Ocean Engineering 111 (2016) 508523

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Precise trajectory control for an inspection class ROV

Serdar Soylu, Alison A. Proctor n, Ron P. Podhorodeski, Colin Bradley, Bradley J. Buckham
Department of Mechanical Engineering, University of Victoria, P.O. Box 1700, Victoria, BC, Canada V8P 5C2

art ic l e i nf o

abstract

Article history:
Received 21 January 2013
Accepted 26 August 2015
Available online 7 December 2015

This work addresses the design, implementation and testing of a new precision guidance and control
system for an inspection class remotely operated underwater vehicle (ROV). A new multi-input multioutput control law, composed of a model-based equivalent control signal and two adaptive signals, is
presented. In the controller, the rst adaptive signal is a PID signal with a novel adaptation law that
enhances the controller performance and allows efcient ne tuning of the controller. The second
adaptive signal continuously estimates the upper bound on the lumped uncertainty vector and acts as a
corrective term for the equivalent control law. A Lyapunov based guidance algorithm is implemented
that can tolerate signicant uncertainties in the system kinematics. The effectiveness of the system on an
ROV is demonstrated through eld trials in sheltered waters. For the experimental work, an extended
Kalman lter is used to for navigation, blending the on-board sensor measurements with a process
model to produce an estimate of the vehicle dynamics. The collective guidance and navigation system are
validated using high precision optical motion capture data. The system achieves decimetre-level precision, signicantly extending the capabilities of the ROV for tasks requiring high precision position and
velocity control.
& 2015 Elsevier Ltd. All rights reserved.

Keywords:
Control system
Remotely operated underwater vehicles
Adaptive
Sliding mode
Guidance
Falcon ROV

1. Introduction
Remotely Operated underwater Vehicles (ROVs) have widespread

need for automatic correction of erroneous motion caused by currents and tether disturbances.
Traditional linear controllers cannot provide high precision

uses in scientic and industrial investigations including bottom sur-

motion control capabilities, as their performance degrades sig-

veys, seaoor imaging, oceanographic and geological data acquisition

and robotic interactions with natural features or man-made infra-

nicantly in the face of highly non-linear ROV hydrodynamics and,

more importantly, tether disturbances. More advanced control

structure. In typical ROV operations, the motion of the ROV is con-

techniques are needed and several nonlinear control techniques

trolled by a human pilot stationed on a surface vessel. The human

have been applied in the ROV paradigm including: the H1

approach in Conte and Serrani (1998); adaptive control techniques
in Fossen and Sagatun (1991), Hsu et al. (2000), Antonelli et al.

pilot's commands are sent to the submerged ROV through a tether.

The human presence makes complex multi-objective underwater
missions possible: humans can react to sudden changes in a mission
plan caused by the unpredictable nature of the ocean environment.
However, the limitations in sensory feedback to the ROV pilot make
certain operations, such as high precision trajectory following,
impossible without some form of machine intelligence. This is
especially true for low-cost inspection ROVs where there is an acute

Abbreviations: DVL, Doppler velocity log; EKF, extended Kalman lter; IMU,
inertial measurement unit; NED, north, east, down reference frame; ROV, remotely
operated underwater vehicle; SBL, short baseline acoustic system; UUB, uniformly
ultimately bounded; wrt, with respect to
n

Corresponding author. Fax: 1 250 721 6035.

E-mail addresses: serdar@uvic.ca (S. Soylu), proctora@me.uvic.ca (A.A. Proctor),

podhoro@me.uvic.ca (R.P. Podhorodeski), cbr@me.uvic.ca (C. Bradley),

bbuckham@me.uvic.ca (B.J. Buckham).

(2003, 2001) and Zhao and Yuh (2005); sliding-mode control in

Yoerger and Slotine (1985), Yoerger et al. (1986), Olgac et al.
(1991), da Cunha et al. (1995), and Soylu et al. (2008); fuzzy logic
control in Ju et al. (2002); neural network control in van de Ven
et al. (2005) and Kim and Inman (2003); combined approaches
such as H1 and sliding-mode control in Soylu et al. (2009),
adaptive, fuzzy, and sliding-mode control in Trebi-Ollennu et al.
(1995) and sliding-mode control and neural network in Liu et al.
(2008); and other nonlinear Lyapunov-based systems such as in
Caccia and Veruggio (2000a) and Lau et al. (2003).
In the current work, a new controller is presented and implemented on an inspection-class ROV, for high precision trajectory
and way-point following. Among the aforementioned controllers,
the work of Soylu et al. (2008) was chosen as the basis of the
controller presented herein for its simplicity and effectiveness.

http://dx.doi.org/10.1016/j.oceaneng.2015.08.061
0029-8018/& 2015 Elsevier Ltd. All rights reserved.

S. Soylu et al. / Ocean Engineering 111 (2016) 508523

509

However, the Lyapunov stability of that controller depends on the

time rate of change of the difference between the known and the
unknown ROV dynamics, referred to as the lumped uncertainty

proposed guidance, navigation and control scheme in obtaining

high precision ROV motion.
The remainder of this article is organized as follows: Section 2

vector. This makes the evaluation of the Lyapunov stability of that

introduces the test facility that was built for this research and

controller inconclusive for the stop-and-go style manoeuvres

that are common in ROV operations. Since sudden changes in the
motion of an ROV is inevitable when a pilot's intent suddenly

provides the brief description of the experimental ROV platform.

changes, a method to make the controller Lyapunov stable in the

Section 4 describes the three layer adaptive control system being

presence of these dynamics is required.

In addition, existing adaptive controllers can be prone to a

proposed and Section 5 discusses the guidance and navigation

parameter drift problem that occurs when the persistency of

ROV and the experimental verication of the vehicles dynamic

model and navigation algorithm. Finally, Section 6 demonstrates

excitation condition is not satised. This problem occurs when the

reference input does not excite the internal system dynamics such
that the adaptation mechanism can stably estimate the adaptive
parameters; roughly speaking, this means that the reference input

Section 3 discusses the ROV kinematics and details the dynamic

model used in both the controller and navigation calculations.

algorithms, and outlines the hydrodynamic characterisation of the

the controller performance through a series of eld trials that

include waypoint and trajectory following tasks and Section 7
provides the authors' conclusions.

must be rich enough that the parameters are observable to the

adaptation mechanism. If the persistency of excitation condition is
not met then it can result in an unboundedness in the parameters
that are being adaptively estimated. These drifting parameters
can result in limit cycles or instability. This problem is particularly
challenging with ROVs because there are often long periods of
inactivity where the vehicle is simply waiting in a hover while the
operator performs a task.
To address these shortcomings, in the current work, the
adaptation from Soylu et al. (2008) is changed to estimate the
upper bound of the unknown dynamics in lieu of the actual

2. Experimental facilities
2.1. Underwater research facility
The experimental test facility is shown in the photographs of
Fig. 1 and is comprised of a modied boathouse in a commercial
marina. The boathouse, which sits on pontoons on top of the
water, has an opening to the test area that is 5.25 m long and 3 m

lumped uncertainty vector. This adjustment removes the dependence of the Lyapunov stability on the time rate of change of the
lumped uncertainty vector, which in turn, leads to a universally
stable controller, in the Lyapunov sense, even in the presence of
sudden vehicle motion. In addition, the parameter drift problem
was eliminated using a leakage term based on a variant of the
s-modication method (Ioannou and Kokotovic, 1983). Finally, the
adaptive PID layer with a novel adaptation mechanism was also
incorporated to further enhance the controller performance. This
PID layer is an intuitive tool to ne tune the controller behaviour
and eliminate any small residual errors.
To facilitate implementation, this new controller must be
complemented by a trajectory generator and by a sensor feedback
loop. In this work, the two-layered hierarchical architecture proposed in Aicardi et al. (1995) is employed. Firstly, the outer guidance loop algorithm receives target way-point locations and
generates motion directives based on current vehicle state, as
estimated by an Extended Kalman Filter (EKF). The motion
directives are subsequently fed to the inner control loop, where
the controller described in this work uses the motion directives
and the same EKF state feedback to generate thruster commands.
The navigation sensors used in this work are housed in an
auxiliary navigation enclosure, which is attached to the ROV. The
enclosure contains a Systron-Donner Inertial Measurement Unit
(IMU), an RDI Explorer Doppler Velocity Log (DVL), a Sparton
Compass, and a SouthStar Short Base Line (SBL) acoustic positioning system. The SouthStar SBL system exploits the umbilical
cable to trigger acoustic signals from the vehicles emitter. In this
conguration, the acoustic signal only needs to travel in one
direction minimizing errors and allowing faster pinging rates to be
achieved. In order to conclusively evaluate the performance of the
navigation system, a high precision optical motion capture system
is utilized. The optical system provides an absolute measurement
of the true vehicle state for comparison to the EKF estimation,
allowing for a direct evaluation of the performance rather than
inferring performance based on covariance matrices produced in
the EKF calculations. The controller performance was tested
through a series of eld trials based on the trajectory following
and way point tasks. The results indicate the effectiveness of the

510

Fig. 1. Wet-Test facility entrance (top) and exit (bottom).

S. Soylu et al. / Ocean Engineering 111 (2016) 508523

wide. The opening is used to launch and recover the ROV and for
making measurements; however, once underwater, the ROV can
venture outside of these boundaries. The water depth in the test

area ranged between 3.5 m and 5 m deep depending on the tide.

While operating in the boathouse, the ROV is subject to tides,
currents, wind, waves, water stratication, marine life, turbidity
and all of the other factors present in the ocean.
2.2. Positioning systems
The test facility is equipped with an underwater acoustic
positioning system as well as a camera based motion capture
system for system validation, illustrated in Fig. 2.
The acoustic positioning system is a SouthStar SBL system that
provides centimetre level subsea positioning at an update rate
between 1 and 2 Hz. The camera based motion capture system is a
VZ3000 Visualeyez vision-based position tracking system by
Phoenix Technologies Inc. The Visualeyez system can provide
millimetre-level precision at an update rate of approximately
80 Hz for objects above the surface of the water. The Visualeyez
system is used to generate truth data for position verication.
The vision system comes with two cameras and a set of optical
markers (LED light sources). Using a mast to keep the markers
above water, the cameras were able to track the motion of the ROV
while it performed manoeuvres up to 1 m below the surface of the
water.
The camera units each contain three high-speed high-resolution imaging sensors that can estimate the position of the optical

Fig. 3. Saab-SeaEye Falcon ROV system components. Image courtesy of Zand

(2009).

2.3. Experimental ROV platform

The ROV used in this experimental work is the inspection class,
Saab-Seaeye Falcon ROV shown in Fig. 3. The vehicle is rated to a
depth of 300 m and has a mass of 75 kg. The ROV has 5 electric
thrusters; the rst four thrusters control motion in the vehicle's
horizontal plane, and the fth one controls vertical motion in the

markers within their eld of view. In the experimental set-up, the

camera units are mounted high up on the boathouse walls such
that they capture a wide swath of the water surface. To estimate
the position and orientation of the vehicle while underwater,

water column. The ROV is tted with a 350 m bre-optic tether to

enable the system to support additional instrumentation.
The stock ROV has a depth sensor, a compass with pitch and

markers were afxed to a lightweight mast in such a way as to

applications described herein, an external instrument suite was

form a coordinate frame and attached to the ROV. In this conguration, the output of the camera system can be used to track the
position and orientation of the vehicle while it is manoeuvring
underwater.
The four SBL reference stations are located in the corners of the
boathouse at a depth of roughly 0.8 m, approximately the same
depth as the ROV operates with the Visualeyez mast attached. This

developed that contains the following sensors: a Systron-Donner

conguration provides the most accurate positioning inside the

boathouse in the plane containing the reference stations. The
positioning information from the SBL is available outside of the
boathouse as well, but the accuracy of the estimate decreases as
the vehicle moves farther away.

roll, and a rate gyro on the Z-axis. To support the advanced control

IMU, an Explorer DVL, a Sparton Compass, and a SouthStar SBL.

All of the software, including sensor drivers, and the navigation,
guidance and control algorithms, was written in LabVIEW and runs
on a National Instruments PXI-1032B real-time data acquisition and
control chassis that is connected to the ROV's surface control unit.

3. Modelling
3.1. ROV kinematics
Two reference frames are used to express the motion of the
ROV, an inertial reference frame that is coincident with the NorthEast-Down (NED) frame and a body-xed frame afxed to the
vehicle centre of mass as shown in Fig. 4. The ROV's state vector
with respect to (wrt) its body-xed frame is dened as
q = [u v w p q r ]T where the rst three elements are the
translational velocities and the second three elements are the
rotational rates of the vehicle about the x y and z axis, respectively.
The vehicle state wrt the inertial frame is given by
= [ X Y Z ]T . The spatial transformation matrix
between the inertial frame and ROV's body-xed frame is given by
J, which includes the angular velocity transformation matrix J1 and
the linear velocity transformation matrix J2. The transform J2 is a
rotation matrix obtained using the typical ZYX ( ) yaw,
pitch and roll Euler rotation sequence and J 1 is the associated
angular velocity transformation matrix.

Fig. 2. SBL and camera layout.

= Jq, J =

03 3 J2

J1 03 3

(1)

Thank you for evaluating AnyBizSoft PDF to Word.

You can only convert 3 pages with the trial version.

To get all the pages converted, you need to purchase the software from:

http://www.anypdftools.com/buy/buy-pdf-to-word.html