http://underwatertrex.blogspot.com/

http://underwatertrex.blogspot.com/

Sensing networks and beyond

Sensing networks and beyond
The elements previously described are the most immediate
requirements for unmanned platforms in the underwater domain.
However, other long-term subsea applications are appearing
posing other problems and challenges. The planning
community can help to overcome them.
Ocean observatories in oceanology, all year round subsea
field inspection for the energy industry and continous
harbour and coast patroling for the Navy are emerging, requiring
a more permanent presence of robotic and sensing
tools underwater. These applications are demanding underwater
networks of fixed sensors in combination with fleets
of AUVs and gliders. These sensing webs require decision
making algorithms in order to be able to optimize the management
of heterogeneous assets and resources, to couple
observation systems and ocean models, to minimize error
in the reconstruction of ocean fields and to provide fast dynamic
response to events.

Collaborative planning

Collaborative planning

Mission plan adaptation can be extended to multiple platforms.

The potential benefits of multi-vehicle operations

include force-multiplier effects, redundancy avoidance and

the utilisation of heterogeneous vehicle configurations to reduce

risk to expensive assets. But to achieve the full benefits,

suitable architectures that facilitate true cooperative autonomous

planning for task splitting and allocation must be

realised (Evans et al. 2006).

In this case, it is necessary to extend the single vehicle

mission plan recovery to a mission plan recovery for a group

or team of vehicles performing a collaborative mission. A

shared situation awareness is required for the team of vehicles

(SAT ) to which every team member possess the SAS

required for its responsibilities (Cartwright et al. 2007). The

main challenge in this case is to deal with the acoustic communication

limitations associated to the underwater environment

(Sotzing, Evans, and Lane 2007).

Collaborative planning

Collaborative planning

Mission plan adaptation can be extended to multiple platforms.

The potential benefits of multi-vehicle operations

include force-multiplier effects, redundancy avoidance and

the utilisation of heterogeneous vehicle configurations to reduce

risk to expensive assets. But to achieve the full benefits,

suitable architectures that facilitate true cooperative autonomous

planning for task splitting and allocation must be

realised (Evans et al. 2006).

In this case, it is necessary to extend the single vehicle

mission plan recovery to a mission plan recovery for a group

or team of vehicles performing a collaborative mission. A

shared situation awareness is required for the team of vehicles

(SAT ) to which every team member possess the SAS

required for its responsibilities (Cartwright et al. 2007). The

main challenge in this case is to deal with the acoustic communication

limitations associated to the underwater environment

(Sotzing, Evans, and Lane 2007).

Mission planning

Mission planning
Single platform
Current mission plan solutions for underwater platforms
are procedural and static (Hagen 2001). If behaviors are
added (Pang et al. 2003), they are only to cope with possible
changes that are known a-priori by the operator. In order
to achieve higher levels of autonomy, plan adaptability
should be available not only to procedural waypoint-based
approaches on trajectory planning but also to a declarative
goal-based solution for adaptive mission planning.
On a goal-oriented command and control, decision tools
can be developed providing support for adaptive data sampling,
optimization of resources and fault recovery capabilities
of the platform (Bellingham, Kirkwood, and Rajan
2006).
The potential benefits of the planning capabilities for
adaptive data sampling have been studied by Rajan (Rajan et
al. 2007) and Fox (Fox et al. 2007). An example of this application
is the T-Rex system, an adaptive planning architecture
with notions of timing and onboard resource management
responsiveness to opportunistic science events (Mc-
Gann et al. 2007).

In a similar way, fault-tolerant mission planning adaptation
can also contribute towards providing more mission
flexibility and robustness. When active and passive measures
fail to protect the vehicle, or unexpected hardware failures
occur, the focus of the mission should shift to ’reconfigure’
itself to use alternative combinations of the remaining
resources (see Fig. 9). In the underwater environment, autonomous
embedded recoverability, is a key capability for
vehicle’s endurance. This can be achieved via adaptation of
the vehicle’s mission plan (Patr´on et al. 2008c). The aim
is to provide with a mission plan as close as possible to the
original. In such case, plan optimality during the adaptation
process can be sacrified for plan stability (Fox et al. 2006).
This level of control can also benefit from the use of
knowledge frameworks. Adaptation is limited to the quality
and scope of the available information. Therefore, to adapt
mission plans due to unforeseen and incipient faults, it is
required that accurate information is available, to recognize
that a fault has occurred and deduce the root cause of the
failure.
Current robotic solutions are generally equipped with
Failure Diagnosis and Detection (FDD) systems based on
damage control that results in the vehicle resurfacing in
the event of any fault in the system. But future operations
will require long-endurance unattended systems that operate
even while being partially damaged or while having constrained
capabilities. Hence, it is important to develop system
which not only detect failure in the component but also
provide mission plan modules capable of adapting and recovering
from these failures. This exchange of meaningful
and reliable knowledge between FDDs and mission planners
can leverage on the use of semantic knowledge frameworks.
Once the knowledge has been transferred from the FDD,
the adaptation approach combines robust execution and mission
plan repair in order to maximize system performance
and response time. Two levels of repair are possible. If the
fault coming from the FDD is incipient or non-critical, a repair
at the executive level is performed. This entails adapting
the intances and parameters of the action performing the
task without changing the action itself. If a critical fault on
a component is diagnosed, a repair at the plan level is performed
that changes the action requiring of the component

and the constraints on the action. During this decision making
process, a combination of refinement and unrefinement
phases of the constraints of a partial ordered mission plan
can be used (Patr´on et al. 2008b) (see Fig. 10).

Obstacle avoidance

Obstacle avoidance
Once the navigation and mapping error has been bounded,
an adaptation of the trajectory and waypoints is required
during mission when unexpected events on the planned trajectory
are sensed. Collision avoidance and escape is a key
capability for underwater vehicle navigation.
Traditional approaches to collision avoidance and escape
are purely reactive and usually involve an algorithmic formulation
of response dictated by the objects geometries and
locations and the relative location of the vehicle. Recent approaches
in rules of collision (Benjamin et al. 2006) and
obstacle avoidance and escape scenarios (Evans et al. 2007)
have focused on reducing susceptibility by looking at the
adaptation of the vehicle’s trajectory plan. These deliberative
behaviours provide the defined and bounded event responses
that link to geometric methods of trajectory planning.
At the trajectory planning level, lifelong planning incremental
search methods have been found to decrease computation
time while still locating the shortest trajectory between
vehicle and goal (Koenig, Likhachev, and Furcy

2004). Instead of starting from scratch every time a trajectory
must be adapted, they reuse data found in previous
searches to save on computation time. Wave propagation
techniques can also provide with smoother and shorter trajectories
than classic discrete search approaches. This fast
marching methods are capable of dealing with uncertainty,
dynamic objects and kinematics of the vehicle during the
adaptation process (Pˆetr`es et al. 2007). A real application of
these techniques in the underwater environment is shown in
Fig. 8.

Trajectory planning

Trajectory planning

Navigation and mapping

Existing geographical information for trajectory planning

can be represented using knowledge-based frameworks.

However, mapping of the environment depends on the uncertainty

in the position of the vehicle when the obstacles

are observed and mapped. Navigation in unknown underwater

environments entails high levels of uncertainty.

Geolocating a vehicle on theWorld’s surface can be accurately

performed by using Global Positioning System (GPS)

receivers. But GPS does not work underwater as the radio

signals on which it depends cannot pass through water. Dead

reckoning is the standard technique used underwater and involves

use of Inertial Measurement Units (IMUs), a combination

of accelerometers and gyroscopes. Abbe error, magnification

of angular error over distance, can be detrimental

to dead reckoning. Inertial Navigation Systems (INS) integrate

accelerations and rates to provide a Kalman filtered

navigation solution. They use inputs from acoustic transmitters

and receivers to pin-point the platform’s location. However,

the range of these systems is limited to no more than a

few nautical miles thus restricting the vehicles autonomy. A

better approach involves aiding the INS with a Doppler Velocity

Log (DVL) sensor that measures the displacement rate

over the seabed. No matter how accurate the sensors are, the

errors in these systems grow with time and the platform becomes

progressively lost unless it is able to obtain external

references from acoustic devices or from a GPS receiver on

the surface. This uncertainty in the vehicle’s location gets

passed to the geolocation of the observed elements in the

environement during the mapping process.

A promising alternative is Simultaneous Localisation And

Mapping (SLAM). Using SLAM a platform maps the environment

and uses the map to localise itself in it. The map can

be georeferenced if the platform maintains an estimate of its

absolute position when the process of mapping is started.

A recent development has shown that it is possible to postprocess

and smooth the SLAM solution using the Rauch-

Tung-Striebel smoother (see Fig.7). This new technique,

coined SLAM-RTS, has been used to create better maps of

the environment and to help the trajectory planning systems

to accurately locate the sensed obstacles in the map (Patr´on

and Tena-Ruiz 2006).

Knowledge representation and transfer

Knowledge representation and transfer

At present, knowledge representation is embryonic and targets

simple mono-platform and mono-domain applications,

therefore limiting the potential of multiple coordinated actions

between agents. Consequently, the main application

for autonomous underwater vehicles is information gathering

from sensor data. In a standard mission flow, data is

collected during mission and then post-processed off-line.

However, in order to be able to let decision making technologies

evolve towards providing higher levels of autonomy

and control, embedded service-oriented agents require

access to higher levels of knowledge representation or abstraction

(see Fig. 5). These higher levels will be required to

provide knowledge representation for contextual awareness,

temporal awareness and behavioural awareness.

Two sources can provide this type of information: the domain

knowledge extracted from the original expert (orientation)

or the inferred knowledge from the processed sensor

data (observation). In both cases, it will be necessary for the

information to be stored, accessed and shared efficiently by

the deliberative agents while performing a mission. These

agents, providing different capabilities, might even be distributed

among the different platforms working in collaboration.

Semantic frameworks have recently raise interest providing

hierarchical distributed representation of knowledge for

multidisciplinary agent interaction (Patr´on et al. 2008a).

They provide with a common machine understanding representation

of knowledge between embedded agents that is

generic and extendable. They also include a reasoning interface

for inferring new knowledge from the observed data and

knowledge stability by checking for inconsistencies. These

frameworks improve local (machine level) and global (system

level) situation awareness and context for mission and

trajectory behavior. They can therefore act as enablers for

autonomy and on-board decision making.

There are currently several institutions and consortiums

developing standards for knowledge representation under

these frameworks. Particular attention is taking the effort

describing the concepts and relationships for the domains of

unmanned platforms (jau 2008) and the underwater environments

(mmi 2008). An example is shown in Fig. 6.

Towards adaptive autonomy

Autonomous adaptation can release the operator from decision

making tasks at the trajectory and mission planning levels.

These, in consequence, can require less communication

with the consequent power saving. Adaptation plays an important

role in providing autonomy. The aim is to be effective

and efficient and a plan costs time to prepare. This time

has been already invested once (to compute the plan that is

now failing), so it might be more efficient to try to reuse previous

efforts by repairing it. Also, commitments might have

been made to the current plan: trajectory reported to other

intelligent agents, assignment of resources or assignment of

part of mission plan to executors, etc. Repairing an existing

plan ensures that as few commitments as possible are invalidated.

Finally, several planners (usually autonomous and

human planners combined) could be performing together to

achieve the goals. In such cases, it is more likely that a similar

mission plan will be accepted by the operator than one

that is potentially completely different.

Autonomous adaptation requires an autonomous understanding

of the environment. The human capability of dealing

and understanding highly dynamic and complex environments

is known as situation awareness (SAH). SAH

breaks down into perception of the environment, comprehension

of the situation and projection of the future status.

Decision making occurs in a cycle of observe-orient-decideact

(OODA) (Boyd 1995). The Observation component

corresponds to the perception level of SAH. The Orientation

component contains the previously adquired knowledge

and understanding of the situation. The Decision component

represents the SAH levels of comprehension and projection.

This last stage is the central mechanism enabling adaptation

before closing the loop with the final Action stage. Note that

it is possible to take decisions by looking only at orientation

inputs without making any use of observations.

Based on the autonomy levels and environmental characteristics,

SAH definitions can be directly applied to

the notion of unmanned vehicle situation awareness

(SAV ) (Adams 2007). Increasing the levels of situation

awareness for individual unmanned vehicle systems (SAS)

can help the transfer from current full human control to fully

autonomous unmanned capabilities (see Fig. 4). This knowledge

representation is the focus of the next section.

Robotic platforms

Unmanned underwater vehicles can be classified in Remote
Operated Underwater Vehicles (ROVs) (see Fig. 2), Autonomous
Underwater Vehicles (AUVs) (see Fig. 3) and
Underwater Gliders. They differ on the power capability,
power endurance and the task complexity that they have
been designed for.
Underwater vehicles have become a standard tool for data
gathering for Maritime applications. In these environments,
mission effectiveness directly depends on vehicle’s operability.
Operability underlies the vehicle’s final availability,
affordability and acceptance. Two main vehicle characteristics
can improve the vehicle’s operability: reliability relates
to vehicle failures due to the internal hardware components
of the vehicle, and survivability relates to vehicle failures
due to external factors or damages.
Each of these characteristics can be improved by providing
autonomous adaptation of the mission plan and autonomous
adaptation of the trajectory plan respectively.
Both require access to the correspondent levels of perception
in order to build their own situation awareness.
In current implementations, the human operator constitutes
the decision making phase. When high-bandwidth
communication links exist, the operator remains in the loop
during the mission execution. Examples of this approach are
existing ROVs. However, when the communication is poor,
unreliable or not allowed, the operator tries, based only on
Figure 2: Typical ROV inspection operation of a riser with a
fluorometer sensor.
Figure 3: AUV recovery after finishing a mission (courtesy
of SeeByte Ltd.).
the initial orientation or expertise, to include all possible behaviours
to cope with execution alternatives. This has unpredictable
consequences, in which unexpected situations can
cause the mission to abort and might even cause the loss of
the vehicle. Examples of this architecture are current implementations

for AUVs and gliders.

Robotic platforms

Unmanned underwater vehicles can be classified in Remote
Operated Underwater Vehicles (ROVs) (see Fig. 2), Autonomous
Underwater Vehicles (AUVs) (see Fig. 3) and
Underwater Gliders. They differ on the power capability,
power endurance and the task complexity that they have
been designed for.
Underwater vehicles have become a standard tool for data
gathering for Maritime applications. In these environments,
mission effectiveness directly depends on vehicle’s operability.
Operability underlies the vehicle’s final availability,
affordability and acceptance. Two main vehicle characteristics
can improve the vehicle’s operability: reliability relates
to vehicle failures due to the internal hardware components
of the vehicle, and survivability relates to vehicle failures
due to external factors or damages.
Each of these characteristics can be improved by providing
autonomous adaptation of the mission plan and autonomous
adaptation of the trajectory plan respectively.
Both require access to the correspondent levels of perception
in order to build their own situation awareness.
In current implementations, the human operator constitutes
the decision making phase. When high-bandwidth
communication links exist, the operator remains in the loop
during the mission execution. Examples of this approach are
existing ROVs. However, when the communication is poor,
unreliable or not allowed, the operator tries, based only on
Figure 2: Typical ROV inspection operation of a riser with a
fluorometer sensor.
Figure 3: AUV recovery after finishing a mission (courtesy
of SeeByte Ltd.).
the initial orientation or expertise, to include all possible behaviours
to cope with execution alternatives. This has unpredictable
consequences, in which unexpected situations can
cause the mission to abort and might even cause the loss of
the vehicle. Examples of this architecture are current implementations
for AUVs and gliders.

The underwater environment: a challenge for planning

Abstract

This paper reviews the applications and challenges of robotic

systems in the underwater domain. It focuses on the challenges

for achieving embedded situation awareness, adaptive

trajectory planning and adaptive mission planning. These are

required elements for providing true autonomy for delegation

of tasks to unmanned underwater vehicles. The paper analyses

current approaches to tackling these challenges and how

planning plays a vital role in overcoming them. It includes a

description of some key applications and future concepts of

operations.

Introduction

In the last few decades, increasing interest in oceans has resulted

in unprecedented attention being focussed on them.

Although they cover 71% of the Earth’s surface, humankind

has sent more astronauts to the Moon than scientists to the

deepest parts of our seas. It was almost 10 years after reaching

the surface of the Moon, that the deepest parts of the

oceans were finally reached. Since then, goverments and

industry have become more and more interested in understanding

and managing our planet and they have realised

how important the underwater regions, two thirds of the total

Earth’s surface, are. Nowadays, it is not only the need

to discover, but also to observe, map and protect our oceans

that motivates further exploration of underwater regions.

Unfortunately, access to these regions is not straightforward.

The underwater environment is a hostile environment

for humans and human technology. It can challenge some of

the capabilities that are now taken for granted in other domains

such as the Earth’s surface, the atmosphere or outer

space. Some of the most representative and specific challenges

underwater are high pressure, corrosion and signal

processing issues related to data transmission and sensing.

Even though the underwater domain presents such challenges,

several maritime disciplines still require access to

this environment. The most relevant ones are:

Oceanography: Scientists are faced with the need of

gaining access to the most remote parts of the oceans,

The study reported in this paper is partly funded by the

Project RT/COM/5/059 from the Competition of Ideas and by

the Project SEAS-DTC-AA-012 from the Systems Engineering

for Autonomous Systems Defence Technology Centre, both established

by the UK Ministry of Defence.

from deep trenches to fresh water lakes under the polar ice

caps. They have to collect information in order to be able

to understand issues such as climate change, the melting

of the polar ice caps and to forecast weather conditions,

hurricanes and tsunamis.

Energy industry: In current offshore oil fields the tasks

of Inspection, Repair and Maintenance (IRM) comprise

up to 90% of the related field activity. This inspection is

dictated by the vessels availability and the weather conditions.

Additionally, the deep sea is still un-exploited.

Gaining access to these deeper levels can provide access

to new sources of minerals and energy.

Military: A priority to current Navy operations is to

maintain clear access to ship passages and to protect vessels,

harbours and coastal waters. Achieving these capabilities

without compromising personnel safety due to foe

actions is still unsolved.

For all these disciplines, robotic platforms are proven to

be very useful in de-risking human activity in the hostile

underwater environment. The main challenges that robotic

systems have to deal with underwater are:

Power : Robots are highly dependant on their battery life

in a domain without possibility of extending it from other

external sources.

Communication : Sound is the media use for sensing and

communicating underwater. Low bandwidth, long delays

and high-power requirements impose many restrictions.

Perception : Visual methods are poor while acoustic

methods come with many false positives. They are affected

by temperature, pressure and salinity making them

very noisy. Range is inversely related to the frequency

and normally quite reduced (see Fig. 1).

 Additionally, raw

sensor data has to be ultimately processed into conceptual

knowledge in order to build the awareness of the environment.

Navigation : The underwater environment is a GPSdenied

area. Existing underwater maps are still quite innacurate.

Together, these make localization and orienteering

for navigation a very hard problem.

Delegation: Autonomous adaptation of different tasks

to changes in the environment has not yet been fully

achieved. Without it, it is necessary that the operator remains

in the loop, observing and taking decisions. Autonomous

adaptation to sensed changes is necessary to

gain the operator’s trust and acceptance and for them to

delegate tasks to the robotic platform (Johnson, Patr´on,

and Lane 2007).

Although new solutions are already being developed to

extend power autonomy and communication requirements,

the other three issues (perception, navigation and delegation)

are still a real challenge for achieving true autonomy

in robotics in the underwater environment. This paper describes

each of these challenges and provides an overview on

how reasoning tools and autonomous planning approaches

can contribute to overcoming them.

Robotic platforms