1 INTRODUCTION

1.1 Motivation

Oceans are resourceful. Ocean operations are vastly

increasing the last decades, and so is the interest in

unmanned surface vehicles (USV) and autonomous

ships. The ocean’s extreme weather and far distances

can result in high-risk-high-cost work conditions.

The world’s economy is mainly defined by three

areas: energy, transportation and communication

(Rifkin, 2012). Ocean industries push the boundaries

of these three areas to the limits. Unmanning of

maritime assets by excessive automation and remote

control could reduce or eliminate the risk imposed on

crew; however, infrastructure cost will increase. A

huge safety potential is accompanied with more

benefits; by removing crew from the assets, crew

related costs are, in theory, removed, costs such as

cooling, heating and ventilation. Accommodation

spaces are, in theory, no longer required, less power

consumption is projected and the chain of promises

goes on.

The drivers for unmanning maritime assets are

developing into motivations for building and

operating autonomous vessels. One example is, Yara

Birkeland, a 120 TEU open-top zero emissions

autonomous containership, planned launch is

expected before 2020, the ship is under construction

with Kongsberg technology. Another example is the

car ferry Falco that was built using Rolls-Royce

technology, launched late 2018.

As operators are moved from the far end of the

operation to shore control centers, their experiences

are changed, their feelings and senses while on duty

from one hand, their toolboxes and control authority

on the distant ship from the other hand. The current

remote-control technologies and their limitations are

subject to discussion. In this paper, the challenges of

Is Full-autonomy the Way to Go Towards Maximizing

the Ocean Potentials?

R. Zghyer, R. Ostnes & K.H. Halse

Norwegian University of Science and Technology, Trondheim, Norway

ABSTRACT: Growth prospects for ocean economy are promising because ocean industries are addressing

challenges such as food security, energy security and climate change. However, safety and efficiency are the

general challenges of ocean operations. Increased automation is believed to solve these problems. This paper

discusses the impact of automation on safety and efficiency. A literature review of ‘Human factors’ mainly from

the aviation and maritime industries is presented to untangle the human-machine relationship characteristics

when increased automation is introduced to operators. A literature review of Hydrodynamics, Guidance,

Navigation and Control (GNC) technologies is presented to introduce the state-of-art and associated

limitations. It is concluded that, if the industry’s drive is safety and efficiency, then full-autonomy is, at present,

not the way to go. Remote control, instead, could facilitate a feasible future, while focused research and

development are in need.

http://www.transnav.eu

the International Journal

on Marine Navigation

and Safety of Sea Transportation

Volume 13

Number 1

March 2019

DOI: 10.12716/1001.13.01.02

unmanning maritime assets and the transfer towards

full-autonomy will be discussed.

1.2 Introduction

This literature review is part of a PhD study with the

objective of “evaluation of technology using

simulators”. Main research area is safety and

efficiency of semi-autonomous vessels and the

research scope includes Hydrodynamics; simulation

and testing; and semi-autonomous maneuvering in

close proximity to structures. The simulator facilities

at NTNU include a variety of simulators used for

teaching and research. The use of simulators enables

operators-in-the-loop testing, connecting technology

to humans. The author is studying the man-machine

semi-autonomous maneuvering problem from both

the technology side and the human side. The

technology side is broken down to four scientific

fields: Hydrodynamics; Guidance; Navigation; and

Control. Those four fields reflect the state-of-art in ship

motion prediction and enhancement of automation

level in ship maneuvering. Whereas the field of

Human Factors (relevant to remote operators) is the

field representing the human side of the problem.

These five fields are reviewed briefly in this paper.

The terms may have multiple definitions, therefore, in

this review, the main fields are defined as follows:

 Hydrodynamics field in this review refers to

methods that describe the motions and responses

of a ship moving in water using maneuvering and

seakeeping theories such as unified models (Skejic

and Faltinsen, 2008).

 Guidance, navigation and control (GNC) is a well-

established technical term used in engineering and

control (cybernetics) fields in topics related to

traveling vehicles; cars, ships, or planes.

 Guidance module is the brain of the robotic

controller that is responsible for trajectory

planning, collision avoidance and conforming

to protocol (such as COLREG) (Fossen, 2011).

 Navigation module is responsible for

estimating own state, that is, identify own

position and motion information using sensors

and GNSS signals, as well as estimating external

situation, including environment perception

(wind, waves, water depth, etc.) and obstacle

state estimation, that is, identify obstacle

position and motion information (Farell, 2008).

 Control is the translation of guidance (desired

trajectory) into actuator instructions that result

in an actual trajectory as close as possible to the

desired one and provides stability to the vehicle

(Pérez, 2005).

 Human factors refers to reflections from human

operators as more automation is introduced to

their operations. Sections below include reviews of

each of the fields separately.

2 LITERATURE REVIEW

After the fields of interest were defined, 59 relevant

articles were reviewed in those fields of interest.

Challenges and conflicts are presented. The literature

is found in two ways: Education and search.

Education literature is based on relevant courses and

their relevant references. While Search literature is

based on digital databases search of the following

keywords: Hydrodynamics; seakeeping; maneuvering;

ship simulation; semi-autonomous vessels; unmanned

surface vehicles; guidance; navigation; control; and human

factors. Search results were filtered based on relevance

to the already defined subjects of interest.

2.1 Topic 1: Hydrodynamics

Dynamics is broken down by the studies of

kinematics and kinetics, the former deals with

geometrical aspects of motion and the latter deals

with forces causing the motion. This review is

concerned with ship dynamics, therefore this section

starts with the maneuvering and the seakeeping

theories as foundation for ship dynamics models. The

former is the study of ship moving in constant speed

in calm waters with the assumption that ship motion

is frequency independent, that is, no wave excitation

takes place. The latter is the study of ship motion at

zero or constant speed in waves using frequency

dependent hydrodynamic coefficients.

An overview of methods for describing

maneuvering and seakeeping are grouped into

experimental methods, unified methods, two-time

scale methods and direct calculations by

Computational Fluid Dynamics (CFD) tools

(Quadvlieg et al., 2014). The research is focused on

real-time simulations and on including dynamics-in-

the-loop for marine control systems, therefore the

interest lies in fast mathematical methods such as the

unified methods and the two-time scale methods.

CFD tools are high computationally demanding and

not suitable for real-time simulations. Both examples

presented below, the unified model and the two-time

scale method, are suitable for real-time simulations.

The unified model is a vectorial model that

describes both the maneuvering and seakeeping ship

motions and dates back to 1991 (Fossen, 1991) and is

considered by the international community as a

“standard model” for marine control systems design.

The “standard model” is an upgrade of an earlier

model (the “classical model”) that represents the ship

motion in a component form instead of vector form

and is mostly used in hydrodynamic modeling where

isolated effects are studied.

The 6 degrees-of-freedom (6-DOF) model is

represented as (Fossen, 2011):

  





wind wave

Mv C v v D v v g η

ττ τ

(1)

where





,,,,,

xyz





 and





,, , ,,

v uvwpqr

are vectors of position / Euler angles and velocities

respectively.

τ vectors are vectors of environment

and control forces and moments. The model matrices

 

, andMC v Dv are inertia, Coriolis and

damping matrices respectively. While



g η is a

vector representing gravitational and buoyancy forces

and

is a representation of ballast restoring forces

and moments. The model is formulated in the time

domain using the Cummins equation that considers the

impulse response function over the past history of the

excitation force, known as fluid memory effects

(Cummins, 1962).

The two-time scale method was proposed by Skejic

and Faltinsen in 2008. It is also a vectorial unified

model that describes both the maneuvering and

seakeeping ship motions. The time domain of the

simulation is divided into two time scales, a slowly

and a rapidly varying one associated with the

maneuvering and the seakeeping respectively. This

method estimates the mean second-order wave loads

(that result in lateral drift caused by incident waves

and wind) “as accurate as possible and at the same

time to be able to simulate real-time maneuvers with

acceptable CPU time.” (Skejic and Faltinsen, 2008, p.

374). The model is represented in a 4-DOF (surge,

sway, roll and yaw) form as follows:

(2)

The main advantage of the two-time scale model is

that it captures the second-order lateral drift

phenomenon. It has better performance in incident

waves, where the mean second-order wave loads

heavily influence the maneuvering behavior. As it

considers theories covering the whole range of

important wavelengths.

For both methods, the potential theory is the main

tool for calculating the hydrodynamic coefficients and

thus forces. This theory assumes water flow across the

rigid body as constant, irrotational, and

incompressible. Chapter 5 of Fossen (2011) covers

hydrodynamic concepts and numerical approaches.

The most common numerical approaches for

calculating the hydrodynamic coefficients are;

 Strip Theory; a 2-D theory that considers the flow

variation in the longitudinal-section is much

smaller than that of the cross-section plane of the

ship.

 Panel Methods; 3-D integration method that

divides the surface of the ship and the

surrounding water into discrete panels, assigned a

distribution of sources and sinks that fulfil the

Laplace equation.

A comparison of the unified model and the two-

time scale method is of interest for this research,

because the hydrodynamic differences affecting ship

control require further research (Liu et al., 2016).

Several examples of unified numerical models have

been developed in the last three decades and here is a

summary of the latest progress. The method proposed

by Skejic and Faltinsen in 2008 was verified and

validated for calm water. This method is further

developed in a study on ship-to-ship hydrodynamic

interaction effects between two ships going ahead in

regular waves, it highlights critical maneuvering

situations and it still requires experimental validation

(Skejic and Berg, 2010). In 2013, the two-time scale

model was applied to irregular seas and validated for

a container ship (Skejic and Faltinsen, 2013).

Hermundstad and Hoff (2009) implemented a time

domain unified model on submarines and compared

with experimental results. It was argued that the used

unified model did not describe the diving maneuvers

correctly because the depth dependency of the

coefficients was not incorporated. A practical method

for ship motion simulation using the two-time scale

method is presented by Yasukawa and Nakayama

(2009) that derives 6-DOF equations of motion for the

high frequency problem and 4-DOF equations of

motions for the low frequency problem. Wave

induced motions for turning maneuver are predicted

for a container ship of geometry S-175 and the

predictions resulted in rough agreement with free

model tests. Yasukawa, Amri Adnan and Nishi

(2010)

compared, numerical estimates of hydrodynamic

forces and wave-induced motions taking into account

lateral drift, with experiments showing that drift

effects are not negligible and that the method is able

to capture them. Seo and Kim (2011) extended the

WISH (computer program for nonlinear Wave

Induced load and Ship motion analysis) by coupling

the maneuvering and the seakeeping models, and

verified it by comparing with published experiment

data in calm weather and regular waves. The

simulations showed fair agreement of overall

tendency in maneuvering trajectories.

Beside lateral drift, the broaching phenomenon is

another challenge for hydrodynamic models, it

concerns loss of stability while sailing in following

seas where the kinetic energy of the ship along the

forward axis transfers to roll motion and leads to

strong heel, loss of heading, even capsize (Wu,

Spyrou and McCue, 2010). Generally, maneuvering in

waves is a challenge for both experimental and

numerical modelling. For simulating ship motion in

waves, forces and hydrodynamic coefficients need to

be calculated dependent on wave frequency, ship

heading and angle of attack angle between wave

direction and ship course (Kim et al., 2014).

2.2 Topic 2: Guidance

The guidance system receives information about the

world, both internal information concerning the ship

maneuvering and engine status, and external

concerning the surroundings, environmental loading

and nearby target ships and other objects and

translates this information into instructions to

controllers. Guidance is responsible for path planning,

including collision avoidance. Fossen (2011) defines

motion objectives categories. The guidance system

together with the control system should fulfill the

motion objectives according to one of the following

categories:

1 Setpoint regulation: heading angle is constant with

no consideration of time.

2 Path following: heading angle is variable,

following a path, no consideration of time.

3 Trajectory tracking: heading angle is variable,

following a trajectory in both space and time.

4 Maneuvering: considers the overall feasibility of the

path, often with more importance to space than

time. To incorporate COLREG, the guidance

system shall consider both space and time

parameters because velocities of maneuvers are

critical.

The guidance system tasks are grouped into two:

global and local path planning. The global path

planning approach is the deliberate part of the

guidance system. It is an optimized plan of the path

from starting point of the trip to the end point, it

includes known information about traffic, weather

forecast, ship properties, land/islands, shallow waters

and buoys. This is a multi-objective optimization

problem and usually done offline and requires large

computational requirements, in which, optimization

methods and heuristic search algorithms are the two

main methods. While local path (re)planning

approach is the reflexive part of the guidance system,

it takes charge of planning local deviations from the

global plan, in case the navigation system detected an

approaching object. A characteristic requirement of

local path re-planning is the low computational

requirements, where real-time methods such as line-

of-sight (LOS) and potential fields are common.

Polvara et al. (2018) presented a review of global

and local planning methods including a section for

advanced computing-based methods. The author

stated that almost all of the methods reviewed did not

consider uncertainties due to environment loads and

vehicle dynamics. A recent review of trajectory

planning and tracking review (for autonomous

driving systems) concluded that even most advanced

guidance and control algorithms, with today’s

available sensor technology, work well under

regulated environments assuming knowledge of

surroundings and weather conditions. It also states

that the inclusion of vehicle dynamics and

environmental loads increases the effectiveness of

such controllers (Dixit et al., 2018). Lately, Wiig et al.

proposed an integral line-of-sight law in the presence

of constant ocean currents (2018).

LaValle in his tutorials points out that “the basic

problem of computing a collision-free path for a robot

among known obstacles is well understood and

reasonably solved; however, deficiencies in the

problem formulation itself and the demand of

engineering challenges in the design of autonomous

systems raise important questions and topics for

future research” (LaValle, 2011, p. 108)

Polvara et al in their recent review stated the

following: “It has been concluded that almost all the

existing methods do not address sea or weather

conditions, or do not involve the dynamics of the

vessel while defining the path. Therefore, this

research area is still far from being considered fully

explored.” (Polvara et al., 2018, p. 241).

2.3 Topic 3: Navigation

The navigation system collects data from various

sources such as sensors, cameras and satellites, and

transfers the data into information of two kinds, state

estimation and environment perception. State

estimation is information about the ship’s motion,

mainly location and velocities. Environment

perception is weather information, wind, waves,

currents, and information about the surrounding as

well, including state of target ships and objects. The

scope of this system vastly increases as level of

automation increases; the number of datasets, their

resolution, frequency, quality and size are vastly

increasing in remotely controlled vessels comparing

to conventional ones. Moreover, since making sense

of the collected data is considered part of the

navigation system, its scope should then include

advanced computing methods in order to deliver a fit-

for-purpose output. Methods such as machine

learning, sensor fusion, computer vision, prediction,

and anomaly detection are now used within the

navigation system for making sense of the collected

data.

On board sensors are susceptible to disturbances

that come from the environment, ship motion and

other noise sources. The disturbances cause

uncertainties in the perception model. This leads to

control errors that accumulate over time, and result in

undesired control behavior. Therefore, data from

multiple sources are correlated against each other to

calculate position and velocity estimates as accurate

as possible. Data sources involved in a navigation

system are:

1 Inertial measurement unit (IMU) is an onboard

three-dimensional navigation system that

comprises of three mutually-orthogonal

accelerometers and three gyroscopes to give the

position, velocity and altitude of own ship. IMU is

often used with (and aided by) satellite positioning

to provide drift-free positioning.

2 Automatic Identification System (AIS) is a very-

high frequency communication system used by

ships to transmit their identity, position, velocity,

destination and other information and in return

they receive information of nearby ships. Even

though AIS is mandatory for commercial vessels,

not all boats have it onboard!

3 GNSS is a global positioning solution system. It

transmits radio signals from satellites orbiting the

planet to the ship. There are a number of GNSS

solution providers including GPS, GLONASS,

Beidou and Galileo.

4 Radar, an acronym for radio detection and

ranging, uses radio waves to detect ships and

obstacles within a long range but its capability of

detecting small moving targets is limited. Radar

wavelength passes through fog and rain and it

provides nearly all-weather data imagery.

5 Lidar, an acronym for light detection and ranging,

is a high resolution and accuracy object detection

sensor for near-range.

6 Sonar, an acronym for sound navigation ranging,

detects submerged objects such as reefs, sunken

ships and submarines. The sonar transmits

ultrasonic pulses, receives the reflected echoes and

displays a picture of the detected objects.

7 Other types of sensors and tools are used for

navigation purposes such as cameras, infrared

sensors, compass systems, navigation lights and

ship whistles.

Most common method for fusing the navigation

data as of today is the Kalman filter. The Kalman

filter, invented by Kalman in 1960, is a real-time

Bayesian estimation algorithm that uses all available

measurements over time, and uses knowledge of

deterministic and statistical properties of the system

parameters in order to provide optimal minimum-

error state estimations (Groves, 2013).

Examples of recent perception technologies in

navigation systems are:

1 Non-linear observers: Advanced alternatives to the

well-established Kalman filter, with proven

stability properties and lower computational

demands (Fossen and Strand, 1999; Aschemann,

Wirtensohn and Reuter, 2016; Bryne, 2017).

2 Extended Kalman filter (EKF) for position and

velocity estimation using GPS and compass

measurements (Caccia et al., 2008; Bibuli et al.,

2009; Tran et al., 2014).

3 Unscented Kalman filter (UKF) for state estimation

without previous knowledge of noise

characteristics (Peng, Han and Huang, 2009;

Vasconcelos, Silvestre and Oliveira, 2011).

4 Inverted Kalman filer (IKF) bounds model

uncertainties that come from environment

variability (Motwani et al., 2013).

5 The eXogenous Kalman filter (XKF) for providing

covariance estimates for the estimated states

generated by non-linear observes (Johansen and

Fossen, 2017).

6 Wave information perception using camera (Liu

and Wang, 2013). Stereo vision system that

generate probabilistic hazard maps and provide

estimates for speed and heading of target objects

(Huntsberger et al., 2011).

2.4 Topic 4: Control

The control system is responsible to translate the

information collected from the guidance system and

communicate it with the actuators as commands.

Actuators such as propellers, thrusters, and rudder

receive commands from the control system and

execute actions producing forces and moments that

affect the state of the ship, approaching the desired

state. The control system is responsible to make sure

that the generated actuator commands are practical

for the underactuated ship given the actuator

limitations and ship dynamics.

Control literature is rich with control design

approaches that extend from the classical

proportional-integral-derivative (PID) controllers to

the more advanced artificial-intelligence (AI) based

controllers. Practical ship control often applies a

combination of different control methods. PID control

approaches are the most favored, they are, however,

suitable for single-input-single-output cases such as

heading control (Minorsky, 1922). This approach

could suffer severe actuator damage caused by high

waves. Simultaneous control of velocity and heading

solves this problem. Multivariable control was

realized by multi-loop PID control (Lefeber, Pettersen

and Nijmeijer, 2003) and fuzzy adaptive control

techniques (Le et al., 2003).

Multivariable control has been widely approached

by optimal control techniques such as H-infinity and

Linear quadratic optimal techniques. Nonetheless,

Linear Quadratic Regulator (LQR) controller suffers

from the assumption that all states are measurable

and known, which is not the case. Linear Quadratic

Gaussian (LQG) controller together with a Kalman

filter estimates in real-time the unknown states,

however, suffers from instability. Instability outside

predefined domain and discontinuities are major

drawbacks of adaptive linear control methods (Liu et

al., 2016). Non-linear methods, such as Fuzzy logic

control, Neural networks and Lyapunov-based

methods argue that they can potentially overcome

stability related issues while maintain smooth time-

parametrized trajectories (Aguiar and Hespanha,

2003).

2.5 Topic 5: Human Factors

In this section the definition of levels of automation

(or autonomy; since both terms are used

interchangeably) is presented and followed by

explanations of the human factors faced by operators

introduced to increased automation in their

operations.

2.5.1 Levels of automation (LOA)

Levels of automation were developed in the 1978.

They were used to describe systems and aid the

communication in the design phase of automated

systems (Sheridan and Verplank, 1978). Multiple

versions of LOAs have been issued since then. In the

ship industry, LOA proposals exist from multiple

sources such as Bureau Veritas, Lloyd’s Register, the

Norwegian Forum for Autonomous Ships (NFAS),

Rolls-Royce, and others. Table 1 shows the LOAs as

proposed by NFAS. General agreement exists in the

different definitions as they range from human-

operated ship (lowest level) to fully autonomous ship

(highest level).

Explicitly, all the different variations of LOA

classifications, agree that, on the highest level of

automation, the machine decides and acts, and

requires no communication with the human.

2.5.2 Increased automation

Automation is intended to increase safety and

efficiency, however, in complex tasks (dynamic

environments involving many variables) it changes

the nature of the human-role in the task, it affects

areas such as workload and cognitive demands.

Moreover, the resultant impact of (increased)

automation turns out to be more complex than

anticipated. The changes are qualitative in context

rather than quantitative and uniform (Woods et al.,

1996). Main human factors involved in the operator-

technology interface are summarized as follows,

including responsibility, surprises of automation,

management by exception and communication:

1 Responsibility: Decisions that the human operator

is used to take and implement will be routinely

delegated to machines. However, can

responsibility be delegated as well? Responsibility

perception and calibration of trust between

humans and machines are important to safe

autonomous operations (Muir, 1987).

Jordan was one of the first to stress out that “we

can never assign them (i.e., the machines) any

responsibility for getting the task done;

responsibility can be assigned to man only”

(Jordan, 1963, p. 164). As suggested by Billings,

human operators bear ultimate responsibility for

operational goals, they must be in command, well

involved and well informed about ongoing

autonomous activities (Billings, 1991).

Table 1. LOAs as proposed by NFAS (Rødseth and Nordahl,

2017).

_______________________________________________

Level LOA Description

name

_______________________________________________

1 Decision This corresponds to today’s and

support tomorrow’s advanced ship types with

relatively advanced anti-collision radars

(ARPA), electronic chart systems and

common automation systems like

autopilot or track pilots. The crew is still

in direct command of ship operations

and continuously supervises all

operations. This level normally

corresponds to "no autonomy".

2 Automatic The ship has more advanced

automation systems that can complete

certain demanding operations without

human interaction, e.g. dynamic

positioning or automatic berthing. The

operation follows a pre-programmed

sequence and will request human

intervention if any unexpected events

occur or when the operation completes.

The shore control centre (SCC) or the

bridge crew is always available to

intervene and initiate remote or direct

control when needed.

3 Constrained The ship can operate fully automatic in

autonomous most situations and has a predefined

selection of options for solving

commonly encountered problems, e.g.

collision avoidance. It has defined limits

to the options it can use to solve

problems, e.g. maximum deviation from

planned track or arrival time. It will call

on human operators to intervene if the

problems cannot be solved within these

constraints. The SCC or bridge

personnel continuously supervises the

operations and will take immediate

control when requested to by the

system. Otherwise, the system will be

expected to operate safely by itself.

4 Fully The ship handles all situations by itself.

autonomous This implies that one will not have an

SCC or any bridge personnel at all. This

may be a realistic alternative for

operations over short distances and in

very controlled environments.

However, and in a shorter time

perspective, this is an unlikely scenario

as it implies very high complexity in

ship systems and correspondingly high

risks for malfunctions and loss of

system.

_______________________________________________

2 “Automation surprises”: It could be difficult for

the operators to follow up with the autonomous

vehicle and understand the grounds for its

decisions. When the actions of the “machine” are

not similar to what the human operator would do

if placed in the same situation then the human

would lose track and fail to predict next steps. A

simulator experiment to evaluate pilots’ mode

awareness was carried out that confirmed that

“automation surprises” are experienced even by

operators with extensive amount of line experience

on similar highly autonomous aircrafts. It was

shown that in non-normal situations, more

problems related to “automation surprises”

occurred (Sarter and Woods, 1994). A previous

study by Wiener, who conducted a survey of B-757

pilots, resulted that 55% of respondents were still

being surprised by the automation after more than

one year of line experience on the aircraft (Wiener,

1989). Norman referred to the phenomenon of

human operator losing track of machine’s behavior

as ‘breakdown in mode awareness’ which has been

linked strongly to the following factors:

automation surprises, increased error possibilities,

new cognitive demands, and failure to intervene

appropriately. Increased automation would also

cause surprise to the ship designers and owners

who experience unexpected consequences because

their automated system fails to behave as was

intended (Norman, 1988).

3 Management by exception: A remote operator,

whether monitoring or supervising, is in a double

bind dilemma with the machine. A dilemma

between trust and takeover. Dekker and Woods

explained this phenomenon in their work titled

“To intervene or not to intervene: the dilemma of

management by exception” (Dekker and Woods,

1999). Supervisory control places the operator in a

decision-making situation. A trade-off between

intervening too early, before enough evidence is

collected about the situation, and intervening too

late, after it escalades into an irreversible crisis.

The operators, for every moment in time, must

assess the criticality of the situation and decide

whether to intervene or not. Late decisions are

catastrophic. Early decisions are not justified.

Decision aids and prediction tools are required,

but how much should they be trusted? (Sheridan,

2000).

Human-machine interaction is changing in nature.

Increased automation reduces workload in

normal-times and increase them dramatically in

non-normal times. In non-normal times the

‘automation surprises’ factor is higher, the ‘mode

awareness’ factor is lower, the attentional

demands and the cognitive demands are highly

increased. Given the dilemma, this setting is

critical in non-normal times as it leads to less

situational awareness (SA) and less intervention

capabilities. Thus, safety is a big concern if things

went wrong in non-normal times. Sarter et al

define the term Mode awareness as “the ability of

a human operator to track and anticipate the

behavior of automated system” (Sarter, Woods

and Billings, 1997, p. 6). Situational awareness,

according to Endsley, is “the perception of the

elements in the environment within a volume of

space and time, the comprehension of their

meaning, and the projection of their status in the

near future” (Endsley, 1995, p. 36).

4 Communication: For example, the grounding of

the Royal Majesty is referred to as a loss of

situational awareness problem; a communication

problem because of increased automation. Among

other factors, the GPS has failed, positioning

information were incorrect, autopilot used the

faulty information, the ship drifted and that was

not apparent to the crew. They believed that the

sailing was flawless but in fact, it lead to a

grounding (Lützhöft and Dekker, 2002).

Researchers emphasized on the value of

communication and collaboration with the

machine for safer autonomous navigation.

Effective communication and coordination

between humans and machines is believed to be

key for successful operations (Sarter, Woods and

Billings, 1997).

3 RESULTS

This review covers topics concerning the future of

autonomous vessels from three perspectives. First

from the side of the technology advancements that

make such a future possible. Second from the human

operator side and the challenges faced while teams

are operating highly autonomous systems, remotely.

Third from the levels-of-automation side, multiple

versions of LOA definitions for the maritime industry

that classify human-machine relationship as

automation increases. Trying to answer the article’s

question.

One may argue that the supporting technologies

are already available, as there are booming examples

of domain-specific advancement, but this review

identifies a shortage in the studies that show how

well these building blocks work out together, and

under uncertainties. Analysis and breakdown of this

identified shortage follows:

There is interaction and signal flow among the

GNC and hydrodynamics fields, as shown in Figure

1. One publication proposing a novel path planning

method would have built-in assumptions regarding

(and pre-selections of) ship dynamic models,

navigation methods, and control design approaches.

For example, Liu, Bucknall and Zhang (2017)

proposed a guidance “fast marching” method for a

USV, and presented their results of full-scale

experiments. They used a preselection of navigation

methods (Kalman filter), control methods (PID

autopilot), and vehicle dynamics model (3-D model)

as in (Motwani et al., 2013).

Figure 1. GNC module interaction and signal flow (Fossen,

2011, p. 233)

Given the interrelation, applications of semi-

autonomous vessels, both real (full-scale) and virtual

(simulators), require a package of GNC and

hydrodynamics technologies interacting together. It is

widely agreed that the performance of methods is

largely altered by uncertainties coming from

environmental loads and ship dynamics (LaValle,

2011; Liu et al., 2016; Polvara et al., 2018).

“Automation could increase sources of error”.

Precautionary perspective is necessary in research

and development. Porathe et al. (2018) includes a

fictive story that predicts a possible future scenario in

one of the Norwegian fjords and provides a forecast

of the risk picture in the maritime industry.

Human operators face challenges with highly

automated systems. In the future, as autonomous

ships become reality, advancing through the LOA

scale, until eventually, full-autonomous vessels are

realized in a safe and efficient manner, remote control

will be essential. Safe and efficient remote operations

are as important as, or even more important than, no-

human-interaction type of control (according to LOA

definitions of full autonomy). There is a literature

shortage in this multidisciplinary field of “ship

remote control”. It should cover remote control-

centered topics of ship design, GNC systems design,

human-machine interaction, navigation functions,

interface, and remote control center design.

4 DISCUSSIONS

4.1 Disagreements

Viewpoints such as “A ship must follow and adhere

to the international regulations for preventing

collisions at sea (COLREGS)” are common in GNC

technology research. However, these viewpoints

oversimplify the problem. They inherently assume

that traffic in the sea is well regulated and all players

follow the rules. In reality, operators and crew do

violate procedures, for different reasons, as shown by

a research that collected 1262 questionnaires from

tankers and bulk carriers crew (Oltedal, 2011).

Some collision avoidance methods enable manual

input of waypoints by a supervisor operator for

replanning the path and avoiding approaching

obstacle. As Campbell et al describe them: “This is not

the most efficient method for avoidance and is subject

to operator error” (Campbell, Naeem and Irwin,

2012). This view is common. It promote two points.

First, that human operators are subject to more errors

than machines. Second, researchers are oriented to

develop technologies with high automation level and

low human interaction, to avoid human errors. This

view conflicts with the status quo of technology,

because also machines are subject to error, and it

conflicts with human factors research, that having less

human interaction with highly automated system

introduces the dilemma of management by exception

and it can be avoided by having human input and

authority over the machines even for highly

autonomous systems.

In a recent survey on communication technologies

(Zolich et al., 2018) a relation of LOAs with

communication requirements was presented. It says,

basically, that the higher the LOA is, the lower the

amount of data the ship would require to

communicate with land. This view conflict with the

human operator’s requirements for safe and efficient

monitoring, supervision and control of the

autonomous remote asset.

The definition of full-autonomy, in all the

variations of LOA scales, emphasizes on “no human

interaction; machine ignores human; no human

input”. These definitions favor automation over safety

and efficiency of the asset because full-autonomous

ships need to be remotely controlled, on demand,

upon the decision of the supervisor in charge. In such

a dynamic multi-objective shipping task, the option of

remote control is necessary; the reason for this desire

of remote control could be any of the following

examples:

 Business and market fluctuation

 Environment regulations and emission related

rules

 Cyber-attacks, piracy and hijacking

 Environment loads and extreme weather

 Incidents at ports such as fires or chain-reaction

accidents

 Customer relations; cargo health; maintenance

issues and etc.

4.2 Main challenges

Main challenges from the different perspectives are

summed up in this section as follows.

Motion coupling: control advancements consider

a simplified ship model, similar to that of a 3-D

unicycle model. The effect of motion coupling to

stability requires further analysis.

Ship motion in waves: Describing ship motion in

harsh weather is a challenge; there is no standard way

of doing it. Hydrodynamic research considers that

ship motion in calm water is assumed to converge to

an underlying true trajectory. The maneuvering

committee of the 27

ITTC address this issue as a

challenge for both experimental work and numerical

modelling (Quadvlieg et al., 2014, sec. 6.4).

Co-simulation of digital models: The

development of GNC algorithms has boomed lately. It

is challenging to know how they will work together

under the influence of stochastic environmental

loading and uncertainties. In addition, how will the

human (remote) operator experience those

advancements?

Remote operator input: How well does these

technologies workout together? Research towards

enhancing the performance of man-machine systems

in dynamic control tasks is crucial in design and

operation of future maritime operations. Effects of

LOA towards situation awareness and mental

workload are researched (Kaber and Endsley, 2004).

However, it is challenging to judge automation based

on the LOA scale because the whole scale is course;

massive variations could be possible within one LOA

level. Variations in terms of interface, controllers,

inputs, outputs and engagement level are expected

for each level.

4.3 Full Autonomy

The main challenges of the previous section maps

man-machine challenges that are valid for the

maritime industry as of today. Worldwide research

and development projects will certainly tackle them

and innovations will pave the way, gradually, to

realizations of higher levels of ship autonomy. The

progress will be gradual, evolutionary rather than

revolutionary, because of the political, legal and

financial inertia involved in such industry.

Systematic bias: Assume that “we are dealing

with a transition towards fully autonomous systems”

with the main objectives “safety and efficiency”. The

way the developers perceive the future is key in

determining the safety and efficiency of that future.

The definitions of LOAs form a huge anchoring bias

that weakens the focus on the objectives and

strengthens the following views:

1 The ultimate goal is full autonomy

2 Full autonomy is that systems run by themselves

with no human interaction

3 Human input is a negative contribution to system

objectives

And those views are expanding systematically

within and across industries and can be seen popular

in technical scientific disciplines and among the youth

in societies of most industrial countries. Broek et al

(2017) mentioned the need of a man-machine

“collaboration framework” even for fully autonomous

systems.

Towards full-autonomy: As it brings value to

other industries, the values of advanced technology

must be harvested in the shipping industry as well.

We strive for fewer accidents, less social and

environmental impact by the utilization of tools such

as data analytics, decision support aids, and advanced

autopilots. Surprisingly, the GNC literature shows

that technology is being developed towards a future

with no human interaction. However, I think that the

values of full-autonomy cannot be harvested unless

the technology becomes developed towards a future

with full human interaction.

5 CONCLUSIONS

If the industry’s drive is safety and efficiency, then

full-autonomy is, at present, not the way to go.

Remote control, instead, could facilitate a feasible

future, while focused research and development are

in need. From the technology side, the literature

shows that uncertainties coming from environmental

loads and ship dynamics largely affect the

performance of GNC technologies in a semi-

autonomous vessel. Thus, accurate modeling and

prediction of semi-autonomous maneuvering is

fragile under uncertainties. From the human side, the

literature shows that as automation is increased and

interaction is decreased the operators face the

management by exception dilemma. Operators

undertaking safe and efficient ship remote control,

even for highly autonomous ships, require high

interaction and high authority over the system.

Automation is promising because of the possible

reduction of cost and risk involved in maritime

operations, nevertheless; it could bring in new sources

of error, while human operators face serious

challenges dealing with highly automated systems.

There is a rush of technology-related research but

there is a lack of holistic research focusing on “ship

remote control”. Research that tests the GNC

technologies under uncertainties with human-

operator in-the-loop is needed. Digital advancements

enable virtual experiment environments with human

interaction such as simulators. Those safe

environments could be the only tools available, for

now, to enable us research whether it is full-

autonomy the right way to go for exploiting the ocean

potentials.

REFERENCES

Aguiar, A. P. and Hespanha, J. P. (2003) ‘Position Tracking

of Underactuated Vehicles’, Acc 2003. doi:

10.1109/ACC.2003.1243366.

Aschemann, H., Wirtensohn, S. and Reuter, J. (2016)

‘Nonlinear Observer-Based Ship Control and

Disturbance Compensation’, IFAC-PapersOnLine.

Elsevier B.V., 49(23), pp. 297–302. doi:

10.1016/j.ifacol.2016.10.358.

Bibuli, M. et al. (2009) ‘Path-following algorithms and

experiments for an unmanned surface vehicle’, Journal of

Field Robotics. doi: 10.1002/rob.20303.

Billings, C. E. (1991) Human-Centered Aircraft Automation: A

Concept and Guidelines, NASA Technical Memorandum

103885. doi: 10.1207/s15327108ijap0104_1.

Broek, V. der et al. (2017) ‘Approaching full autonomy in the

maritime domain: paradigm choices and Human Factors

challenges’, in MTEC 2017, pp. 1–11.

Bryne, T. H. (2017) Nonlinear Observer Design for Aided

Inertial Navigation of Ships. Available at:

http://hdl.handle.net/11250/2440519.

Caccia, M. et al. (2008) ‘Basic navigation, guidance and

control of an Unmanned Surface Vehicle’, Autonomous

Robots, 25(4), pp. 349–365. doi: 10.1007/s10514-008-9100-

Campbell, S., Naeem, W. and Irwin, G. W. (2012) ‘A review

on improving the autonomy of unmanned surface

vehicles through intelligent collision avoidance

manoeuvres’, Annual Reviews in Control. Elsevier Ltd,

36(2), pp. 267–283. doi: 10.1016/j.arcontrol.2012.09.008.

Cummins, W. E. (1962) ‘The impulse response function and

ship motions’, Symposium on Ship Theory. Hamburg.

Dekker, S. W. . and Woods, D. . (1999) ‘To intervene or not

to intervene the dilemma of management by exception’,

Cognition, Technology & Work. doi:

10.1007/s101110050035.

Dixit, S. et al. (2018) ‘Trajectory planning and tracking for

autonomous overtaking: State-of-the-art and future

prospects’, Annual Reviews in Control. Elsevier Ltd, 45,

pp. 76–86. doi: 10.1016/j.arcontrol.2018.02.001.

Endsley, M. R. (1995) ‘Toward a Theory of Situation

Awareness in Dynamic Systems’, Human Factors: The

Journal of the Human Factors and Ergonomics Society. doi:

10.1518/001872095779049543.

Farell, J. A. (2008) Aided Navigation: GPS with High Rate

Sensors, The McGraw-Hill Companies. doi:

10.1017/CBO9781107415324.004.

Fossen, T. I. (1991) ‘Adaptive control of nonlinear

underwater robotic systems’, Modeling, Identification and

Control, 12(2), pp. 95–105.

Fossen, T. I. (2011) Handbook of Marine Craft Hydrodynamics

and Motion Control, John Wiley & Sons Ltd.

Fossen, T. I. and Strand, J. P. (1999) ‘Passive nonlinear

observer design for ships using Lyapunov methods’,

Automatica, pp. 3–16.

Groves, P. (2013) Principles of GNSS, Inertial, and Multisensor

Integrated Navigation Systems, Journal of Chemical

Information and Modeling. doi: 10.1109/MAES.2014.14110.

Hermundstad, E. M. and Hoff, J. R. (2009) ‘Time domain

simulation of the maneuvering of a vessel in a seaway’,

in Proceedings of the International Conference on Offshore

Mechanics and Arctic Engineering - OMAE. doi:

10.1115/OMAE2009-79946.

Huntsberger, T. et al. (2011) ‘Stereo vision-based navigation

for autonomous surface vessels’, Journal of Field Robotics.

doi: 10.1002/rob.20380.

Johansen, T. A. and Fossen, T. I. (2017) ‘The eXogenous

Kalman Filter (XKF)’, International Journal of Control. doi:

10.1080/00207179.2016.1172390.

Jordan, N. (1963) ‘Allocation of functions between man and

machines in automated systems’, Journal of Applied

Psychology. doi: 10.1037/h0043729.

Kaber, D. B. and Endsley, M. R. (2004) ‘The effects of level of

automation and adaptive automation on human

performance, situation awareness and workload in a

dynamic control task’, Theoretical Issues in Ergonomics

Science. doi: 10.1080/1463922021000054335.

Kim, Y. et al. (2014) Seakeeping Committee Final Report and

Recommendations to the 27th ITTC. Copenhagen.

Available at:

https://ittc.info/downloads/proceedings/27th-

conference-copenhagen-2014/volume-1/ on 19/12/2018.

LaValle, S. (2011) ‘Motion Planning: Part II’, IEEE Robotics &

Automation Magazine, 18(2), pp. 108–118. doi:

10.1109/MRA.2011.941635.

Le, M. D. et al. (2003) ‘A new and effective fuzzy PID

autopilot for ships’, in Proceedings of IEEE International

Symposium on Computational Intelligence in Robotics and

Automation, CIRA. doi: 10.1109/CIRA.2003.1222204.

Lefeber, E., Pettersen, K. Y. and Nijmeijer, H. (2003)

‘Tracking control of an underactuated ship’, IEEE

Transactions on Control Systems Technology. doi:

10.1109/TCST.2002.806465.

Liu, K. K. and Wang, J. H. (2013) ‘A Method of Detecting

Wave Grade Based on Visual Image Taken by USV’,

Applied Mechanics and Materials. Trans Tech Publications,

291–294, pp. 2437–2441. doi: 10.4028/

www.scientific.net/AMM.291-294.2437.

Liu, Y., Bucknall, R. and Zhang, X. (2017) ‘The fast marching

method based intelligent navigation of an unmanned

surface vehicle’, Ocean Engineering. Elsevier Ltd,

142(April), pp. 363–376. doi: 10.1016/

j.oceaneng.2017.07.021.

Liu, Z. et al. (2016) ‘Unmanned surface vehicles: An

overview of developments and challenges’, Annual

Reviews in Control. Elsevier Ltd, 41, pp. 71–93. doi:

10.1016/j.arcontrol.2016.04.018.

Lützhöft, M. H. and Dekker, S. W. A. (2002) ‘On your watch:

Automation on the bridge’, Journal of Navigation. doi:

10.1017/S0373463301001588.

Minorsky, N. (1922) ‘Directional Stability of Automatically

Steered Bodies’, Journal of the American Society for Naval

Engineers. doi: 10.1111/j.1559-3584.1922.tb04958.x.

Motwani, A. et al. (2013) ‘Interval Kalman filtering in

navigation system design for an uninhabited surface

vehicle’, Journal of Navigation. doi: 10.1017/

S0373463313000283.

Muir, B. M. (1987) ‘Trust between humans and machines,

and the design of decision aids’, International Journal of

Man-Machine Studies. doi: 10.1016/S0020-7373(87)80013-5.

Norman, D. A. (1988) The Psychology of Everyday Things, New

York, Basic Books. New York, Basic Books. doi:

10.2307/1423268.

Oltedal, H. A. (2011) Safety culture and safety management

within the Norwegian-controlled shipping industry.

University of Stavanger.

Peng, Y., Han, J. and Huang, Q. (2009) ‘Adaptive UKF based

tracking control for Unmanned Trimaran Vehicles’,

International Journal of Innovative Computing, Information

and Control, 5, pp. 3505–3515.

Pérez, T. (2005) Ship Motion Control: course keeping and roll

stabilisation using rudder and fins. Springer. doi: 10.1007/1-

84628-157-1.

Polvara, R. et al. (2018) ‘Obstacle Avoidance Approaches for

Autonomous Navigation of Unmanned Surface

Vehicles’, Journal of Navigation, 71(1), pp. 241–256. doi:

10.1017/S0373463317000753.

Porathe, T. et al. (2018) ‘At least as safe as manned shipping?

Autonomous shipping, safety and “human error”’, in

Safety and Reliability – Safe Societies in a Changing World.

Proceedings of ESREL 2018, June 17-21, 2018, Trondheim,

Norway.

Quadvlieg, F. et al. (2014) Manoeuvring Committee: Final

Report and Recommendations to the 27th ITTC.

Copenhagen. Available at:

https://ittc.info/downloads/proceedings/27th-

conference-copenhagen-2014/volume-1/ on 19/12/2018.

Rifkin, J. (2012) The Third Industrial Revolution- How Lateral

Power Is Transform-ing Energy, The Economy, and the

World, World Future Review (World Future Society). doi:

978-0230341975.

Rødseth, Ø. J. and Nordahl, H. (2017) Definitions for

autonomous ships by NFAS. Available at:

http://nfas.autonomous-ship.org/resources/autonom-

defs.pdf.

Sarter, N. B. and Woods, D. D. (1994) ‘Pilot Interaction With

Cockpit Automation II: An Experimental Study of Pilots’

Model and Awareness of the Flight Management

System’, The International Journal of Aviation Psychology.

doi: 10.1207/s15327108ijap0401_1.

Sarter, N., Woods, D. D. and Billings, C. E. (1997)

‘Automation surprises’, Handbook of human factors and

ergonomics. doi: 10.1207/s15327108ijap0204_5.

Seo, M. G. and Kim, Y. (2011) ‘Numerical analysis on ship

maneuvering coupled with ship motion in waves’, Ocean

Engineering. doi: 10.1016/j.oceaneng.2011.09.023.

Sheridan, T. B. (2000) ‘HCI in supervisory control: Twelve

dilemmas’, in Human Error and System Design and

Management. doi: 10.1038/scientificamerican0798-64.

Sheridan, T. B. and Verplank, W. L. (1978) Human and

Computer Control of Undersea Teleoperators.

Skejic, R. and Berg, T. E. (2010) ‘Combined seakeeping and

maneuvering analysis of a two ship lightering

operation’, Proceedings of the 29th Annual International

Conference on Offshore Mechanics and Arctic Engineering

(OMAE 2010). doi: 10.1115/OMAE2010-20616.

Skejic, R. and Faltinsen, O. M. (2008) ‘A unified seakeeping

and maneuvering analysis of ships in regular waves’,

Journal of Marine Science and Technology, 13(4), pp. 371–

394. doi: 10.1007/s00773-008-0025-2.

Skejic, R. and Faltinsen, O. M. (2013) ‘Maneuvering behavior

of ships in irregular waves’, in Proceedings of the ASME

32nd International Conference on Ocean, Offshore and Arctic

Engineering OMAE 2013. Nantes, France.

Tran, N.-H. et al. (2014) Tracking Control of an Unmanned

Surface Vehicle, AETA 2013: Recent Advances in Electrical

Engineering and Related Sciences. Edited by I. Zelinka, V.

H. Duy, and J. Cha. Berlin, Heidelberg: Springer Berlin

Heidelberg.

Vasconcelos, J. F., Silvestre, C. and Oliveira, P. (2011)

‘INS/GPS Aided by Frequency Contents of Vector

Observations With Application to Autonomous Surface

Crafts’, IEEE Journal of Oceanic Engineering, 36(2), pp.

347–363. doi: 10.1109/JOE.2011.2126170.

Wiener, E. L. (1989) Human Factors of Advanced Technology

(‘Glass Cockpit’) Transport Aircraft, NASA-CR-177528. doi:

10.13246/j.cnki.iae.2015.02.004.

Wiig, M. S. et al. (2018) ‘An Integral Line-of-Sight Guidance

Law with a Speed-dependent Lookahead Distance’, in

European Control Conference 2018 (ECC 2018). Limassol,

Cyprus, pp. 1269–1276. doi: 10.23919/ECC.2018.8550243.

Woods, D. D. et al. (1996) ‘Bridging The Gap Between User-

Centered Intentions and Actual Design Practice’, in

Proceedings of the Human Factors and Ergonomics Society

40th Annual Meeting-1996, pp. 967–971.

Wu, W., Spyrou, K. J. and McCue, L. S. (2010) ‘Improved

prediction of the threshold of surf-riding of a ship in

steep following seas’, Ocean Engineering. doi:

10.1016/j.oceaneng.2010.04.006.

Yasukawa, H., Amri Adnan, F. and Nishi, K. (2010) ‘Wave-

Induced Motions on a Laterally Drifting Ship’, Ship

Technology Research. doi: 10.1179/str.2010.57.2.001.

Yasukawa, H. and Nakayama, Y. (2009) ‘6-DOF motion

simulations of a turning ship in regular waves’, in

International Conference on Marine Simulation and Ship

Manoeuvrability.

Zolich, A. et al. (2018) ‘Survey on Communication and

Networks for Autonomous Marine Systems’, Journal of

Intelligent and Robotic Systems: Theory and Applications,

pp. 1–25. doi: 10.1007/s10846-018-0833-5