返回
Generalized Model Learning for Reinforcement Learning on a Humanoid Robot Austin Villa 2010

Generalized Model Learning for Reinforcement Learning on a Humanoid Robot Austin Villa 2010

ID：40363279

大小：1.52 MB

页数：6页

时间：2019-08-01

Generalized Model Learning for Reinforcement Learning on a Humanoid Robot Austin Villa 2010 _第1页
Generalized Model Learning for Reinforcement Learning on a Humanoid Robot Austin Villa 2010 _第2页
Generalized Model Learning for Reinforcement Learning on a Humanoid Robot Austin Villa 2010 _第3页
Generalized Model Learning for Reinforcement Learning on a Humanoid Robot Austin Villa 2010 _第4页
Generalized Model Learning for Reinforcement Learning on a Humanoid Robot Austin Villa 2010 _第5页
资源描述：

《Generalized Model Learning for Reinforcement Learning on a Humanoid Robot Austin Villa 2010 》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、InIEEEInternationalConferenceonRoboticsandAutomation(ICRA2010),Anchorage,Alaska,May2010.GeneralizedModelLearningforReinforcementLearningonaHumanoidRobotToddHester,MichaelQuinlan,andPeterStoneDepartmentofComputerScienceTheUniversityofTexasatAustinAustin,TX78712{todd,mquinlan,

2、pstone}@cs.utexas.eduAbstractReinforcementlearning(RL)algorithmshavelongbeenpromisingmethodsforenablinganautonomousrobottoimproveitsbehavioronsequentialdecision-makingtasks.Theobviousenticementisthattherobotshouldbeabletoimproveitsownbehaviorwithouttheneedfordetailedstep-by-

3、stepprogramming.However,forRLtoreachitsfullpotential,thealgorithmsmustbesampleefficient:theymustlearncompetentbehaviorfromveryfewreal-worldtrials.Fromthisperspective,model-basedmethods,whichuseexperientialdatamoreefficientlythanmodel-freeapproaches,areappealing.Buttheyoftenreq

4、uireexhaustiveexplorationtolearnanaccuratemodelofthedomain.Inthispaper,wepresentanalgorithm,ReinforcementLearningwithDecisionTrees(RL-DT),thatusesdecisiontreestolearnthemodelbygeneralizingtherelativeeffectofactionsacrossstates.Theagentexplorestheenvironmentuntilitbelievesith

5、asareasonablepolicy.ThecombinationofthelearningapproachwiththetargetedFig.1.Oneofthepenaltykicksduringthesemi-finalsofRoboCup2009.explorationpolicyenablesfastlearningofthemodel.WecompareRL-DTagainststandardmodel-freeandmodel-basedlearningmethods,anddemonstrateitseffectiveness

6、onanValue-functionmethodscanthemselvesbedividedintoAldebaranNaohumanoidrobotscoringgoalsinapenaltykickmodel-freealgorithms,suchasQ-LEARNING[6],thatarescenario.computationallycheap,butignorethedynamicsoftheworld,thusrequiringlotsofexperience;andmodel-basedalgo-I.INTRODUCTIONr

7、ithms,suchasR-MAX[7],thatlearnanexplicitdomainAsthetasksthatwedesirerobotstoperformbecomemodelandthenuseittofindtheoptimalactionsviasim-morecomplex,andasrobotsbecomecapableofoperatingulationinthemodel.Model-basedreinforcementlearning,autonomouslyforlongerperiodsoftime,wewilln

8、eedtothoughcomputationallymoreintensive,givestheagentthemovefromhand-codeds

当前文档最多预览五页，下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页，下载文档查看全文
温馨提示：
1. 部分包含数学公式或PPT动画的文件，查看预览时可能会显示错乱或异常，文件下载后无此问题，请放心下载。
2. 本文档由用户上传，版权归属用户，天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容，确认文档内容符合您的需求后进行下载，若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误，付费完成后未能成功下载的用户请联系客服处理。