Académique Documents
Professionnel Documents
Culture Documents
Abstract: An autonomous navigation algorithm for marine vehicles is proposed in this paper us-
ing fuzzy logic under COLREG guidelines. The VFF (Virtual Force Field) method, which is
widely used in the field of mobile robotics, is modified for application to the autonomous navi-
gation of marine vehicles. This Modified Virtual Force Field (MVFF) method can be used in ei-
ther track-keeping or collision avoidance modes. Moreover, the operator can select a track-
keeping pattern mode in the proposed algorithm. The collision avoidance algorithm has the abil-
ity to handle static and/or moving obstacles. The fuzzy expert rules are designed deliberately un-
der COLREG guidelines. An extensive simulation study is used to verify the proposed method.
Keywords: Modified virtual force field, collision avoidance, marine vehicles, fuzzy logic,
moving obstacles, COLREG.
1. INTRODUCTION mobile robot while the obstacle repels it, as in the case
of electric charges. This method works well for static
The autonomous navigation of marine vehicles is obstacles. Marine vehicles, however, frequently face
gaining increasing attention due to the inherent diffi- moving objects. Moreover, there usually exists a pre-
culties in their manual navigation and control. To this determined track (usually the shortest path between
end, the research reported in the literature generally neighboring points), which the marine vehicle must
uses methods based on modern control theory [2,4,5, follow as precisely as possible. Furthermore, COL-
11,16]. These methods, however, require precise REG1 guidelines must be followed for secure colli-
mathematical models of the dynamic behavior of the sion avoidance. The original VFF method, however,
given marine vehicle and its immediate environment. also fails to offer the flexibility and robustness needed
Conversely, the operating environment of marine ve- to address this concern, which is important in the
hicles is often complicated and dynamic modeling of navigation of marine vehicles.
the vehicles themselves is surrounded in uncertainty.
Therefore, an approach using artificial intelligent and
soft computing may be a promising choice as reported
in the literature [1,3,6,8,13-15]. The approach pro-
posed herein attempts to combine the heuristic per-
spectives offered in these works with the notion of
VFF (Virtual Force Field), which has been success-
fully used in mobile robotics, especially in addressing
Goal
the problem of obstacle avoidance [7,12]. The basic
concept, as shown in Fig. 1, is that the goal attracts the
_________
Manuscript received July 16, 2003; revised January 7, 2004; Obst
accepted January 9, 2004. Recommended by Editorial Board acle
member Jin Young Choi under the direction of Editor Keum-
Shik Hong. This work was supported by the Agency for Defense
Development (ADD) through ADD-00-6-2.
Sang-Min Lee and Kyung-Yub Kwon are with the Dept. of
Control & Instrumentation Eng, Graduate School of Changwon
National University, 9 Sarim-dong, Changwon, Kyungnam, 641- Robot
773, Korea (e-mail: smlee@iname.com, kykwon@changwon.
ac.kr ).
Joongseon Joh is with the Dept. of Control & Instrumentation Fig. 1. The basic concept of the VFF method.
Eng, Changwon National University, 9 Sarim-dong, Changwon,
Kyungnam, 641-773, Korea (e-mail: jsjoh@sarim.changwon. 1
The Rules of the Road International Regulations for Avoiding Col-
ac.kr). lisions
1
at Rules
The Sea: of the Road International Regulations for Avoiding Col-
* Corresponding author. http://www.boatsafe.com/nauticalknowhow/boating/colregs.html.
lisions at Sea:
172 International Journal of Control, Automation, and Systems Vol. 2, No. 2, June 2004
Ship
Track- ψd Course
Keeping δ Ship ψ Ship
Desired Track Keeping
Controller Controller Model Dynamics
ψd
x X0
Ship
→
X0
Fa
δ Obstacl
e
Desired track
r Waypoint 1 Waypoint 2
v u →
Ship ea X0
ψd → S Fa
Y ep
ψ tk
( Max − S) F p
X
Y0 Desired track
Waypoint 1 Waypoint 2
Desired Track
ψ ca
→
Obstacle
Fig. 8. Block diagram for the proposed algorithm.
Fca
Safe region
0 H
Desired path
d
Escape
Dead Yes
zone ? mode
α = 0, β = 1
Fig. 12. Term set for α and β.
No
Track
α then the marine vehicle will move towards the de-
keeping sired path as soon as possible. Therefore, (6) indicates
mode that the emphasis in the motion of the marine vehicle
is to return to the pre-determined desired path when
Course
the marine vehicle has moved too far from the given
keeping path. Alternatively, the next way-point receives more
mode emphasis when the marine vehicle is near the pre-
determined desired path. Fuzzy logic provides a good
blending between these two extreme cases.
Ship
model
4. DETERMINATION OF ψ ca
Fig. 10. Escape motion in the danger region. Basically, the VFF method yields good collision
avoidance performance, particularly for static obsta-
cles. However, collision situations for marine vehicles
β = 1 as illustrated in Fig. 10. In the so called safe re-
are generally complex as noted in COLREG guide-
gion, α and β are, however, are determined via fuzzy
lines. For instance, Rule 8 in COLREG guidelines
rules:
requires strict safety precautions in view of both the
direction of motion of the vehicles involved as well as
R1: IF d is Near THEN α is B and β is S , their relative speed. While it is generally difficult to
R2 : IF d is Middle THEN α is M and β is M , (6) translate these guidelines into a computationally trac-
R3 : IF d is Far THEN α is S and β is B, table strategy, this section is meant to illustrate the fact
that the proposed approach addresses the key aspects
of these guidelines. In particular, Fig. 13 shows the
where the linguistic variable d denotes the shortest
space surrounding a marine vehicle.
distance of the given marine vehicle from the pre-
Representative locations of possible static or mov-
determined desired path and has {Near, Middle, Far}
ing obstacles are marked as circled numbers. The front
its term set as shown in Fig. 11. The term set for α and
half of the space to which the marine vehicle advances
β is {S(mall), M(edium), B(ig)} and the corresponding is divided more precisely than the rear half space since
membership functions are represented in Fig. 12. the majority of collisions occur in this space. The ob-
Note that if α is much greater than β then the ma- stacles can be static or dynamic. The direction of the
rine vehicle will head to the next waypoint as shown moving obstacles can be towards or away from the
in Fig. 6. On the other hand, if β is much greater than marine vehicle. The original VFF is not sufficiently
International Journal of Control, Automation, and Systems Vol. 2, No. 2, June 2004 175
2
Ship
10degs
6 3
5 Obstacle
Fig. 15. Term set for dobs (nm).
0 x X0
ship
y
Ship
5nm
Danger regio
far
H
d
middle
10nm Safe regio
near
0
waypoint 1 waypoint 2
Fig. 17. Term set for θ (degree). Fig. 19. Danger and safe regions.
5. SIMULATIONS
This means that if the distance to the obstacle is where a , b , c , f , W , S , g1 , and g 2 are the ship
‘Near’, the relative velocity between the ship and ob model parameters and given as a = 1.084/min, b =
The proposed algorithm in this paper is able to set the 0.62min, c = 3.553/min, f = 0.86/min, W = 0.067nm/
collision avoiding distance r. (Refer to Fig. 22.) The rad2, S = 0.215nm/min2, g1 = -0.0375nm/min, and g2 =
position of the marine vehicle is the center of the cir- 0 and where nm is the unit of nautical mile and 1 nm
cle of radius r and the circle plays the role of the onset is equal to 1,852 meters.
of the collision avoiding action. This means that the This model is cited from [11]. The angle and rate of
algorithm ignores obstacles outside the circle but it the rudder is limited as –35 ~ +35 degree and –120 ~
activates its collision avoiding action whenever the +120 degree/min for simulation. The radius of the col-
stacle is ‘Small’, the obstacle is located at ‘Left Front’, lision avoiding circle is set to 4 nm. It is assumed that
and the moving object is approaching from ‘Left Side’, there exist only surge, sway, and yaw motions.
then the vector for collision avoidance is directing
22.5 degree. 5.1. Simulations for track-keeping mode
The proposed algorithm in this paper is able to set Track-keeping ability of the proposed algorithm is
the collision avoiding distance r. (Refer to Fig. 22.) investigated in this subsection. Way points used are as
The position of the marine vehicle is the center of the follows,
International Journal of Control, Automation, and Systems Vol. 2, No. 2, June 2004 177
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
1 0.0 0.0 22.5 0.0 0.0 67.5 0.0 0.0 45.0 67.5 0.0 0.0 22.5 0.0 0.0 0.0 22.5 0.0 0.0 67.5 0.0
2 0.0 22.5 45.0 22.5 22.5 67.5 45.0 0.0 67.5 90.0 67.5 0.0 22.5 0.0 0.0 22.5 45.0 22.5 0.0 67.5 45.0
3 22.5 0.0 0.0 22.5 0.0 22.5 0.0 67.5 0.0 0.0 67.5 0.0 22.5 0.0 22.5 0.0 0.0 22.5 0.0 22.5 0.0
4 22.5 45.0 22.5 0.0 45.0 67.5 0.0 67.5 90.0 67.5 0.0 0.0 22.5 0.0 22.5 45.0 22.5 0.0 45.0 67.5 0.0
5 0.0 22.5 0.0 0.0 0.0 67.5 0.0 0.0 67.5 45.5 0.0 0.0 22.5 0.0 0.0 22.5 0.0 0.0 0.0 67.5 0.0
6 0.0 45.0 0.0 0.0 22.5 45.5 0.0 0.0 90.0 0.0 0.0 0.0 0.0 0.0 0.0 45.0 0.0 0.0 22.5 45.0 0.0
7 0.0 45.0 22.5 0.0 45.0 67.5 45.0 22.5 90.0 67.5 0.0 0.0 22.5 0.0 0.0 45.0 22.5 0.0 45.0 67.5 45.0
8 0.0 22.5 0.0 45.0 0.0 0.0 45.0 22.5 22.5 0.0 0.0 0.0 0.0 0.0 0.0 22.5 0.0 45.0 0.0 0.0 45.0
9 22.5 45.0 0.0 45.0 67.5 45.0 0.0 67.5 90.0 0.0 0.0 22.5 0.0 0.0 22.5 45.0 0.0 45.0 67.5 45.0 0.0
10 0.0 45.0 0.0 0.0 45.0 22.5 0.0 0.0 90.0 0.0 0.0 0.0 0.0 0.0 0.0 45.0 0.0 0.0 45.0 22.5 0.0
pected.
26 26
24 24
22 22
20 20
) 18 )
Y(nm)
Y(nm)
m m 18
n( n(
Y 16 Y 16
14
14
12
12
10
10
8
5 10 15 20 25 30 35 40 45 10 15 20 25 30 35
X (nm) X (nm)
16
20
15
18
14
16
) 13 )
Y(nm)
Y(nm)
m m 14
n( n(
Y 12 Y
12
11
10
10
8
9
10 12 14 16 18 20 22 24 26 28 30 15 20 25 30 35 40
X (nm) X (nm)
10.4
11.5
10.2
Obstacle
10 11
9.8 10.5
) Ship
Y(nm)
m
n( 9.6 )
Y(nm)
Y m 10
n(
Ship
9.4 Y
9.5
9.2
9
9
Obstacle
8.5
38 40 42 44 46 48
X (nm) 36 38 40 42 44 46 48 50
X (nm)
Fig. 23. Avoiding an obstacle approaching from the Fig. 25. Avoiding an obstacle approaching from the
Front. right.
12 12
11.5 11.5
11 11
10.5 10.5
) )
Y(nm)
Y(nm)
m m 10
n( 10 n(
Y Ship Y Ship
9.5 9.5
9 9
Obstacle
Obstacle
8.5 8.5
8 8
36 38 40 42 44 46 48 38 40 42 44 46 48 50 52
X (nm) X (nm)
12.5
12.5 Obstacle
Obstacle 12
12
11.5
11.5
11
11
) 10.5
10.5 )
Y(nm)
Y(nm)
m
n( Ship m Ship
n(
Y 10 10
Y
9.5 9.5
9 9
8.5 8.5
8 8
38 40 42 44 46 48 50 52
36 38 40 42 44 46 48 50 52
X (nm)
X (nm)
Fig. 27. Avoiding an obstacle approaching from the Fig. 29. Avoiding an obstacle approaching from the
back left. front left.
13
11
Obstacle
12
10.5
11
Ship Obstacle
10
) Ship
Y(nm)
)
Y(nm)
m 10 m
n( n(
Y Y
9.5
9
9
8
8.5
38 40 42 44 46 48 40 45 50 55
X (nm) X (nm)
Fig. 28. Avoiding an obstacle approaching from the Fig. 30. Passing a moving vehicle.
left.
It is further demonstrated that either static or mov-
6. CONCLUDING REMARKS ing obstacles can be avoided with the proposed algo-
rithm. In particular, obstacles in stationery state or
In this paper, an algorithm for collision avoiding
moving in various directions relative to the given ve-
autonomous navigation of marine vehicles is proposed
hicle are simulated in this study. It is detailed that the
under the COLREGS guidelines. In particular, in view
proposed approach offers a viable means of avoiding
of the advantages and limitations of the so called po-
catastrophic collisions in these cases by strictly fol-
tential field approach, a modified algorithm, i.e. the
lowing the COLREGS guidelines.
MVFF (Modified Virtual Force Field) method is
The complete system involves some two hundred
newly devised. The algorithm incorporates two behav-
fuzzy rules with four linguistic variables in the prem-
ior parameters, namely α and β, as well as a mode
ise part, which presents a certain challenge from a
number S, thus providing a certain level of flexibility
computational standpoint. It is conceivable, however,
in the selection of track-keeping mode. α reflects the that the rule set can be streamlined and hence reduced
desire to approach the next waypoint while β reflects in complexity.
the need to maintain a certain predetermined path.
Typical behavior of the ship in the “danger” and REFERENCES
“safe” regions is implemented by designing fuzzy [1] E. Alfaro-Cid, E. W. McGookin, and D. J.
logic rules yielding α and β. Furthermore, the concept Murray-Smith, “Genetic algorithm optimization
of mode number S, reflects the operator’s selection of of a supply ship propulsion and navigation sys-
return mode, i.e. the degree to which the operator re- tems,” OCEANS, 2001: MTS/IEEE Conference
quires the ship to return quickly to its predetermined and Exhibition, vol. 4, pp. 2645-2652, 2001.
path. This is critical in situations where the ship may [2] R. S. Burns, G. Blackwell, and S. Calvert, “An
be close to the borders of a neighboring territory, i.e. automatic guidance, navigation and collision
the so called “danger” region. avoidance system for ships at sea,” IEE Collo-
International Journal of Control, Automation, and Systems Vol. 2, No. 2, June 2004 181
quium on Control in the Marine Industry, pp. [10] A. Lozowicki and Z. Zwierzewica, “On Adapta-
3/1-3/3, 1988. tion of H ∞ Control To The Ship Trajectory
[3] Y. M. Enab, “Intelligent controller design for the Tracking Problem,” Proc. of the 3rd IFAC Work-
ship steering problem,” IEE Proceedings Control shop on Control Applications in Marine Systems,
theory & Applications, vol. 143, no. 1, pp. 17-24, pp. 131-137, 1995.
1996. [11] A. A. Masoud and M. M. Bayoumi, "Robot navi-
[4] T. I. Fossen, Guidance and Control of Ocean gation using the vector potential approach," Proc.
Vehicles, John Wiley & Sons Ltd., 1994. of IEEE international Conference on Robotics
[5] T. Holzhuter, “LQG approach for the high- and Automation, vol. 1, pp. 805-811, 1993.
precision track control of ships,” IEE Proceed- [12] M. Stamenkovich, “An application of artificial
ings Control theory & Applications, vol. 144, no. neural networks for autonomous ship navigation
2, pp. 121-128, 1997. through a channel,” Position Location and Navi-
[6] X. Hong, C. J. Harris, and P. A. Wilson, gation Symposium, ‘500 Years After Columbus -
“Autonomous ship collision free trajectory navi- Navigation Challenges of Tomorrow’, IEEE
gation and control algorithms,” Proc. 7th IEEE PLANS ‘92, pp. 346-352, 1992.
International Conference on Emerging Tech- [13] T. W. Vaneck, “Fuzzy guidance controller for an
nologies and Factory Automation, ETFA ‘99, vol. autonomous boat,” IEEE Control Systems, vol.
2, pp. 923-929, 1999. 17, no. 2, pp. 43-51, 1997.
[7] K.-Y. Im and S.-Y. Oh, “An extended virtual [14] N. A. J. Witt, R. Sutton, and K. M. Miller, “A
force field based behavioral fusion with neural track keeping neural network controller for ship
networks and evolutionary programming for mo- guidance,” Proc. of the 3rd IFAC Workshop on
bile robot navigation,” Proc. of the 2000 Con- Control Applications in Marine Systems, pp.
gress on Evolutionary Computation, vol. 2, pp. 385-392, 1995.
1238-1244, 2000 [15] R. J. Zhang, Y. B. Chen, Z. Q. Sun, F. C. Sun,
[8] S. B. Lee and I. Kim. “A ship automatic naviga- and H. Z. Xu, “Path control of a surface ship in
tion in dynamic environment using artificial in- restricted waters using sliding mode,” IEEE
telligence,” Proc. of the Fourth International Trans. on Control Systems Technology, vol. 8, no.
Conference on Knowledge-Based Intelligent En- 4, pp. 722-732, 2000.
gineering Systems and Allied Technologies, vol. [16] X. K. Zhang, X. L. Jia, and X. C. Wang, “Robust
1, pp. 418-422, 2000. track-keeping autopilot for ships”, Proc. of the
[9] J. D. Lee and T. F. Sanquist, “Augmenting the 3rd IFAC Workshop on Control Applications in
operator function model with cognitive opera- Marine Systems, pp. 138-142, 1995.
tions: assessing the cognitive demands of techno-
logical innovation in ship navigation,” IEEE
Trans. on Systems, Man and Cybernetics, Part A,
vol. 30, no. 3, pp. 273-285, 2000.
Sang-Min Lee received the M.S. de- Joongseon Joh (M’95) was born in
gree in Control and Instrumentation Hong-Sung, Korea. He received the B.S.
Engineering from Changwon National degree in Mechanical Engineering from
University, Changwon, Korea, in 1998. Inha University, Korea, in 1981, the
He is currently pursuing a Ph.D. His M.S. degree in Mechanical Design and
research interests include intelligent Production Engineering from Seoul
computer systems and control, neural National University, Korea, in 1983,
networks, soft computing techniques, and the Ph.D. in Mechanical Engineer-
etc. ing from the Georgia Institute of
nology in 1991. From 1983~1986, he was with the central
research center of Daewoo Heavy Industries. He was also
with the Agency for Defense Development from 1991~1993.
Kyung-Yub Kwon received the M.S.
Since 1993, he has been with the Department of Control and
degree in the Department of Control
Instrumentation Engineering, Changwon National Univer-
and Instrumentation Engineering from
sity, where he is now a Professor. He received the out-
Changwon National University, Chang-
standing paper award from the Trans. on Fuzzy Systems,
won, Korea, in 2001. He is currently
IEEE in 2001. His research interests include intelligent con-
pursuing a Ph.D. His research interests
trol, development of DSP-based controllers for servo sys-
include intelligent systems, develop-
tems using BLDC motors, and the control of electro-
ment of DSP-based controllers for
hydraulic servo systems.
servo systems using BLDC motors, etc.