In thispaper,apracticalprocedureforlinearparameter-varying(LPV)modelingandidentificationof
a roboticmanipulatorispresented,whichleadstoasuccessfulexperimentalimplementationofan
LPV gain-scheduledcontroller.Anonlineardynamicmodelofatwo-degrees-of-freedommanipulator
containingallimportanttermsisobtainedandunknownparameterswhicharerequiredtoconstructan
LPV modelareidentified.Animportanttoolforobtainingamodelofcomplexitylowenoughtobe
suitableforcontrollersynthesisistheprinciple-component-analysis-basedtechniqueofparameterset
mapping.Sincetheresultingquasi-LPVmodelhasalargenumberofaffineschedulingparametersanda
large overbounding,parametersetmappingisusedtoreduceconservatismandcomplexityin
controllerdesignbyfindingtighterparameterregionswithfewerschedulingparameters.Asufficient
a posteriori condition isderivedtoassessthestabilityoftheresultingclosed-loopsystem.Toevaluate
the applicabilityandefficiencyoftheapproximatedmodel,apolytopicLPVgain-scheduledcontrolleris
synthesizedandimplementedexperimentallyonanindustrialrobotforatrajectorytrackingtask.The
experimentalresultsillustratethatthedesignedLPVcontrolleroutperformsanindependentjointPD
controllerintermsoftrackingperformanceandachievesaslightlybetteraccuracythanamodel-based
inverse dynamicscontroller,whilehavingasimplerstructure.Moreover,itisshownthattheLPV
controllerismorerobustagainstdynamicparameteruncertainty.