CORE Metadata, citation and similar papers at core.ac.uk

Provided by Sapientia

P A P E R

Widely Scalable Mobile Underwater SonarTechnology: An Overview of the H2020WiMUST Project

A B S T R A C TThe Widely scalable Mobile Underwater Sonar Technology (WiMUST) project is

an H2020 Research and Innovation Action funded by the European Commission.The project aims at developing a system of cooperative autonomous underwatervehicles (AUVs) for geotechnical surveying and geophysical exploration. Thepaper describes the main objectives of the project, gives an overview of the meth-odologies adopted to achieve them, and summarizes the work done in the first yearof R&D work.Keywords: marine systems, cooperative AUV control, geophysical exploration,geotechnical surveying

A U T H O R SPedro AbreuLaboratory of Roboticsand Engineering Systems(LARSyS), ISR/IST,University of Lisbon, Portugal

Gianluca AntonelliFilippo ArrichielloUniversity of Cassino e del LazioMeridionale–ISME Node

Andrea CaffazGraal Tech S.r.l., Genova, Italy

Andrea CaitiUniversity of Pisa–ISME Node

Giuseppe CasalinoUniversity of Genova–ISME Node

Nicola Catenacci VolpiUniversity of Hertfordshire

Ivan Bielic de JongGeo Marine Survey Systems,Rotterdam, The Netherlands

Daniela De PalmaUniversity of Salento–ISME Node

Henrique DuarteGeoSurveys-Consultores emGeofísica, Lda., Aveiro Portugal

Joao Pedro GomesLaboratory of Roboticsand Engineering Systems(LARSyS), ISR/IST,University of Lisbon

Jonathan GrimsdaleCGG, Massy, France

Giovanni IndiveriUniversity of Salento–ISME Node

Sergio JesusSiPLAB, FCT–University of Algarve

Konstantin KebkalEvoLogics, Berlin, Germany

Elbert KelholtGeo Marine Survey Systems,Rotterdam, The Netherlands

July/A

Antonio PascoalLaboratory of Roboticsand Engineering Systems(LARSyS), ISR/IST,University of Lisbon

Daniel PolaniUniversity of Hertfordshire

Lorenzo PolliniUniversity of Pisa–ISME Node

Enrico SimettiUniversity of Genova–ISME Node

Alessio TurettaGraal Tech S.r.l., Genova, Italy

Introduction

WiMUST (Widely scalable MobileUnderwater Sonar Technology) is anongoing project funded by the Euro-pean Community within the H2020framework (Work Programme 2014–2015, LEIT-ICT, 5. Leadership inenabling and industrial technologies–

Information and CommunicationTechnologies). The ultimate goal ofthe project is to design and test a sys-tem of cooperating autonomousunderwater vehicles (AUVs) able toperform innovative geotechnical sur-veying operations. In particular, theWiMUST system will be composed

by a small fleet of AUVs towingstreamers with hydrophones to ac-quire sub-bottom profiling acousticdata. In contrast with classical tech-nology based on ship-towed streamers(Figure 1), the WiMUST solutionwill allow a change in the geometryof the acoustic antenna, something

ugust 2016 Volume 50 Number 4 1

that has not been achieved in practiceand holds potential to significantlyimprove ocean surveying.

This paper gives a brief overview ofthe project (Al-Khatib et al., 2015a;Indiveri &Gomes, 2014). In particular,this paper is an updated and extendedversion of Al-Khatib et al. (2015b).Concept and Approach addressesthe main concepts and technical ap-proaches. Distributed Sensor Arrayfocuses on the distributed sensorarray (DSA) issues to be faced in theproject. Cooperative Control describesthe AUV cooperative control activities,whereas Mission Planning gives anoverview of the mission planning is-sues. Communications addresses thecommunication activities within theproject, and Integration and Experi-mentation gives a brief account forthe necessary system integration and ex-perimentation activities. Finally, Con-clusions contains concluding remarks.

Concept and ApproachThe WiMUST project (Al-Khatib

et al., 2015b) aims at conceiving,designing, and engineering an intelli-gent, manageable, distributed, and re-

2 Marine Technology Society Journal

configurable underwater acoustic arraythat could significantly improve the ef-ficacy of the methodologies used toperform geophysical and geotechnicalacoustic surveys at sea (see Figure 1).The use of the WiMUST system willbe beneficial in a vast number of appli-cations in the fields of civil engineeringand the oil and gas industry, where sea-bed mapping, seafloor characterization,and seismic exploration are fundamen-tal operations. The novel key feature oftheWiMUST system consists in the useof a team of cooperative autonomousmarine robots, acting as intelligent sens-

ing and communicating nodes of a re-configurable moving acoustic network.The vehicles are equipped with hydro-phone streamers of small aperture,such that the overall system behaves asa distributed acoustic array capable ofacquiring acoustic data obtained by illu-minating the seabed and the ocean sub-bottom with strong acoustic waves sentby one (or more) acoustic sources in-stalled onboard a support ship/boat(Figure 2). By actively controlling thegeometry of the robot formation, it be-comes possible to change the shape ofthe acoustic array according to theneeds of a specific application. Theresulting operational flexibility holdspotential advantages, as it allows im-proving the seabed and sub-bottomresolution and obtaining sidelobe rejec-tion at almost any frequency and forany plane. The availability of the pro-posed system, other than improvingthe quality of the acquired data, willalso greatly facilitate the operations atsea because of the lack of a physicalconnection between a surface vesseland the data acquisition equipment.

As a preliminary step for the re-search work planned in the project, theactivities have started with an analysisof the reference scenario, which needs

FIGURE 1

Ship-Geophysical survey using hydrophone streamers towed by a support ship carrying an acous-tic source (photo courtesy of CGG).

FIGURE 2

Artist’s rendition of the WiMUST system for subbottom profiling with source-receiver decoupling.

to be characterized in terms of desiredfunctionalities and expected behaviorof the WiMUST system. In particular,the reference scenarios cover both 2Dand 3D active geophysical explorationand geotechnical seafloor characteriza-tion scenarios. As a result, specificationsfor all the WiMUST subsystems (mis-sion planning, communications,distributed sensing, navigation, andcoordination) were prepared in detail.

Distributed Sensor ArrayOne of the most groundbreaking

proposals in WiMUST is that theAUV-based acoustic sensing array isphysically disconnected from thesource system, which allows for a newfreedom in sensing array geometry. Ifwe abstract from the physical con-straints relative to vehicle navigation,speed, positioning, and other limita-tions, the WiMUST sensing systemmay be viewed as a DSA free to adoptany geometry. Moreover, this geo-metry may change along time, uponrequest, based on the analysis of theseismic data collected online. Thus,in terms of data processing, the funda-mental question to be tackled is whetherthe seismic data acquisition and pro-cessing protocol established for ship-towed streamers is also suited for themobile AUV-based acoustic sensingsystem ? And, if not, in which situationand how should it be modified?

Determining the geophysical prop-erties of the ocean bottom is an old andnontrivial problem with scientific aswell as engineering implications. Evenif electromagnetic approaches exist(Ziolkowski & Wright, 2012), thelarge majority of the survey methodsare based on the analysis of bottomacoustic returns complemented within situ observations and historical in-formation. The problem is complicated

by the fact that sound velocity variesstrongly in consolidated sediments,which makes it difficult to determineat the same time bottom layering (i.e.,layer thickness), based on the time-of-flight of acoustic arrivals, and compres-sional velocities on each layer wherethose arrivals propagate. This makesthe problem nonlinear. There are twomain classes of methods to address thisproblem: one is based on the matchingof the observed data with the acousticfull-field calculated for a physicalmodel of the bottom, including layer-ing and layer physical properties. Thisis the so-called matched-field inversion(MFI) class of methods. The other isbased on transforming the nonlinearproblem into two linear steps: in thefirst step, the time-of-flight of identifiedbottom returns are matched assuminga given velocity model (based on apriori knowledge); in the second step,arrival times are successively matchedwith those obtained from the dataand, as later arrivals are considered,the grazing angle changes and thefocal point moves deeper into the bot-tom, which requires successive adjust-ment of the velocity model until a goodmatch is obtained. This is the approachused in classic time-of-arrival (ToA)seismic surveying (Yilmaz, 1987). Forvarious reasons, MFI is more oftenused with vertical arrays and for highresolution of surficial sediments inshallow water, whereas ToA is usedwith horizontal (towed) arrays, bothin shallow and deep water, for low-resolution bottom imaging of largehigh-intensity reflectors deeply buriedinto the bottom. The latter is the typicalproblem faced in oil and gas explorationwhere the objective is to detect a saltcrust, sometimes one or more kilo-meters below the ocean bottom.

In order to motivate the approachtaken in this work, we start by consid-

July/A

ering a highly conceptual view wherewe attempt to “explain” the data ac-quired d with a model m(p0) that de-pends on a parameter vector p0. Thesought performance criterion (?) are ob-viously dependent on how close, in theleast squares sense, parameter vector es-timate p is of true vector p0. Therefore,the problem may be stated as follows

min jjp� p0jj2 s:t: d ¼ m(p) ð1Þ

which cannot be solved for p since p0 isunknown. One way to circumvent thisis to separate p into two parts: the layer-ing x and the remaining parameters θ,namely the sound velocities in the con-sidered layers. With this assumption,modelmmay be written under a linearform as m = A(θ)x. Although un-known, the number of effective layersin x is much smaller than the actualmeasurements in m (or d), thanmodel m = A(θ)x may be viewed as asparse system of equations and theproblem in Eq. (1 rewritten as

min jjxjj1 s:t: d ¼ A θð Þx ð2Þ

where now theminimization is over thenonzero components of sparse vector x.If there is noise in the observations d,now taken as y = d + n, where n isthe noise, (Eq. (2) reduces to

min jjxjj1 s:t: jjy�A θð Þxjj2 < δ

ð3Þ

For θ known (or estimated) and A fullrank, even if the number of observationsis much smaller than the dimensionof vector x, Equation (2 is a problemthat can be solved by linear program-ming using an efficient basis-pursuitalgorithm subject to fitting constraints.The solution depends on the structureof matrix A that has less rows than

ugust 2016 Volume 50 Number 4 3

columns and therefore should guaran-tee, with a high probability, that thenonzero entries of x are observable ind (or y). This is the problem faced incompressed sensing where it wasshown that the condition is that Ashould have a low coherence (Candes& Romberg, 2007). Low coherencemeans that A should be as “flat” as pos-sible. The practical problem in our caseis that matrixA is formed by the acous-tic propagation model that connectsthe source, the bottom layers, and thereceivers, i.e., the Green function ofthe media (constrained in parameterθ) and which cannot be changed. Inpractice, the coherence of matrix Amay be lowered by selecting observa-tions (the rows of A) at random posi-tions (Fannjiang et al., 2009; Jesus,2016). Low coherence means high di-versity, which is simply equivalent toplacing the sensors of our DSA at posi-tions where the bottom received acous-

4 Marine Technology Society Journal

tic field is the most diverse or where itshows the lowest coherence. In otherwords, our conviction is that the criteriafor determining the optimal DSAgeometry should follow the line oflow coherence or high diversity within,of course, the operational and practicalconstraints of the WiMUST AUV-based platform.

Figure 3 illustrates the results ob-tained in a 2D example where a syn-thetic environment composed of a140-m-deep water layer, two 10- and20-m-thick sediment layers over asub-bottom half space with variablecompressional speed from 1,550 upto 2,500 m/s, excited by a 500-Hzsound source at 5-m depth. The re-ceiver is a linear 100 sensor arraywith 2,500-m aperture also at 5-mdepth. A bottom grid of 50 × 10 sam-ples was defined for a bottom coverageof 60-m depth × 2,500-m range. Anumber of up to K = 30 bottom scat-

ters were simulated in a peak shapeddeeper sediment layers. The acousticpropagation model OASES (Schmidt,2011) was used to generate the arrayreceived field where an SNR of 10 dBwas used. The 100 receivers were ran-domly distributed, and a coherence of1.3 was obtained for the exampleshown (note that the used normaliza-tion implies that the coherence variesbetween 1 and square root of the num-ber of search sample, in this case 22).Three optimization algorithms fromthe YALL package (Zhang et al.,2010) were tested, and the resultsare shown in Figure 3 (left) for threedifferent algorithms compared withthe true solution (blue curve): the l1norm showing a large number ofspurious peaks (red), the l1 − l2 normunconstrained where still some lowamplitude estimates at wrong depthscan be seen (green), and the l1 − l2norm constrained, corresponding to

FIGURE 3

2D seismic results on synthetic example with two range-variable thickness bottom layers using the YALL (Zhang et al., 2010) package for the l1norm (red), for the l1 norm l2 unconstrained (green), and for the l1 norm l2 constrained (black) algorithms and true values (blue) (left) and bottomreconstruction with the l1 norm l2 constrained (right). Water depth is 140 m, first layer within 150 m and second layer within 170m at a range of 0 km.(Color versions of figures are available online at http://www.ingentaconnect.com/content/mts/mtsj/2016/00000050/00000004.)

(Eq. (3), with the lowest mean squareerror (black). At each range, the threedifferent amplitudes denote the reflec-tor strength decrease with depth intothe bottom. The results obtainedwith the l1 − l2 norm constrainedwere used to reconstruct the 2D bottomestimate along range and depth, givinga visual perception of the bottom layer-ing structure (Figure 3, right).

The above methodology requiresfurther testing for variable sensing arraygeometries, broadband signals, and ro-bustness testing to various environmen-tal parameter θ mismatch. A simulatorfor performance prediction in realisticdata will normally follow with applica-tion to real data when available.

Cooperative ControlCooperative navigation, guid-

ance, and control of an AUV roboticvehicle team is one of the pillars of theWiMUST system. The WiMUSTvehicles will need accurate relative nav-igation and control capabilities, ensur-ing intervehicle collision avoidance andyielding a virtual structure to accuratelyposition each hydrophone streamer,thereby shaping the formation of theresulting sonar receiving array.

Acquiring seismic data with well-defined characteristics requires theacoustic source and each of the acous-tic receivers to maintain a specific geo-metric formation. In the scope ofWiMUST, this translates into a clearcontrol task: steer each vehicle so thatall vehicles, the streamers being towed,and the acoustic source (usually asparker) maintain a specified forma-tion geometry. We make the simplify-ing yet realistic assumption that thetrajectory followed by the source (usu-ally towed from a support vessel) isknown with an adequate degree of ac-curacy, usually obtained from a DGPSor RTK GPS installed on the source.

At this point in the project, twotypes of candidate solutions arise forthis control problem, each with its ad-vantages and disadvantages. We nowgive a brief overview of each of thesesolutions, as they currently stand andpropose modifications and applicationscenarios in the WiMUST setting.

Cooperative Path FollowingThe cooperative path following

(CPF) control algorithm assumes thateach vehicle has full knowledge of thepath to be followed in a close future,with no associated explicit timinglaw, and that the paths of all vehiclesare parameterized with an along-pathcoordinate γ. The goal of the controlalgorithm is that all vehicles follow thepath in a synchronized fashion, i.e., thevehicles are at “corresponding” points(points with the same γ coordinate)on their own paths at each instant oftime. Therefore, the parameterizationof the paths must take into accountthe desired formation geometry. Theabsence of an associated timing law im-plies that the algorithm does not im-pose where each individual vehicleshould be at each instant in time—just

July/A

that all vehicles should be at corre-sponding points on their paths. How-ever, in CPF, there is room to specifydesired speeds as functions of the pa-rameter γ and not of time explicitly;this is called speed assignment.

Themain idea behind the algorithm,as described formally in Ghabchelooet al. (2009), where a proof of conver-gence is given, is that each vehicle ikeeps track of its own “virtual target”(a point on the path, with an associatedalong-path coordinate γi). Althougheach vehicle attempts to steer towardits target, the targets themselves are ad-justed (slowly) based on the targets ofthe other vehicles in the formation,so that

(i) they converge to the same valueof γ on their respective paths;(ii) their speeds dγ/dt along the re-spective paths converge to desiredvalues, as determined by the specificmission being executed (Figure 4).In the scope of WiMUST, one of

the elements in the formation to bemaintained is the acoustic source. Asthe source is towed by a human-pilotedvessel, this element should be consid-ered not-cooperative in the sense that

FIGURE 4

A schematic drawing of a formation running the CPF algorithm andmaintaining a triangular geometry.

ugust 2016 Volume 50 Number 4 5

its “virtual target” cannot be adjustedin real time in response to the motionof the vehicles. As such, in order to usea CPF-like algorithm, a modificationmust be introduced. A possible ideais to consider that the formation in-cludes all the vehicles, but not theacoustic source, and replace require-ment (ii) above with a prescribedvalue on position (along-path coordi-nate) as a function of time, renderingthe algorithm closer to a tracking ap-proach, as described later. This willbe subject of further investigation, be-cause it is not known whether intro-ducing this requirement compromisesthe desirable robustness and stabilitycharacteristics of the CPF algorithm.

The main advantage of this algo-rithm lies in its robustness to failureof a single element. The particularlysimple case of a formation consistingof several vehicles moving along a com-mon straight line is an example of spe-cial importance. In this case, somekind of intervehicle cooperationwould ideally exist, so that if one ofthe vehicles in front of the formationfails and stops, the remaining vehiclesbehind it also eventually slow downand stop, regardless of the motion ofthe support vessel towing the source.

In contrast, the main disadvantagesare that

- a priori information about thepaths to be followed is needed;- the algorithm requires the vehiclesto exchange information (theiralong-path coordinates γi) at a con-siderably high rate, so that the γ co-ordinates can be updated regularlyto achieve synchronization.

Moreover, communication strategiesinvolving frequent information ex-change among all vehicles cannot beused in an underwater environmentwithout compromising scalability—the frequency of this information ex-

6 Marine Technology Society Journal

change would decrease linearly as thenumber of vehicles increases. Therefore,a solution purely based on CPF-likealgorithms can be used only whenall the vehicles move at the surface,so that this information exchange canoccur via WiFi instead of acousticnetworks.

Target/Trajectory TrackingThe target/trajectory tracking (TT)

family of algorithms, described and an-alyzed formally in Abreu & Pascoal(2015) and then applied to specific sce-narios in Abreu et al. (2015a, 2015b),resorts to a fundamentally different ap-proach. The goal is for a single vehicleto follow a trajectory with an associatedtiming law, i.e., enforcing where thevehicle should be at specific times.The idea behind this is that, again, avehicle tries to steer toward its “virtualtarget,” but this target moves solelybased on a specific timing law. Thisimplies that, for each time instant t,there is a prescribed desired positionfor the vehicle, pd(t), and optionallyother references such as desired head-ing and depth. The algorithm canthen be naturally extended (Abreuet al., 2015b) to achieve geometric for-mations in multiple-vehicle missionsby choosing the trajectory (includingthe spatial path and its associated tim-ing law) for each vehicle appropriatelyfor the desired geometry. Then, whenall vehicles eventually converge to theirown targets, the formation moves in asynchronized fashion, and the desiredgeometry is maintained.

In practice, a major drawback ofthis algorithm when compared topath-following is its lack of robustness,in the sense that if some unplannedevent happens (e.g., one of the vehiclescannot follow its target) the desiredgeometry can no longer bemaintained.However, some characteristics of this

algorithm seem to fit the particularWiMUST case, namely:

- no a priori knowledge of the tra-jectory is needed—it suffices thatat each time instant each vehiclecan compute the position and veloc-ity vector of its virtual target, basedon information about the motion ofthe acoustic source, for example;- the “time-dependent” nature ofthe algorithm suits the control ob-jective: the desired position ofeach vehicle, at any specific time in-stant, depends on the position ofthe acoustic source at that time. Inother words, each vehicle should beat the right place at the right time(there is no point in following thesame path as the acoustic source ifnot with the correct timing);- no high-frequency informationexchange between the vehicles isneeded, as each vehicle can com-pute the position and velocity itsvirtual target based solely on infor-mation about its own position andvelocity and those of the acousticsource. This allows TT-like algo-rithms to be applied even in an un-derwater setting where the vehiclescan communicate only via theacoustic network and, as stated be-fore, high-frequency informationexchange between all vehicles isnot possible for scalability reasons.

Hybrid ApproachesHaving stated the main characteris-

tics of these two types of geometric for-mation control algorithms, it seemsthat some hybrid approach would bedesirable in the context of WiMUST.Ideally, the algorithm should meet thehigh-level control objective—to keepthe vehicles in a desired geometrywith respect to the acoustic source—while borrowing some of the robust-ness characteristics of the CPF and,

for the underwater case, avoiding theneed for frequent intervehicle commu-nication. The application of possiblehybrid approaches will certainly bea topic of further study within the pro-ject. An example of an alternative to beinvestigated is the case where the fu-ture trajectory of the acoustic sourceis predicted on a time window, basedon past and present position and veloc-ity vectors, and a CPF-like algorithmis run using this predicted path.

Single-Range Navigationand Guidance

During the initial phase of eachmission, when the vehiclesmove towarda given waypoint in order to reach thedesired formation, a single-range aidednavigation could be used to improvethe localization performance. Indeedsingle range aided navigation tech-niques are believed to have a potential-ly significant impact on the WiMUSTmotion control architecture as AUVscommunicating with a given nodewill be able to estimate their range toit based on time of arrival measure-ments. This information can beexploited to improve a relative localiza-tion estimate to be used in the motioncontrol schema. For the time beingwithin the project several single-range-related navigation approacheshave been explored: a novel strategyhas been proposed in De Palma et al.(2015b) that addresses the problemof driving an underwater vehicle to adesired position while optimizing theperformance of a single-range-basedlocalization system. In fact, the perfor-mance of single-range-based localiza-tion techniques is strongly dependenton the vehicle’s motion that needs tobe sufficiently rich to yield a goodmea-sure of observability. A prioritizedtask-oriented technique to combine apoint-to-point guidance task with an

observability optimization task can beused. In particular, the minimum sin-gular value of the Fisher InformationMatrix is adopted as the observabilitymetric. Task prioritization is achievedthrough a null-space-based projectiontechnique that is also combined with asaturation management algorithm tocope with velocity command satura-tions. The stability of the resultingcontrol law is analyzed following aLyapunov-based approach. Robustnessagainst uncertainty in the initial posi-tion of the vehicle can be also handledresorting to a procedure that maximisesthe minimum (worst-case) observabil-ity metric. Numerical simulationsreported in (De Palma et al. (2015b))confirm the effectiveness of the pro-posed strategy. Notice also that theuse of a prioritized task-oriented tech-nique can be exploited to include otherimportant tasks in the framework asthe collision avoidance. Indeed, suchtask is relevant for the WiMUST sys-tem, because it is composed by a fleetof vehicles.

As for relative AUV navigation, rel-ative range information can be embed-ded in the navigation system of eachindividual vehicle in the form of a geo-metric constraint, depending on therange measurements and on the com-munication topology. Preliminarypromising results exploiting thisapproach have been reported byDe Palma et al. (2015a).

Mission PlanningThe execution of WiMUST mis-

sions will require the availability offlexible mission planning algorithmsfor the deployed robotic units. An in-novative approach toward missionplanning is developed, which extendsclassical algorithms for mission plan-ning using novel intelligent methodol-

July/A

ogies. The proposed approach is basedon the information-theoretic quantityempowerment (Klyubin et al., 2005,2008), which is associated with thestate of a robot and can be exploitedto generate preferred behaviors with-out having to resort to specializedcost functions, which usually varyfrom task to task and have to behand-designed. Technically speaking,empowerment is defined as the Shannonchannel capacity between the robotcontrols and its states in subsequenttime steps; therefore, it is measured inbits. Its computation is based on adap-tive evaluation of the acting-sensing in-teraction of the robot group with itsenvironment. Within the WiMUSTproject, empowerment is used for thefirst time in the context of marine ro-botics. In this context, its definitionmakes explicit use of marine systemsdynamicmodeling, taking into accountsingle-vehicle hydrodynamic effects(added masses, viscous frictions, re-storing torques and forces, currents,etc.). Empowermentwas already provento identify desirable states for agents,in term of intrinsic survivability androbustness drives, in a wide spec-trum of scenarios (Jung et al., 2011).In fact, it can be considered as aninformation-theoretic generalizationof the concept of mobility (in principle,a quantitative metrics of the distinctcontrollably accessible states). Essen-tially it measures to what extent anagent can influence the environmentby its controls: it is zero if, regardlesswhat the robot does, the outcome ofthe controls will be the same; it is max-imal if every control will have a distinctoutcome. InWiMUST, we investigatehow empowerment can be used as avehicle-intrinsic metric to evaluateAUVs’ desirable trajectories in termsof mobility and survivability. Hence,while performing team formation

ugust 2016 Volume 50 Number 4 7

control, the overall WiMUST missionplanner may benefit by the possibilityof monitoring online the empower-ment of the state of each vehicle.Because empowerment is zero in situ-ations where the vehicles have no mo-bility at all, one rationale for its use isto detect the robots being in or ap-proaching undesirable conditions asbeing stuck or crashed. In addition,empowerment can play the role ofa purely intrinsic utility function thatserves to enhance the control of thevehicles during a mission; we utilizeits property as a powerful heuristic toimprove situation-adapted and flexiblereaction in safe-critical scenarios. Thiscan be particularly relevant in emer-gency situations when the vehicle hasno access to an explicitly specified ex-ternal cost function to minimize (e.g.,because it lost communication withthe base ship). To this end, we developan efficient empowerment-based con-troller for vehicles that will eventuallyfollow their own empowerment gradi-ent (i.e., they will choose controls thatlead to successor states with the highestempowerment). The key innovation ofthe empowerment-based approach isto move away from inflexible hand-designed algorithms to a flexible ap-proach that has, in many examplescenarios, demonstrated the ability to in-corporate situation-specific response ca-pability. To this end, we intend to showthat the utilization of empowerment-based mission control can be used tomake the vehicles improve their ownsurvivability autonomously, withouthaving to explicitly specify this require-ment within the mission definition.We hypothesize that this approachwill endow the WiMUST missionplanner the “graceful degradation”property, so as to prevent the basic func-tionalities of the vehicles to breakdown entirely when subcomponents

8 Marine Technology Society Journal

of the WiMUST control system maybe inoperative.

CommunicationsThe ma in ob j ec t i v e s o f the

communication-related research withinWiMUST are the development ofalgorithms and procedures for accuratedistance measurements (up to the cen-timeter scale) between AUVs movingin formation, reconstruction of theirrelative positions in 3D space, and pre-cise synchronization of their localclocks. These are required to assigntemporal and spatial tags to acquireddata that are consistent across the for-mation of vehicles and to support somemodalities of navigation that requireexplicit relative positions to controlthe shape of the formation of vehicles.Moreover, it will be necessary toachieve an estimation of practicaldata throughput boundaries for differ-ent AUV formations. Based on suchestimates, specific work will addressthe development of communicationalgorithms and procedures to supportthe different operative and environ-mental conditions. The work in thisarea will be organized along two majorlines: long-range and short-range com-munication issues.

As for the short-range communica-tion requirements, high bit rate com-munication of AUVs at short range isa prerequisite for the design of cooper-ative teams of underwater vehicles forthe operations envisioned in theWiMUST applications. Typical bitrates by conventional acoustic modemsare rather low, usually reaching, inpractice, effective bit rates on theorder of several hundred bits per sec-ond having nominal bit rates of severalkilobits per second. Recent modemdevelopments made by EvoLogicsprovided a significant increase of prac-

tically achievable bit rates. The objec-tive is to focus on the evaluation ofpractically achievable bit rates withinthe necessary team geometries of practi-cal interest, which will enable successfulimplementation of the navigation andmotion control tasks within the project.

Accounting for specific AUV for-mations and typical environmentalconditions, the achievable bit rateswill be firstly obtained by modelingand simulation and then comparedwith the results of practical experi-ments. In this manner, the basic esti-mates of currently achievable bit ratescan be accounted for the needs of theproject. To enable a reliable communi-cation necessary to adequately controlthe AUVs formation, a robust datatransmission with the use of advancednetworking protocols will be imple-mented and tested. Moreover, specificwork will aim at developing a decentra-lized clock synchronization algorithmbased on the data exchange usingshort range link. The algorithm willmake it possible to keep mutual clockdrifts much smaller than for traditionalcomputer clocks. At the moment,synchronization experiments havedemonstrated the feasibility of decen-tralized clock synchronization viahydroacoustic channel for stationarynodes (Kebkal et al., 2015). Synchro-nization algorithms are currently in de-velopment for mobile nodes.

The overall short-range commu-nication system will make it possibleto extend the existing point-to-pointalgorithms of distance measurementin the WiMUST team of AUVs.The challenge here is to provide suffi-ciently accurate distance measure-ment in the distributed network ofmoving vehicles.

While improving the data ratesand reducing the packet overhead inpoint-to-point communications is of

paramount importance, an equally rel-evant issue is to design decentralizedlocalization and clock synchronizationalgorithms for efficient use of the trans-mission medium (see, e.g., Furfaro &Alves, 2014; Ludovico et al., 2014;Vermeij & Munafo, 2015). Thisneed arises due to the potentiallylarge number of vehicles that mustshare the acoustic channel, whichmay quickly lead to a saturation of re-sources in the network as a whole evenif individual packets are transmittedas fast as the channel allows. It is there-fore necessary to develop streamlinedversions of distributed estimationalgorithms that take advantage of thebroadcast nature of the medium toreduce the amount of data that eachnode must share with others topreserve valuable channel resources(in addition to all optimizations thatmay be undertaken at the level of net-working protocols), without undulysacrificing estimation accuracy. Al-though active research is being per-formed within the project relative tothe clock synchronization issue, it isplanned that the WiMUST AUVswill be equipped with atomic clocksenabling the WiMUST system toproperly function for the duration ofa typical mission in spite of possiblepoor clock synchronization algorithmperformance.

As for the long-range communica-tion requirements of the overall sys-tem, the exploitation of advantagesgiven by sweep-spread communicationtechnology of EvoLogics opens theway to practically achievable bit ratesof about a several to tens kilobit/s inlong range. While the range of datatransmission and the capacity of theunderwater acoustic channel maychange vigorously in consequence ofchanges in environmental conditions,an efficient implementation of an

acoustic communication network willstrongly depend on the capability ofthe modem to automatically adjust itsbit rate to the actual channel condi-tions. Because parallel/asynchronoustransmission of many differently prior-itized data streams from one or manydata sources represents one of standardfeatures of EvoLogics modems, thecommunication bit rate is alwaysadapting to the WiMUST operatingconditions.

Integration andExperimentation

As described in Al-Khatib et al.(2015b), specific activities for systemintegration are planned within the pro-ject: indeed, all the subsystems andfunctionalities developed within theproject must be integrated to formthe final WiMUST system. More-over, the overall system will be vali-dated through experimental testsat sea. The vehicles constituting the

July/A

WiMUST system will need to performcooperative guidance, navigation (local-ization), and control by implementingthe solutions and methods derived inthe course of the project. In the finalexperiment, the WiMUST vehicleswill be required to exhibit a sufficientdegree of autonomy and intelligencein controlling the required formationwhile concurrently performing opera-tional tasks related to the intravehi-cle distance and quality of service ofvehicle-to-vehicle communication.The final experimental tests will thusserve as a validation for all projectgoals. Such tests are planned to beexecuted either in the Setubal area(Portugal) or close to the Elba islandin Italy (Figures 5 and 6).

With reference to the system integra-tion activities, these will initially dealwith mechatronic integration of thesensing payloads and the communica-tion devices on board the AUVs. Succes-sively, the software modules concerninggroup navigation and coordination

FIGURE 5

Setubal (Portugal) area: one of the candidate sites for the final WiMUST tests.

ugust 2016 Volume 50 Number 4 9

will be finalized and inserted in thefinal AUVs’ software architecturetogether with the distributed sonaralgorithms and the communicationprotocols and strategies.

Once the basic functionalities willbe validated in the lab through “wet”engineering experiments, sea trials ofthe overall WiMUST system will beexecuted. Data gathered during exper-iments will be finally analyzed for

10 Marine Technology Society Journa

obtaining indications on the real per-formance of the WiMUST system.

Preliminary tests of the MEDUSAclass AUVs from the WiMSUT part-ner Instituto Superior Técnico (IST,Lisbon, Portugal) towing streamers(from partner GEOMARINE, The

l

Netherlands) were performed in late2015/early 2016. The streamers usedfor these tests were actually “dummy”streamers, namely without any inter-nal cabling, hydrophones, and pre-amplifiers. Other than this, they useexactly the same materials and havethe same geometry that is envisionedfor the WiMUST project . Twodummy streamers were manufacturedspecifical ly for this purpose byGEOMARINE and kindly madeavailable to IST. The mass of eachstreamer was 19.0 kg. The preliminarytest experiments in Lisbon involvedmissions with single and cooperativemarine vehicles of the MEDUSA-class (shown in Figure 7). Two mis-sions were carried out in which aMEDUSA vehicle operating at the sur-face, towing a streamer, followed circu-lar trajectories of radii 12 m and 8 m,respectively. A small, human-pilotedsupport boat followed the tip of thestreamer, and a GPS antenna fixed toa bar was manually kept on top of it.This produced GPS position data forboth the MEDUSA vehicle and the

FIGURE 7

Aerial view of twoMEDUSA vehicles (IST) per-forming a cooperative path following maneu-ver while towing streamers along the curvedportions of the paths.

FIGURE 8

Position data for the MEDUSA vehicle and the tail of the streamer, obtained experimentally withGPS in RTK mode.

FIGURE 6

Elba island (Italy) area: one of the candidate sites for the final WiMUST tests.

tip of the streamer. For each of the cir-cular missions, the streamer tip was in-deed found to follow a circumferenceof a smaller radius than that of theMEDUSA, and a radii difference be-tween the two was computed for eachof the missions (12 and 8 m for theMEDUSA). Figure 8 shows the trajec-tories obtained experimentally for radiiof 12 and 8 m. The acquired data willbe instrumental to identify a dynamicmodel of the AUV-streamer systemthat is of great importance for naviga-tion, guidance, and control of the over-all system. Experiments were also donewith two MEDUSA vehicles perform-ing a cooperative path following ma-neuver while towing streamers alongthe paths (Figure 7).

ConclusionsThe v i s i on unde r l y i n g t h e

WiMUST project is that of developingadvanced cooperative and networkedcontrol/navigation systems to enablea large number (tens) of marine robotsto interact by sharing information as acoordinated team (not only in pairs).The WiMUST system may be envi-sioned as an adaptive variable geometryacoustic array. This paper has providedan overview of the WiMUST objec-tives and the methods adopted toachieve them.

Corresponding Author:Giovanni IndiveriUniversity of Salento, Departmentof Innovation EngineeringVia per Monteroni, 73100 Lecce, ItalyEmail: giovanni.indiveri@unisalento.it

ReferencesAbreu, P.C., & Pascoal, A.M. 2015. Forma-

tion Control in the scope of the MORPH

project. Part I: Theoretical Foundations. In:

4th IFAC Workshop on Navigation, Guidance

and Control of Underwater Vehicles (NGCUV

2015). Girona, Spain: IFAC-PapersOnLine,

48(2): 244-9.

Abreu, P.C., Bayat, M., Pascoal, A.M.,

Botelho, J., Góis, P., Ribeiro, J., … Silva, H.

2015a. Formation Control in the scope of

the MORPH project. Part II: Implementa-

tion and Results. In: 4th IFAC Workshop

on Navigation, Guidance and Controlof

Underwater Vehicles (NGCUV 2015).

Girona, Spain: IFAC-PapersOnLine, 48(2):

250-5.

Abreu, P.C., Bayat, M., Botelho, J., Gois, P.,

Gomes, J., Pascoal, A.M., … Silva, H. 2015b.

Cooperative formation control in the scope

of the EC MORPH project: Theory and

experiments. In: OCEANS 2015-Genova,

pp.1-7. Genoa, Italy: IEEE.

Al-Khatib, H., Antonelli, G., Caffaz, A.,

Caiti, A., Casalino, G., de Jong, I.B., …

Polani, D. 2015a. Navigation, guidance and

control of underwater vehicles within the

Widely scalable Mobile Underwater Sonar

Technology project: an overview. In: 4th IFAC

Workshop on Navigation, Guidance and

ControlofUnderwaterVehicles (NGCUV2015).

Girona, Spain: IFAC-PapersOnLine. 48(2):

189-93.

Al-Khatib, H., Antonelli, G., Caffaz, A.,

Caiti, A., Casalino, G., de Jong, I.B., …

Polani, D. 2015b. The Widely scalable Mobile

Underwater Sonar Technology (WiMUST)

project: an overview. In: OCEANS 2015 -

Genova, pp.1-7. Genoa, Italy: IEEE.

Candès, E.M., & Romberg, J. 2007. Sparsity

and inchoerence in compressive sampling.

Inverse Problem. 23(3):969-85. http://dx.doi.

org/10.1088/0266-5611/23/3/008.

De Palma, D., Indiveri, G., & Parlangeli, G.

2015a. Multi-vehicle relative localization

based on single range measurements. In:

3rd IFAC Workshop on Multivehicle

Systems (MVS 2015). Genoa, Italy: IFAC-

PapersOnLine. 48(5):17-22. http://dx.doi.

org/10.1016/j.ifacol.2015.06.457.

De Palma, D., Indiveri, G., & Pascoal,

A.M. 2015b. A Null-Space-Based Behavioral

July/Au

Approach to Single Range Underwater

Positioning. In: 10th IFAC Conference on

Manoeuvring and Control of Marine Craft

(MCMC 2015). Copenhagen, Norway:

IFAC-PapersOnLine. 48(16):55-60.

Fannjiang, A.C., Yann, P., & Strohmer, T.

2009. Compressed remote sensing of sparse

objects. arXiv:0904-3994.

Furfaro, T., & Alves, J. 2014. An application

of distributed long baseline—Node ranging

in an underwater network. In: Under-

water Communications and Networking

(UComms’14), pp.1-5. Sestri Levante, Italy:

IEEE. http://dx.doi.org/10.1109/ucomms.

2014.7017126.

Ghabcheloo, R., Aguiar, A.P., Pascoal, A.M.,

Silvestre, C., Kaminer, I., & Hespanha, J.

2009. Coordinated Path-Following in the

Presence of Communication Losses and Time

Delays. SIAM J Control Optim. 48(1):234-

65. http://dx.doi.org/10.1137/060678993.

Indiveri, G., & Gomes, J. 2014. Geophysical

surveying with marine networked mobile

robotic systems: The WiMUST project. In:

WUWNET’14 Proceedings of the Inter-

national Conference on Underwater Networks

& Systems. New York, NY: Association for

Computing Machinery. http://dx.doi.org/

10.1145/2671490.2677084.

Jesus, S.M. 2016. Distributed sensor array

for bottom inversion. In: IEEE/OES China

Ocean Acoustics’2016. Harbin, China: IEEE/

OES. Available at: http://www.siplab.fct.ualg.

pt/pubs/sjesus1.16.pdf.

Jung, T., Polani, D., & Stone, P. 2011. Em-

powerment for continuous agent-environment

systems. Adapt Behav. 19(1):16-39. http://dx.

doi.org/10.1177/1059712310392389.

Kebkal, K.G., Kebkal, V.K., Kebkal, O.G.,

& Petroccia, R. 2015. Underwater Acoustic

Modems (S2CR Series) for Synchronization

of Underwater Acoustic Network Clocks

During Payload Data Exchange. IEEE J

Oceanic Eng. 99:1-12.

Klyubin, A.S., Polani, D., & Nehaniv, C.L.

2005. Empowerment: A universal agent-

centric measure of control. In: IEEE

Congress on Evolutionary Computation

gust 2016 Volume 50 Number 4 11

(Volume:1), pp. 128-35. Edinburgh,

Scotland: IEEE. http://dx.doi.org/10.1109/

cec.2005.1554676.

Klyubin, A.S., Polani, D., & Nehaniv, C.L.

2008. Keep your options open: An information-

based driving principle for sensorimotor

systems. PloS ONE. 3(12):e4018. http://dx.

doi.org/10.1371/journal.pone.0004018.

Ludovico, V., Gomes, J., Alves, J., & Furfaro,

T. 2014. Joint localization of underwater

vehicle formations based on range difference

measurements. In: Underwater Communica-

tions and Networking (UComms’14), pp.1-5.

Sestri Levante, Italy: IEEE. http://dx.doi.

org/10.1109/ucomms.2014.7017127.

Schmidt, H. 2011. OASES 3.1–User guide

and reference manual. Cambridge, MA: MIT.

Vermeij, A., & Munafo, A. 2015. A Robust,

Opportunistic Clock Synchronization Algo-

rithm for Ad Hoc Underwater Acoustic

Networks. IEEE J Oceanic Eng. 40(4):

841-52. http://dx.doi.org/10.1109/JOE.2015.

2469955.

Yilmaz, O. 1987. Seismic Data Processing.

Tulsa: SEG.

Zhang, Y., Yang, J., & Yin, W. 2010. YALL

package. Rice University.

Ziolkowski, A., & Wright, D. 2012. The

potential of the controlled source electromag-

netic method. IEEE Signal Proc Mag. 29(4):

36-52. http://dx.doi.org/10.1109/MSP.2012.

2192529.

12 Marine Technology Society Journal