IoT-based systems will require a high level of decision making automation both in terms of infrastructure management and within the logic of the IoT applications themselves. Decision support systems (DSSs) are an essential tool in this context on account of their ability to enhance human decision-making processes with machine intelligence. The cognitive automation framework that we have created at Ericsson speeds up the development and deployment of intelligent DSSs by reusing as much knowledge as possible, including domain models, behaviors and reasoning mechanisms. The framework substantially reduces operational costs for IoT-based system management by enabling DSSs to make decisions about how to adapt to changes in their context and environment with minimal or no human interaction.
Handwritten Text Recognition for manuscripts and early printed texts
Ericsson Technology Review: Tackling IoT complexity with machine intelligence
1. THE IoT AND COGNITIVE AUTOMATION ✱
1APRIL 6, 2017 ✱ ERICSSON TECHNOLOGY REVIEW
ERICSSON
TECHNOLOGY
Observer
Connectedvehicles&
passengermobilephones
S
re
Legend:
Raw data streams
Component under development
Implemented component
Component in place – open source software
C H A R T I N G T H E F U T U R E O F I N N O V A T I O N | # 4 ∙ 2 0 1 7
COGNITIVE
AUTOMATION
ASANIoT
ENABLER
2. ✱ THE IoT AND COGNITIVE AUTOMATION
2 ERICSSON TECHNOLOGY REVIEW ✱ APRIL 6, 2017
ANETA VULGARAKIS
FELJAN, ATHANASIOS
KARAPANTELAKIS,
LEONID MOKRUSHIN,
RAFIA INAM, ELENA
FERSMAN, CARLOS
R. B. AZEVEDO,
KLAUS RAIZER,
RICARDO S. SOUZA
Internet of Things (IoT) applications transcend traditional telecom to
include enterprise verticals such as transportation, healthcare, agriculture,
energy and utilities. Given the vast number of devices and heterogeneity
of the applications, both ICT infrastructure and IoT application providers
face unprecedented complexity challenges in terms of volume, privacy,
interoperability and intelligence. Cognitive automation will be crucial to
overcoming the intelligence challenge.
The IoT is built on the concept of cross-
domain interactions between machines that
can communicate with each other without
human involvement. These interactions
generate a vast number of heterogeneous
data streams full of information that must be
analyzed, combined and acted on.
■ Tobesuccessful,providersofICTinfrastructure
andIoTapplicationsneedtoovercomeinterlinking
challengesrelatingtovolume,privacy,interoperability
andintelligence.Doingsorequiresamultifaceted
approachinvolvingconceptsandtechniquesdrawn
frommanydisciplines,asillustratedinFigure1.
Researchin5Gradio,networkvirtualizationand
distributedcloudcomputingprimarilyaddresses
thevolumechallenge,byevolvingnetwork
infrastructureandapplicationarchitecture
designtoincreasetheamountofresources
availabletoapplications(throughput,compute
andstoreresources,forexample).Meanwhile,
theprivacyandinteroperabilitychallengesare
beingaddressedwithamixofR&Deffortsand
standardizationactivitiesthatareleadingtonovel
conceptsandtechniques,suchasdifferential
privacy,k-anonymityalgorithms,securemultiparty
computation,ontologymatching,intelligentservice
discovery,andcontext-awaremiddlewarelayers.
Addressingtheintelligencechallengerequires
thedevelopmentanduseofintelligentdecision
TacklingIoT
complexityWITH
MACHINE
INTELLIGENCE
3. THE IoT AND COGNITIVE AUTOMATION ✱
3APRIL 6, 2017 ✱ ERICSSON TECHNOLOGY REVIEW
supportsystems(DSSs)–interactivecomputer
systemsthatuseacombinationofartificial
intelligenceandmachinelearningmethodsknown
asmachineintelligence(MI)toassistandenhance
humandecisionmaking.Inresponsetothis
need,Ericssonhasbuiltacognitiveautomation
frameworktosupportdevelopmentandoperation
ofintelligentDSSsforlarge-scaleIoT-based
systems.Theusecasesarenotonlythetraditional
telecomones(suchasoperationssupportsystems/
businesssupportsystemsautomation),butnew
onesindomainssuchastransportation,robotics,
energyandutilities,aswell.
Themainbenefitofourcognitiveautomation
frameworkisthatitreducesDSSdevelopmentand
deploymenttimebyreusingasmuchknowledge
aspossible(suchasdomainmodels,behaviors,
architecturalpatterns,andreasoningmechanisms).
ItalsocutsoperationalcostsforIoT-basedsystem
managementbymakingitpossibleduringruntime
foraDSStodecideautomaticallyhowtoadapt
tochangesinitscontextandenvironment,with
minimalornohumaninteraction.
WhileDSSspredominantlyaddressthe
intelligencechallenge,theycanalsobeusedto
addressprivacy,volumeandinteroperability
challenges.Forexample,IoTdevicesneed
intelligencetolearntotrusteachother.
Managementofnetworkinfrastructuremayalso
benefitfromintelligentsystems,tohandlethe
largevolumesoftransmitteddatamoreefficiently.
Finally,systemsthatusedifferentstandardsand/
orAPIscanuseMItointerfacewitheachother,by
usinglanguagegames,forexample.
Architecturalcomponentsandstructure
oftheknowledgebase
Ahigh-levelconceptualviewofthearchitectureof
ourframeworkisdepictedin Figure2.Themain
componentsaretheobserver,theknowledgebase,
thereasonerandtheinterpreter.
Theobserverisresponsibleforpushing
knowledgefromtheenvironmenttotheknowledge
base.Thisisdonebyfirstderivingasymbolic
representationofdatausingmodeltransformation
Figure 1
Key challenges (in gray)
and approaches to
addressing them (in pink)
Volume challenge:
large increase in data traffic
Privacy challenge:
sharing and processing sensitive data
Interoperability challenge:
talking the same language
Intelligence challenge:
sensing and making sense
Standardization
and common APIs
Cloud computing models
Machine intelligence
Cloud computing models
Distributed trust models
4. ✱ THE IoT AND COGNITIVE AUTOMATION
4 ERICSSON TECHNOLOGY REVIEW ✱ APRIL 6, 2017
rulesthatidentifytherelationbetweenrawdata
comingfromtheIoT-basedsystemandasymbolic
representationofthecaptureddata.Thenextstepis
toperformETL(extract,transform,load),followed
byasemanticanalysisoftheobtaineddata.
Theknowledgebasecontainsaformalized
descriptionofgeneralconceptsthatcanbeused
acrossdomainstofacilitateinteroperability,as
wellasdomain-specificconcepts(transportation,
forexample).Inaddition,theknowledgebase
containsthepossiblediscretestatesofthesystem,
potentialtransitionsbetweenstates,meta-
reasoningexpertiseandmodeltransformation
rules.OntologiesinRDF/OWLsemanticmarkup
languageareamongthepossibleformatsofthe
knowledgestoredintheknowledgebase[1].
Domainexpertscanenterdomainandreasoning
expertisedirectlyintotheknowledgebase.Since
thebehavioroftheenvironmentisdynamicand
unpredictable,theknowledgebaseiscontinuously
updatedwithknowledgecomingfromtheobserver
andthereasoner.
Mostoftheframeworkintelligencecomesfromthe
reasoner,whichcontainsgeneralpurposeinference
mechanismsthatallowittodrawconclusionsfrom
information(propositions,rules,andsoon)stored
intheknowledgebase,andproblem-specifictools
thatimplementvariousMItaskssuchasmachine
learning,planning,verification,simulation,and
soon.Byrelatingagoal/missionquerytometa-
reasoningexpertise,theinferenceengineselectsan
appropriatemethodorproblemsolver,andderives
Figure 2
A high-level, conceptual
view of the framework
architecture
Reasoner
Interpreter
Observer
Inference
Connectedthings
Actuation
Training set
Domain expert User
Generated
knowledge
Meta-
reasoning
expertise
Model
transformation rules
Cross-domain
concepts
Reusable states
and transitions
Use-case-specific
states and
transitions
Domain-specific
concepts
Symbolic data
representations
Raw data
streams
Explanation
What-if analysis
Property check
Strategy
Domain and
reasoning
expertise
Mission/goal
Planning
Verification
. . .
Simulation
Machine learning
Knowledgebase
Approver
6. ✱ THE IoT AND COGNITIVE AUTOMATION
6 ERICSSON TECHNOLOGY REVIEW ✱ APRIL 6, 2017
Automatedschedulingoftrainlogisticsservices
Inthisproject,wecreatedaconceptforafully
automatedlogisticstransportationsystem.
Thesystemconsistedofseveralactors(suchas
trains,loadingcranesandrailroadinfrastructure
elements)equippedwithsensorsandactuators,
andcoordinatedbyintelligentsoftware.Themain
goalwastoinvestigatethepotentialbenefitsof
combiningfactualandbehavioralknowledgeto
raisethelevelofabstractioninuserinteraction.A
user-specified,high-levelobjectivesuchas“deliver
cargoAtopointBintimeC”isautomatically
brokendowntosubgoalsbytheintelligent
managementsystem,whichthencreatesand
executesanoptimalstrategytofulfillthatobjective
throughitsabilitytocontrolcorresponding
connecteddevices.
Smartmetering
Weusedourframeworktocreateanintelligent
assistanttohelpEricsson’sfieldengineers
manageEstonia’snetworkofconnectedsmart
electricitymeters.Itstaskwastoautomate
troubleshootingandrepairproceduresbyutilizing
knowledgecapturedfromthedomainexperts.
Byseparatingknowledgefromthecontrollogic,
wemadethesystemextendablesothatnew
knowledgecanbeaddedtoitovertime.The
representationcomprisesseveraldecisiontrees
andworkflowsthatencodetherootcausesof
variouspotentialtechnicalissues,processesfor
diagnosingandidentifyingthem,andthestepsin
thecorrespondingrepairprocedure.Ourproject
revealedthatinthistypeofusecase,thebulkofthe
workisinthemanualacquisitionandencodingof
expertknowledge.
Ticketbook
Whenmanagingandoperatingmobilenetworks,
supportengineersnormallycreatetickets(issue
descriptions)forproblemsclassifiedasnon-
trivial.Basedontheassumptionthatthesolution
toanissuesimilartoanotherthathaspreviously
beendealtwithwillbesimilar,wecreateda
systemforsemanticindexingandsearching
historicaltickets.Weusedavectorspacemodel
tocalculateasimilarityscoreandsortedsearched
ticketsaccordingtotheirrelevancetothecurrent
case.Asimilartechniquewasusedtofind
technicaldocumentationrelevanttothecase
currentlybeingsolved.Thisusecasereliesonthe
availabilityofhistoricaldata(previoustickets)
toautomaticallybuildaknowledgebasethat
facilitatessubsequentinformationretrievaland
relevancyranking.
XaaS
InourXaaSproject,wedevelopedanMI-assisted
platformforservicelifecyclemanagement.By
leveragingtechnologiessuchassemanticweb
andautomatedplanning,wepresentedhow
consumerservicescanbedeliveredautomatically
usingunderlyingmodularcomponentsknown
asmicroservicesthatcanbereusedacross
differentapplicationdomains.Thisautomation
providesflexibilityinservicedeploymentand
decommissioning,andreducesdeploymenttime
frommonthsandweekstominutesandseconds.
Theknowledgebasecomprisedaspecification
ofthedomainusedforservicerequirement
formulation,predefinedservicedefinitions,the
metadataoftheservicefunctionalcomponents,
andthedescriptionsofthemicroservicesthathad
beenusedtoimplementthosecomponents.
CONSUMER SERVICES
CAN BE DELIVERED
AUTOMATICALLY USING
UNDERLYING MODULAR
COMPONENTS KNOWN AS
MICROSERVICES THAT CAN
BE REUSED ACROSS
DIFFERENT APPLICATION
DOMAINS
7. THE IoT AND COGNITIVE AUTOMATION ✱
7APRIL 6, 2017 ✱ ERICSSON TECHNOLOGY REVIEW
Prototypeimplementation–transportation
planning
Theautomationofatransportationplanning
processisanexcellentexampleofhowourcognitive
automationframeworkcanbeusedtocreatean
intelligentDSS.Inthisexample,theDSSisusedto
automaticallydeterminehowtotransportpassengers
orcargoatminimalcost(takingintoaccountthe
distancetraveled,thetimeneededtotransport
eachpassengerandpieceofcargo,orthenumberof
busesrequiredfortransportation)usingconnected
vehicles(busesortrucksrespectively).Theanswer
foraparticulartaskcouldbeaplanthatconsists
ofasequenceofstepsforthesystemtoperformto
reachthegoalstate.Ifthesystemdeterminesthat
thetaskcannotbeperformed,theanswerfromthe
framework-basedDSScouldbethereasonwhy,
aswellasapossiblesolutionsuchasincreasingthe
numberofvehicles.Themotivationbehindusing
theframeworkforthisparticularusecasewasto
demonstratethedecreaseinthecostofdesignby
buildingandreusingcross-domainknowledge
specificallybetweenpeopleandcargologistics.
Figure3providesanoverviewofthepartially
implementedprototypeimplementationof
ourframework-basedDSSfortransportation
planning.(Workisstillongoingtoimplement
missingcomponentsconnectingtheframeworkto
theenvironment.)Theknowledgebasecontains
modeltransformationrulesforPDDLandstates
andtransitionmodelsthatarebasedonaset
ofontologiescontaininginformationforthe
particulartransportationdomain.Theontologies
areorganizedinmultiplelayersofabstraction:one
Reasoner
Interpreter
Observer
Inference
PDDLgenerator
Connectedvehicles&
passengermobilephones
Actuation
Symbolic data
representations
use case data +
domain ontology
(CPSs, ITSs) +
transformation rules
Knowledge input:
Logistics task API
Legend:
Transport plan output:
vehicle schedules
Raw data
streams
PDDL
transformation rules
CPSs ontology
States and
transition library
for transport
logistics
ITSs ontology
Meta-
reasoning
expertise
People
logistics
use case
Goods
logistics
use case
Planning
PDDLplanner
Component under development
Implemented component
Component in place – open source software
Figure 3
Overview of partially
implemented
framework-
based DSS for
transportation
planning
8. ✱ THE IoT AND COGNITIVE AUTOMATION
8 ERICSSON TECHNOLOGY REVIEW ✱ APRIL 6, 2017
whichiscross-domain(thecyber-physicalsystems
ontologyinFigure3);andonewhichisspecificto
intelligenttransportationsystems(ITSs),usedfor
reasoningabouttransportationproblems.
Thisimplementationassumesthattheinference
enginehasalreadydeducedthataplannershould
beusedtosolvethetransportationplanning
problemusingmeta-reasoningexpertise(fromthe
knowledgebase)andauser-specifiedgoal.Theother
implementedcomponentisthePDDLgenerator,
which–giventhePDDLtransformationrules,states
andtransitionfilesasinput–generatesfilesthatare
understandableforaPDDLplanner.Thesource
codeoftheimplementationisavailableonline[3].
Asalways,thereisadirectrelationshipbetween
theamountofknowledgetobeformalizedand
theeffortrequiredtoformalizeit.Putsimply,this
meansthatthemorenon-formalizedknowledge
thereis,thehighertheoperationalcostswill
beintermsoftime,humanresourceallocation
andmoney.Inthecaseofatransportschedule
generationprocess,thebenefitofusingthe
frameworktolowerthecostofdesignisthat
itreducestheeffortrequiredtoformalizethe
knowledgeneededforthesystemtobeautomated.
Figure4showsthedifferenttypesofknowledge
usedtogenerateatransportationplanfortwouse
cases:transportationofpassengersinbusesand
transportationofcargointrucks.Thetransport
networkisviewedasagraph,withpointsof
interest(POIs)asverticesandPOI-connecting
roadsasedges.The“transportableentities”canbe
passengersorcargo,dependingontheusecase.
Transitionsaresimilarregardlessoftheroute,
numberandtypeoftransportagents(busesor
trucks)andtransportableentities.Thismeans
thatreusingtransitionsacrossdifferenttransport
planningusecases,andstoringthemaspartofthe
“transitionlibrary”(seeFigure3),canreducethe
costofdesign.
Figure 4
Knowledge for
transport plan
generation
DOMAIN KNOWLEDGE
Route
Transport agents
Transportable entities
Initial state
Transitions
Goal state
DESCRIPTION
Specification of route(s) as graph(s), which includes vertices, edges
and edge-traversal costs (such as travel time and fuel spent).
Number of vehicles, vehicle IDs and vehicle capacity.
Number and ID of transportable entities (passengers, cargo).
The starting location of the transport agents
and transportable entities.
The transitions that transport agents can perform. Current
prototype implementation contains three transitions: pickup,
drop, and move-to-next-coordinate.
Which criteria need to remain constant for the transport service
to complete the specified route (usually this means that all
transportable entities are picked up and dropped off at specific
points along the route).
9. THE IoT AND COGNITIVE AUTOMATION ✱
9APRIL 6, 2017 ✱ ERICSSON TECHNOLOGY REVIEW
Weuseasimplemetriccalledthereusability
indextomeasurereductionsinthecostofdesign.
Thereusabilityindexistheratioofreusedentities
versustotalentitiesintheknowledgeinput
availabletothePDDLgeneratortogeneratethe
plan.Theratiowas0.364forthebuscaseand0.251
forthetruckusecase.Thismeansthatoutofthe
totalnumberofentitiescreatedforeachcase,36.4
percentwerealreadyavailableinthelibraryforthe
buscaseand25.1percentforthetruckcase.The
contrastbetweenthetwocanmainlybeattributed
tothedifferenceoftheroutespecificationentities,
asagentsonbothroutesfolloweddifferentpaths.
Astheknowledgebasebecomesmorepopulated,
thereusabilityindexwillincrease,whichwillin
turnleadtoadecreaseinthedevelopmenttimeof
furthercustomITSsolutions.Ourmeasurements
alsoindicatethatthereusabilityindexcanriseifthe
routespecificationispartofthePDDLgenerator,
whichcanautomaticallygeneratethespecification
usingdatafromamappingserviceinconjunction
witharoutinglibrary.Theknowledgeengineer
couldthenonlyspecifythedesiredwaypoints(bus
stops,forexample),andtheroutesandgraphswould
begeneratedautomaticallybythePDDLgenerator.
Conclusion
IoT-basedsystemsrequirealargenumberof
decisionstobemadeinashorttime,whichin
turnrequiresautomation–notonlyintermsof
infrastructuremanagement,butalsowithinthe
logicoftheIoTapplicationsthemselves.ADSSis
anessentialtoolinthiscontext,owingtoitsability
toenhancehumandecision-makingprocesseswith
MIbasedonbehavioralmodelsandinformation
containedintherelevantdatastreams.
Anysystemthatrequiresautomationinthe
decision-supportprocesscouldbeenhanced
byourcognitiveautomationframework.Ithasa
domain-agnostic,fixedarchitectureinwhichthe
onlyvariablecomponentsaretheknowledgebase
andthesetofreasoningmethods.Byseparating
cross-domainknowledgefromuse-case-specific
knowledge,theframeworkmakesitpossibleto
maximizereuse,enablingsubstantialsavings
intermsofdesigncost,andshorteningtimeto
marketforcustom-developedsolutions.Theset
ofreasoningmethodsisextensibleandcanbe
populatedwiththemethodsandtoolsthatare
bestsuitedtospecifictasks.Theframework’s
self-adaptationissupportedbymeta-reasoning
andcontinuousreinforcementoftheinternal
knowledgeovertime.Inthefuture,weplanto
extendtheframeworkwithanalysisofhypothetical
situationsandanintuitiveuserinterfaceforboth
expertsandlessexperiencedusers.
Terms and abbreviations
API – application programming interface | CPS – cyber-physical system | DSS – decision support system | ETL –
extract, transform, load | IoT – Internet of Things | ITS – intelligent transportation system | MI – machine intelligence
| OWL – Web Ontology Language | PDDL – Planning Domain Definition Language | POI – point of interest | RDF –
Resource Description Framework | TAaaS – Test Automation as a Service | XaaS – Everything as a Service
THE FRAMEWORK’S
SELF-ADAPTATION IS
SUPPORTED BY META-
REASONING AND
CONTINUOUS
REINFORCEMENT OF THE
INTERNAL KNOWLEDGE
OVER TIME
10. ✱ THE IoT AND COGNITIVE AUTOMATION
10 ERICSSON TECHNOLOGY REVIEW ✱ APRIL 6, 2017
Aneta Vulgarakis
Feljan
◆ has been a senior
researcher in the area of
MI at Ericsson Research
in Sweden since 2014. Her
Ph.D. in computer science
from Mälardalen University
in Västerås, Sweden,
focused on component-
based modeling and
formal analysis of real-time
embedded systems. Feljan
has coauthored more than
30 refereed publications
on software engineering
topics, and served as an
organizer and reviewer of
many journals, conferences
and workshops.
Athanasios
Karapantelakis
◆ joined Ericsson in 2007
and currently works as a
senior research engineer in
the area of MI at Ericsson
Research in Sweden.
He holds an M.Sc. and
Licentiate of Engineering
in communication systems
from KTH Royal Institute of
Technology in Stockholm,
Sweden. His background is
in software engineering.
Leonid Mokrushin
◆ is a senior researcher in
the area of MI at Ericsson
Research in Sweden.
His current focus is on
creating and prototyping
innovative concepts for
telco and industrial use
cases. He joined Ericsson
in 2007 after starting
postgraduate studies
at Uppsala University
focused on the modeling
and analysis of real-time
systems. He holds an M.Sc.
in software engineering
from St. Petersburg State
Polytechnical University
in Russia.
Rafia Inam
◆ has been a researcher
at Ericsson Research in
Sweden since 2015. Her
research interests include
5G cellular networks,
service modeling and
virtualization of resources,
reusability of real-time
software and ITS. She
received her Ph.D. from
Mälardalen University
in Västerås, Sweden, in
2014. Her paper, Towards
automated service-oriented
lifecycle management for
5G networks, won her the
best paper award at the
IEEE’s 9th International
Workshop on Service
Oriented Cyber-Physical
Systems in Converging
Networked Environments
(SOCNE) in 2015.
Elena Fersman
◆ is a research leader in
the area of MI at Ericsson
Research and an adjunct
professor in CPSs at the KTH
Royal Institute of Technology
in Stockholm, Sweden. She
received a Ph.D. in computer
science from Uppsala
University, Sweden, in
2003. Her current research
interests are in the areas
of modeling, analysis and
management of software-
intensive intelligent systems
applied to 5G and industry
and society.
theauthors
11. THE IoT AND COGNITIVE AUTOMATION ✱
11APRIL 6, 2017 ✱ ERICSSON TECHNOLOGY REVIEW
1. PascalHitzler,MarkusKrötzsch,BijanParsia,PeterF.Patel-SchneiderandSebastianRudolph(December
2012)OWL2WebOntologyLanguagePrimer(SecondEdition).W3CRecommendation,availableat:
https://www.w3.org/TR/owl2-primer
2. DrewMcDermott,MalikGhallab,AdeleHowe,CraigKnoblock,AshwinRam,ManuelaVeloso,DanielWeldand
DavidWilkins(1998)PDDL–ThePlanningDomainDefinitionLanguage.TechReportCVCTR-98-003/DCSTR-
1165,YaleCenterforComputationalVisionandControl,availableat:http://homepages.inf.ed.ac.uk/mfourman/
tools/propplan/pddl.pdf
3. PDDLGeneratorforTransportationLogisticsScenarios,basedonCPSandITSontologies,availableat:
https://github.com/SSCIPaperSubmitter/ssciPDDLPlanner
References
Further reading
〉〉 KMARF: A Framework for Knowledge Management and Automated Reasoning presented at a
Development aspects of Intelligent Adaptive Systems (DIAS) workshop (February 2017), available at:
https://arxiv.org/abs/1701.03000
〉〉 The Networked Society will not work without an automation framework, Ericsson Networked Society Blog
(July 2015), available at: https://www.ericsson.com/thinkingahead/the-networked-society-blog/2015/07/09/the-
networked-society-will-not-work-without-an-automation-framework/
Carlos R. B. Azevedo
◆ joined Ericsson Research’s
Brazilian team in 2015.
He is a researcher in the
area of MI whose work is
devoted to supporting and
automating sequential
decision processes by
anticipating and resolving
conflicts. He holds a Ph.D. in
electrical engineering from
the University of Campinas in
Brazil, and his doctoral thesis
was awarded the Brazilian
Ministry of Education’s
2014 Thesis National
Prize in Computation and
Automation Engineering.
Klaus Raizer
◆ is a researcher in the area
of MI at Ericsson Research
in Brazil. He joined the
company in 2015. Raizer
holds a Ph.D. in electrical
and computer engineering
from the University of
Campinas in Brazil, where
he is a founding member
of the IEEE Computational
Intelligence Chapter. His
research interests include
computational intelligence,
machine learning, cognitive
architectures, CPSs,
robotics and automation.
Ricardo S. Souza
◆ joined Ericsson Research
in Brazil in 2016, where he
is a researcher in the area
of MI. He holds an M.Sc. in
electrical and computer
engineering from the
University of Campinas in
Brazil, and he is currently
finalizing his Ph.D. at
the same institution. His
research interests include
distributed systems,
networked and cloud
robotics, shared control
and computational
intelligence.
theauthors