高技术通讯2018，Vol.28Issue(1)：58-66,9.DOI:10.3772/j.issn.1002-0470.2018.01.008
基于深度学习的机器人抓取位置检测方法
A method for robotic grasping position detection based on deep learning
摘要
Abstract
The research on intellignet grasping of robots is conducted,and a method for detection of the robotic grasping position of an object based on deep learning is proposed.The method learns the optimal grasping position of an ob-ject by using the mode of combining unsupervised learning and supervised learning,with the multimodal features of a target object as the training data.In the unsupervised learning process,it uses a contractive auto encoder(CAE) to pre-train the neural network layer by layer,and then the whole network is fine-tuned by using the back propaga-tion algorithm in the supervised learning process.The simulation results verify that the proposed method can give the accurate optimal grasping position of any object.The experiment on Universal Robot 5 shows that the grasping success rate is very high,indicating that the proposed method can be applied to robotic grasping.关键词
深度学习/机器人抓取/位置检测/压缩自动编码器(CAE)Key words
deep learning/robotic grasping/position detection/contractive autoencoder(CAE)引用本文复制引用
闫哲,杜学丹,曹淼,蔡莹皓,鲁涛,王硕..基于深度学习的机器人抓取位置检测方法[J].高技术通讯,2018,28(1):58-66,9.基金项目
国家自然科学基金(61503381)和北京市科技计划(Z171100000817009)资助项目. （61503381）
