reading
the GPS, compass and wind sensor. The other is responsible for logging data to
an SD card and transmitting it via a RockBlock Iridium satellite transceiver. The
control system board periodically sends data via an I2C bus to the other board, which
wakes from its sleep mode, records the data and if enough time has elapsed sends
a message via Iridium. The Iridium messages contain a summary of the compass
heading, roll and pitch angles, battery voltages and currents since the last message.
It also sends the current location, waypoint data and temperature. This design keeps
the communications/logging board in sleep mode most of the time, the non time
deterministic and blocking operation of sending Iridium messages is also moved

MaxiMOOP: A Multi-Role, Low Cost and Small Sailing Robot Platform


away from the critical control system simplifying concurrency and ensuring that the
control system does not freeze or stop due to problems with the communications

5 Sea Trials
The versatility of the MaxiMOOP design is illustrated by the variety of tasks accomplished so far. Dewi successfully competed in the 2013 SailBot competition
and completed all events, including a 6 hour long triangular course for the long
distance event. Figure 10 shows the route taken by Dewi at the 2013 SailBot competition. This race took place on June 13th 2013 in Gloucester, Massachusetts, USA.
Conditions were quite challenging for a small boat, with the wind blowing at 15-20
kts from the North West and with wave heights of around 50-75 cm. This figure
illustrates that during the upwind legs of the course an average speed of 1 knot was
achieved, while on the downwind leg a speed of between 1.5 and 2 kts was achieved.

Fig. 10 Map showing the speed of Dewi during the triangular long distance course at the
2013 SailBot competition

Fig. 11 Map showing ABoat Time’s route during her 2014 Microtransat attempt from Cape
Cod, Massachusetts to Fenit, Ireland. The blue line indicates a measurement of 165 nautical
miles to the East of Cape Cod and the purple line represents the path followed by the boat.
The text at each point indicates the date (MM-DD format) and time (HH:MM format, in
UTC) that the position report was transmitted.


P. Miller

In addition to Dewi’s successful participation in the 2013 SailBot competition and
Morwyn’s experiments in Wales, MLC has served as a successful test platform for
a variety of experiments.
From May 16th-22nd 2014 AT attempted a transatlantic voyage from Cape Cod,
Massachusetts to Fenit, Ireland. The voyage ended early when she was accidentally
caught in a scallop dragger’s net and hauled aboard, during which her aft solar panel
was damaged. During her 220 NM voyage she saw winds to 35 knots and seas to
20 feet along with two days of calm winds and fog. This exceeded the previous
Microtransat record of 123 NM set by Breizh Spirit DCNS in 2012 [1]. Figure 11
shows her track and figure 12 is a picture taken on board the dragger. The crew
reported that she was sailing well at the time and when opened she was dry inside.

Fig. 12 ABoat Time on board the M/V Atlantic Destiny. Her damaged solar panel is evident
and the taped-over wind indicator was necessary to stop her sail and rudder from adjusting
when she was hauled aboard.

Fig. 13 Proposed lines of an improved MaxiMOOP

MaxiMOOP: A Multi-Role, Low Cost and Small Sailing Robot Platform


A similar fate is believed to have been suffered by at least two previous Microtransat entries (Breizh Spirit DCNS in 2012 and Erwan 1 in 2013) and highlights the
need for path planning away from fishing areas, mechanisms to make sailing robots
easily detectable and autonomous collision avoidance strategies.

6 Future Work
While the MaxiMOOP design has fulfilled its goals, it too can be improved. The
keel is not as streamlined as it could be and stability is limited due to the relatively
shallow draft. This has been particularly evident when multiple pieces of equipment
are mounted on the deck. Based on observation of the boat while sailing, a little more
freeboard would also be beneficial. A hypothetical second-generation MaxiMOOP,
including all the original characteristics but with some improvements, including 75
mm more draft and 10 mm more freeboard is shown in Figure 13. This model also
has added volume for an additional 2 kg of payload. The VPP indicates a roughly 49% performance improvement in the voyaging condition. To date the MaxiMOOPs
effectiveness as an oceanographic sensing platform has not been tested, although
a salinity probe and water temperature sensor have been installed in the keel of

1. History of the microtransat, http://www.microtransat.org/history.php
(accessed May 20, 2014)
2. Alves, J., Ramos, T., Cruz, N.: A reconfigurable computing system for an autonomous
sailboat. Journal of the Österreichische Gesellschaft für Artificial Intelligence (Austrian
Society for Artificial Inteligence) 27, 18–24 (2008)
3. Alvira, M., Barton, T.: Small and inexpensive single-board computer for autonomous
sailboat control. In: Sauze, C., Finnis, J. (eds.) Robotic Sailing 2012, vol. 121, pp. 105–
116. Springer, Heidelberg (2013)
4. Bruget, K., Clement, B., Reynet, O., Weber, B.: An arduino compatible CAN bus architecture for sailing applications. In: Le Bars, F., Jaulin, L. (eds.) Robotic Sailing 2013,
vol. 142, pp. 37–50. Springer, Heidelberg (2014)
5. Cabrera-Gámez, J., de Miguel, A.R., Domı́nguez-Brito, A.C., Hernández-Sosa, J.D.,
Isern-González, J., Fernández-Perdomo, E.: An embedded low-power control system for
autonomous sailboats. In: Le Bars, F., Jaulin, L. (eds.) Robotic Sailing 2013, vol. 142,
pp. 65–77. Springer, Heidelberg (2014)
6. Gibbons-Neff, P., Miller, P.: Route planning for a micro-transat voyage. In: Schlaefer, A.,
Blaurock, O. (eds.) Robotic Sailing, vol. 79, pp. 183–194. Springer, Heidelberg (2011)
7. Martin, D.E., Beck, R.F.: Pcsail, a velocity prediction program for a home computer. In:
Chesapeake Sailing Yacht Symposium (2001)
8. Miller, P., Hamlet, M., Rossman, J.: Continuous improvements to USNA sailbots for
inshore racing and offshore voyaging. In: Sauze, C., Finnis, J. (eds.) Robotic Sailing
2012, vol. 121, pp. 49–60. Springer, Heidelberg (2013)


P. Miller

9. Sauzé, C., Neal, M.: Design considerations for sailing robots performing long term autonomous oceanography. In: Proceedings of the International Robotic Sailing Conference, Breitenbrunn, Austria, May 23-24, pp. 21–29 (2008)
10. Sauzé, C., Neal, M.: MOOP: A miniature sailing robot platform. In: Schlaefer, A., Blaurock, O. (eds.) Robotic Sailing, vol. 79, pp. 39–53. Springer, Heidelberg (2011)
11. Schröder, C., Hertel, L.: Development of a low-budget robotic sailboat. In: Le Bars, F.,
Jaulin, L. (eds.) Robotic Sailing 2013, vol. 142, pp. 13–24. Springer, Heidelberg (2014)
12. Stelzer, R., Dalmau, D.E.: A study on potential energy savings by the use of a balanced
rig on a robotic sailing boat. In: Sauze, C., Finnis, J. (eds.) Robotic Sailing 2012, vol. 121,
pp. 87–93. Springer, Heidelberg (2013)

Towing with Sailboat Robots
Luc Jaulin and Fabrice Le Bars

Abstract. Moving huge objects floating at the surface of the ocean (such as containers or icebergs) with boats requires many human operators and a lot of energy.
This is mainly due to the fact that when humans operate such equipment, time is
costly. Now, when we have time (as when robots operate, for instance), it is possible
to move arbitrarily large objects, for over long distances, with a limited quantity of
energy. This is a consequence of the fact that in the water, the friction forces are
proportional to the square of the speed (i.e., when we go slowly, we have almost
no friction). This paper proposes the use of a sailboat robot to tow large objects. It
shows which control law could be used is order to (i) avoid loops inside the towing
cable, (ii) avoid collisions between the robot and the towed object, and (iii) move
the object toward the desired direction. The control law is validated on a simulation
where the object to be towed has to follow a trajectory corresponding to a large

1 Introduction
From an operational point of view, sailboat robots (see e.g. [20] [19] [5] [1] [4]) can
be viewed as floating objects moving toward some desired waypoints and taking the
energy for propulsion from its environment. A sailboat robot has to be small (otherwise it may become dangerous) and therefore cannot be used for transportation.
These vessels can be used as sensors to collect measurements [22] [8], [18] for scientific surveys [15] [25] or as relays for communication [24]. They can even be used
as a windmill in order to produce the required energy [14]. However, the application
field remains limited and it is important to find some other domains where sailboat
robots can become useful.
This paper proposes the use of sailboat robots to tow a large object (named
the load). For such a heavy load (more than 10 times the weight of the sailboat)
Luc Jaulin · Fabrice Le Bars
Lab-STICC/CID/IHSEV, STIC/OSM, ENSTA Bretagne, 2 Rue F. Verny, 29806 Brest, France
e-mail: {Luc.Jaulin,Fabrice.Le Bars}@ensta-bretagne.fr
c Springer International Publishing Switzerland 2015

F. Morgan and D. Tynan (eds.), Robotic Sailing 2014,
DOI: 10.1007/978-3-319-10076-0_3



L. Jaulin and F. Le Bars

floating at the surface of the ocean, the trajectory is mainly governed by the wind
and the currents. Now, towing the load with a sailboat will have a small impact
against the wind and currents. Thus, the main propulsion of the load will be governed by the current and the sailboat can be used to influence the direction of the
load perpendicular to the current. In this way, the sailboat acts as a train switch, able
to move the load from one current path to another. The use of the sailboat we can
expect is to move the load from one point to another taking advantage of the currents. A zero effort path has first to be found using techniques from optimal control
theory and the sailboat will play the role of a regulator controlling the load around
the reference trajectory. The technique is similar to the strategy used to send a space
explorer outside the solar system. To achieve the goal, two subproblems need be
solved (1) control the motion of the load using the sailboat, neglecting the forces
generated by the currents to the load (2) finding a route consistent with the currents
to bring the load to the target.
This paper investigates subproblem (1), i.e., assumes that the influence of the
currents/wind on the load is negligible. The paper proposes a simple and robust
controller which is able to move the load from one point to another. The main difficulty for the sailboat is taking into account the wind for its own propulsion, pulling
the load towards the right direction, and maintaining the towing cable straight.
The paper is organized as follows. Section 2 proposes a state space model which
includes the sailboat, the load and the interaction between both. Section 3 describes
the controller which tunes the rudder angle and the length of the mainsheet in order
to accomplish the mission. A test-case is presented in Section 4. In this test-case, the
sailboat is controlled in order to tow the load on a circular path. Section 5 concludes
the paper.

2 State Space Model
Consider the sailboat represented in Figure 1 with one rudder and one sail. The
sailboat has to tow a load using a cable which is attached both to the boat and to the
load, as illustrated by Figure 2.
The model of the system including the sailboat and the load is given by the following state space equations.
(i) ṁ =
θ̇ =
⎨ (iii) v̇ =
(iv) ω̇ =
(v) ṡ =
(vi) ṅ =

v.uθ + p1aψ uψ − fc uα

f s sin δs − f r sin u1 −p2 v.|v|− f c cos(α −θ )
f s (p6 −p7 cos δs )−p8 f r cos u1 −p3 ω v
f c .uα −p12 .s.s

where the link variables are given by



Towing with Sailboat Robots


Fig. 1 Sailboat that has to tow an object

Fig. 2 The sailboat robot has to tow an heavy load. The angle α (here represented in a vector
form) corresponds to the direction of the cable

(a) wap
(b) ψ ap
(c) γs
(d) δs
(e) fs
(f) fr
(h) α


aψ cos (ψ − θ ) − v
aψ sin (ψ − θ )
atan2(wap )
cos ψ ap + cosu2

π + ψ ap
if γs ≤ 0
−u2 sign (sin ψ ap ) otherwise
p4 wap  sin (δs − ψ ap )
(force on the sail)
p5 v sin u1
(force on the rudder)
exp (m − n − L0 )
(force on the cable)
atan2(m − n)

This model is close to the models developed in [11] and [14] except that it incorporates the load (object to be towed). The following describes all variables involved
in this model.
Inputs. The sailboat has two inputs. The first input u1 = δr is the angle between
the rudder and the sailboat. The second input u2 corresponds to the length of the
mainsheet. More precisely, u2 corresponds to the absolute value of the maximal
angle δs that could reach the sail when the mainsheet is tight.


L. Jaulin and F. Le Bars

State variables. The state variables occurring in the proposed model (1) are m, θ , v,
ω , s, n where m = (mx , my ) are coordinates of the robot, θ is its heading, v is its
speed along the main axis, ω is its angular speed. The state variables associated
with the load should also be included, i.e., the position n = (nx , ny ) and the speed
s = (sx , sy ) of the load.
Parameters. In the model, p1 is the drift coefficient, p2 is the tangential friction, p3
is the angular friction, p4 is the sail lift, p5 is the rudder lift, p9 is the mass of the
boat and p10 is the mass moment of inertia. The distances p6 , p7 , p8 are represented
in Figure 1. Also add the mass of the load p11 , the friction coefficient p12 . All
parameters pi are assumed to be known exactly. Two other quantities should also
be considered as parameters: the speed aψ of the wind and its direction ψ . All
quantities are expressed using the international unit system.
Link variables. These variables are used to shorten the expression of the state equations. (a) The vector wap corresponds to the apparent wind expressed in the robot
frame. (b) The angle of wap (in the robot frame) is denoted by ψ ap . (c) The coefficient γs is positive if the mainsheet is tight. (d) When the mainsheet is not tight,
the angle of the sail δs , is equal to π + ψ ap and it behaves as a flag. Otherwise, the
angle corresponds to ±u2 . (e) fs represents the force of the wind on the sail, (f) fr
is the force of the water on the rudder. (g) The force on the cable is obtained from
the expression of the potential energy of an elastic cable: E p = exp (m − n − L0 ),
where L0 is the nominal length of the cable. (h) α is the angle of the cable which is
assumed to be straight.
State equations. The first equation (i) of (1) expresses that the boat follows its
heading uθ , where uθ = (cos θ , sin θ )T , but always loses some advance with respect
to the wind aψ uψ (where uψ = (cos ψ , sin ψ )T is the direction vector of the wind
and aψ is the speed of the wind). The motion of the boat is also influenced by the
cable through the term fc uα where uα = (cos α , sin α )T . Equations (iii) and (iv) are
obtained using the Newton laws applied to the boat. Equation (v) corresponds to the
Newton law applied to the load.
Note that this model for the sailboat could be made more realistic by adapting the
modeling tools described by Fossen in the context of marine vessel [6] to sailboats
(see [28]). To the authors’ best knowledge a model with a sailboat towing a load has
never been proposed before.

3 Controller
There exist different approaches to control sailboat systems [10] [27] [9] [3] [7] [2]
[16] [26]. This section proposes a pragmatic approach (as in [12] [17]) to have a
simple controller for the towing sailboat, with few parameters and easy to debug.
The system to be controlled is composed of two subsystems: the load and the sailboat. The sailboat can be interpreted as a complex actuator which is able to generate
forces that will influence the motion of the load. To find the controller we decompose into six steps described as follows. (a) Since the load is typically a second order

Towing with Sailboat Robots


system moving on a plane, we first propose a basic proportional control to move it
toward the desired direction. (b) The proportional control is adapted to make the
load move with respect to a desired vector field. (c) We introduce a new concept of
segment for a sailboat which consists of an arc where the robot can move maintaining the cable taut. (d) Unfeasible segments are projected to get the nearest feasible
segment. (e) The heading of the boat is oriented in order to pull the cable as much as
possible. (f) Control the rudder and the sail in order to have the right heading. Once
all these six steps are described, we propose a controller implementation.
(a) Moving the load with a proportional control. Consider a load moving at the
surface of the water. Assume that the load can be pull with respect to the direction α ∗
A proportional like control pulls toward the desired direction α0 , but also corrects by
pulling proportionally to the heading error. More precisely, the proportional control
is given by
α ∗ = α0 + sin (α0 − atan2(ṅ)) .
Here α0 −atan2(ṅ) represents the heading error of the load and the sine function
eliminates the 2π discontinuities. Note that, for this application, it is the sailboat
which pulls. It has its own dynamics and cannot pull toward arbitrary direction at
(b) Vector field. The desired dynamics for the load is described by a vector field
[21] [23]. Classically, a vector field is a function from Rn to Rn (since the load
moves in the plane, n = 2). This vector field associates with a position (nx , ny ) of
the load, the vector (speed and direction) to be followed. Now, in a sailboat context,
the amplitude of the vector has no meaning: we do not want to control our speed
to a given value. Instead, the boat has to do its best to go to the right direction
and to go as fast as possible. As a consequence, a vector field will be a function
from R2 to [−π , π ]. It associates to a position n of the load, the direction α0 to
be followed. The corresponding vector is (cos α0 , sin α0 ). The proportional control
(2) can be used to make the load going in the direction α0 . Again here, we do not
consider that the load is towed by a sailboat and that all directions for α ∗ are not
(c) Feasible segment. We now consider that the load is towed by a sailboat.
Assume that the cable is long and that the speed of the load is small with respect to
the speed of the boat. To maintain the cable taut, the boat has to move in a circle of
radius L0 centered in n (see Figure 3). Define a segment as an arc of this circle with
an half angle ζ0 . A segment is identified by its middle angle ᾱ as shown by Figure
3. A segment is feasible if the two directions (direct and indirect) along this arc are
feasible. A segment with angle ᾱ is thus feasible if
|sin (ᾱ − ψ )| ≤ cos (ζ + ζ0) ,
where ζ is the close hauled angle of the sailboat (typically ζ = π4 ). When the boat
decides to follow a segment, it has to scan this segment back and forth. A Boolean
state variable ε ∈ {−1, 1} has thus to be introduced so that the controller knows
in which direction it is scanning the selected segment. When ε = 1 the segment is
scanned in the direct trigonometric sense. The value for ε is allowed to change only


L. Jaulin and F. Le Bars

Fig. 3 Six segments are represented. Since none of them intersects the gray sectors, they are
all feasible, i.e., the boat can move back and forth, on each segment.

once the boat has reached the border of the segment, i.e., when cos (ᾱ − α ) < cos ζ0 .
In such a case, ε takes the value 1 if sin (ᾱ − α ) > 0 and −1 otherwise.
(d) Projection. Recall that the projection x̄ of a point x onto a subset X of a metric
space corresponds to one point of X which minimizes its distance to x. A point x may
have several projections. Now, for simplicity, we assume that projection is unique.
Proposition. Given a set of angles B = {β such that cos |β | > cos β0 } where β0 ∈
0, π2 , the projection ᾱ of an angle α onto B (i.e., the closest angle in B to α ) is
obtained using the following analytical expression

ᾱ = atan2 (min (|sin α | , sin β0 ) .sign (sin α ) , max (|cos α | , cos β0 ) .sign (cos α )) ,
where atan2(y, x) returns the horizontal angle of the vector (x, y)T .
Proof. The projection ȳ of a real y onto an interval [−b, b] is ȳ = min(|y| , b)·sign(y) .
The projection x̄ of a real x onto the complementary of the interval [−a, a] is
x̄ = max(|x| , a)·sign(x) . To obtain formula (3), it suffices to project sin α onto
[− sin β0 , sin β0 ] and cos α on the set [−1, − cos β0 ] ∪ [cos β0 , 1]. The corresponding
angle (obtained using atan2) is the projection. 

Towing with Sailboat Robots


Fig. 4 Illustration of the projection on the set of angles represented by the hatched sectors.
The αi are projected into ᾱi . Since α5 is inside the section, α5 = ᾱ5 .

Fig. 5 Graph of the angular projection function. The frame box is [−10, 10] × [−4, 4].


L. Jaulin and F. Le Bars

Fig. 6 (a),(c) the boat follows the segment but does not pull the cable; (b),(d) the boat pulls
the load

Figure 4 gives an illustration of the projection onto the set
{β such that cos |β | > cos β0 }.
For β = 0.5, the graph of the projection function (3) is given by Figure 5.
(e) The robot has to pull. If the segment ᾱ is fixed, the robot now has to be
controlled in order to follow its segment. At first glance, it could be thought that the
robot has to follow the direction θ̄ = α + ε π2 . Now, for such a direction, the boat
does not pull anything, as illustrated by Figure 6 (a),(c). To pull the load, the robot
has to look at outside the circle as in Figure 6 (b),(d). The desired heading for the
boat will thus be chosen as θ̄ = α + ε instead of θ̄ = α + ε π2 .
(f) Heading control. To control the low level actuators (here the rudder and the
length of the mainsheet), a heading controller similar to [13] is used. This heading
control is given by
if ( cos θ − θ̄ ≤ 0)
.sign sin θ − θ̄
u1 = π4
4 . sin θ − θ̄

π cos ψ − θ̄ + 1
u2 = .
Here, to avoid doing any loop, always tack with the cable on the back, i.e.,
.sign sin α − θ̄
if ( cos θ − θ̄ ≤ 0) (tacking)
u1 = π4
Controller. The resulting controller is as follows

Towing with Sailboat Robots


Controller. in: m, n, ṅ, θ , ψ ; out: u1 , u2 ; inout: ε
1 α0 = field(n) ; α ∗ = α0 + sin (α0 − atan2(ṅ)) ;
2 α̃ = α ∗ − ψ ; β
0 = 2 − ζ − ζ0 ;

max (|cos α̃ | , cos β0 ) .sign (cos α̃ )
3 ᾱ = ψ + atan2
min (|sin α̃ | , sin β0 ) .sign (sin α̃ )
4 α =atan2(m − n);
5 if cos (ᾱ − α ) < cos ζ0 then ε = sign(sin (ᾱ − α ));
6 θ̄ = α + ε ;
7 if ( cos θ− θ̄ ≤ 0) then
 u1 = 4 .sign sin α − θ̄ else u1 = 4 . sin θ − θ̄ ;
cos(ψ −θ̄ )+1
8 u2 = π2 .
The inputs of the controller are the position m of the boat, the position of the load
n, its speed ṅ, the heading of the boat θ and the direction of the wind ψ . The output
of the controller are the rudder angle u1 and the length u2 of the mainsheet. There is
only one binary state variable ε ∈ {−1, 1}. From n, ṅ, Step 1 computes the direction
α ∗ we want to pull from the desired vector field (b). This direction is obtained using
the proportional control law (a). Since α ∗ may be unfeasible, Steps 2 and 3 project
(see (d)) this angle to get the nearest feasible direction ᾱ . This direction gives the

Fig. 7 The limit cycle Cρ of this vector field is the bold circle


L. Jaulin and F. Le Bars

segment (c) to be followed back and forth. When the boundary of the segment is
reached the scanning direction ε changes at Step 5. The desired heading is corrected
at Step 6 in order to pull the cable (see (e)). Steps 7 and 8 correspond to the low
level control (f) for the rudder and the sail.

4 Test-Case
Assume that the load has to follow a circle Cρ with radius ρ . Thus choose a vector
field so that the circle Cρ corresponds to a limit cycle of the field. An expression of
such a vector field in given by

n − ρ
ϕ (n) = atan2 (n) + + atan

Fig. 8 The load (bold curve) follows the desired path (dotted circle). The sailboat trajectory is
mainly composed with small arcs (or segments) in order the keep the cable taut. The starting
point for the load is the center of the target circle.

Towing with Sailboat Robots


where δ is the required accuracy (typically the GPS accuracy). Figure 7 shows such
a vector field. Taking two circles bracketing the limit circle Cρ (such as the two
dotted circles Cρ− , Cρ+ of Figure 7), the corresponding dynamical system, is captured by the corridor delimited by these two circles. This is the case even for small
For simulation, the same coefficients as [17] are used for the sailboat. For the
cable, a length of L0 = 50m is chosen. The mass of the load is p11 = 2000Kg. The
friction coefficient of the load is p12 = 10 Kg.s.m−1 . For the controller, a segment
with half radius ζ0 = 0.2rad is applied. For the initialization, the load is placed at
the center of the circle. For a circle with radius ρ = 150m, the trajectories painted in
Figure 8 are obtained. This figure illustrates the desired circle Cρ for the load (dotted
circle), the trajectory of the load (bold curve), and the trajectory of the sailboat
(thin curve). The trajectory of the boat is mainly composed of many segments (i.e.,
arcs with half angle ζ0 = 0.2rad  11 deg). The big arcs (with angle around π2 )
correspond to situations where the load wants to go perpendicular to the wind and

Fig. 9 Trajectories of the load and the sailboat, when the radius of the circle to be followed
is ρ = 500m


L. Jaulin and F. Le Bars

thus two different projections exist in (3). The sailboat thus has to alternate between
these two projections.
Remark. Figure 8 contains some arcs that would not be considered as feasible. See,
e.g., at the very top (before the trajectory of the towed load finishes the circle). The
sailboat is performing a large arc windwards. This phenomenon is due to the model
we considered: the dynamic of the boat and that of the load are considered almost
independent. More precisely, in the unfeasible arcs, the boat is in a closed hauled
mode whereas the cable maintains it and the arc. This makes the boat going upwind
(which is not possible). In practice, the switching would require tacking.

For a circle with radius ρ = 500m, we get the simulation illustrated by Figure 9. Note that now, due to the zoom out effect, the trajectory for the load looks
much more accurate. Once the initialization has been performed and that the load
has reached the desired circle, the distance of the load to the circle is always less
than 20m.

5 Conclusion
This paper has presented a controller for a sailboat robot for towing an heavy load
along a desired trajectory. To our knowledge, this problem has never been considered before. The controller is simple to implement (about 10 lines of C++ code) and
only includes easy-to-tune parameters (such as the length of the cable or the angle of
the segments). The feasibility and the robustness of the controller has been tested on
some simulations. It remains to validate the principle on experiments involving an
actual sailboat robot (towing a zodiac, for instance). Note that the controller could
also be used for towing objects that are not as large as addressed in the paper: in
practical application a sailing robot could be used to tow a floating or submerged
sensing device.
All C++ codes associated with the test-case can be found at

1. Brière, Y.: The first microtransat challenge. ENSICA (2006),
2. Cabral, H.M.P., Alves, J.C., Cruz, N.A., Valente, J.F., Lopes, D.M.: MPL — A Mission
Planning Language for Autonomous Surface Vehicles. In: Le Bars, F., Jaulin, L. (eds.)
Robotic Sailing 2013, vol. 142, pp. 133–143. Springer, Heidelberg (2014)
3. Cabrera-Gámez, J., Isern-González, J., Hernández-Sosa, D., Domı́nguez-Brito, A.C.,
Fernández-Perdomo, E.: Optimization-Based Weather Routing for Sailboats. In: Sauze,
C., Finnis, J. (eds.) Robotic Sailing 2012, vol. 121, pp. 23–34. Springer, Heidelberg
4. Cruz, N., Alves, J.: Ocean sampling and surveillance using autonomous sailboats. In:
International Robotic Sailing Conference, Austria (2008)
5. Erckens, H., Büsser, G., Pradalier, C., Siegwart, R.: Navigation Strategy and Trajectory
Following Controller for an Autonomous Sailing Vessel. IEEE RAM 17, 47–54 (2010)

Towing with Sailboat Robots


6. Fossen, T.: Guidance and Control of Ocean Vehicles. Wiley, New York (1995)
7. Gal, O.: Multi-agents Decision Making Concept for Multi-missions Applications in Marine Environments. In: Le Bars, F., Jaulin, L. (eds.) Robotic Sailing 2013, vol. 142,
pp. 123–132. Springer, Heidelberg (2014)
8. Gorgues, T., Ménage, O., Terre, T., Gaillard, F.: An innovative approach of the surface
layer sampling. Journal des Sciences Halieutique et Aquatique 4, 105–109 (2011)
9. Guillou, G.: Architecture multi-agents pour le pilotage automatique des voiliers de
compétition et extensions algébriques des réseaux de petri. PhD dissertation, Université
de Bretagne, Brest, France (2011)
10. Herrero, P., Jaulin, L., Vehi, J., Sainz, M.A.: Guaranteed set-point computation with application to the control of a sailboat. International Journal of Control Automation and
Systems 8(1), 1–7 (2010)
11. Jaulin, L.: Représentation d’état pour la modélisation et la commande des systèmes (Coll.
Automatique de base). Hermès, London (2005)
12. Jaulin, L., Le Bars, F.: A simple controller for line following of sailboats. In: Sauze,
C., Finnis, J. (eds.) Robotic Sailing 2012, vol. 121, pp. 117–130. Springer, Heidelberg
13. Jaulin, L., Le Bars, F.: An Interval Approach for Stability Analysis; Application to Sailboat Robotics. IEEE Transaction on Robotics 27(5) (2012)
14. Jaulin, L., Le Bars, F.: Sailboat as a windmill. In: Le Bars, F., Jaulin, L. (eds.) Robotic
Sailing 2013, vol. 142, pp. 79–89. Springer, Heidelberg (2014)
15. Klinck, H., Stelzer, R., Jafarmadar, K., Mellinger, D.K.: AAS Endurance: An Autonomous Acoustic Sailboat for Marine Mammal Research. In: 2th International Robotic
Sailing Conference, Matosinhos, Portugal (2009)
16. Langbein, J., Stelzer, R., Frühwirth, T.: A Rule-Based Approach to Long-Term Routing
for Autonomous Sailboats. In: Schlaefer, A., Blaurock, O. (eds.) Robotic Sailing, vol. 79,
pp. 195–204. Springer, Heidelberg (2011)
17. Le Bars, F., Jaulin, L.: An experimental validation of a robust controller with the
VAIMOS autonomous sailboat. In: Sauze, C., Finnis, J. (eds.) Robotic Sailing 2012,
vol. 121, pp. 73–84. Springer, Heidelberg (2013)
18. Miller, P., Sauzé, C., Neal, M.: Development of ARRTOO: A Long-Endurance, HybridPowered, Oceanographic Research Vessel. In: Le Bars, F., Jaulin, L. (eds.) Robotic Sailing 2013, vol. 142, pp. 53–64. Springer, Heidelberg (2014)
19. Miller, P.H., Hamlet, M., Rossman, J.: Continuous improvements to USNA sailbots for
inshore racing and offshore voyaging. In: Sauze, C., Finnis, J. (eds.) Robotic Sailing
2012, vol. 121, pp. 49–60. Springer, Heidelberg (2013)
20. Neumann, T., Schlaefer, A.: Feasibility of basic visual navigation for small robotic sailboats. In: Sauze, C., Finnis, J. (eds.) Robotic Sailing 2012, vol. 121, pp. 13–22. Springer,
Heidelberg (2013)
21. Petres, C., Ramirez, M.R., Plumet, F.: Reactive Path Planning for Autonomous Sailboat.
In: IEEE International Conference on Advanced Robotics, pp. 1–6 (2011)
22. Sauze, C., Neal, M.: An Autonomous Sailing Robot for Ocean Observation. In: Proceedings of TAROS 2006, Guildford, UK, pp. 190–197 (2006)
23. Schmitt, S., Le Bars, F., Jaulin, L., Latzel, T.: Obstacle Avoidance for an Autonomous
Marine Robot - A Vector Field Approach. In: 7th International Robotic Sailing Conference. Springer, Irland (2014)
24. Stelzer, R., Dalmau, D.E.: A study on potential energy savings by the use of a balanced
rig on a robotic sailing boat. In: Sauze, C., Finnis, J. (eds.) Robotic Sailing 2012, vol. 121,
pp. 87–93. Springer, Heidelberg (2013)


L. Jaulin and F. Le Bars

25. Stelzer, R., Jafarmadar, K.: The Robotic Sailing Boat ASV Roboat as a Maritime Research Platform. In: Proceedings of 22nd International HISWA Symposium on Yacht
Design and Yacht Construction, Amsterdam, The Netherlands (2012)
26. Stelzer, R., Proll, T., John, R.: Fuzzy Logic Control System for Autonomous Sailboats. In: Proceedings of IEEE International Conference on Fuzzy Systems, London,
UK (2007)
27. Xiao, K., Sliwka, J., Jaulin, L.: A wind-independent control strategy for autonomous
sailboats based on voronoi diagram. In: CLAWAR 2011 (best paper award), Paris (2011)
28. Xiao, L., Jouffroy, J.: Modeling and nonlinear heading control of sailing yachts. IEEE
Journal of Oceanic Engineering (2013)

Part II

Power Management and Mission Planning

Power Management Strategies
for an Autonomous Robotic Sailboat
Kjell Dahl, Anton Bengsén, and Matias Waller

Abstract. Different modes of operation for an autonomous sailing vessel are
introduced. The modes of operation are motivated from a nautical as well as
an energy efficiency point of view. Based on the modes of operation, an efficient way to turn on and off sensors and/or actuators using a microcontroller
and transistor-based relays is applied.

1 Introduction
The paper presents a flexible solution for managing the power consumption of
microcontrollers, sensors and actuators for an autonomous sailing robot. The
main objective in the design of the control system is to enable autonomous
missions with propulsion by sails alone over long periods, i.e., from hours to
months. The focus is further on small sailing vessels between one and four
meters in length with corresponding limited capacities for solar panels and
accumulators for supplying electricity to the control system.
Although all energy for propulsion will naturally be generated by the sails,
the control unit, sensors and actuators still need electrical power. In order
to deal with a limited and varying supply of electricity, heuristics are used
to define different operational modes. These modes are then used as a basis
for power management. Detailed solutions for operating different sensors and
actuators at different, and varying, sampling rates and the microcontroller
at different, and varying, clock rates are presented. Measurements of power
consumption for individual components are presented.
An overall strategy for the long-term operation of an autonomous sailing
vessel is thus presented. Future work will focus on actual field experiments
and their assessment as well as a more thorough review of relevant literature.
Kjell Dahl · Anton Bengsén · Matias Waller
Åland University of Applied Sciences
e-mail: firstname.lastname@ha.ax
c Springer International Publishing Switzerland 2015

F. Morgan and D. Tynan (eds.), Robotic Sailing 2014,
DOI: 10.1007/978-3-319-10076-0_4



K. Dahl, A. Bengsén, and M. Waller

2 Microcontroller and Sensors
Åland Sailing Robots (http://www.sailingrobots.ax/) is a project at the
Åland University of Applied Sciences concerned with various aspects of autonomous sailing. In this project, several different types of small sailing vessels
ranging from one to four meters will be considered. For this reason, different
actuators for rudder and sails will be used since specific choices for rudder
and sails are likely to depend on the vessel used. The focus of the current
study is on strategies for control, power management and choice of sensors
since these aspects are independent of sailing vessel, rudder and sails.
For the choice of microcontroller, low power consumption is one of the
main criteria. Some features for five ATSAM4 microcontrollers based on the
Cortex M4 processor core are presented in Table 1. Due to its very low power
consumption, ATSAM4L is chosen for the prototype considered in the current
Table 1 Comparison between different ATSAM4 microcontrollers

Clock rate

120 MHz
2048 KB
170 KB
200 µA/MHz

100 MHz
1024 KB
80 KB
170 µA/MHz

120 MHz
1024 KB
128 KB
200 µA/MHz

48 MHz
512 KB
64 KB
90 µA/MHz

48 MHz
256 KB
64 KB
103 µA/MHz

Some actual measurements of power consumption for the ATSAM4L in
normal and standby mode were also conducted and are presented in Table 2.
It can be noted that the microcontroller can be run at a lower clock rate along
with other alternative configurations to further reduce power consumption.
Table 2 Measured power consumption for ATMSAM4L
Power consumption
Clock rate

Normal (running) Wait (a standby mode)
3.3 V
3.3 V
4.3 mA
4.7 µA
14 mW
0.015 mW
48 MHz

For a brief comparison, the affordable Raspberry Pi which has been used
for autonomous sailing vessels [4] is considered. At 700 MHz and with 256
MB or 512 MB flash memory for model A and model B respectively, the
models are rated at 300 mA and 700 mA at 5V, i.e., a power consumption of
1.5 W and 3.5 W respectively. With respect to power consumption only and
in active mode, the factor is thus 100-250 in advantage of using ATMSAM4L.

Power Management Strategies for an Autonomous Robotic Sailboat


An advantage of the Raspberry Pi is that it comes readily equipped with one
or two USB-interface(s) for model A and model B respectively. The efforts to
configure, connect and use USB-sensors (e.g., GPS) and USB-actuators (e.g.,
servos) are thus quite modest. Some drawbacks of the Raspberry Pi is that
it can only be run at full clock rate and it has no mode for standby. Also,
it is not intended for industrial use, e.g., not suited for the possibly harsh
environments encountered on a sailing vessel. Furthermore, the Raspberry Pi
includes, among others, a graphics processing unit, a sound card and inputs
and outputs for video and sound. Although extensions using audio and video
can be useful, it is doubtful whether such applications are vital for crossing
the Atlantic with an autonomous sailboat.
A possible drawback of basing a dedicated unit on an ATMEL ARM
SAM4L is that a circuit for connecting sensors and actuators needs to be
designed and created. In this study, a prototype for such a circuit is considered. As sensors, the following are considered:
• GPS, for course over ground (COG), speed over ground (SOG) and
• Compass for heading
• Rudder angle
• Wind vane
• Sheet angle
In addition, voltages of batteries and charging current of solar panels will be
monitored and used for choosing different modes of operation.

3 Different Modes of Operation and Basic Strategies
for Control
As an essential feature of the system, different modes of operation are defined
and routines for automatic switching between different modes are developed.
On a logical level, the modes are classified in five essential categories:

Startup mode
Shore mode
Open water high power
Open water low power
Fault mode

The startup mode is used in order to set up the boat for sailing according to a set heading based mainly on output from compass and wind vane.
Since it can be expected that compass and wind vane readings might benefit
from filtering, these are read at full frequency according to sensor specifications and microcontroller capacity operating at full clock rate. In this mode,
GPS measurements are monitored, and with a simple model, combined with
compass and wind readings in a Kalman filter in order to obtain reliable
estimates of heading, position, COG and SOG. The exact frequency for the


K. Dahl, A. Bengsén, and M. Waller

other sensors and for actuators can depend on the type of vessel and will
be determined based on trial and error. As a starting point, 1 Hz will be
Once reliable estimates of heading, position, COG and SOG have been
obtained, the system automatically switches to shore mode. In shore mode,
a setpoint for COG is used, possibly connected in cascade to the heading
control system. Again, exact frequencies for sensors and actuators in shore
mode will be determined based on experiments, but between 0.1 Hz and
1 Hz might be reasonable based on our experience of small sailing vessels.
Shore mode is also always applied in the vicinity of islands and/or other
obstacles such as (underwater) reefs. Once the vessel is in open water, based
on comparisons between estimated position and predetermined limits for the
mission in question, the system automatically switches between shore and
open water mode. In open water mode, the set point is managed in a similar
fashion to shore mode. In addition, the setpoint can be adjusted according to
the wind direction. This means that instead of simply following a set course,
a course relative wind direction can be chosen. The choice between using a set
COG versus a set course relative to the direction of the wind is determined
based on estimates of heading, COG, SOG, drift, etc.
Depending on voltage measurements of the accumulator, either low power
or high power is chosen within open water mode. In high power mode, i.e.,
when the solar panels are generating more than sufficient electrical power,
the actuators and sensors can be operated fairly often, e.g., at 0.1-0.01 Hz.
In low power mode, on the other hand, sensors and actuators are shut down
for longer periods, e.g., hours to days depending on position, and the microcontroller goes into wait mode. In open water mode, it is deemed that
significant improvement in the surveillance can be achieved by using dead
reckoning (DR) as a complement to the infrequent measurements. Similar
strategies are not uncommon in navigational GPS applications [2].
Fault mode is used when some inconsistencies are accounted, e.g., rudder
indicator appears inconsistent with heading and SOG, DR calculations deviate significantly from infrequent measurements, estimates of the Kalman
filter are highly improbable, etc. In fault mode further calculations are used
in order to determine the likely cause of the inconsistencies. Information is
then gathered and transmitted. In addition, rough sea conditions can be detected with an accelerometer type compass and the circumstances can thus
also invoke special transmissions.

4 Operating Sensors and Actuators
The setup considered in the current study is in many ways similar to an
earlier approach [1], with a similar microcontroller and different sampling
rates for different sensors/actuators. Also in [3], different sampling frequencies
for actuators, sensors, communication devices were used. In the current study,

Power Management Strategies for an Autonomous Robotic Sailboat


Table 3 Measured power consumption and startup times for different sensors
Wind vane
Wind vane
Wind vane and speed CV7
GPS module
Telit GM862-GPS

3.3 V
5.0 V
12 V
3.7 V

16.5 mA
17.3 mA
9.3 mA
85 mA

54 mW
86 mW
112 mW
315 mW

Startup time
75 s

however, different modes of operation that use different sampling frequencies
are introduced. The main reason for introducing different modes is that the
energy consumption of sensors can be significant compared to the ATSAM4L.
Some observations are summarized in Table 3.
The possibility of employing different and varying sampling frequencies for
the sensors by turning them on and off depending on the mode of operation
thus seems to be an attractive alternative from a power management point
of view. Clearly, this requires energy-efficient means to turn off and on actuators/sensors. A possibility is to use transistors for this purpose [5]. A strategy
that uses the microcontroller directly to easily and efficiently turn on and off
sensors (and actuators) via MOS-FET transistors Q1 and Q2 is illustrated
in Fig. 1.
The connection uses a digital output of the microcontroller to switch on
and off sensors. For an output corresponding to a logical one, i.e., approximately 3 V for the ATMSAM4L, the connected transistor is fully saturated,
i.e., the transistor switch is closed and the sensor is on. The reason for using
R1 is a precaution to shield the microcontroller from high, transient currents
due to internal capacitors in the transistor. A typical measurement of closing
the switch is illustrated in the left hand side of Fig. 2. As the figure reveals,
a sensor is switched on in approximately 5 μs.
R2 is needed in order to discharge internal capacitors and open the transistor switch. A larger R2 corresponds to a longer time to switch off. With
R2 = 470kΩ the switch using the transistor considered is opened in approximately 1 ms. A typical measurement of the switch being turned off is
illustrated in the right hand side of Fig. 2. As the figure reveals, a sensor is
witched off in approximately 3 ms.
When this strategy is employed, the sensors will in practice consume zero
power when they are switched off. When they are switched on, the steady
state relays have a voltage drop due equal to the sensor current times the
internal resistor of the transistor, RDS,on ≈ 0.05Ω, where D stands for drain
and S for source, see Fig. 1. The steady state current at logical high voltage
from the microcontroller is negligible as it is given by the logical high voltage
divided by R1 + R2 , i.e., 3V/(471kΩ) ≈ 6μA. For the ATMSAM4L, with a
positive logical voltage of 3 V, power losses are summarized in Table 4 for
the MTD20N03HDL transistor. IDS is the current used by the sensor and
UDS is the voltage drop over the transistor switches. For the transistor the


K. Dahl, A. Bengsén, and M. Waller

Fig. 1 Using transistors (solid state relays) to easily switch on and off sensors/actuators (loads). The figure illustrates the case of two loads with obvious
possibilities for expansion to an arbitrary number of loads.

Fig. 2 Voltage measurements when switching on (left hand side) and switching
off (right hand side) the transistor switch. Blue line is the voltage drop over the
transistor (low when the switch is on) and yellow line is the output from the microcontroller, i.e., the voltage controlling the switch (high to turn the switch on).

maximum voltage for UDS is 30 V and maximum current IDS is 20 A. As a
numeric example for the typical operation of the system, the following can be
considered: 5 seconds is required for switching on the wind vane and speed

Power Management Strategies for an Autonomous Robotic Sailboat


Table 4 Measured power consumption for transistor switches when sensor is on
using the transistor MTD20N03HDL
IDS (mA) UDS (mV) Power loss (mW)

sensor (CV7), acquiring data, moving the microcontroller from standby to
running mode, switching off the sensor, performing reasonable calculations
and setting the microcontroller in standby mode. The energy consumption
for a typical operation cycle will thus be less than 700 mWs. More detailed
measurements of different operation cycles will be considered in future work.

4.1 Some Notes on the GPS Device
The GPS module clearly provides central data for the success of autonomous
sailing missions. However, it has significant power consumption and also a
very long startup time. The manual for the device provide little detail regarding internal (selective) amplifiers and filtering algorithms. Possibly, a
fair estimate of the position at startup acquired, e.g., by DR, might significantly decrease startup times. Given the key role GPS data plays, it might be
motivated to develop a system with greater redundancy and better insight
into and possibilities to control the internal algorithms. Further studies of
these possibilities and alternative GPS modules are required. As an alternative, COG readings based on GPS readings could also be used in an outer
control loop that is updated at a much lower frequency than an inner heading
control loop. The outer loop could thus be a slower, integral type controller
that only infrequently adjusts the set point for the heading.

4.2 Some Notes on the Actuators
Although the focus of the current paper is on power management of microcontrollers