融合物理与几何先验的无抓取标注6-DoF抓取检测方法

doi:10.16182/j.issn1004731x.joss.25-0502

系统仿真学报 ›› 2026, Vol. 38 ›› Issue (5): 1290-1302.doi: 10.16182/j.issn1004731x.joss.25-0502

• • 上一篇

融合物理与几何先验的无抓取标注6-DoF抓取检测方法

石敏¹, 郭诗盛¹, 王素琴¹, 李兆歆²^,³, 朱登明⁴^,⁵

^1.华北电力大学控制与计算机工程学院，北京 102206
^2.中国农业科学院农业信息研究所，北京 100081
^3.农业农村部农业大数据重点实验室，北京 100081
^4.中国科学院计算技术研究所，北京 100190
^5.太仓中科信息技术研究院，江苏太仓 215400

收稿日期:2025-06-03 修回日期:2025-08-11 出版日期:2026-05-21 发布日期:2026-05-29
通讯作者: 王素琴
第一作者简介:石敏(1975-)，女，副教授，博士，研究方向为具身智能、三维场景感知与重建。
基金资助:
国家自然科学基金(61972379);国家自然科学基金(62172392);山东省新旧动能转换重大产业攻关项目(2021-55);苏州市科技计划前沿技术研究(SYG202327);苏州市科技计划(SYC2022048)

Annotation-free 6-DoF Grasp Detection Method Integrating Physical and Geometric Priors

Shi Min¹, Guo Shisheng¹, Wang Suqin¹, Li Zhaoxin²^,³, Zhu Dengming⁴^,⁵

^1.School of Control and Computer Engineering, North China Electric Power University, Beijing 102206, China
^2.Agricultural Information Technology Research Laboratory, Chinese Academy of Agricultural Sciences, Beijing 100081, China
^3.Key Laboratory of Agricultural Big Data, Ministry of Agriculture and Rural Affairs, Beijing 100081, China
^4.Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China
^5.Taicang-CAS Institute of Information and Technology, Taicang 215400, China

Received:2025-06-03 Revised:2025-08-11 Online:2026-05-21 Published:2026-05-29
Contact: Wang Suqin

摘要/Abstract

摘要：

为提升复杂堆叠场景中抓取姿态估计的稳定性与跨类别泛化能力，提出一种融合物理规则与几何结构先验的无标注6-DoF抓取检测方法。在离线阶段，基于多物理约束，构建可行抓取姿态模板库，无需依赖人工抓取标注。在网络设计中，引入物体结构对称性与空间重叠关系建模，设计具备遮挡感知与暴露度建模能力的几何引导机制，并结合关键点回归实现目标物体的稳健姿态对齐。构建多类型堆叠数据集IPA-Stack++，并在单类与混合类工业零部件抓取任务中进行测试。实验结果表明：所提方法在单类型和多类型场景中的抓取成功率明显优于现有其他同类方法，展现出良好的抓取稳定性与通用性。

关键词: 六自由度抓取检测, 无标注学习, 物理规则约束, 几何结构先验, 多任务点云感知

Abstract:

To improve the stability and cross-category generalization capability of grasp pose estimation in complex stacked scenes, an annotation-free 6-DoF grasp detection method integrating physical rules and geometric structure priors was proposed. In the offline stage, a template library of feasible grasp poses was constructed based on multi-physical constraints, without relying on manual grasp annotations. In the network design, the modeling of structural symmetry of objects and spatial overlap relationships was introduced; a geometric guidance mechanism with occlusion perception and exposure modeling capabilities was designed, and robust pose alignment of target objects was achieved by combining keypoint regression. A multi-type stacked dataset IPA-Stack++ was constructed, and tests were conducted in grasping tasks of single-category and mixed-category industrial parts. Experimental results show that the grasp success rate of the proposed method in single-type and multi-type scenes is significantly superior to other existing methods of the same type, demonstrating good grasp stability and versatility.

Key words: 6-DoF grasp detection, annotation-free learning, physical rule constraint, geometric structure prior, multi-task point cloud perception

中图分类号:

TP391.9

石敏,郭诗盛,王素琴等 . 融合物理与几何先验的无抓取标注6-DoF抓取检测方法[J]. 系统仿真学报, 2026, 38(5): 1290-1302.

Shi Min,Guo Shisheng,Wang Suqin,et al . Annotation-free 6-DoF Grasp Detection Method Integrating Physical and Geometric Priors[J]. Journal of System Simulation, 2026, 38(5): 1290-1302.

图/表 16

图1

图2

图3

图4

图5

图6

图7

图8

表1

表2

表3

表4

表5

表6

表7

图9

参考文献 22

[1]	徐健, 刘高峰, 赵一剑, 等. 装配机器人的数字孪生虚实同步及抓取方法[J]. 系统仿真学报, 2024, 36(9): 2181-2192.
	Xu Jian, Liu Gaofeng, Zhao Yijian, et al. The Synchronous Grasping Method of Virtual-real Assembly Robot Based on Digital Twin[J]. Journal of System Simulation, 2024, 36(9): 2181-2192.
[2]	Zhang Jing, Tao Dacheng. Empowering Things with Intelligence: A Survey of the Progress, Challenges, and Opportunities in Artificial Intelligence of Things[J]. IEEE Internet of Things Journal, 2021, 8(10): 7789-7817.
[3]	Wang Shengfan, Jiang Xin, Zhao Jie, et al. Efficient Fully Convolution Neural Network for Generating Pixel Wise Robotic Grasps with High Resolution Images[C]//2019 IEEE International Conference on Robotics and Biomimetics (ROBIO). Piscataway: IEEE, 2019: 474-480.
[4]	Mahler J, Liang J, Niyaz S, et al. Dex-Net 2.0: Deep Learning to Plan Robust Grasps with Synthetic Point Clouds and Analytic Grasp Metrics[EB/OL]. (2017-08-08) [2025-03-24]. .
[5]	Yu Sheng, Zhai Dihua, Xia Yuanqing, et al. SE-ResUNet: A Novel Robotic Grasp Detection Method[J]. IEEE Robotics and Automation Letters, 2022, 7(2): 5238-5245.
[6]	Morrison Douglas, Corke Peter, Leitner Jürgen. Closing the Loop for Robotic Grasping: A Real-time, Generative Grasp Synthesis Approach[EB/OL]. (2018-05-15) [2025-02-14]. .
[7]	Kaynar Furkan, Rajagopalan Sudarshan, Zhou Shaobo, et al. Remote Task-oriented Grasp Area Teaching by Non-experts Through Interactive Segmentation and Few-shot Learning[EB/OL]. (2023-03-17) [2025-03-24]. .
[8]	Liu Jirong, Zhang Ruo, Fang Haoshu, et al. Target-referenced Reactive Grasping for Dynamic Objects[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 8824-8833.
[9]	Fang Haoshu, Wang Chenxi, Gou Minghao, et al. GraspNet-1Billion: A Large-scale Benchmark for General Object Grasping[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2020: 11441-11450.
[10]	Zhao Binglei, Zhang Hanbo, Lan Xuguang, et al. REGNet: REgion-based Grasp Network for End-to-end Grasp Detection in Point Clouds[C]//2021 IEEE International Conference on Robotics and Automation (ICRA). Piscataway: IEEE, 2021: 13474-13480.
[11]	Ren Guangli, Geng Wenjie, Guan Peiyu, et al. Pixel-wise Grasp Detection via Twin Deconvolution and Multi-dimensional Attention[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2023, 33(8): 4002-4010.
[12]	Liang Hongzhuo, Ma Xiaojian, Li Shuang, et al. PointNetGPD: Detecting Grasp Configurations from Point Sets[C]//2019 International Conference on Robotics and Automation (ICRA). Piscataway: IEEE, 2019: 3629-3635.
[13]	Qin Yuzhe, Chen Rui, Zhu Hao, et al. S4G: Amodal Single-view Single-shot SE(3) Grasp Detection in Cluttered Scenes[C]//Proceedings of the Conference on Robot Learning. Chia Laguna Resort: PMLR, 2020: 53-65.
[14]	Wang Chenxi, Fang Haoshu, Gou Minghao, et al. Graspness Discovery in Clutters for Fast and Accurate Grasp Detection[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2021: 15944-15953.
[15]	徐胜军, 崔智伟, 史亚, 等. 多尺度区域注意力堆叠物体抓取检测网络[J]. 激光与光电子学进展, 2025, 62(10): 224-235.
	Xu Shengjun, Cui Zhiwei, Shi Ya, et al. Regional-attention Stacked-object Grasp Detection Network[J]. Laser & Optoelectronics Progress, 2025, 62(10): 224-235.
[16]	Shi Min, Hou Jingzhao, Li Zhaoxin, et al. NG-net: No-grasp Annotation Grasp Detection Network for Stacked Scenes[J]. Journal of Intelligent Manufacturing, 2025, 36(2): 1477-1490.
[17]	石敏, 侯京召, 朱登明, 等. 无抓取标注的堆叠物体抓取位姿检测方法[J]. 计算机辅助设计与图形学学报, 2025, 37(9): 1632-1642.
	Shi Min, Hou Jingzhao, Zhu Dengming, et al. No-label Stacked Object Grasp Pose Detection Algorithm[J]. Journal of Computer-Aided Design & Computer Graphics, 2025, 37(9): 1632-1642.
[18]	Huang Dingtao, Lin Ente, Chen Lipeng, et al. SD-net: Symmetric-aware Keypoint Prediction and Domain Adaptation for 6D Pose Estimation in Bin-picking Scenarios[C]//2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Piscataway: IEEE, 2024: 2747-2754.
[19]	Dong Zhikai, Liu Sicheng, Zhou Tao, et al. PPR-net: Point-wise Pose Regression Network for Instance Segmentation and 6D Pose Estimation in Bin-picking Scenarios[C]//2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Piscataway: IEEE, 2019: 1773-1780.
[20]	Zeng Long, Weijie Lü, Dong Zhikai, et al. PPR-Net++: Accurate 6-D Pose Estimation in Stacked Scenarios[J]. IEEE Transactions on Automation Science and Engineering, 2022, 19(4): 3139-3151.
[21]	Kleeberger Kilian, Landgraf Christian, Huber Marco F. Large-scale 6D Object Pose Estimation Dataset for Industrial Bin-picking[C]//2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Piscataway: IEEE, 2019: 2573-2578.
[22]	楚红雨, 冷齐齐, 张晓强, 等. 融入注意力机制的多模特征机械臂抓取位姿检测[J]. 控制与决策, 2024, 39(3): 777-785.
	Chu Hongyu, Leng Qiqi, Zhang Xiaoqiang, et al. Multi-modal Feature Robotic Arm Grasping Pose Detection with Attention Mechanism[J]. Control and Decision, 2024, 39(3): 777-785.

方法	Pepper	Tless20	Tless29	Gear	GearsSaft
PointNetGPD	84	75	74	80	82
NG-Net	98	80	96	93	97
No-Label^[17]	99	91	94	98	98
本文算法	99	98	99	99	99

方法	Pepper	Tless20	Tless29	Gear
PointNetGPD	72	64	52	56
NG-Net	92	80	96	88
No-Label	80	76	88	82
本文算法	98	96	98	98

方法	单类型场景下					多类型场景下
方法	Pepper	Tless20	Tless29	Gear	Gearshaft	多类型场景下
PointNetGPD	3.381	4.862	4.415	4.363	3.553	9.080
No-Label	1.083	1.407	1.435	3.655	1.222	3.516
本文算法	1.071	1.464	1.481	3.653	0.951	3.371

抓取模板数量	单类型场景下					多类型场景下
抓取模板数量	Pepper	Tless20	Tless29	Gear	Gearshaft	多类型场景下
180	0.123 5	0.127 2	0.215 5	0.141 5	0.146 6	0.171 3
200	0.158 2	0.137 3	0.237 0	0.167 0	0.158 6	0.189 9
220	0.162 6	0.156 6	0.258 5	0.179 9	0.193 5	0.211 4

场景	方法	Pepper	Tless20	Tless29	Gear	Gearshaft
单类型	基于关键点	99	98	99	99	99
单类型	直接回归	96	95	94	97	98
多类型	基于关键点	98	96	98	98
多类型	直接回归	94	94	92	96

融合物理与几何先验的无抓取标注6-DoF抓取检测方法

Annotation-free 6-DoF Grasp Detection Method Integrating Physical and Geometric Priors

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 16

参考文献 22

相关文章 15

编辑推荐

Metrics

本文评价

场景	方法	Pepper	Tless20	Tless29	Gear
单类型	Baseline+得分	99	98	99	99
单类型	Baseline	99	93	96	99
多类型	Baseline+得分	98	96	98	98
多类型	Baseline	96	96	92	88

[1]	周子聪, 曾俊杰, 胡越, 朱正秋, 尹全军. 基于次优示例引导的兵棋推演多智能体强化学习方法[J]. 系统仿真学报, 2026, 38(5): 1277-1289.
[2]	李权, 苏鹏, 万海英, 张承玺, 何志坚, 倪艺洋, 赵忠盖, 刘飞. 基于多阶段LHS-EPRCC方法的青霉素发酵过程建模[J]. 系统仿真学报, 2026, 38(5): 1255-1276.
[3]	梅华威, 杨鹏慧, 余洋. 计及数据漂移改进PatchTST的超短期光伏功率预测[J]. 系统仿真学报, 2026, 38(5): 1239-1254.
[4]	苏泓嘉, 张成, 刘飞. 基于模糊功能依赖网分析的体系效能评估方法[J]. 系统仿真学报, 2026, 38(5): 1224-1238.
[5]	刘银钢, 马明, 张荣华. 基于大语言模型的兵棋推演动态任务规划[J]. 系统仿真学报, 2026, 38(5): 1187-1204.
[6]	李校男, 晁涛, 马萍, 杨明, 王玉轩. 基于期望最大化方法的非线性SSM黑箱鲁棒辨识[J]. 系统仿真学报, 2026, 38(5): 1146-1158.
[7]	董志明, 胡忠奇, 戴浩然, 高建成. 基于大语言模型的作战仿真想定自动化生成方法[J]. 系统仿真学报, 2026, 38(5): 1129-1145.
[8]	南焱栋, 朱金达, 路鑫彬, 秦志英, 齐丹丹, 丁智恒. 数字孪生驱动的核电连接套质检设备状态监测[J]. 系统仿真学报, 2026, 38(4): 1004-1017.
[9]	孙晓哲, 符振宇, 徐兆可, 李键新. 电动加载系统多余力矩机理分析[J]. 系统仿真学报, 2026, 38(4): 1018-1029.
[10]	尚佳策, 张锐, 戴娅, 朱华, 胡思奇, 葛林强, 师晨光. 基于MBSE的低轨通信卫星设计建模方法[J]. 系统仿真学报, 2026, 38(4): 1080-1093.
[11]	赵拓, 刘玉乔, 马远, 成云, 李燊. 基于DoDAF的靶弹目标特性体系结构建模及仿真[J]. 系统仿真学报, 2026, 38(4): 1094-1105.
[12]	李文龙, 桑书涵, 刘玉生, 何海燕, 梁赞, 袁文强, 牛彪, 罗伟峰. 动态模型驱动的模块化航空炸弹系统验证框架[J]. 系统仿真学报, 2026, 38(4): 1106-1118.
[13]	陈鑫杭, 凌晓冬, 郎程畅, 郑师军, 汤伊奇. 基于改进趋近律的共轴双旋翼无人机控制研究[J]. 系统仿真学报, 2026, 38(4): 1119-1128.
[14]	彭莱春阳, 叶飞, 郭晓明, 周靖林. X语言仿真大模型：体系架构、关键技术与典型应用[J]. 系统仿真学报, 2026, 38(4): 869-888.
[15]	王国荣, 邓海顺, 寇子明, 闫宣宣, 黄志祥. 考虑模糊需求与时间容忍度的井工矿车辆路径优化研究[J]. 系统仿真学报, 2026, 38(4): 903-915.