基于多源信息融合的协作机器人演示编程及优化方法

doi:10.13973/j.cnki.robot.180164

摘要
图/表
参考文献
相关文章 (15)

全文: PDF (1070 KB) HTML ( ) 输出: BibTeX \| EndNote (RIS)
引用本文:
王斐, 齐欢, 周星群, 王建辉. 基于多源信息融合的协作机器人演示编程及优化方法[J]. 机器人, 2018, 40(4): 551-559.DOI: 10.13973/j.cnki.robot.180164. WANG Fei, QI Huan, ZHOU Xingqun, WANG Jianhui. Demonstration Programming and Optimization Method ofCooperative Robot Based on Multi-Source Information Fusion. ROBOT, 2018, 40(4): 551-559. DOI: 10.13973/j.cnki.robot.180164.

摘要为解决现有机器人装配学习过程复杂且对编程技术要求高等问题，提出一种基于前臂表面肌电信号和惯性多源信息融合的隐式交互方式来实现机器人演示编程.在通过演示学习获得演示人的装配经验的基础上，为提高对装配对象和环境变化的自适应能力，提出了一种多工深度确定性策略梯度算法（M-DDPG）来修正装配参数，在演示编程的基础上，进行强化学习确保机器人稳定执行任务.在演示编程实验中，提出一种改进的PCNN（并行卷积神经网络），称作1维PCNN（1D-PCNN），即通过1维的卷积与池化过程自动提取惯性信息与肌电信息特征，增强了手势识别的泛化性和准确率；在演示再现实验中，采用高斯混合模型（GMM）对演示数据进行统计编码，利用高斯混合回归（GMR）方法实现机器人轨迹动作再现，消除噪声点.最后，基于Primesense Carmine摄像机采用帧差法与多特征图核相关滤波算法（MKCF）的融合跟踪算法分别获取X轴与Y轴方向的环境变化，采用2个相同的网络结构并行进行连续过程的深度强化学习.在轴孔相对位置变化的情况下，机械臂能根据强化学习得到的泛化策略模型自动对机械臂末端位置进行调整，实现轴孔装配的演示学习.

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	王斐
	齐欢
	周星群
	王建辉

关键词 ：演示编程, 多源信息融合, 隐式人机交互, 深度学习, 强化学习

Abstract：In order to solve the problems of complex assembly and learning process of robot and high requirements for the programming technology, an implicit interaction method based on the fusion of forearm sEMG (surface electromyography) and inertial multi-source information is proposed to realize robot demonstration programming. Based on the assembly experience obtained from the demonstration learning by presenters, a multiple deep deterministic policy gradient (M-DDPG) algorithm is proposed to modify the assembly parameters in order to improve the adaptability to the changes of assembly objects and environment. On the basis of demonstration programming, reinforcement learning is carried out to ensure that the robot performs its tasks stably. In demonstration programming experiment, an improved PCNN (parallel convolution neural network) is proposed, named 1-D PCNN (1D-PCNN), which automatically extracts feature inertia and EMG through 1-dimensional convolution and pooling and enhances the generalization performance and accuracy of gesture recognition. In the demonstration reproduction experiment, the Gaussian mixture model (GMM) is used to statistically encode the demo data, and Gaussian mixture regression (GMR) is used to reproduce the robot's trajectory and eliminate noise points. Finally, the environmental changes in the X-axis and Y-axis directions are acquired respectively with Primesense Carmine camera by using the fusion tracking algorithm based on the frame difference method and the multiple kernel correlation filter (MKCF) algorithm. Two identical network structures are used to concurrently carry out the deep reinforcement learning of continuous processes. When the relative position of the peg and hole changes, the robotic arm can automatically adjust the end-effector position according to the generalization strategy model learned by reinforcement learning so as to realize the demonstration learning of the peg-in-hole assembly.

Key words： demonstration programming multi-source information fusion implicit HCI (human-computer interaction) deep learning reinforcement learning

收稿日期: 2018-02-17 录用日期: 2018-05-24

TP241.3

基金资助:装备预研教育部联合基金（6141A02022605）；中央高校基本科研业务费专项资金（N150308001，N172608005）；辽宁省科学事业公益研究基金（20170021）

通讯作者: 王斐,wangfei@mail.neu.edu.cn E-mail: wangfei@mail.neu.edu.cn

作者简介: 王斐(1974-),男,博士,副教授.研究领域:协作机器人,人机交互,生肌电融合.

链接本文:

https://robot.sia.cn/CN/10.13973/j.cnki.robot.180164 或 https://robot.sia.cn/CN/Y2018/V40/I4/551

[1] 马静,李宇.工业机器人自动装配生产线的研制[J].电子技术与软件工程,2017(8):132. Ma J, Li Y. Development of an automatic assembly line based industrial robots[J]. Electronic Technology & Software Engineering, 2017(8):132.
[2] 潘立,鲍官军,胥芳,等.六自由度装配机器人的动态柔顺性控制[J].浙江大学学报(工学版),2018,52(1):125-132. Pan L, Bao G J, Xu F, et al. Dynamic compliant control of six DOF assembly robot[J]. Journal of Zhejiang University (Engineering Science), 2018, 52(1):125-132.
[3] 党宏社,候金良,强华,等.基于视觉引导的SCARA机器人自动装配系统[J].电子技术应用,2017,43(5):21-24. Dang H S, Hou J L, Qiang H, et al. SCARA automatic assembly system based on vision guided[J]. Application of Electronic Technique, 2017, 43(5):21-24.
[4] 沈程慧,白瑞林,李新.视觉引导的装配机器人平面定位补偿方法[J].激光技术,2017,41(1):79-84. Shen C H, Bai R L, Li X. Plane positioning compensation method for assembly robot with visual guiding[J]. Laser Technology, 2017, 41(1):79-84.
[5] Ngeo J G, Tamei T, Shibata T. Continuous and simultaneous estimation of finger kinematics using inputs from an EMG-to-muscle activation model[J]. Journal of Neuroengineering & Rehabilitation, 2014, 11:No.122.
[6] 佟丽娜,侯增广,彭亮,等.基于多路sEMG时序分析的人体运动模式识别方法[J].自动化学报,2014,40(5):810-821. Tong L N, Hou Z G, Peng L, et al. Multi-channel sEMG time series analysis based human motion recognition method[J]. Acta Automatica Sinica, 2014, 40(5):810-821.
[7] Schwung D, Csaplar F, Schwung A, et al. An application of reinforcement learning algorithms to industrial multi-robot stations for cooperative handling operation[C]//IEEE International Conference on Industrial Informatics. Piscataway, USA:IEEE, 2017:194-199.
[8] Rozo L, Jiménez P, Torras C. Robot learning from demonstration of force-based tasks with multiple solution trajectories[C]//IEEE 15th International Conference on Advanced Robotics. Piscataway, USA:IEEE, 2011:124-129.
[9] Rozo L, Jiménez P, Torras C. A robot learning from demonstration framework to perform force-based manipulation tasks[J]. Intelligent Service Robotics, 2013, 6(1):33-51.
[10] Toris R, Suay H B, Chernova S. A practical comparison of three robot learning from demonstration algorithms[C]//ACM/IEEE International Conference on Human-Robot Interaction. NewYork, USA:ACM, 2012:261-262.
[11] Wang F, Zhou J, Lin J, et al. A comparative study on sign recognition using sEMG and inertial sensors[C]//6th lEEE Annual International Conference on Cyber Technology in Automation, Control, and Intelligent Systems. Piscataway, USA:IEEE, 2016:290-295.
[12] Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks[C]//26th Annual Conference on Neural Information Processing Systems. Canada:Neural Information Processing System Foundation, 2012:1097-1105.
[13] LeCun Y, Bengio Y, Hinton G. Deep learning[J]. Nature, 2015, 521(7553):436-444.
[14] Goodfellow I, Bengio Y, Courville A. Deep learning[M]. Cambridge, USA:MIT Press, 2016.
[15] Shen X J, Zhu X C, Jiang X P, et al. Visualization of non-metric relationships by adaptive learning multiple maps t-SNE regularization[C]//IEEE International Conference on Big Data. Piscataway, USA:IEEE, 2017:3882-3887.
[16] Zhang R, Gong W, Grzeda V, et al. An adaptive learning rate method for improving adaptability of background models[J]. IEEE Signal Processing Letters, 2013, 20(12):1266-1269.
[17] Holding T, Lestas I. On the convergence to saddle points of concave-convex functions, the gradient method and emergence of oscillations[C]//53rd IEEE Annual Conference on Decision and Control. Piscataway, USA:IEEE, 2014:1143-1148.
[18] Chen J, Lau H Y K, Xu W, et al. Towards transferring skills to flexible surgical robots with programming by demonstration and reinforcement learning[C]//8th International Conference on Advanced Computational Intelligence. Piscataway, USA:IEEE, 2016:378-384.
[19] Sheng W, Thobbi A, Ye G. An integrated framework for human-robot collaborative manipulation[J]. IEEE Transactions on Cybernetics, 2015, 45(10):2030-2041.
[20] Gu S, Holly E, Lillicrap T, et al. Deep reinforcement learning for robotic manipulation with asynchronous off-policy updates[C]//IEEE International Conference on Robotics and Automation. Piscataway, USA:IEEE, 2016:3389-3396.
[21] Silver D, Lever G, Heess N, et al. Deterministic policy gradient algorithms[C]//31st International Conference on International Conference on Machine Learning. International Machine Learning Society, 2014:605-619.