面向雕刻任务的机器人技能学习应用研究  

Application Research on Learning from Demonstration for Robot Carving Task

作　　者：占宏[1] 黎志毅 杨辰光[1] ZHAN Hong;LI Zhiyi;YANG Chenguang(School of Automation Science and Engineering,South China University of Technology,Guangdong Guangzhou 510641,China)

机构地区：[1]华南理工大学自动化科学与工程学院,广东广州510641

出　　处：《机械设计与制造》2025年第1期272-275,共4页Machinery Design & Manufacture

基　　金：广东省高等教育教学研究和改革项目(粤教高函[2023]4号,20220399);华南理工大学第十批探索性实验项目(C9238440)。

摘　　要：缺乏自主作业能力是制约机器人在工业制造领域大规模应用的主要因素之一。机器人通过学习操作技能,可有效加强自主作业能力,进一步提升智能化水平。针对雕刻任务,提出了一种机器人技能学习方法以实现高效作业。该方法首先对任意轨迹进行分割和特征提取,利用梯度提升决策树多分类方法进行轨迹分类,并基于改进型动态运动原语模型的技能学习方法生成待雕刻字样轨迹。同时,在所搭建的机器人雕刻实验平台,采用艾力特机械臂完成了雕刻英文任务,实验结果表明雕刻效果良好,实现了机器人雕刻技能学习。该方法效率高,能够适应不同任务需求,在工业制造领域有着广阔应用前景。Lack of autonomous operation ability is one of the main factors restricting the large-scale application of robots in in⁃dustrial manufacturing.By learning operational skills,the robot can effectively strengthen the ability to operate autonomously and further improve the level of intelligence.A robotic learning from demonstration strategy for carving task is proposed.Firstly,trajectory segment and feature extraction are performed on any trajectories,the multi classification based on gradient boosting de⁃cision tree(GBDT)is used,and the new trajectory is generated by the robot skill learning method based on the improved dynamic motion primitives model.Finally,the carving task is completed by the robot carving experiment platform with good carving re⁃sults,and the robot carving skill learning is realized.This proposed strategy has high efficiency and can adapt to different task re⁃quirements,which has broad application prospects in the field of industrial manufacturing.

关 键 词：机器人 雕刻任务 技能学习 动态运动原语 梯度提升决策树 

分 类 号：TH16[机械工程—机械制造及自动化] TP242[自动化与计算机技术—检测技术与自动化装置]