THE JOURNAL OF NAVIGATION (2016), 69, 246–256.

© The Royal Institute of Navigation 2015

doi:10.1017/S0373463315000612

Design of Ship Course-Keeping

Autopilot using a Sine Function-Based
Nonlinear Feedback Technique
Xian-ku Zhang and Guo-qing Zhang
(Navigation College, Dalian Maritime University (DMU), Liaoning 116026 China)
(E-mail: [email protected])

Course keeping for ships is the core of automatic navigation in sea transportation. Much work
has concentrated on developing novel control strategies for closed loop systems. We have
turned our attention the other way to improve the control performance of marine autopilots
in this work by “modulating” the control error using a sine function while the construction of
the controller is not changed. The nonlinear feedback signal thus obtained is sent to the con-
troller to replace the control error itself, which used to be the deviation between the output
response and the reference input of the system. Such a control scheme is called “nonlinear
feedback control” hereafter. Theoretical analysis by using a describing function and robust
control theory shows that the same control quality is guaranteed with minor control
actions for the nonlinear feedback scheme. Simulation experiments were carried out for the
ship Yulong of Dalian Maritime University. It is shown that the method postulated in this
paper has advantages of safety and energy saving in navigation; the maximum initial
rudder angle is reduced by 31·2% with satisfactory control effect.

K E Y WO R D S
1. Navigation. 2. Closed-loop gain shaping. 3. Course keeping. 4. Nonlinear feedback.
5. Sine function. 6. Describing function.

Submitted: 16 September 2014. Accepted: 24 July 2015. First published online: 2 September 2015.

1. INTRODUCTION. In recent years, research into linear feedback control has

been richly developed and matured. To compensate the nonlinearity existing in the
real plant, nonlinear control provides a unified approach for control design such as
the Backstepping method, exact feedback linearisation, etc (Krstic et al., 1995). The
value of these strategies is verified by the fact that many papers in the field are pub-
lished. However, few works about nonlinear feedback control are apparent through
searching the Web of Knowledge. In ship motion control, a small rudder amplitude
and slow rudder ratio mean energy saving and abrasion reduction in the steering
motor during manoeuvring. In heavy sea states, steering at large rudder angles can
bring about increased rolling amplitudes, causing dangerous navigation situations
(Johansen and Fossen, 2013). Therefore, the initial rudder angle and the steering
NO. 2 S I N E F U N C T I O N - B A S E D N O N L I N E A R F E E D B AC K T E C H N I Q U E 247
frequency are required to be as small as possible when the control algorithm of auto-
pilot for ships is designed (Aarsaether and Moan, 2010).
Course keeping control for ships can be taken as a benchmark problem in the study
of ship motion control (Fossen, 2011). It is used as a test bed to demonstrate the control
effect when a new control algorithm is developed. In Du et al. (2007), an adaptive
course controller for time-varying parametrically uncertain nonlinear ships with com-
pletely unknown time-varying bounded control coefficients was developed and the
design method did not require any a priori knowledge of the sign of the unknown
time-varying control coefficient. Unlike the existing works (Bian et al., 2009; Ho
et al., 2010), the heading autopilot in Velasco et al. (2013) was investigated based
on an autonomous In-Scale Fast Ferry. The physical control system was implanted
by a PC using WiFi communications, and the results are very valuable for course-
keeping autopilot design. Satpati et al. (2008) presented the design of a robust
course controller for a cargo ship interacting with an uncertain environment using
Particle Swarm Optimisation (PSO) -enabled automated Quantitative Feedback
Theory (QFT). The plant dynamics were described as a second order Nomoto
model with structure parametric variation. Simulation experiments showed the valid-
ity of the algorithm. A ship course-keeping control scheme of Robust Least Squares
Support Vector Machine (RLSSVM) was proposed in Liu et al. (2010), which made
full use of the nonlinear mapping ability, self-learning adaptability and parallel infor-
mation processing of the least squares support vector machine. Combined with the H2/
H∞ robust control method, simulation results showed that the control system pos-
sessed good adaptive ability with regard to sea condition variations. All of the above
research used linear feedback control, including the works on other control tasks,
for example ship collision avoidance (Szlapczynski, 2011) and fault-tolerant control
(Bong, 2015).
A conventional closed-loop system is a linear feedback control scheme; the input
signal led to the controller is proportional to the control error. As a result, for
smaller control error, the control action produced by the controller may be not
enough, on the other hand, for larger control error, the control action may be too
strong to eliminate the error (Thomas et al., 2013). Motivated by this argument, a non-
linear feedback control scheme is considered in this research, for example, a cubic func-
tion of the error is applied to the controller for a test. However, the control effect is not
as much improved as expected in the simulation for an actual ship. When a cubic non-
linear feedback is replaced with a sine function, satisfactory results appear: the control
action is obviously reduced while the control quality is kept almost the same as before.
In this paper, theoretical analysis and simulation experiments for course keeping man-
oeuvring of ships using nonlinear feedback control are developed.

2. CONTROL PROBLEM AND NONLINEAR FEEDBACK TECHNIQUE.

Consider a course keeping problem for ships, the controlled plant G is taken as the
nominal Nomoto model when the controller K is designed using the first order
closed loop gain shaping algorithm without considering the nonlinear feedback
(Zhang and Wang, 2010; Aarsaether and Moan, 2010). A robust controller for a stand-
ard feedback system is solved below under the following three predetermined condi-
tions: the bandwidth frequency of the closed system being 1/T1 (1/T1 should be the
crossover frequency in the strict sense, and is approximately regarded as bandwidth
248 XIAN-KU ZHANG AND GUO-QING ZHANG VOL. 69

frequency for the sake of easy analysis), the largest singular value being unity, and the
high frequency asymptote slope being −20 dB/dec. The frequency spectrum of the
closed-loop system is made equal to the frequency spectrum of a first-order inertial
system with the largest singular value approximately one (Zhang and Wang, 2010), i.e.
1 GK
¼
T1 s þ 1 1 þ GK
The course keeping controller is then solved as
1
KðsÞ ¼ ð1Þ
GT1 s
The ship model being a standard Nomoto model is expressed in Equation (2) where ψ
is the heading angle, δ is the rudder angle, K0 and T0 are the manoeuvrability indices of
the ship.
ψ K0
GðsÞ ¼ ¼ ð2Þ
δ sðT0 s þ 1Þ
To eliminate the steady state error using the closed loop gain shaping algorithm, a
minor constant ɛ (0·01) is added into the denominator of the Nomoto model. Then,
Equation (3) is obtained
K0
GðsÞ ¼ ð3Þ
T0 s2 þ s þ ε
Actually, ɛ can derive the integral effect for the designed controller and it also repro-
duces the effect of uncertain constant disturbance upon the closed-loop system. When
ɛ is too small, the control law may be with the static error or the long setting time.
When ɛ is too large, the system output tracks the reference signal with the overshooting
dynamics. Therefore one should trade off the two terms by selecting the parameter ɛ
properly.
Thus, substitute Equation (3) into Equation (1), according to closed loop gain
shaping algorithm, a linear Proportional-Integral-Differential (PID) controller is
obtained
1 ε 1 T0
KðsÞ ¼ þ þ s ð4Þ
K0 T1 K0 T1 s K0 T1
In actual application, we discover that the settling time is relatively long for ships with
large time constants such as oil tankers etc. The dynamic performance of course
keeping control systems for ships can be improved greatly when a positive constant
ρ is added to the proportional part of the control law, Equation (4). Finally, the
actual controller is presented in Equation (5). The corresponding theoretical analysis
and simulation test are given in reference Zhang (2012).
1 ε 1 T0
KðsÞ ¼ þρþ þ s ð5Þ
K0 T 1 K0 T 1 s K0 T 1
A nonlinear feedback system driven by a sine function is shown in Figure 1. Contrary
to the standard feedback configuration, sin (ω1(r − y)) where ω1 is the design parameter
substituted for (r − y). Note that the block diagram of sin (ω1(r − y)) shown in Figure 1
NO. 2 S I N E F U N C T I O N - B A S E D N O N L I N E A R F E E D B AC K T E C H N I Q U E 249

Figure 1. The block diagram of a nonlinear feedback system.

does not conform to its standard graphical representation. How to find a stable K with
fine control performance in δ = K(r − y) is the main work in the previous research no
matter whether K is linear or nonlinear, but in this section the main work is how to test
the better control performance of the nonlinear feedback control in the form of δ =
K sin (ω1(r − y)) under the same controller K.
The effects of nonlinear feedback on the dynamic and static performance are ana-
lysed by using sin (ω1(r − y)) ≈ ω1(r − y) when the error is small. The demonstration
process is presented in Zhang (2011). In some situations, this assumption may not
be tenable when the error is large. The effects of nonlinear feedback driven by sine
function to the closed system can be analysed by Taylor series expansion:

sinðω1 ðr  yÞÞ ≈ ω1 ðr  yÞ  ðω1 ðr  yÞÞ3 =3!

ð6Þ
þðω1 ðr  yÞÞ5 =5!    

Let the error e = r − y, Equation (6) is simplified up to the third order, then

ω1 3 e3
f ðeÞ ¼ ω1 e  ð7Þ
6
According to Hu (2007), if the error input e of Equation (7) is Asin ω0t, then the output
of the nonlinear system in Equation (7) can be approximated by its first order har-
monic element, and the equivalent frequency characteristics is the describing function
of the nonlinear system.
Let the output of Equation (7) be f(t) under the sine input Asin ω0t, then its output
can be expressed as Equation (8) using its first order harmonic element of the Fourier
series (Ciaurri et al., 2010).

f ðtÞ ¼ A0 þ A1 cos ω0 t þ B1 sin ω0 t ð8Þ

where A0 is the DC component, A1, B1 are the first order harmonic components, and
8
> 1 2π
>
> A1 ¼ ∫0 f ðtÞ cos ω0 tdω0 t;
>
> π
<
1 2π
B1 ¼ ∫0 f ðtÞ sin ω0 tdω0 t;
>
> π
>
>
>
: A ¼ 1 ∫2π f ðtÞdω t
0 0
2π 0
Under the action of sine input signal e in Equation (7), the complex ratio of its first
order harmonic element in the steady state output to its input signal is referred to
250 XIAN-KU ZHANG AND GUO-QING ZHANG VOL. 69

Figure 2. Equivalent block diagram of a nonlinear feedback system.

the describing function which is expressed as N(A).

B1 þ jA1
NðAÞ ¼
A
Equation (7) is an odd function, thus A0 = 0. When e = Asin ω0t

ω1 3 A3 3
f ðtÞ ¼ ω1 A sin ω0 t  sin ω0 t ð9Þ
6
Equation (9) is also an odd function of t, so A1 = 0. Because of the semi-cyclic sym-
metry property of f(t), then
4 π=2
B1 ¼ ∫ f ðtÞ sin ω0 tdω0 t
π 0
 
π=2 ω1 A
3 3
4 π=2
¼ ∫0 ω1 Asin ω0 tdω0 t ∫0
2
sin ω0 tdω0 t
4
π 6
  ð10Þ
4 π ω1 3 A3 3 π
¼ ω1 A    
π 4 6 8 2
B1 ω1 3 A2
NðAÞ ¼ ¼ ω1 
A 8
In the light of the physical meaning of frequency characteristics, the system in Figure 1
is equivalent to the system shown in Figure 2. Effects of the nonlinear feedback driven
by sine function can be analysed as follows.
2.1. Effect on the steady state of the closed loop system. Let the reference input
be a step signal, its amplitude is r, the steady state error to the step input is obtained
directly by the final value theorem as given below:
1 r
eð∞Þ ¼ lim s
s!0 1 þ GKNðAÞ s

r
¼ lim ¼0
s!0 1 K0
1þð þρ Þω1
T1 s sðT0 s þ 1Þ
Therefore the nonlinear feedback driven by sine function has no extra effect to the
steady state of the system.
2.2. Effect on the dynamic performance of the closed loop system. Based on the
block diagram of the nonlinear feedback system shown in Figure 2, the transfer
NO. 2 S I N E F U N C T I O N - B A S E D N O N L I N E A R F E E D B AC K T E C H N I Q U E 251
function from the input r (i.e. the setting course ψr) to the output y of the system (i.e.
the heading angle ψ) can be obtained as Equation (11).
y GKNðAÞ
¼ ð11Þ
r 1 þ GKNðAÞ
For the ship course keeping problem, wave action is a high frequency disturbance
whose frequency spectrum lies in the range of 0·3∼1·25 rad/s. Generally 1/T1 < 0·3
rad/s is taken in Equation (10) to shy away from the wave frequency spectrum.
Suppose the range of course changing is between 0~2π rad, then 0 < N(A) ≤ ω1. The
Loop Shaping algorithm of H∞ robust control theory is a type of open-loop gain
shaping method (Zhang, 2012). Its key point lies in finding a controller K to make
the gains σðGKÞ and σðGKÞ of the open-loop transfer matrix GK satisfy robust per-
formance in the low frequency zone and robust stability in the high frequency zone,
i.e. high gain in the low frequency zone and low gain in the high frequency zone.
The loop shaping algorithm implements the closed-loop performance of the system
through selecting weighting functions to shape the open-loop frequency characteristic
curve, and obtains an acceptable performance/robustness trade-off. According to the
loop shaping theory, if Equation (11) is compared to the closed loop transfer function
GK/(1 + GK) of a standard feedback system, the introduction of N(A) has less effect on
the dynamic performance of the system because of the high gain of GK in the low fre-
quency zone and 0 < N(A) ≤ 1. Thus the proper trade-off between the stabilising per-
formance and robustness can be obtained by selecting the corresponding parameter
setting appropriately.
2.3. Effect on the output of the closed loop system. The transfer function from the
input r to the output δ of the controller (i.e. the rudder angle) is
δ KN ðAÞ
¼ ð12Þ
r 1 þ GKN ðAÞ
According to Hu (2007), K has the first order large gain, GK has the second order large
gain. Similar analysis as for Equation (11) can proceed showing that the introduction
of N(A) (i.e. the parameter ω1) can bring about the benefit of reducing the output of the
controller. In addition, the sine function can bring the maximum and minimum values
of the error input of the controller within ±1. A similar processing technique can be
seen in fuzzy control, neural network and optimising Genetic Algorithms (GA).
Certainly, the precondition is to ensure the input of the nonlinear function ω1e ∈
[−π/2, π/2] to acquire the preferable effect.

3. NUMERICAL EXAMPLES AND DISCUSSIONS. Taking the Yulong train-

ing ship of the Dalian Maritime University as an example, the corresponding ship
particulars are: Length between perpendiculars L = 126 m, Beam B = 20·8 m, dis-
placement ∇ ¼ 14278:1m3 , draught D = 8·0 m, block coefficient Cb = 0·681, distance
from centre of gravity to the origin of x coordinate axis (i.e. the geometric centre) xG =
−3·38 m, ship speed U0 = 15 kn, rudder area Aδ = 18·8 m2. The manoeuvrability
indices of the Nomoto model can be calculated from the above parameters (Zhang,
2012): K0 = 0·48 s−1, T0 = 216·58 s. In this simulation, the linear Nomoto model is
employed to derive the control law. The parameter settings ρ = 2, T1 = 0·3 s are
252 XIAN-KU ZHANG AND GUO-QING ZHANG VOL. 69

Figure 3. Simulation diagram of Simulink.

employed, which makes the effective working bandwidth frequency of the course
keeping controller 1/3 rad/s to avoid overlapping with the wave disturbance range.
The nonlinear feedback parameter is selected as ω1 = 0·25. While the nonlinear math-
ematical model Equations (13) of ship dynamics are considered as the plant to illus-
trate the effectiveness of the proposed control scheme, which can reflect conditions
similar to the real motion of the ship (Fossen, 2011).
8  
>
> ðm þ mx Þu_  m þ my vr  mxG r2 ¼ XH þ XP þ XR
>
>
>
> m þ my v_ þ ðm þ mx Þur þ mxG r_ ¼ YH þ YP þ YR
<
ðIzz þ Jzz Þ_r þ mxG ðv_ þ urÞ ¼ NH þ NP þ NR
ð13Þ
>
> ψ_ ¼ r
>
>
>
> x_ ¼ u cos ψ  v sin ψ
: 0
y_ 0 ¼ v cos ψ þ u sin ψ
Where the rudder servo system is also considered in the simulation, the steering engine
is modelled as a system with single hydraulic circuit analogue control variable (Zhang,
2012), the maximum rudder ratio is ± 5°/s and the saturation rudder angle is ± 35°. u, v,
r, ψ, x0, y0, n, δ denote linear and angular velocities, heading, positions, propeller ro-
tational speed, rudder angle (expressed in rad) respectively; m and Izz are the ship’s
mass and mass moment of inertia, respectively. mx, my, Jzz are added mass and
added moment of inertia. XH, YH,NH, XP, YP, NP, XR, YR, NR are hydrodynamic
forces and moments acting on hull, propeller and rudder, respectively. For more
details about the mathematical model, please refer to the references (Jia and Yang,
1999; Fossen, 2011).
When the ship navigates at sea, the sway motion and heading deviation are caused
mainly by marine environmental disturbances. In the simulation, one considers sea
wind and irregular wind-generated waves. These are simulated by fusing the physical-
based mathematical model. The wind speed (Beaufort No.7) Vwind = 14·25 m/s, wind
direction ψwind = 85°. The Joint North Sea Wave Project (JONSWAP) spectrum is
adapted to produce the corresponding wind-related waves, which has been defined
as an International Towing Tank Conference (ITTC) standard.
The simulation diagram implemented in Simulink is shown in Figure 3, the setting
course is 50°. To provide quantification, three popular performance specifications,
Equation (14) are employed to evaluate the corresponding algorithm. That is Mean
Absolute Error (MAE), Mean Control Input (MCI) and the Total Variation (TV) of
the control. MAE is used to measure the performance of the system response, and
NO. 2 S I N E F U N C T I O N - B A S E D N O N L I N E A R F E E D B AC K T E C H N I Q U E 253

Figure 4. Simulation results for the nominal system: (a) System response, (b) Control input.

MCI and TV measure properties of energy consumption and smoothness. As to the

nominal plant, the simulation results are presented in Figure 4. It is noted that the
control effect of the nonlinear feedback driven by sine function is almost the same
as that in the standard feedback (with the same maximum overshoot and settling
time) while the initial maximum rudder angle is reduced to 24·02° from the original
34·96° with the drop of 31.2%. There is a 41.4% drop in the average rudder angle
which is decreased on average to 0·0826° from 0·1409°. In order to illustrate the
robust performance of the proposed algorithm, the perturbed plant is considered
with the speed variation −50%. The corresponding response dynamics become
worse than that of the nominal system. Figure 5 presents the simulation result for
the perturbed condition. It is clear to note that saturation is generated in the initial
stage using the standard feedback scheme but not under the nonlinear feedback
scheme. The other performance indices of the nonlinear feedback scheme are still su-
perior to that of the standard one. In Figures 4(b) and 5(b), the subfigure is a local
zoom of the plot, which is to show the control dynamics clearly.
1 t∞
MAE ¼ ∫ jrðtÞ  yðtÞjdt;
t∞  t0 t0
1 t∞
MCI ¼ ∫t0 uðtÞdt; ð14Þ
t∞  t0
1 t∞
TV ¼ ∫ juðt þ 1Þ  uðtÞjdt
t∞  t0 t0
254 XIAN-KU ZHANG AND GUO-QING ZHANG VOL. 69

Figure 5. Simulation results for the perturbed system: (a) System response, (b) Control input.

Table 1. Quantitative comparison of the simulation results.

Plant MAE MCI TV

The standard feedback Nominal 0·09753 −0·04320 0·03281

The nonlinear feedback 0·08079 −0·04347 0·01338
The standard feedback Perturbed 0·11655 −0·03889 0·03085
The nonlinear feedback 0·09208 −0·03967 0·01337

Table 1 gives a quantitative comparison of the simulation results, and this verifies the
effectiveness of the nonlinear feedback algorithm. The performance specifications of
MAE and MCI are approximated, the smoothness index of the control input is obvi-
ously lower than that of the standard feedback method. Under the circumstance of
heavy sea state, steering with large rudder angles can increase the amplitude of roll
and thus increase the probability of cargo damage and decrease the comfort index
of seafarers as well as the safety coefficient of the ship. Therefore reducing the ampli-
tude of rudder angle means that the ship will navigate more safely in addition to being
more energy efficient.
In addition, Figure 6 gives the comparisons of the modulating functions N(e) = e,
N(e) = ω1e and N(e) = sin (ω1e). As shown in Figure 6, it can be concluded that: the
control performance of the nonlinear feedback N(e) = sin (ω1e) is equivalent to that
of the linear feedback with an extra constant gain ω1 when the error e = r − y is
small; the performance of the nonlinear feedback is superior to that of the linear feed-
back with an extra constant gain ω1 when the error e is medium; the nonlinear feed-
back technique cannot work effectively when the error e is too large. It is a very
NO. 2 S I N E F U N C T I O N - B A S E D N O N L I N E A R F E E D B AC K T E C H N I Q U E 255

Figure 6. Comparison of the modulating functions (schematic diagram): N(e) = e, N(e) = ω1e and
N(e) = sin (ω1e).

important conclusion that the improvement of control performance is at the cost of the
reduction of the system robustness for both schemes by further simulation experiments.

4. CONCLUSION. In this paper a novel idea is presented, that the control error
between the reference input and the output is modulated by a sine function and
then is fed back to the controller instead of the original direct error, the former is es-
sentially the so-called nonlinear feedback. Nonlinear feedback control can achieve the
same control effect with minor control action under the unchanged control law. Taking
ship Yulong as an example, simulation results are given using the nonlinear feedback
controller driven by the sine function of course error when the set course is 50° under
the wind scale of Beaufort No.7. The rise time is 128 s, the maximum initial rudder
angle is reduced to 24·02° from 34·96°with 31·2% drop, the average rudder angle is
decreased to 0·0826° from 0·1409° with 41·4% drop, while the control effect is
almost the same as that in the linear feedback control. The algorithm has the advan-
tages of energy saving and safety in navigation. The same conclusion can be drawn
when the nonlinear feedback is used in some other examples. Hence the algorithm
has some universality. However, prudent use needs to be made of the nonlinear feed-
back technique when the set value is too large.

AC K N O W L E D G M E N T
This work was supported by the National Natural Science Foundation of China (Grant
No.51109020) and the Fundamental Research Funds for the Central University (Grant
No.2011QN093 and 3132014302).The authors would like to thank Jia Xinle for his insightful
remarks on this note, and the anonymous reviewers for their valuable comments to improve
the quality of this paper.
256 XIAN-KU ZHANG AND GUO-QING ZHANG VOL. 69

REFERENCES
Aarsaether, K.G. and Moan, T. (2010). Adding the human element to ship manoeuvring simulations. The
Journal of Navigation, 63, 695–716.
Bian, X., Wang, Y., Zhang, X. and Xia, G. (2009). Research on simulation of course control for large ship
with wave disturbances in different see conditions. MTS/IEEE Biloxi-Marine Technology for Our Future:
Global and Local Challenges, 1–5.
Bong, S.P. (2015). Neural network-based fault-tolerant control of underactuated surface vessels.
Mathematical Problems in Engineering, Article ID 903759, in press.
Ciaurri, O., Perez, M., Reyes, J.M. and Varona, J.L. (2010). Mean convergence of Fourier-Dunki series.
Journal of Mathematical Analysis and Applications, 372(2), 470–485.
Du, J., Guo, C., Yu, S. and Zhao, Y. (2007). Adaptive autopilot design of time-varying uncertain ships with
completely unknown control coefficient. IEEE Journal of Oceanic Engineering, 32(2), 346–352.
Fossen, T.I. (2011). Handbook of Marine Craft Hydrodynamics and Motion Control. John Wiley & Sons,
New York.
Ho, W.H., Hsieh, C.H. and Chou, J.H. (2010). Optimal course handing control for nonlinear ship maneu-
vering systems, International Journal of Innovative Computing, Information, and Control, 10(6), 4379–4388.
Hu, S. (2007). Automatic control principle (Version No.5). Science Press, Beijing.
Jia, X. and Yang, Y. (1999). Ship Motion Mathematic Model (the mechanism modeling and the identification
modelling), Dalian Maritime University Press.
Johansen, T.A. and Fossen, T.I. (2013). Control allocation—a survey. Automatica, 49(5), 1087–1103.
Krstic, M., Kanellakopoulos, I. and Kokotovic, P. (1995). Nonlinear and Adaptive Control Design. Wiley,
New York.
Liu, S., Song, J. and Li, G. (2010). Robust LSSVM control for ship course-keeping system. Control and
Decision, 25(4) 551–556.
Satpati, B., Bandyopadhyay, I., Koley, C. and Ojha, S.K. (2008). Robust controller design for course
changing/course keeping control of a ship using PSO enabled automated quantitative feedback theory.
TENCON 2008 IEEE Region 10 Conference, 1–6.
Szlapczynski, R. (2011). Evolutionary sets of safe ship trajectories: a new approach to collision avoidance.
The Journal of Navigation, 64, 169–181.
Thomas, B., Sebastian, H. and Knut, G. (2013). Nonlinear model predictive control of a magnetic levitation
system. Control Engineering Practice, 21(9), 1250–1258.
Velasco, F.J., Herrero, E.R., Lopez, E. and Moyano, E. (2013). Identification for a heading autopilot of an
autonomous In-Scale Fast Ferry. IEEE Journal of Oceanic Engineering, 38(2), 263–274.
Zhang, G. and Zhang, X. (2012). Stability analysis and design of integrating unstable delay processes using
the mirror mapping technique. Journal of Process Control, 24(7), 1038–1045.
Zhang, X. (2011). Control Algorithm of Autopilot Driven by Sine Function of Course Deviation. Navigation
of China, 34(1), 1–5.
Zhang, X. (2012). Concise Robust Control for Marine Ships. Science Press, Beijing.
Zhang, X. and Wang, X. (2010). Concise nonlinear robust control of course keeping for ships. ICIC Express
Letters, 4(4), 1263–1268.