基于模板模仿学习的对称物体抓取策略研究

打开文本图片集

中图分类号：TP242.6 文献标识码：A 文章编号：2096-4706（2025）11-0183-04

Research on Grasping Strategies for Symmetrical Objects with Template Imitation Learning

DING Cheng （JAKARoboticsCo.，Ltd.，Shanghai200241，China）

Abstract：Traditional robot grasping involves pose estimation of objects and generation of grasping strategies.For symmetricalobjects，itisnecessary tospecifythe principal axis direction to eliminate pose uncertaintyand determine he grasping direction.Thisresearchaims toenable non-profesionals toachieverobot programmingthroughasingledemonstration to reduce developmentcosts.The paper proposes recording therelative coordinate relationships betweenobjectsdurng the demonstrationprocess as a template，and designing symmetry detection and template matching modules based onthe rotation groupto achieve optimal template matching.Toverifythe efectivenessof theproposedalgorithm，the paperconducts grasping experiments on tree categories ofobjects with diffrentsymmetries.The experimentalresultsshowthat the proposed method is feasible and efficient.

Keywords： robot grasping; demonstration learning; symmetry detection; pose estimation; template matching

0 引言

随着小批量生产的兴起，缩短开发周期的需求变得迫切。（剩余5535字）