THE JOURNAL OF NAVIGATION (2002), 55, 117–136.

# The Royal Institute of Navigation

DOI : 10.1017\S0373463301001631 Printed in the United Kingdom

The Integrated Design of Fuzzy

Collision-Avoidance and
H_-Autopilots on Ships
Cheng-Neng Hwang
(National Cheng Kung University, Taiwan)

Collision avoidance remains the most important concern for ships at sea. Despite the
electronic equipment now fitted on ships to support the mariner, expert experience is still
essential when a ship is in danger of colliding with the others. To include these experts’
experiences to resolve the problems of collision, we have designed a fuzzy collision-avoidance
expert system that includes a knowledge base to store facts and rules, an inference engine to
simulate experts’ decisions and a fuzzy interface device. Either a quartermaster or an
autopilot system can then implement the avoidance action proposed in the research. To
perform the task of collision-avoidance effectively, a robust autopilot system using the state
space H_ control methodology has been designed to steer a ship safely for various conditions
at sea in performing course keeping, course-changing and route-tracking more robustly. The
integration of fuzzy collision-avoidance and H_autopilot systems is then proposed in this
paper.
KEY WORDS
1. Collision avoidance. 2. Autopilots. 3. Automation.

1. I N T R O D U C T I O N. In recent years, . shipping has rapidly developed in

marine nations to meet growing economic demands. In order to remedy the shortage
of personnel and to improve the safety of navigation, vessel systems are becoming
more and more automatic and intelligent. In this paper, we use the concepts of the
fuzzy set theory and fuzzy inference method to design a fuzzy collision-avoidance
system. To fit time-varying environments for ship navigation, static obstacles with
no prior position information and moving ships with unknown trajectories are
considered in this study. Fuzzy logic is applied to guide a ship from a starting point
toward the target trajectory without colliding with any obstacles or other ships.
Intuitive actions of human beings are modelled into fuzzy rules such that the ship has
the capability, like human beings, of avoiding obstacles or other moving ships. These
fuzzy rules can be dynamically weighted according to the proximity of the found
obstacles or target ships. Furthermore, the proposed approach can also be used for
navigation of multiple ships, with few modifications to the original algorithm.
When a ship navigates at the sea, the influence of ship speed, the depth of water and
the draft of ship will cause changes in its dynamic properties. Besides, the influence
of the currents, winds and waves will cause extra inputs to this system. These
uncertain factors may also make a closed-loop system unstable. To eliminate the ill
effects of these uncertain factors, we applied H_ theory to design an auto-pilot in this
paper to find an optimal control law such that the system still has certain robustness
118 C H E N G -N E N G H W A N G A N D O T H E R S VOL. 55

Figure 1. Graphical interpretation of DCPA and TCPA.

under the worst exogenous input, while it keeps the closed loop system stable and
ensuring a certain degree of accuracy in tracking target trajectories. This paper is
organized as follows : Section 2 illustrates the development of ship collision-
avoidance, ship and obstacle safety domains, traffic separation schemes and avoid
actions. Section 3 describes the applications of fuzzy theory and basic configuration
of fuzzy logic. Section 4 proposes the design process of an H_ autopilot, including the
theorem for obtaining the state space solution of H_ optimal control. Section 5
presents the computer simulation results for the proposed integrated system, which
is a combination of fuzzy collision-avoidance system and the H_ autopilot system, to
demonstrate the feasibility of the proposed integrated system. Finally, a summary
and main conclusion of this study are described in Section 6.

2. C O L L I S I O N -A V O I D A N C E O F S H I P S . Benefiting from the devel-

opment of modern science, high technology is now widely used in the field of
navigation. Satellite navigation and communication systems have been successfully
applied to minimize the problems faced by the sailor and, as a result, the problem of
collision-avoidance becomes relatively more important. Moreover, collision accidents
are increasing as ships increase in size, in speed and in number. The problem of
collision-avoidance has thus become an urgent issue.
For the above reasons, the major maritime countries of the world have given a
great deal of attention to this problem. One solution to the problem is to establish
navigation regulations, to strengthen traffic control, and to improve the technical
level of seafarer training, as well as to study collision-avoidance systems.
To obtain the maximum economic benefit, newly built ships are tending to be get
bigger and to be operated more automatically. COLREGS (The International
Regulations for Preventing Collisions at Sea) are the legal provisions to co-ordinate
the behaviour of ships when there is a risk of collision at sea. However, many of these
rules are qualitative and can only be used after quantifying the situation. In practice,
this causes some difficulties for sailors in the implementation of the COLREGS. For
this reason, many researchers in western maritime countries began to study the
quantitative methods in the early fifties and sixties. Their results established some
specific terms concerning ship collision such as the distance at the closest point of
approach (DCPA) and the time to reach the closest point of approach (TCPA).
NO. 1 INTEGRATED DESIGN OF FUZZY COLLISION AVOIDANCE 119
Figure 1 describes the concepts of DCPA and TCPA. It can be used to determine
the possibility of collision between two ships and the remaining time for taking
collision-avoidance action to avoid the risk. Note that to avoid a possible collision,
the DCPA and TCPA must be considered simultaneously. The appraisal index
proposed by Kearon (1997) is the weighted sum of the squares of DCPA and TCPA :

λi l (a D C P Ai)=j(b T C P Ai)=
i l 1, 2, 3, ….n,

where :
λi is the appraisal index,
a, b are weightings,
i is the number of target ships.
When λi reaches a preset threshold value, collision-avoidance action must be taken.
Statistic analysis of experimental data is explored by Holmes as an alternative method
to appraise the danger of ship collision. Shimizu uses the method of fuzzy reasoning
and fuzzy control to establish a model for determining the time needed for taking
collision-avoidance action. This research uses basic fuzzy theory to design a fuzzy
collision-avoidance system, which is an effective one. In recent years, because of the
increase of international trade, the current commodities and the flat-top building for
oil exploration have made traffic more complex at the sea. In order to resolve the
increased danger of collision, researchers have begun to use the concepts of ‘ Ship and
Obstacle Safety Domain Theory. ’ The definition of a ship domain proposed by
Goodwin is : ‘ The surrounding affective waters that the navigator of a ship wants to
keep clear of other ships or fixed objects. ’ Basically the shape or the size of a ship
safety domain is affected by the following factors :

(a) Physical factors : the size of ships, traffic density and relative speed, etc.,
(b) Environment factors : weather, visibility, etc., and
(c) Psychology factors : navigator’s work record, etc.

As to the size of a ship safety domain, researchers in many countries have different
results, but they have the same viewpoint, that is ‘ encounters by ships travelling in
different directions have different safety domains because of the difference of their
sizes, speeds, relative positions and directions. ’ The safety domain defined by
Goodwin (1975) can be divided into three sectors :

Sector 1, starboard sector : 0  θ  112:5

Sector 2, port sector : 247:5 θ 360
Sector 3, astern sector : 112:5 θ 247:5

The radius of each sector in Dover Strait and North Sea is listed in Table 1.

Table 1. The value of each sector’s radius in different waters (nm).

Sector 1 Sector 2 Sector 3

Dover Strait 0:82 0:75 0:10

North Sea 0:85 0:70 0:45
120 C H E N G -N E N G H W A N G A N D O T H E R S VOL. 55

Because this ship safety domain is not continuous, it will increase the complexity
in computer simulation. Thus, Davis et al. (1982) improves upon Goodwin’s safety
domain to make the domain boundary continuous. Then, the Japanese scholar Fuji
(Toyoda and Fuji, 1971) completed an experiment on ship domains for large, middle
and small ships, and he realized that a ship’s safety domain has connection with its
length and that the domain is not symmetric. The starboard sector is the largest one,
the port sector is the next and the astern sector is the smallest one. However, the size
of the astern sector increases as a ship increases in length. Thus, considering a
conservative ship safety domain, a circle with radius eight times the ship’s length is
used to describe a ship’s safety domain in this research. If other ships enter the ship’s
safety domain, the proposed fuzzy collision-avoidance system will provide advice to
avoid possible collision.
In recent years, there has been a considerable increase in the number of structures
used for offshore oil exploitation. Thus, an obstacle safety domain should also be
built to avoid collision. The obstacle’s size and shape, the depth of surrounding water
and the draft of the ship determine the safety domain size of an obstacle. For
simulation convenience, we select a circle with radius 1:2 times the obstacle’s length
radius as our obstacle’s safety domain.
The first traffic separation schemes in the world were introduced in the Dover Strait
in 1967. A traffic separation scheme is mainly composed of a separation line, a
separation zone and a course borderline. For various types of terrain, the separation
lines and separation zones can be designed as follows :
(a) Use natural obstacles and geographic features to separate ships travelling in
opposite directions.
(b) Utilize the inshore traffic zone to separate the ships travelling in opposite
directions.
(c) Construct a fan-shaped zone for a port to separate the course lane.
(d) Construct a ring-shaped separation zone close to conjunction points.
The width of a course lane in open waters is set to be about 5-3-5, i.e., both sides
of a course fairway have a width of 5 miles respectively and a width of 3 miles is used
as the centre separation zone, and can be up to 20-10-20 (Gung, 1990).
According to the COLREGS, when a ship encounters others at sea, the burden or
the privilege ship can be identified by the strategy listed in Table 2 (Leo, 1979).

Table 2. The adopted strategy of the privilege ship and the burden ship.

Heading- on Crossing Overtaking By- overtaking

encounter encounter encounter encounter

Privilege ship G G
Burden ship G G G

In any potential collision situation, the navigator faces two questions ; do I face a
collision of collision? If so, should I take avoiding actions and what actions should
be taken while considering all vessels in the vicinity? When an encounter involves a
risk of collision (i.e., the DCPA is less than the radius of the ship’s safety domain),
the actions should be taken to avoid the collision. In Figure 1, ship A is the burden
NO. 1 INTEGRATED DESIGN OF FUZZY COLLISION AVOIDANCE 121
ship and ship B is the privileged one. If both ships keep their current course and speed,
ship B will pass the CPA with a relative speed VBA. Since the DCPA in this case is less
than the distance needed for safety, the navigator of ship A must take action and he
has two choices ; he can either turn right to obtain sufficient DCPA before passing the
CPA or, he can change speed. At a slower speed VAh , the encounter situation is as
shown in Figure 2 and will be safe in the sense that the DCPA is greater than or equal

Figure 2. Graphical interpretation of avoiding action.

to the safety distance. For the cases of crossing traffic lanes in a TSS, other regulations
should be followed ; this is explained in references Gung (1990) and Leo (1979). To
offer proper advice to avoid collision between ships, the collision-avoidance system is
designed based on Fuzzy set theory and is described in the next section.

3. F U Z Z Y C O L L I S I O N -A V O I D A N C E S Y S T E M . During the past dec-

ade, fuzzy logic control has emerged as one of the most active and fruitful areas for
research in the application of fuzzy set theory, fuzzy logic and fuzzy reasoning
(Kearon, 1977). Since fuzzy reasoning can be done in linguistic ways, which can
effectively simplify the complexity in modelling system dynamics especially for non-
linear and ill-defined systems like ships, we used fuzzy logic to design the ship
collision-avoidance system described in this paper. The basic operation of a fuzzy set
can be illustrated as follows :
(a) When U (the universe of discourse) is discrete, a fuzzy set A can be represented
as :
µ (χ ) µ (χ ) µ (χ )
Al A < j A = j … j A n ,
χ χ χn
< =
where : (µA(χi)\χi) represents the relationship of the generic element χi of U and
its grade of membership µA(χi) (Lee, 1999, and Lin and Pun (1994)).
(b) Fuzzy intersection ; the membership function µc(x) of the intersection A E B is
defined for all µ ? U by :
µC(χ) l minoµA(χ), µB(χ)q l µA(χ) F µB(χ)
(c) Fuzzy union ; the membership function µc(x) of the union A D B is defined for
all µ ? U by :
µC(χ) l maxoµA(χ), µB(χ)q l µA(χ) G µB(χ)
122 C H E N G -N E N G H W A N G A N D O T H E R S VOL. 55

(d) Fuzzy complement ; he membership function µA- (x) of the complement of a

fuzzy set A is defined for all µ ? U by :
µA` (χ) l 1kµA(χ)
(e) Fuzzy relation ; if A and B are fuzzy relation in xMy and yMz, respectively, the
composition of A and B is a fuzzy relation denoted by A @ B and the
membership function µC(x, z) of the composition A and B is defined by :
µC(x, z) l µA @ B(x, z) l supomin[ µA(x, y), µB(y, z)]q,
or cij l Goaik F bkjq
k
Based on the above fuzzy operation concepts, the basic configuration of a fuzzy
logic controller (FLC) is proposed and shown in Figure 3, which comprises four

Figure 3. Basic configuration of fuzzy logic controller.

principal components : a fuzzification interface, a knowledge base, an inference engine

and a de-fuzzification interface. The main functions of these four components can be
described as follows :
(a) The fuzzification interface involves the following functions :
(i) It receives the state variables from the plant.
(ii) It transfers the range of values of input variables into corresponding
universes of discourse.
(iii) It performs the function of fuzzification that converts input data into
suitable linguistic values.
(b) The knowledge base consists of a ‘ data base ’ and a ‘ linguistic control rule
base ’ :
(i) The database provides necessary definitions that are used to define
linguistic control rules and fuzzy data manipulation in an FLC.
(ii) The rule base characterizes the control policy and control goals of the
domain experts by means of a set of linguistic control rules.
NO. 1 INTEGRATED DESIGN OF FUZZY COLLISION AVOIDANCE 123
(c) The inference engine is the most important kernel, and it is the decision-making
centre of a FLC, which is designed by simulating a human thinking model.
(d) The de-fuzzification interface performs the following functions :
(i) It yields a non-fuzzy control action from an inferred fuzzy control action.
(ii) It converts the range of values of output variables into corresponding
universes of discourse.

Fuzzification is related to the vagueness and imprecision in a natural language. It

is a subjective valuation to transform measurement data into valuation of a subjective
value. Hence it can be defined as a mapping from an observed input space to labels
of fuzzy sets in a specified input universe of discourse. Since the data manipulation
in a FLC is based on fuzzy set theory, fuzzification is necessary and desirable at an
early stage. In fuzzy control applications, the observed data are usually crisp. A
natural and simple fuzzification approach is to convert a crisp value X into a fuzzy
;
singleton A within the specified universe of discourse. That is, the membership
function µA(x) of A is equal to 1 at the point X and zero at other places.
;
A fuzzy system is characterized by a set of linguistic statements based on expert
knowledge. The expert knowledge is usually as ‘ if-then ’ rules, which are easily
implemented by fuzzy conditional statements in fuzzy logic. Fuzzy control rules have
the form of fuzzy conditional statements that relate the state variables in the
antecedent and process control variables in the consequence. Many experts have
found that fuzzy control rules provide a convenient way to express their domain
knowledge. This explains why most FLC are based on the knowledge and experience
that are expressed in the language of fuzzy ‘ if- then ’ rules. The general form of the
fuzzy control rules in the case of two-input single-output systems is :

IF x is A and y is B THEN z is C
< < <
IF x is A and y is B THEN z is C
= = =
…
IF x is An and y is Bn THEN z is Cn

Where x, y and z are linguistic variables representing the process state variable and
control variable, respectively? An, Bn and Cn are the linguistic values of the linguistic
variables x, y and z in the universe of discourse U, V, and W. In what follows, we
consider some useful properties of the FLC inference engine (Lee, 1990 ; Wan and Ho,
1992 and Klir and Folger, 1988).

Theorem 1
n
(Ah, Bh) @ U Ri  µCh(z) l ( µAh(x), µBh( y)) @ max( µR (x, y, z), … , µR (x, y, z))
i=< x, y, z < n

l sup maxomin[( µAh(x), µBh(y)), µR (x, y, z)], … , min[( µAh(x), µBh(y)), µR (x, y, z)]q
x, y x, y, z < n

l maxo[( µAh(x), µBh( y)) @ µR (x, y, z)], … ,[( µAh(x), µBh(y)) @ µR (x, y, z)]q
x, y, z < n

n n
 l U C hi l U (Ah, Bh) @ Ri
i=< i=<
124 C H E N G -N E N G H W A N G A N D O T H E R S VOL. 55

Figure 4. Graphical interpretation of fuzzy inference under minimum rule.

Theorem 2
For the intersection operation of fuzzy sets, the minimum and the product
methods are formulated as follows :
If µA ×B l µA F µB then
i i i i
(Ah, Bh) @ (Ai and Bi Ci) l [Ah @ (Ai)  Ci] E [Bh @ (Bi  Ci)]

If µA ×B l µA iµB then
i i i i
(Ah, Bh) @ (Ai and Bi  Ci) l [Ah @ (Ai  Ci)]i[Bh @ (Bi  Ci)]

The above two formulae imply that we need to make a combination of the
membership function operation and the logic operation. Because Ai and BiCi is not
easily operated, we partition it into two parts and evaluate them separately.

Theorem 3
If the inputs are fuzzy singletons, namely, Ah l x , Bh l y , based on the minimum
; ;
operation and the product operation rules, we have the following four different
operations.
αFi
F µC (z)
i
αF : µ (z) αF l µA (x ) F µB ( y )
i Ci where : i i ; i ;
αi F µC (z)
:
αi l µA (x ) : µB ( y )
:
i i ; i ;
α:i : µC (z)
i

The above theorems explain the process of fuzzy inference. Figure 4 gives a graphic
interpretation of Theorem 3 in terms of minimum operation rule, while Figure 5
offers a graphic interpretation of Theorem 3 in terms of product operation rule.
Basically, de-fuzzification is a mapping from a space of fuzzy control actions
defined over an output universe of discourse into a space of non-fuzzy control actions.
It is employed because a crisp control action is required in many practical
NO. 1 INTEGRATED DESIGN OF FUZZY COLLISION AVOIDANCE 125

Figure 5. Graphical interpretation of fuzzy inference under product rule.

Figure 6. The general output of fuzzy controller.

applications. At present, the commonly used de-fuzzification strategies may be

described by the method of the centre of area or the mean of maximum (Lee, 1990) :
(a) The Centre of Area Method (COA). The widely used COA strategy generates
the centre of gravity of the possibility distribution of a control action. In the case
of a discrete universe, this method yields :

n
 µ (z )z
i=< C i i
zCOA l .
n
 µC(zi)
i= <
The notation n in the above equation is the number of quantitative levels of the
output.
(b) The Mean of Maximum Method (MOM). The MOM strategy generates a
control action that represents the mean value of all local control actions whose
126 C H E N G -N E N G H W A N G A N D O T H E R S VOL. 55

Figure 7. The flow chart of fuzzy collision avoidance.

membership functions reach the maximum. In the case of a discrete universe, the
control action may be expressed as :

m z
zMOM l  i
i = <m

In the above equation, zi is the support value. At this value, the membership
function reaches the maximum value µC(zi), and m is the number of the support
values. Figure 6 shows a graphic interpretation of the various defuzzification
strategies mentioned above.
These fuzzy approach skills are now applied to design a ship collision-avoidance
system, which is composed of five fuzzy-controlled modules : the detecting obstacle or
NO. 1 INTEGRATED DESIGN OF FUZZY COLLISION AVOIDANCE 127
ship near-far module, the keeping away from the static obstacle module, the avoiding
encountering ship module, the tracking target-course module and the ship speed-
control module. The fuzzy rules of each of the control modules are derived from
human being’s intuitive methods of adopting collision-avoidance actions and how
they apply the above fuzzy theorems and operation processes.
When one ship encounters obstacles or other ships, its radar will detect whether
they are in left front, right front or directly in front of the ship. The data obtained
from radar are then the inputs of the detecting obstacle or ship near-far module. The
output of this module is the nearness degree of the found objects, which is used to
determine each of the weights of the keeping away from static obstacle module, the
avoiding encountering ship module, the tracking target-course module and the speed-
control module. If one of the obstacles or target ships detected by radar is closer to
the ship, it will have a larger weight. When the value is greater than a preset threshold
for any of these modules, which is decided by the ship safety domain and the fuzzy
rules mentioned before, the module would then be initiated to take action to avoid
a possible collision. The design process of fuzzy collision-avoidance is illustrated in
Figure 7.
When own ship is in the danger of colliding with obstacles or encountering ships,
the fuzzy collision-avoidance system proposed above will advise a proper avoidance
action to resolve the risk. The avoidance action can then be implemented by either a
quartermaster or an autopilot system. The objective of H_ control problem is to
obtain an H_ optimal control law such that the transfer function between the
exogenous input and the controlled output is minimum while keeping the closed-loop
system stable. In the next section, we will design an autopilot with H_ theory to
ensure that the ship can avoid collision while keeping a good performance even under
the worst exogenous input. The avoidance action advised by the fuzzy collision-
avoidance system can be implemented by H_ autopilot system. In other words, the
integration of fuzzy collision-avoidance and H_ autopilot systems will be proposed
below.

4. D E S I G N O F H _ A U T O P I L O T. The earlier autopilots are of mechanical

construction and are only able to provide simple control actions. The PID-type
controllers were introduced to replace the mechanical devices ; these electrical
and electronic equipments make the autopilots more flexible but have to be
adjusted manually when the ship’s dynamics or disturbances change. Katebi and
Byrne (1988) used adaptive control theory to develop the adaptive auto-pilot system
whose parameters can be adjusted automatically when the ship’s situation or
disturbances change. A lot of research has been undertaken in the field of artificial
intelligence (AI) and its application to control problems, such as the development of
the AI knowledge-based and expert systems, the neural-networked and self-organizing
fuzzy controllers.
However, when a ship navigates at sea, its performance may be influenced by its
own speed, its draft, the depth of water, the encountering situations and the
surrounding currents, winds and waves, etc., which cause extra inputs to the system
dynamics. These uncertain factors may also cause the system to become unstable. For
this reason, the H_ autopilot system is a feasible alternative because of its excellent
disturbance rejection capability.
H_ control theory considers the worst-case of the inputs. In other words, it ensures
128 C H E N G -N E N G H W A N G A N D O T H E R S VOL. 55

a good performance of the closed-loop system even in the worst situation. H_ norm
in the frequency domain is defined as :

RG RH_ l supσmaxQG ( jω)Q (1)

< ω <
G (s) : Transfer Function’s l jω.
<
The objective of the H_ control problem is to obtain an H_ optimal control law
such that the transfer function between exogenous input and controlled output is
minimum while keeping the whole closed loop system stable. The ‘ Standard Problem ’
is considered in the state space form as follows :

Xc (t) l AX(t)jB u(t)jB w(t)

= <
Y(t) l C X(t)jD w(t) (2)
= =<
Z(t) l C X(t)jD u(t)
< <=
In (2), X(t) ? R , Y(t) ? R , u(t) ? R , w(t) ? Rr and Z(t) ? Rl denote the state, the
n p m
measurement output, the control law, the exogenous input and the control output,
respectively. We suppose that u(t) l KX(t), where K is the controller constant gain
matrix. The exogenous input w(t) typically consists of reference inputs, disturbance,
and sensor noises. The components of the controlled output z(t) are tracking errors,
control efforts, etc. The objective of the H_ control problem is to obtain an H_
optimal control law u(t) ? L [0, _) such that for any exogenous input w(t) in a pre-
=
specified ball Ω of L [0, _), the controlled output Rz(t)R is minimum, i.e., the
= =
transfer function between exogenous input and controlled output is minimum in
L [0, _). By defining Ω in the normalized form :
=
Ω l ow(t)Qw(t) ? L [0, _),RW(t)R  1q, (3)
= =
the theorem proposed by Hwang (1993) can now be applied.
Theorem 1. For a plant in the standard form of (2), suppose (A, B ) is controllable,
=
(C , A) is observable, DT D l I, DT C l 0 (the orthogonal assumption). Then, the
= <= <= <= <
H_ optimal control law u(t) minimizing RZR under the worst exogenous input in a
=
pre-specified ball in L [0, _) is given by :
=
u(t) l kBTK X(t) (4)
= <
where K is the positive definite solution of the Algebraic Riccatic Equation (ARE) :
<
ATK jK AjK (B BTkB BT)K jCTC l 0. (5)
< < < < < = = < < <
If (4) and (5) are under the assumption of a white Gaussian input, u(t) is given by :
u(t) l kBTKX(t), (6)
=
where K is the positive definite solution of the ARE :
ATKjKAkKB BTKjCTC l 0. (7)
= = < <
However, in many cases it is difficult to form the standard problem satisfying the
orthogonal condition, DT C l 0. Therefore, to facilitate use of the H_ approach, the
<= <
orthogonal assumption must be removed. For this reason, the following theorem
leading to a more general solution of the time-varying H_ optimal control problem
is developed.
NO. 1 INTEGRATED DESIGN OF FUZZY COLLISION AVOIDANCE 129
Theorem 2. For a plant in the standard form of (2), suppose (A, B ) is controllable,
=
(C , A) is observable, DT D l I, then, the H_ optimal state feedback control law u(t)
= <= <=
minimizing RZR under the worst exogenous input in a pre-specified ball in L [0, _)
= =
is given by :
u(t) l k(BTK jDT C )X(t) (8)
= < <= <
where K is the positive definite solution of the ARE :
<
0 l (AkB DT C )TK jK (AkB DT C )jK (B BTkB BT)K
= <= < < < = <= < < < < = = <
jCT(IkD DT )(IkD DT )C (9)
< <= <= <= <= <
For a linear time-invariable system P(s) expressed in the following form :

xc S(t) l ASxS(t)jBSu(t)jGSd(t)
(10)
y(t) l CSxS(t)
It can easily be transformed into a standard form as shown in (2) by the formulation
methods mentioned in Hwang (1993). The above theorems can then be applied to
obtain an optimal controller for the proposed autopilot.
Based on the above strategy, the system block diagram of the proposed H_
autopilot system is designed and shown in Figure 8. By integrating the proposed fuzzy
collision-avoidance and H_ autopilot systems, the avoidance action advised by fuzzy
collision-avoidance system can then be implemented by H_ autopilot system.

5. C O M P U T E R S I M U L A T I O N R E S U L T S . In this section, the meth-

odology developed in the previous sections is applied to an oil tanker (Cheng, 1994),
whose system parameters are known, so as to prove the feasibility of the proposed
integrated system. The ship is of length 331 m, width 52 m, mould depth 26 m, mould
draft 20 m and weight 285 944 tons.
The dynamic equation of the ship at 15 knots is :

xc S(t) l ASxS(t)jBSu(t)jGS(t)DS(t)
yS(t) l CSxS(t)
where :
A
k0.2019*10−< k0.1199*10= 0
C

As l k0.3679*10−? k0.3996*10−< 0 (11)

B
0 1 0 D

A
0.1415*10;
C

Bs l k0n9786*10−>
B
0 D
A C
Cs l B
0 0 1 D (12)
A
7.6703*10−?sin 4t
C
0
Gs l 0 1.1069*10−@sin 6t
B
0 0 D

The state vector xs l [v r φ]T represents the sway velocity, the yaw angle velocity
and the yaw angle of the ship, respectively. The control input u(t) denotes the rudder
angle of the ship. Disturbance DS(t) contains the sidelong force and yaw moment
130 C H E N G -N E N G H W A N G A N D O T H E R S VOL. 55

Figure 8. Diagram of the integration of fuzzy collision-avoidance and H_ autopilot system.

induced by sea waves. In simulation, the disturbance is chosen as a unit step input.
The dynamic equation of the rudder can be expressed as :
δ 1
l ,
δC 1jτs
where δ is the output rudder angle, δC is the input rudder signal of the steering system,
and τ is the time constant of the system. Therefore, the plant, which includes the ship’s
model and the steering system, can be rearranged as :
A C A C
A C AS BS A C A C 0
xc S(t) xS(t) GS(t)
l 1 j DS(t)j 1 δC(t)
δc (t) 0 k δ(t) 0
B D
B
τ D
B D B D
B
τ D (12)
A C
xS(t)
y(t) l [CS 0]
B
δ(t) D

To study the course tracking performance of the ship, the servo compensator and
the weighting function are chosen as (AC, BC, CC) l (k0n001, 850, 1) and (A , B , C )
> > >
l (k0n001, 0n22,1), respectively. To ensure good disturbance-rejection capability for
all kinds of exogenous inputs, other than pure white Gaussian signals, the proposed
H_ autopilot formulation is used in this example. It is assumed that the positions, the
NO. 1 INTEGRATED DESIGN OF FUZZY COLLISION AVOIDANCE 131

Figure 9. Suggested path for avoiding obstacles (case 1).

Figure 10. Relative distance from the obstacles for case 1.

heading angles and the speeds of own ship and the encountering ships can be obtained
from the on-line outputs of a Radar, a Loran-C, a GPS, a Doppler Log, an
electromagnetic log or an automatic radar plotting apparatus (ARPA). When a ship
navigates at sea, there are numerous encountering cases, fifty of which are discussed
in detail by Imazu. To demonstrate the feasibility of the proposed integrated system,
three typical and complex encountering cases among these fifty situations are
explored. They represent the encountering cases of avoiding obstacles, crossing a
single lane and crossing double lanes, which are shown in Figures 9, 11, and 13
respectively. They are denoted as the case 1, the case 2 and the case 3 in these figures.
For the encountering obstacles, the computer simulation results reveal that the
proposed fuzzy collision-avoidance system advises a dotted-line ship path as shown
in Figure 9. Figure 10 gives the relative distance between the ship and obstacles, which
clearly indicates that the system can effectively avoid a possible collision with these
obstacles. The path advised by the proposed fuzzy collision-avoidance system is then
executed by the H_ auto-pilot, which achieves a perfect route-tracking as shown in
Figure 15 under the surrounding disturbances while the details of the path tracking
errors are given in Figure 16.
The encountering case shown in Figure 11 shows how the fuzzy collision-avoidance
system instructs the ship 1 to avoid the collision with the ships 2 and 3 in crossing a
single lane. The relative position between the encountering ships shown in Figure 12
132 C H E N G -N E N G H W A N G A N D O T H E R S VOL. 55

Figure 11. Suggested path to avoid collision in the case of crossing a single alley (case 2).

Figure 12. Simulation of crossing single alley for case 2.

Figure 13. Suggested path to avoid collision in the case of crossing double alleys (case 3).
NO. 1 INTEGRATED DESIGN OF FUZZY COLLISION AVOIDANCE 133
indicates that the avoidance action is successful. By applying the H_ autopilot to the
system, Figure 17 shows that an excellent tracking result can be achieved even under
a persistent disturbance mentioned before. The tracking errors of the suggested path
are small and are shown in Figure 18.

Figure 14. Relative position between the encountering ships for case 3.

Figure 15. Path tracking executed by H_-autopilot for case 1.

Figure 16. The tracking errors in case 1.

134 C H E N G -N E N G H W A N G A N D O T H E R S VOL. 55

Figure 17. Path tracking executed by H_-autopilot for case 2.

Figure 18. Tracking errors in case 2.

Figure 19. Path tracking executed by H_-autopilot for case 3.

In Figure 13, ship 1 meets four ships and two obstacles when it intends to cross the
double lanes. The navigating path suggested by the fuzzy collision-avoidance system
is shown in Figure 13, which demonstrates its feasibility of the fuzzy collision-
NO. 1 INTEGRATED DESIGN OF FUZZY COLLISION AVOIDANCE 135

Figure 20. Tracking errors in case 3.

avoidance system in Figure 14. The path advised by the proposed fuzzy collision-
avoidance system is then executed by the H_ autopilot, which can still achieve a good
route-tracking precision as shown in Figure 19 even under various surrounding
disturbances. The details of the path tracking errors are given in Figure 20.

6. CONCLUSION. As international trade increases, vessels tend to be larger and

to be operated more automatically, yet collision accidents are also increasing as ships
increase in size, in speed and in number. The problem of collision-avoidance,
therefore, becomes an urgent issue. The fuzzy collision-avoidance system, cooperating
with navigators and explored in this paper, could be used to avoid collisions between
ships. When own ship is in the danger of colliding with obstacles or target ships, the
fuzzy collision-avoidance system will advise a proper avoidance action to resolve the
risk. Either a quartermaster or an autopilot system can then implement the avoidance
action.
The objective of H_ control problem is to obtain an H_ optimal control law such
that the transfer function between the exogenous input and the controlled output is
minimum while keeping the closed-loop system stable. In this research, we designed
an autopilot with H_ theory to ensure that the closed-loop system still has certain
robustness under the worst exogenous input.
The integration of fuzzy collision-avoidance and H_ autopilot systems is proposed
in this paper. The avoidance action advised by fuzzy collision-avoidance system can
then be implemented by H_ autopilot system.
Computer simulation results reveal that, with the aid of the integration of fuzzy
collision-avoidance system and H_ auto-pilot system designed in this paper, ships can
undertake a proper avoidance action to avoid the risk of collision at the right time
and can track the desired path within allowable range to reach the target port.

ACKNOWLEDGEMENT
The National Science Council, R.O.C., under Grant NSC 85-2611-E-006-024, supported this
work. The authors would like to extend their gratitude to NSC for the support.
136 C H E N G -N E N G H W A N G A N D O T H E R S VOL. 55

REFERENCES
Amerongen, J. V. (1980). An adaptive auto-pilot for track keeping. Ship Operation Automation III,
pp. 105–114.
Buckley, J. W. Siler, and Tucker, D. (1986). A fuzzy expert system. Fuzzy Sets and Systems, Vol. 20,
pp. 1–16.
Chen, S. T. (1994). The H_ control design for linear time-varying system and it’s applications on ships.
Thesis Published at National Cheng Kung University, Tainan, Taiwan, R.O.C.
Davis, P. V., Dove, M. J. and Stockel, C. T. (1982). A computer simulation of multi-ship encounter. This
Journal, Vol. 35, No. 2, pp. 347–352.
Gasos, J., Carcia, M. C. and Garcia, R. (1991). Fuzzy strategies for the navigation of autonomous mobile
robots. Proc. I.F.E.S., pp. 1024–1034.
Goodwin, E. M. (1975). A statistical study of ship domains. This Journal, Vol. 28, No. 3, pp. 328–341.
Gung, W. S. (1990). The research of safety coursing and collision avoidance between ships in traffic
separation schemes under the situations of crossing and overtaking. Thesis published at National Taiwan
Marine University, Taiwan.
Hwang, C. N. (1993). Formulation of H and H_ optimal control problems – a variational approach.
=
Journal of the Chinese Institute of Engineers, Vol. 16, No. 6.
Iwasaki, H. and Hara, K. (1987). A fuzzy reasoning model to decide the collision avoidance action. Journal
of the Japanese Institute of Navigation, Vol. 75, No. 2, pp. 121–129.
Jingsong, Z., Wang, F. and Wu, Z. L. (1992). The development of ship collision avoidance automation.
This Journal, Vol. 45, No. 1, pp. 107–113.
Katebi, M. R. and Byrne, J. C. (1988). LQG adaptive ship auto-pilot. Trans. Inst. MC Vol. 10, No. 4, pp.
187–197.
Kearon, J. (1977). Computer program for collision avoidance and track keeping. Proc. Conference on
Mathematics Aspects of Marine Traffic, pp. 229–242.
King, P. J. and Monolanj, E. H. (1977). The application of fuzzy control system to industrial processes.
Automatic, Vol. 13, No. 3, pp. 235–242.
Klir, J. and Folger, T. A. (1988). Fuzzy sets, uncertainty and information. Prentice-Hall.
Kosko, B. (1992). Neural networks and fuzzy systems. Prentice-Hall.
Lee, C. (1990). Fuzzy logic in control systems. Fuzzy Logic Controller, Part II and IEEE Trans. Systems,
I Man. And Cybernetics, Vol. 20, No. 2, pp. 404–433.
Leo, G. P. (1979). The International Regulations for Preventing Collision at Sea in 1972. Hi Wug Publishing
Company.
Lin, S. C. and Pun, C. P. (1994). Analysis of fuzzy theory. The Third Wave Culture, Inc.
Lin, T. S. (1996). The use and applications of Matlab. Lu Lin Publishing Company.
Maeda, M., Maeda, Y. and Murakami, S. (1991). Fuzzy drive control of an autonomous mobile robot.
Fuzzy and Systems, Vol. 39, pp. 195–204.
Mamdani, H. (1974). Applications of fuzzy algorithms for simple dynamic plant. Proc. IEE, Vol. 121, No.
12, pp. 1585–1588.
Nagai, Y. and Enomoto, N. (1988). Fuzzy control of a mobile robot for obstacle avoidance. J. Inf. Sci.,
Vol. 45, No. 2, pp. 231–248.
Patrick, M. (1996). Graphics and GUIs with MATLAB. CRC Press, Inc.
Sugeno, C. (1985). An introductory survey of fuzzy control. Inform. SCI., Vol. 36, pp. 59–83.
Tanscheit and Scharf, E. M. (1988). Experiments with the use of a rule-based self-organizing controller for
robotics applications. Fuzzy Sets and Systems, Vol. 26, pp. 195–241.
Toyoda, S. and Fuji, Y. (1971). Marine traffic engineering. This Journal, Vol. 24, No. 1, pp. 24–30.
Uan, P. C. (1989). The implications of fuzzy control in industries. The Centre of Chinese Productivity.
Wan, C. H. and Ho, G. S. (1992). Fuzzy engineering. Cheng Wa Technology Library, Inc.
Witt, N. A. J., Sutton, R. and Miller, K. M. (1994). Recent technological advances in the control and
guidance of ships. This Journal, Vol. 47, No. 2, pp. 236–241.
Yang, S. T. and Yei, F. P. (1991). Post modern control and design. O Ya Publishing Company.
Zadeh, L. A. (1965). Fuzzy sets. Information and Control, Vol. 8, pp. 338–353.