607
1 INTRODUCTION
Autonomousshipping refers tothe concept of using
advanced technology, such as cybernetics and
robotics, to operate ships with less or even without
human intervention. The research on autonomous
shipping has been around for a long time, but it is
gainingmoreandmoreattentionrecentlyasa
result
oftherapidadvancementsintechnology,decreasing
costof requiredon‐boardsensorsand the increasing
demand for more efficient, more environmentally
friendlyandsafershippingpractices.
Simulators are important tools for the
development and validation of automatic control
algorithms.They rely on mathematical manoeuvring
models, which are often
based on physical scale
modeltestsespeciallyinshallowandconfinedwater,
tosimulateshipmanoeuvringbehaviours,providing
testing in a digital world [1]. They allow for the
realistic testing of a wide range of scenarios and
conditions in a safe and controlled environment.
Differentkindsofwaterwayscanbeimplemented
in
thesimulationenvironmenttoverifythepracticability
oftrackkeepingautopilots.Weatherconditions,such
as waves, tides, and wind [2] can be modelled in a
simulator and their intensity adjusted. Furthermore,
theeffectofshallowandconfinedwatermayalsobe
takenintoaccountbyusingcorrespondingmodels[3].
At the same time, simulators reduce the risks
associatedwithphysicaltestsonrealships,especially
in challenging conditions. They allow developers to
iterate on their designs and test different scenarios
quicklyandeasily.
Testing in simulators is typically more cost‐
effectivethanphysicaltestingonrealships.Physical
testingcan
becarriedoutintwoways:atfullscaleor
A Ship Manoeuvring Desktop Simulator for Developing
and Validating Automatic Control Algorithms
H.He
1
,E.Lataire
1
,T.VanZwijnsvoorde
2
&G.Delefortrie
1
1GhentUniversity,Ghent,Belgium
2FlandersHydraulics,Antwerpen,Belgium
ABSTRACT:Thispaperpresentsauser‐friendlysimulatordevelopedbasedonWindowsFormsanddeployed
as a test bed for validating automatic control algorithms. The effectiveness of some of the integrated track
controllershasbeentestedwithfreerunningexperimentscarried
outintheTowingTankforManoeuvresin
ShallowWaterinOstend,Belgium.Thecontrollersenableashiptofollowpredefinedrandompathswithhigh
accuracy.Ship‐to‐shipinteractionisconsideredinsomecases.Simulatorenvironmentsprovideusefultoolsfor
extending the number of validation scenarios, supplementing the work
performed in the towing tank. The
simulatorispresentedwithagraphicaluserinterface,aimingatprovidingagooduserexperience,numerous
test scenarios and an extensively‐validated library of automatic control algorithms. With the usage of the
simulator, further evaluation of developed control algorithms by implementing extensive test runs with
different ships and waterways could be made. Case studies are shown to illustrate the functionality of the
simulator.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 17
Number 3
September 2023
DOI:10.12716/1001.17.03.12
608
at model scale. In a full‐scaletest, the experiment is
appliedonarealshipandconductedinafairwayor
at sea. Testers have no control over hydro‐
meteorologyandthereforeshouldbeluckyorpatient.
Asignificantamountofhumanresourcesisrequired
to cooperate in the
test. A model scale test is
performed in an experimental environment (e.g.
towing tank). Waves and wind can be made by a
wavemakerandafanbutalsothewaterdepthand
bathymetryareunderfullcontrol.Theenvironmentis
controllable but the costs of installing and utilizing
these
facilities are high. By simulation, the costs of
timeandmoneyarereallylow.Abatchofsimulation
runs with different settings can be performed in a
short time, which helps researchers test developed
methods extensively before deployment. In this
process,issuesmaybeidentified.Inaddition,bugsin
the control
algorithms can be identified in the
debuggingmodeandparametersettingscanbewell‐
tunedbeforeimplementation.Allofthishelptospeed
upthedevelopmentprocessandensurethatthefinal
productisrobustandeffective.
This paper aims to elaborate on a ship
manoeuvring desktop simulator applied to
the
development and validation of automatic control
algorithms, which is developed recently by the
Maritime Technology Division of Ghent University
and Flanders Hydraulics. The layout of its user
interface and the functions of each panel on the
interface are shown in Section 2. The library of
automatic control algorithms embedded in
the
simulatorisintroducedinSection3.Thislibrarycould
beextendedtotestnewmethodsinthesimulatorand
couldbeexportedtouseinphysicaltesting.Section4
describesseverallegacymathematicalmodelsofship
manoeuvringutilizedtogeneratemotiondata.Section
5 gives five case studies with
the applicationof this
simulator.Thecasesaremainlydistributedintothree
tasks:pathfollowing,trajectorytrackingandcollision
avoidance.
2 GRAPHICALUSERINTERFACE
2.1 Layout
Adedicatedgraphicaluserinterface(GUI)iscreated
tofacilitate performing simulations. Itis a Windows
Forms application developed on .NET Framework
4.7.2. The GUI
of the simulator – of which a print
screen is shown in Figure 1 – consists of several
panels, e.g. a map panel (Map A+Map B), a gauge
panel,amonitorpanelandconfigurationpanels.
Figure1.Graphicaluserinterfaceoftheshipmanoeuvring
simulatorbasedonWindowsForms[4].
2.2 MapPanel
The function of the map panel has been developed
basedontheopensourcelibrary,GMap.netWindows
Forms&Presentation[5].Itprovidescontrolstoselect
different map providers and define the centre and
direction of the map. Currently, map providers,
Google Satellite, Google Terrain, Bing Terrain, Bing
SatelliteandOpenSea,areavailableinthesimulator.
The OpenSea map is highly recommended as it
provides nautical information. On this map panel,
twomapwindowsofdifferentsizesanddirectionsare
included. Map A rotates with the ship because it
directs to the ship heading, while Map B always
points North. By default, the map centre is the
positionoftheship,somapsmovewiththeship.
Worldwide waterways, such as the Western
ScheldtandYangtzeRiverasshowninFigure2,can
bechosenasthecurrentnavigationenvironment.This
provides a number of realistic test scenarios, as
a
result the practicality of the automatic control
algorithmscanbevalidated.
Figure2. Western Scheldt, Netherlands (left) [4] and
ZhenjiangsectionofYangtzeRiver,China(right)[6].
In addition, a custom navigation area, shown in
Figure 3, has been created to represent a test basin
environment.Simulationrunscarriedoutwithinthis
area can then be referenced to free running tests
performed in a towing tank, and the difference
betweenasimulatedandarealtestcanbe
compared
tovalidateandimprovemathematicalmodels.
Figure3. Custom navigationarea representingthe Towing
TankforManoeuvresinShallowWater.
Over the map layer, layers made up of markers
andlinesareaddedtoillustrate waypoints, planned
route,ownshipcontour,ownship’shistorytrajectory
and predictive trajectory, and encountered ship’s
contouranditstrajectory,asshowninFigure4.
609
Figure4Waypoint(bluebullet),plannedroute(yellowline),
own ship contour (red polygon), own ship’s history
trajectory (red line), own ship’s predictive trajectory (blue
line), encountered ship (black polygon) and encountered
ship’strajectory(blackline)displayedonthemap[4].
Route planning is achieved by picking up
waypointsonthemaplayerandgeneratingacurved
route through curve fitting and interpolation, as
showninFigure5.Aroutecanbemodifiedbyadding
anddeletingwaypoints.Threemethodsareprovided
for generating a route: Piecewise Cubic Hermite
Interpolating Polynomial (PCHIP),
Cubic Spline
Interpolation (SPLINE) or Akima Piecewise Cubic
Hermite Interpolation (AKIMA).,.SPLINEleadstoa
smoother result, PCHIP produces less oscillation
when data are not smooth, and AKIMA tries to
combinetheadvantagesoftheprevioustwomethods.
In an oscillatory environment (on a track with more
bends), PCHIP creates
a flattened curve, while
SPLINEandAKIMAbothprovidesmoothcurves(see
Figure6).AsseenfromFigure7,inaflatregion(ona
trackwithfewerbends),SPLINEdeliversacurvewith
sort of oscillation, while PCHIP and AKIMA could
reduceit.
Figure5.PlannedroutefromwaypointselectioninYangtze
River[7].
2.3 GaugePanel
Onthegaugepanel(seeFigure1),whichisdeveloped
basedonaWindowsFormpackage,WinFormGauge
[8], the ship’s speed, heading, water depth, track
error,rudderangle,propellerrateandsamplingtime
are displayed. Buttons to launch and suspend the
simulation process as well as switch
control modes
between manual control and automatic control are
present.Whenmanualcontrolisactivated,thesliders
canbemovedtoadjusttheinputsofrudderangleand
propellerrate.
2.4 MonitorPanel
Themonitorpanelplotsthetimehistorycurvesofthe
controlactions(rudderangleandpropellerrates)
and
someoftheshipstates(e.g.heading),andcrosstrack
error, as an example is shown in Figure 8. The
windowcanbezoomedinandoutandcurvescanbe
hidden to highlight the remaining ones. This
informationcouldhelpuserstodebugtheirprograms,
evaluate control performance and
tune control
parameters.
Figure8Monitortoshowsystem’sinputandoutput
2.5 Configuration
Simulationfrequency(thespeedofthesimulatorthat
is asymptotically limited by calculation time) and
samplingfrequency(atwhichfrequencyanewmath
model evaluation is required) can be set
independently to speed up or slow down the
simulationprocess.Accordingto theperformance of
thehost computerand
thecomplexityofthecontrol
algorithm chosen, there is a limitation to the
maximumsimulationfrequency.Thecalculationtime
for each simulation step is displayed at the right‐
bottomcorner of the gauge panel, and therefore the
reciprocalofthesimulationfrequencyshouldalways
belargerthanthisvalue.
Currently,
threeshipmodels,Tanker,Marinerand
Container (see in Chapter 4), can be chosen as the
controlplant.Theirmaindimensionsandinitialstates
are displayed and can be modified. Two control
schemesareavailableinthesimulator.The essential
parameters of each scheme are listed on the
“Configuration” page and
can be tuned before and
duringthe simulation. Ship parameters andobstacle
(encountered ships when considering ship‐ship
interaction)parametersarealsoadjustedonthispage.
Figure6Pathgenerationinanoscillatoryenvironment(lefttoright:SPLINE,PCHIP,AKIMA)[7].
610
Figure7Pathgenerationinaflatregion(lefttoright:SPLINE,PCHIP,AKIMA)[4].
2.6 Otherfeatures
Fast‐time simulation can also be performed to save
time if the user is not interested to the simulation
process. Only the final result will be shown on the
monitorandmap.
Simulationresultscanbesaved,whichprovidesa
spreadsheet containing the values of the ship states
andimagesvisualizingthetrajectoryandtimehistory
oftheheading,rudderangle,andpropellerrate.
Figure9.“Communication”pagetoconfiguretheserialport
The simulator is able to communicate with other
devices through serial port communication. This is
achievedbytheconnectingthesimulatorandtheship
(avirtualshiponanothercomputerorarealship)via
a serialport line and talkingto each other based on
the predefined communication protocol. A
“Communication” page is available to configure the
serialport,asshowninFigure9.
3 GUIDANCE,NAVIGATIONANDCONTROL
3.1 Outline
Figure10illustratestheoutlineofthesimulator,from
which it can be seen that the simulator consists of
threemainparts,theGNC(guidance,navigationand
control) library,
the GUI and mathematical
manoeuvringmodels.
Figure10Frameworkofthesimulator
The GNC library contains automatic control
algorithms developed by Ghent University and
FlandersHydraulics,mostofthemarevalidatedwith
freerunningtestsintheTowingTankforManoeuvres
in Shallow Water, Ostend, Belgium. The library is
writteninC#asintheframeworkoftheGNCsystem
proposedbyFossen
[2].
3.2 Guidancesystem
Theguidancesystemtakesinputsfromthenavigation
system(shipstates)andtheuserinterface(waypoints)
andproducesreferencestothecontrolsystem.Based
on the waypoints provided, a curved route for the
ship to follow is then generated with fitting and
interpolation (not only the
waypoints but also the
entirecurveistakenintoaccountinthetrackcontrol).
A point selected on the map is presented in
latitude and longitude. However, it is much more
convenienttoconsidermotiononatwo‐dimensional
surface. Mercator projection is then implemented to
map a point on the
sphere surface to a two‐
dimensionalsurface[9]:
0
1sin
ln
21sin
m
yR
R
xm
(1)
where(𝜄,𝜆)representsalatitude‐longitudepoint,(𝑥
m,
𝑦
m)denotesaWebMercatorpoint,𝜆0isthelongitude
ofanarbitrarycentralmeridianthatisusuallythatof
Greenwich (i.e., zero),.and 𝑅 is the radius of earth,
which is 6378137m. Latitude and longitude are in
radians.
Theinversetransformation ofthe aboveequation
is
611
0
1
2tan exp
2
m
m
x
R
y
R
(2)
Figure11showsthereferencesystemsadoptedin
the simulator. A local ground‐fixed frame system
(LGF, north‐east‐down) whose origin locates at the
beginning point of the planned route is used for
simplifyingcalculation,ascoordinatesintheMercator
frame (MF) are usually huge numbers. The ship
information relative to the route and path
information, e.g. cross track error, along track error,
relative heading/ course, desired positions on the
routeandcurvature,ismoredifficulttoobtainwhen
followingacurvedroutethanfollowingstraightlines.
Therefore,amovingpath‐tangentialreferencesystem,
the Serret‐Frenet frame (SF)
[10], is then utilized to
solve this. Based on the guidance law, line‐ of‐sight
[11], a reference heading is provided to reduce the
cross track error in path following. However, some
commercialautopilotsrequirethereferenceyawrate,
whichcouldbederivedfromthereferenceheading.In
addition,in
caseofemployingpredictivecontrol,the
desired trajectory (including position, course and
velocities)inaperiodshouldbegivenasthereference
tothecontrolsystem.
Figure11Referencesystemsadoptedinthesimulator
3.3 Navigationsystem
Thenavigationsystemplaysaroleingatheringsensor
dataaboutshipstatesandenvironmentalperception,
processing the raw data and delivering fused ship
states and environmental information. For the
processing,alow‐passfilterisemployedtofilterthe
noise,whiletheextendedstateobserver(ESO)is
used
forestimatingshipstatesanddisturbance.
Prediction can also be produced with the
implementation of a predictive model based on the
Nomotomodel:
~
cos
sin
rr
rK
r
TT
xU
yU
(3
where𝜒isthecourse,𝑟istheyawrate,𝛽isthedrift
angle,𝐾and𝑇arethegainandtimeconstantinthe
Nomotomodel,𝛿isrudderangle,𝑈isthespeed,and
(𝑥,𝑦)isthecoordinate.Supposing thespeedanddrift
angle
remains constant in a prediction horizon, the
ship’sfuturetrajectorywiththecurrentrudderangle
couldbeforecasted.InFigure12,thebluedashlines
representthepredictivetrajectoriesin150secondsfor
atankerclassship(length:304.8m,speed:15kn).
Figure12.Predictedtrajectorieswithrudderangles:‐29deg,
‐15deg,1.6deg,10.7degand23deg(lefttoright)[4].
3.4 Controlsystem
Thecontrolsystemisthecoreofthedecision‐making
processoftheship.Basedonthereferencestatesand
real/predicted states, it proposes control actions for
theshiptofollowthereference. Therearetwomain
routes, model‐free control and model‐based control,
to design a
control scheme. The model‐ free control
does not require an inherited model of the control
plant,as thecontrol laws are proposed according to
the current error between the reference and
measurement. Therefore, a correction appears after
theexistingdeviation.However,shipmodels, which
reflect the dynamic response to the
control input of
theship,areutilizedinmodel‐basedcontrol.Dynamic
behaviourisconsideredinthecontrollaw,leadingto
advanced control. A correction before deviation
becomes possible if the prediction on model basis is
consideredinthealgorithm.
In this simulator, two control methods are
available.
Active disturbance
rejection control (ADRC), the
frameworkofwhichisshowninFigure13.Ittakes
the reference heading (𝜓
d) as input, employs an
ESOtoestimatetheheading(𝑧
1),yawrate(𝑧2)and
disturbance(𝑧
3),usesatrackingdifferentiator(TD)
to smoothen the reference (𝑣
1) and gets its
differential (𝑣
2), and then produces feedback
control(𝛿
0)(e.g.PDcontrol).AnotherkeyinADRC
isdynamiccompensation(𝛿𝛿d),whichequipsthe
controllerabilitytocounteracttheinnerandouter
disturbance(𝑏=𝑇/𝐾).
Figure13.FrameworkofADRC
Modelpredictivecontrol(MPC),theframeworkof
whichisshowninFigure14MPCisastate‐of‐the‐
art control method, which enables the system to
take action in advance by looking ahead to the
nearfuture[12].Thisrequiresa predictive model
oftheship(e.g.aNomoto
model,orasimplified3‐
DOFmanoeuvringmodel).Itoptimizesthecontrol
action in a feasible region to minimize the
612
differencebetweenthereferencetrajectoryandthe
predicted trajectory. The feasible region is
constrained by the maximum control input and
collision‐free condition. The optimization is
performed at each time step, providing updated
prediction and control action. This boosts the
robustnessofthecontrolsystem,withtheabilityto
tolerate
the error from its predictive model and
otheruncertainties,andatthesametimeimproves
controlaccuracy.
Figure14.FrameworkofMPC
3.5 Systemidentification
The GNC system is enhanced with system
identification to make it more practical. Appropriate
ship manoeuvring models are required in the
navigation system to design observers and in the
control system for the design of controllers. These
modelsincludeprincipallytheNomotomodelandthe
3‐DOF models,
Abkowitz and MMG models.
However,determiningthecoefficientsin the models
isquitedifficult.Thesystemidentificationtechnique
can quickly estimate the coefficients based on
systematic input‐output data collected in
manoeuvring tests or a navigation database. A
number of parameter estimation methods can be
found in literature, for instance,
least square [13],
Kalman filter [14] and support vector machine [15].
The simulator employs the least square method to
identifytheNomotomodel.Duringasimulationrun,
theusercouldpausetheprocessandgettheidentified
modelonthe“Identification”page.
4 SHIP MANOEUVRINGMODEL
The computer numerical simulation necessitates
accuratemathematicalmanoeuvringmodelsofships.
To describe realistic manoeuvring behaviours,
dedicated hydrodynamic manoeuvring models are
utilized. They are critical to the simulator because
theydeterminethesimulationqualityandtherealism
of the behaviour of the ships. The hydrodynamic
modelsarederivedthroughthetheoryofkinematics
anddynamics
andreflectshipmotionresponseunder
theactionofhydrodynamicforcesandcontrolforces.
Modelsaretypicallydividedintotwotypes:modular
models and integrated models. In an integrated
model,suchastheAbkowitzmodel[16]andNorrbin
model [17], the hydrodynamic forces on the hull,
propeller and rudder are considered
as a whole,
which is then expanded to terms with respect to
velocities and rudder angle by using Taylor series
expansion.However,inamodularmodel,forinstance
theMMGmodel[18],theforcesonhull,propellerand
rudder are modelled separately but with the
considerationofinteractionamongthem.
Severalshipmodelsareavailableinthesimulator:
Mariner
AMarinerclassvessel(afastcargoship)operating
atitsdesignspeedispresentedintheformofthe
Abkowitz model. This model takes into account
only the rudder angle as an input parameter,
making it suitable for verifying a
path‐following
(constant propeller rate) algorithm. The detailed
model and its hydrodynamic coefficients can be
foundin[19],[20].
Tanker
The Norrbin model is preferable for simulating
largeships.WiththeusageoftheNorrbinmodel,
the manoeuvring motion of the ESSO Osaka
190000 dwt crude oil tanker is
simulated. This
modeltakesbothrudderangleandpropellerrate
as inputs and is suitable in deep and shallow
waters, with a concise form. Therefore, this ship
model could be employed to test algorithms of
trajectory tracking (the propeller needs to be
controlled) and to investigate the shallow water
effects on
controllers. The formulas and
parametersofthemodelareshownin[20],[21].
Container
Son and Nomoto [22] presented a nonlinear
rolling‐coupledmanoeuvringmodelofacontainer
carrierthathasadisplacementof21,222m3.Itisa
typeoftheseparatedmodelwithinputsforrudder
angle
and propeller rate, so it can be used for
validating trajectory tracking. Coefficients of the
model can be found in the above‐mentioned
literatureand[20].
The main dimensions and the design speed of
theseshipsarelistedinTable1.
Table1.Principalparametersofshipsutilizedinthe
simulator
________________________________________________
Ship[m] Length Breadth Design Design Design
between draft displa‐ speed
perpendicularscement
[m] [m][m][m]
3
[nm/h]
________________________________________________
Tanker 304.80 47.17 18.46 220,000 16
Mariner 160.93 23.17 8.23 18,541 15
Container 175.00 25.40 8.50 21,222‐
________________________________________________
5 APPLICATION
The simulator is implemented to model kinds of
scenarios to validate and test the automatic control
algorithms.Inthefollowing,fivecasestudiesapplied
to various ship models, navigation scenarios and
controllers will be introduced to elaborate on the
applicationofthesimulator.Thissectionismerelyto
show the applicability, because as depicted above,
manoeuvring models that are a bit outdated and
simplified are used as provided and have not been
correctedforshalloworconfinedwater.Asdiscussed
in Chapter 6, they will be updated with the models
proposedbyFlandersHydraulics.
613
5.1 Case1:ADRCpathfollowing
Figure 15 shows the result of Container’s path
following by implementing the ADRC controller in
the Western Scheldt, The Netherlands. With the
definition of 13 waypoints, a route was created and
theContainercouldfollowthepredefinedroutewith
highaccuracy(theredline
agreeswiththeyellowline
well).However,differences couldstillbeseenatthe
corners of the route where the ship requires a
(unachievable) large yaw rate to make a turn (for
instance,asseeninFigure15(b)).FromFigure15(c),it
can be seen that the predicted trajectory of
150
seconds inputted with the current rudder command
(for the ADRC algorithm, the predicted trajectory is
just for indication to visualize the decision‐making)
wasstilldirectedtotheoutsideoftheplannedroute
atthecorner.Thismeansthatthereisalateturningof
theshiparoundthecorner,
whichisdue tothatthe
shipcould notmakeactioninadvancebasedonthe
informationofthefuturepath.
Figure15. ADRC path following of the Container with a
constant80rpmpropellerrateintheWesternScheldt[4].
5.2 Case2:MPCpathfollowing
Figure16illustratesContainer’spathfollowingresult
byusingtheMPCcontroller intheWestern Scheldt,
Netherlands.Thepredefinedrouteisthesameasthe
one in Case 1. With the application of the MPC, the
Containerwasabletofollowtheroutewithvery
high
accuracy(seeinFigure16(a))evenaroundthecorners
of the route, where the tracking error was largely
reduced compared with that when using the ADRC
controller(seeinFigure15(b))andFigure16(b)).This
isbecauseoftheactionoftheshipinadvancewiththe
implementation of
a predictive model (a Nomoto
model). As seen from Figure 16(c), the Container’s
predicted trajectory (only 30 seconds was shown as
the prediction horizon used in the MPC was 30
seconds) was very close to the planned route.
Therefore, the performance of the MPC controller is
betterthanthatofthe
ADRCcontroller.
Figure16. MPC path following of the Container with a
constant80rpmpropellerrateintheWesternScheldt[4].
The MPC algorithm was also validated with free
runningtestsintheTowingTankforManoeuvresin
ShallowWaterinOstend,Belgium.AKVLCC2model
ship a length of 4.267 meters, whose detailed main
dimensionscanbefoundin[23],wasadoptedasthe
control plant. Figure 17 shows the result
of a free
running test, where in can be seen that the MPC
controllerprovidesreallyhightrackingaccuracy.
Figure17. Validation of the MPC controller with a free
runningtestoffollowingaDNA‐shapedpathintheTowing
Tank for Manoeuvres in Shallow Water (400 RPM,
predictionhorizonof8seconds)
5.3 Case3:MPCtrajectorytracking
Path following only set spatial constraints, while
trajectorytrackinghastimelimitationsapartfromthe
spatialconstraints.Inotherwords,thespeedneedsto
be controlled in trajectory tracking. Figure 18 shows
theresultoftrajectorytrackingoftheTankerbyusing
theMPCcontroller.
Theunderkeelclearanceissetto
50%ofthe draft.Thesimulationisperformedinthe
Nanjing section of the Yangtze River, China. A
desiredspeedof16knotswasrequiredfortheshipto
keep.Inthiscase,a simplifiedandidentified3‐DOF
predictivemodelwasemployed
tocouplethesurge,
swayandyawmotions,thereforeruddercontroland
propellercontrolwerethencoupled.Fromthefigure,
itiseasytoseethattherealtrajectoryalmostoverlaps
the planned route, indicating high accuracy in
followingthepath.Atthesametime,thespeedgauge
presentsameasured
speedofnearlyaround16knots,
meaningthattheshiphadreachedthedesiredspeed
withtheautomaticcontrol.
614
Figure18. MPC trajectory tracking of the Tanker with a
desirespeedof16knintheNanjingsectionofYangtzeRiver
[24].
5.4 Case4:half‐automatedcollisionavoidanceby
modifyingthepredefinedroute
Figure19showsthecollisionavoidanceprocessofthe
Mariner ship in the Norderelbe, Germany by
implementing MPC and modifying the predefined
route manually. The propeller rate of the Mariner
remained at 80 rpm, which results in her speed
of
around 15 knots. Figure 19(a) illustrates that the
Mariner met a ship (length: 200 m, speed: 8 knots)
around her predefined route in a head‐on situation,
and there is a high risk of collision. Based on the
COLREGs[25],theMarinerneededtogiveherwayto
her
starboard side in this situation. As shown in
Figure 19(b), the next waypoint was then moved to
her starboard side with a distance manually by the
user and the planned path was updated then. In
Figure 19(c), (d) and (e), the Mariner went ahead to
thenewrouteandfinallystayed
onitandsheavoid
theconflictsuccessfully.
5.5 Case5:fullyautomatedcollisionavoidancebysetting
constraints
In the MPC algorithm, distance constraints to the
obstacles (encountered ships) can be set in its
optimization function. In this way, while following
thepredefinedroute,asafetydistancetotheobstacle
isalsorequired.TobeaccordancewiththeCOLREGs,
constraintsoftheyawratesshouldalsobesettomake
the ship turn to its specific side (starboard or port).
After that, three classic scenarios of collision,
overtaking, meeting and crossing, were modelled.
Figure 20 and Figure 21 shows the simulated
overtakingandcrossingprocessoftheTankerinthe
Western Scheldt by using the constrained MPC
algorithm,whileFigure22showsthemeetingprocess
in the Nieuwe Maas, Netherlands. The under keel
clearanceissetto50%ofthedraft.
Figure19.Half‐automatedcollisionavoidanceoftheMarinerbymodifyingthepredefinedrouteintheNorderelbe[26].
Figure20.AutomatedovertakingprocessoftheTankerintheWesternScheldtbyusingconstrainedMPC[4].
615
Figure21.AutomatedcrossingprocessoftheTankerintheWesternScheldt(atthecrossingsectionofthemainfairwayand
thesecondaryfairwaybetweenBathandHansweert)byusingconstrainedMPC[4].
Figure22.AutomatedmeetingprocessoftheTankerintheNieuweMaasbyusingconstrainedMPC[27]
6 CONCLUSION
Theproposedshipmanoeuvringsimulatorprovidesa
wide range of test scenarios for the validation of
automatic control algorithms. Users can select
differentwaterways,allaroundtheworld,togenerate
desired routes of different kinds as input of the
algorithms and create custom collision scenarios (by
generating dynamic obstacles)
for ship‐ship
interaction. As a result, the developed control
algorithms can be verified with well‐rounded and
extensivetestingwiththissimulator.
The process of autonomous navigation is
simulated and visualized via the user interface.
Duringtheprocess,theparametersofthecontrollers
canbe tunedbasedonthe
informationdisplayedon
themonitorpanelandthemap,whichfacilitatesthe
investigation of the principles for parameter tuning.
Inthedebuggingmode,issuesinthealgorithmscan
beidentifiedeasilyandthencorrected.
Theautomaticcontrolalgorithmsarewrappedina
module,increasingcodereusabilityandextendibility.
Extensively‐tested algorithms
can be extracted from
thesimulatorandprovidedtoclientstoimplementin
physicaltesting.Newalgorithmscanbeverifiedwith
thesimulator.Onewayistoestablishcommunication
with the simulator through a serial port connection.
And another way is to incorporate them into the
moduleofautomaticcontrol
algorithms.
Although the presented simulator is a powerful
tool to bring together manoeuvring models and
control algorithms, rather simple and outdated
manoeuvringmodelshavebeenadoptedtoshowthe
applicabilityinthispaper.Theenvironmentprovided
in the simulator is limited to calm water and deep
water(exceptfortheTanker
model).Thismeansthat
presentlytheonlypossibleenvironmentaldisturbance
is shallow water. Flanders Hydraulics has more
accurate shallow and confined manoeuvring models
in calm water and recently disturbance from waves,
currentandwindisconsideredinsomeofthemodels.
The coupling of the current simulator and the
Flanders
Hydraulics’shipmodelAPIwillbedonein
thenearfuture,bringingthistoolanditsoutputtoa
muchhigherlevel.
REFERENCE
[1]O. F. Sukas, O. K. Kinaci, and S. Bal, “Theoretical
background and application of MANSIM for ship
maneuvering simulations,” Ocean Eng., vol. 192, Nov.
2019.
[2]T. I. Fossen, Handbook of marine craft hydrodynamics
andmotioncontrol.Wiley,2011.
[3]M. Vantorre, K. Eloot, G. Delefortrie, E. Lataire, M.
Candries,
and J. Verwilligen, “Maneuvering inshallow
and confined water,” in Encyclopedia of Maritime and
OffshoreEngineering,JohnWiley&Sons,Ltd,2017,pp.
1–17.
[4]OpenSeaMap, “Western Scheldt,” 2023. [Online].
Available: https://map.openseamap.org/. [Accessed: 06‐
Mar‐2023].
[5]J. D. L. Rodriguez, “GMap.NET.Windows.” [Online].
Available: https://github.com/judero01col/GMap.NET.
[Accessed:22‐Feb‐2023].
[6]
Google Maps, “Zhenjiang section of Yangtze River,”
2023.
[7]GoogleMaps,“NanjingsectionofYangtzeRiver,”2023.
[Online]. Available:
https://www.google.be/maps/place/Nanjing,+Jiangsu,+C
hina.[Accessed:06‐Mar‐2023].
[8]Code Art Engineering, “WinForm Gauge,” 2022.
[Online]. Available:
616
https://winformgauge.codearteng.com/. [Accessed: 06‐
Mar‐2023].
[9]Wikipedia, “Mercator projection.” [Online]. Available:
https://en.wikipedia.org/wiki/Mercator_projection.
[Accessed:22‐Feb‐2023].
[10]F.Frenet,“Surlescourbesàdoublecourbure.,”J.Math.
PuresAppl.,pp. 437–447, 1852. [11]T.I.Fossen,“Line‐
of‐sight path‐following control utilizing an extended
Kalman filter for estimation of
speed and course over
groundfromGNSSpositions,”J.Mar.Sci.Technol.,vol.
27,no.1,pp.806–813,2022.
[12]M. Abdelaal, M. Fränzle, and A. Hahn, “Nonlinear
model predictive control for trajectory tracking and
collision avoidance of underactuated vessels with
disturbances,” Ocean Eng., vol. 160, pp. 168–180, Jul.
2018.
[13]
W. Raman‐Nair and R. Gash, “Least squares
identification of linear sway‐yaw manoeuvring
coefficients and drag‐area parameters of ships,” Proc.
Inst.Mech.Eng.PartMJ.Eng.Marit.Environ.,vol.235,
no.3,pp.809–815,2021.
[14]M. A. Abkowitz, “Measurement of hydrodynamic
characteristics from ship maneuvering trials
by system
identification,”1980.
[15]X.G.ZhangandZ.J.Zou,“IdentificationofAbkowitz
model for ship manoeuvring moti on usingε‐support
vectorregression,”J.Hydrodyn.,vol.23,no.3,pp.353–
360, Jun. 2011. [16]M. A. Abkowitz, “Lectures on ship
hydrodynamics‐‐steeringandmanoeuvrability,”1964.
[17]N.H.Norrbin,
“Theoryandobservationontheuseofa
mathematicalmodelforshipmaneuveringindeepand
confinedwaters,”Proc.8thSymp.Nav.Hydrodyn.,vol.
2,no.August1970,pp.807–904,1970.
[18]H.YasukawaandY.Yoshimura,“IntroductionofMMG
standard methodfor ship maneuvering predictions,” J.
Mar.Sci.Technol.,
vol.20,no.1,pp.37–52,2015.
[19]M. Chislett and J. Strom‐Tejsen, “Planar motion
mechanism tests and full scale steering and full scale
steering maneuvering predictions for a mariner class
vessel,”1965.
[20]FossenT,Guidanceandcontrolofoceanvehicles.1994.
[21]T. Cimen, “Development and validation
of a
mathematical model for control of constrained non‐
linear oil tanker motion,” Math. Comput. Model. Dyn.
Syst.,vol.15,no.1,pp.17–49,Feb.2009.
[22]K.‐H.SonandK.Nomoto,“Onthecoupledmotionof
steeringandrollingofashipinfollowingseas,”J.Soc.
Nav.Archit.
Japan,Jan.1982.
[23]“SIMMAN2020,” 2020. [Online]. Available:
https://simman2020.kr/contents/Ship_data_summary.ph
p.[Accessed:06‐Mar‐2023].
[24]OpenSeaMap,“NanjingsectionofYangtzeRiver,”2023.
[Online]. Available: https://map.openseamap.org/.
[Accessed:06‐Mar‐2023].
[25]IMO, “COLREGs,” 1972. [Online]. Available:
https://www.imo.org/en/About/Conventions/Pages/COL
REG.aspx.[Accessed:06‐Mar‐2023].
[26]OpenSeaMap, “Norderelbe,” 2023. [Online]. Available:
https://map.openseamap.org/.[Accessed:06‐Mar‐
2023].
[27]OpenSeaMap, “Nieuwe Maas,” 2023. [Online].
Available: https://map.openseamap.org/. [Accessed: 06‐
Mar‐2023].
