基于改进行为克隆算法的机器人运动控制策略

打开文本图片集

中图分类号：TP242.6 文献标志码：A 文章编号：1001-3695（2025）06-009-1668-08

doi：10.19734/j.issn.1001-3695.2024.11.0474

Robot motion control strategy based on improved behavior cloning algorithm

Huang Xiaoxia， Yang Bo† ，Xiang Xin，Chen Ling，Chen Zhongxiang，Sun Shunyao， Xiao Hongfeng （College of Engineering& Design，Hunan Normal University，Changsha 41oo81，China）

Abstract：This paper addressed the challenges of complex motion control strategy training，ineffcientand imprecisepath planing executioninrobotsperformingfineoperationssuchasdual-armcooperativeinsertiontasks.Itproposedanimproved robotbehavior cloning algorithmbasedonmulti-scale feature pyramidsandatention mechanisms.Thealgorithmcombinedresidual networksandfeaturepyramidstodesignthebackbone network，extractingandfusingmulti-scaleimagefeatures，which enhancedtherobot’senvironmentalperceptionand visualfeedback capabilities.Itintroducedanactionsegmentation module to improvetheauracyandsmoohnessofcontrol strategies，reducingcompounderrors inbehaviorcloning.Additionally，the algorithmtrainedthecontrol strategyasaconditional varitionalautoencoder（CVAE）using theatentionmechanism to learn thedistributionofdemonstrationdataandcapturethecorelationbetween image features andactions.This appoachimproved thegeneralizationabilityandadaptabilityofthestrategyinunfamiliarenvironments.Simulationresultsshowthattheproposed algorithmoutperformsfivebaselinemodels intermsofsuccessrateandtrajectorysmoothnessintwofineoperationtasks.These results demonstrate thatthe algorithm can execute precise robot fine operation tasks through simple training.

Key words：precision operation of robot；motion control strategy；behavioral cloning；action sequence

0 引言

随着机器人技术的不断进步，双臂机器人在工业、医疗、家庭等多个领域的应用日益广泛。（剩余19534字）