摘要
为了在种类多样、位姿随机、背景散乱的堆叠场景下得到机器人抓取目标物体的最优次序与抓取位姿,本文提出一种结合实例分割与抓取筛选的堆叠目标抓取方法,包含抓取次序推理与抓取位姿检测两部分.抓取次序推理部分,首先设计基于自注意力与边界细化的实例分割网络,利用自注意力模块提升特征提取能力,并通过边界细化模块提高堆叠场景下物体边界的分割精度;其次提出一种抓取筛选方法,运用判断物体间完整掩码是否重叠与优先抓取分数排序的策略,筛选出目标最上层未被遮挡的物体为待抓取目标,以减少机器人多余抓取动作.抓取位姿检测部分,为适应不同尺度物体的检测,设计融合多尺度密集残差模块的抓取位姿检测网络,以保留物体的多尺度特征,从而适应不同尺度物体的检测.实验结果表明,本方法能够有效推理出堆叠场景下的抓取次序,并准确检测出物体的位姿,以实现目标的抓取.
In order to obtain the optimal sequence and grasping pose of the robot for grasping the target object under the heap scene with diverse categories,random poses and scattered background,a stacked object grasping method combining instance segmentation and grasp filtering is proposed in this paper,which contains two parts:grasping sequence reasoning and grasping pose detection.For the grasping sequence reasoning part,an instance segmentation network based on self-attention and boundary refinement is designed to improve the feature extraction capability by using self-attention module and enhance the segmentation accuracy of the object boundary in the heap scene by boundary refinement module.Then,a grasping selection method is proposed to use the strategy of judging whether the complete masks of objects overlap and priority grasping score sorting to select the objects on the top of the target as the objects to be grasped,in order to reduce the unnecessary grasping actions of the robot.For the grasping pose detection part,a grasping pose detection network integrating multi-scale dense residual module is designed to retain the multi-scale features of the object,so as to adapt to the detection of different scale objects.The experimental results show that the proposed method can effectively infer the grasping sequence in the heap scene and accurately detect the pose of the object to realize the grasping of the target.
作者
刘光辉
宋鑫
孟月波
徐胜军
LIU Guanghui;SONG Xin;MENG Yuebo;XU Shengjun(College of Information and Control Engineering,Xi′an University of Architecture and Technology,Xi′an 710055,China;Xi′an Key Laboratory of Intelligent Technologyfor Building Manufacturing,Xi′an 710055,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2024年第7期1648-1655,共8页
Journal of Chinese Computer Systems
基金
陕西省重点研究计划项目(2021SF-429)资助。
