基于启示图的软包装货物的机器人抓取  

Robotic Grasping of Soft-packed Goods Based on Affordance Graph

作　　者：于瑞涛 侯言旭 李俊[1,2] Yu Ruitao

机构地区：[1]东南大学自动化学院,江苏,南京210096 [2]复杂工程系统测量与控制教育部重点实验室,江苏南京210096

出　　处：《工业控制计算机》2020年第9期12-13,16,共3页Industrial Control Computer

摘　　要：为了实现密集无序环境中对软包装货物进行机器分拣,设计了基于启示图的软包装货物抓取系统。系统采用先抓取后识别的两阶段策略,首先机器人将软包装货物从分拣箱中取出,再用YOLOv3对取出的货物识别;抓取阶段使用基于启示图的吸取点检测网络,可以对已知和未知软包装货物的吸取点进行检测。实验表明,在密集无序的环境中,使用该系统对软包装货物的分类的准确率达到93.4%,抓取的成功率达到91%,验证了机器人抓取系统的有效性。In order to realize the robot sorting of soft-packed goods in a dense and disordered environment,this paper designs a soft-packed goods grasp system based on the affordance graph.The system adopts a two-stage strategy of grasping first and then recognizing.First,the robot takes the soft-packed goods out of the sorting box,and then uses YOLOv3 to recognize the goods;in the grasping stage,this paper uses an affordance-based suction point detection network to detect the suction point of the known and unknown soft-packed goods.In a dense and disordered environment,the grasping system has a 93.4% accuracy rate in recognizing the goods and a 91% success rate in grasping soft-packed goods.The data verified the effectiveness of the robotic grasping system in a dense and disordered environment.

关 键 词：软包装货物抓取 启示图 卷积神经网络 深度学习 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]