Automatic simulation of ship navigation

Yanzhuo Xue
a,n
, D. Clelland
b
, B.S. Lee
c
, Duanfeng Han
a
a
College of Shipbuilding Engineering, Harbin Engineering University, Heilongjiang, China
b
Department of Naval Architecture & Marine Engineering, University of Strathclyde, 100 Montrose Street, Glasgow, UK
c
Safety at Sea Ltd., 280 St. Vincent Street, Glasgow, UK
a r t i c l e i n f o
Article history:
Received 22 March 2011
Accepted 16 October 2011
Editor-in-Chief: A.I. Incecik
Keywords:
Automatic ship navigation systems
Potential eld method
Route nding
Collision avoidance
a b s t r a c t
Automatic simulation programs of ship navigation can be a powerful tool for operational planning and
design studies of waterways. In such a simulation system the key tasks of autonomous route-nding
and collision-avoidance are performed by the simulation program itself with no or minimum
intervention of a human navigator. This is in many ways similar to automatic navigation systems in
that they are designed to carry out autonomous navigation safely and efciently without the need for
human intervention or to offer advice to the navigator regarding the best course of action to take in
certain situations. There are two key tasks of automatic ship navigation systems: route nding and
collision avoidance. This paper presents an effective and practical method for nding safe passage
for ships in possible collision situations, based on the potential eld method. The general steps of
implementing the potential eld method applied to automatic ship navigation are described. The
algorithm is fairly straightforward to implement, and is shown to be effective in automatic ship
handling for ships involved in complex navigation situations.
& 2011 Elsevier Ltd. All rights reserved.
1. Introduction
In normal maritime navigation situations and pilot-controlled
navigation simulation systems it is the responsibility of the
human pilot to assess the trafc situation, decide the safest and
most direct route to the destination and issue the requisite ship
control commands, adjusting them as and when necessary in
response to the changing trafc and environmental situations.
Simulation of ship navigation in congested waterways is a very
useful tool in the design and operational planning of such waterways
as inland waterways, approaches to harbours and narrow channels.
Most of the current simulation systems rely on experienced human
navigators to control the ship. Fast track simulation is difcult to use
in these systems and consequently they are time- and resource-
intensive to use. On the other hand, automatic simulation of ship
navigation will be able to overcome much of these problems.
Due to increases in trafc, speeds and sizes of modern vessels,
todays waterways and harbours are becoming busier and the
navigation environment is becoming more complex. Increasing
congestion in navigation channels has led to continued unaccep-
tably high levels of occurrences of marine accidents despite
considerable advances in navigational aids and equipment. It is
without question that collision is one of the most severe marine
accidents with potentially catastrophic consequences. The ships
operational procedures for collision avoidance are not very com-
plex, but require the full attention and a good judgement of the
navigator. This is especially true in areas of heavy trafc, such as
coastal zones and narrow sea passages. In these areas collision
avoidance takes on increased signicance, and the increased
threat of possible collisions gives navigators signicantly more
pressure and work load.
An automatic ship navigation system, probably used as an
advisory tool to start with, would be an effective assistant to the
crew contributing considerably to safe and efcient navigation. It
would be able to guide the navigator in determining the safe and
near-optimum trajectory for the ship. In the future, it is quite
possible that trustworthy intelligent machines may be developed
to navigate ships within waterways and ports without human
supervision (Statheros et al., 2008).
This paper is concerned with automatic simulation of ship
navigation only. However, in either an automatic simulation system
of ship navigation or an automatic ship navigation system, one of
the key elements is an intelligent decision-making capability. The
problem of intelligent decision-making is connected with collision
avoidance manoeuvres and route planning of vessels. In recent years,
intensive research on automatic ship navigation has been carried out
along with developments in computer hardware and software algo-
rithms. Iijima and Hagiwara (1994) applied expert systems in ship
autonomous navigation to assess the collision situation and make
a decision. Smierzchalski (1996) adopted evolutionary algorithms to
Contents lists available at SciVerse ScienceDirect
journal homepage: www.elsevier.com/locate/oceaneng
Ocean Engineering
0029-8018/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.oceaneng.2011.10.011
n
Corresponding author. Tel.: 86 451 8251 9193; fax: 86 451 8251 8443.
E-mail address: xueyanzhuo@gmail.com (Y. Xue).
Ocean Engineering 38 (2011) 22902305
develop a ship guidance system for possible collision scenarios. Ito
et al. (1999) employed genetic algorithms to compute the collision
avoidance navigation path. Harris et al. (1999) proposed an intelligent
guidance and control system using a neuro-fuzzy network model for
ship obstacle avoidance. Hwang et al. (2001) developed a system for
collision avoidance and track-keeping employing fuzzy logic. Liu and
Shi (2005) developed a fuzzy-neural inference network model for
ship collision avoidance. Yang et al. (2007) developed the system
of Vessel Intelligent Collision Avoidance Decision-making (VICAD)
based on articial intelligence methods which combines the princi-
ples of expert systems, analytic geometry and fuzzy logic.
In general, numerous attempts have been made in the past using
different approaches to develop an automatic navigation system.
Most of the reported studies can generate collision-free paths for
the vessel, but the work is by no means complete. For example,
some methods are relatively complex and time-consuming and some
systems either ignore collision prevention regulations (COLREGS) or
are incapable of describing the complex encounter situations in
detail. There are three important problems which have not been
effectively addressed in existing research work (Xue et al., 2009):
(a) Navigational rules, including regulations of preventing colli-
sions at sea and the normal practice of mariners, are usually
not taken into consideration in route planning.
(b) Most of the proposed approaches consider encounters with
other vessels (strange ships) in an open sea environment only
(i.e. there is no land involved in the process of route planning)
and assume that the target ships do not change their courses.
(c) Most research work is able to simulate movement of the own
ship, but assumes that all other trafc maintains a known
direction and speed of travel throughout the simulation.
To address the deciencies mentioned above and simulate
realistic situations, such as when many vessels use conned
waterways simultaneously, a new approach is required. Ideally,
this method should be simple and effective whilst taking advan-
tage of the state-of-the-art technology in this area of research.
2. Manoeuvring model used
Accurate and safe navigation requires precise knowledge of
the manoeuvring behaviour of the ship. In order to represent a
manoeuvring ship fully in space a mathematical model with six
degrees of freedom is required. To simplify the problem, it is
assumed that the steering of a ship can be regarded as a rigid-
body motion on the horizontal plane, as is customary. Thus,
the mathematical model is reduced to three degrees of freedom.
Nomenclature
a angle between the relative position and the relative
speed
b angle between the tangential line and the relative
position
a positive scalar parameter
f factor used in collision avoidance distance
m positive constant determining the shape of the
destination
n positive constant
n
obs
number of obstacles
Z positive scaling factor
y parameter used in cubic spline
r
RO
shortest distance between the robot and the obstacle
r
O
limit distance of the repulsive potential eld
inuence
c ships heading
~
c ships heading error
o
n
natural frequency
x relative damping ratio
l factor
u,v surge speed and sway speed, respectively
_ u, _ v surge and sway acceleration, respectively
r, _ r yaw rate and yaw acceleration, respectively
~ r yaw rate error
d rudder angle
(x
i
(y),y
i
(y)) position of the vessel
C
A
collision avoidance distance
C
E
position evaluation error
C
R
checking collision range
C
S
safe passing distance
F
!
att
attractive force
F
!
rep
repulsive force
F
!
total
total force
I
Z
yaw moment of inertia of the ship
K gain constant
K
d
derivative gain constant
K
i
integral gain constant
K
p
proportional gain constant
L
OW
length of own ship
L
pp
length of ship between perpendiculars
L
TA
length of strange ship
N yaw moment
p
!
point on the water surface
p
!
t position of ship at time t
p
!
d
destination position of ship
p
O
positive constant describing the inuence range of
the obstacle
p
S
shortest distance between the ship and the obstacle
surface
P
!
O
position vector of own ship
P
!
T
position vector of the strange ship
P
!
OS
distance between own ship and the point of
intersection
P
!
OT
distance between own ship and the strange ship
T time constant
U
!
p
!
gravity potential energy
U
!
att
p
!
attractive potential energy
U
!
rep
p
!
repulsive potential energy
V
!
O
speed vector of own ship
V
!
T
speed vector of target ship
V
!
OT
relative speed between own ship and target ship
X force applied on the ship in the x-direction
Y force applied on the ship in the y-direction
Abbreviations
COLREGS International Regulations for Preventing Collision at
Sea
GNRON Goal Non-Reachable with Obstacle Nearby
VICAD Vessel Intelligent Collision Avoidance Decision-
making
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2291
With the global and ship coordinate systems shown in Fig. 1, the
equation of motion can be written as (Fossen, 1994)
Surge m _ uvrx
G
r
2
X 1
Sway m _ vur x
G
r
2
Y 2
Yaw I
Z
_ r mx
G
_ vur N 3
where m is the mass of the ship; u, v represent the surge speed
and the sway speed, respectively; _ u, _ v represent the surge and the
sway acceleration, respectively; r, _ r are the yaw rate and the yaw
acceleration, respectively; X is the force applied on the ship in the
x-direction; Y is the force applied on the ship in the y-direction;
I
Z
is the yaw moment of inertia of the ship; N is the yaw moment.
The forces X, Y and moment N can be expressed as functions of
the state variables u, v, r, their time derivatives _ u, _ v, _ r and the
rudder angle d:
X Xu,v,r, _ u, _ v, _ r,d 4
Y Yu,v,r, _ u, _ v, _ r,d 5
N Nu,v,r, _ u, _ v, _ r,d 6
This model of the ship that includes surge, sway and yaw will
yield sufcient information to show the manoeuvring behaviour
of the ship.
3. Potential eld method applied to route nding for a ship
The potential eld method was rst used by Khatib (1986) for
robot path planning in the 1980s. In this method, a potential eld is
dened in the conguration space such that it has a minimum
potential at the goal conguration. While the target is ideally at the
minimum, all obstacles, or walls, are treated as high potential hills. In
such a potential eld, the robot is attracted to its goal position and
repulsed from any obstacles. This method is particularly attractive
because of its mathematical elegance and simplicity. Furthermore,
from a computational point of view, no prior processing is required
and the method is capable of automatically indicating dynamic
behaviour necessary to avoid all obstacles. It allows real-time robot
operations in a complex environment and is currently widely used
for path planning of mobile robots. When this type of route planning
is used at every time step, for example, in a dynamic environment,
the process can be called route nding.
In ship autonomous navigation, Lee et al. (2004) introduced a
fuzzy logic autonomous navigation algorithm based on virtual
eld force (VFF) which is derived from the concept of potential
eld method. Shi et al. (2007) adopted stream function algorithm,
which is another category of potential eld method, to solve
automatic ship navigation.
Ship route nding is, in a sense, similar to the path planning of
a mobile robot. Consider an obstacle in the way of a direct route
between the starting point and the destination point of a ship. The
shortest route for the ship to follow is shown in blue line (desired
track) in Fig. 2. However, the actual safe route will be something
like that shown as the actual track. This actual track can be
determined by applying the potential eld method.
Since the ship is pulled towards the destination point, the
potential energy responsible for it acts in a way reminiscent of
gravitational potential energy. The total potential energy of any
point is the sum of the attractive potential due to the destination
point and the repulsive potential due to the obstacle:
U
!
p
!
U
!
att
p
!
U
!
rep
p
!
7
where U
!
p
!
is the total potential energy; U
!
att
p
!
is the potential
energy due to attraction towards destination point; U
!
rep
p
!
is the
potential energy due to repulsion of the obstacle; p
!
denotes a
point on the water surface.
The ship then is subjected to a force which is derived from this
potential eld as follows:
F
!
F
!
att
F
!
rep
8
where F
!
att
grad U
!
att
p
!
, F
!
rep
grad U
!
rep
p
!
. F
!
att
may be
called the attractive force as it pulls the ship towards the
destination; F
!
rep
is repulsive force as it pushes the ship away
from the obstacle. The feasible path can now be found by
following the direction of the total force at any given position.
More than one obstacle can be accounted for by summing all the
repulsive forces due to the obstacles.
3.1. Attractive potential function
The attractive potential is a function of the relative distance
between the ship and the destination point. The main character-
istics of this function should be that the value is high when they
are far apart, but then reduces gradually until it becomes nil at
the destination. The attractive potential function can, therefore,
be written as follows:
U
!
att
p
!
a: p
!
d
p
!
t:
m
9
where p
!
d
and p
!
t denote the destination position and the position
of ship at time t, respectively; : p
!
d
p
!
t: is the Euclidean distance

y
0G
x
0G
G

V
O
x
0
x
y
0
y
r
Fig. 1. Global and ship coordinate systems.
Start
Desired Track
Obstacle
Destination
A
t
t
r
a
c
t
i
v
e

F
o
r
c
e
Total Force
Repulsive
Force
Actual Track
Fig. 2. Potential eld in ships route nding. (For interpretation of the references
to colour in this gure, the reader is referred to the web version of this article.)
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2292
between the ship at time t and the destination position; a is a scalar
positive parameter; m is a positive constant.
The corresponding virtual attractive force is dened as the
negative gradient of the attractive potential:
F
!
att
p
!
rU
!
att
p
!

@U
!
att
p
!

@ p
!
10
Substituting (9) into (10),
F
!
att
p
!
ma: p
!
d
p
!
t:
m1
11
The shape of attractive potential function can be modied by
varying the value of m, and the strength of the attractive potential
eld can be modied by changing the value of a.
3.2. Repulsive potential function
Unlike the attractive potential of the destination, the repulsive
potential of an obstacle is local in that the area of inuence of an
obstacle is limited to its vicinity. Furthermore, it is logical to
assume that the strength of the eld will increase from 0 at the
border of inuence to a maximum (an innity) at the obstacle
itself.
When the potential eld method was used for path planning of
mobile robots, a problem known as GNRON (Goal Non-Reachable
with Obstacle Nearby) was found in some cases. In applying the
potential led method for route nding in ship navigation, the
GNRON problem is likely to be encountered frequently and a
provision has to be made for such a situation. For example, the
destination point (target) can be near other structures (such as
jetty), and consequently the point with the global minimum
potential may not be at the destination position. As a result, the
ship cannot reach the correct target point. In order to alleviate
this problem, the following form of repulsive potential function
was used and was found to be satisfactory in this regard:
U
!
rep
p
!

1
2
Z
1
p
s

1
p
o
_ _
2
: p
!
tp
!
d
:
n
if p
s
rp
o
0 if p
s
4p
o
_
_
_
12
where U
!
rep
p
!
denotes the repulsive potential generated by the
obstacle; Z and n are the constants; p
s
is the shortest Euclidean
distance between the ship and the obstacle surface; p
o
is a
positive constant describing the inuence range of the obstacle.
Similar to the denition of the attractive force, the correspond-
ing repulsive force is dened as the negative gradient of the
repulsive potential in terms of position:
F
!
rep
p
!
rU
!
rep
p
!

@U
!
rep
p
!

@ p
!
13
Substituting (12) into (13),
F
!
rep
p
!

F
!
rep1
F
!
rep2
if p
s
rp
o
0 if p
s
4p
o
_
14
where
F
!
rep1
Z
1
p
s

1
p
o
_ _
1
p
2
s
: p
!
tp
!
d
:
n
F
!
rep2

n
2
Z
1
p
s

1
p
o
_ _
2
: p
!
tp
!
d
:
n1
15
3.3. Total potential function
The total potential and the total virtual force can be obtained
from Eqs. (7) and (8).
In the case where there are multiple obstacles, the repulsive
force is given by
F
!
rep
p
!

n
obs
i 1
F
!
rep

i
16
where n
obs
is the number of obstacles and F
!
rep

i
is the repulsive
force generated by the ith obstacle.
The ow chart of potential eld method described in the above
sections is given in Fig. 3.
3.4. Representation of obstacles
One of the main tasks in applying the potential eld method
for ship route nding concerns the representation of obstacles.
They can be represented by a number of primitives, including
point, line and arc. Primarily for the reason of simplicity and
superior exibility, point primitives are selected in this work.
Larger obstacles, such as coastlines and islands, can be repre-
sented as a series of point obstacles judiciously placed on the
boundaries as shown in Fig. 4. A utility program to generate the
point obstacles semi-automatically was developed and sample
screen shots of the program outputs are shown in Fig. 5.
3.5. Limitations of the potential eld method
When the potential eld method was applied to ship route
nding, several limitations inherent in the potential eld method
came to light under certain conditions and of these local minima
problem and oscillations in narrow passages were two main
deciencies (Koren and Borenstein, 1991).
As an example of local minima limitation, consider the case
shown in Fig. 6 where the ship is proceeding towards its
destination and a point obstacle exists exactly in line with the
destination. The repulsive force, F
rep
, and the attractive force, F
att
,
Ship state parameter
inputs P(t), Pd and U
Calculation of attractive
force Fatt
Calculation of the distance
between the ship and the
obstacle surface Ps
Calculation of avoidance
direction
If Ps < Po
Calculation of repulsive
force Frep
Repulsive force
Frep= 0
Calculation of total
force F
N
Y
Calculation of ships
dynamic position
Fig. 3. Flow chart of the potential eld method.
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2293
will be directly in line and there will be no component normal to
the ships heading. No safe route can be found in this case and the
ship will simply get stuck at the bottom of the local minimum.
This problem is overcome simply by giving a small initial devia-
tion in its heading to break it out of its trap, if the ship has an
initial non-zero speed. If, for any reason, the initial speed is zero
and the local minimum problem occurs, the ship is given a small
displacement sideways without changing the heading.
For the limitation of oscillations in narrow passages, consider
the case shown in Fig. 7. The ships start position is at (0,0) (km),
the destination is at (50,45) (km) and the speed is 6.2 m/s. The
value of P
o
is set 3 km.
In this particular case, it was found that the ship got stuck after
a few oscillations. This limitation could be overcome through the
following two methods:
(a) Modify the obstacle points as shown in Fig. 8. In this case, one
obstacle point was removed and this appears to have solved
the problem in this particular situation.
(b) In Fig. 9, the strength of repulsive eld was modied and this
appears to have resolved the problem as well.
Although these limitations do exist in ship navigation, they are
uncommon because most waterways have well-dened routes
often strictly regulated by local laws. If, for any reason, these
situations do occur in a simulation, there are a few adjustments
that can be made, although it means an operator intervention.
4. Automatic collision avoidance
As mentioned previously, many methods of collision avoid-
ance either do not take the international regulations for prevent-
ing collision into consideration or are incapable of describing the
Fig. 5. Sample screen shots of the program generating obstacle points. (a) The built map and (b) the map with the obstacle points placed automatically by the program.
Fig. 6. Local minimum problem.
Fig. 7. Ship oscillates in a narrow passage. Fig. 4. Coastline is represented as a series of point obstacles.
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2294
complex encounter situations at sea in detail. Sometimes the
route recommended by these systems is against what the good
seamanship or the regulations of collision avoidance would
dictate. Furthermore, many methods are associated with optimi-
sation algorithms whose goal is to identify an optimum or near
optimum route between the provided waypoints which are
determined a-priori. However, in some situations, such as busy
channels and harbours, the number of moving obstacles in
addition to static ones might be quite signicant, making the
pre-determined waypoints almost useless in a situation of micro-
navigation. In such cases it may not be meaningful to designate
waypoints in advance, or adhere to them rigidly when they are
designated. Therefore, the objective of this study is not to solve an
optimisation problem for a pre-dened set of waypoints, but to
achieve collision avoidance in real time. This, after all, is how the
ship masters operate.
For ships at sea, navigational rules, such as collision regula-
tions (COLREGS) will also have to be obeyed.
4.1. Determining the situation of possible collision
The COLREGS dene the rules for navigation and collision
avoidance. They are the highway codes at sea and are essential for
collision avoidance and as such referenced throughout this paper.
Further details concerning these rules and others can be found in
Crockcroft and Lameijer (1996).
According to COLREGS, the navigator has to decide if a risk of
collision exists and, if it does, determine what manoeuvre to carry
out in order to avoid a collision. Similarly an automatic navigation
system rstly has to decide by itself whether such a risk exists
and then, if it does, take actions for avoiding it. Since there are no
clear criteria for determining when the risk of collision is high
enough to cause concern, a collision detection algorithm has to be
formalised. For this some parameters need to be dened.
(a) Collision avoidance distance: Collision avoidance distance C
A
is
used to dene navigational boundaries. It is the smallest
possible distance between two passing vessels which must
be maintained for avoiding collision, dened here as
C
A
L
OW
L
TA
17
where L
OW
and L
TA
are the lengths of own ship and strange
ship, respectively. This criterion can also be interpreted as the
centre point of the own ship not crossing the circle of radius
of (L
OW
L
TA
) with the centre at the centre point of the
strange ship.
(b) Safe passing distance: In real navigation it is common practice
to allow for uncertainties, such as weather conditions and
current, to give any obstacles as wide a berth as prudence
would dictate. This can be simply represented using a safety
factor, an allowance for the relative speed of the two ships
and an allowance for possible error in position xing as
follows:
C
S
f C
A
C
E
V
r
T 18
where C
S
is the position evaluation error, f is a safety factor, V
r
is
the relative approach speed of the two ships and T is the time
that operator takes to make a decision. Different values for f may
be adopted depending on the situation and visibility (Hilgert and
Baldauf, 1997). C
S
determined in this manner, when plotted
around the strange ship, will depict a misshapen circle, like the
outline of an egg perhaps. The safe passing distance is dynami-
cally updated at every time step of the simulation.
(c) Range of collision checking: In some situations, a number of
ships may be in the vicinity of the own vessel. However, not
all of them may be in a potential collision situation. Therefore,
an important issue in collision avoidance is to determine
under what circumstances the risk of collision needs to be
evaluated so that the ships which are quite safe from any
danger of collision, can be ltered out. It is reasonable to
assume that the most crucial factor determining this will be
the distance between the two ships. The distance at which a
ship begins to receive a close scrutiny to evaluate the collision
risk is termed collision checking range C
R
here. The magnitude
of C
R
depends on the prevailing weather conditions, sailing
area and the speed of the own ship.
4.2. Strategy for collision avoidance
As described previously in Section 3, the potential eld
method has been found to be effective in avoiding stationary
Fig. 8. One obstacle point is removed.
Fig. 9. Field strength of one obstacle was modied.
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2295
obstacles. However, the situation becomes more complex when
the obstacle is moving, as for example, when two ships are sailing
towards the same point at the same time. In this situation,
COLREGS state that one ship should maintain course and speed
(stand-on vessel), while the other is responsible for the avoidance
manoeuvre (give-way vessel) (COLREGS 16 and 17). The avoiding
manoeuvres to be taken, therefore, should be in accordance with
the regulations which all maritime trafc is required to adhere to.
The pictorial representation of this strategy for collision
avoidance of two ships is given in Fig. 10.
The inside radius of the circle around the strange ship is the
collision avoidance distance C
A
and the outside radius is the safe
passing distance C
S
estimated for the situation given. The magni-
tude of vector P
!
OT
is the distance between the own ship and the
strange ship at time t. P
!
OT
can be written as
P
!
OT
P
!
T
P
!
O
19
where P
!
T
and P
!
O
are the position vectors of the strange ship and
own ship, respectively.
S is the point of intersection of P
!
OT
and a circle of radius C
S
with the centre at the midship of the strange ship. P
!
OS
can be
written as
P
!
OS
P
!
OT
C
S
20
V
!
OT
is the relative velocity of the own ship with respect to that of
the strange ship, i.e.
V
!
OT
V
!
O
V
!
T
21
where V
!
O
and V
!
T
denote the velocities of own ship and strange
ship at time t, respectively.
A straight line is drawn from the position of the own ship
tangential to the circle around the strange ship as shown in the
illustration. Then b is dened as the angle between this tangent
line and the relative position vector P
!
OT
. a is the angle between
the relative position vector P
!
OT
and the relative speed vector V
!
OT
.
If the distance between the own ship and the strange ship
9 P
!
OT
9 is less than the range of collision checking C
R
, then the own
ship starts examining this particular strange ship with a view to
establish if the collision risk exists. The risk of collision exists if
the extension of the relative velocity V
!
OT
of the two ships crosses
a circle of radius C
S
around the strange ship. This condition can be
formalised as aob. Thus the two necessary conditions for the
existence of potential collision risk in two-ship encounter situa-
tion are the following:
(a) the distance between the two ships is less than the collision
checking range, i.e. 9 P
!
OT
9oC
R
;
(b) the extension of the relative velocity V
!
OT
of the two ships
crosses a circle of radius C
S
around the strange ship, i.e. aob.
The strategy for collision detection, therefore, is reduced to
checking angles a and b at each time step. If aob, the situation
can be rectied by changing the velocity of either or both ships. In
practice, this means changing the speed and/or heading of either
or both ships. As a rule, ship masters do not like changing speed
as the primary means of navigation unless it is unavoidable. Thus,
it is more likely that changing heading would be used as the
primary means of avoiding a collision in normal circumstances.
This is the primary strategy adopted for avoiding a collision in this
work, but in emergency situations where a collision cannot be
avoided by changing heading alone, the method of reducing speed
may also have to be used. However, such emergencies only occur
when appropriate preventive action is not taken in good time.
When there are more than two ships in relative proximity, the
situation can become much more complicated as shown in
Fig. 11. To make matters worse, there is no rule dealing speci-
cally with the cases of multi-ship encounter situation. In order
to deal with these problems and also to imbue the automatic
simulation program with more realism than evidenced hitherto,
the software developed in this work nds a safe route not only for
the own ship but also for all the ships concerned. In other words,
the program considers each moving ship in turn as the own ship.
The procedure adopted for this purpose can be summarised as
follows:
At each time step the system examines each ship in turn to see
if it is within the collision checking range and with which
ship(s).
If the ship is within the checking range with one or more ships,
then the collision detection procedure described above is
applied for each strange ship.
If there is more than one strange ship, the system will decide
which ships are faced with the highest threats and give them
the highest priority. This step is fundamental to ensure multi-
ple ship collision avoidance.
This process is repeated throughout the entire simulation
while the ship is in motion.
The prioritisation is carried out as follows:
(1) The existence of a static obstacle: Ships which need to avoid
static obstacles will be given the highest priority to man-
oeuvre, because their avoidance manoeuvre is necessary
regardless of whether or not they face threats from other
ships.
Fig. 11. Multiple ships encounter situation.
V
o P
OT
O
V
OT
b
a
S
T
C
A
C
S
Strange Ship
P
OS
Own Ship
V
T
Fig. 10. Basic strategy of avoiding collision in the case of two ships encountering.
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2296
(2) The distance 9 p
!
OS
9: This distance is used to determine which
pair of ships is in the most imminent danger. The pair of ships
with the shortest distance 9 p
!
OS
9 is given the next highest
priority.
(3) Encounter type: If the above two conditions are the same, then
the encounter type of each pair of ships is taken into
consideration. The highest priority is given to the head-on
situation, followed by the strange ship crossing from port to
starboard, the strange ship overtaking, the strange ship cross-
ing from starboard to port and nally the strange ship being
overtaken.
5. Dynamic route generation and heading control
5.1. Dynamic route generation
In most of the previous studies, the ships waypoints are
dened in advance and the route is xed. In real navigation,
however, the environment in which the ship sails changes
continuously, and, therefore, to simulate ship navigation in real
time, the ship states (including the route to follow) need to be
updated at every time step in the simulation. Consequently,
having a set of xed waypoints is not very helpful except as a
rough guidance to follow, as they will be renewed at every time
step any way. The simulation program developed in this work,
therefore, only generates the part of the route closest to the
current position. At the next time step this manoeuvring require-
ment may have to be altered due to the changing circumstances
or the difculty in implementing the required manoeuvre.
This procedure can thus be called dynamic route generation or
micro planning. Fig. 12 shows how dynamic route generation is
executed.
Since the waypoints are reset at every time step during the
simulation, the ships route dynamically responds to changes in
the navigation environment. This is one major way in which the
current study differs from previous work.
5.2. Cubic spline for route generation
After generating the dynamic waypoints, a cubic spline is used
to interpolate between the waypoints. In this way every segment
of the route can be described by a unique cubic polynomial,
expressed in a parameterised form (Fossen, 2002):
x
i
y a
3
y
3
a
2
y
2
a
1
ya
0
22
y
i
y b
3
y
3
b
2
y
2
b
1
yb
0
23
where (x
i
(y),y
i
(y)) are the positions of a point on the route and y is
the parameter.
There are many textbooks (see e.g. Rogers and Adams, 1990)
and papers describing the basic principles of cubic spline techni-
que, and therefore the methods are not discussed any further
here, except to note that the gradient of the curve at the
waypoints is analogous to the heading (or at least the direction
of travel) the ship is required to take.
After calculating the route through the waypoints, the system
can get the desired heading c
i
(y) along the route at any given
point, by calculating the direction of the tangential vector at that
point:
c
i
y arctan
y
y
i
y
x
y
i
y
_ _
24
where y
y
i
y @y
i
y=@y and x
y
i
y @x
i
y=@y.
Then the desired yaw rate r
i
is
r
i
c
y
i
y
x
y
i
yy
y
2
i
yx
y
2
i
yy
y
i
y
x
y
i
y
2
y
y
i
y
2
25
5.3. PID heading controller
Having computed the viable route with its way-points and the
desired heading at any given point in real time, the ship has to
be steered to achieve this. Modelling of human pilot behaviour
requires a large amount of observational data and complex
modelling which is beyond the scope of this study. Therefore,
the solution adopted for the current study was to use an auto-
matic pilot based on a PID control system.
The PID-controller can be designed as follows (Fossen, 2002):
d
PID
t K
p
~
cK
d
~ rK
i
_
t
0
~
ctdt 26
where d
PID
(t) is the rudder angle required at time t;
~
c cc
i
is
the heading error; K
p
is the proportional gain constant; K
d
is the
derivative gain constant; ~ r rr
i
is the yaw rate error; K
i
is the
integral gain constant.
Destination
Strange ship
Obstacle
Obstacle
Dynamic route
Dynamic waypoint
Start
Own ship
Dynamic route
Obstacle
Obstacle
Strange ship
Destination
Dynamic waypoint
Own ship
New start
Fig. 12. Dynamic route generation.
Ship state parameter
inputs P(t), P
d
and U
Decision of static danger Decision of dynamic danger
Decision of avoidance directions
Calculation of ships dynamic position
Generation of dynamic route
Calculation of desired heading
Calculation of ships real position and speed
Fig. 13. A block diagram of the algorithm.
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2297
The controller gains can be found in terms of the design
parameters o
n
and x, through
K
P

o
2
n
T
K
K
d

2xo
n
T1
K
K
i

o
3
n
T
10K
27
where o
n
is the natural frequency of the system and x is the
relative damping ratio; T and K are the time constant and the gain
constant, respectively. The K
p
term allows the controller to
simulate the restoring action of a manual helmsman to a heading
error. The K
d
term simulates the anticipatory action of a well
trained helmsman and nally the K
i
term simulates the helmsman
compensating for low frequency drift of the vessel heading.
A block diagram of the developed algorithm based on the
method described in the above sections is given in Fig. 13.
6. Case studies
A number of simulation studies were carried out in order to
investigate the performance of the proposed system. All of the
moving ships within the simulation range were assumed to be
Mariner class ships. These vessel types have been studied in detail
in various comparative studies by different authors and detailed
information is available on their manoeuvring characteristics.
Details of this vessel type can be found in Fossen (1994). The
parameters of simulation are given as follows:
m2, a 20, Z 200, n 2, f 2:5, C
g
0:04 Nm,
T 107:3, K 0:185, o
n
0:03 and x 1
Fig. 14. Two ships crossing for the rst scenario (1).
Fig. 15. Two ships crossing for the rst scenario (2).
Table 1
Position and speed of ships crossing each others path.
Start (km) Destination (km) Speed (kn)
Own ship (0,0) (20,20) 15
Strange ship (20,0) (0,20) 15
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2298
Fig. 18. Two ships crossing for the third scenario.
Fig. 17. Two ships crossing for the second scenario.
Fig. 16. (a) Own ships yaw rate, yaw angle, speed and rudder angle and (b) strange ships yaw rate, yaw angle, speed and rudder angle.
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2299
6.1. The fundamental cases
Encounters between two ships including crossing, head-on
and overtaking were simulated in the rst instance. The same
algorithm is applied to both ships featured in the simulation.
(1) Two ships crossing: The situation where the paths of two ships
are likely to cross is simulated primarily to check the
compliance with COLREGS. The position and the speed of
ships are given in Table 1.
As shown in Figs. 14 and 15, the program decides that there is
a risk of collision and identies the ship on the left (own ship)
as the give-way ship according to provisions in COLREGS.
It then manoeuvres to starboard to avoid collision, and the
stand-on ship on the right (strange ship) maintains its course
and speed achieving a port-to-port safe passing manoeuvre.
The own ship and strange ships yaw rate, yaw angle, speed
and rudder angle are given in Fig. 16.
The simulation software has been developed to calculate the
turning angle automatically for each of the ships according to
their speed, the encounter type and the collision distance
based on available practical operational experience (Li and
Wang, 1983). This feature is demonstrated in the second
scenario whereby the own ships speed is reduced to 13 kn
and the strange ships speed is maintained at 15 kn. The
simulation results are presented in Fig. 17.
In the third scenario the own ships speed is 15 kn and strange
ships speed is 12 kn. The simulation progress is shown in Fig. 18.
These case studies have shown that the developed simulation
program can make proper decisions concerning the risk of a
collision. The program has correctly determined which ship(s)
needs to take action, and also calculated the turning angle
required to avoid a collision.
(2) Two ships head-on: In this case, the performance of the
automatic navigation system when two ships are travelling
head-on towards one another is investigated. The position
and the speed of ships are given in Table 2. The results from
the simulation run are shown in Fig. 19.
(3) One ship overtaking another: The position and the speed of
ships are given in Table 3. Because the own ship is faster than
the strange ship, the own ship is to turn to port to overtake
the strange ship in accordance with the rules. The result is
presented in Fig. 20.
From these fundamental cases, it can be seen that the
algorithm is capable of making appropriate decisions to avoid
a collision. Furthermore, the simulation results also show that
the algorithm is compliant with COLREGS.
6.2. Autonomous ship navigation through a channel
In order to investigate the performance of the system further
an extreme case the Strait of Istanbul was chosen for the
next case study of navigation in a narrow channel. This strait
presents one of the greatest challenges for ship navigation as
it snakes through the heart of Asia Minor (Kose et al., 2003).
The part of this strait between two red lines as shown in
Fig. 21(a) was modelled as presented in Fig. 21(b) for this study.
The ships speed was set at 7.5 kn, the starting point was at
(x
s
,y
s
)(0,0) (km) and the destination point was at (x
d
,y
d
)(8,10)
(km). The inuence range of the obstacle p
o
was set at 0.3 km and
the main obstacle points were placed manually to represent the
coastlines as shown graphically in Fig. 22(a). The obstacle point-
generation utility program then calculated and placed additional
points along the coastline to ensure that the coastlines cannot be
crossed as shown in Fig. 22(b).
Fig. 23 shows the contour plot of the potential eld generated
by the discrete obstacle points and the destination point. High
potential level can be seen at and near the obstacles, and the high
potential eld forms a kind of dykes along the coastlines,
preventing the ship from leaking out of the shipping lane.
Fig. 19. Two ships head-on.
Table 3
Position and speed of ships, one overtaking the other.
Start (km) Destination (km) Speed (kn)
Own ship (11, 2) (11,19) 20
Strange ship (11,4) (11,14) 10
Table 2
Position and speed of ships in head-on situation.
Start (km) Destination (km) Speed (kn)
Own ship (11,0) (11,28) 15
Strange ship (10,28) (10,0) 15
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2300
Fig. 20. One ship overtaking another.
Fig. 21. (a) Selected part of the strait and (b) simulation model. (For interpretation of the references to colour in this gure, the reader is referred to the web version of this article.)
Fig. 22. (a) Main obstacle points manually placed and (b) additional obstacle points placed by the utility program.
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2301
Figs. 2426 show the progress of the simulation and the key
manoeuvring parameters are shown against time in Fig. 27.
This case demonstrates that the developed algorithm is capable
of automatically navigating a ship though a channel or waterway
which has narrow sections with many sharp turns. It is true that
the path of the ship thus produced contains a few excessive
manoeuvres which a human pilot can judge to be a bad practice.
However, the main purpose of the current study is to investigate
the methods of implementing automatic simulation of ship navi-
gation and no attempt has been made to optimise the route.
6.3. Navigation in a congested area
Ship navigation in a congested area represents one of the
greatest challenges for the experienced mariner. Here, the simu-
lation system is put to the test where it has to deal with a highly
complex encounter situation involving six ships in the presence of
some static obstacles. The position and the speed of each ship are
given in Table 4. Note that the program treats each ship in turn as
the own ship, as discussed earlier, and nds safe routes for all the
ships concerned.
Figs. 2830 show the progress of the simulation. From the very
beginning it is obvious that a complex encounter situation will Fig. 23. Contour plot of the potential eld.
Fig. 24. Progress of simulation (1).
Fig. 25. Progress of simulation (2).
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2302
develop in due course. Once the program detects the risk of collision,
it attempts to identify which ship should be given the priority. For the
ship(s) with the highest priority the program computes the actions
required to avoid a collision according to the COLREGS. In this way
the system steers all the ships safely to their destinations.
This case study demonstrates the ability of the program in
dealing with a complex encounter situation. In reality this degree
of complexity is not very likely to arise and it does give a degree
of condence that the program will be able to deal with most of
the realistic navigational situations to be expected.
Fig. 26. Progress of simulation (3).
Fig. 27. Ships yaw rate, yaw angle, speed and rudder angle.
Table 4
Position and speed of ships.
Start (km) Destination (km) Speed (kn)
Ship A (green) (0,0) (45,40) 15
Ship B (red) (40,0) (3,40) 14
Ship C (magenta) (45,30) (3,20) 14
Ship D (blue) (0,30) (45,20) 14
Ship E (yellow) (40,50) (25,0) 15
Ship F (black) (4,50) (40,10) 14
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2303
Fig. 28. Progress of simulation in a congested area (1).
Fig. 29. Progress of simulation in a congested area (2).
Fig. 30. Progress of simulation in a congested area (3).
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2304
7. Concluding remarks
This paper has presented a method for automatic route nding
and collision avoidance to be used in an automatic navigation
simulation system. So far as can be ascertained through the case
studies, the potential eld method, despite its limitations, appears
to be well-suited for automatic route nding and collision
avoidance algorithm. The collision avoidance algorithm devel-
oped in this study incorporates the key provisions in COLREGS.
There is a potential to develop this algorithm further so that a
practical automatic or advisory navigation system can be con-
structed. Much work is required to incorporate the knowledge
and skills of experienced mariners so that the actions of the
system can resemble those of human pilots more closely.
The simulation program in its current form does have some
deciencies. For example, the weather conditions are not taken into
consideration; it does not have any optimisation or prediction
ability per se; no extreme encounter cases where normal man-
oeuvres may not be able to resolve the situation without resorting
to emergency action, such as change of speed or reversing are not
catered for. Nevertheless, the developed algorithm has been shown
to be effective in complex navigational situations.
References
Crockcroft, A.N., Lameijer, J.N.F., 1996. A Guide to the Collision Avoidance Rules,
5th edition. BH Newnes, Oxford.
Fossen, T.I., 1994. Guidance and Control of Ocean Vehicles. Wiley, Chichester.
Fossen, T.I., 2002. Marine Control Systems: Guidance, Navigation and Control of Ships,
Rigs and Underwater Vehicles, 1st edition. Marine Cybernetics, Trondheim,
Norway.
Harris, C.J., Hong, X., Wilson, P.A., 1999. An intelligent guidance and control system
for ship obstacle avoidance. Proc. Inst. Mech. Eng. Part IJ. Syst. Control Eng.
213, 311320.
Hilgert, H., Baldauf, M., 1997. A common risk model for the assessment of
encounter situations on board ships. Ocean Dyn. 49 (4), 531542.
Hwang, C.N., Yang, J.M., Chiang, C.Y., 2001. The design of fuzzy collision avoidance
expert system implemented by H-innity autopilot. J. Mar. Sci. Technol. 9,
2537.
Iijima, Y., Hagiwara, H., 1994. Results of collision avoidance manoeuvre experi-
ments using a knowledge-based autonomous piloting system. J. Navig. 47,
194202.
Ito, M., Zhang, F., Yoshida, N., 1999. Collision avoidance control of ship
with genetic algorithms. In: Proceedings of the 1999 IEEE International
Conference on Control Applications, Kohala Coast, Hawaii, USA, August, vol. 2,
pp. 17911796.
Khatib, O., 1986. Real-time obstacle avoidance for manipulators and mobile
robots. J. Robotics Res. 5 (1), 9098.
Koren, Y., Borenstein, J., 1991. Potential eld methods and their inherent limita-
tions for mobile robot navigation. In: Proceedings of the International Con-
ference on Robotics and Automation, Sacramento, California.
Kose, E., Basar, E., Demirci, E., Guneroglu, A., Erkebay, S., 2003. Simulation of
marine trafc in Istanbul strait. Simulat. Modelling Pract. Theory 11 (2003),
597608.
Lee, S.M., Kwon, K.Y., Joh, J., 2004. A fuzzy logic for autonomous navigation of marine
vehicles satisfying COLREGS guidelines. Int. J. Control Autom. Syst. 2 (2), 171181.
Li, J., Wang, F., 1983. The Ship Collision Avoidance at Sea. Peoples Communica-
tions Press, Beijing (in Chinese).
Liu, Y., Shi, C., 2005. A fuzzy-neural inference network for ship collision avoidance.
In: Proceedings of the IEEE Third International Conference on Machine
Learning and Cybernetics, ICMLC2005, Guangzhou, China, pp. 47544759.
Rogers, D.F., Adams, J.A., 1990. Mathematical Elements for Computer Graphics,
2nd edition. McGraw-Hill Publishing Company, New York.
Shi, C., Zhang, M., Peng, J., 2007. Harmonic potential eld method for autonomous
ship navigation. In: Telecommunications, Proceedings of the 7th International
Conference on ITS, 68 June.
Smierzchalski, R., 1996. The decision support system to design the safe manoeuvre
avoiding collision at sea. In: Proceedings of the 14th International Conference
Information on Systems and Synthesis, Orlando, USA.
Statheros, T., Howells, W.G.J., McDonald-Maier, K.D., 2008. Autonomous ship
collision avoidance navigation concepts, technologies and techniques. J. Navig.
61, 129142.
Xue, Y., Lee, B.S., Han, D., 2009. Automatic collision avoidance of ships. Proc.
IMechE Part M: J. Eng. Marit. Environ. 223 (1), 3346.
Yang, S., Li, L., Suo, Y., Chen, G., 2007. Study on construction of simulation platform
for vessel automatic anti-collision and its test method. In: Proceedings of
the IEEE International Conference on Automation and Logistics, Jinan, China,
1821 August.
Y. Xue et al. / Ocean Engineering 38 (2011) 22902305 2305