333
1 INTRODUCTION
1.1 Humansupervisioninautonomouscollision
avoidance
The last decade has shown an increasing interest in
research and development efforts towards use of
autonomy in the maritime industry. The purpose of
increasedautomationisdiverse,butimprovementsin
cost,efficiencyandsafetyforsharp‐endpersonnelare
major
drivers[1]–[3]. YaraBirkeland,andtheASKO
barges are examples of the ambition of the industry
whenitcomestotheapplicationofhighlyautomated
functions to support and/or substitute onboard
personnel[4],[5].Inthisdevelopment,remote‐control
centres are foreseen to play a role from where
operatorscan
performoversightofautonomousships
and can make critical decisions with regards to the
operationsoftheship[6].
The purpose of remote‐control centres is to
provideshore‐sidesupportforautonomousships,to
be compliant with current regulations on minimum
safe manning, and to provide an equivalent level of
safety (or better) compared to conventional ship
operations [7], [8]. The idea is that from a remote‐
control position operators can supervise the ship’s
operationsandmonitor,assist,andtakeoverfromthe
autonomoussystemswhenthecircumstancesrequire
this. In this case, it is assumed that humans can
perceive
andunderstand theinformationconcerning
the ship under supervision such that adequate
situationawarenesscanbeattainedandmaintained.
Akeychallengetoberesolvedinmovingtowards
autonomous,and potentially unmanned, shipping is
howunforeseen circumstances,suchascollision and
grounding situations, are handled without the
presence of navigators onboard
the ship [9]–[12].At
present, navigators determine collision risk and
Operationalising Automation Transparency for
Maritime Collision Avoidance
K.vandeMerwe
1,2
,S.Mallam
2,3
,Ø.Engelhardtsen
1
&S.Nazir
2
1
DNV,Høvik,Norway
2
UniversityofSouth‐EasternNorway,Borre,Norway
3
MemorialUniversityofNewfoundland,St.Johnʹs,Canada
ABSTRACT: Automation transparency is a means to provide understandability and predictability of
autonomoussystemsbydisclosingwhatthesystemiscurrentlydoing,whyitisdoingit,andwhatitwilldo
next. To support human supervision of autonomous collision avoidance systems,
insight into the system’s
internalreasoningisanimportantprerequisite.However,thereislimitedknowledgeregardingtransparencyin
thisdomainanditsrelationshiptohumansupervisoryperformance.Therefore,thispaperaimstoinvestigate
howaninformationprocessingmodelandacognitivetaskanalysiscouldbeusedtodrivethedevelopmentof
transparencyconcepts.Also,realistictrafficsituations,reflectingthevariationincollisiontypeandcontextthat
canoccurinreal‐life,weredevelopedtoempiricallyevaluatetheseconcepts.Together,theseactivitiesprovide
the groundwork for exploring the relation between transparency and human performance variables in the
autonomousmaritimecontext.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 17
Number 2
June 2023
DOI:10.12716/1001.17.02.09
334
perform relevant avoidance manoeuvres supported
by a range of systems, e.g., radar, AIS, and ECDIS.
Also, collision and grounding avoidance requires
knowledge,skills,andexperiencetobeperformedin
accordance with the collision regulations. When this
task is performed by an autonomous Artificial
Intelligence‐powered collision avoidance system,
adequate and
sufficient contextual information is
essential to support human oversight (see Figure 1)
[13].
Figure1.Conceptualizationofcontrolinconventional‐and
supervisedcollisionavoidance.
Anearlierstudyledby thefirstauthoridentified
the information required to supervise the
performance of an autonomous collision avoidance
systemthroughamappingandassessmentofrelevant
cognitive tasks [12], [14]. This study concluded that
adequatelysupervisingan autonomouscollisionand
groundingavoidancesystemrequiresinsightintothe
system’s
information processing to understand its
decisions and actions. Based on the knowledge that
human supervision of automated functions has
challenges in terms of human performance, keeping
humansintheloop,orrather“ontheloop”,becomes
an essential design requirement [15], [16]. Thus,
providingsufficientinformationabouttheautomated
system’s
reasoningprocesshasbeenproposedasone
oftheelementsthatcouldsupporthumansinsucha
role. In other words, by disclosing the system’s
internaldecision‐makingprocesstoitssupervisor,the
systemismadetransparentwithregardstoitsintent,
performance,futureplans,andreasoningprocess[17].
Automation transparency
is concerned with
making the inner reasoning of systems observable,
such that its actions are understandable and
predictable [15], [18], [19]. Therefore, transparency
shouldmakeit clearto humansupervisorswhat the
systemiscurrentlydoing,whyitisdoingit,andwhat
itwilldonext[15].Earlierreviews
haveindicatedthat
transparency has a promising effect on human
performance and situation awareness [20]–[22].
However, there is limited knowledge regarding
transparency in the maritime domain, especially in
relation to autonomous collision and grounding
avoidance. To this end, further work is needed to
investigate the role of transparency in supervised
autonomous
shipping andto exploreits relationship
withhumanperformanceinthiscontext.
This paper discusses ongoing work towards
performinganempiricalevaluationtostudydiffering
levelsandtypesoftransparencyconceptsinarealistic
traffic collision avoidance setting. An empirical
evaluation is planned in which participants take the
role of
a supervisor of an autonomous collision
avoidance system. An approach is used in which
participants are tasked with evaluating traffic
situations for their understandability, whilst being
measured on human performance variables. The
purpose of this evaluation is to better understand
which levels and types of transparency information
supporthumansupervisorsand
howthisknowledge
can be applied to a dynamic collision avoidance
context.Thispaperdescribesthegroundworkforthis
evaluationbydescribingthesystematicdevelopment
process behind the traffic situations, as well as the
levelsandtypesoftransparencyconceptsdeveloped
forthis.
2 DEVELOPINGTRAFFICSITUATIONS
2.1 Definingcriteria
toensurevariation
To provide participants of the planned empirical
evaluation with realistic conflicts, traffic situations
were developed that reflected the variation in
collisiontypeandcontextthatmayoccurinreal‐life.
Also, to avoid familiarisation with the traffic
situations, and thereby unintentionally influencing
the results of the evaluation, multiple
variants of
traffic situations were developed based on a set of
criteria(seeTable1).
Table1.Criteriaforestablishingavariedsetoftraffic
situations.
________________________________________________
CriterionVariation
________________________________________________
Complexityavoidance Low‐Nolimitations
manoeuvreownship High‐Limitationsmanoeuvre
CollisiontypeCR‐Crossing
HO‐Head‐on
OT‐Overtaking/overtaken
NC‐Nocollision
AvoidanceactionsownshipGive‐way
Stand‐on
Restrictionstarget Norestrictions
Restrictedinmanoeuvrability
TrafficdensityFewothershipandobjects
Manyotherships
andobjects
GeographyLand
Openwater
________________________________________________
Variabilitywasensuredthroughdifferinglevelsof
complexity, collision types, the avoidance actions of
own ship, restrictionsto target ships, traffic density,
andgeography.First,inhighcomplexsituations,own
ship was restricted in its avoidance manoeuvring
abilitycomparedtolowcomplexsituations.Thatis,in
low complexity situations, own ship
was free to
manoeuvre in any direction to avoid a collision,
whereas in high complexity situations, there were
obstacles prohibiting own ship to perform certain
manoeuvres. Second, for collision type, traffic
situations consisted of crossing‐, head‐on‐,
overtaking‐/overtaken situations. Also, situations
were developed in which no collision
was present.
Third, for avoidance actions, situations were
developed for which own ship was the give‐way
vessel or the stand‐on vessel. Fourth, for some
situations, target ships were restricted in their
manoeuvrability,e.g.,becauseofongoingbunkering.
335
Fifth, situations were developed withlow‐ and high
traffic densities. Finally, traffic situations were
developed in which contextual factors were varied
that were external to the traffic situations (i.e., land
formationsoropenwater).
To constrain the amount of variation and retain
controllability in the traffic situations, some
limitations were
set in terms of number of ships
posing a collision risk and the number of
simultaneous collision situations. That is, own ship
couldonlybeindirectconflictwithoneothershipfor
onecollisiontype(e.g.,notinacrossingandhead‐on
situation simultaneously), own ship could not
be in
both a give‐way and stand‐on situation
simultaneously,andownshipwasneverrestrictedin
its manoeuvrability. Also, although is it recognised
that grounding avoidance is an essential part of
collisionavoidance,thetrafficsituationsinthispaper
were limited to collision situations only. Finally,
externalfactorsthat
couldaffectthecollisionsituation
or own‐ship’s capabilities, such as weather or
technicalfailures,werenotincluded.
2.2 Developmentprocess
For each criterion in Table 1, two scenarios were
createdresultinginasetof70situations(seeTable2).
The traffic situations were created in a desktop
simulator
fromapopularequipmentmanufacturerby
a navy‐certified navigator with five years of
navigational experience. Upon creating an initial set
of traffic situations, a review was performed with
independentnavigators.
Table2.Thetrafficsituationscreatedbasedonthesetof
criteria.Key:HO=Head‐on,CR=Crossing,OT=
Overtaking/overtaken,NC=Nocollision,L=Low,H=
High,T=Total.*Note:inahead‐onsituationwithone
motorisedtargetshipandnootherexceptions,ownship
cannotbestand‐on.
________________________________________________
HO CROT T
Variant/Complexity L H L H L H
________________________________________________
Type(HO/CR/OT) 5 5 4 4 4 4 26
Type(NC)2 2 2 2 2 2 12
Ownshipstand‐on* 0 0 2 2 2 2 8
Restrictionstarget 2 2 2 2 2 2 12
Geography(land) 2 2 2 2 2 2 12
________________________________________________
Total11 11 12 12 12 12 70
________________________________________________
2.3 Verificationandvalidationworkshop
Thefinal verificationandvalidationwereperformed
with two independent navigators holding active
navigationallicenses(D1/D2),withanaverageof6.5
years of navigational experience (SD=2.1, min=5,
max=8). The reviewwas performed inthe formof a
1,5‐dayworkshop.
Intheworkshoptrafficsituationswereshownona
display and participants were asked to state if own
ship was in a collision situation, if yes, which type
(HO/CR/OT), and the avoidance action required by
own ship (give‐way/ stand‐on). In addition, three
questions were asked, using a 7‐point Likert scale,
probing the situation’s
realism, complexity, and
likelihood of occurrence. With these questions, a
comparison between the situation’s intended
depiction and the navigator’s perception was
obtained. Discrepancies were discussed and
suggestions for improving the design of the traffic
situationswerenoted.Afinalsetoftrafficsituations
were produced, incorporating the inputs from the
workshop(seeFigures2,3,4,and5forexamples).
Figure2.Ownshipisinahead‐onsituationinopenwater
where it is required to give‐way. The situation is of low
complexity as there are no restrictions to own ship’s
avoidancemanoeuvrability.
Figure 3. Own ship is in an overtaking situation in open
waterswhereitisrequiredtogive‐way.Thesituationisof
high complexity as there are restrictions to own ship’s
avoidancemanoeuvrability(bothportandstarboard).
Figure4.Ownshipisinacrossingsituationinopenwaters
where it is required to stand ‐on. The situation is of high
complexity as there are restrictions to the target ship’s
avoidancemanoeuvrability(theshipcrossingatportside).
336
Figure5.Ownshipisinacrossingsituationinopenwaters
where it is required to give‐way. The situation is of high
complexity as there are restrictions to the target ship’s
avoidancemanoeuvrability(abuoy).
3 DEVELOPINGTRANSPARENCYFOR
COLLISIONAVOIDANCE
3.1 Definingtransparencylayers
Anearlierstudy ledby the firstauthor performeda
cognitive task analysis to identify the information
required to perform supervision of a collision
avoidancesystem[12],[14].Theanalysisdescribesthe
information pertaining to the supervisory task and
depicts which
information should be disclosed to
humansupervisorstomaketheinternalreasoningof
thecollisionavoidance system observable.However,
the analysisonly describes what informationshould
bemadeavailableanditdoesnotdictatewhichtype,
orhowmuchoftheidentifiedinformationshouldbe
disclosed. Simply depicting all information
simultaneously will likely put too large a cognitive
burden on the supervisor’s information processing
capabilities,resultinginhighmentalworkload.Atthe
same time, only limiting the information from the
system to single information elements may not
provide the full picture about the system’s internal
reasoningeither.Inaddition,consideringthe
dynamic
natureofthecollisionavoidancetask,theinformation
neededtoeffectivelysupervise thesystem mayvary
giventhecircumstancesandthetaskanalysisdoesnot
definewhich informationshould be disclosedwhen.
As such, providing transparency to supervisors
means making choices as to which information is
madeavailableto
allowsupervisorstounderstandthe
system’sbehaviour.
The rationale for specifying what constitutes
transparency information in a collision avoidance
context, together with how this information can be
categorised into distinct information types is
discussedina separate study [23].Inbrief, asimple
information processing model was used (see Figure
6),
consistingof information acquisition,information
analysis, decision selection, and action
implementationstages,toidentifyandcategorisethe
informationintodiscretesteps[24].Assuch,alayered
approach to transparency was used allowing
supervisors to observe the different facets of the
system’s inputparameters, reasoning,decisions, and
actionspertainingtothe
collisionsituation.
Information
acquisition
Information
analysis
Decision
selection
Action
implementation
Figure6.Asimplemodelofhumaninformationprocessing
adoptedfrom[24].
This model provides, at minimum, a means to
organize the information describing the system’s
information processing into several distinct parts.
However,themodeldoesnotprovideguidanceasto
which information takes priority over the other. A
potentialstartingpointistotryanswerthequestionof
whatinformationsupervisorswould
liketoknowata
minimum, before adding layers of transparency to
allow for increased understandability. A plausible
means for human supervisors to obtain an
understanding of the collision avoidance system’s
performanceistobeinformedwhetherthesystemcan
avoid a potential collision at all. In other words,
supervisors
likely need to be informed about the
system’sdecisionsandactionsfirst,beforeneedingto
“digdeeper”intothesystem’sunderlyingreasoning.
This indicates that the starting point for providing
transparency to supervisors is thus the “decision
selection” step of the information processing model
depicted in Figure 6 and not the
“information
acquisition” step. (Note that in the “action
implementation” step there is no information
processing,onlyexecution.)Furtherunderstandingof
howand whythesystemhas derivedatits decision
andplannedactionscansubsequentlybeobtainedby
“going backwards” through the model. That is, the
“information analysis” stage of
the model provides
the relevant information pertaining to the analysis
that underlie the system’s decisions and actions.
Finally, when the full picture is required for
understandingthesystem’sdecisionsandactions,the
“informationacquisition”stageofthemodelprovides
allthe input data the system usesin its information
processing.
3.2
Developmentprocess
A concept illustration is provided of a radar screen
depictingatrafficsituationinwhichownship,inthe
centre of the radar screen, is involved in a head‐on
situation(seeFigure7).Ownshipdepictsitsintended
avoidance manoeuvre by drawing its planned track
for the
next three manoeuvring steps (each step
corresponds to one vector length and equals six
minutes).Italsostates“GW”indicatingitintendsto
give‐way. Additional information about current and
nextactions,includingspeed,aredepictedontheleft
side of the figure. With this information, minimum
transparency is provided to
allow supervisors to
understand that the system isabout to initiate a 12‐
degreestarboardturnandthatitintendstogive‐way.
The informationprovided inFigure 7 was proposed
as the minimum information needed to obtain an
understanding of the own ship’s decisions and
actions.
337
Figure7. Traffic situation with transparency information
overlaid(decisionselection).
Figure 8 depicts that own ship considers two
targetsas especially relevant in this traffic situation.
Thetargetshipinredisdepictedasthehighestriskas
this ship is the one considered to be on collision
course with own ship (minimum predicted CPA
exceeded).Thetargetinorangeis
alsohighlightedas
own ship has considered this target to be of
importanceduringtheavoidancemanoeuvre.Further
information regarding the targets that own ship
considers is provided through the indicatorsnext to
the targets depicting the conflict situation (e.g., HO
for head‐on, and MV for motor vessel). In
addition,
furtherinformationregardingthesystem’sreasoning
is provided through a manoeuvrability indicator
aroundownshipindicatingwhereitcanmanoeuvre
withinonevectorlength.Finally,tablestotheleftof
theradarscreen depictadditionaltarget information
and the variables own ship has considered in
determiningsafespeed.
Figure8. Traffic situation with transparency information
overlaid(decisionselection+informationanalysis).
Figure 9 provides a depiction of what a
transparentcollisionavoidancesystemcouldlooklike
when all transparency information described in the
task analysis is provided. Here, all targets have
received identifiers (green circles), and initial
classifications (ship types and relevant conflict type
indicators). In addition, information regarding the
status of
the system’s sensors are provided in the
tablestotheleftoftheradarscreen.
Figure9. Traffic situation with transparency information
overlaid (decision selection + information analysis +
informationacquisition).
3.3 Verificationandvalidationworkshop
Thetransparencyconceptsweredevelopedthrougha
seriesofiterationsbasedontheinformationfromthe
task analysis and the information processing model.
Finalverificationandvalidationoftheinterfaceswas
performed in a second workshop with two
independent navigators holding active navigational
licenses with an
average of 12 years of navigational
experience(SD=9.9,min=5,max=19).
The purpose of this second workshop was to
evaluate a selected set of traffic situations that
includedthetransparencylayersasdescribedabove.
Arepresentativesubsetoffivetrafficsituationswere
includedforreviewinthisworkshop,includinghead‐
on,
crossingwithownshipasstand‐on,overtakenby
ashiprestrictedinitsmanoeuvrability,crossingwith
speed‐only as the avoidance manoeuvre and
overtaking a slower ship when approaching a
harbour. A talk aloud protocol was used where
participants were asked to describe their
interpretation of the traffic situation with
primary
focusontheinformationthesystemprovidedthrough
theHumanMachineInterface(HMI).Inotherwords,
thefocusintheworkshopwasonhowtheyperceived
the collision avoidance system would solve the
conflict situation, and not how they would solve it.
The independent navigator’s interpretations were
noted, including
all comments related to
recommendations for improvement, corrections, and
additions which were included in the final
transparencyiteration.
4 SUMMARYANDFURTHERWORK
Whenacollisionsituationoccursthatrequireshuman
intervention,thecollisionavoidancesystemneedsto
facilitatehuman supervisorsingainingSA suchthat
successful decisions can be made.
This paper
described the systematic development of a realistic
and validated foundation for evaluating the
relationship between automation transparency and
human supervisory performance in an autonomous
collision avoidance context. First, a set of traffic
situations were developed based on navigational
experience aimed at capturing the variability
encountered in real‐life
situations. Second, a set of
338
transparency concepts were developed based on a
cognitive task analysis and a model for human
informationprocessing. Together, these preparations
provide the groundwork for the planned empirical
worktoexplorethisrelationship.
Asthemaritimeindustrymovestowardsincreased
useof automation, includingdeployingsystemsthat
can perform (part of)
the collision and grounding
avoidance functions, there is an urgent need to
understand how humans will interact with these
systems.Automationtransparencyhasbeenproposed
as a critical element that can support human
supervisors in obtaining situation awareness of the
system’s behaviours and actions [16]. Conversely,
without transparency, i.e., systems
that have low
degrees of observability and predictability, humans
willbehighlychallengedinunderstandingwhatthe
systemisdoing,whyitisdoingit,andwhatitwilldo
next. As such, given the critical nature of the
supervisory task for autonomous maritime collision
andgroundingavoidancesystems,itis
pertinentthat
furtherunderstanding isneeded with regards tothe
applicationofthetransparencyinthisdomain.
This paper aimed to address this need by
investigating how an information processing model
could be used to drive the development of
transparency layers. Given the dynamic nature of
collision and grounding avoidance
the amount and
typeofinformationneededtounderstandthesystem
may depend on the type of situation, the degree of
human oversight,the complexityofthe situation, or
the time available to intervene. The transparency
concepts discussed in this paper have attempted to
address this. In addition, an empirical evaluation
is
underway in which the relationship between
automation transparency and human performance
variables are evaluated in a collision avoidance
context.Thisway,therelationbetweentransparency
and human performance variables can be explored,
anditspracticalbenefitscanbeassessed.
ACKNOWLEDGEMENTS
The authors would like to express their gratitude to the
navigatorsfortheirparticipationintheworkshops.Also,we
would like to express our sincere gratitude to Koen
Houweling for his contribution in developing the traffic
situationsandthetransparencyillustrations.
REFERENCES
[1]L. Kretschmann, H. C. Burmeister, and C. Jahn,
“Analyzing the economic benefit of unmanned
autonomous ships: An exploratory cost‐comparison
between an autonomous and a conventional bulk
carrier,” Research in Transportation Business and
Management, vol. 25, pp. 76–86, 2017, doi:
10.1016/j.rtbm.2017.06.002.
[2]I. Kurt and M. Aymelek, “Operational and economic
advantages of autonomous ships and their perceived
impactsonportoperations,”MaritEconLogist,vol.24,
no. 2, pp. 302–326, Jun. 2022, doi: 10.1057/s41278‐022‐
00213‐1.
[3]K. Wróbel, J. Montewka, and P. Kujala, “Towards the
assessmentofpotentialimpactofunmannedvesselson
maritimetransportationsafety,”ReliabEng
SystSaf,vol.
165,pp.155–169,2017,doi:10.1016/j.ress.2017.03.029.
[4]Kongsberg, “Kongsberg maritime and Massterly to
equip and operate two zero‐emission autonomous
vessels for ASKO,” Sep. 01, 2020.
https://www.kongsberg.com/maritime/about‐us/news‐
and‐media/news‐archive/2020/zero‐emission‐
autonomous‐vessels/(accessedNov.18,2020).
[5]Yara International, “Yara Birkeland,” 2021.
https://www.yara.com/news‐and‐media/media‐
library/press
‐kits/yara‐birkeland‐press‐kit/(accessedJan.
02,2023).
[6]DNV, “DNVGL‐CG‐0264: Autonomous and remotely
operated ships.” 2018. [Online]. Available:
http://rules.dnvgl.com/docs/pdf/dnvgl/cg/2018‐
09/dnvgl‐cg‐0264.pdf
[7]IMO,ResolutionA.1047(27)Principles ofsafe manning.
2011.
[8]IMO, “Guidelines for the approval of alternatives and
equivalents as provided for in various IMO
instruments,”MSC.1/Circ.1455,Jun.2013.
[9]K.Aylward,R.Weber,M.Lundh,S.N.MacKinnon,and
J.Dahlman,“Navigators’viewsofacollisionavoidance
decision support system for maritime navigation,” J.
Navigation, pp. 1–14, Sep. 2022, doi:
10.1017/S0373463322000510.
[10]E. Hannaford, P. Maes, and E. Van Hassel,
“Autonomous ships and the
collision avoidance
regulations: a licensed deck officer survey,” WMU J
MaritAffairs,vol.21,no.2,pp.233–266,Jun.2022,doi:
10.1007/s13437‐022‐00269‐z.
[11]M. A. Ramos, I. B. Utne, and A. Mosleh, “Collision
avoidance on maritime autonomous surface ships:
Operators’tasksandhumanfailureevents,”SafSci,vol.
116,pp.33–44,2019,doi:10.1016/j.ssci.2019.02.038.
[12]K.vandeMerwe,S.C.Mallam,Ø.Engelhardtsen,and
S. Nazir, “Exploring navigator roles and tasks in
transitioning towards supervisory control of
autonomous collision avoidance systems,” J. Phys.:
Conf. Ser., vol. 2311, no. 1, p. 012017, Jul. 2022, doi:
https://doi.org/10.1088/1742‐6596/2311/1/012017.
[13]
S. N. Mackinnon, Y. Man, M. Lundh, and T. Porathe,
“Command and controlof unmanned vessels:Keeping
shore based operators in‐the‐loop,” 18th International
ConferenceonShipsandShippingResearch,NAV2015,
pp.612–619,2015.
[14]K.vandeMerwe,S.Mallam,Ø.Engelhardtsen,andS.
Nazir, “Supporting human supervision
in autonomous
collision avoidance through system transparency: a
structuredandsystematicapproach,”underreview.
[15]M. R. Endsley, “From Here to Autonomy: Lessons
Learned from Human‐Automation Research,” Hum
Factors, vol. 59, no. 1, pp. 5–27, 2017, doi:
10.1177/0018720816681350.
[16]M. R. Endsley, “Supporting Human‐AI
Teams:Transparency, explainability, and situation
awareness,”ComputersinHumanBehavior,vol.140,p.
107574,Mar.2023,doi:10.1016/j.chb.2022.107574.
[17]J.Y.C.Chen,K.Procci,M.Boyce,J.Wright,A.Garcia,
and M. J. Barnes, “Situation Awareness‐Based Agent
Transparency,” U.S. Army Research Laboratory,
Aberdeen Proving Ground, ARL‐TR‐6905, Apr. 2014.
doi:10.21236/ADA600351.
[18]J.
Y. C. Chen, S. G. Lakhmani, K. Stowers, A. R.
Selkowitz, J. L. Wright, and M. J. Barnes, “Situation
awareness‐based agent transparency and human‐
autonomy teaming effectiveness,” Theor Issues Ergon
Sci, vol. 19, no. 3, pp. 259–282, May 2018, doi:
10.1080/1463922X.2017.1315750.
[19]M.R.Endsley,B.Bolté,andD.
G.Jones,Designingfor
situation awareness: an approach to user‐centered
design.London ;NewYork:Taylor&Francis,2003.
[20]A. Bhaskara, M. Skinner, and S. Loft, “Agent
Transparency: A Review of Current Theory and
339
Evidence,” IEEE Trans Hum Mach Syst, vol. 50, no. 3,
pp.215–224,Jun.2020,doi:10.1109/THMS.2020.2965529.
[21]F. Rajabiyazdi and G. A. Jamieson, “A Review of
Transparency (seeing‐into) Models,” in 2020 IEEE
International Conference on Systems, Man, and
Cybernetics (SMC), Oct. 2020, pp. 302–308. doi:
10.1109/SMC42975.2020.9282970.
[22]K. van de
Merwe, S. Mallam, and S. Nazir, “Agent
Transparency, Situation Awareness, Mental Workload,
and Operator Performance: A Systematic Literature
Review,” Hum Factors, p. 00187208221077804, Mar.
2022,doi:10.1177/00187208221077804.
[23]K.vandeMerwe,S.Mallam,Ø.Engelhardtsen,andS.
Nazir, “Supporting human supervisory performance
through information disclosure: establishing
transparency requirements for
maritime collision
avoidance,” in Proceedings of the Human Factors
SocietyAnnualMeeting,Orlando,FL,submitted.
[24]R.Parasuraman,T.B.Sheridan,andC.D.Wickens,“A
model for types and levels of human interaction with
automation,”IEEETransSystManCybern,vol.30,no.3,
pp.286–297,May2000,doi:10.1109/3468.844354.
