Semantic-mapping-for-mobile-robotics-tasks-A-surve

RoboticsandAutonomousSystems66(2015)86–103ContentslistsavailableatScienceDirectRoboticsandAutonomousSystemsjournalhomepage::AsurveyIoannisKostavelis∗,AntoniosGasteratosLaboratoryofRoboticsandAutomation,ProductionandManagementEngineeringDepartment,DemocritusUniversityofThrace,Vas.Sophias12,GR-67100Xanthi,Greece1highlights•Twolevelnavigation.•Cognitivenavigation.•Spatialsemantics.articleinfoArticlehistory:Received6February2014Receivedinrevisedform7December2014Accepted12December2014Availableonline27December2014Keywords:MobilerobotsSemanticmapTopologicalmapTemporalcoherenceObjectrecognitionPlacerecognitionHuman–robotinteractionKnowledgerepresentationPlanningabstractTheevolutionofcontemporarymobileroboticshasgiventhrusttoaseriesofadditionalconjuncttech-nologies.Ofsuchisthesemanticmapping,whichprovidesanabstractionofspaceandameansforhu-man–robotcommunication.Therecentintroductionandevolutionofsemanticmappingmotivatedthissurvey,inwhichanexplicitanalysisoftheexistingmethodsissought.Theseveralalgorithmsarecatego-rizedaccordingtotheirprimarycharacteristics,namelyscalability,inferencemodel,temporalcoherenceandtopologicalmapusage.Theapplicationsinvolvingsemanticmapsarealsooutlinedintheworkathand,emphasizingonhumaninteraction,knowledgerepresentationandplanning.Theexistenceofpub-liclyavailablevalidationdatasetsandbenchmarking,suitablefortheevaluationofsemanticmappingtechniquesisalsodiscussedindetail.Last,anattempttoaddressopenissuesandquestionsisalsomade.©2014ElsevierB.V.Allrightsreserved.Alocalvillagerknowshiswaybywontandwithoutreflectiontothevillagechurch,tothetownhall,totheshopsandbackhomeagainfromthepersonalpointofviewofonewholivesthere.But,askedtodrawortoconsultamapofhisvillage,heisfacedwithlearninganewanddifferentsortoftask:onethatemployscompassbearingandunitsofmeasurement.Whatwasfirstunderstoodinthepersonaltermsoflocalsnapshotsnowhastobeconsideredinthecompletelygeneraltermsofthecartographer.Thevillager’sknowledgebywont,enablinghimtoleadastrangerfromplacetoplace,isadifferentskillfromonerequiringhimtotellthestranger,inperfectlygeneralandneutralterms,howtogettoanyoftheplaces,orindeed,howtounderstandtheseplacesinrelationtothoseofothervillages.¬GilbertRyle‘‘Abstractions’’∗Correspondingauthor.Tel.:[phone].E-mailaddresses:[e-mail](I.Kostavelis),[e-mail](A.Gasteratos).1[1],howevertoday’sroboticsspecialistshaverealizedthattheyfacethesameproblemasthelocalvillagers,yettheotherwayround.Nowadaysonemayarguethattheproblemofsimultaneouslylo-calizationandmapping(SLAM)hasbeensolved,stilltheoutputofsuchaprocessisonlyperceivablebyamanbearingcompassandunitsofmeasurement.Accordingly,contemporarymobilerobotsbehavelikemachinecartographers,unabletoliaisewithlocalvil-lagers,thatisthehumaninhabitants,whoknowbywonttonav-igatethroughtheownenvironment.Thus,themajorityoftheexistingmappingapproachesaimtoconstructagloballyconsis-tentmetricmapoftherobot’soperatingenvironment.Therobotsbearstateoftheartinstrumentationthatallows,ontheonehand,theconstructionofthemapand,ontheotherhand,theownlocal-izationwithrespecttothismapand,thus,todeterminetheirglobalposewithremarkableaccuracy.Basedonthiscapability,therobots©2014ElsevierB.V.Allrightsreserved.I.Kostavelis,A.Gasteratos/RoboticsandAutonomousSystems66(2015)86–10387Fig.1.Taxonomyofthesemanticmappingmethodologies.Notethatthemetricmappingisconsideredcomplementaryattributeforthesemanticmapping.canplanapathandnavigatetowardsagoal,whichshouldbealsoaspecifiedmetricpositionintheglobalmapreferenceframe.How-beit,forarobottoapprehendtheenvironmentthewayahumandoesand,consequently,toleadastrangerfromplacetoplace,adif-ferentskillthananygeometricalmapcanprovideisrequired.Therobotstocomeshouldbeendowedwithcapacitiestounderstandtheirsurroundingsinahuman-centricterm,i.e.tobeabletotellthedifferencebetweenaroomandacorridorortodiscriminatethedifferentfunctionalityakitchenandalivingroomhave.Therefore,theformationofmapsaugmentedbysemasiologicalattributesin-volvinghumanconcepts,suchastypesofrooms,objectsandtheirspatialarrangement,isconsideredacompulsoryattributeforthefuturerobotsthatshouldbedesignedtooperateinenvironmentsinhabitedbyhumans.Asolutiontothisproblemisofferedbysemanticmapping,aqualitativedescriptionoftherobot’ssurroundings,aimingtoaug-mentthenavigationcapabilitiesandthetask-planning,aswellastobridgethegapinhuman–robotinteraction(HRI),seee.g.[2–4].Especiallytheworkin[4]addressessemanticmappingwithemphasisonHRIbyusingnaturallanguage,thusenablingthemostdirectwayforrobotstosocializewithhumans.Thence,se-manticmappingisaflourishedpioneeringareaencouragingtheelaborationofseveraldoctoraldissertations[5,6].Thetermse-manticderivesfromtheGreekwordσηµαντικ`oς[s¯emantikos],standingforsignificant,whichinturnderivesfromtheverbσηµα`ινϵιν[s¯emainein]