sensors

Article
Modeling and Experimental Testing of an Unmanned
Surface Vehicle with Rudderless Double Thrusters
Chunyue Li 1,† , Jiajia Jiang 1, *,† , Fajie Duan 1 , Wei Liu 2 , Xianquan Wang 1 , Lingran Bu 1 ,
Zhongbo Sun 1 and Guoliang Yang 1
1 The State Key Lab of Precision Measuring Technology and Instruments, Tianjin University,
Tianjin 300072, China; 2016202086@tju.edu.cn (C.L.); fjduan@tju.edu.cn (F.D.); wxq1993@tju.edu.cn (X.W.);
lingranbu@tju.edu.cn (L.B.); zbsun@tju.edu.cn (Z.S.); guoliang_yang@tju.edu.cn (G.Y.)
2 The Department of Electronic and Electrical Engineering, University of Sheffield, Sheffield S10 2TN, UK;
w.liu@sheffield.ac.uk
* Correspondence: jiajiajiang@tju.edu.cn; Tel.: +86-152-2289-2283
† These authors contributed equally to this work.




Received: 13 March 2019; Accepted: 29 April 2019; Published: 2 May 2019


Abstract: Motion control of unmanned surface vehicles (USVs) is a crucial issue in sailing performance
and navigation costs. The actuators of USVs currently available are mostly a combination of thrusters
and rudders. The modeling for USVs with rudderless double thrusters is rarely studied. In this
paper, the three degrees of freedom (DOFs) dynamic model and propeller thrust model of this kind
of USV were derived and combined. The unknown parameters of the propeller thrust model were
reduced from six to two. In the three-DOF model, the propulsion of the USV was completely provided
by the resultant force generated by double thrusters and the rotational moment was related to the
differential thrust. It combined the propeller thrust model to represent the thrust in more detail.
We performed a series of tests for a 1.5 m long, 50 kg USV, in order to obtain the model parameters
through system identification. Then, the accuracy of the modeling and identification results was
verified by experimental testing. Finally, based on the established model and the proportional
derivative+line of sight (PD+LOS) control algorithm, the path-following control of the USV was
achieved through simulations and experiments. All these demonstrated the validity and practical
value of the established model.

Keywords: unmanned surface vehicles; rudderless double thrusters; propeller thrust model;
three-DOF dynamic model; system identification

1. Introduction
As an intelligent offshore platform, unmanned surface vehicles (USVs) have in the recent years
been playing an increasingly significant role in the field of intelligent ships. By carrying a variety of
sensors, it is possible for USVs to continuously monitor various marine phenomena such as ocean
waves, tides and marine meteorology in replacement of traditional human-controlled ships and
operations [1,2]. Given their small size, intelligence, and unmanned nature, they can avoid casualties,
and can therefore be applied to antisubmarine and anti-mine warfare, maritime security protection, and
other military fields [3,4]. In recent years, although the technological development of environmental
sensing, wireless communication, positioning and navigation for USVs has made significant progress,
the motion control aspect of USVs has lagged, such as in autonomous navigation and path-following
control. It plays an essential role in some marine applications, such as seawater monitoring for fixed
points, water sampling for fixed paths and even tracking control of the ordinary merchant ship [5–7].
For effective motion control of USVs, it is necessary to establish a mathematical model about the

Sensors 2019, 19, 2051; doi:10.3390/s19092051 www.mdpi.com/journal/sensors

Sensors 2019, 19, 2051 2 of 22

kinematics and kinetics performance of the USVs [8,9], which is the basis for realizing intelligent
control and maneuverability prediction.
There are two kinds of mathematical models for ship motion. One is the Abkowitz model [10,11],
which is also called the global model. It considers the hull, propeller and rudder as an integrated
whole, and expands the hydrodynamic force acting on the hull into the Taylor series of each motion
variable. The other is the so-called MMG model [12–14], which is also known as the separable model.
It was proposed by the Japanese Mathematical Modeling Group (MMG) based on the Abkowitz
model. The MMG model decomposes hydrodynamic forces into three parts of the hull, propeller and
rudder, and takes the interaction between them into account. The Abkowitz model is mathematically
more complete and rigorous, and therefore widely used in the community. Based on the Abkowitz
model, Fossen et al. [15,16] put forward a method of modeling and control for marine craft, including
ships, high-speed craft, semi-submersibles and floating rigs. They conducted a detailed analysis
of the kinematics and kinetics performance of the marine craft. J. Menoyo Larrazabal et al. [17]
formulated a coupled surge-sway-yaw model to guide the USV maneuvering control and dealt with
the rudder control of a USV system. Based upon a simplified 3 degrees of freedom (DOFs) model
and an actuation model, Petr Švec et al. [18] presented a trajectory planning and tracking approach
to follow a differentially constrained target vehicle operating in an obstacle field. Simulation and
experimental studies were conducted to demonstrate the effectiveness of the developed method.
However, these models are all aimed at USVs with the combination of thrusters and rudders.
In the actuator of this kind of USVs, the thrust value of the thruster controls USV velocity, and the
rotation angle of rudder controls USV steering. The mechanical structure of a combination of thrusters
and rudders is complicated, and the rudder used is often plastic, which easily can be damaged by
the wear of the gear during the steering process, and as a result significantly affects the reliability
of the USV. To deal with this problem, the structure of the rudderless double thrusters is employed,
where the thrust values of port and starboard thrusters control the USV velocity, and the differential
thrust controls the steering of the USV. It simplifies the mechanical structure of the USV, and makes
the steering process more flexible. The structure reduces the control of the rudder, and improves the
reliability of the USV. Besides, the energy conversion rate of the structure is higher than the combined
structure of thrusters and rudders. There has been some research focused on USVs with rudderless
double thrusters. Justin E. Manley et al. [4] introduced the development of USVs for nearly 15 years.
The actuators of AutoCat and ROAZ II USVs both adopted the structure of rudderless double thrusters
for hydrological surveys and oceanographic researches [4]. R. Sutton et al. [19–21] analyzed the control
system of Springer USV with rudderless double thrusters and implemented the nonlinear control of its
motion. A. Savvaris et al. [22] presented the development of the C-Enduro USV with rudderless double
thrusters and the collision avoidance algorithms for the sensing and avoidance system. However,
the modeling of this kind of USV has rarely been studied before. In order to study the motion and
dynamic performance and provide guidance for the motion control of the USVs with rudderless
double thrusters, the three-DOF dynamic model and propeller thrust model are established for a 1.5 m
long, 50 kg USV with rudderless double thrusters. The USV has a double hull configuration and a
custom-designed propulsion system, which consists of two propellers. The model parameters are
obtained through system identification.
The main contributions of this paper can be summarized as follows:

(1) The three-DOF dynamic model and propeller thrust model are established and combined for a
USV with rudderless double thrusters. It guides the modeling of general USVs with rudderless
double thrusters.
(2) Since the direction of thrust generated by the propeller is always consistent with the heading of
the USV, which makes the USV unstressed in the lateral direction, the three-DOF model simplifies
the force analysis and model designing of the USV. It combines the propeller thrust model in
which the unknown parameters are reduced from six to two to represent the thrust in more detail.
Sensors 2019, 19, 2051 3 of 22

(3) System
Sensors 2019,identification
19, x is utilized to get the parameters of the three-DOF model and propeller 3 of 22
thrust model.
(4) (4) Based
Based on on
thethe establishedmodel,
established model, thethe path-following
path-following control
controlofofthe
theUSV
USVis is
achieved through
achieved through
PD+LOScontrol
PD+LOS controlalgorithm.
algorithm. ItIt links
links theory
theory with
with practice,
practice,and
andprovides guidance
provides for for
guidance the the
experimental tests.
experimental tests.
This paper is organized as follows. The USV used in this paper is a fully automatic sensing (FAS)
This paper is organized as follows. The USV used in this paper is a fully automatic sensing (FAS)
USV. Its main characteristics are described in Section 2, with emphasis on the propulsion system and
USV.theItssensing,
main characteristics
communicationare anddescribed
guidance, in Section 2,
navigation, andwith emphasis
control on the propulsion
(GNC) system. In Section 3, system
the
and three-DOF
the sensing,dynamic model and propeller thrust model are derived. In Section 4, the acceleration test, 3,
communication and guidance, navigation, and control (GNC) system. In Section
the three-DOF
circle test dynamic
and zigzag model
test and
are propeller
performedthrust modelthe
to obtain aremodel
derived. In Sectionthrough
parameters 4, the acceleration
system
test,identification.
circle test and Thezigzag
accuracytest aremodeling
of the performed andto obtain the results
identification modelare parameters
verified by through
experimentalsystem
identification.
testing. The The accuracyderivative+line
proportional of the modeling and(PD+LOS)
of sight identification
controlresults are is
algorithm verified
utilizedbyto experimental
realize the
path-following
testing. of the USV
The proportional and the results
derivative+line ofof simulations
sight (PD+LOS) andcontrol
experiments are compared
algorithm is utilizedin Section 5. the
to realize
PD stands for proportional and derivative control. The PD controller is based on
path-following of the USV and the results of simulations and experiments are compared in Section 5. the error of the
system,for
PD stands using the proportional
proportional and derivative
and derivative calculations
control. The PD to control the
controller outputon
is based ofthe
the error
system.
of the
Finally, in Section 6, some concluding remarks are given regarding the results shown and possible
system, using the proportional and derivative calculations to control the output of the system. Finally,
future work.
in Section 6, some concluding remarks are given regarding the results shown and possible future work.
2. THE FAS-01 USV
2. THE FAS-01 USV
The FAS-01 USV designed in this paper is 1.5 m long and weighs 50 kg, as shown in Figure 1.
The FAS-01 USV designed in this paper is 1.5 m long and weighs 50 kg, as shown in Figure 1.
The USV adopts a catamaran structure including two main hulls on the right and left sides
The USV adopts a catamaran
symmetrically. structure
The catamaran includingprevents
construction two mainroll
hulls
andonpitch
the right and left
motions andsides
keepssymmetrically.
the USV
The position
catamaran construction
balanced. prevents roll and pitch motions and keeps the USV position balanced.

Figure 1. The FAS-01 USV.

Figure 1. The FAS-01 USV.
It also
It also affords
affords easy
easy manoeuvrabilityand
manoeuvrability and stability
stability under
under excessive
excessiveload
load[23,24]. TheThe
[23,24]. physical
physical
characteristics of the USV are shown in Table
characteristics of the USV are shown in Table 1. 1.

Table
Table 1.1.Physical
Physicalcharacteristics
characteristics of
ofthe
theFAS-01
FAS-01USV.
USV.

Parameter
Parameter Value
Value
Length overall (LO) 1.5 m
Length overall (LO) 1.5 m
Waterline length (L) 1.3 m
WaterlineDraft (D)(L)
length 0.125
1.3 mm
Beam overall (W) 0.9 m
Draftof(D)
Diameter the propeller 0.125
0.27 mm
Distance between two propellers (B) 0.52 m
Beam overall
Mass(W) 0.9 m
50 kg
USV velocity (max) 1.5 m/s
Diameter of the propeller 0.27 m
Sensors 2019, 19, x; doi: www.mdpi.com/journal/sensors
 Distance between two propellers (B) 0.52 m

Mass 50 kg

Sensors 2019, 19, 2051 USV velocity (max) 1.5 m/s 4 of 22

2.1.
2.1. Propulsion
Propulsion System
System
The
The propulsion
propulsion system
system guarantees
guarantees the
the continuous
continuous output
output ofof USV
USV power
power and
and is
is an
an essential
essential part
part
of the USV. It consists of two direct-current (DC) brushless thrusters powered by a 12 V lithium
of the USV. It consists of two direct-current (DC) brushless thrusters powered by a 12 V lithium battery.
battery. The thrusters
The thrusters are equipped
are equipped with
with two two three-blade
three-blade propellers
propellers with awith a diameter
diameter of 0.27
of 0.27 m, which
m, which are
are adjusted by two rotary speed controllers, respectively. The USV has a differential steering
adjusted by two rotary speed controllers, respectively. The USV has a differential steering mechanism
mechanism
and requiresandtworequires
inputs, two
n1 and n2 ton1adjust
inputs, and n2itstoheading,
adjust itswhere
heading, where
n1 and n1 and
n2 are the n2 arepropeller
two the two
speeds inspeeds
propeller Revolutions Per Second
in Revolutions (RPS). The
Per Second two
(RPS). propeller
The speedsspeeds
two propeller control the USV
control velocity,
the USV and
velocity,
the differential
and speedspeed
the differential controls the steering
controls of theof
the steering USV.
the There
USV. is no rudder
There in the propulsion
is no rudder system,
in the propulsion
as shown
system, asin Figurein2.Figure 2.
shown

Figure 2. Block diagram of the propulsion system.

Figure 2. Block diagram of the propulsion system.

Obviously, straight line motion requires port and starboardstarboard thrusters

thrusters to
to run
run at
at the
the same
same speed,
speed,
and the
and thedifferential
differentialthrust is zero
thrust in this
is zero in case
this [19]. c and n
case n[19]. and nd represent
ncd represent the common
the mode
commonand differential
mode and
mode propeller speeds, respectively. They are defined as:
differential mode propeller speeds, respectively. They are defined as:
nn
c c==((nn11 ++n2n)/
2 )2
/2,
, (1)
(1)

nn = ((nn11−−nn2 )/
d d= 2 )2
/2.
. (2)
(2)
The USV
The USV with
with thethe rudderless
rudderless double
double thrusters
thrusters needs
needs toto generate
generate momentum
momentum by by the
the differential
differential
thrust between
thrust betweenthe theport
portand starboard
and starboard propellers
propellersto change the heading.
to change DuringDuring
the heading. this process, the velocity
this process, the
of the USV also changes. So, there is a coupling between the velocity
velocity of the USV also changes. So, there is a coupling between the velocity and the yaw and the yaw rate. In order to
rate. In
eliminate the effects of coupling and maintain the velocity of the USV, Oe Xe Ye must remain constant at
order to eliminate the effects of coupling and maintain the velocity of the USV, Oe XeYe must remain
all times, and Oe Xe Ye changes around zero depending on the direction of the manoeuvre.
constant at all times, and Oe XeYe changes around zero depending on the direction of the manoeuvre.
2.2. Sensing, Communication and GNC System
2.2. Sensing, Communication and GNC System
Adequate sensing capabilities are generally required to enhance the performance of the USV.
The inertial
Adequate measurement unit (IMU)
sensing capabilities areand global required
generally positioning system (GPS)
to enhance as the basicof
the performance sensors
the USV.are
typically
The used
inertial in combination
measurement unit with
(IMU) the andsystem
globaltopositioning
guarantee the USV(GPS)
system remaining
as theinbasic
goodsensors
operating
are
condition,used
typically and in to combination
improve its performance
with the system [25].toThe IMU/GPS
guarantee theisUSV
usedremaining
to estimate inthe
goodposition and
operating
orientationand
condition, of the USV during
to improve sailing. It has
its performance a resolution
[25]. The IMU/GPS of 1 degree
is usedintothe heading.
estimate the The GPS and
position can
provide up of
orientation to 1the m USV
accuracies
duringinsailing.
both latitude
It has and longitude,
a resolution of depending
1 degree inon thecloud coverThe
heading. andGPSsatellite
can
availability.
provide up toBesides, the USV has
1 m accuracies ultrasonic
in both latitude sensors to providedepending
and longitude, information onon surrounding
cloud cover andobstacles.
satellite
Cameras, radar,
availability. sonar,the
Besides, as well
USVashas other kinds of sensors
ultrasonic sensors are optionally
to provide adopted, depending
information on surroundingon the
specific task
obstacles. at hand.radar, sonar, as well as other kinds of sensors are optionally adopted, depending
Cameras,
on the specific
The task at hand.system includes not only wireless communication with the ground
communication
Thestation
control communication
and other system
vehiclesincludes
to perform not only wirelesscontrol,
cooperative communication
but also with the ground
onboard control
wired/wireless
station and other
communication withvehicles
a varietytoofperform cooperative
sensors, actuators, andcontrol, but also onboard
other equipment. Therefore,wired/wireless
the reliability
communication
of the communication with a variety
system of is sensors,
crucial for actuators,
USV. Due and toother
the equipment. Therefore, the reliability
different communication rates and
of the communication
distances required for data system is crucial
and image for USV. between
transmission Due to the the different
USV and communication
the ground control rates and
station,
distances
we adopt arequired
differentfor data and image
communication transmission
scheme. The data between the USV
transmission usesand
433the
MHz ground
radiocontrol station,
frequency (RF)
we adopt a different
communication. communication
The image scheme. The take
and video transmissions dataatransmission
5 GHz wireless uses 433 MHz
bridge. radio frequency
The sensors, actuators,
and other equipment are communicated with the microcontroller through universal asynchronous
Sensors 2019, 19, x; doi:
receiver/transmitter (UART) in order to obtain real-time status informationwww.mdpi.com/journal/sensors
of the USV.
As the most vital component of a USV, the GNC system is generally composed of an onboard
microcontroller and some algorithms. Firstly, the GNC system obtains the past and current states of
(RF) communication. The image and video transmissions take a 5 GHz wireless bridge. The sensors,
actuators, and other equipment are communicated with the microcontroller through universal
asynchronous receiver/transmitter (UART) in order to obtain real-time status information of the USV.
As the most vital component of a USV, the GNC system is generally composed of an onboard
Sensors 2019, 19, 2051and some algorithms. Firstly, the GNC system obtains the past and current states
microcontroller 5 of of
22

the USV (such as position, orientation, velocity, and acceleration), and the environmental information
including
the USV (such the ocean currents
as position, and windvelocity,
orientation, speed from the onboard and
and acceleration), sensors. Secondly, based
the environmental on actual
information
environmental conditions and task requirements, it generates a series of desired
including the ocean currents and wind speed from the onboard sensors. Secondly, based on paths and points or
actual
obtains the desired
environmental paths from
conditions the ground
and task station.it Finally,
requirements, generates thea GNC
seriessystem produces
of desired paths the
andcontrol
points
or obtains the desired paths from the ground station. Finally, the GNC system produces theforward
command to the actuator based on control algorithms to complete the specified steering and control
and achieve
command topath-following andon
the actuator based autonomous navigation
control algorithms of the USV.
to complete the specified steering and forward
and achieve path-following and autonomous navigation of the USV.
3. Three-DOF Dynamic Model and Propeller Thrust Model
3. Three-DOF Dynamic Model and Propeller Thrust Model
3.1. Three-DOF Dynamic Model
3.1. Three-DOF Dynamic Model
The availability of a sufficiently accurate USV model enabling effective control design is
The availability
imperative of a sufficiently
for both control accurate
methodology USV and
design model enabling study
simulation effective control [24].
purposes design is imperative
This requires a
for
priorboth control methodology
investigation design
of both a precise and simulation
mathematical USV study
modelpurposes [24]. This
with reasonable requires
system a prior
parameters.
investigation
Generally, theofstudy
both aof
precise mathematical
a standard USV model
USV dynamic modelwith
canreasonable
be divided system parameters.
into two Generally,
parts: kinematics,
which
the treats
study of aonly geometrical
standard aspectsmodel
USV dynamic of motion,
can beand kinetics,
divided into which is the
two parts: analysis of
kinematics, the forces
which treats
causing
only the motion.
geometrical aspects of motion, and kinetics, which is the analysis of the forces causing the motion.

3.1.1. Kinematic
3.1.1. Kinematic
Within
Within the
the field
field of
of kinematics,
kinematics, an an ocean vessel moves
ocean vessel moves inin six
six DOFs,
DOFs, which
which are
are defined
defined by:
by: surge,
surge,
sway, heave, roll, pitch, and yaw. However, the six-DOF model is complicated. Because
sway, heave, roll, pitch, and yaw. However, the six-DOF model is complicated. Because some DOFs some DOFs in
the USV model have inherent stability, the model can be simplified. The buoyancy
in the USV model have inherent stability, the model can be simplified. The buoyancy of the USV of the USV stabilizes
the regular
stabilizes themotion
regularsomotion
that heave can
so that be ignored.
heave BecauseBecause
can be ignored. of the sufficient longitudinal
of the sufficient and lateral
longitudinal and
metacentric height of the USV, the motion of roll and pitch can also be ignored. Therefore,
lateral metacentric height of the USV, the motion of roll and pitch can also be ignored. Therefore, the the six-DOF
model
six-DOF can be simplified
model to a three-DOF
can be simplified model to describe
to a three-DOF model to thedescribe
planar motions of USV
the planar in surge,
motions of USVsway
in
and yaw
surge, [25],and
sway as shown
yaw [25],in Figure
as shown3. in Figure 3.

Figure 3. Schematic
Schematic diagram
diagram of
of three
three DOF of the USV.

To
To determine
determinethe theequations
equations of motion, twotwo
of motion, reference coordinate
reference systems
coordinate are considered:
systems the inertial
are considered: the
or fixed to earth frame O X Y O
thatX Y
may be taken to coincide with the USV fixed
inertial or fixed to earth frame e e e that may be taken to coincide with the USV fixed coordinates
e e e coordinates in some
initial condition and the body-fixed frame Os Xs Ys in Figure 3. Since the motion of the earth hardly
in some initial condition and the body-fixed frame Os XsYs in Figure 3. Since the motion of the earth
affects the USV (different from air vehicles), the earth-fixed frame Oe Xe Ye can be considered to be
hardly affects
inertial. The bodythe USVaxes(different
Os Xs andfrom Os Ysair vehicles),
coincide thethe
with earth-fixed
principalframe
axes ofOeinertia
XeYe canand be are
considered
usually
defined as follows:
to be inertial. The body O X is
s axes
s theOs Xs and OY
longitudinal axis (directed from aft to fore);
s s coincide with the principal axes of
O Y is
s inertia
s the transverse axis
and are usually
(directed to starboard).
defined as follows: Os Xs is the longitudinal axis (directed from aft to fore); OY s s is the transverse axis
The typical USV kinematic model [16] in planar motion and without the presence of disturbances
(directed to starboard).
can be expressed as:
.
Sensors 2019, 19, x; doi: η = J(η)υ www.mdpi.com/journal/sensors (3)
T
where η = [ x y ψ ] is the position (x, y) and yaw (ψ) of the USV in the earth-fixed frame,
. . . . T . . .
η=[ x y ψ ] describes the vehicle linear North(x), East( y), and Z-axis angular velocities(ψ) in the
Sensors 2019, 19, 2051 6 of 22

T
earth-fixed frame, υ = [ u v r ] is the vehicle surge velocity (u), sway velocity (v) and yaw rate (r)
and the transformation matrix J(η), related to yaw (ψ), is given by:

 cos ψ − sin ψ 0 
 

J(η) =  sin ψ cos ψ 0 

 
(4)
 
0 0 1

3.1.2. Kinetic
In addition to the kinematic model, it is also necessary to analyze the force of the USV to set up
the kinetic model. When designing the control system of the USV, there are clear advantages using the
vectorial model [5] instead of the component forms of the Taylor-series expansions in the Abkowitz
model [8,9]. The main reasons are that system properties such as symmetry, skew-symmetry and
positiveness of matrices can be incorporated into the stability analysis [16]. The kinetic model of the
USV is given by
.
Mυ+C(υ)υ+D(υ)υ=τ+τE (5)
 
 m − Xu. 0 −my g 
 
M =  0 m − Yv. mx g − Yr.  (6)
 
−my g mx g − Nv . Iz − Nr .

 Y . +Nv. 
−m ( x r + v ) + Y .v+ r
0 0 g v 2 r 
 
 
C(υ) =  0 0 ( m − Xu ) u
. 
 (7)
 Yr. +Nv.
m(x g r + v) − Yv v − 2 r −(m − Xu )u 0
. .

   
 Xu 0 0   Xu|u| |u| 0 0 
   
D(υ) = D+Dn (υ) = − 0 Yv Yr  −  0 Yv|v| |v| + Yv|r| |r| Yr|v| |v| + Yr|r| |r|  (8)
   
0 Nv Nr 0 Nv|v| |v| + Nv|r| |r| Nr|v| |v| + Nr|r| |r|
where M is the mass matrix, and x g and y g are the coordinates of the USV center of gravity in the
body-fixed frame. The model designates the body-fixed origin at the center of gravity and assumes
horizontal symmetry, leading to x g = 0 and y g = 0; Iz denotes the moment of inertia about the Zs -axis.
The notation of SNAME [26] for hydrodynamic derivatives is used in this expression. For instance,
.
the hydrodynamic added mass force Y along the y-axis due to the acceleration u in the x-direction is
written as
. .
Y = −Yu. u Yu. := ∂Y/∂u (9)

where C(υ) is the Coriolis and centripetal matrix, D(υ) is the drag matrix, including the linear drag
term (D) and the nonlinear drag term (Dn (υ)), τ is the vector of forces and moments generated by the
h iT
propulsion system, and τE = τuE τvE τrE is the vector of forces and moment caused by the
disturbance.
For the USV with rudderless double thrusters, the thrust generated by the port and starboard
thrusters is always in the same direction as the heading of the USV, and there is no force generated by
the rudder. Therefore, the vector τ can be expressed as
h iT
τ= τu 0 τr (10)

where τu = XP1 + XP2 . XP1 , Xp2 denote the thrust from the port and starboard thrusters, respectively.
Torque by the port and starboard thrusters is given by

Mr = XP1 × (lp i + dp j) + XP2 × (−lp i − dp j) (11)

Sensors 2019, 19, 2051 7 of 22

where i and j are unit vectors in the Xe and Ye directions. dP is the transverse distance from the
centerline of the USV to the centerline of each thruster, which equals 0.26 m in the paper. lp is the
longitudinal distance from the thruster to the center of gravity. Because the direction of the torque by
the port and starboard thrusters is both in the Zs -axis, to simplify the calculation, the scalars of the
torque are taken in the model. τr is the magnitude of Mr , that can be expressed as;
q dp
τr = (XP1 − XP2 ) · lp 2 + dp 2 q = (XP1 − XP2 ) · dP (12)
lp 2 + dp 2

It can be seen that the propulsion of the USV is provided entirely by the resultant force generated by
the port and starboard thrusters and that the rotational moment is related to the differential thrust and
the distance of two thrusters. This cancels the function of the rudder to make the calculation simpler.
Modeling analysis of the propeller thrust is detailed in the next part.
In order to simplify controller design, we make the following assumptions of the kinetic model:

(1) When the USV is slow (the maximum velocity of the USV is 1.5 m/s.), the effect of nonlinear drag
term (Dn (υ)) can be ignored.
(2) Because the effect of off-diagonal terms on the USV’s dynamics is smaller than the diagonal terms,
the off-diagonal terms of M and D can be ignored.
(3) Assuming the coincident center of added mass and gravity Nv. can be replaced by Yr. .
A combination of approximate fore-aft symmetry and light draft suggests that the sway force
arising from yaw rotation and the yaw moment induced by the acceleration in the sway direction
are much smaller than the inertial and added mass terms. Therefore, in the Coriolis and centripetal
matrix C(υ), Nv. = Yr. = 0.
(4) Assuming the USV sails in a calm environment, the environmental disturbances τE can
be neglected.

Based on these assumptions, the kinetic model can be simplified to

.
Mυ+C(υ)υ+D(υ)υ=τ (13)
 
 m − Xu. 0 0 
 
M =  0 m − Yv. 0 
 (14)

0 0 Iz − Nr.

 
 0 0 −(m − Yv. )v 
 
C(υ) =   0 0 (m − Xu. )u 
 (15)

(m − Yv. )v −(m − Xu. )u 0

 
 Xu 0 0 
 
D(υ) = D = − 0 Yv 0  (16)
 
0 0 Nr
Therefore, the three-DOF model of the USV with rudderless double thrusters can be expressed by
( .
η = J(η)υ
. (17)
Mυ+C(υ)υ+D(υ)υ=τ
Sensors 2019, 19, 2051 8 of 22

h iT
where τ = XP1 + XP2 0 (XP1 − XP2 ) · dP . The expanded form of three-DOF equations of
motion is: .
x = u cos ψ − v sin ψ



 .
y = u sin ψ + v cos ψ



.


ψ=r



(18)

 .


 (m − Xu. )u − (m − Yv. )vr + Xu u = XP1 + XP2
 .
(m − Yv. )v + (m − Xu. )ur + Yv v = 0




 .
(Iz − Nr. )r − ((m − Xu. ) − (m − Yv. ))uv + Nr r = (XP1 − XP2 ) · dp



By designing the three-DOF model of the USV with rudderless double thrusters, it reduces the
analysis of the force generated by the rudder. Since the direction of thrust generated by the propeller is
always consistent with the heading of the USV, which makes the USV unstressed in the lateral direction,
it also simplifies the force analysis and model designing of the USV.

3.2. Propeller Thrust Model

The thrust generated by the rudderless double thrusters can be decomposed into two forces:
the longitudinal force to keep the USV moving forward and the steering moment to change the heading
of the USV. In USVs with a combination of thrusters and rudders, the steering moment is the effect of
the rudders. However, in this paper, the USV is not equipped with a rudder. The differential thrust
generated by double thrusters produces the steering moment. It simplifies the mechanical structure of
the USV, and makes the steering of the USV more flexible. Besides, it greatly improves the reliability
and energy conversion rate of the USV.
The propeller thrust model can be expressed as [25]

XP = (1 − tP )ρD4 KT ( J0 )n|n| (19)

where ρ is water density, D is the diameter of the propeller, and tP represents the coefficient of thrust
reduction. When the propellers are working behind the USV, the water flow velocity at the stern is
increased due to the suction effect, resulting in an increase in the hull pressure resistance and frictional
resistance. The thrust portion used to offset this increased resistance ∆T is the thrust deduction
tP = (XP − ∆T )/XP . The advance ratio J0 is given by

J0 = V (1 − ω)/(nD) (20)
√
where V = u2 + v2 is the velocity of the USV, ω is the wake fraction number with ω < 1 accounting
for the relative reduction in flow velocity into the propeller due to the USV’s wake, and n is the
propeller speed. In Equation (17), KT is a thrust coefficient depending on the advance ratio J0 which is
well-modeled as a linear function [27,28], given by

KT = α1 + α2 J0 (21)

where α1 and α2 are two constants.

Based on Equations (19–21), the propeller thrust XP can be expressed as

XP = cVn + d|n|n (22)

where c = (1 − tP )(1 − ω)ρD3 α2 , and d = (1 − tP )ρD3 α2 .

From Equation (22), we can obtain the propeller thrust model by identifying two parameters c and
d. It is simpler than identifying six parameters ω, ρ, D, α1 , α2 , tP in Equations (19–22). This simplifies
the model and reduces the workload of system identification.
 simplifies the model and reduces the workload of system identification.

4. System Identification and Experimental Validation

The goal of system identification is to obtain a simplified understanding of how the vehicle
Sensors 2019, 19, 2051 9 of 22
behaves, so that the control algorithm can accurately calculate control outputs to produce the desired
vehicle response. System identification that requires a specific set of input time histories, extracts
4.thrust
System and hydrodynamic
Identification parameters from
and Experimental standard motion measurements, which include the
Validation
acceleration test, circle test and zigzag test [29,30]. These tests are all open loop maneuverability tests.
The goal of system identification is to obtain a simplified understanding of how the vehicle
The data can be obtained from these tests by setting the speeds of the port and starboard propellers
behaves, so that the control algorithm can accurately calculate control outputs to produce the desired
manually. System identification results are presented in this section and the three-DOF dynamic
vehicle response. System identification that requires a specific set of input time histories, extracts thrust
model and the propeller thrust model are determined.
and hydrodynamic parameters from standard motion measurements, which include the acceleration
test,
4.1.circle testPerformance
Thruster and zigzag Experiment
test [29,30]. These tests are all open loop maneuverability tests. The data
can be obtained from these tests by setting the speeds of the port and starboard propellers manually.
SystemInidentification
order to achieve motion
results control ofin
are presented the USV,
this we need
section to know
and the the relationship
three-DOF between
dynamic model andmotor
the
commands
propeller andmodel
thrust the thrust on the USV. For this test, the USV was tied to a tension meter, connected
are determined.
to a fixed pole. Port and starboard thrusters were controlled by the same motor commands to apply
4.1.
theThruster Performance
thrust uniformly in Experiment
the surge direction. When the line connecting the tension meter and the USV
wasIntight, that is, when
order to achieve motion the tension
controlwas stable,
of the USV, the tension
we need meter the
to know reading could be
relationship obtained.
between motorThe
commands and the thrust on the USV. For this test, the USV was tied to a tension meter, connected to the
reading was the total thrust of the propulsion system. Three experiments were performed under a
same experimental conditions, and the average of three readings of the tension meter was used as
fixed pole. Port and starboard thrusters were controlled by the same motor commands to apply the
the final experimental result.
thrust uniformly in the surge direction. When the line connecting the tension meter and the USV was
The corresponding thrust under different motor commands is listed in Table 2, and the
tight, that is, when the tension was stable, the tension meter reading could be obtained. The reading
relationship between motor commands and thrust is shown in Figure 4.
was the total thrust of the propulsion system. Three experiments were performed under the same
experimental conditions, and the average of three readings of the tension meter was used as the final
Table 2. The corresponding thrust values under different motor commands.
experimental result.
The corresponding thrust under different motor commands is listed in Table 2, and the relationship
Motor
between motor
command −100 commands
−90 −80and thrust
−70 is
−60shown −50in Figure
−40 4. −30 30 40 50 60 70 80 90 100
(%)
Table 2. The corresponding thrust values under different motor commands.
ThrustMotor −110
(N) Command −88
(%) −64−90
−100 −45
−80 −30 −60 −14
−70 −50 −40−8 −30 −130 240 950 1560 3970 5480 120
90 177
100 227
Thrust (N) −110 −88 −64 −45 −30 −14 −8 −1 2 9 15 39 54 120 177 227

Sensors 2019, 19, x; doi: Figure 4. The relationship between motor commands and thrust.
www.mdpi.com/journal/sensors

From Figure 4, when the motor commands are from −30% to 30%, the thrust of port and starboard
thrusters is zero, which can be called the “dead zone”. As shown in Table 2, we can find that under the
same motor commands, the propeller thrust is different when the propeller rotates forward and reverse,
and the thrust is larger under the forward rotation than the reverse rotation. The different performance
of the thrusters makes the velocity and steering ability different when moving forward and backward.
Sensors 2019, 19, 2051 10 of 22

4.2. Propeller Thrust Model Identification

In straight line motion v = r = 0, V = u, and the three-DOF dynamic model can be simplified to
a one-DOF model [27]. According to the propeller thrust model in Section 3, the one-DOF model is
given by:
.
(m − Xu. )u − (m − Yv. )vr + Xu u = XP1 + XP2 (23)

When applying the same motor commands to the port and starboard thrusters, the thrusts of
them are the same. Because the performance of port and starboard thrusters is the same, in straight
line motion, the thrust XP1 = XP2 , XP1 + XP2 = 2XP . Then, Equation (21) can be further expressed as:

. Xu 2XP
u= u+ (24)
( m − Xu )
. (m − Xu. )

When the USV is in the steady state with constant velocity u0 ,

Xu 2XP0
0= u0 + (25)
(m − Xu. ) (m − Xu. )

substituting perturbation values u = u0 + ∆u, XP1 = XP0 + ∆XP into Equation (22),

. Xu 2(XP0 + ∆XP )
∆u = (u0 + ∆u) + (26)
(m − Xu. ) (m − Xu. )

According to Equation (23), in the initial limit,

. ∆Xp c∆u∆n + d|∆n|∆n

u(0) = 2 =2 (27)
(m − Xu. ) (m − Xu. )

When the USV is stationary, i.e., u0 = 0, (25) can be simplified to

.
u(0) = e
cun + d|n|n
e (28)

2c 2d
where: ec = (m−X . , d = (m−X . ) .
e
u) u
Therefore, through collecting initial acceleration, surge velocity and propeller speed, we can
obtain the parameters e c and debased on the recursive least squares method. The detailed steps are
shown in Figure 5. The initial value of repeated test number n is 0.
Initial acceleration, surge velocity and propeller speed were recorded during execution of the
steps to extract ec and dein Equation (28). Figure 6 shows the fit of the relationship among the initial
acceleration, surge velocity, and propeller speed, where it can be seen that the fitting result is satisfactory
when ec = −6.402 × 10−6 (s−1 · RPS−1 · kg−1 ), de= 2.014 × 10−4 (ms−2 · RPS−2 · kg−1 ). Thus, we can obtain
the propeller thrust parameters c and d further.
Sensors 2019, 19, 2051 11 of 22
Sensors 2019, 19, x 11 of 22

Figure 5. The flow diagram of the test for collecting initial acceleration, surge velocity and propeller speed.

Initial acceleration, surge velocity and propeller speed were recorded during execution of the
steps to extract c and d in Equation (28). Figure 6 shows the fit of the relationship among the initial
acceleration, surge velocity, and propeller speed, where it can be seen that the fitting result is
satisfactory when c = −6.402 ×10 (s ⋅ RPS ⋅ kg ) , d = 2.014 ×10 (ms ⋅ RPS ⋅ kg ) . Thus, we can
−6 −1 −1 −1 −4 -2 -2 −1

Figure
obtain 5. The flow
the propeller thrustdiagram of thec and
parameters test dfor collecting initial acceleration, surge velocity and
further.
Figure 5. The flow diagram of the test for collecting initial acceleration, surge velocity and propeller speed.
propeller speed.
Initial acceleration, surge velocity and propeller speed were recorded during execution of the
steps to extract c and d in Equation (28). Figure 6 shows the fit of the relationship among the initial
acceleration, surge velocity, and propeller speed, where it can be seen that the fitting result is
satisfactory when c = −6.402 ×10 (s ⋅ RPS ⋅ kg ) , d = 2.014 ×10 (ms ⋅ RPS ⋅ kg ) . Thus, we can
−6 −1 −1 −1 −4 -2 -2 −1

obtain the propeller thrust parameters c and d further.

Figure
Figure 6.
6. Relationship
Relationship of
of initial
initial acceleration,
acceleration, surge
surge velocity,
velocity, and
and propeller speed.
propeller speed.

Three-DOF Model Identification

4.3. Three-DOF
Using the estimate of the added
added mass
mass terms
terms presented
presented in [31,32],
[31,32], the
the hydrodynamic
hydrodynamic added mass
terms in the three-DOF model can be expressed as:

m m11==mm−−XX.u ≈
≈mm+
+0.05m ==50.05kg
0.05m 50.05kg (29)
(29)
Figure 6. Relationship11of initial acceleration,
u surge velocity, and propeller speed.
m22 = m −Yv ≈ m + 0.5(ρπ D2 L) = 84.36kg (30)
4.3. Three-DOF Model Identification 2
m22 = m − 2Yv ≈ 2m + 0.5(ρπD2 L) = 84.36kg
. (30)
Using the estimate m( L + W ) + 0.5(0.1mB +inρπ D2 L3 ) the hydrodynamic added mass
m33of=the
I −added
Nr ≈m(Lmass terms presented [31,32],
2 + W 2 ) + 0.5(0.1mB2 + ρπD2 L3 ) = 17.21kg (31)
terms in the three-DOF m33 =model ≈ be expressed as: 12
I − Nr. can = 17.21kg (31)
12
According to the empirical value of−the
m11 = m Xu ≈added mass,
m + 0.05 the propeller thrust parameters can be
m = 50.05kg
According to the empirical value of the added mass, (29)
the propeller thrust parameters can be
obtained as follows:
obtained as follows:
m22 = m −. Yv ≈ m + 0.5(ρπ D2−4
L) = 84.36kg (30)
Sensors 2019, 19, x; doi: c =e c(m − Xu )/2 = −1.60 × 10 (s−1 · RPS−1 ) www.mdpi.com/journal/sensors (32)
m( L2 + W 2 ) + 0.5(0.1mB2 + ρπ D2 L3 )
m33 = I − Nr ≈ = 17.21kg (31)
d = de(m − Xu. )/2 = 5.0412 × 10−3 (ms−2 · RPS−2 ) (33)
According to the empirical value of the added mass, the propeller thrust parameters can be
obtained as follows:

Sensors 2019, 19, x; doi: www.mdpi.com/journal/sensors

 c = c(m − Xu ) / 2 = −1.60 ×10−4 (s−1 ⋅ RPS−1 ) (32)

Sensors 2019, 19, 2051 d = d (m − X u ) / 2 = 5.04 ×10−3 (ms-2 ⋅ RPS-2 ) 12(33)

of 22

In order to identify the unknown hydrodynamic parameters Xu ,Yv , Nr in the three-DOF model,
In order
a series to identify the
of experimental unknown
tests hydrodynamic
were conducted parameters
for system Xu , Yv , Nr including
identification, in the three-DOF model, a
the acceleration
series of experimental tests
test, circle test and zigzag test.were conducted for system identification, including the acceleration test,
circleIn test and zigzag
order test. the linear drag coefficient X in the surge direction, the acceleration test
to estimate u
In order to estimate the linear drag coefficient Xu in the surge direction, the acceleration test
was conducted. The USV was accelerated from the velocity of zero to a different velocity under the
was conducted. The USV was accelerated from the velocity of zero to a different velocity under the
specified motor command, and sailed at a steady-state velocity for about 60 s. The range of motor
specified motor command, and sailed at a steady-state velocity for about 60 s. The range of motor
command was
command was 60%
60% to
to 100%.
100%. When
When the
the USV
USV achieved
achieved steady-state
steady-state velocity,
velocity, thethe drag force FFu =
drag force =XXuuu
u u
in the
in the surge
surge direction
direction was
was equal
equal to
tothe
thethrust.
thrust. Therefore,
Therefore, wewe could obtain XXuu=
could obtain = 151.46 by fitting
151.46 by fitting the
the
relationship
relationshipof ofthe
thesurge
surgevelocity
velocityand
andthe
thedrag
dragforce,
force,as
asshown
shownin inFigure
Figure7.7.

Figure 7.
Figure 7. Relationship of surge
Relationship of surge velocity
velocity and
and drag
drag force
force in
in surge
surge direction
direction for
for USV.
USV.

The circle
The circle test
test waswas conducted
conducted by by first
firstestablishing
establishing aa steadysteady state
state condition
condition andand then
then setting
setting port
port
and starboard
and starboardmotorsmotorstotodifferent
differentspeeds.
speeds.The TheUSV USVbegan
began tracking
tracking a circular
a circular pattern
pattern in the
in the absence
absence of
of wind
wind and current
and current for open forloop
open loop for
testing testing
a minimumfor a ofminimum
360 degrees.of 360 degrees.
However, However,
it was recommendedit was
recommended
to conduct the test to conduct the testtofor
for 720 degrees 720environmental
assess degrees to assess environmental
effects. This test waseffects.
conducted Thisattest was
several
conducted
motor speedsat several
for bothmotor port speeds for bothturns.
and starboard port and starboard
First, the portturns. First, the port
and starboard motor and starboard
commands
motorsetcommands
were to 100% for were10 sset
to to
keep100%theforUSV 10 stationary
s to keep the USV stationary
in maximum surgeinvelocity.
maximum Then,surge
both velocity.
motor
Then, bothwere
commands motor setcommands
to −100% and were set to
100% for−100% and 100% for
20 s, respectively. In20 s, test,
this respectively.
the USV was In this
abletest, the USV
to spin in a
was able
circle to spin
around in a circle
its center aroundwith
of gravity its center
minimal of gravity
surge andwithswayminimal surge which
velocities, and swaywere velocities, which
approximately
were Therefore,
zero. approximately zero. Therefore,
the kinetic model of USV the kinetic
in yaw model
direction of can
USVbeinsimplified
yaw direction
to Nrcan
r = be
(XP1simplified
− XP2 ) · dto
p.
The = ( Xrate
Nr ryaw P1 − X
(r)P2 )
was⋅ d p
recorded
. The yaw in Figure
rate (r) 8a.
was The drag
recorded moment
in Figure coefficient
8a. The from
drag yaw
moment rate N can
coefficient
r then be
from
estimated. The rotation radius is 0.77 m, which is about half of the length of the USV. Similarly, when
yaw rate N can then be estimated. The rotation radius is 0.77 m, which is about half of the length of
the USV wasr stationary in maximum surge velocity, the port and starboard motor commands were set
the0%
to USV.
andSimilarly,
100% for when the USV wasInstationary
20 s, respectively. this test, thein maximum
USV was surge able tovelocity, the port
steer around and starboard
a turning radius.
motor commands . were set to 0% and 100% for 20 s, respectively. In this test, the USV was able to
The acceleration v in sway direction, the surge velocity u, the sway velocity v and the yaw rate r were
steer around
recorded. The ayawturning radius.
rate (r) The acceleration
was recorded in Figure 8a.inAccording
sway direction, the surge
to the kinetic velocity
model in sway, direction
the sway
velocity . and the yaw rate were recorded. The yaw rate (r) was recorded in Figure 8a. According
(m − Yv. )v + (m − Xu. )ur + Yv v = 0, the drag coefficient in the sway direction from sway velocity Yv
to the
was kinetic
then model in
estimated. sway
The direction
rotation radius (mis−Yabout
v )v + (m
 2.56 − Xm.
u )ur +Yvcircle
The v = 0 ,paths
the drag
for coefficient
estimatingin Nthe
r andsway
Yv
direction from sway velocity Yv was then estimated. The rotation radius is about 2.56 m. The circle
were shown in Figure 8b. Due to the environmental disturbances, such as wind waves and water
currents, there are deviations in the circle paths in the first and the second 360 degrees. As shown in
paths for estimating N and Yv were shown in Figure 8b. Due to the environmental disturbances,
Figure 8b, the maximumr deviations of the circle path for estimating Nr and Yv are 0.055 m and 0.071 m,
such as windThe
respectively. waves and water
deviations can becurrents,
ignoredthere are to
relative deviations
the rotation in the circle of
radiuses paths in the
the two firstpaths.
circle and the
second 360 degrees. As shown in Figure 8b, the maximum deviations of the circle path for estimating

Sensors 2019, 19, x; doi: www.mdpi.com/journal/sensors

Sensors 2019, 19, x 13 of 22

Nr and Yv are 0.055 m and 0.071 m, respectively. The deviations can be ignored relative to the
rotation
Sensorsradiuses of the two circle paths.
2019, 19, 2051 13 of 22

(a) (b)

Figure 8. The experimental results of the circle test: (a) the yaw rates during circle tests; (b) the circle
Figure 8. The experimental results of the circle test: (a) the yaw rates during circle tests; (b) the circle
paths for estimating Nr and Yv in circle tests.
paths for estimating Nr and Yv in circle tests.
Combining the identification result of the propeller thrust model, the propeller thrust XP1 and
XP2 in the three-DOF
Combining model canresult
the identification be expressed as:
of the propeller thrust model, the propeller thrust X and P1

XP2 in the three-DOF model can be expressed

p as:
Xpi = −1.60 × 10−4 u2 + v2 ni + 5.04 × 10−3 |ni |ni , i = 1, 2 (34)
X = − 1.60 × 10 − 4 u 2 + v 2 n + 5.04 × 10 − 3 n n , i = 1, 2 (34)
Then the three-DOFpimodel identification is iachieved, given
i i
by:
Then the three-DOF model identification . is achieved, given by:
 x = u cos ψ − v sin ψ
.

 sinψψ−+v sin ψψ

yx == uucos v cos



 .


ψ=r


 y = u−4
sin√ψ +2 v cos2ψ

(35)

.
 u + v (n1 + n2 ) + 5.04 × 10−3 (|n1 |n1 + |n2 |n2 )

 50.05 u − 84.36vr + 151.57u = −1.60 × 10
ψ = r + 132.5v = 0


.


84.36v + 50.05ur


 17.21r. − (50.05 − 84.36)uv + 34.56r = (−1.60 × 10−−4 √


 (35)
50.05u − 84.36vr +151.57u = −1.60 ×10 4 u2u+2 v+2 (vn21(+n1n2−) +
n25.04 10−3×( n10
) + ×5.04 1 n1 + n12 n12 )
−3 (|n |n − |n |n )) · 0.26

 2 2
 84.36v + 50.05ur + 132.5v = 0
A zigzag test was conducted to evaluate the model by comparing experimental data with
17.21r −For the− 84.36) + 34.56r = (−1.60 ×10 −4
u2 +thrusters,
v2 (n1 − n2 ) +the ×10−3 ( n1 test
n1 − ncannot

simulations. (50.05 USVuv with rudderless double 5.04 zigzag 2 n2 )) ⋅ 0.26
be carried out
through setting up the rudder angle, like 20◦ /−20◦ , 10◦ /−10◦ . The test should be conducted by varying
A zigzag test was conducted to evaluate the model by comparing experimental data with
propeller speeds [30]. Moreover, this is similar to a 20◦ /−20◦ zigzag test, except that the angles may be
simulations. For the USV with rudderless double thrusters, the zigzag test cannot be carried out
different. The test adopted five combinations of motor commands, as shown in Table 3.
through setting up the rudder angle, like 20°/−20°, 10°/−10°. The test should be conducted by varying
propeller speeds [30]. Moreover, thiscombinations
Table 3. Five is similar toofa motor
20°/−20° zigzag test,
commands in theexcept
zigzagthat
test.the angles may be
different. The test adopted five combinations of motor commands, as shown in Table 3.
Test High Value (%) Low Value (%)
Table 3. Five combinations
1 of
100motor commands in 80
the zigzag test.
2 90 75
Test3 High 85Value (%) Low Value
65 (%)
4 75 57
1 5 100
65 80
50

2 90 75
Each alternating differential thrust motor command combination was executed with a period of
8 s in a series of 32 s runs. An3 example of the port
85 and starboard motor
65 commands is shown in Figure 9.
Figure 10 is the experimental 4 results of the zigzag
75 test, while Figure
57 10a is the comparison result
between the simulation path and the actual path. It can be seen from Figure 10a that the maximum
distance
Sensors 2019, 19,error
x; doi:between the actual path and simulation path is about 0.4 m, which may be due to
www.mdpi.com/journal/sensors
environmental disturbances. The yaw rate is an important parameter when conducting the zigzag test
 Each alternating differential thrust motor command combination was executed with a period of
8 s in Each
a series of 32 s runs.
alternating An example
differential of the
thrust portcommand
motor and starboard motor commands
combination is shown
was executed within Figure 9.
a period of
Figure
8 s in a10 is the
series of 32experimental results of the
s runs. An example the port
zigzag
andtest, whilemotor
starboard Figurecommands
10a is theiscomparison result9.
shown in Figure
between
Figure 10 theis simulation path and
the experimental the actual
results of thepath. It can
zigzag test,bewhile
seen Figure
from Figure
10a is10a
thethat the maximum
comparison result
distance error between the actual path and simulation path is about 0.4 m,
between the simulation path and the actual path. It can be seen from Figure 10a that the maximum which may be due to
Sensors 2019, 19, 2051 14 of 22
environmental
distance error disturbances.
between the actualThe yaw rate
path andis simulation
an important parameter
path is about when
0.4 m,conducting
which maythe be zigzag
due to
test because it isdisturbances.
environmental later used inThe model
yaw development, specifically
rate is an important in determining
parameter the maneuvering
when conducting the zigzag
coefficients
test because
because pertaining
it is laterit is intomodel
later
used how in
used fast the USV
model changes
development,
development, heading.
specifically Figurein
specifically
in determining 10b shows
determining
the the result
maneuvering for both the
the maneuvering
coefficients
simulation
coefficients
pertaining toand
how actual
pertaining yaw
fast the rates,
to USV
how fastwhere
changes they
the USV match
changes
heading. each other
heading.
Figure 10b well,
Figure
shows theverifying
10b forthe
shows
result the correctness
bothresult of the
for both
the simulation the
model and
simulation the
and identification
actual yaw results.
rates, where they match each other well, verifying
and actual yaw rates, where they match each other well, verifying the correctness of the model and the the correctness of the
model and the
identification identification results.
results.

(b)
(a)
(b)
(a)
Figure 9. An example of port and starboard motor commands (85%–65%): (a) port motor command
Figure9.9.An
Figure
(85%–65%); An example
(b) ofofport
starboard
example portand
motor starboard
command
and motor
motorcommands
(65%–85%).
starboard commands(85%–65%):
(85%–65%):(a)
(a)port
portmotor
motorcommand
command
(85%–65%); (b) starboard motor command (65%–85%).
(85%–65%); (b) starboard motor command (65%–85%).

(a) (b)
Figure10.
Figure 10.The (a)
Theexperimental
experimental resultsof
results ofthe
thezigzag
zigzagtest:
test:(a)
(a)the
the zigzag (b) path
zigzagsimulation
simulation pathand
andactual
actualpath;
path;
(b)
(b) theresult
the
Figure resultofofsimulation
10. The simulationand
and
experimental actualof
actual
results yaw
yaw rateininzigzag
therate
zigzag zigzag
test: test.
test.
(a) the zigzag simulation path and actual path;
(b) the result of simulation and actual yaw rate in zigzag test.
Based
Basedon onzigzag
zigzagtest, test,thetheparameters
parametersin inthethethree-DOF
three-DOFmodel model werewereidentified
identifiedas asananUSVUSVwithwith
vector propulsion
vectorBased
propulsion
on zigzag[25]. ItItmainly
[25].test, mainly used
usedthe
the parameters
. .
thedatadata
in.
u,uthree-DOF
the r
v,, r,v n, and
, nδmodel
in the
and δ zigzag
in the
were test. δ was
zigzag
identified test.
as anδ
theUSVrudder
was the
with
vector propulsion [25]. It mainly used the data u , v , r, n and δ in the zigzag test. δ was the
angle.
rudder Based
angle. on the
Based sampling
on the time,
sampling u, v and
time, u
r , v
were and 
r
calculated.
were The recursive
calculated. The least squares
recursive least method
squares
was usedangle.
method
rudder to identify
was used
Based tothe parameters
identify
on of time,
the model.
the parameters
the sampling However,
u ,ofv theand rwerein
model. the system
However,
calculated. identification
inThetherecursive method
system identification
least in
squares
the paper,inwas
method
method the
theacceleration
paper,tothe
used test and
thecircle
acceleration
identify testtest
andwere
parameters ofconducted
circle the were to
testmodel. obtain data
conducted
However, for
theidentifying
toinobtain the model
data foridentification
system identifying
parameters.
the model
method The established
inparameters.
the paper, the The model was
established
acceleration simplified
model
test and circle to one
was simplified DOF in each
to one DOF
test were conducted test, which made
in eachdata
to obtain the
test,for
which amount
made
identifying
of data
the small
theamount and the calculation
of data smallThe
model parameters. simple.
andestablished
the calculation A
model circle
simple. test was conducted
A circle test
was simplified towas to compare
one conducted
DOF in each the
to comparetwo
test, which models
themade
two
identified
models by the identification method in the paper and the method
the amount of data small and the calculation simple. A circle test was conducted to compare theand
identified by the identification method in the paper and thein [25].
method The inport
[25].and
The starboard
port two
motor
models commands
starboard identified were
motor commands
by theset to 85% and
were 55%
set to
identification 85% forand
method 20 s,inrespectively.
55% theforpaper Thethe
simulation
20 s, respectively.
and method The results[25].of
simulation
in Thea round
results
port andof
of
the two identification
starboard motor commands methodwere are shown in Figure
set to 85% and 55% 11. for
It can
20 s,berespectively.
seen from Figure 11 that the
The simulation resultof
results
Sensors
of 2019, 19,
the circle x; doi:
test based on the identification method in the paper is similar to www.mdpi.com/journal/sensors
the result based on the
method in [25].
Sensors 2019, It demonstrates that the accuracy of the identification method
19, x; doi: in the paper is the
www.mdpi.com/journal/sensors
same as the method in [25]. Therefore, the identification method in the paper can not only ensure the
accuracy, but also has lower computational load computation complexity than the method in [25].
a round of the two identification method are shown in Figure 11. It can be seen from Figure 11 that
the result of the circle test based on the identification method in the paper is similar to the result
based on the method in [25]. It demonstrates that the accuracy of the identification method in the
paper is the same as the method in [25]. Therefore, the identification method in the paper can not
only
Sensorsensure
2019, 19,the
2051accuracy, but also has lower computational load computation complexity than the
15 of 22
method in [25].

Figure 11.
Figure The simulation
11. The simulation results
results of
of aa circle
circle test
test of
of the
the two
two identification
identification methods.
methods.

5. PD+LOS Control Algorithm

5. PD+LOS Control Algorithm
For the USV motion control system, the design of the path-following controller is an important
For the USV motion control system, the design of the path-following controller is an important
part. Due to the simple structure and functional stability of PD control, it is currently used more in
part. Due to the simple structure and functional stability of PD control, it is currently used more in
autonomous navigation of the ship. Based on the established USV model, the USV motion control
autonomous navigation of the ship. Based on the established USV model, the USV motion control
system through PD+LOS guidance is designed [33–35].
system through PD+LOS guidance is designed [33–35].
The advantages of the LOS algorithm lie in that it is light in computation and easy for
The advantages of the LOS algorithm lie in that it is light in computation and easy for
implementation. The main principle of the LOS guidance algorithm is to imitate the behavior of
implementation. The main principle of the LOS guidance algorithm is to imitate the behavior of a
a helmsman, which steers the vehicle towards a lookahead distance ahead of the projection point of the
helmsman, which steers the vehicle towards a lookahead distance ahead of the projection point of
vehicle along the path, as shown in Figure 12. The USV assumes a tracking target on the tracking path
the vehicle along the path, as shown in Figure 12. The USV assumes a tracking target on the tracking
and then sails along the line of the USV and the hypothetical tracking target. As the USV approaches the
path and then sails along the line of the USV and the hypothetical tracking target. As the USV
desired path, the heading deviation is slowly reduced, and the desired path can be accurately tracked.
approaches the desired path, the heading deviation is slowly reduced, and the desired path can be
In Figure 12, (xk , yk ) and (xk+1 , yk+1 ) are two adjacent desired waypoints, and the line connecting them
accurately tracked.
is the desired path. In (x, Figure
y) is the12,
( xk , yposition
actual k ) and ( x +1, yk +measured
ofkUSV 1 ) are twoby
adjacent
GPS. ψ(desired
ti ) is thewaypoints, and
actual heading
the line connecting
measured by inertialthem is the desired
measurement units path. ) is
( x , yor
(IMU) the actual
compass. position
(xLOS of USV
(ti ), yLOS measured
(ti )) is by GPS.
the hypothetical
ψ (ti ) is target.
tracking The line
the actual connecting
heading (xLOS (tiby
measured ), yLOS (ti )) and
inertial the USV is the
measurement “line(IMU)
units of sight”.
or The angle
compass.
between the “line of sight” and the north direction is the desired heading of the USV, ψd (ti ), given by
( xLOS (ti ), yLOS (ti )) is the hypothetical tracking target. The line connecting ( xLOS (ti ), yLOS (ti )) and the
USV is the “line of sight”. The angle between theψ“line
ψd ( t i ) = ofψsight”
p ( ti ) + r (ti ) and the north direction is the desired
(36)
heading of the USV, ψ d (ti ) , given by
where ψp (ti ) is the path-tangential angle. ψr (ti ) is the velocity-path relative angle, ψr (ti ) =
arctan(−d(ti )/∆(ti )) It ensures that the velocity ψ d (ti ) =ψispdirected
(ti ) +ψ r (ttoward
i) (36)
a point on the path that is located
a lookahead distance ∆(ti ) > 0 ahead of the direct projection of (xk , yk ) on to the path [36–38].
where
The lookahead ψ p (ti ) distance
is the ∆(path-tangential
ti ) is dependent on angle. ψr (ti ) is error.
the cross-track the velocity-path relative
This results in lower angle,
values for
ψ = arctan(
∆(r t(it)i )(and thus−ad(more
ti ) / Δagile
(ti )) . and aggressive
It ensures response)
that the velocity when the USV
is directed is far from
toward theon
a point desired path,
the path and
that is
greater ∆(ti ) values when the USV is closer to the path and less abrupt behavior is needed so as to
located a lookahead distance Δ(ti ) > 0 ahead of the direct projection of ( xk , yk ) on to the path [36–
avoid oscillating around the path. The selection of ∆(ti ) refers to [36].
38]. The lookahead distance Δ(ti ) is dependent on the cross-track error. This results in lower values

Sensors 2019, 19, x; doi: www.mdpi.com/journal/sensors

Sensors 2019, 19, x 16 of 22

for Δ(ti ) (and thus a more agile and aggressive response) when the USV is far from the desired path,
and greater Δ(ti ) values when the USV is closer to the path and less abrupt behavior is needed so as
Sensors 2019, 19, 2051 16 of 22
to avoid oscillating around the path. The selection of Δ(ti ) refers to [36].

Figure 12. The

Figure 12. The principle
principle of
of the
the LOS
LOS guidance
guidance algorithm.
algorithm.

According to
According tothe
theEquation
Equation(17),
(17),the
thecontrol
control law
law forfor
thethe
USVUSV with
with rudderless
rudderless double
double thrusters
thrusters can
can be expressed
be expressed as as
. . .. .. .
J(η)−1 η+J
M(M(J(η) -1
 (η)η)+C
η+J(η)η) (υ)(J(η)−1
+C(υ)(J(η)-1 η)+D(υ)(J(η-1)−1 η)=τ
 +D(υ)(J(η) η)
η)  =τ (37)
(37)
T
where η = [ x y ψ T ] is related to the position (x, y) and yaw (ψ) of the USV. The input of the
where η = [ x y ψ ] is related to the position ( x , y ) and yaw ( ψ ) of the USV. The input of the
controller is the heading error between the yaw (ψ) and desired heading ψd , which is influenced by
controller
η. is the heading
The controller adjustserror between
the actual the yaw
heading ψ and(ψ )theandposition
desired(heading
x, y) of the ψd ,USV
which byiscontrolling
influencedthe by
η . Theerror.
heading controller adjustsand
The Coriolis thecentripetal
actual heading
matrix ψ Cand
(υ), the drag
position
matrix( x , yD) (of
υ) theandUSV by controlling
the vector of forces
and moments generated by the propulsion system τ will be impacted further,
the heading error. The Coriolis and centripetal matrix C (υ ) , the drag matrix D (υ ) and the vector which can be seen fromof
forces and moments generated by the propulsion system τ will be impacted further, which can be
Equations (15)–(18). It also realizes the control of the port and starboard propeller thrust through the
control
seen from of the velocity(15)–(18).
Equations of the USV and realizes
It also the speedsthe of two thrusters
control of the port according
and starboard to Equation
propeller(22) thrust
in the
established
through thepropeller
control ofthrust model.ofThrough
the velocity the USV changing
and the speedsthe force andthrusters
of two momentaccording of the USV, to the controller
Equation (22)
can achieve the motion control of the USV.
in the established propeller thrust model. Through changing the force and moment of the USV, the
For adopting
controller can achieve the PD thecontroller, on theofone
motion control thehand,
USV. at first, a proportional integral derivative (PID)
h iT
controller was carried
For adopting the PD outcontroller,
in the experimental testing.
on the one hand, However,
at first, because of
a proportional τ = derivative
integral τu 0 τr (PID) in
τ = [τdirection.
0 τ r ] Itinwas
T
the USV with
controller wasrudderless
carried outdouble in thethrusters,
experimental the USV wasHowever,
testing. not stressed in theofsway
because u
the
found that the lateral positioning errors were relatively large due to the
USV with rudderless double thrusters, the USV was not stressed in the sway direction. It was found less control effect in the sway
direction. Underpositioning
that the lateral the influenceerrors
of integral
were effect, the lateral
relatively large positioning
due to the errors
less control were accumulated,
effect in thewhich sway
resulted in that the integral effect was strong and the desired
direction. Under the influence of integral effect, the lateral positioning errors were control forces and moments skewed
accumulated,
towards regulating
which resulted the lateral
in that errors.effect
the integral This made the actuator
was strong and theconfigurations
desired control difficult
forcestoand achieve,
momentsand
the path-following
skewed controller the
towards regulating was lateral
unableerrors.
to achieveThisa made
steadythe state condition.
actuator The implementation
configurations difficult ofto
anti-saturation techniques was not sufficient to mitigate these problems.
achieve, and the path-following controller was unable to achieve a steady state condition. On the other hand, because
The
the GPS positioning
implementation has errors itself,
of anti-saturation the controller
techniques was not is set to starttotracking
sufficient mitigatethe nextproblems.
these waypointOn whenthe
the USV
other hand, arrives
because within the specified
the GPS positioning range of theitself,
has errors tracking waypoint.is set
the controller Therefore, when the
to start tracking theerror
next
between
waypointthe whenactual
theheading
USV arrivesand the desired
within heading range
the specified is within a specific
of the tracking small range, Therefore,
waypoint. the requirement when
of
the error between the actual heading and the desired heading is within a specific small range,state
path-following is considered to be achieved. This makes the compensation for the steady the
error unimportant
requirement in the path-following
of path-following controltoofbethe
is considered USV. Hence,
achieved. This amakes
PD controller, which is found
the compensation to
for the
perform adequately in simulation and experiment, is used instead.
steady state error unimportant in the path-following control of the USV. Hence, a PD controller,
whichThe is USV
foundmotion
to performcontrol system based
adequately on PD+LOS
in simulation andguidance
experiment,is shown
is used in Figure
instead. 13. According to
the ψThe
d ( t ) obtained by the LOS algorithm, the heading error is defined
USV motion control system based on PD+LOS guidance is shown in Figure 13. According as:

to the ψd (t) obtained by the LOS algorithm,

e(t) the
= ψheading
(t) − ψ (error
t) is defined as: (38)
d

e(t) =ψ (t) −ψ (t) (37)

The output of the PD controller is the differentiald speed of the port and starboard thrusters,
given by:
∆n = KP e(t) + KD (e(t) − e(t − 1)) (39)
Sensors 2019, 19, x; doi: www.mdpi.com/journal/sensors
 The output of the PD controller is the differential speed of the port and starboard thrusters,
given by:

Δn = KPe(t) + KD (e(t) − e(t −1)) (38)

Sensors 2019, 19, 2051 17 of 22
In the PD controller, the import of the derivative part improves the dynamic characteristics of
the system, but is particularly sensitive to noise. Therefore, the first order low-pass filter is added to
1
In the PD controller,
the derivative the import
part. The transfer of the derivative
function part improves
of the derivative thebe
filter can dynamic as h ( s ) = of the,
characteristics
expressed
1 +
system, but is particularly sensitive to noise. Therefore, the first order low-pass filter is added toTthe fs

Tf is the filter
derivative part.time
The constant.
transfer function of frequency
Since the the derivative filter
of the cansignal
input be expressed
in the PD (s) = 1+1Tiss ,10
as hcontroller
f
T f Hz,
is the
to
filter time constant. Since the frequency of the input signal in the PD controller is 10 Hz, to reduce
reduce noise effects, Tf is set to 0.1. The output of the PD controller after adding the derivative filter noise
effects, T f is set to 0.1. The output of the PD controller after adding the derivative filter is given by
is given by
T fT f KKDD
∆n =Δ nKP=eK (t ) +
(tP)e+ uDu D(t( t−−11)) +
+ ( e((et () t−) e−( te−(t1))
− 1)) (40)
(39)
T f T+f +TT TTf f++
TT

where uuDD((tt −
where −1)1)isisthe
theoutput
outputofof the
the derivative
derivative part
part at
at the
the last moment. TT is the period of the PD
last moment. PD
controller, which equals to 0.1
controller, which equals to 0.1 s. s.

Figure 13. The

Figure 13. The USV
USV motion
motion control
control system.
system.

Based on
Based on the
the established
established USV USV model
model and
and the
the PD+LOS
PD+LOS control algorithm, we adjusted the PD
parameters
parameters in simulation for
in simulation for aapreliminary
preliminaryselection
selectionofofPDPDparameters.
parameters.The The gainsKKP Pand
gains and KD of the
D of the
PD controller were manually adjusted. The detailed steps were as follows: (1) we first set KD = 0,
PD controller were manually adjusted. The detailed steps were as follows: (1) we first set KD = 0 , and
and adjusted KP gradually from small to large until the closed-loop response curve appeared equal
adjusted Koscillation;
amplitude P gradually(2) from
we small to large
recorded until the
the current closed-loop
value KP and the response curveperiod
oscillation appeared equal
To of the
amplitude oscillation; (2) we recorded the current value K P and the oscillation period PTo of the
closed-loop response curve. Based on the Ziegler-Nichols rules [39], we calculated the gains K and K D;
(3) we then set PD parameters as the values calculated in step (2). Based on the system response, we
closed-loop response curve. Based on the Ziegler-Nichols rules [39], we calculated the gains K P and
optimized the PD parameters further. The initial selection of PD parameters in the simulation provided
KD ; (3) we
guidance forthen set PD parameters
the setting of PD parametersas the in
values calculated
experiments, in reduced
and step (2). Based on the system
the corresponding response,
adjustment
we optimized
workload. Thethe PD parameters
finalized PD parametersfurther.were
The Kinitial selection
P = 5.3, KD = of 5.6.PDWeparameters
conductedinthe thesimulations
simulation
provided
and guidance
experiments for the
about setting of PDofparameters
path-following in experiments,
the USV, including andcircle
line path, reducedpaththe
andcorresponding
zigzag path.
Different
adjustment from the line, circle
workload. and zigzag
The finalized PDtests in systemwere
parameters KP = 5.3, Kthe
identification, D = 5.6 . We conducted the
line path, circle path and
zigzag path were closed loop maneuvers which were three tracking
simulations and experiments about path-following of the USV, including line paths as examples generated
path, circle path andby
setting the desired waypoints. There were other paths such as rectangular path, eight-shaped
zigzag path. Different from the line, circle and zigzag tests in system identification, the line path, path, etc.
The three
circle pathpaths were served
and zigzag path wereas examples to show
closed loop the performance
maneuvers which were ofthree
the path-following.
tracking paths as Inexamples
addition,
complicated time-dependent disturbances τ were considered as [33,40]:
generated by setting the desired waypoints.EThere were other paths such as rectangular path, eight-
shaped path, etc. The three paths were served as examples to show the performance of the path-
τuE = 1 + 0.1 sin(0.1t) + 0.3 cos(0.3t)

τ

following. In addition, complicated τvE time-dependent
= 1 + 0.2 sin(0.5tdisturbances
) + 0.1 cos(0.2tE) were considered as [33,40]: (41)





 τ = 1 + 0.3 sin(0.2t) + 0.1 cos(0.4t)

τ uE = 1 + 0.1sin(0.1t ) + 0.3cos(0.3t )
rE

The time-dependent disturbances τ vE = 1 +τ0.2sin(0.5t ) + 0.1cos(0.2t )
E were added to the simulations of path-following.
(40)
τ = 1 + 0.3sin(0.2t ) + 0.1cos(0.4t )
The conditions of simulations and  rE
experiments were set in the same. In the line path-following,
the direction and the distance of the desired line path were identical. The starting direction and the
radius of the desired circle path were the same in the circle path-following. Inwww.mdpi.com/journal/sensors
Sensors 2019, 19, x; doi:
the zigzag path-following,
the angle of each rotation and the duration of the desired zigzag path were identical. Besides, the initial
headings of the USV in the three path-following were the same. Figure 14 gives simulation and
experimental results for path following. The actual experimental environment of the lake is shown in
 The time-dependent disturbances E were added to the simulations of path-following. The
conditions of simulations and experiments were set in the same. In the line path-following, the
direction and the distance of the desired line path were identical. The starting direction and the radius
of the desired circle path were the same in the circle path-following. In the zigzag path-following, the
angle of each rotation and the duration of the desired zigzag path were identical. Besides, the initial
Sensors 2019, 19, 2051 18 of 22
headings of the USV in the three path-following were the same. Figure 14 gives simulation and
experimental results for path following. The actual experimental environment of the lake is shown
in Figure 15. TheThe root mean square error (RMSE)
(RMSE) of of the
the distances
distances between
betweenthe
thedesired
desiredwaypoints
waypointsandandthe
theactual
actualpositions
positionsofofthe
theUSV
USVisisadopted
adoptedtotoevaluate
evaluatethetheperformance
performanceofofthe
thecontroller,
controller,given
givenbyby
n


v
n
u
t (( xi((−xxdi−) 2x+ ()2yi+ 2
− (yydi )− ) y )2 )
u P
i di i di (41)
d RMSE = i =1
i=1
dRMSE = n (42)
n
(x , y ) is the
where di(x di, y is) the
where position of the
position i desired
of the waypoint.
i desired n isnthe
waypoint. number
is the of the
number desired
of the points.
desired
(x , y )
points.i (xi , y )
di di i i
is the i actual
is the i actualposition
positionof of
thetheUSV.
USV.
AsAsshown in Figure 14a,
shown in Figure 14a, the the distance
distanceerror is large
error when
is large whenstarting to track
starting thethe
to track desired waypoint.
desired waypoint.
However, the deviation is corrected by the adjustment of the controller,
However, the deviation is corrected by the adjustment of the controller, and the path-following and the path-following
performance
performance is ishigh
highininthe
the latter half of
latter half ofthe
theline
linepath.
path.InIn
thethe circle
circle andand zigzag
zigzag path-following,
path-following, the
the RMSEs
RMSEs of the distances between the desired waypoints and the actual positions
of the distances between the desired waypoints and the actual positions are larger than the RMSEs are larger than the
RMSEs in the line path-following, both in simulations and experiments.
in the line path-following, both in simulations and experiments. It indicates that the performance It indicates that the of
performance
the controller of the
in a controller
straight line in path-following
a straight line is path-following
better than that is in
better
curvethan that in curveHowever,
path-following. path-
following. However, because the RMSEs in the three path-following are all small,
because the RMSEs in the three path-following are all small, the controller still maintains excellent the controller still
maintains
performance excellent
in both performance
line and curve in path-following.
both line and curve Due topath-following.
the existence of Due to the existence
disturbances of
in simulations
disturbances in simulations and experiments in Figure 14, there are distance
and experiments in Figure 14, there are distance deviations between the simulation path, actual path deviations between the
simulation path, actual
and the desired path. path and the
Moreover, thedesired
RMSEs in path.
the Moreover,
experiments theof RMSEs in the
line, circle, andexperiments of line,
zigzag path-following
circle, and zigzag
are 0.33 m, 0.86path-following are 0.33 m, They
m, 0.54 m, respectively. 0.86 m, are0.54 m, respectively.
larger than those inThey are larger than
the simulations, thoseare
which
in 0.25
the simulations,
m in line path, 0.57 m in circle path, 0.42 m in zigzag path. Considering the accuracy ofpath.
which are 0.25 m in line path, 0.57 m in circle path, 0.42 m in zigzag 1 m of
Considering the accuracy
GPS positioning, of 1 m
the distance of GPS positioning,
deviations in simulations theand
distance deviations
experiments can be in accepted
simulations and
in practice.
experiments
Therefore, can be accepted
clearly satisfactory in practice. Therefore,
path following clearly
has been satisfactory
achieved by thepathUSV following has been
in both simulations
achieved by the
and experiments. USV in both simulations and experiments.

(a) (b)

Figure 14. Cont.

Sensors 2019, 19, x; doi: www.mdpi.com/journal/sensors

 Sensors 2019, 19, x 19 of 22
Sensors 2019, 19, x 19 of 22
Sensors 2019, 19, 2051 19 of 22

(c)
(c)
Figure 14.
Figure 14. The
The simulation
simulation and
and experimental
experimental results
results of the simulation
of the simulation path,
path, actual
actual path
path and
and desired
desired
Figure
path:14.
(a)The
linesimulation
path; (b) and path;
circle experimental
(c) results
zigzag path. of the simulation path, actual path and desired
path: (a) line path; (b) circle path; (c) zigzag path.
path: (a) line path; (b) circle path; (c) zigzag path.

Figure 15. The experimental environment.

Figure 15. The experimental environment.
Figure 15. The experimental environment.
6. Conclusions
6. Conclusion
6. Conclusion
The motion control problem for USVs has been studied with a focus on those equipped with
The motion
rudderless doublecontrol
thrustersproblem
due tofor USVs
their simplehasmechanical
been studied with a flexible
structure, focus onsteering
those equipped
capabilitywithand
The motion
rudderless control
double problem
thrusters due for
to USVs
their has
simple been studied
mechanical with a
structure, focus on
flexible those equipped
steering capabilitywithand
some other associated benefits. In particular, a propeller thrust model was derived first, with only two
rudderless
some other double thrusters
associated due to In
benefits. their simple mechanical
particular, a propeller structure,
thrust flexible
model wassteering
derived capability
first, withandonly
unknown parameters, which reduces the workload of system identification. Secondly, a three-DOF
some
two other associated
unknown benefits.
parameters, In particular,
which reduces a propeller
the workload thrust
of modelidentification.
system was derived Secondly,
first, with aonly three-
dynamic model was established, which combines the propeller thrust model to represent the thrust
twoDOFunknown
dynamic parameters, which reduceswhich
the workload of the
system identification. Secondly, a three-
in more detail. model
Moreover, was established,
the propulsion of thecombines
USV is provided propeller
by thethrust model
resultant to represent
force generatedthe by
DOF dynamic
thrust model was established, which combines the propeller thrust model to represent the
the portinand
more detail. thrusters
starboard Moreover, the propulsion
completely, and the ofrotational
the USV momentis provided by the
is related to resultant force
the differential
thrust in more
generated detail.
portMoreover, the thrusters
propulsion of the USV andistheprovided by the resultant force
thrust andby thethedistance andofstarboard
two thrusters. completely,
It provides guidance rotational
for moment
the modeling ofisgeneral
related to the
USVs
generated by
differential the port
thrust and
and starboard
the distance thrusters
of two completely,
thrusters. It and
providesthe rotational
guidance moment
for the is related
modeling ofto the
general
with rudderless double thrusters. Then, the model parameters were identified and verified through
differential thrust
USVs standard
with and the double
rudderless distancethrusters.
of two thrusters.
Then, the It provides
model guidance for
parameters the identified
modeling of general
some motion measurements, including the acceleration test,were
circle test andand verified
zigzag test.
USVs with
throughbasedrudderless
some on standard double thrusters.
motionmodel, Then,
measurements, the model
including parameters were identified and verified
Finally, the dynamic the PD+LOS controlthe acceleration
algorithm was test, circle test
employed and zigzag
to perform the
through some standard
test. Finally, based onmotion measurements,
the dynamic model, theincluding
PD+LOSthe acceleration
control algorithm test, circle
was test andtozigzag
employed perform
path-following control of the USV. The parameters in the PD controller were initially selected based on
test.
theFinally, based on the
path-following dynamic
control of themodel, the PD+LOS
USV. The parameters control
in thealgorithm
PD controllerwas employed to perform
were initially selected
simulations and then tuned during experimental tests, which reduced the corresponding adjustment
thebased
path-following control of the USV. The parameters in the PD controller
on simulations and then tuned during experimental tests, which reduced the corresponding were initially selected
workload in the experiments. The main goal of this work is to establish and identify a three-DOF
based on simulations
adjustment workload andinthen tuned duringThe
the experiments. experimental
main goaltests, of thiswhich
workreduced the corresponding
is to establish and identify a
dynamic model for the USV with rudderless double thrusters and validate the performance of the
adjustment workload in the experiments. The main goal of this
three-DOF dynamic model for the USV with rudderless double thrusters and validate work is to establish and identify athe
USV concerning the established model through simulations and experiments. Through a series of
three-DOF dynamic model for the USV with rudderless double
performance of the USV concerning the established model through simulations and experiments. thrusters and validate the
simulations and experiments, the RMSEs of distances between the desired waypoints and the actual
performance of the USV concerning the established model through
Through a series of simulations and experiments, the RMSEs of distances between the desired simulations and experiments.
Through a series of simulations and experiments, the RMSEs of distanceswww.mdpi.com/journal/sensors
Sensors 2019, 19, x; doi:
between the desired
Sensors 2019, 19, x; doi: www.mdpi.com/journal/sensors
Sensors 2019, 19, 2051 20 of 22

positions in path-following were small, which demonstrated that the motion control of the USV was
achieved successfully. Moreover, the controller could maintain good performance in both line and
curve path-following. It indicated the validity and practical value of the established models. It links
theory with practice, and the established model provides guidance for the experimental tests.

Author Contributions: C.L. and J.J. conceived and developed the idea and wrote this paper. These authors
contributed equally to this work; F.D., W.L. and X.W. designed and implemented experiments; L.B., Z.S. and G.Y.
provided some improved idea for this paper and revised the language.
Funding: This research was funded by the TianJin Natural Science Foundations of China under Grant No.
17JCQNJC01100, Open Project (6142218081811) of Key Laboratory of Underwater Information and Control, National
Natural Science Foundations of China under Grant No. 61501319, 51775377, 61505140, National key research and
development plan (2017YFF0204800), Young Elite Scientists Sponsorship Program By Cast of China under Grant
No. 2016QNRC001, Open Project (MOMST2015-7) of Key Laboratory of Micro Opto-electro Mechanical System
Technology, Tianjin University, Ministry of Education, Photoelectric Information and Instrument-Engineering
Research Center of Beijing Open Project No.GD2015007.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Specht, C.; Świtalski, E.; Specht, M. Application of an Autonomous/Unmanned Survey Vessel (ASV/USV) in
Bathymetric Measurements. Polish Marit. Res. 2017, 24, 36–44. [CrossRef]
2. Jorge, V.A.; Granada, R.; Maidana, R.G.; Jurak, D.A.; Heck, G.; Negreiros, A.P.; dos Santos, R.H.;
Gonçalves, L.M.G.; Amory, A.M. A survey on unmanned surface vehicles for disaster robotics:
Main challenges and directions. Sensors 2019, 19, 702. [CrossRef]
3. Yan, R.; Pang, S.; Sun, H.; Pang, Y. Development and missions of unmanned surface vehicle. J. Mar. Sci. Appl.
2010, 9, 451–457. [CrossRef]
4. Manley, J.E. Unmanned Surface Vehicles, 15 Years of Development. In Proceedings of the OCEANS 2008,
Quebec City, QC, Canada, 15–18 September 2008.
5. Geng, X.; Wang, Y.; Wang, P.; Zhang, B. Motion Plan of Maritime Autonomous Surface Ships by Dynamic
Programming for Collision Avoidance and Speed Optimization. Sensors 2019, 19, 434. [CrossRef]
6. Zhang, G.; Zhang, X. Concise robust adaptive path-following control of underactuated ships using DSC and
MLP. IEEE J. Ocean. Eng. 2014, 39, 685–694. [CrossRef]
7. Sarda, E.I.; Qu, H.; Bertaska, I.R.; von Ellenrieder, K.D. Station-keeping control of an unmanned surface
vehicle exposed to current and wind disturbances. Ocean Eng. 2016, 127, 305–324. [CrossRef]
8. Liu, Z.; Zhang, Y.; Yu, X.; Yuan, C. Unmanned surface vehicles: An overview of developments and challenges.
Annu. Rev. Control 2016, 41, 71–93. [CrossRef]
9. Jin, J.; Zhang, J.; Liu, D. Design and verification of heading and velocity coupled nonlinear controller for
unmanned surface vehicle. Sensors 2018, 18, 3427. [CrossRef]
10. Abkowitz, M.A. Lectures on Ship Hydrodynamics-Steering and Manoeuvrability; Hydro-Og Laboratorium:
Lyngby, Denmark, 1964.
11. Abkowitz, M.A. Measurement of Hydrodynamic Characteristics from Ship Maneuvering Trials by System
Identification. SNAME Trans. 1980, 88, 283–318.
12. Asmara, I.P.S.; Kobayashi, E.; Pitana, T. Simulation of Collision Avoidance in Lamong Bay Port by Considering
the PAW of Target Ship using MMG Model and AIS Data. Mar. Eng. Front. 2014, 2, 31–38.
13. Yasukawa, H.; Yoshimura, Y. Introduction of MMG standard method for ship maneuvering predictions.
J. Mar. Sci. Technol. 2015, 20, 37–52. [CrossRef]
14. Yang, L.; Tao, Y. CFD-Based calculation of Regular Transverse Wave Force in the Maneuvering Mathematical
Modeling Group (MMG) Model. Appl. Mech. Mater. 2013, 380, 1716–1720. [CrossRef]
15. Fossen, T.I.; Strand, J.P. Tutorial on nonlinear backstepping: Applications to ship control.
Model. Identificat. Control 1999, 20, 83–134. [CrossRef]
16. Fossen, T.I. Handbook of Marine Craft Hydrodynamics and Motion Control, 1st ed.; John Wiley & Sons Ltd.:
Chichester, West Sussex, UK, 2011; pp. 122–171.
17. Larrazabal, J.M.; Peñas, M.S. Intelligent rudder control of an unmanned surface vessel. Expert Syst. Appl.
2016, 55, 106–117. [CrossRef]
Sensors 2019, 19, 2051 21 of 22

18. Švec, P.; Thakur, A.; Raboin, E.; Shah, B.C.; Gupta, S.K. Target following with motion prediction for unmanned
surface vehicle operating in cluttered environments. Auton. Robot 2014, 383–405. [CrossRef]
19. Sharma, S.K.; Naeem, W.; Sutton, R. An Autopilot Based on a Local Control Network Design for an Unmanned
Surface Vehicle. J. Navig. 2012, 65, 281–301. [CrossRef]
20. Sutton, R.; Sharma, S.; Xao, T. Adaptive navigation systems for an unmanned surface vehicle. J. Mar.
Eng. Technol. 2014, 10, 3–20. [CrossRef]
21. Naeem, W.; Sutton, R.; Chudley, J. Modelling and control of an unmanned surface vehicle for environmental
monitoring. In Proceedings of the UKACC International Control Conference, Glasgow, Scotland,
30 August–1 September 2006.
22. Savvaris, A.; Niu, H.; Oh, H.; Tsourdos, A. Development of collision avoidance algorithms for the c-enduro
usv. In Proceedings of the 19th IFAC world congress (IFAC 2014), Cape Town, South Africa, 24–29 August
2014.
23. Murphy, R.R.; Steimle, E.; Hall, M.; Lindemuth, M.; Trejo, D.; Hurlebaus, S.; Medina-Cetina, Z.; Slocum, D.
Robot-Assisted Bridge Inspection. J. Intell. Robot. Syst. 2011, 64, 77–95. [CrossRef]
24. Sutresman, O.; Syam, R.; Asmal, S. Controlling Unmanned Surface Vehicle Rocket using GPS Tracking
Method. Int. J. Technol. 2017, 8, 709–718. [CrossRef]
25. Mu, D.; Wang, G.; Fan, Y.; Sun, X.; Qiu, B. Modeling and Identification for Vector Propulsion of an Unmanned
Surface Vehicle: Three Degrees of Freedom Model and Response Model. Sensors 2018, 18, 1889. [CrossRef]
26. Subcommittee, S.H. Nomenclature for Treating the Motion of a Submerged Body through a Fluid.
In Proceedings of the American Towing Tank Conference, New York, NY, USA, 11–14 September 1950.
27. Sonnenburg, C.R. Modeling, Identification, and Control of an Unmanned Surface Vehicle. Ph.D. Dissertation,
Virginia Polytechnic Institute and State University, Blacksburg, VA, USA, 2012.
28. Fossen, T.I. Nonlinear modelling and control of underwater vehicles. Ph.D. Dissertation, Dept. of Marine
Technology, Norwegian Institute of Technology, Trondheim, Norway, 1991.
29. Zheng, J.; Meng, F.; Li, Y. Design and Experimental Testing of a Free-Running Ship Motion Control Platform.
IEEE Access 2018, 6, 4690–4696. [CrossRef]
30. Marquardt, J.G.; Alvarez, J.; von Ellenrieder, K.D. Characterization and System Identification of an Unmanned
Amphibious Tracked Vehicle. IEEE J. Ocean. Eng. 2014, 39, 641–661. [CrossRef]
31. Muske, K.R.; Ashrafiuon, H.; Haas, G.; Mccloskey, R. Identification of a Control Oriented Nonlinear Dynamic
USV Model. In Proceedings of the American Control Conference, Seattle, WA, USA, 11–13 June 2008;
pp. 562–567.
32. Fossen, T.I. Guidance and Control of Ocean Vehicles, 1st ed.; Wiley: New York, NY, USA, 1994; pp. 32–42.
33. Mu, D.; Wang, G.; Fan, Y.; Sun, X.; Qiu, B. Adaptive LOS Path Following for a Podded Propulsion Unmanned
Surface Vehicle with Uncertainty of Model and Actuator Saturation. Appl. Sci. 2017, 7, 1232. [CrossRef]
34. Breivik, M.; Fossen, T.I. Path following of straight lines and circles for marine surface vessels. In Proceedings of
the IFAC Conference on Computer Applications in Marine Systems - CAMS 2004, Ancona, Italy, 7–9 July 2004.
35. Fossen, T.I.; Breivik, M.; Skjetne, R. Line-of-sight path following of underactuated marine craft. In Proceedings
of the 6th IFAC Conference on Manoeuvring and Control of Marine Craft (MCMC 2003), Girona, Spain,
17–19 September 2003.
36. Lekkas, A.M.; Fossen, T.I. A time-varying lookahead distance guidance law for path following.
In Proceedings of the 9th IFAC Conference on Manoeuvring and Control of Marine Craft, Arenzano, Italy,
19–21 September 2012.
37. Fossen, T.I.; Pettersen, K.Y.; Galeazzi, R. Line-of-sight path following for dubins paths with adaptive sideslip
compensation of drift forces. IEEE Trans. Control Syst. Technol. 2015, 23, 820–827. [CrossRef]
38. Lekkas, A.M.; Fossen, T.I. Integral LOS path following for curved paths based on a monotone cubic hermite
spline parametrization. IEEE Trans. Control Syst. Technol. 2014, 22, 2287–2301. [CrossRef]
Sensors 2019, 19, 2051 22 of 22

39. Åström, K.J.; Hägglund, T. Revisiting the Ziegler-Nichols step response method for PID control.
J. Process Control. 2004, 14, 635–650. [CrossRef]
40. Pan, C.Z.; Lai, X.Z.; Yang, S.X.; Wu, M. An efficient neural network approach to tracking control of an
autonomous surface vehicle with unknown dynamics. Expert Syst. Appl. 2013, 40, 1629–1635. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).