You are on page 1of 14

J Mar Sci Technol (2010) 15:257–270

DOI 10.1007/s00773-010-0089-7

ORIGINAL ARTICLE

Collision risk assessment for ships


CheeKuang Tam • Richard Bucknall

Received: 27 May 2009 / Accepted: 26 March 2010 / Published online: 23 April 2010
Ó JASNAOE 2010

Abstract Efficient maritime navigation through dynamic and distance (DCPA); however, such approaches only
obstructions at close range is still a serious issue faced by provide one-dimensional information on the traffic situa-
mariners. There have been studies focusing on collision tion. A two-dimensional assessment of the collision risk,
risk assessment in the past, but the majority were based on the ship domain concept, was introduced by Fujii et al. [1]
the first person perspective, with area-based ship domain and Goodwin [2], using an area around either the OS or
concepts that are defined around either the ownship or the TS1 to indicate the risk of collision, which is easily visu-
obstacle. Such methods are acceptable for encounters alised in 2-D space. There are numerous subsequent studies
where the ownship is required to manoeuvre according to [3–5] that have focused on collision risk assessment using
the collision regulations (COLREGs), but they will not such area-based concepts, but which employ a modified
work correctly if the ownship is the stay-on party. This model or different methodologies to generate the boundary
article presents an alternative method of assessing the of the safety area (i.e. fuzzy logic or artificial neural net-
collision risk for surface ships in close-range encounters works); these studies have been reviewed and discussed in
that is compliant with the COLREGs as well as other ships Tam et al. [6] and Thomas et al. [7].
from different perspectives. Overall, the majority of these studies are based on the
ship domain concept and define a safety area around either
Keywords Collision risk  Collision avoidance  the OS or TS which represents the region where other ships
Evasive manoeuvre should not enter so as to avoid the need to make evasive
manoeuvres. These studies have not incorporated the
COLREGs explicitly in such a way that all obstacles have
1 Motivation areas that the OS should not enter, or an area around the OS
that all other obstacles should keep out of. The effects of
Efficient maritime navigation through dynamic obstacles at COLREGs were partially realised by employing a specially
close range is still one of the many problems faced by constructed geometry of the safety area, i.e. the approaches
mariners, especially in terms of determining the manoeu- of Smierzchalski [8] and Davis et al. [4], where the safety
vres necessary to avoid a potential collision that is com- area is enlarged on the starboard side so that a longer
pliant with the collision regulations (COLREGs). In the distance is needed if the navigation path is around the
past, studies have been conducted to assess the collision starboard side. This makes navigation around the port side
risk using parameters based on the properties related to the a more favourable manoeuvre, mimicking the effect of the
closest point of approach (CPA), such as the time (TCPA) COLREGs in certain types of encounter. Such methods are
acceptable for encounters where the OS is legally obliged

C. Tam (&)  R. Bucknall


Department of Mechanical Engineering,
1
University College London, In this article, a ship that is in direct control is referred as an
Torrington Place, London WC1E 7JE, UK ownship (OS), while any other ship besides an OS is referred as the
e-mail: c_tam@meng.ucl.ac.uk target ship (TS) or obstacle.

123
258 J Mar Sci Technol (2010) 15:257–270

to give way to the TS, but will not interpret the traffic designed to be as general as possible while maintaining the
scenario correctly otherwise. flexibility to be customised to different ship dynamic
Other studies, which have used other techniques such as properties.
reinforced learning [9] or fuzzy logic based [10] collision Furthermore, it is worth emphasising that this study is
risk assessments, usually contain explicit requirements of not intended to recommend a specific dimension of the
the COLREGs based on evaluating the direction of safety area, as this topic has been well documented by
approach of the TS. However, such approaches are gen- some recent studies [11, 12]. In addition, the method used
erally not suitable for path-planning algorithms, as they in the generation of a navigation path that is COLREGs
solely determine the safest manoeuvre for a single obstacle compliant will not be discussed in this article, but will be
at a particular instance—typically the obstacle with the the subject of the next publication from us.
highest risk of collision without considering other obstacles
with a lower risk of collision. This way of assessing the
2.2 Overview
collision risk could lead the OS into an unfavourable sit-
uation at a later stage, as it is not considering the overall
Figure 1 shows the overall concept and processes involved
picture of the traffic scenario.
in assessing the risk of collision. It is assessed based on the
In addition, most studies have taken the first person
discretised navigation path of the OS, where the type of
view, where the OS is the only manoeuvring party while
instantaneous encounter and the risk of collision are eval-
the other object remains at the detected bearing; hence, the
uated. The assessment process involves computing the
resulting safety area would not be compatible if the colli-
dimensions and shape of the safety area, which is consid-
sion risk is evaluated from different perspectives. The
ered the region where ships should not normally enter, as
result of such an approach is that the OS would avoid all
the relative distance between the ships is too low to allow
other obstacles even though the OS is not legally obliged to
safe operation in this area; this is similar to the concept
give way. This article reports a novel method of assessing
proposed by Goodwin [2]. The risk of collision is assessed
the risk of collision for ships, which was developed spe-
in two main steps (italics in Fig. 1): determining the type of
cifically for close-range encounters that address the short-
encounter and determining the dimensions of the safety
comings identified above.
area, which is explained in Sects. 2.3 and 2.4 respectively.
This article is structured into three main sections: Sect. 2
The collision risk assessment process effectively works in a
explains the concept as well as the assumptions of the
loop in such a way that it evaluates all of the discretised
proposed method in assessing the risk of collision, which is
points along the navigation path for risk of collision;
determined based on two major conditions—the encounter
however, as mentioned in Sect. 2.1, it only indicates the
type (Sect. 2.2) and the dimensions of the safety area (Sect.
suitability of the navigation path for the traffic scenario; it
2.3). Simulation results are presented in Sect. 3; they are
does not provide suggestions regarding the best evasive
grouped according to type of encounter. The results are
manoeuvre.
followed by a discussion in Sect. 4. Section 5 is a con-
clusion that focuses on the main findings and explains
aspects of the next publication on an extension to this 2.3 Encounter type for each obstacle
study.
The safety area developed in this study will be located
only on selected obstacles that the OS is mandated to give
2 Concept way to according to the COLREGs. There is no safety area
on the OS; the benefits of this approach are discussed in
2.1 Simplification Sect. 3.
In this study, the collision risk between the OS and the
In order to reduce the computational complexity and obstacle is assessed by an area-based method similar to that
resources, all ships (including the OS and all TS) are of Davis et al. [4]. However, the safety area will be located
reduced to point objects, since the ratio of the distance on the obstacle, and the dimensions and geometry of the
traversed across the water to the ship’s dimensions is safety area are determined using a different principle. The
normally large, even in close-range encounters. safety area will be computed at a fixed temporal interval
Due to the fact that there are no explicit guidelines or for all obstacles. The overall approach can be explained in
regulations on safe distances, there are bound to be dis- two steps: the first step involves determining the type of
agreements regarding the ‘‘appropriate’’ dimensions of the encounter with the obstacle of concern; the second step
safety area among navigators due to different interpreta- involves calculating the dimensions of the safety area as
tions of the traffic scenario; hence, the safety area is necessary.

123
J Mar Sci Technol (2010) 15:257–270 259

Fig. 1 Flow chart of the


collision risk assessment Collision risk assessment
process

No risk of collision Risk of collision exists

No Yes

Else
Check whether OS is in safety area?
Navigation path of OS

Determine the type of Determine


If needed the dimensions
Next time step encounter for all TS
of the safety area

Discretised at fixed intervals At each point


All TS

Determinine
Based on projection
the instantaneous
of initial velocity vector
position and heading

OS_ s fHO1 ; HO2 ; OT1 ; OT2 g ¼ fp=8; 15=8p; 5=8p; 11=8pg:


HO2 HO1
The values of OT1 and OT2 are based on Rule 13 of the
R1 COLREGs, which defines an overtaking encounter. How-
ever, there is no explicit guideline in Rule 14 of the
COLREGs that defines a head-on encounter, except when
R6 R2 discussing the visibility of the masthead light and side-
lights; sidelight visibility is defined in Annex I 9(a) of the
3 COLREGs to be small (1–3°). In this study, instead of
2
R5 R3
2 the recommended values, the HO1 and HO2 values were
increased (to angles of p/8 radians), and they will be dis-
cussed later.
OT2 OT1
R4 The obstacle is also further categorised based on its
relative heading with respect to the heading of the OS,
based on the same principle used to categorise the instan-
taneous position. The categorising regions TSR1 to TSR6
are again arbitrary regions, as shown in Fig. 3.
Fig. 2 Regions used to categorise the position of the obstacle; the OS
Finally, the encounter type will be determined based on
is located at the centre
the combination of the instantaneous categorisation of the
obstacle’s relative position and its heading with respect to
The need for a safety area around an obstacle is deter- the position and heading of the OS. The overall idea is
mined by the type of encounter associated with it; each illustrated in Fig. 4, where categories of relative TS
obstacle is categorised into a particular encounter type heading are placed on top of categories of relative TS
based on its direction of approach as well as its relative position with respect to the position and heading of the OS.
bearing with respect to the heading of the OS (OS_hs). The Short descriptions of each possible encounter type are lis-
obstacle is first categorised based on its instantaneous ted in Table 1; the difference between a ‘‘stay-on’’ and a
position with respect to the heading and position of the OS ‘‘safe’’ encounter is that, in a ‘‘stay-on’’ encounter, the OS
according to the regions defined in Fig. 2. The regions R1 is within the collision range of the obstacle, and the
to R6 are arbitrary regions created using data from the obstacle is expected to avoid such encounters by initiating
COLREGs, where: an evasive manoeuvre according to the rules of the

123
260 J Mar Sci Technol (2010) 15:257–270

OS_ s COLREGs; on the other hand, in a ‘‘safe’’ encounter, there


is no close range contact, and hence the obstacle can be
safely disregarded so long as both the OS and the obstacle
maintain the initial heading.
TSR1
As mentioned earlier, the HO1 and HO2 values are larger
HO2 HO1
(angles of p/8 radians) than those recommended, as it has

TSR6 TSR2
3 Table 1 Abbreviations for and brief descriptions of encounter types
2 2 for obstacles
Abbreviation Description
TSR5 TSR3
HO Head-on encounter
OT Overtaking encounter
TSR4
SO Stay-on encounter
SF Safe encounter
OT2 OT1 GW Give-way encounter
ST Static obstacle
Fig. 3 Regions used to categorise the heading of the obstacle

Fig. 4 Chart used to determine


the encounter type. An obstacle
is associated with a different OT
encounter type depending on its
bearing and position relative to GW SO
the OS. For example, if the
obstacle is located in the region
GW SO
R2, and the heading of the
obstacle is in the zone TSR1, HO
the resulting encounter type for
the obstacle will be an
overtaking (OT) encounter, OT OT
according to the label OS_ s
SF SO GW SF

SF SO GW SF
R1
HO HO
R6 R2

R5 R3

OT R4 OT

SF SO GW SF

SF SO GW SF
SF OT SF

GW SO

SF SF
SF

123
J Mar Sci Technol (2010) 15:257–270 261

been found from simulations that the algorithm behaves Table 2 Dimensions of the safety area for different encounter types
better with increased angles. With small angles, the Dimensions (shape) Condition
encounter type changed from HO to either GW or SO
rather sensitively upon small changes in the OS heading. 0 Encounter type = SO OR SF
This is because the region R1 was too narrow. Note that CSA (circular) Encounter type = OT AND TS_U B OS_U
there is a drastic difference between HO and GW or SO in ESA (half-elliptical) Encounter type = HO
term of legal status: only one of the ships needs to perform 0 Encounter type = OT AND TS_U [ OS_U
the evasive manoeuvre in a GW or SO while the other ESA (half-elliptical) Encounter type = GW
maintains its course, whereas both ships must perform StaticRadius (circular) Encounter type = ST
evasive manoeuvres in an HO. On the other hand, the StaticRadius is the minimum safe relative distance between the OS
enlarged R1 also provides an additional buffer against and a static obstacle, CSA refers to a circular safety domain, ESA
uncertainties when deciding upon the type of encounter refers to a half-elliptical safety domain, while TS_U and OS_U are the
occurring; as stated in Rule 14 (c), ‘‘when a vessel is in any speeds of the TS and OS, respectively
doubt as to whether such a situation exists she shall assume
that it does exist and act accordingly’’. Therefore, the distance covered by a TS travelling at 30 kn in 30 s, which
approach adopted in this study is biased towards the safe is considered sufficient for most evasive manoeuvres. The
and conservative side; it considers marginal HO and GW geometry of the safety area for such an encounter is
encounters to be HO encounters, where both ships should deemed to be circular, because such a shape maintains the
perform evasive manoeuvres. safe distance at the stern section of the TS, while also
ensuring that the safe distance from the side of the TS is
2.4 Dimensions of the safety area maintained if the TS fails to notice the OS overtaking from
stern. The safety area at the bow section of the TS is
Collision risk assessment is based on safety areas around insignificant for two reasons; first, since TS_U B OS_U, a
each obstacle, as this is the most computationally practical circular safety area with a radius proportional to TS_U is
and popular method. The dimensions and shape of the considered sufficient to ensure safety. Second, once the OS
safety area depend on the type of encounter as well as the has successfully overtaken the TS, the encounter type
relative speeds of the OS and the obstacle of concern, as changes and hence the circular safety area alters according
shown in Table 2. to the new traffic configuration, meaning that its role is not
StaticRadius is the minimum safe relative distance as significant under such conditions.
between the OS and a static obstacle, while TS_U and For HO and GW encounters, the safety area is half-
OS_U are the speeds of the TS and OS, respectively. CSA elliptical, and it is different from previously published
is a function that computes the dimensions of a circular studies in such a way that the safety area at the fore section
safety domain in an encounter in which the OS overtakes of the TS is elliptical while that at the aft section is circular.
the TS, and is defined as follows: The dimensions of this half-elliptical area are computed by


TS U  OTScaling if TS U  OTScaling  MinSAD;
CSA ¼
MinSAD otherwise;

where OTScaling (= 1.0 min) is the safety area scaling two common functions, namely ESAA and ESAF, which
factor for a specific overtaking encounter, which is intro- determine the radii for the aft and fore sections of the
duced as a way to customise the shape and dimensions of safety area, respectively. The reason for dividing the
the safety area. MinSAD is the minimum safe distance that elliptical domain into fore and aft sections is to reduce
must be maintained between OS and TS for safety pur- complexity when modelling the geometry of the safety
poses. It is defined as 0.25 nmi, computed based on the area. ESAA is defined as follows:


RadiusA þ DTScaling if RadiusA  MinSAD  MinSAD;
ESAA ¼
MinSAD otherwise;

123
262 J Mar Sci Technol (2010) 15:257–270

where DTScaling = ||Dt|| 9 DTScale (in min) is the function Referring to Fig. 5, ESAF returns a similar output to
that relates ||Dt|| (the size of the time step, Dt) to the dimen- ESAA at low speeds (up to the speed where the size of the
sions of the safety area, such that a large ||Dt|| will give a safety area reaches SASLimit); at high speeds, ESAA is
slightly larger safety area in order to prevent tunnelling. capped and gradually reduces to MinSAD, while ESAF
DTScale (= 0.5) is the predefined dimensionless scaling factor increases in proportion to TS_U, as more emphasis is
for the magnitude of the time step, and RadiusA computes the placed on the fore section of the safety area or the direction
safety area’s aft-section radius, defined as follows: of travel of the TS at high TS_U.


TS U  SAScaling TS U  SAScaling\SASLimit;
RadiusA ¼
2 SASLimit  ðTS U  SAScalingÞ otherwise;

where SAScaling (= 1.0 min for ESAA and 1.5 min for ESAF) The combined outputs of ESAA and ESAF are depicted
is the generic scaling variable of the safety area, which in Fig. 6. At low TS_U, the safety area is circular;
depends on the type of encounter (HO or GW). SASLimit assuming that the ship has high manoeuvrability and low
(*0.7 nmi) is a predefined scalar property that limits the inertia at low speed, the TS can easily turn in any direction,
maximum allowable safety area radius on the side and stern so the probability of existence is evenly distributed around
sections; it depends on the manoeuvrability of the TS and will the TS. As TS_U increases, the safety area gets larger
be explained in detail later. Similar to OTScaling, the while maintaining a circular shape up to a certain TS_U
parameters DTScale, DTScaling, SAScaling and RadiusA are value (the peak of the dotted line, TS_U & 0.5), which is
introduced to the process in order to act as customising referred as SASLimit. Practically speaking, the value of
parameters; the values used in this study are based on educated TS_U at which SASLimit occurs should be specific to each
guesses for the performance of a typical 10 t ship. The changes TS, as different ships have different characteristics and
in the magnitudes of the radii at the aft and fore sections are manoeuvrability; however, for simplicity, all ships were
collectively shown in Fig. 5; the dotted line represents the assumed to have the same dynamic properties in this study
output of the ESAA function, which starts with a constant (i.e. a 10 t displacement vessel).
value of MinSAD when RadiusA is lower than MinSAD, so When TS_U [ 0.5, the output of the ESAA function
that a minimum clearance distance is maintained between the reduces while the output of ESAF increases, such that the
OS and TS. Once RadiusA is greater than MinSAD, it grows fore section of the safety area becomes elliptical while the
linearly with TS_U, reaching a peak at SASLimit; then it aft section remains circular but diminishes in radius. Such a
decreases linearly before settling at MinSAD, the minimum change in geometry is designed to emulate the behaviour
allowable size of the safety area.The function that determines and manoeuvrability of a typical displacement ship. When
the safety area’s fore section (ESAF) is defined as: it is travelling above a certain speed, its manoeuvrability


ðTS U  SAScalingÞ þ DTScaling if ðTS U  SAScalingÞ þ DTScaling  MinSAD;
ESAF ¼
MinSAD otherwise;

Safety area outputs, nm deteriorates and hence the ship is more likely to travel in
ESAF the direction of the initial velocity vector, so the safety area
1.5
has an elliptical shape that follows its velocity vector, with
a higher probability of the OS existing directly in front of
1. its velocity vector. Also note that the minor axis of the
safety area has a radius similar to that of the aft section, so
SASLimit
that the safety area always has a continuous boundary.
0.5
ESAA The radius of the safety area at the aft section reduces as
MinSAD
an indication of TS diminishing probability of existence at
0.2 0.4 0.6 0.8 1.0 1.2 1.4
TS_U, kn its side and aft sections. This trend persists until ESAA
reaches MinSAD, where ESAF continues to increase
Fig. 5 Safety area outputs for increasing TS_U. The dotted line according to the magnitude of TS_U while ESAA remains
represents the output for the aft section, while the dashed line
at MinSAD, thus ensuring that, regardless of the magnitude
represents the fore section of the TS

123
J Mar Sci Technol (2010) 15:257–270 263

2 2 2

1 1 1

0 0 0

1 1 1
0.2 nm min , 12. kn 0.45 nm min , 27. kn 0.7 nm min , 42. kn
0.37 aft, 0.37 fore 0.62 aft, 0.62 fore 0.47 aft, 0.87 fore

2 2 2
2 1 0 1 2 2 1 0 1 2 2 1 0 1 2

2 2 2

1 1 1

0 0 0

1 1 1
0.95 nm min , 57. kn 1.2 nm min , 72. kn 1.45 nm min , 87. kn
0.25 aft, 1.1 fore 0.25 aft, 1.4 fore 0.25 aft, 1.6 fore

2 2 2
2 1 0 1 2 2 1 0 1 2 2 1 0 1 2

Fig. 6 The changes in the shape of the combined outputs of ESAA indicate the speed of the TS in different units (nm/min and kn), while
and ESAF (i.e. the safety area) as TS_U is increased. Note that a the bottom row of numerical values shows the dimensions of the
scaling factor of 1 was used. The black arrows indicate the magnitude safety areas for the aft and fore sections, respectively. The values on
and direction of TS_U. The top row of numerical values in each figure the X and Y axes are in nmi

of TS_U, a minimum clearance between OS and TS is Since this study focuses on close-range encounters, we
maintained at the instantaneous positions at all times. As can also increase the size of MinSAD in order to maintain
explained earlier, the OT encounter type has a circular a healthy distance from the TS during a close-range
safety area because, in an overtaking encounter, the OS encounter, thus preventing the performance of the OS
approaches the TS from aft, so a safety area with a from being dynamically affected by the pressure field of
diminishing aft section is not suitable for such an the TS.
encounter.
As mentioned earlier, the safety area is a concept with
no established geometrical or dimensional standards in 3 Simulations
collision risk assessment; it is open to alternate interpre-
tations in terms of the traffic scenario, and different people The proposed method of collision risk assessment was
may be prepared to perform riskier manoeuvres. For tested with a range of typical traffic scenarios that were
this reason, all of the parameters (OTScaling, DTScale, constructed specifically to emulate different types of
SAScaling, SASLimit and MinSAD) can be altered to encounter, and which can be grouped according to the
accommodate the effects of changes in manoeuvrability initial position and heading of the obstacle. These simu-
due to changes in speed, different ship types, or to account lations were setup to study the variation in the safety area
for different personal judgements. over time, as both the OS and obstacles move according to
In addition, the dimensions and geometry of the safety the navigation path. As mentioned earlier, the method used
area can be further enhanced by using data on the ship’s to generate the navigation path, which is COLREGs
length, manoeuvrability and dynamic characteristics. compliant, will not be discussed here.

123
264 J Mar Sci Technol (2010) 15:257–270

42. 42. 42.


4 4 4

2 2 2

TS OS
0 0 0

2 2 2

4 4 4
0.000 1.500 3.000
4 2 0 2 4 4 2 0 2 4 4 2 0 2 4

42. 42. 42.


4 4 4

2 2 2

0 0 0

2 2 2

4 4 4
4.500 6.000 7.500
4 2 0 2 4 4 2 0 2 4 4 2 0 2 4

Fig. 7 Motion simulation of a port-crossing encounter at selected corresponding to that particular figure. The green circle represents the
times. There is no safety area for the TS (in gold; the arrow indicates initial position while the blue circle shows the final position of the
its velocity vector) since the OS (in red) is the stay-on party in this OS, and the yellow line is the navigation path of the OS. The values
traffic scenario. The numerical value in the top left corner of each on the axes are given in nmi, which is also true of all other figures in
figure is the instantaneous speed of the OS in kn, and the numerical this article unless otherwise stated
value in the lower left corner shows the instantaneous time in min

3.1 Crossing encounters The second crossing scenario was similar to the first,
except that the roles of the OS and TS were reversed, such
The first test cases were crossing encounters where the TS that the TS approaches from the OS’s starboard while the
approached the OS to port, and where both the OS and TS other properties remain unchanged. This is essentially the
had the same initial speeds. These crossing encounters previous crossing test case but considered from the TS’s
allowed us to evaluate the conceptual risk assessment perspective. A motion simulation of this test case is shown
method’s interpretation of Rule 15 of the COLREGs for a in Fig. 8, and the aim of this test was similar to the aim of
crossing situation. the first test, but the OS is now the manoeuvring party and
A motion simulation of this scenario is shown in Fig. 7. is therefore required to avoid the TS by passing it on the
According to Rule 15 of the COLREGs, in such a traffic stern side while the TS stays on, in accordance with the
scenario the OS is the passive party and is not required to COLREGs.
alter course in a stay-on (SO) encounter, whereas the TS is The starboard manoeuvre is a better option since it is
the active party that has the responsibility to alter course, shorter in length; a port manoeuvre involves a longer path
crossing the OS at the stern. This explains why there is no to avoid the larger fore section of the TS’s safety area. The
safety area for the TS in the figures. Figure 7 only shows safety area is rendered in dark grey in the figure in order to
the traffic from the OS’s perspective; the TS’s interpreta- distinguish it from the safety area of HO, which is shown in
tion will be explained next, and it is the combination of red in Fig. 10. The safety area exists at the initial step
the navigation paths from the OS and TS perspectives (t = [0.0, 1.5]) because the TS has an encounter of type
that will resolve the potential collision shown in Fig. 7 at GW (give way) according to the conditions shown in
t = 3.0 min. Fig. 4. Once the TS has crossed the path of the OS, this

123
J Mar Sci Technol (2010) 15:257–270 265

42. 39.7832 42.


4 4 4

2 2 2

TS
0 0 OS 0

2 2 2

4 4 4
0.000 1.500 3.000
4 2 0 2 4 4 2 0 2 4 4 2 0 2 4

31.3504 34.3204 37.2904


4 4 4

2 2 2

0 0 0

2 2 2

4 4 4
4.500 6.000 7.500
4 2 0 2 4 4 2 0 2 4 4 2 0 2 4

Fig. 8 Motion simulation of the starboard crossing at selected times. the safety area of the TS. From t = 3.33 onwards, the encounter type
There is a safety area around the TS initially because the TS is in the changes to safe SF according to the conditions shown in Fig. 3. The
R2 and TSR5 regions, which results in a GW encounter type. The OS changes in the speed of the OS at t = 1.50 and 4.50 are due to losses
is required to alter its heading to starboard, otherwise it would enter of momentum by the OS after it changes heading

5
assessment method, since they essentially reverse the roles
of the OS and the TS under the same traffic scenario.
4
Figure 9 shows the combined navigation paths from these
3 Positions of OS crossing scenarios. It shows that the interpretations of the
traffic scenario from different perspectives are compatible
2 with the proposed collision risk assessment, where both
parties perform manoeuvres that are compliant with Rule
1 Positions of TS
15 of the COLREGs, i.e. that a ship that has a ship
0
approaching from its starboard side should manoeuvre and
avoid passing ahead of the other party. If we had used
-1 previously proposed methods where a safety area exists for
-4 -2 0 2 4
all encounter types, there would be a safety area on the TS,
Fig. 9 Comparison of the navigation paths (circles represent ship and the OS in the first test case would need to manoeuvre to
positions at different times) from the two crossing scenarios. The port to avoid the TS even though it is not legally obliged to,
straight line from right to left is the navigation path from port
crossing, while the curved path from centre to top is the starboard
so the combination of navigation paths viewed from the
crossing. The ship positions are colour coded according to the scale perspectives of the OS and the TS would be impractical.
on the right, where each number represents a time step
3.2 Head-on encounter
changes to safe (SF) because both the OS and the TS are
moving away from each other. The second test scenario was a head-on encounter with the
These two crossing scenarios were constructed specifi- TS approaching the OS head-on. The objective of this test
cally to verify the consistency of the proposed collision risk was to evaluate the collision risk assessment concept in

123
266 J Mar Sci Technol (2010) 15:257–270

42. 38.624 41.594


4 4 4

TS
2 2 2

0 0 OS 0

2 2 2

4 4 4
0.000 1.500 3.000
4 2 0 2 4 4 2 0 2 4 4 2 0 2 4

28.2489 31.2189 34.1889


4 4 4

2 2 2

0 0 0

2 2 2

4 4 4
4.500 6.000 7.500
4 2 0 2 4 4 2 0 2 4 4 2 0 2 4

Fig. 10 Motion simulation of a head-on encounter at selected times. the encounter changes to type SF, as both ships are moving away from
From t = 0.00 to t = 4.50 there is a safety area on the TS because the each other. The safety area is shown in red in order to differentiate it
encounter is of type HO according to the conditions in Fig. 4. Once from GW, which is shown in dark grey
the OS has passed the TS at t = 6.00, the safety area vanishes because

5
interpreting Rule 14 of the COLREGs, which dictates that
the ships should pass each other port to port. Figure 10
4
shows a motion simulation of the OS performing a star-
board manoeuvre in order to pass the TS on its port side. 3
Similar to previous simulations, the TS is not manoeuvring
in the simulation because we are viewing the scenario from 2 Positions of TS

the OS’s perspective. The method used to ensure the star- Positions of OS
1
board manoeuvre will be discussed in a subsequent publi-
cation from the authors. 0
Since the initial traffic conditions are the same from
either the OS’s or the TS’s perspective, given that the -1
-4 -2 0 2 4
initial headings of both ships cause them to be directly
head-on, observing the navigation path from the OS’s Fig. 11 Ship positions based on the navigation paths during a head-
perspective alone is sufficient to investigate the interpre- on encounter viewed from different perspectives. The navigation path
from centre to top refers to the OS’s perspective, while the other path
tation of the COLREGs from a different perspective, since is based on the TS’s perspective. The positions of the ships are also
the TS’s perspective of the traffic is simply a mirror image colour coded according to the scale on the right, where each number
of the OS’s perspective, as shown in Fig. 11. In the figure, represents a time step
both ships are manoeuvring according to Rule 14 of the
COLREGs, such that both ships pass each other port to port because the navigation paths were generated by assuming
in a head-on encounter. The relative distance between the that the other party was maintaining course. However,
two ships may be excessive for such a head-on encounter other parameters such as the magnitude of MinSAD or

123
J Mar Sci Technol (2010) 15:257–270 267

42. 42. 42.


4 4 4

2 2 2
TS
0 0 OS 0

2 2 2

4 4 4
0.000 1.500 3.000
4 2 0 2 4 4 2 0 2 4 4 2 0 2 4

39.2842 42. 42.


4 4 4

2 2 2

0 0 0

2 2 2

4 4 4
4.500 6.000 7.500
4 2 0 2 4 4 2 0 2 4 4 2 0 2 4

Fig. 12 Motion simulation of the overtaking encounter at selected visible) throughout the simulation because the encounter type of the
times from the OS’s perspective. There is a safety area around the TS OS remains OT according to the conditions defined in Fig. 4 and
(shown in light brown; the velocity vector of the TS is too small to be Table 2

SAScaling can be adjusted to reduce the safety margin Figure 13 shows a motion simulation for the same sce-
between the ships, hence producing a more realistic and nario from the TS’s point of view. There is no safety area
practical navigation path. around the OS in this encounter because the velocity of the
OS is greater than that of the TS throughout the encounter,
3.3 Overtaking encounter so the TS should maintain course while the OS manoeuvres
according to Rule 13 of the COLREGs.
The third test scenario was an overtaking scenario where Figure 14 shows the combined navigation paths from
the OS approaches the TS from TS’s stern. The aim of this the perspectives of the OS and the TS, where both ships
test case was to evaluate the method’s interpretation of manoeuvre as dictated by Rule 13 of the COLREGs, i.e.
Rule 13 of the COLREGs. A motion simulation of this the overtaking (or faster) ship keeps a safe distance away
traffic scenario is shown in Figs. 12 and 13 from the from the slower TS.
perspectives of the OS and the TS, respectively. In
Fig. 12, which evaluates the situation from the OS’s per-
spective, there is a safety area around the TS because its 4 Discussion
encounter type is OT (overtaking), meaning that the OS
needs to manoeuvre to either port or starboard to avoid Two important assumptions were adopted as the basis of
entering this safety area. Since the velocity of the OS is collision risk assessment, namely the availability of navi-
greater than that of the TS throughout the simulation, the gational information on all obstacles and universal adher-
safety area remains in place well after the OS has over- ence to COLREGs. The availability of navigational
taken the TS, because the encounter type of the OS information on all obstacles can be justified by noting
remains OT according to the conditions defined in Fig. 4 the increasingly widespread implementation of ARPA
and Table 2. and AIS, which provide this positional information. The

123
268 J Mar Sci Technol (2010) 15:257–270

15. 15. 15.


4 4 4

2 2 OS 2

TS
0 0 0

2 2 2

4 4 4
3.000 6.000 9.000
4 2 0 2 4 4 2 0 2 4 4 2 0 2 4

15. 15. 15.


4 4 4

2 2 2

0 0 0

2 2 2

4 4 4
12.000 15.000 18.000
4 2 0 2 4 4 2 0 2 4 4 2 0 2 4

Fig. 13 Motion simulation of the overtaking encounter at selected times from the TS’s perspective. There are no safety areas for the OS in this
encounter because OS_U [ TS_U, so the TS must maintain course while the OS manoeuvres according to Rule 13 of the COLREGs

5 manoeuvre against the established regulations and expect


others to harmonise (unless, of course, when the navigation
4 is centrally controlled). The proposed collision risk
Positions of OS
assessment method will only work if all ships follow the
3
COLREGs or are collaborating such that all ships are
2 aware of each others’ intents.
Positions of TS
One of the main differences between the proposed
1 collision risk assessment method and others [2, 8] is that
the collision risk is continuously assessed at fixed time
0
intervals in the proposed method, rather than solely based
-1
on the initial configuration. Unlike a sweep volume-based
method, the proposed approach creates a temporary safety
-2 area on a particular TS that is deemed necessary
-4 -2 0 2 4
according to the COLREGs, based on the relative
Fig. 14 Ship positions based on the navigation paths during an instantaneous positions and velocities of the OS and the
overtaking encounter viewed from different perspectives. The straight TS in question. The benefits of such an approach are that
navigation path is based on the TS’s perspective, while the other is
based on the OS’s perspective. The positions of the ships are also
the safety area can easily be computed at each time step,
colour coded according to the scale on the right, where each number and the safety area is designed to automatically switch on
represents a time step or off depending on the ship’s instantaneous legal status
(whether to stay on or give way) according to the COL-
COLREGs are effectively the framework that dictates REGs. This means that the COLREGs are interpreted on
movements during all surface ship encounters, and have the fly for a particular navigation path. Such properties
therefore been widely adopted. Thus, it would be unwise to are useful for a path-planning algorithm for close-range

123
J Mar Sci Technol (2010) 15:257–270 269

manoeuvres, because unnecessary manoeuvres can be On the other hand, there are additional benefits of the
eliminated since it is the responsibility of the TS to ini- proposed method that have not been addressed in this
tiate the evasive manoeuvres shown in the crossing and article, but will be discussed in our next publication. For
overtaking simulations when the legal status of the OS is example, it is easier to compute the variation in OS_U
‘‘stay on’’. under the influence of environmental force fields (i.e. wind
In contrast to ‘‘traditional methods’’ [3, 4], there is no or current flow) with a discretised path by computing the
need to offset the safety area (to port or starboard) in order OS_U at each time step. In addition, since the encounter
to mimic the effect of the COLREGs in the proposed type is examined for each obstacle at each time step, we are
method, as it is built into the assessment process such that not limited to assessing the collision risk for a single
there is no safety area if the ship is on a COLREGs-com- obstacle; we can examine the overall traffic scenario. The
pliant path. A number of scaling parameters have been safety area was placed around the point of the TS instead of
introduced when modelling the safety area so that the using the more common CPA because the collision risk
dimensions and shape of the safety area can be customised, assessment was developed in association with a path-
thus making it easy to add supplementary properties such planning algorithm that utilises another parameter for the
as changes in ship manoeuvrability in different environ- TS which generates a vector field where the direction of
ments to the model. In addition, the safety area generated rotation is determined by the COLREGs. It is therefore
was consistent from different perspectives as well as more computationally convenient for the safety area to be
COLREGs compliant in the simulations. located around the TS, since it allows the ‘‘cost’’ of a
On the other hand, such an interpretation of traffic may particular OS manoeuvre with respect to the COLREGs
create a risky situation for the stay-on ship, as, according to vector field to be calculated more easily.
Rule 17 (b) of the COLREGs, ‘‘when, from any cause, the
vessel required to keep her course and speed finds herself
so close that collision cannot be avoided by the action of 5 Conclusion
the give-way vessel alone, she shall take such action will
best aid to avoid collision’’. As mentioned earlier, one of A conceptual collision risk assessment for ships in close-
the main objectives of this study was to produce a collision range encounters has been presented and discussed. The
risk assessment method that eliminates ‘‘unnecessary’’ risk of collision is assessed based on a safety area assigned
manoeuvres if the ship has stay-on status, as it is the to any obstacle that the OS is legally obliged to give way
responsibility of the ship that gives way to initiate an to. The safety area is generated based on the COLREGs
avoidance manoeuvre and avoid any risk of collision. using the instantaneous properties of the traffic configura-
However, one could always set up a special reactive con- tion. Compared with previous methods, the proposed
dition to initiate an emergency avoidance manoeuvre by method is more precise in interpreting the COLREGs,
the stay-on ship if the ship that gives way passes too close capable of generating safety areas around obstacles
to the stay-on ship. While this is outside the scope of this whenever appropriate (as deemed by the COLREGs), and
study, it will be addressed in future studies. consistent from different perspectives (meaning that the
Since the collision risk is assessed based on the rela- navigation paths of all ships involved are compatible).
tive instantaneous positions of the ships, it is only as In addition, the geometry of the safety area does not
good as the accuracy of the detected bearing or possible require offsets like those used in most previous methods to
future course deviations (if known), because the method mimic the effects of the COLREGs, because these are
is based on linear time projection of the velocity vector. incorporated into the collision risk assessment process. The
The elliptical shape of the safety area, on the other hand, shape of the safety area also depends on the obstacle’s
includes the probability of existence, as shown in Fig. 6, velocity vector, which provides the probability of existence
in order to reduce the uncertainty over ship positional for the obstacle in question, therefore reducing positional
information. This effectively incorporates statistical uncertainties. In the next publication, we will discuss how
properties of the TS by varying the safety area according this collision risk assessment method can be used in a path-
to the magnitude and direction of TS_U from its pre- planning algorithm for ships in close-range encounters.
dicted instantaneous position, which constrains the num-
ber of possible practical manoeuvres available to the TS. Acknowledgments The authors would like to express their grati-
As the future positions of the TS are computed based on tude for financial support from the ACCeSS group. The Atlantic
linear projection of its velocity vector at discrete time Centre for the Innovative Design and Control of Small Ships
(ACCeSS) is an ONR-NNRNE programme with grant number
steps, it is also possible to incorporate known future N0014-03-0160; the group consists of universities and industrial
changes in the TS velocity vector into the collision risk partners that perform small-ship-related research. The authors also
assessment. wish to thank Alistair Greig for his comments.

123
270 J Mar Sci Technol (2010) 15:257–270

References 7. Thomas S, Howells G, McDonald-Maier K (2008) Autonomous


ship collision avoidance navigation concepts, technologies and
1. Fujii Y, Tanaka K (1971) Studies in marine traffic engineering: techniques. J Navig 61:129–142
traffic capacity. J Navig 24:543–552 8. Smierzchalski R (1999) Evolutionary trajectory planning of ships
2. Goodwin E (1975) A statistical study of ship domains. J Navig in navigating traffic areas. J Mar Sci Technol 4:1–6
28:328–341 9. Iijima Y, Hagiwara H (1991) Results of collision avoidance
3. Colley BA, Curtis RG, Stockel CT (1983) Manoeuvring times, manouvre experiments using a knowledge-based autonomous
domains and arenas. J Navig 40:355–365 piloting system. J Navig 44:194–204
4. Davis PV, Dove MJ, Stockel CT (1980) A computer simulation 10. Hwang CN, Yan J-M, Chiang CY (2001) The design of fuzzy
of marine traffic using domains and arenas. J Navig 33:215– collision-avoidance expert system implemented by H-autopilot.
222 J Mar Sci Technol 9:25–37
5. Zhu X, Xu H, Lin J (2001) Domain and its model based on neural 11. Pietrzykowski Z, Uriasz J (2009) The ship domain—a criterion of
network. J Navig 54:97–103 navigational safety assessment in an open sea area. J Navig
6. Tam CK, Bucknall R, Greig A (2009) Review of collision 62:93–108
avoidance and path planning methods for ships in close range 12. Szlapczynski R (2006) A unified measure of collision risk derived
encounters. J Navig 62(3):455–476 from the concept of a ship domain. J Navig 59:477–490

123

You might also like