基于改进Faster RCNN的机器人运动控制与视觉抓取研究  

Research on Robot Motion Control and Visual Grasping Based on Improved Faster RCNN

作　　者：王成 李敏[2] 王克宇 WANG Cheng;LI Min;WANG Ke-yu(Hu’nan Vocational and Technical College of Finance and Economics,Hu’nan Hengyang 421002,China;University of South China,Hu’nan Hengyang 421001,China;Hu’nan University of Technology,Hu’nan Zhuzhou 412007,China)

机构地区：[1]湖南财经工业职业技术学院,湖南衡阳421002 [2]南华大学,湖南衡阳421001 [3]湖南工业大学,湖南株洲412007

出　　处：《机械设计与制造》2024年第8期339-345,共7页Machinery Design & Manufacture

基　　金：2019年度湖南省技工教育重点课题(JYKT201906)。

摘　　要：智能机器人执行高精度目标抓取操作的关键是目标视觉识别,且目标视觉识别亦是影响智能机器人工业化应用的核心问题之一。为提升物品抓取机器人的运动控制效果和物体抓取精度,对机器人运动控制和物品视觉抓取进行研究,提出了基于改进Faster RCNN的机器人运动控制与物品视觉识别新方法。为保证智能机器人对物品的识别和抓取精度,该方法通过最小二乘法对机器人本体零位进行标定;采用公开数据集多标签图像对Faster RCNN的卷积层和池化层进行改进,提高目标图像检测识别的准确率和效率;通过基于多目标、多类检测深度学习框架的抓取算法,实现机器人手臂的运动控制建模和求解,提高物品抓取机器人的运动控制效果。为验证所提出的基于改进Faster RCNN的机器人运动控制与物品视觉识别新方法的有效性,对其开展了实际环境中物体的视觉识别和抓取实验。实验结果表明,基于改进Faster RCNN的机器人运动控制与物品视觉识别新方法大大提升了对物品的识别精度,可有效提高机器人对物品的抓取效率。Target visual recognition is the key of intelligent robot to perform high-precision target grasping operation and one of the key problems affecting the industrial application of intelligent robots.To improve the motion control effect and object grasping accuracy of object grasping robot,the robot motion control and object visual grasping are studied,and a new method of robot motion control and object visual recognition based on the improved Faster RCNN is proposed in this research.To ensure the target recognition and grasping precision of intelligent robot,the zero position of robot body is calibrated by the least square method.The convolution layer and pooling layer of Faster RCNN are improved by using multi-label images of open data sets to improve the accuracy and efficiency of target image detection and recognition.Through grasping algorithm based on multi-objective and multi class detection deep learning framework,the motion control modeling and solving of robot arm are realized,and the motion control effect of object grasping robot is improved.For verifying the effectiveness of the proposed method based on improved Faster RCNN for robot motion control and object visual recognition,visual recognition and grasping experiments in real environment are carried out.The experimental results show that the new method of robot motion control and object visual recognition based on improved Faster RCNN can greatly improve the accuracy of object recognition and effectively improve the efficiency of robot grasping object.

关 键 词：目标视觉识别与抓取 机器人零位标定 运动控制 深度学习 

分 类 号：TH16[机械工程—机械制造及自动化] TN98[电子电信—信息与通信工程]