基于GMR-GPR技能泛化优化策略研究

打开文本图片集

中图分类号：TP24 文献标识码：A 文章编号：2096-4706（2025）18-0039-06

Abstract：Learning from Demonstration （LfD）refers torobots learning humanoperations，anditis simpler than traditional scriptcommandoperationrobots.Robots need to leam human demonstrations toquicklyadapttochanges in theenvironment. Forskill generalzationafterlaingtheobot'sotiontrajectory，thetrajectorygeneratedbyGMR-GPRcannotaccuratelypass throughthe startingconstraint pointsofeach part.Basedonthe GMR-GPRalgorithm，it is improvedtoperform incremental GMR-GPR skill generalization.While keeping the true intentionof the teaching action as muchas possible，itcan also pass through theconstraintpointsatthesametime.Thegeneratedtrajectoryfurtherimprovesthegeneralizationintheconstraintpoint area.Inaddition，thefeasibilityof theproposedmethodisverifedandanalyzedbyperformingletersimulationand UR5erobot experiments.Theresults show thatthe improved method can provide areliable solution to restore thelossof the instructor's correction intention while retaining the true demonstration diagram as much as possible.

Keywords： Learning from Demonstration （LfD）; incremental GMR-GPR; skill generalization; UR5e

0 引言

LfD通过人类示范教会机器执行任务，已广泛应用于行为模型的训练，推动了智能系统的进步。（剩余6682字）