Robot Learning by Demonstration.pdf

《Robot Learning by Demonstration.pdf》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、SchoolofComputerScienceInternshipProjectReportRobotLearningbyDemonstrationSupervisor:Author:Dr.GerhardLakemeyerKaushikSubramanianHeadofKnowledgeBasedSystemsRutgersUniversityGroupAugust31,2009AbstractInthisreport,twosystemshavebeendevelopedforrobotbehavioracq

2、uisi-tionusingkinestheticdemonstrations.TherstenablesahumanoidrobottoimitateconstrainedreachinggesturesdirectedtowardsatargetusingalearningalgorithmbasedonGaussianMixtureRegression.Theimitationtrajectorycanbereshapedinordertosatisfytheconstraintsofthetaskan

3、ditcanadapttochangesintheinitialconditionsandtotargetdisplace-mentsoccurringduringthemovementexecution.Thesecondisfocusedonbehaviorlearningandwalk-gaitoptimizationbysimulationusingSwarmIntelligence.Thetnessofeachswarmparticleisevaluatedusingasim-ulatoruntil

4、theexpectedbehaviorisreproducedandthentestedontherealrobot.ThepotentialofthesemethodsisevaluatedusingexperimentsinvolvingAldebaran'sNaohumanoidrobotandFawkes,anopensourcerobotsoftwarebytheKBSGatRWTHUniversity.Contents1Introduction21.1Motivation..............

5、..............21.2Contents..............................32TaskSpaceBehaviorLearningusingGMR42.1SystemInputs...........................42.1.1KinestheticDemonstrations...............42.1.2TaskConstraints.....................62.2GaussianMixtureRegression.......

6、...........72.3TrajectoryReproduction.....................92.4ExperimentsandResults.....................102.4.1Observations.......................142.5Limitations............................143BehaviorLearningviaSimulationusingSwarmIntelligence173.1Introduct

7、ion............................173.2BehavioralControlPoints....................173.2.1PrincipalComponentAnalysis.............173.3ParticleSwarmOptimization..................183.4FitnessEvaluationusingWebotsSimulator..........193.5ExtensiontoWalkGaitOptimizat

8、ion.............214Conclusion22Bibliography24AComputerprograms261Chapter1Introduction1.1MotivationToteachtherobotnewbehaviorswithoutextensiveprogramming.ThereareanumberofissuesthatpersistinLearn