283
1 INTRODUCTION
Scientific and technological progress in electronics
andrelatedareasoftechnologyaswellastheneedto
increasethesafety of sea travel have resulted in the
development of devices supporting the work of the
navigator.Specializedradardevicesappearedonthe
market in the form of ARPA
(Automatic Radar
Plotting Aids) anti‐collision systems, ECDIS
(Electronic Chart Display and Information System),
AIS (Automatic Identification System) and many
others. They provide information about the location
of: detected foreign objects (destination, course and
speed), obstacles in the form of shallows or wrecks,
danger zones [1]‐[4]. Currently, ECDIS systems
connect all navigation devices, thus creating an
integrated information system about ship motion
parameters.Theyareagreatsupportforthenavigator
who, on the basis of the information provided by
them,makesappropriatedecisionsrelatedtosafesea
travel. Despite the facilities offered by modern
devices,shipcollisionsstilloccur,
oftencausedbythe
human factor. Many marine disasters can be
prevented by improving and using computer‐aided
methods for safe ship motion control, such as:
differential games, fuzzy control, expert systems,
genetic algorithms, neural networks, etc. The
improvement of these methods can lead to the
automation of the object
tracking process based on
informationobtainedon‐boardanti‐collisionsystems.
These algorithms can be used to support the
navigatorʹs decisions or as an addition to the
automatic control of the ship in a collision situation
[5]–[11].Choosingtheright anti‐collisionmanoeuvre
at the right moment will eliminate the
navigatorʹs
error and thus increase the safety of navigation. In
recentyears,artificialintelligencetechnologyhasbeen
rapidly developing in many areas of life, including
industryinseaandlandtransport,etc.[12]‐[14].Many
scientists are trying to improve and introduce new
solutionstoshipsystemsresponsiblefor
seatravelso
that vessels become safer, economical and even
partially autonomous. The implemented methods of
automatic steering are able to facilitate the work of
navigatorsbyperformingcalculations,estimatingthe
safetyoftheselectedsearoute, taking controlofthe
ship and making optimal decisions or assisting the
Modern Method Based on Artificial Intelligence for
Safe Control in the Marine Environment
M.Mohamed‐Seghir
GdyniaMaritimeUniversity,Gdynia,Poland
ABSTRACT:Thearticlepresentsanapproachtoformulatingashipcontrolprocessmodelinordertosolvethe
problemofdeterminingasafeshiptrajectoryincollisionsituations.Fuzzyprocesspropertiesareincludedin
themodeltobringitclosertoreality,asinmanysituations
thenavigatormakesasubjectivedecision.Aspecial
neural network was used to solve the presented problem. This artificial neural network is characterized by
minimum and maximum operations when set. In order to confirm the correctness of the operation of the
proposedalgorithm,theresultsofthesimulationsobtainedwere
presentedandandiscussionwasconducted.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 17
Number 2
June 2023
DOI:10.12716/1001.17.02.03
284
crew in servicing the shipʹs infrastructure during
normalseavoyageoperation[15].
Thepurposeofthisstudyistodevelopamethod
tohelpthenavigatorinmakingdecisionsincollision
situationsandtoshowthepossibilitiesofformulating
amodeloftheprocessofsafeshipcontrol
inafuzzy
environment and solving it with the use of artificial
intelligence.
2 ASIMPLIFIEDKINEMATICMODELOFTHE
PROCESS,TAKINGINTOACCOUNTTHE
DYNAMICPROPERTIESOFTHESHIP
In the general case, the behaviour of the ship is
interpreted as the motion of a rigid body with a
specific mass of accompanying water, with six
degreesoffreedom.Forthepurposesofshipmotion
controlincollisionrisksituations,theconsiderationis
limitedtotheshipmotion inthehorizontalplaneof
the movable or stationary coordinate system. The
parametersoftransmittanceorovertakingtimet
wand
angular velocity ω
z are used to assess the shipʹs
dynamicproperties.Theseparametersareselectedon
the basis of ship dynamics tests and depend on the
angle of the rudder deflection, speed, degree of
loading and external conditions. Neglecting the
decrease in speed during course manoeuvring, the
kinematic relative motion of the
ship, taking into
account its dynamic properties after the manoeuvre,
can be approximately described by the following
equations[16].
00
**
*
00
00
**
00
sin sin
sin sin
sin sin
2
jjjj w
jj
jj
w
Xt X V V t
VV
tan V V t
(1)
00
**
*
00
00
**
00
sin sin
sin sin
sin sin
2
jjjj w
jj
jj
w
Yt Y V V t
VV
tan V V t
(2)
where:
X
j,Yj,Xj(t),Yj(t)–coordinatesofthej‐thobjectbefore
andafterthemanoeuvre,respectively,
V
j,
j – the speed and course of the j‐th object,
respectively,
V
0,
0,V0
*
,
0
*
‐speedandcourseofthevesselbefore
andafterthemanoeuvre,respectively.
In the case where maneuvering is velocity, the
equationsofmotiontakethefollowingform
00
**
000 0
sin sin
sin sin
on
kw
on
jjjj on
tT
t
T
jj
T
Xt X V V T
VVVVe dt
(3)
00
**
000 0
sin sin
cos cos
on
kw
on
jjjj on
tT
t
T
jj
T
Yt Y V V T
VVVVe dt
(4)
where:T
on–delaytime;Tkw–thetimeconstantofthe
shipʹshullandaccompanyingwater.
Taking into account the presented assumptions,
the output values of the model in the form of CPA
(Closest Point of Approach) and TCPA (Time to
Closest Point of Approach) can be derived from
equations(3and4).The
modeldescribedinthisway,
withtheassumptionsmade,isamathematicalsystem
thatwillbethebasisforsolvingthisproblem[17].
3 MODELOFTHEPROCESSWITHITSFUZZY
PROPERTIES
Inordertoobtainamodernanti‐collisionsystem,itis
necessary to study the process that makes up
the
navigation environment and manoeuvrability of the
ship, as well as the subjectivity of the navigator in
makingdecisions.Theprocessofsafeshipcontrolin
collisionrisksituationscanbedescribedbyageneral
model of multi‐stage decision making in a fuzzy
environment. Certain constraints are imposed on
all
possibledecisions,sonotalldecisionsareadmissible,
therefore the optimal decision is sought among the
possible solutions. The ship manoeuvres in
accordancewithCOLREGs(InternationalRegulations
for Preventing Collisions at Sea) in relation to the
most dangerous object encountered, i.e. The shipʹs
dynamic properties are represented by the
angular
velocityω
zandtheovertakingtimetw.It isassumed
that the controlled process is a deterministic system
definedbytheequationofstate,andTheequationsof
state and output with discretetimeare described as
follows.
1,
x
tfttXXU
(5)
1,
y
tfttYYU
(6)
where:X,Y, U‐asetofstates,asetofoutputsanda
setofcontrols,respectively.
Taking into account the above assumptions, the
process of determining the shipʹs safe trajectory is
treated as a decision‐making process in a fuzzy
environment. On the decision set U,
a fuzzy set of
goals G and a fuzzy set of constraints C with
appropriatemembershipfunctionsaredefined.
The membership function of the fuzzy goal set
definedwithvaluesintherange[0,1]isasfollows:
22
1
10
dt
G
CPA TCPA
for T CPA
e
(7)
However, for TCPA<0, the value of the function
equals1,
where:
CPA‐ClosestPointofApproach,
TCPA‐TimetoClosestPointofApproach,
d,
t – parameters of the navigatorʹs subjectivity in
assessingtheshipʹssafety.
The way loss is presented as a fuzzy constraint
whosemembershipfunctionhasthefollowingform:
285
00
cos
1
Cz z z
C
VV T
e
(8)
where:V
z,
z–servicespeedandsetcourseoftheship,
C – parameter of subjectivity of the navigator in
assessingthelossoftheway,
T
z–thetimeremainingtotheturningpointoftheset
courseortoreachingtheservicespeedinthecaseofa
speedmanoeuvre.
In the presented work, the ship collision risk
membershipfunctionwasformulatedinthefollowing
form:
22
1
0
rd rt
R
CPA TCPA
for T CPA
e
(9)
However, for TCPA <0, the value of the function
equals0,
Where:
rd,
rt – navigator subjectivity parameters in ship
collisionriskassessment.
The values of the coefficients
d,
t,
C,
rd,
rt
remained established on the basis of empirical
research among a selected group of navigators.
Characteristic collision situations were presented to
thenavigatorsintheformofsimulations,whichwere
determined on the basis of previous studies of ship
traffic. For example, the values of the goalʹs
membership function should
be given for each
situationandtypeofvessel,andforgoodandlimited
visibility.Thenavigatorcanchooseavaluefrom0to1
at his discretion, 1 is absolute safety, 0 is absolute
danger.
Tosolvetheproblemformulatedabove,amethod
based on anartificial neural network was
proposed,
asbelow.
3.1 Algorithmbasedonaspecialneuralnetwork
Theaim of this workisto solvetheproblemofsafe
ship control in collision conditions [18], [19]. This
problem is solved by a method based on neural
networks, which solves the problem of dynamic
programmingina
fuzzyenvironment.thestructureof
eachartificialneuralnetworkdependsonthenumber
of layers and the rules of connections between
neurons, its size, speed of operation, and above all,
theeffectivenessofactionsthatareatoolforsolvinga
givenproblemdependonit.Inthiscase,the
problem
is the optimal control of the ship in collision
situations, and more precisely, determining its safe
trajectory in a blurred environment. To fix the
problem,startwiththelaststepandthengobackto
the previous steps. Referring to dynamic
programming,whenreturningfromstageNtostage
0, there are two phases in each stage: minimization
andmaximization. Suchoperations canbepresented
usingaspecialneuralnetwork,which,amongothers,
was proposed in work [20]. This neural network is
characterizedbyminimumandmaximumoperations
after a finite set. Taking into account the above
information, the algorithm
solving the problem of
optimal multi‐stage control in a fuzzy environment
canbedescribedbyemulatingdynamicprogramming
usinganeuralnetworkandpresentedintheformof
pseudocode.
Algorithm‐Pseudocode
________________________________________________
Begin
1 calculationofconstraintandgoalmembership
functionsatallstages;
2 forallneuronsofthemaximumtypeatstagek=0set;
0
j
R
qM
;
3 forallneuronsoftheminimumtypeatstagekset
i
R
k
qm
;
4 forallneuronsofthemaximumtypeatstagekset
j
T
k
qM
;
5 IFk:=Nthank=k+1goreturntoposition3,inanother
casecontinue
6 set:k=1,j=1,i=1;
7 calculationof
,
j
ii i
CT NkRT
kk kk
qm qm f qmqM andsetl=1;
8 IF
1
li
RT
kk
qM qm
thanset
1
,1
li
kk
WM m
andcalculate
1
min , , ;
j
iil
CR G
kkkk
ym q m M yM
00
Nk
j
j
kR
G
uM qM
9 IF
1k
llastM
thanl=l+1andreturnto8
ElseIF
k
llastm
thani=i+1andreturnto6
ElseIF
k
jlastM
thanj=j+1andreturnto6
ElseIF
kN
thank=k+1andreturnto6
Elsego10
10sett=0andj=1and
0
j
Nt
yM
;
11calculate
i
max ,
jj
i
Nt
Nt Nt
yM ym yM ;
12IF
Nt
ilastm
thani=i+1andreturnto11
ElseIF
Nt
jlastM
thanj=j+1andreturnto10
ElseIF
tN
thant=t+1returnto10
Elsego13;
13determiningthebestsolution(way).
End
________________________________________________
3.2 Theresultsobtainedfromtheproposedmethod
Inordertochecktheadequacyoftheproposedmodel
andthecorrectoperationofthealgorithm,anumber
of tests were carried out. This section presents the
followingthreenavigationsituations.
1. Situation when the ships are heading straight
towardseach
other.
In Figure 1A, the algorithm determined the anti‐
collision manoeuvre that allows the object to be
passedontheportside.Thechoiceismainlydue
to the navigatorsʹ subjectivity coefficients, which
take into following the recommendations of the
COLREG(InternationalRegulationsforPreventing
Collisionsat Sea) [21]–[24].Then, when
the ships
areatanequalaltitude,thecourseisleveled,and
286
after moving away from the foreign object, they
return to the original trajectory. Both trajectories
weredeterminedcorrectlyandinaccordancewith
COLREG. This is not a complicated collision
situation, therefore the determined trajectories
differslightly.However, theinfluenceofweather
conditions on determining the safe anti‐collision
trajectorycan
beobserved,Figure1B.
Theownshipcoordinatesareasfollows:position
(x,y)(0,0);Course45.0
0
;speed10.0kn.
Table1.Thecoordinatesoftheobjectinthecaseofships
headingstraighttowardseachother
________________________________________________
ob. Nj Dj Ψ V TCPA CPA μR
[°] [nm][°] [kn] [min] [nm]
________________________________________________
1 45.0 4.0 225.0 7.0 14.12 0.00 0.997
________________________________________________
Figure1. Comparison of trajectories with good and bad
visibilityinthecaseofshipsheadingstraighttowardseach
other,A‐fgoodvisibility,B‐limitedvisibility.
2. Situationwhencrossingcoursesatrightangles,the
objectisonthestarboardside.
From Figure 2 it can be concluded that the
algorithm determined the optimal trajectory by
changingtheheadingto59.0°.Whentheownship
isatthecrossoveraltitude,itreturnstoitsoriginal
course. From
the report it can be read that the
smallest possible approach of ships is 0.66 nm.
Thus, the safety condition was met, additionally,
theshipspassedeachotherinaccordancewiththe
rightofsearoute.
Theownshipcoordinatesareasfollows:position
(x,y)(0,0);Course45.0
0
;speed10.0kn.
Table2.Thecoordinatesoftheobjectinthecaseofcrossing
coursesatrightangles,theobjectonthestarboardside
________________________________________________
ob. Nj Dj Ψ V TCPA CPA μR
[°] [nm][°] [kn] [min] [nm]
________________________________________________
1 90.0 6.0 315.0 10.0 25.46 0.0 0.9976
________________________________________________
3. Situationwhencrossingcoursesatrightangles,the
objectisonthestarboardside.
Suchasituationinwhichnineobjectsparticipated
in it, three of which affect the safety of the own
shipʹsvoyage.Three ofthemstand still whilethe
restmoveindifferentdirections.ObjectNo.
2isa
direct threat‐the value of the collision risk
functionis0.5973.Otherdangerousobjectshavea
muchlower riskand practically do notaffect the
trajectoryoftheownship.
BasedonFigure3,itcanbeseenthattheownship
changed the original trajectory due
tothe second
object with which the collision was most likely.
Then the situation was so safe that the own ship
returned to its original course and stayed on it
untilthelaststage.
Theownshipcoordinatesareasfollows:position
(x,y)(0,0);Course160.0
0
;speed10.0knandgood
visibility.
Table3.Coordinatesofobjectswhenpassing9objects
________________________________________________
ob. Nj Dj Ψ V TCPA CPA μR
[°] [nm] [°] [kn] [min] [nm]
________________________________________________
1 90.0 1.600 85.0 12.7‐5.17 1.11 0.0200
2 204.0 3.600 40.0 12.0 11.86 0.55 0.5973
3 190.0 3.000 86.0 12.8 10.74 1.81 0.0035
4 231.0 5.000 0.0 0.0 11.23 4.73 0.0000
5 264.0 5.500 0.0 0.0 ‐ 9.18 5.34 0.0000
6 216.0 4.700 0.0 0.0 18.13 3.90 0.0000
7 288.0 3.600
45.0 8.8 ‐ 1.35 3.58 0.0000
8 176.0 2.600 46.0 9.6 9.62 0.84 0.1057
9 203.0 5.900 40.0 7.2 24.66 1.64 0.0099
________________________________________________
Figure2. The optimal trajectory for avoiding a dangerous
objectthatcrossestheshipʹscourseatrightangles.
Figure3.Theshipoptimalsafetrajectoryinpassingwith9
objects.
For the situation shown in Figure 1A, Table 4
presents a comparison of four algorithms for
determining the optimal safe ship in a collision
situation in a fuzzy environment. These algorithms
287
are: Branchand bound (BB); Dynamic Programming
(DP);EvolutionaryAlgorithm(EA);NeuralNetworks
(NN). After a brief analysis based on the results
included,itcanbeseenthat:
for the NN algorithm, the calculation time
(simulation duration) is much shorter than for
othermethodsandamountsto1.2s,
minimalspeedchangeonlyinonecasefortheNN
algorithm,
manoeuvre speed is the fastest for the EA
algorithm.
Table4.Comparisonoftheresultsoffouralgorithmsfor
solvingthesituationinFigure1A
________________________________________________
BBDPEANN
________________________________________________
P. C S C S C S C S
[°] [kn] [°] [kn] [°] [kn] [°] [kn]
________________________________________________
1 59 10 67 10 75 10 60 10.3
2 59 10 59.6 10 64 10 60 10.3
3 59 10 56 10 59.1 10 60 10.3
4 59 10 53.7 10 56 10 45 10
5 59 10 51.2 10 53.8 10 45 10
6 31 10 45 10 52.6
10 45 10
7 31 10 30 10 45 10 45 10
8 31 10 42 10 45 10 30 10
9 31 10 44 10 45 10 30 10
10 45 10 45 10 45 10 30 10
________________________________________________
C–Course;S–Speed
4 CONCLUSIONS
This study was to show the possibilities of
formulatingamodeloftheprocessofsafeshipcontrol
inablurredenvironmentandsolvingitwiththeuse
ofartificialintelligence.Thealgorithmisabletosolve
much more complicated situations, but sometimes
theremaybemanoeuvresinwhich
thepathlossesare
large. For this reason, the algorithm needs to be
refined towards a more precise definition of the
membership function of the set of constraints. The
advantageofthealgorithmisthattheshipʹstrajectory
returnstoitsoriginalcourseifthereisnolongerany
danger.Inaddition,theresultofthepresentedneural
networkisselectedfrommanyconnections,thanksto
whichthedeterminedtrajectoryisthebestpossibleto
obtain in a given situation. To sum up, the created
algorithmcan be usedas a decision support tool by
thenavigatorinorderto
maintainsafeseanavigation.
The obtained simulation results are promising and
showthegreatpotentialofthealgorithm.
REFERENCES
[1]D.C. Donderi, R. Mercer, M. B. Hong, and D.Skinner,
‘Simulated navigation performance with marine
electronic chart and information display systems
(ECDIS)’,J.Navig.,vol.57,no.2,pp.189–202,May2004,
doi:10.1017/S0373463304002668.
[2]A.WeintritandK.Stawicki,‘OperationalRequirements
for Electronic Chart Display and Information Systems
(ecdis). Risk of Overreliance on Ecdis’, Transp. Probl.,
vol. 3, no. 2, pp. 75–82, 2008, Accessed: Apr. 26, 2023.
[Online]. Available:
https://www.webofscience.com/wos/woscc/summary/78
1e4681‐1996‐48ac‐bba3‐e9c24d9d4f0e‐
85823306/relevance/1
[3]S. Zuskin, D. Brcic, and D. Sabalja, ‘A Contribution to
Improving the Standards of ECDIS Training’, Pomor.‐
Sci.J.Marit.Res.,vol.
27,no.1,pp.131–148,Jun.2013,
Accessed: Apr. 26, 2023. [Online]. Available:
https://www.webofscience.com/wos/woscc/summary/78
1e4681‐1996‐48ac‐bba3‐e9c24d9d4f0e‐
85823306/relevance/1
[4]T.Abramowicz‐GerigkandJ.Jachowski,‘ShipBerthing
and Unberthing Monitoring System in the Ferry
Terminal’, Sensors, vol. 22, no. 23, p. 9133, Dec. 2022,
doi:10.3390/s22239133.
[5]J. Lisowski,
‘Sensitivity of Safe Trajectory in a Game
Environment on Inaccuracy of Radar Data in
AutonomousNavigation’,Sensors,vol.19,no.8,p.1816,
Apr.2019,doi:10.3390/s19081816.
[6]A. Lazarowska, ‘Comparison of discrete artificial
potential field algorithm and wave‐front algorithm for
autonomousshiptrajectoryplanning’,IEEEAccess,vol.
8,pp.
221013–221026,2020.
[7]R.ZacconeandM.Martelli,‘Arandomsamplingbased
algorithm for ship path planning with obstacles’, in
Proceedings of the International Ship Control Systems
Symposium(iSCSS),2018,p.4.
[8]R.Śmierzchalski and A. Witkowska, ‘Advanced Ship
Control Methods’,in AutomaticControl, Robotics, and
InformationProcessing,Springer,
2021,pp.617–643.
[9]S.Ni,Z.Liu,Y.Cai,andX.Wang,‘Modellingofship’s
trajectory planning in collision situations by hybrid
geneticalgorithm’,Pol.Marit.Res.,vol.25,no.3(99),pp.
14–25,2018.
[10]R. Szlapczynski and H. Ghaemi, ‘Framework of an
Evolutionary Multi‐Objective Optimisation Method for
Planning a Safe Trajectory for a Marine Autonomous
SurfaceShip’,Pol.Marit.Res.,vol.26,no.4,pp.69–79,
2019,doi:10.2478/pomr‐2019‐0068.
[11]Z. Pietrzykowski et al., ‘The autonomous navigation
system of a sea‐going vessel’, Ocean Eng., vol. 261, p.
112104,Oct.2022,doi:10.1016/j.oceaneng.2022.112104.
[12]A.
Weintrit and T. Neumann, Advances in Marine
NavigationandSafetyofSeaTransportation.CRCPress,
2019.
[13]K.Janczyk,J.Rumiński,T.Neumann,N.Głowacka,and
P. Wiśniewski, ‘Age Prediction from Low Resolution,
Dual‐EnergyX‐rayImagesUsingConvolutionalNeural
Networks’, Appl. Sci., vol. 12, no. 13,
Art. no. 13, Jan.
2022,doi:10.3390/app12136608.
[14]M. Rybczak and W. Gierusz, ‘Maritime Autonomous
Surface Ships in Use with LMI and Overriding
TrajectoryController’,Appl.Sci.‐Basel,vol.12,no.19,p.
9927,Oct.2022,doi:10.3390/app12199927.
[15]R. Szlapczynski and J. Szlapczynska, ‘A ship domain‐
basedmodelof
collisionriskfornear‐missdetectionand
CollisionAlertSystems’,Reliab.Eng.Syst.Saf.,vol.214,
p.107766,Oct.2021,doi:10.1016/j.ress.2021.107766.
[16]M. Mohamed‐Seghir, K. Kula, and A. Kouzou,
‘Artificial Intelligence‐Based Methods for Decision
SupporttoAvoidCollisionsatSea’,Electronics,vol.10,
no. 19, p. 2360, Oct.
2021, doi:
10.3390/electronics10192360.
[17]J.Lisowski,phamTiep,andM.Mohamed‐Seghir,‘Safe
ship automatic control taking into consideration fuzzy
propertiesoftheprocess’,Pol.Marit.Res.,vol.1,pp.25–
32,Jan.1994.
[18]P. Borkowski, Z. Pietrzykowski, and J. Magaj, ‘The
AlgorithmofDetermininganAnti‐CollisionManoeuvre
Trajectory
Based on the Interpolation of Ship’s State
Vector’,Sensors,vol.21,no.16,p.5332,Aug.2021,doi:
10.3390/s21165332.
[19]Y.‐Y. Chen, M.‐Z. Ellis‐Tiew, W.‐C. Chen, and C.‐Z.
Wang,‘FuzzyRiskEvaluationandCollisionAvoidance
ControlofUnmannedSurfaceVessels’,Appl.Sci.‐Basel,
vol. 11,
no. 14, p. 6338, Jul. 2021, doi:
10.3390/app11146338.
[20]R. F. Romero, J. Kacprzyk, and F. Gomide, ‘Neural
Network Based Algorithm for Dynamic System
Optimization’, Asian J. Control, vol. 3, no. 2, pp. 131–
142,2001,doi:10.1111/j.1934‐6093.2001.tb00052.x.
288
[21]Y. Cho, J. Han, and J. Kim, ‘Efficient COLREG‐
compliant collision avoidance in multi‐ship encounter
situations’,IEEETrans.Intell.Transp.Syst.,2020.
[22]F. Deng, L. Jin, X. Hou, L. Wang, B. Li, and H. Yang,
‘COLREGs:Compliant Dynamic ObstacleAvoidanceof
USVsBasedontheDynamicNavigationShip
Domain’,
J. Mar. Sci. Eng., vol. 9, no. 8, p. 837, Aug. 2021, doi:
10.3390/jmse9080837.
[23]L. Li, D. Wu, Y. Huang, and Z.‐M. Yuan, ‘A path
planning strategy unified with a COLREGS collision
avoidance function based on deep reinforcement
learningandartificialpotentialfield’,Appl.OceanRes.,
vol.
113,p.102759,2021.
[24]J. Ning, H. Chen, T. Li, W. Li, and C. Li, ‘COLREGs‐
Compliant Unmanned Surface Vehicles Collision
Avoidance Based on Multi‐Objective Genetic
Algorithm’,IeeeAccess,vol.8,pp.190367–190377,2020,
doi:10.1109/ACCESS.2020.3030262.
