一种基于高斯过程混合模型的机械臂抓取方法  

A Manipulator Grasping Method Based on Mixture of Gaussian Processes Model

作　　者：陈友东[1] 刘嘉蕾 胡澜晓 CHEN Youdong;LIU Jialei;HU Lanxiao(School of Mechanical Engineering and Automation,Beihang University,Beijing 100191,China)

机构地区：[1]北京航空航天大学机械工程及自动化学院

出　　处：《机器人》2019年第3期343-352,共10页Robot

基　　金：国家科技支撑计划(2015BAF01B04);北京市科技计划(D161100003116002)

摘　　要：为了避免现有的基于视觉的机械臂抓取方法中存在的标定繁琐和求逆困难的不足,提出一种基于高斯过程混合模型的机械臂抓取方法.在学习阶段,利用高斯过程混合模型直接构建目标物体的位姿到机械臂关节角度的映射.在抓取阶段,通过相机获取目标物体的位姿,分别计算各个高斯分量下该位姿的生成概率,选取后验概率最大的高斯分量对应的高斯过程回归计算相应的机械臂关节角度.定位容差为20 mm时,仿真抓取成功率达到93.3%,实际抓取成功率达到了88.3%,对于精度要求不高的抓取作业,该方法可以实现机械臂的快速部署和使用.A manipulator grasping method based on the mixture of Gaussian process(MGP) model is proposed, in order to avoid the shortcomings of the commonly used vision-based method, such as the cumbersome visual calibration and the difficulties in the inverse kinematics solution. In the learning phase, the MGP model is used to directly construct the mapping from the pose of the target object to the joint angles of the manipulator. In the grasping phase, the pose of the target object is firstly captured by the camera. Secondly, the generating probability of the pose under each Gaussian component is calculated respectively. Finally, the Gaussian process regression is selected to calculate the corresponding joint angles,which is corresponding to the Gaussian component with the maximum posterior probability. When the positioning tolerance is 20 mm, the success rate of grasping simulation reaches 93.3%, and the success rate of actual grasping reaches 88.3%. For the grasping with low precision, this method can realize the rapid deployment and the use of the robot.

关 键 词：机械臂抓取 高斯过程混合模型 机器视觉 机器学习 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]