generating manipulation trajectory using motion harmonics · motion learning and generating •...

GeneratingManipulationTrajectoryUsingMotion

HarmonicsYongqiangHuangandYuSun

RobotPerceptionandActionLab(RPAL)ComputerScienceandEngineering

UniversityofSouthFlorida

1

MotionLearningandGenerating

• Observehumanmanipulationmotions– Differenttypesofmanipulationmotiontrajectories

• Representmotion– Essenceofthemotion:spatial-temporalpatterns– Flexibilitytoadapt

• Generatemotiontrajectoryfornewconstraints– Robot’skinematicanddynamicmodels– Environmentconstraints– obstacles– Newstartandgoal

2

MotionRepresentations

• HiddenMarkovmodel(HMM)• Directedgraph,hierarchicalgraph,motiongraph• Principalcomponentanalysis(PCA)• Lineardynamicalsystem(LDS)• Gaussianprocess(GP)+Newtoniandynamics• Movementprimitive• Functionaldataanalysis– primarilyinmotionanalysis

• Manyothers

3

ApproachOverview

4

FunctionalMotionDataAnalysisandRepresentation

5

M(t)=a0+a1f1(t)+a2f2(t)+a3f3(t)

fi(t)– motionharmonic– containsasetofbasisfunctions:B-splineorFourier

DemonstratedMotionTrajectories

6

MotionClustering

7

MotionTrajectoryGenerating

8

{αdist[(Mrobot(a,t),Mdemo(t)]+(dist[Mrobot(a,t),Ci])}

Subjectto

Mrobot (t)∈ [pmin, pmax ],!Mrobot (t)∈ [vmin,vmax ],!!Mrobot (t)∈ [amin,amax ]

Mrobot(t)=a0+a1f1(t)+a2f2(t)+a3f3(t)

i=1

m

∑mina∈R3

Similaritytodemonstratedtrajectories

Passingthroughconstraints

Evaluation

• Weevaluatetwoaspectsofourapproach1. Howwelldoesthenewtrajectorymeetthe

twogoalsoftheoptimization?• resemblingthedemonstratedtrajectories• passtheviapointsatspecifiedtime.

2. Cantheviapointsguidethenewtrajectoryaroundobstacles?

9

Metrics

10

ErrorandSimilarity

11

Evaluation:SuccessatClearingObstacleswithviaPoints

12

Evaluation withNAO• Weusedtherightupper

armofNAOasthekinematicschain

• Werandomlygeneratedsetsofstartandendpoints

• WecomparewiththeLinearSegmentParabolicBlend(LSPB)algorithm,andtheRRTalgorithmfromOMPL

13

VisualComparisons

14

Summary• Representfunctionalmotionswithmotionharmonics

• Keepspatial-temporalmotionpatternsandmeetconstraints

• Usedissimilaritybetweenmotionanddistancetotheconstraintstoevaluate

• Workwithsample-basedmotionplanners• ThismaterialisbaseduponworksupportedbytheNationalScienceFoundationunderGrantNo.1421418.

15

References1. Lin,Y.andSun,Y.,2016.Task-orientedgraspplanningbasedondisturbancedistribution.InRoboticsResearch(pp.577-592).Springer

InternationalPublishing.2. Sun,Y.,YunLin,andYongqiang Huang(2016)RoboticGraspingforInstrumentManipulations,URAI,1-33. Paulius,D.Huang,Y.,Milton,R.,Buchanan,W.D.,SamJ.,andSun,Y.(2016)FunctionalObject-OrientedNetworkforManipulation

Learning,IROS,3655-36624. Huang,Y.andSun,Y.(2015)GeneratingManipulationTrajectoriesUsingMotionHarmonics,IROS,4949-4954.5. Lin,Y.andSun,Y.(2015)Task-BasedGraspQualityMeasuresforGraspSynthesis,IROS,485-490.6. Lin,Y.andSun,Y.,(2015)GraspPlanningtoMaximizeTaskCoverage,Intl.JournalofRoboticsResearch,34(9):1195-1210.7. Lin,Y.,andSun,Y.(2015)RobotGraspPlanningBasedonDemonstratedGraspStrategies,Intl.JournalofRoboticsResearch,34(1):26-

42.8. Sun,Y.,andY.Lin.ModelingPairedObjectsandTheirInteraction."InNewDevelopmentinRobotVision,pp.73-87.SpringerBerlin

Heidelberg,2015.9. Sun,Y.,Ren,S.,andLin,Y.(2014)Object-ObjectInteractionAffordanceLearning,RoboticsandAutonomousSystems,62(4),487-49610. Lin,Y.,Sun,Y.(2014)GraspPlanningBasedonGraspStrategyExtractionfromDemonstration,IROS,pp.4458-4463.11. Dai,W.,Sun,Y.,Qian,X.,(2013)FunctionalAnalysisofGraspingMotion,IROS,pp.3507-3513.12. ChristineBringes,YunLin,YuSun,RedwanAlqasemi (2013)DeterminingtheBenefitofHumanInputinHuman-in-the-LoopRobotic

Systems,IEEEROMAN2013,pp.1-8.13. RenS.,SunY.(2013)Human-Object-Object-InteractionAffordance,IEEEWorkshopinRobotVision(WoRV)/WinterVisionMeeting

(WVM),pp.1-6,201314. LinY.,SunY.(2013)GraspMappingUsingLocalityPreservingProjectionsandKNNRegression,IEEEIntl.ConferenceonRoboticsand

Automation,pp1068-107315. LinY.,RenS.,ClevengerM.,andSunY.(2012)LearningGraspingForcefromDemonstration,IEEEIntl.ConferenceonRobotics and

Automation,pp.1526-1531.16. LinY,SunY(2011)5-DForceControlSystemforFingernailImagingCalibration,IEEEIntl.ConferenceonRoboticsandAutomation,pp.

1374-137917. SunY(2011)FingertipForceandContactPositionandOrientationSensor,IEEEIntl.ConferenceonRoboticsandAutomation,pp.1114-

111916