1095
1 INTRODUCTION
Improper lookouts by navigators have been identified
as a primary cause of frequent maritime accidents.
Failure to maintain a proper lookout has been
identified as a major cause of ship collisions, and is
considered a situational awareness error by navigators.
Grech et al. analyzed maritime accident reports and
found that 71% of human errors were related to SA [1].
SA is a critical factor in risk assessment. Endsley
proposed a situational awareness global assessment
technique (SAGAT) [2] to measure the SA of an aircraft
pilot in a cockpit. Okazaki proposed a method that
could be measured by the SAGAT to measure the SA
of a ship navigator's SA in pilot training using a ship
maneuvering simulator [3]. An analytical method for
maritime accident analysis that uses a bridge simulator
to identify the critical factors contributing to
navigational watchkeeping and to assess a navigator's
decision-making processes has been proposed [4], [5].
To reduce the number of maritime accidents,
researchers have developed ship support systems to
provide information for ship collision avoidance.
However, the effectiveness of such systems may vary
depending on the navigator's level of experience.
Specifically, the information provided may be
insufficient for less experienced navigators, whereas it
may be overly detailed for highly experienced
navigators, potentially affecting decision-making
efficiency. Ship collision risk modeling and risk
analysis have recently become the focus of several
studies [6]. One of the ship collision risk models is the
collision risk index (CRI), which is a computational tool
used to reflect the risk of collision based on dynamic
navigational factors such as the DCPA, TCPA, relative
distance, relative bearing, and velocity ratio between
A Study on a Situation Awareness Model for Navigators
in Congested Waters
T. Endoo, C. Nishizaki & O. Tadatsugi
University of Marine Science and Technology, Tokyo, Japan
ABSTRACT: In recent years, most maritime accidents have been caused by deficiencies in navigators’ situational
awareness. Previous studies have evaluated the navigators' situation awareness (SA) through the application of
the Situation Awareness Global Assessment Technique (SAGAT) in ship maneuvering simulations. Researchers
have developed collision avoidance support systems and collision risk assessment models to mitigate maritime
accidents. However, existing models often apply conventional weight parameters for collision risk factors, which
may not be appropriate for navigators with different experience levels. To refine these weight parameter sets
tailored to each navigator level, based on each navigator's significant SA, derived from experimental navigators'
situation awareness measurement. In addition, grid search-based weight aggregation was employed to
systematically refine the weight distributions, optimize the impact of collision risk factors, and ensure improved
model accuracy. The results demonstrate that the proposed weight parameters improve the detection rate of
significant targets according to navigator’s experience level in congested waters.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 19
Number 4
December 2025
DOI: 10.12716/1001.19.04.06
1096
two ships through fuzzy inference to evaluate the
probability and severity of collision risk with multiple
ships [6]. In addition, the SA model is a cognitive
process by which navigators perceive critical
environments and current risks, and project future
states to support effective decision-making [7].
Consequently, the collision risk and SA models are
interdependent, as an accurate SA enables navigators
to interpret and make decisions regarding collision
risks, while the collision risk model provides a real-
time risk indicator that supports the perception of risk
and projects the future state of the SA. By integrating
the SA model into the collision risk model, collision
risk recognition can be made more compatible with the
actual decision-making of navigators to improve the
applicability of CRI models across different navigator
experience levels.
However, the existing CRI models typically rely on
fixed-weight parameters for risk factors, which may
not align with the decision-making strategies of
navigators with different expertise levels. Thus, an
optimized weighting approach is required.
Traditionally, a grid search has been employed in
machine learning for hyperparameter tuning by
systematically exploring a predefined set of parameter
values to identify the optimal configuration [8]. In this
study, we adapted the grid search approach by
applying it to weight aggregation rather than to
hyperparameter tuning. This method systematically
evaluates different combinations of weight parameters
and identifies the optimal set that best aligns with the
SA patterns of the navigators. This adaptation ensures
that the collision risk model accurately reflects the
decision-making of navigators across various ranks,
thereby improving its effectiveness in real-world
maritime operations.
2 NAVIGATOR SITUATIONAL AWARENESS
This study focuses on ship navigation in congested
waters, where vessel density and collision risks are
significantly higher than those in the open sea. In
addition, navigators operating under such conditions
require advanced practical skills, knowledge,
situational awareness, and decision-making to
encounter a multiple-target ship scenario. To
systematically evaluate navigators' SA, decision-
making processes, and risk prioritization strategies
under congested water conditions. The SA
measurements and risk prioritization of the navigators
were conducted during the simulated experimental
scenario, and the adapted SAGAT was applied using a
radar-plotting chart. In addition, experiments were
conducted at the National Maritime Research Institute
in a bridge simulator, which contains general
navigational equipment such as radar, ECDIS,
compass, binoculars, and a steering control unit.
2.1 SAGAT
The SAGAT is an effective technique for measuring SA
in simulation experiments. The Originally developed
and proposed by Endsley to measure aircraft pilots’ SA
in simulator cockpits, and procedural standards for the
SAGAT were established [2]. An adapted SAGAT
modifies the original SAGAT measurement method
and procedural standard to measure the SA of a ship
navigator in a ship maneuvering bridge simulator [3].
During the experiment, the SAGAT was used to
measure the SA. The experiment was interrupted
immediately. In addition, the simulator screen
temporarily turned black while the subject answered
questions to measure the recognized situation. After
completion of the questionnaire, the simulations were
resumed. However, if the simulation is interrupted for
an excessive duration, experimental continuity
decreases. Therefore, interruption exceeding two
minutes were found to negatively affect experimental
continuity. However, it is difficult to keep the
questions answered within two minutes by oral
answers. Additionally, there are ambiguities in oral
answers regarding recognized targets in a congested
sea area, particularly expressing their positions [4].
In this study, the adapted SAGAT was used to
measure navigator SA. In addition, we adapted a
method in which subjects filled in the recognized
targets on the radar plotting chart. This method was
proposed as an evaluation method for pilot marine
trainees [9]. We set the interruption time to one minute
to enable the subjects to answer questions based on
radar-screen information using radar-plotting charts.
Figure 1 illustrates the radar plotting chart, in which
the subjects marked the recognized targets used in the
experiment.
Figure 1. Example of RADAR plotting chart for SAGAT.
2.2 Result of the previous study on SA.
In a previous study by Nishizaki [4], the SA and risk
priority measurements of navigators were conducted
using SAGAT in scenarios involving multiple target
ships in congested waters. A previous study used a
bridge simulator equipped with general navigational
equipment. In addition, four participants had
previously participated in a real-world ship in the
captain position. The results of the four captains' SA
from a previous study are presented in Table 1.
2.3 Comparison of evaluation risk area and experimental
results
Nakamura and Nishizaki proposed a method for
evaluating the safety of an automatic collision
avoidance system capable of evaluating risk areas such
1097
as danger, caution, attention, and safety [10], [4]. The
evaluation method aims to ensure compliance with the
maritime rules of the road (COLREGs), which contain
three encounter situations: crossing, head-on, and the
same way. In addition, the evaluation method was
based on the relationship between the rate of change in
the bearing and the relative distance. The evaluation
risk area method can potentially identify risk targets
and categorize them into risk categories such as
danger, caution, attention, and safety. In congested
waters, there are so many target ships that it is difficult
to recognize all those requiring attention. According to
the results of the captains' SA in a previous study
compared with the evaluation risk area method, as
illustrated in Table 1. In the table, the first column and
first row indicate the ship ID and SAGAT
measurement time, respectively, and the fourth to last
rows indicate the ships recognized by the subjects. The
total number of recognized ships at each measurement
time point is indicated in the third row. The lighter-
grey-highlighted cells indicate ships recognized by
more than half of the subjects and were significant
recognition targets in the study. Grey highlighted cells
represents ships captured by the evaluation risk area
method, and dark highlighted cells indicate ships
captured by more than half of the subjects and the
evaluation risk area method.
Table 1. Result of the captains' SA in the previous study [4].
2.4 Concept of the proposed method.
The recognized risk targets for navigators vary
depending on the level of the navigator. The capability
to recognize collision risk targets varies among
navigators at different levels with varying knowledge,
experience, and skills, as is evident from the results of
the captains' SA, which illustrates the target
recognition patterns in Table 1. In addition,
navigational safety relies on the recognition of
surrounding ships, particularly in multiple encounter
situations. Navigators are tasked with assessing
RADAR, automatic identification system (AIS) data,
and visualized lookouts. Humans have limitations in
recognizing collision risk targets in congested waters.
Consequently, the proposed concept gathers the SA of
the navigators from the previous study and the current
study, which were conducted at the captain’s level and
officer’s level, respectively, to optimize the collision
risk model. On the one hand, the collision risk model
and the evaluation risk area method have the potential
to evaluate the collision risk of surrounding target
ships. However, the evidence from the evaluation risk
area method, illustrated in Table 1, indicates that the
capability may not capture all significant target ships
that the subjects identified. Therefore, the proposed
concept involves weight optimization in the collision
risk model by adapted grid-search weight optimization
based on the gathered SAs of the navigators.
3 COLLISION RISK MODEL
3.1 Outline
The proposed method aims to enhance the CRI by
optimizing its weight parameters to identify the target
recognition more effectively based on each navigator's
SA result. In a previous study, Nishizaki proposed a
model of the evaluation risk area method, such as the
attention area, to evaluate the target ships that
navigators pay attention to through a simulated
experiment in congested water and situation
awareness measurements. The proposed method
contains three components. First, CRI is the main
formula used to evaluate target risk levels. The
collision risk index is formulated as fuzzy interference
using multiple navigational factors, such as DCPA,
TCPA, relative distance, relative bearing, and velocity
ratio [6]. Second, we adapted the grid search technique
to generate weight combinations to optimize the
importance of each factor using the recorded simulated
experimental data and navigator SA results. Third, the
optimal weight combination was placed into the
collision risk index model to assess the risk targets and
compare the identified risk targets with the navigator's
SA. In addition, the evaluation was based on
classification performance metrics to determine how
well the optimized weight combination related to the
navigator's significant targets.
3.2 Collision risk index model
The CRI was defined as the basis for decision-making
in collision avoidance to evaluate a target that has the
possibility and severity of collision risk. The CRI
evaluates the degree of collision risk by
comprehensively considering the DCPA, TCPA,
relative distance, relative bearing, and velocity ratio
between two ships [6]. As illustrated in Figure 2, the
DCPA and TCPA can be obtained through geometric
calculations of the ship encounter situation, where the
dashed line represents the extended line of the relative
velocity (VR). The position coordinates, velocity, and
course of the own ship are SO (xO,yO), VO, and φO. The
position coordinates, velocity, and course of the target
ship were ST (xT,yT), VT, and φT. Additionally, those of
DR, VR, and φR are the relative distance, velocity, and
course, respectively. aT is the azimuth of the target ship,
and θT is the relative bearing. The equations for DCPA,
TCPA, DR, φR, VR and θT are defined in Eqs. (3)-(10) [6].
( )
sin
R R T
DCPA D a
= − −
(1)
1098
( )
cos /
R R T R
TCPA D a V
= − −
(2)
( ) ( )
22
R T O T O
D x x y y= − + −
(3)
1
1
1
1
tan , 0, 0
tan 90 , 0, 0
tan 180 , 0, 0
tan 270 , 0, 0
Rx
Rx Ry
Ry
Rx
Rx Ry
Ry
R
Rx
Rx Ry
Ry
Rx
Rx Ry
Ry
V
VV
V
V
VV
V
V
VV
V
V
VV
V
−
−
−
−
+
=
+
+
(4)
Figure 2. Diagram of the two-ship encounter geometry [6].
Rx Tx Ox
Ry Ty Oy
V V V
V V V
=−
=−
(5)
sin sin
,
cos cos
Ox O O Tx T T
Oy O O Ty T T
V V V V
V V V V
==
==
(6)
22
R Rx Ry
V V V=+
(7)
T T O
a
=−
(8)
The crucial factors (DCPA, TCPA, DR, θT, and VT/VO)
are considerable for evaluating the collision risk index.
However, evaluating these factors is difficult because
the relationship between collision risk factors is
complex and ambiguous, and each factor has a
different impact on the evaluation of the risk of
collision depending on the encounter situation. Owing
to the complex relationship between collision risk
factors, fuzzy synthesis judgment is introduced to
express knowledge and experience with ambiguous
boundaries [11], [12]. Therefore, this study was
conducted to predict the CRI through fuzzy inference
according to various encounter situations using
membership functions defined as follows [6]:
1. The membership function of DCPA:
1
21
12
21
2
1,
0.5 0.5sin ,
2
0,
DCPA
d DCPA
dd
u DCPA d DCPA d
dd
d
+
= − −
−
DCPA
(9)
where uDCPA denotes the membership function of the
DCPA, d1 denotes the minimum encounter distance
between the two ships, and d2 denotes the safe
encounter distance. d1 and d2 are defined as [6].
( )
( )
1
21
0.2
1.1 , 0 112.5
0.4
1.0 , 1 12.5 180
0.4 2
1.0 , 1 80 247.5
0.2 2
1.1 , 247.5 360
2
T
T
T
T
T
T
T
T
d
dd
−
−
=
−
−
−
−
=
(10)
2. The membership function of TCPA:
1
2
2
12
21
2
1,
,
0,
TCPA
t TCPA
t TCPA
u t TCPA t
tt
t TCPA
−
=
−
(11)
where uTCPA is the membership function of the TCPA, t1
represents the ship collision time, and t2 represents the
time taken to start paying attention to the target ship.
t1 and t2 are expressed as [6].
( )
( )
22
1
1
1
1
1
22
2
2
2
2
2
,
,
,
,
R
R
R
R
D DCPA
DCPA D
V
t
D DCPA
DCPA D
V
D DCPA
DCPA D
V
t
D DCPA
DCPA D
V
−
=
−
−
=
−
(12)
3. The membership function of DR [6]:
1
2
2
12
21
2
1, 0
,
0,
R
R
R
DR
R
DD
DD
u D D D
DD
DD
−
=
−
(13)
where uDR is the membership function of the relative
distance and D1 represents the minimum distance to be
avoided by the give-way ship. Generally, D1 is set as
12–14 times the ship length, D2 represents the safe
distance for collision avoidance. D1 and D2 are
expressed as [6].
1099
( )
( ) ( )
1
2
2
12 14
1.7 cos 19 4.4 2.89 cos 19
TT
DL
D
=−
= − + + −
(14)
4. The membership function of θT:
( ) ( )
2
440 5
0.5 cos 19 cos 19
289 17
= − + + − −
T
θ T T
u θθ
(15)
where uθT is the membership function of relative
bearing.
5. The membership function of VT/VO :
2
1
2
1
1 2 sin
u
C
=
+
++
(16)
where u
is the membership function of relative ratio
velocity,
represents VT/VO , C ∈[0°,180°] is collision
angle. C can be expressed as follows [6].
( )
sin sin
TO
C
=−
(17)
The CRI value between the two ships was
calculated based on the membership functions above.
The CRI has a value between 0 and 1 for the degree of
collision risk by simultaneously reflecting various
condition risks. The collision risk value is considered
high when fCRI is greater than 0.6 [6]. Therefore, the
collision risk model is defined as follows [6]:
( )
, , , ,
R
RT
T
DCPA
TCPA
D
CRI DCPA TCPA D
u
u
u
f W U w w w w w
u
u
= =
(18)
where fCRI is the collision risk function, W is the weight
matrix in which each weight belongs to (0,1) and the
sum of the weights is 1. wDCPA, wTCPA, wDR, wθT and wε
are weight of membership functions, which are usually
set as 0.400, 0.367, 0.133, 0.067 and 0.033 accordingly
[6]. In addition, U is the membership function matrix.
3.3 Proposed method
The collision risk function consists of two components:
a weight parameter matrix and membership function
matrix. The proposed method focused on wDCPA, wTCPA,
wDR, wθT and wε that are typically set to 0.400, 0.367,
0.133, 0.067 and 0.033, respectively. The navigators' SA
was obtained in both a previous study involving
captains and the current study involving officers. The
SAs of the navigators showed that the target
recognition patterns of the captains in the previous
study and officers in the current study were different,
as shown in Table 1. and Table 3. This allows the CRI
to identify the target recognition patterns of captains
and officers separately. In addition, to determine
which parameters place more emphasis, the weight
parameters must be optimized separately for each
navigator group.
This study adopted a grid search technique, which
is typically used to determine the optimal
hyperparameters in machine learning [8]. The grid
search-based weight aggregation optimization is
illustrated in Figure 3. to determine the optimal weight
parameter set to compute the collision risk function
during maritime navigation. The proposed method
systematically explores the weight space to maximize
the model performance, ensuring accurate risk
classification based on the SA of navigators. We used
combinations of weights with up to three decimal
places, from 0.0 to 1.0, and differed by 0.001 within
each grid. Because the number of possible weight
combinations is extremely large, we applied the grid
search refinement level technique, which divides the
refinement into three levels, starting with a resolution
of 0.1, and increasing to 0.01, and 0.001, respectively,
with each level of refinement capturing the majority of
weights and using them in the next refinement level.
Moreover, the weight combinations are constrained to
sum to 1.0. The grid search weight aggregation process
continues until all possible weight combinations are
evaluated. Each combination is applied to the collision
risk function to obtain fCRI by using the recorded
navigational simulation data. To evaluate the
effectiveness of each weight combination, the
significant targets of the navigators were used as
ground-truth risk targets. The classification of risk or
non-risk is under the condition that if fCRI is greater than
0.6, as was set in a previous study [6], the target will be
identified as a risk target. In addition to evaluating
classification performance, standard evaluation
metrics such as the F1-score, precision, recall, and
accuracy were applied. Grid search weight aggregation
was performed to determine the best weight
parameters until all possible combinations were
accomplished.
Figure 3. Grid search method for finding the best weights
combination.
1100
4 EXPERIMENT
4.1 Bridge Simulator
The experimental bridge simulator setup was adapted
from a previous study [4], with the same equipment
configuration and procedure used to assess the SA of
navigators using the SAGAT. The purpose of the
simulator experiment was to obtain the SA results of
the navigators in watchkeeping, which were measured
by the SAGAT. Experiments with the SAGAT were
conducted using the full mission bridge simulator of
the National Maritime Research Institute. The bridge
simulator installed general navigational equipment,
such as a compass, binoculars, radar, ECDIS, and a
steering stand with a steering wheel. The simulator
analysis systems recorded the subjects’ responses in
the navigational order and encountered situations.
4.2 Subjects
In this study, 12 subjects with different onboard
experience from those in the previous study were
employed. The details of the subjects in previous and
current studies are shown in Table 2 [4].
Table 2. Comparison of subjects' details between the
previous and current study.
Experimental session
Subject ID
Proficiency
Previous study
Sub.A
Captain
Sub.B
Captain
Sub.C
Captain
Sub.D
Captain
Current study
Sub.E
2/O
Sub.F
2/O
Sub.G
Captain
Sub.H
Captain
Sub.I
Captain
Sub.J
Captain
Sub.K
C/O
Sub.L
2/O
Sub.M
2/O
Sub.N
2/O
Sub.O
Captain
Sub.P
2/O
In this study, the subjects' watchkeeping used
general navigational equipment including a compass,
binoculars, radar, ECDIS, and steering instruments.
Furthermore, the participants were instructed to
maintain their own simulated ship on a steady course
and speed if possible. During the experiment, when the
participants perceived an imminent risk of collision,
they were instructed to avoid it.
Before the experiment, to conform with ethical
standards in human research, we requested all the
subjects to fill out informed consent forms for human
research, which all subjects accepted and signed to
signify their informed consent.
4.3 Experimental Scenario and Measurement Method for
navigators' SA
According to the navigator was measured using the
SAGAT in a previous study [4]. In this study, it was
necessary to perform measurements under the same
experimental conditions. Therefore, the open sea was
applied as the sea area for the experimental scenario,
and there were 24 target ships, each having various
encounter situations with the simulated own ship of
the subjects. Ship route tracking of the ship and target
ships is shown in Figure 4. Because the experiment was
conducted using different bridge simulators and ship
model versions, the ship model was changed from a
cargo ship to a container ship with different ship
characteristics such as ship speed and ship
maneuvering ability. Therefore, to measure the SA of
the navigators under the same encounter situations,
the own ship speed was decreased, and the
measurement times for the SAGAT were adjusted.
Figure 4. Example tracking of own ship and target ships.
In the current study, the SA of the navigators was
measured using the SAGAT under the same
experimental conditions as in the previous study. In
addition, the scenario spanned approximately 30 min,
including the interruption time for the SAGAT, and the
experimental scenario was suspended five times after
0, 3, 8, 13, and 22 min. The first measurement was set
at the beginning, the second at 3 min, and the third to
fifth measurements were conducted every 5 min.
During the interruption time, the navigator's SA
was measured by the report that the subjects filled in
based on the displayed radar plot chart (Figure 1). The
participants filled in the recognized ships and priority
rankings in the report. Table 3 illustrates the officers'
SA obtained in the current study. In the table, the first
column shows the target ship IDs, the second column
shows the relationship between the own-ship and
target ships, and the initial encounter is mentioned in
the second column. In addition, the third column
illustrates the result of the SA of the navigators, where
the cells marked with a circle represent the recognized
targets, and the cells marked with a backslash
represent no navigator SA measurements because the
navigators performed collision avoidance before the
SA measurement time. Moreover, the gray-highlighted
cells represented significant target recognition, as
assumed in this study, if more than half of the subjects
recognized the target.
1101
Table 3. Result of officers' SA.
5 ANALYSIS
5.1 Analysis of Weighting Parameters in Collision Risk
Model
In this study, significant recognition targets of captains
and officers and simulated navigational data were
analyzed by executing a collision risk function through
resimulation. This study determined that when the fCRI
was > 0.6, it was identified as a significant target.
Moreover, the analysis method is a grid-search-based
weight parameter optimization, where the sum of
weights must equal one. The results of the normalized
proposed method for the captain and officer compared
to the conventional weight parameters are illustrated
in Table 4, where the first row indicates the developed
weight set of the captain, the second row indicates the
developed weight set of officers, and the third row
indicates the conventional weight set proposed in a
previous study [6].
Table 4. Normalized Proposed Method Weight Parameters
Result for Captain and Officer.
Weights set
Subject
W(DCPA)
W(TCPA)
W(Relative
Distance)
W(Relative
Bearing)
W(Vo/Vt)
Developed
Captain
0.370
0.190
0.120
0.150
0.170
Officer
0.246
0.110
0.200
0.333
0.111
Conventional
0.400
0.367
0.133
0.067
0.033
Figure 5 shows the radar chart of the comparison of
the normalized proposed method weight parameters
on different parameters between the captain and
officer, which indicates the distribution of each
parameter. In the radar chart, the solid line represents
the captain’s optimized weight parameter results and
the dashed line represents the officer's optimized
weight parameter results. Table 4 presents the
normalized weight parameters of the proposed
method and shows the distribution of each parameter
for the captain and officer. In addition, the developed
weight parameter results show the differences in each
parameter compared with the conventional weight
parameters. Furthermore, the developed weight
parameters for the captain and officer illustrate
differences in the distribution of the weight
parameters.
Figure 5. Comparison of normalized proposed method
weight parameters on different parameters between Captain
and Officer.
Table 5 indicates that the developed weight
parameters outperform the conventional weight
parameters at both navigator levels. In particular, the
developed weight parameters for the captain achieved
accuracy, precision, recall, and F1-score higher than
those of conventional weights. Similarly, the
developed weights for the officer achieve accuracy,
precision, recall, and F1-score, which are higher than
the conventional weights. In addition, precision
measures the proportion of correctly identified
positives out of all predicted positives, recall reflects
the ability of the weights set to identify all actual
positives, and the F1-score is the harmonic means of
precision and recall that balances false positives and
false negatives. These improvements suggest that the
developed weights align better with expert decision-
making patterns.
Table 5. Performance comparison between developed
weights and original weights for Captain and Officer.
Subject
Weights set
Accuracy
Precision
Recall
F1-score
Captain
Developed
0.875
0.818
0.692
0.750
Conventional
0.802
0.769
0.385
0.513
Officer
Developed
0.909
0.773
0.680
0.723
Conventional
0.847
0.615
0.320
0.421
Exp.
min
1
(0 min)
2
(3 min)
3
(8 min)
4
(13 min)
5
(22 min)
Sub.
E
F
G
H
I
J
K
L
M
N
O
P
E
F
G
H
I
J
K
L
M
N
O
P
E
F
G
H
I
J
K
L
M
N
O
P
E
F
G
H
I
J
K
L
M
N
O
P
E
F
G
H
I
J
K
L
M
N
O
P
Ship
Encounter
5
7
1
3
3
4
5
8
3
6
1
1
7
9
0
5
5
0
0
7
4
0
2
2
6
0
0
7
7
0
0
9
3
0
1
2
4
0
0
7
0
5
0
0
0
0
2
0
6
0
0
0
0
0
6
3
3
0
0
0
1
S to P*1
Cross
O
O
O
2
S
Overtake
O
O
O
O
O
O
O
O
O
O
O
O
3
S to P
Cross
O
O
O
O
O
O
O
O
O
O
O
O
4
S to P
Cross
O
O
O
O
O
O
O
O
O
O
O
5
P to S
Cross
O
O
O
O
O
O
O
O
6
P to S
Cross
O
O
O
O
O
O
O
O
O
7
P to S
Cross
O
O
O
8
P to S
Cross
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
9
S to P
Cross
O
10
S to P
Cross
O
O
O
O
O
O
O
O
11
S to P
Cross
12
P to S
Cross
O
O
O
O
O
13
P to S
Cross
O
O
O
O
O
14
P to S
Cross
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
15
S to P
Cross
O
O
O
16
S to P
Cross
17
P to S
Cross
O
18
P
Head on
O
O
O
O
O
O
O
O
19
S
Parallel
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
O
20
S to P
Cross
O
21
S
Head on
O
22
S
Parallel
O
O
O
O
O
23
S
Head on
O
O
O
O
O
O
O
24
S to P
Cross
O
O
O
*1 “S” denotes the starboard side and “P” denotes the port side.
1102
5.2 Comparison of Weight Parameters Performance
Between Proposed Weight Parameters and
Conventional Weight Parameters
Figures 6 and 7 show a comparison of the performance
of the developed weight parameters and conventional
parameters for the captain and officer, respectively.
The vertical axis indicates the number of significant
targets identified and the horizontal axis represents the
time intervals. The solid line indicates the SA of the
navigators; that is the targets selected for more than
half of the subjects were considered as significant
targets, and the dashed line indicates the targets
captured by executing the collision risk function with
the developed weight parameters. In addition, the
dotted line indicates the significant targets captured by
executing the collision risk function using conventional
weight parameters. As shown in the figures, the
developed weight parameters consistently
outperformed conventional parameters. The
developed weight parameters achieved recognition of
more significant targets throughout the interval time
compared to conventional weight parameters, which
had lower performance in recognizing the significant
target.
Figure 6. Comparison of performance between developed
weight parameters and conventional parameters for the
captain.
Figure 7. Comparison of performance between developed
weight parameters and conventional parameters for officer.
6 CONSIDERATION
Through the officers' simulation experiment using
SAGAT and adapted grid-search weight optimization,
the study revealed the differences in the weight
parameters of navigational risk factors at each
navigator level. The analysis of the weighting
parameters in the collision risk model illustrates that
the developed weight parameters captured significant
target ships measured at each navigator level more
effectively than the conventional weight parameters.
Notably, the results revealed a distinction in emphasis
between the two navigator levels. First, captains have
more advanced experience, which tends to emphasize
higher weights for DCPA, TCPA, and the relative
velocity ratio (Vo/Vt), indicating a specific risk-
collision strategy. However, officers placed more
emphasis on the relative bearing and distance factors.
This suggests that officers may rely more on observable
target ship positions, whereas captains may focus on
the estimated trajectory approach of the target ships.
Moreover, the experimental findings support the
hypothesis that situational awareness varies not only
in strategies for collision risk assessment but also in
emphasis on navigational risk factors across navigator
ranks. The analysis of the SA results of the navigators
illustrates that the significant recognition between the
captain and officer levels was different in the target
ship that passed from the port side to the starboard
side, and that the relative bearing of the target ship
affected the SA of the navigators. Notably, the captains
emphasized the target ships, which have a larger
relative bearing on the starboard side and the
possibility of changing their course across the heading
on the starboard side, than the target ships that are
closer to their heading. Figure 8 shows the RADAR
display at the SAGAT measurement timing in relative
motion mode, and the dashed lines represent the trail
tracking in two minutes. The RADAR display range
was six nautical miles. In addition, the black triangles
represent the target ships, and the red and blue
triangles represent the target ships that were identified
as SA by the captain and officer groups, respectively.
Moreover, the green triangles represent the target ships
identified as situationally aware by both.
Figure 8. RADAR display at SAGAT measurement.
According to a previous study, the rank-order level
of navigators mainly focuses on three types of
information to decide the priority level of target ships:
1103
the rate of bearing change, distance, and type of
encounter situation. In addition, the experimental
interview in a previous study illustrated evidence that
navigators placed more emphasis on crossing
situations from the starboard than on other encounter
situations [4]. Similarly, these findings support our
findings that the target ships were primarily identified
in crossing situations. Furthermore, this study's
evidence illustrates that the officers emphasized not
only crossing situations from the starboard side but
also crossing situations from the port side.
However, the limitation remains in that the officer
group was not categorized into official maritime ranks,
such as chief, second, or third, which may have led to
a mixture of navigator experience levels in SA. In
addition, the size of the captains’ SA data was limited.
To address this limitation, future studies should
provide deeper insights into the role of navigator
experience. In addition to determining more versatile
weights and developing weight parameters, it is
necessary to explore the SA data derived from future
experiments and combine them with onboard
experience.
7 SUMMARY
This study proposes an optimized set of weight
parameters for the CRI that is compatible with two
levels of navigators to distinguish their varying
emphasis on navigational factors in maritime
operations.
The experiment was conducted with officers using
the SAGAT to measure the SA of navigators. In
addition, through resimulation using recorded
experimental navigational data and the navigators' SA
as the ground truth, grid search-based weight
optimization was applied to find the optimal weight
parameters. The CRI weight parameters were refined
to better capture the significant target ships recognized
at each navigator level. The findings demonstrated that
the optimized weight parameter configurations could
effectively identify risk targets based on the
experimental SA data of the navigators. Remarkably,
while both the captains and officers recognized some
significant target overlap, the variation in the
recognition targets indicated that each navigator level
emphasized different navigational collision risk
factors. These findings suggest that the collision risk
model can be developed for specific navigator
experiences, thereby improving the risk target
recognition model for congested water.
This study has certain limitations. The availability
of navigators' situational awareness (SA) data for
navigators in both prior and current studies remains
limited in terms of quantity. To validate these findings,
experiments across different navigator levels are
necessary to collect more comprehensive SA data.
Additionally, it is necessary to validate the finding that
an extended evaluation of more diverse navigator
profiles is required. Furthermore, officer-level
experiments did not categorize officers into three
distinct ranks, although maritime navigation classified
ship officers into three levels: third, second, and first
officers. Therefore, the officers’ target recognition was
mixed. Addressing these limitations in future research
will provide deeper insights into the role of navigator
experience and the diversity of navigator profiles.
ACKNOWLEDGEMENT
This work was supported by JSPS KAKENHI (grant number
24K07900).
REFERENCES
[1] M. R. Grech, T. Horberry, and A. Smith, "Human Error in
Maritime Operations: Analyses of Accident Reports
Using the Leximancer Tool," Proceedings of the Human
Factors and Ergonomics Society Annual Meeting, vol. 46,
no. 1, pp. 1613-1617, Sept. 2002.
[2] Mica R. Endsley, "Situation awareness global assessment
technique (SAGAT)," IEEE National Aerospace and
Electronics Conference, vol.3, pp.789-795, 1988.
[3] T. Okazaki and C. Nishizaki, "A Study on Situation
Awareness of Ship Maneuvering Simulator Training,"
International Journal of Emerging Trends in Engineering
& Technology (IJETET), vol. 3, no. 1, pp. 21-26, 2015.
[4] C. Nishizaki, T. Takemoto, and Y. Kunieda, "A
Characteristic of a Navigator's Situation Awareness for
Crossing Ships," TransNav: International Journal on
Marine Navigation and Safety of Sea Transportation, vol.
11, no. 2, pp. 295-302, June 2017.
[5] C. Nishizaki, H. Itoh, K. Yoshimura, K. Hikida, and N.
Mitomo, "Development of a method for marine accident
analysis with bridge simulator," in 6th International
Conference on System of Systems Engineering (SoSE), pp.
31-36, 2011.
[6] Y. Hu, A. Zhang, W. Tian, J. Zhang, and Z. Hou, "Multi-
Ship Collision Avoidance Decision-Making Based on
Collision Risk Index," Journal of Marine Science and
Engineering, vol. 8, no. 9, p. 640, Aug. 2020.
[7] M. R. Endsley, “Toward a Theory of Situation Awareness
in Dynamic Systems,” Human Factors: The Journal of the
Human Factors and Ergonomics Society, vol. 37, no. 1, pp.
32–64, Mar. 1995.
[8] J. Wu, X.-Y. Chen, H. Zhang, L.-D. Xiong, H. Lei, and S.-
H. Deng, "Hyperparameter optimization for machine
learning models based on Bayesian optimization,"
Journal of Electronic Science and Technology, vol. 17, no.
1, pp. 26-40, 2019.
[9] Okazaki, T. and Ohya, "A Study on Situation Awareness
of Marine Pilot Trainees in Crowded Sea Route".
Proceedings of IEEE International Conference on System,
Man and Cybernetics. pp. 1525‐1530, 2012.
[10] S. Nakamura and N. Okada, "Safety evaluation of
automatic collision avoidance systems," Journal of the
Japan Institute of Navigation, vol. 142, no. 1, pp. 18-28,
2020.
[11] B. Wu, T. Cheng, T. L. Yip, and Y. Wang, "Fuzzy logic-
based dynamic decision-making system for intelligent
navigation strategy within inland traffic separation
schemes," Ocean Eng, vol. 197, p. 106909, 2020.
[12] B. Wu, T. L. Yip, X. Yan, and C. Guedes Soares, "Fuzzy
logic-based approach for ship-bridge collision alert
system," Ocean Eng, vol. 187, p. 106152, 2019.
