基于力位感知技能学习的轴孔柔顺装配方法

doi:10.11996/JG.j.2095-302X.2024020250

图学学报 ›› 2024, Vol. 45 ›› Issue (2): 250-258.DOI: 10.11996/JG.j.2095-302X.2024020250

• 数字化设计与制造专刊 • 上一篇下一篇

基于力位感知技能学习的轴孔柔顺装配方法

路龙飞(), 王峻峰(), 赵世闻, 李广, 丁鑫涛

华中科技大学机械科学与工程学院，湖北武汉 430074

收稿日期:2024-01-05 修回日期:2024-01-26 出版日期:2024-04-30 发布日期:2024-04-29
通讯作者: 王峻峰(1970-)，男，教授，博士。主要研究方向为数字化智能化装配和人机交互与协作等。E-mail：wangjf@hust.edu.cn
作者简介:路龙飞(2000-)，男，硕士研究生。主要研究方向为机器人装配。E-mail：llf13673586024@163.com
基金资助:
国防基础科研计划资助(JCKY2021203B072)

Peg-in-hole compliant assembly method based on skill learning of force-position perception

LU Longfei(), WANG Junfeng(), ZHAO Shiwen, LI Guang, DING Xintao

School of Mechanical Science and Engineering, Huazhong University of Science and Technology, Wuhan Hubei 430074, China

Received:2024-01-05 Revised:2024-01-26 Online:2024-04-30 Published:2024-04-29
Contact: WANG Junfeng (1970-), professor, Ph.D. His main research interests cover digital and intelligent assembly, robot-human interaction and collaboration, etc. E-mail：wangjf@hust.edu.cn
About author:LU Longfei (2000-), master student. His main research interest covers robot assembly. E-mail：llf13673586024@163.com
Supported by:
Defense Industrial Technology Development Program(JCKY2021203B072)

摘要/Abstract

摘要：

针对传统机器人轴孔装配方法建立精确几何接触模型难、学习方法需要样本大和初始姿态偏差大且成功率低的问题，提出了一种基于力位感知装配技能学习的机器人轴孔柔顺装配方法。在搜孔阶段均匀采集轴未入孔的力和力矩样本数据，构建力-动作数据集，搭建多层感知机(MLP)和注意力模块网络进行监督学习、生成力-动作映射判别模型，根据装配过程中的六维力信号预测下一步装配动作，减小轴中心线与孔中心线的夹角和距离，完成轴孔对准操作；在插孔阶段设计了一种以位置控制为内环的柔顺控制算法，通过设置轴端面的期望接触力，在六维力传感器数据反馈的作用下以主动顺应方式实时调整轴的位置和方向。以最小间隙为0.1 mm的单轴孔为对象，设计了100组装配实验，在平均时间为15.1 s内的装配成功率为94%。通过与其他装配方法比较，提高了轴孔装配的效率和成功率。

关键词: 轴孔装配, 力位感知, 技能学习, 注意力机制, 阻抗控制

Abstract:

Traditional methods for robot peg-in-hole assembly face challenges in constructing accurate geometric contact models and learning methods that require large samples with a high initial attitude deviation leading to a low assembly success rate. A compliant robot peg-in-hole assembly method was proposed based on the skill learning of force-position perception. During the hole search stage, the force and torque sample data for the peg missing the hole were uniformly collected, constructing a force-action dataset. A multi-layer perceptron and an attention module network were constructed for supervised learning, generating a discriminant model for mapping force to action. Based on the six-dimensional force signal in the assembly process, the method predicted the next assembly action, while reducing both the angle and distance between the peg center line and hole center line to achieve proper alignment of the peg and the hole. During the hole insertion stage, a compliance control algorithm was designed with position control as its inner loop. By setting desired contact forces on the end face of the peg, real-time adjustments were made to both the position and orientation of peg parts using active compliance techniques based on feedback from a six-dimensional force sensor. To validate its effectiveness, 100 sets of assembly experiments were conducted using a single axle hole with a minimum clearance of 0.1 mm. The method achieved an average success rate of 94% within an average time of 15.1 seconds. Comparative analysis with other assembly algorithms demonstrated that the force-position perception assembly method based on skill learning significantly enhanced efficiency and success rate in peg-in-hole assemblies.

Key words: peg-in-hole assembly, force-position perception, skill learning, attention mechanism, impedance control

中图分类号:

TP391

路龙飞, 王峻峰, 赵世闻, 李广, 丁鑫涛. 基于力位感知技能学习的轴孔柔顺装配方法[J]. 图学学报, 2024, 45(2): 250-258.

LU Longfei, WANG Junfeng, ZHAO Shiwen, LI Guang, DING Xintao. Peg-in-hole compliant assembly method based on skill learning of force-position perception[J]. Journal of Graphics, 2024, 45(2): 250-258.

图/表 12

图1 两阶段轴孔装配调姿过程

Fig. 1 Pose adjustment process of two-phase peg-in-hole assembly

图2 轴孔柔顺装配算法流程图

Fig. 2 Peg-in-hole compliant assembly algorithm flow chart

图3 力样本采集与标注((a)样本采集；(b)标签标注)

Fig. 3 Force sample collection and direction label annotation ((a) Sample collection; (b) Label annotation)

图4 某次样本采集时X轴六维力变化图

Fig. 4 X-axis l force change curve during a sample collection

图5 多层感知机和自注意力机制网络

Fig. 5 Multilayer Perceptron and Self-attention networks

图6 缩放点积注意力模块

Fig. 6 Scaling dot product attention module

图7 阻抗控制框图

Fig. 7 Impedance control algorithm diagram

图8 机器人轴孔装配系统架构图

Fig. 8 Robot peg-in hole assembly system architecture diagram

图9 实验平台组成

Fig. 9 Composition of experimental platform

图10 装配过程状态((a)轴孔接近；(b)轴孔接触；(c)姿态调整；(d)位置调整；(e)轴孔对准；(f)轴插入孔)

Fig. 10 Assembly process status ((a) Peg approaches hole; (b) Peg contacts hole; (c) Posture adjustment; (d) Position adjustment; (e) Peg and hole alignment; (f) Peg inserts hole)

图11 轴孔装配过程力、力矩与时间曲线((a)力；(b)力矩)

Fig. 11 Curve of force and torque changing with time during peg and hole assembly process ((a) Force; (b) Torque)

表1 不同策略下轴孔装配实验结果

Table 1 Experimental results of peg-in-hole assembly under different strategies

装配策略	装配成功率/%	平均时间/s	平均最大接触力/N
几何模型^[3]	91	26.4	12.47
SVM^[6]	86	17.8	21.62
GMR^[8]	84	16.9	14.56
本文方法	94	15.1	16.35

参考文献 20

[1]	WANG S, CHEN G D, XU H, et al. A robotic peg-in-hole assembly strategy based on variable compliance center[J]. IEEE Access, 1809, 7: 167534-167546. DOI URL
[2]	JIN S Y, ZHU X H, WANG C H, et al. Contact pose identification for peg-in-hole assembly under uncertainties[C]// 2021 American Control Conference. New York: IEEE Press, 2021: 48-53.
[3]	ZOU P, ZHU Q G, WU J, et al. An approach for peg-in-hole assembling based on force feedback control[C]// 2019 Chinese Automation Congress. New York: IEEE Press, 2019: 3269-3273.
[4]	甘亚辉, 段晋军, 戴先中. 非结构环境下的机器人自适应变阻抗力跟踪控制方法[J]. 控制与决策, 2019, 34(10): 2134-2142.
	GAN Y H, DUAN J J, DAI X Z. Adaptive variable impedance control for robot force tracking in unstructured environment[J]. Control and Decision, 2019, 34(10): 2134-2142 (in Chinese).
[5]	LIU Z, SONG L B, HOU Z M, et al. Screw insertion method in peg-in-hole assembly for axial friction reduction[J]. IEEE Access, 2019, 7: 148313-148325. DOI URL
[6]	HUANG H B, CHENG H K, WANG T T, et al. Hole-finding learning strategy for a robot assembly with keyed circular peg[J]. IEEE Access, 2023, 11: 21488-21497. DOI URL
[7]	罗威, 李明富, 赵文权, 等. 基于力-位图像学习的工业机器人柔顺装配方法研究[J]. 机械工程学报, 2022, 58(21): 69-77. DOI
	LUO W, LI M F, ZHAO W Q, et al. Research on compliant assembly method for industrial robots based on force-position image learning[J]. Journal of Mechanical Engineering, 2022, 58(21): 69-77 (in Chinese). DOI
[8]	潘柏松, 颜天野, 胡鑫达, 等. 基于几何约束与隐马尔可夫链模型的轴孔装配策略[J]. 计算机集成制造系统, 2022, 28(12): 3766-3776.
	PAN B S, YAN T Y, HU X D, et al. Peg-in-hole assembly strategy based on geometric constraint and hidden Markov model[J]. Computer Integrated Manufacturing Systems, 2022, 28(12): 3766-3776 (in Chinese).
[9]	GAO X, LING J, XIAO X H, et al. Learning force-relevant skills from human demonstration[EB/OL]. [2024-03-06]. https://www.researchgate.net/publication/330844389_Learning_Force-Relevant_Skills_from_Human_Demonstration.
[10]	KOBER J, GIENGER M, STEIL J J. Learning movement primitives for force interaction tasks[C]// 2015 IEEE International Conference on Robotics and Automation. New York: IEEE Press, 2015: 3192-3199.
[11]	WANG G H, YI J J, JI X C. Research on intelligent peg-in-hole assembly strategy based on deep reinforcement learning[C]// 2023 3rd Asia-Pacific Conference on Communications Technology and Computer Science. New York: IEEE Press, 2023: 468-473.
[12]	INOUE T, DE MAGISTRIS G, MUNAWAR A, et al. Deep reinforcement learning for high precision assembly tasks[C]// 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems. New York: IEEE Press, 2017: 819-825.
[13]	CONNOLLY T H, PFEIFFER F. Neural network hybrid position/force control[C]// 1993 IEEE/RSJ International Conference on Intelligent Robots and Systems. New York: IEEE Press, 2002: 240-244.
[14]	HOU Z M, LI Z H, HSU C, et al. Fuzzy logic-driven variable time-scale prediction-based reinforcement learning for robotic multiple peg-in-hole assembly[J]. IEEE Transactions on Automation Science and Engineering, 2022, 19(1): 218-229. DOI URL
[15]	LIU Y K, XU H, LIU D, et al. A digital twin-based sim-to-real transfer for deep reinforcement learning-enabled industrial robot grasping[J]. Robotics and Computer-Integrated Manufacturing, 2022, 78: 102365. DOI URL
[16]	朱子璐, 刘永奎, 张霖, 等. 基于深度强化学习的机器人轴孔装配策略仿真研究[EB/OL]. (2023-07-20) [2023-12-01]. https://doi.org/10.16182/j.issn1004731x.joss.23-0518.
	ZHU Z L, LIU Y K, ZHANG L, et al. Simulation study on robot shaft-bore assembly strategy based on deep reinforcement learning[EB/OL]. (2023-07-20) [2023-12-01]. https://doi.org/10.16182/j.issn1004731x.joss.23-0518 (in Chinese).
[17]	LUO J L, SOLOWJOW E, WEN C T, et al. Reinforcement learning on variable impedance controller for high-precision robotic assembly[C]// 2019 IEEE International Conference on Robotics and Automation. New York: IEEE Press, 2019: 3080-3087.
[18]	ZHANG X M, ZHENG Y L, OTA J, et al. Peg-in-hole assembly based on two-phase scheme and F/T sensor for dual-arm robot[J]. Sensors, 2017, 17(9): 2004. DOI URL
[19]	李歆雨, 张强, 文闻, 等. 一种基于阻抗控制的机械臂双轴孔插孔柔顺控制方法研究[J]. 空间控制技术与应用, 2022, 48(2): 96-102.
	LI X Y, ZHANG Q, WEN W, et al. A compliance control method for double peg-in-hole assembly of manipulator based on impedance control[J]. Aerospace Control and Application, 2022, 48(2): 96-102 (in Chinese).
[20]	张立建, 胡瑞钦, 易旺民. 基于六维力传感器的工业机器人末端负载受力感知研究[J]. 自动化学报, 2017, 43(3): 439-447.
	ZHANG L J, HU R Q, YI W M. Research on force sensing for the end-load of industrial robot based on a 6-axis force/torque sensor[J]. Acta Automatica Sinica, 2017, 43(3): 439-447 (in Chinese).

基于力位感知技能学习的轴孔柔顺装配方法

Peg-in-hole compliant assembly method based on skill learning of force-position perception

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 20

相关文章 15

编辑推荐

Metrics

本文评价

[1]	李跃华, 仲新, 姚章燕, 胡彬. 基于改进YOLOv5s的着装不规范检测算法研究 [J]. 图学学报, 2024, 45(3): 433-445.
[2]	张相胜, 杨骁. 基于改进YOLOv7-tiny的橡胶密封圈缺陷检测方法 [J]. 图学学报, 2024, 45(3): 446-453.
[3]	李滔, 胡婷, 武丹丹. 结合金字塔结构和注意力机制的单目深度估计 [J]. 图学学报, 2024, 45(3): 454-463.
[4]	郭宗洋, 刘立东, 蒋东华, 刘子翔, 朱熟康, 陈京华. 基于语义引导神经网络的人体动作识别算法[J]. 图学学报, 2024, 45(1): 26-34.
[5]	翟永杰, 赵晓瑜, 王璐瑶, 王亚茹, 宋晓轲, 朱浩硕. IDD-YOLOv7：一种用于输电线路绝缘子多缺陷的轻量化检测方法[J]. 图学学报, 2024, 45(1): 90-101.
[6]	古天骏, 熊苏雅, 林晓. 基于SASGAN的戏剧脸谱多样化生成[J]. 图学学报, 2024, 45(1): 102-111.
[7]	崔克彬, 焦静颐. 基于MCB-FAH-YOLOv8的钢材表面缺陷检测算法[J]. 图学学报, 2024, 45(1): 112-125.
[8]	魏陈浩, 杨睿, 刘振丙, 蓝如师, 孙希延, 罗笑南. 具有双层路由注意力的YOLOv8道路场景目标检测方法[J]. 图学学报, 2023, 44(6): 1104-1111.
[9]	丁建川, 肖金桐, 赵可新, 贾冬青, 崔炳德, 杨鑫. 基于脉冲神经网络的复杂场景导航避障算法[J]. 图学学报, 2023, 44(6): 1121-1129.
[10]	张丽媛, 赵海蓉, 何巍, 唐雄风. 融合全局-局部注意模块的Mask R-CNN膝关节囊肿检测方法[J]. 图学学报, 2023, 44(6): 1183-1190.
[11]	张驰, 张效娟, 赵洋, 杨帆. 基于调色板的半交互式低照度唐卡图像增强[J]. 图学学报, 2023, 44(6): 1202-1211.
[12]	杨陈成, 董秀成, 侯兵, 张党成, 向贤明, 冯琪茗. 基于参考的Transformer纹理迁移深度图像超分辨率重建[J]. 图学学报, 2023, 44(5): 861-867.
[13]	宋焕生, 文雅, 孙士杰, 宋翔宇, 张朝阳, 李旭. 基于改进教师学生网络的隧道火灾检测[J]. 图学学报, 2023, 44(5): 978-987.
[14]	李利霞, 王鑫, 王军, 张又元. 基于特征融合与注意力机制的无人机图像小目标检测算法[J]. 图学学报, 2023, 44(4): 658-666.
[15]	李鑫, 普园媛, 赵征鹏, 徐丹, 钱文华. 内容语义和风格特征匹配一致的艺术风格迁移[J]. 图学学报, 2023, 44(4): 699-709.