511
1 INTRODUCTION
Large autonomous merchant vessels are still on not
forreal.However,inNorwaythebuildingcontractis
alreadysignedforYARABirkeland,thefirstMari‐time
Autonomous, Surface Ship (MASS), an un‐manned
container feeder, scheduled to start tests in 2020
(Kongsberg,2019).LackingIMOregulations,the
tests
will have to commence in national waters, which in
this case means the Grenland area of Porsgunn and
Larvik in southern Norway with complex narrow,
inshorearchipelagonavigation.Itisabusyindustrial
areawherealargeportionoftheshiptrafficconsists
ofgascarriersandvesselswithhazardous
cargoand,
summertime,anabundanceofsmallleisurecraftsand
kayaks.Theseatrafficintheareaismonitoredbythe
BrevikVTSwhichin2015made623“interventions,”
meaningthattheVTSaskedforsomealterationfrom
the planned sailing route (Statistics Norway, 2016).
Conductingautonomousnavigationinsuch
anareais
ahugechallenge.
The project is ambitious. The 80 meters long,
unmanned,autonomousvessel,taking120containers
with a fully electric propulsion system, will replace
some 40,000 truck hauls every year. Thus moving
heavy traffic from road to sea, from fossil fuel to
hydrogeneratedelectricity.Theplan
iscurrentlythat
shewillstarttestrunsin2020. First with a manned
bridgeonboard, thenwiththesame bridgeliftedoff
to the quay side, remotely controlling the vessel,
beforefinallyattemptingtogoautonomouslyin2022
(Kongsberg,2019).
1.1 Unmanned,automaticandautonomous
Todays manned ships may
be thought of as
“manual.” However, the level of automation is in
many ships quite high. With an autopilot in “track‐
following” mode, set so that the ship can execute
turns along a pre‐planned route without
acknowledgment from the Officer of the Watch
(OOW)‐given that the voyage plan is
correct and
Maritime Autonomous Surface Ships (MASS) and the
COLREGS: Do We Need Quantified Rules Or Is “the
Ordinary Practice of Seamen” Specific Enough?
T.Porathe
NorwegianUniversityofScienceandTechnology,Trondheim,Norway
ABSTRACT:MaritimeAutonomousSurfaceShips(MASS)iscurrentlyontheagendainseveralcountriesand
alsointheIMO.InNorwaya120TEUcontainerfeederisbeingbuildandwillstartsailingautonomouslyin
2022. The challenge is huge. One question is
whether or not the present, quantitative, collision regulations
needstobeupdatedtoruleswhereexpressionsas“early”and“substantial”arequantified?Orifshipscansail
autonomouslyunderthepresentrules?AnotherquestionisifMASSshouldbemarkedtosignalthattheshipis
inautonomousmode?Or
ifitisenoughthatshefollowsCOLREGS?Thisdiscussionpaperwilltakeacloser
look at these questions and advocate automation transparency, meaning that the behavior of an autonomous
vesselhastomakesenseandbeunderstandabletohumanoperatorsonothermannedshipsandcrafts.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 13
Number 3
September 2019
DOI:10.12716/1001.13.03.04
512
validated for a set under keel clearance. This is the
way the Norwegian coastal express Hurtigruten
navigates during most of its inshore route from
Bergen to Kirkenes (Porathe, pers. comm.). But the
OOWstillhastobepresentonthebridgetolookout
forandhandleencounterswithother
shipsandcrafts.
Whatisneededtoremovetheoperatorcompletelyis
different sensors that can see and identify moving,
uncharted obstacles in the sea, and an autopilot
connected with a collision avoidance module
programmed with the International Regulations for
Preventing Collisions at Sea, COLREGS for short
(IMO,1972).With
suchasystemitisspeculatedthata
ship in autonomous mode may navigate
automatically.
However,suchan“automaticship”doesnotneed
tobeunmanned.Itmaycontainamaintenancecrew,
or even a reduced number of navigators who take
manualwatchesduringdifficultconditions,ormaybe
daytime watches in
good conditions, saving the
automation for the long boring night watches or
uneventful oversee passages. With such a partly
manned bridge the ship would have a “periodically
unattended bridge” according to IMO’s latest
definitions,(IMO,2018).
The watch can also be handed over to a Shore
ControlCentre(SCC)that
canaccesstheshipssensors
and communication, ready to wake up the OOW if
something unexpected happens (in which case the
shipis“remotelymonitored”).Or,theSCCcouldbe
grantedaccesstotheautopilot,inwhichcasetheship
willbe“remotecontrolled”.Itisreasonabletothink
that this
will be a gradual evolution towards higher
and higher levels of automation,maybe a
combination of remote monitoring and control, and
autonomy.
It can also be useful to consider the concept
“Operational Design Domain” (ODD) used by the
self‐driving car industry(Rodseth & Nordahl, 2017).
Inthemaritimedomain,
itwouldmeanthattherewill
be certain shipping lanes and fairways were the
automation has been specifically trained and which
have been specifically prepared, maybe with
designated lanes, or by specific technical
infrastructure. In these areas, a ship may navigate
autonomously, while the ship in other areas must
navigate manually
with a manned bridge or remote
controlledfromtheshore.
TheconceptofOODalsohasdeeperimplications
intothecultureofvesseltrafficinspecificareas.More
onthislater.
Forthe discussion inthis paperthe focus will be
on ships in “autonomous mode”, regardless of
whetherit
ispermanentoronlyperiodically.With“in
autonomousmode”Imeanthatacomputerpro‐gram
is navigating, taking decisions and executing them,
regardlessofwhetheranOOWisstandingbyonthe
bridge,orthecaptainisinhiscabinonboardorina
remote centre ashore. The focus here
is on how the
ship automation can handles interaction with other
ships,and particularlyhowitcouldfollowtherules
oftheroad,theCOLREGS.
2 THECOLREGS
For several centuries ships came and went, sailing
withthesamewindandtideanditwasnotuntilthe
steam ships
turned up in the beginning of the 19th
century that collision regulations became vital
(Crosbie, 2006). In 1840 the London Trinity House
drewupasetofregulations,oneofwhichrequireda
steam vessel passing another vessel in a narrow
channeltoleavetheotheronherownporthand.
The
other regulation relating to steam ships required
steam vessels on different crossing courses, so as to
involve risk of collision, to alter course to starboard
and pass on the port side of each other. The two
TrinityHouserulesforsteamvesselswerecombined
into a single rule and
included in the Steam
NavigationActof1846.Duringtheyearsanumberof
iterations and internationalizations, through what is
nowthe International Maritime Organization (IMO),
led to the latest revision of the International
Regulations for Preventing Collisions at Sea
(COLREGS)onaninternationalconferenceconvened
inLondonin1972.
One
may ask if maybe new rules are needed for
autonomous ships? Or maybe there should be ma‐
chine‐to‐machine negotiations in every individual
case of conflicting courses? The final answer to that
questionisunknown,butitismyfirmopinionthatas
longasMASSwillinteractwith
humansonmanned
ships there has to be a limited number of common
andeasy tounderstandrulesknownto, and obeyed
by,allvesselsatsea.Onecandreamupotherrules,
but what we got, and need to adhere to, is the
COLREGS. Having said that, one might consider
if
extensionsorrevisionsmaybeneeded.
2.1 Qualitativerules
The collision regulations are, like legal text often is,
writteninageneralmannersoastobeapplicablein
as many situations as possible. The precise
interpretation has to be made in the context of the
actualsituationjudged
notonlyonknowledgeofthe
rules, but also on experience and culture, what the
rules call “the ordinary practice of seamen,” as is
statedalreadyinthesecondrule.
The qualitative nature of COLREGS will be a
problemfortheprogrammerwhoistowritecodefor
the collision avoidance
algorithms of autonomous
navigation modules. I will in this section point to
some these “soft,” qualitative, clauses where these
problemswillbecomeapparent.
2.2 Rule2:theordinarypracticeofseamen
Rule2oftheCOLREGSisaboutresponsibility.Ithas
twosections.Section(a)state“NothingintheseRules
shall exonerate any vessel, or the owner, master or
crewthereof,fromtheconsequencesofanyneglectto
comply with these Rules or of the neglect of any
precautions which may be required by the ordinary
practiceofseamen,orbythespecialcircumstancesof
thecase.”
Section (b) of the
same rule states that “In
construing and complying with these Rules due
513
regardshall be hadtoalldangers of navigation and
collisionand toany special circumstances,including
the limitations of the vessels involved, which may
makeadeparturefromtheseRulesnecessarytoavoid
immediatedanger.”
Whatthisrulebasicallysays is that you mustal‐
ways follow these rules,
but that you must also
deviatefromtheseruleswhennecessarytoavoidan
accident. In essence, if you have an accident it is a
good chance that you have violated one or both of
thesesections.Theproblemforthenavigatorishow
long, or close into an encounter,
he or she should
followtheRulesandwhenitistimetoskiptherules
anddowhateverisnecessarytoavoidacollision. The
answeris:itdependsonthecircumstances.TheRules
give no hint as to the number of cables or miles,
minutesorseconds.Itdoes
noteventrytodefinethe
“ordinarypracticeofseamen.”
Similarsoftenumerationsarefoundforinstancein
Rules15,16and17.
2.3 Rule15to17,riskofcollision
Rule 15 of the COLREGS talks about “crossing
situations”: “When two power‐driven vessels are
crossing so as to
involve risk of collision, the vessel
whichhastheotheronherownstarboardsideshall
keepoutofthewayandshall,ifthecircumstancesof
the case admit, avoid crossing ahead of the other
vessel.”
Calculatingwhenacrossingsituationmayleadto
acollision isprettystraitforward
giventhatpresent
courseandspeedcanbeextrapolated.(Thisis,how‐
ever,inrealitynotalwaysthecaseastheintentionsof
the other ship may not be known.) If the bearing to
theothershipisconstantovertime,itcanbeassumed
thatthereexistsarisk
ofcollision.Rule15alsodefines
which vessel should take action to avoid collision.
“Theonewhichhas theotheron her ownstarboard
side.”
The following rule then defines how this action
should be done by the “give‐way” vessel (Rule 16):
“Everyvesselwhichisdirectedtokeepout
oftheway
of another vessel shall, as far as possible, take early
andsubstantialactiontokeepwellclear.”
Thisactioncouldbeachangeofspeedorachange
of course, but for the software programmer the
problematic keywords here are “early and
substantial”.Thereisnosuggestion
inmilesorclock
minutes what constitutes “early”, neither how large
course change or speed change constitutes
“substantial”.
Rule17definestheactionsoftheship that isnot
obligedtoyield,“thestand‐on”vessel:“(a),(i)Where
oneoftwovesselsistokeepoutofthewaythe
other
shallkeephercourseandspeed.(ii)Thelattervessel
may, however, take action to avoid collision by her
maneuver alone, as soon as it becomes apparent to
herthatthevesselrequiredtokeepoutofthewayis
not taking appropriate action in compliance with
these Rules.
(b) When, from any cause, the vessel
requiredtokeephercourseandspeedfindsherself so
closethatcollisioncannotbeavoidedbytheactionof
thegive‐wayvesselalone,sheshalltakesuchaction
aswillbestaidtoavoidcollision.(c)Apower‐driven
vessel which takes
action in a crossing situation in
accordance with subparagraph (a)(ii) of this Rule to
avoid collision with another power‐driven vessel
shall,ifthecircumstancesatthecaseadmit,notalter
courseto portforavesselonherown port side. (d)
ThisRuledoesnotrelievethegive‐
wayvesselofher
obligationtokeepoutoftheway.”
Thisruleaddstothecomplexitybyusingqualitative
definitions like “as soon as it becomes apparent,”
“findsherselfsoclosethatcollisioncannotbeavoided
bytheactionofthegive‐wayvessel alone,”“actionas
will best aid
to avoid collision” and “if the
circumstancesatthecaseadmit.”
Figure1.ThiswasthetrafficsituationatSkagen,thenortherntipofDenmarkat15:00on5November2018.Onemayreflect
onthedifficultiesofCOLREGalgorithmsneededtodocollisionavoidanceinsuchanareawheregivingwaytooneship
mightleadintoanotherconflictsituationin
anunpredictable,cascadingmanner(screenshotfromMarineTraffic.com).
514
For a programmer programming the collision
avoidance module of an autonomous navigation
software the difficulty is not only in judging which
action, but also when to execute it “early” and
“substantially”.Theanswerwillbethesameasitwas
in the previous section: it depends on the
circumstances. Are
there only two ships meeting
alone on the high seas the task might be relatively
simple,butattheotherendofthespectrum,inahigh
complexity situation, e.g. in a constrained and
intensely trafficked area like the Straits of Malacca
and Singapore, the task is of an entirely
different
dimension.Not onlydoesthelargenumberofships
ina limited space changethe value of variables like
“early” and “substantial,” but an evasive maneuver
foroneshipmayleadintoa closequarterssituation
with another ship and so on, in a cascading
interactioneffectwithunpredictableresults.
Figure1
shows the complicated traffic situation around
SkagenonthenortherntipofDenmark.
Insomeareastherecanalsobeadifferentculture
ofhowthingsaredone(sometimesquitecontraryto
COLREGS).WhenthehighspeedferryStenaCarisma
trafficked the Gothenburg‐Fredrikshavn line in 30+
knots,
anofficerIspoketosaid“Wealwayskeepout
ofthewayofeverythingthatmovesbecauseweare
so fast and maneuverable” (Porathe, pers. comm.).
AlsointheSoundbetweenSwedenandDenmarkthe
Helsingborg‐Helsingor ferries has a culture of
keeping out of the way in most situations
(Porathe,
pers.comm.).
A possible strategy for a programmer trying to
catch“earlyandsubstantial”aswellas“theordinary
practiceofseamen”foraspecificarea(anODD)could
be to study large amounts of AIS (Automatic
Identification System) data for the specific area in
questionsandfromthat
datadeducelimitsof“early”
and“substantialaction”.Ausefulconceptcouldthen
beships“safetyzones”whichisthezonearoundones
shipthatnavigatorstendnottoletothershipswithin.
“A zone around a vessel within which all other
vessels should remain clear unless authorized,”
(IALA,2008).This
zonetendstobelargerontheopen
sea than in narrow waters or in a port and can be
studied using AIS data. Using such AIS studies,
establishmentofazoneoutside which anaction can
be considered “early” could be attempted. But the
contextisimportant,notonly
thestaticgeographical
context, but also the time dependent traffic density
context.
TheNauticalInstitutementionsthat“Asageneral
guideline,attempttoachieveaCPA(closestpointof
approach)of2(nautical)milesintheopenseaand1
mileinrestrictedwaters”(Lee&Parker,2007,p.35).
If
all ships in such a complex situation where
autonomousandgovernedbycleveralgorithmsthere
isachancethatsuchacollisionavoidanceapplication
could be successful, but in a mixed situation where
mostormanyoftheshipsarecontrolledbyhumans,
whicharelesspredicta ble,theriskof
abadoutcome
isevident.
2.4 Rule19,restrictedvisibility
ThefinalrulethatIwanttobringuphereisRule19,
“Conductofvesselsinrestrictedvisibility.”Thisisa
quitlengthyrulewhichsays:
“(a)ThisRuleappliestovesselsnotinsightofone
another when navigating
in or near an area of
restrictedvisibility.”
Further “(b) Every vessel shall proceed at a safe
speed adapted to the prevailing circumstances and
conditions of restricted visibility. A power‐driven
vessel shall have her engines ready for immediate
maneuver.”
“(c) Every vessel shall have due regard to the
prevailing
circumstancesandconditionsofrestricted
visibilitywhencomplyingwiththeRulesofSectionI
ofthisPart.”
“(d) A vessel which detects by radar alone the
presenceofanothervesselshalldetermineifaclose‐
quarters situation is developing and/or risk of
collisionexists.Ifso,sheshall takeavoidingaction
in
ampletime,providedthatwhensuchactionconsists
of an alteration of course, so far as possible the
followingshallbeavoided:(i)analterationofcourse
toportforavesselforwardsofthebeam,otherthan
for a vessel being overtaken; (ii) an alteration of
coursetowards
avesselabeamorabaftthebeam.”
“(e) Except where it has been determined that a
risk of collision does not exist, every vessel which
hearsapparentlyforwardsofherbeamthefogsignal
of another vessel, or which cannot avoid a close‐
quarterssituationwithanothervesselforwardsofher
beam, shall reduce her speed to the minimum at
which she can be kept on her course. She shall if
necessary take all her way off and in any event
navigate with extreme caution until danger of
collisionisover.”
TheDutchCouncilofTransportationhasaddedan
amplificationto
thisruleforDutchmariners:“During
aperiodofreducedvisibilityunexpectedbehaviorof
othervesselsshouldbeanticipated.Thespeedandthe
correlated stopping distance must correspond with
thissituation,”(vanDokkum,2016).
The big difference with this rule versus Rule 15
above is that in restricted visibility both vessels
are
suddenlygive‐wayvesselsand the responsibilityfor
avoidingacollisionisshared.Theproblemsherefora
quantitative approach lies in soft terms like “safe
speed,” “due regard to the prevailing circumstances
and conditions of restricted visibility” and “take
avoiding action in ample time.” But also in the
problem of defining “restricted visibility.” As a
meteorological phenomenon “restricted” is not
defined,noris“safespeed”,althoughanassumption
mightbethatthevesselshouldbeabletostopwithin
the distance that can be overlooked. An assumption
that cannot always be followed as in many parts of
the world
ships regularly navigate in conditions of
visibilitywhereeventheownshipsforecastle(front)
cannotbeseenfromthebridge.
Another reflection is that “restricted visibility”
refers to human visibility of the eye, which in the
autonomouscasecanbetranslatedtothevisibilityof
the day‐light cameras. Section
(d) in Rule 19 which
refers to when ships are detected “by radar alone”
515
wasaddedin1960afteranumberof“radarassisted
accidents”(themostwell‐knownwastheStock‐holm‐
Andrea Doria accident in 1956). An autonomous
vessel will most probably, apart from day‐light
cameras, AIS and radar, also have infrared cameras
andmaybeLIDAR.Butevenifsensorresources
onan
autonomousshipcouldbejudgedasbeingbetterthan
thehumaneye,thisrulemakesitnecessarytoinclude
visibility sensors to decide if Rule 19, “restricted
visibility,”ortherules11to18,“conductofvesselsin
sight of each other,” should apply. A confounding
factorhere,
thatneedstobetakenintoconsideration,
isthatfogoftenappearsinpatchesorbanks,soeven
if the autonomous ship itself may be in an area of
goodvisibility,theothervesselmightbehiddenina
fog bank, in which case Rule 19 apply. A possible
solutionfor
theMASSmightbetocompareradarand
cameraimages.
Aphenomenon worth taken into consideration is
that while an autonomous vessel will weigh its
different sensor inputs in an objective manner
resultingina sightingwithaprobabilitymeasure,the
human operator on a manual vessel has a cognitive
systemthatprefervisualegocentricinputthroughthe
eyes as compared to exocentric images from radar
andelectronicchartsthatneedstobementallyrotated
tobeaddedtotheinnermentalmap,(Porathe,2006).
An example of this is the allision of the container
vessel Cosco Busan in 2007 with
the San Francisco
Oakland Bay Bridge in heavy fog but with fully
workingradarandGNSS/AISsupport(NTSB,2009).
Thehumancognitivesystemhasotherlimitations
such as e.g. “normality bias” and “confirmation bi‐
as.” (Porathe et al., 2018). With this, together with
otherhumanshortcomingslikefatigue,aninclination
towards short‐cuts, and sometimes sheer viola tions,
the risk is that the list of potential interaction
problems between human and machine guided
navigationwillbelong.
3 QUANTITATIVECOLREGS
Thecodeforacollisionavoidancesoftwarethatisto
coverallpossiblesituationswillhavetobeverylong
and he
would still not suffice. The unknown
unknowns,blackswans,wouldkeepappearing.
From a computer programmer’s point of view, it
mightseem helpfulif allqualitative, soft,
enumerations of COLREGS could be quantified into
nauticalmiles,degreesofarcandclockminutesonce
and for all. This would greatly facilitate
the
development of the necessary algorithms that will
governfuturecollisionavoidancesystems.However,
suchaquantifiedregulatorytextwould,inthesame
way, have to be very lengthy and it would still not
cover all possible situations. Instead COLREGS, like
otherlegaltextwillneedtohaveageneral
formatthat
isopentointerpretationsinacourtofmaritimelaw,
andtheoppositeof“theordinarypracticeofseamen,”
i.e.“goodseamanship,”includejuridicaloptionssuch
as “negligence” and “gross negligence”, (van
Dokkum, 2016). Ships technical performance and
maneuverability, experience and training of seamen,
allevolvewithtime,so
fortherulesoftheroadtobe
validtheymustbewritteninageneralmanner.
Insteaditis thealgorithmsof collisionavoidance
applicationsthatneedtobepreciseandquantitative.
By using AIS data and large scale simulations,
applicationscan bemade to learn the most effective
and efficient way of maneuvering in different
situations, still following the COLREGS. It would
probably be beneficial if such machine learning was
ongoing “lifelong” for the AI (Artificial Intelligence)
on the bridge, which then would become more and
more experienced through the years. However, it is
unlikely that the IMO would
accept an AI on the
bridgewhichwasnotcertifiedandwhobehavedina
precisely predetermined way for a specific situation
(evenifthiscouldbedefendedbycomparing theAI
toatrainedandlicensedthirdmateworkinghisway
up through the ranks gaining more and more
experience).
Anotherpointtopayattentiontoisthat,aslongas
there are manual ships governed by humans on the
sea, the actions of autonomous ships has to be
predictable for these humans. Autonomous
navigation,supportedbyartificialintelligenceonthe
bridge, has a number of advantages compared to
human,
manual navigation: improved vigilance,
improvedsensingandperception,longerendurance,
anabilitytolookfurtherintothefutureandtokeep
more alternative options open during the decision
makingprocess.Forinstance,bykeepingtrackofall
shipmovementsonaverylongrangeanAImightbe
able to
predict a possible close quarters situation
several hours ahead of a human navigator but may
therefor make maneuvers which might not make
sense to an OOW on a manual ship in the vicinity.
Therefore, it is of outmost importance that
autonomousshipsarepredictableandtransparentto
humans.
4 AUTOMATIONTRANSPARENCY
4.1 Anthropomorphism
Every one of us that are struggling with the
complexity of digital tools know that they do not
always do what we want or assume they will do.
They“think”differentlyfromus.Aninnatetendency
of human psychology is to attribute human traits,
emotions,orintentionsto
non‐humanentities.Thisis
calledanthropomorphism.Wedosobecauseitgivesus
a simple (but faulty) method to “understand”
machines. However, the chance is that if we know
that MASS always will follow COLREGS, we can
learntoknowtheirbehaviorandinahumanmanner
be able
to understand their working. This in
opposition to normal, manned ships, where you
alwayshave to be cautious of misunderstandings or
violations.
4.2 Identificationlight
In my opinion it is therefore important that ships
navigationinautonomousmode show some kindof
identificationsignal.Itcouldbean“A”addedto
their
AISiconinECDISorontheradarscreen.Duringdark
alightsignalcouldbeadded(e.g.apurplemast‐head
all‐aroundlight,seeFig.2).
516
Theassumptionaboveisthatifautonomousships
always follow COLREGS their behavior will be a
hundred per cent predictable. But as we have seen
above,thismightnotbetrueife.g.thespectrometers
onboard the autonomous ship does not interpret
“restricted visibility” the same way we do (and
thereforeRule19shouldorshouldnotbeused).
Figure2: Should ships navigating in autonomous mode
carry a special identification light? The behavior of the
navigationAImaybedifferentfromthebehaviorofnormal,
manned ships. The light could be purple which is a color
that is not used for other purposes. The same discussion
andcolorchoiceis
debatedintheautonomouscarindustry.
4.3 Intentions
Anotherimportantissueisunderstandingintentions.
Interpretingthe intentions of other ships correctlyis
imperative to rule following. An old accident in the
English Channel 1972 can serve as an example of
what misinterpreted intentions (and therefore
applyingthewrongrules)mayleadto:
The ferry St. Germain,
coming from Dunkirk in
FranceanddestinedforDover,wasturningslowlyto
port,awayfromthestraitwesterlycoursetoDo‐ver.
Insteadher captain intended to take her south‐west,
downontheoutsideoftheTrafficSeparationScheme
(TSS),intheInshoreZone, in order to finda
clearer
placetocrosstheTSSata“rightangle”accordingto
Rule10oftheCOLREGS.ThebulkcarrierAdartewas
headingnortheastuptheTSSto‐wardstheNorthSea.
Thepilotonboardrecognizedtheradartargetasthe
Dunkirk‐Dover ferry and assumed, quite wrongly,
thatshe
wouldcross aheadofhimandthattherenow
existedariskofcollision(Rule15).Adartewouldthen
bethegive‐wayshipandwasobligedtogive‐wayby
turning to starboard. At the same time St. Germain
started her port turn, the pilot on Adarte started to
made
aseriesofsmallcoursealternationstostarboard
togiveway(quitecontrarytothe“substantialaction”
required by Rule 16). But instead St. Germain
continuedherportturnandthetwoshipscollided.St.
Germainsank,killinganumberofpassengers(Lee&
Parker,2007).
This accident is retold
to illustrate the need to
understandintentionsandthisgoesforbothmanned
andunmannedships.Iftheintentionoftheothership
is not understood, the risk is that COLREG will not
saveasituation.Itisimportantthatautomationshare
information about its workings, its situation
awareness and its
intentions. Questions like: What
does the autonomous ship know about its
surroundings?Whatothervesselshasbeenobserved
byitssensors?Thesequestionscouldbeansweredby
e.g. a live chart screen accessible on‐line through a
webportalbyothervessels,VTS,coastguardetc.See
Figure3.
Based on
its situation awareness the automation
willmakedecisionsonhowitinterpretstherulesof
collision avoidance. It would be a benefit if the
intentionsofshipscouldbecommunicated,asargued
in Porathe & Brodje (2015). Large ships obey under
IMO’sSOLASconvention.ASOLASship(asdefined
in Maritime
Rule Part 21) is any ship to which the
InternationalConventionfortheSafetyofLifeatSea
(SOLAS) 1974 applies; namely: a passenger ship
engaged on an international voyage, or a non‐
passenger ship of 500 tons’ gross tonnage or more
engagedonaninternationalvoyage(IMO,1980).
SOLAS
ships must transmit their position and
some other information using AIS. In addition,
SOLAS ships are usually big and make good radar
targets, which will provide a second source of
information.Furthermore,allSOLASshipmustmake
a voyage plan from port to port. Several passed and
ongoing projects aim at
collecting voyage plans and
coordinating ship traffic for reasons of safety and
efficiency (e.g. EfficienSea, ACCSEAS, MONALISA,
SMART navigation,SESAME, andthe STM
Validationprojects).Theseattemptsinrouteexchange
wouldmakeitpossibleforSOLASships–alsoMASS
‐tocoordinatetheirvoyagesandshowintentionswell
aheadof
timetoavoidenteringintoaclosequarters
situationwheretheCOLREGswillapply.
Fig.3: Exampleof automation transparency:An on‐line chartportal showing thesituation awareness ofthe autonomous
ship(hereAutomatExpress),whereshethinkssheis,whatothershipsandobjects shehasobserved,andwhatintentionsshe
hasfortheclosefuture.An“A”isaddedtotheAIS
symbolfor“Iamnavigatingautonomously”andtheintendedroute
showncouldalsobevisibleinECDISandradarsofshipsinthevicinity.
517
Route exchange would for instance allow each
ship to send a number of waypoints ahead of the
shipspresentpositionthoughAIStoallshipswithin
radio range. All ships can then see other ships
intendedroute(asinFig.3).IntheACCSEASproject
2014asimulatorstudy was
madewith11professional
British, Swedish and Danish bridge officers, harbor
masters, pilots and VTS operators with experience
from complex traffic in the test area which was the
HumberEstuary.Thefeedbackfromtheparticipants
on the benefits of showing intentions were overall
positive(Porathe&Brodje,2015).
5 CONCLUSIONS
Ihave inthis discussion pointed at some challenges
facingdevelopersofcollisionavoidancesoftware.
Muchofthishastodowiththequalitativenatureof
COLREGSvisaviethequantitativeneedsofreallife
situations.
However,alsotheinteractionbetween traditional
ships in “manual mode” is from time to time
problematic. The introduction of autonomous ships
which in their navigation follows a machine
interpretationofCOLREGSmightleadtomanymore
problemsifnotimplementedcarefully.
It is of great importance that the maneuvers of
autonomousshipsarepredictabletohumanoperators
on manual ships. The AI onbord has a
potential to
becomemuch“smarter”thanhumans,andtobeable
to extrapolate further into the future and thereby
behave in a way that might surprise people
(“automationsurprise”).Insteadthesoftwareshould
focusonbehavinginahumanlikemanner.
Such automation transparency might consist of
MASSshowingitsnavigationmode
(thepurplemast‐
headlight=inautonomousmode),thecontentofits
situationawareness(whichvesselsareobserved–and
thereby which are not observed) and its intentions.
Intentions can be shared e.g. using route exchange
technologydevelopedinrecente‐Navigationprojects
likeEfficienSea,ACCSEASandMONALISA.
Only
if other mariners can understand the
workingsofMASS,apeacefulcoexistenceispossible.
ACKNOWLEDGEMENTS
Thisdiscussionpaperbuildsonapreviouspaperpublished
in the COMPIT 2019 conference. However, the solutions
and argumentation has been further developed. The
research is conducted within the SAREPTA (Safety,
autonomy, remote control and operations of transport
systems) project funded by the Norwegian Research
Council,whichisherebygratefullyacknowledged.
REFERENCES
Crosbie,J.W.2006.LookoutVersusLights:SomeSidelights
on the Dark History of Navigation Lights. Journal of
navigation59(1):1‐7.
van Dokkum, K. 2016. The COLRGS Guide (6th ed.).
Vlissingen,The Netherlands: DokMarMaritime
Publisher.
IALA,theInternationalAssociationforAidstoNavigation
and Lighthouse Authorities. 2008. VTS Manual.
https://www.iala
‐
aism.org/wiki/dictionary/index.php/Ship_Safety_Zone
[Acc.2019‐02‐10]
IMO, the International Maritime Organization. 1972.
Convention on the International Regulations for Preventing
CollisionsatSea,1972,(COLREGs). London:IMO.
IMO.1980.TheInternationalConventionfortheSafetyofLifeat
Sea (SOLAS) 1974. London: IMO.
http://www.imo.org/en/About/Conventions/ListOfConv
entions/Pages/International‐Convention‐for‐the‐Safety‐
of‐Life
‐at‐Sea‐(SOLAS),‐1974.aspx[Acc.2019‐02‐10]
IMO.2018.Regulatoryscopingexercise fortheuse ofMaritime
Autonomous Surface Ships (MASS). MSC 100/5/6, 12
October2018.London:IMO.
Kongsberg. 2019. Autonomous ship project, key facts about
YARA Birkeland. https://www.km.kongsberg.com/ks/
web/nokbg0240.nsf/AllWeb/4B8113B707A50A4FC12581
1D00407045?OpenDocument[Acc.2019‐02‐10]
Lee,G.W.U.&
Parker,J.2007.ManagingCollisionAvoidance
atSea.London:NauticalInstitute.
NTSB,NationalTransportation SafetyBoard. 2009.Allision
of Hong Kong‐Registered Containership M/V Cosco Busan
with the Delta Tower of the San Francisco–Oakland Bay
Bridge San Francisco, California November 7, 2007.
Washington DC: Accident Report NTSB/MAR‐09/0,
PB2009‐91640.
Porathe, T. 2006. 3‐D Nautical Charts and Safe Navigation.
Vasteras:MalardalenUniversityPress.
Porathe,T.&Brodje,A.2015.HumanFactorAspectsinSea
Traffic Management. In V. Bertram (ed.), Proceedings of
the14thInter‐nationalConferenceonComputerApplications
and Information Technology in the Maritime Industries
(COMPIT ‘15).
Technische Universität Hamburg‐
Harburg.
Porathe,T., Hoem, A.,Rodseth,O., Fjoltoft, K. & Johnsen,
S.O. 2018. At least as safe as manned shipping?
Autonomous shipping, safety and “human error. In S.
Haugenetal.(eds),SafetyandReliability–SafeSocietiesin
aChangingWorld.London:Taylor&FrancisGroup.
StatisticsNorway.2016.Variousindicatorsfromtheoperational
area of VTS centres. https://www.ssb.no/191461/various‐
indicators‐from‐the‐operational‐area‐of‐vts‐centres[Acc.
2019‐02‐10]
Rodseth O.J. & Nordahl H. 2017. Definition for autonomous
merchantships. Version1.0,October10.2017.Trondheim:
Norwegian Forum for Autonomous Ships.
http://nfas.autonomous‐ship.org/resources‐en.html
[Acc.
2019‐02‐10].