机器人示教缝纫动作的学习方法  

The learning method of robot teaching sewing motion

作　　者：王皞燚 王晓华[1] 王文杰 WANG Haoyi;WANG Xiaohua;WANG Wenjie(School of Electronics and Information,Xi’an Polytechnic University,Xi’an 710048)

机构地区：[1]西安工程大学电子信息学院,陕西西安710048

出　　处：《西安工程大学学报》2022年第1期76-84,共9页Journal of Xi’an Polytechnic University

基　　金：国家自然科学基金项目(51905405);陕西省重点研发计划(2019ZDLGY01-08);陕西省重点研发计划项目(2022GY-276)。

摘　　要：提出了一种基于高斯混合模型(Gaussian mixture model,GMM)-高斯混合回归(Gaussian mixture regression,GMR)的机器人动作学习方法,以实现机器人对示教缝纫动作的学习。采用改进的OPENPOSE模型识别示教缝纫动作,并运用标签融合方法更正关节点标签,解决缝纫过程中因布料遮挡造成的关节定位失败问题。以人体上肢关节的坐标变化作为缝纫动作训练样本,采用时间间隔将轨迹样本分割成运动基元,并运用GMM对每段运动基元和时间进行混合编码,得到高斯分量的回归函数。应用GMR对运动基元进行运动预测,生成缝纫动作轨迹,更新回归函数的高斯参数,实现工人上肢缝纫动作的学习。通过轨迹跟踪的仿真实验以及与Kalman方法进行实验对比,验证了本文缝纫动作学习方法的平稳性和有效性。In order to realize the robot′s learning of teaching sewing motion,a robot motion learning method based on Gaussian Mixture Model(GMM)-Gaussian Mixture Regression(GMR)was proposed.The improved OPENPOSE model was used to recognize the teaching sewing movements,and the label fusion method was used to correct the joint point labels,so as to solve the problem of joint positioning failure caused by cloth occlusion during the sewing process.Taking the coordinates of the human upper limb joints as training samples of sewing motion,the trajectory samples were divided into motion primitives by time interval,and each motion primitive and its corresponding time were mixed-encoded using GMM to obtain the regression function of the Gaussian component.Besides,GMR was used to predict the connection among motion primitives to generate the sewing motion trajectory,update the Gaussian parameters of regression function,and realize the learning of worker′upper limbs sewing motion.Through the simulation experiment of trajectory tracking and experiment comparison with Kalman method,the stability and effectiveness of the sewing motion learning method were verified.

关 键 词：缝纫机器人 OPENPOSE模型 示教动作 高斯混合模型 高斯混合回归 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]