###
工程科学与技术:2024,56(1):1-10
←前一篇   |   后一篇→
本文二维码信息
码上扫一扫!
面向协作机器人的零力控制与碰撞检测方法研究
(1.东北大学 信息科学与工程学院,辽宁 沈阳 110819;2.新松机器人自动化股份有限公司,辽宁 沈阳 110168;3.东北大学 机器人科学与工程学院,辽宁 沈阳 110169)
Research on Zero-force Control and Collision Detection for Cooperative Robots
(1.School of Info. Sci. & Eng., Northeastern Univ., Shenyang 110819, China;2.SIASUN Robot & Automation Co., Ltd., Shenyang 110168, China;3.Faculty of Robot Sci. and Eng., Northeastern Univ., Shenyang 110169, China)
摘要
图/表
参考文献
相似文献
附件
本文已被:浏览 643次   下载 180
投稿时间:2022-09-16    
中文摘要: 在3C(计算机、通信和消费电子)行业中,对协作机器人的安全、交互、精度、柔顺等方面有严格要求。为了解决协作机器人柔顺交互控制问题,对机器人的零力控制和碰撞检测方法进行了深入研究。首先,建立了一种分析冗余协作机器人牛顿-穆尔彭罗斯(Newton-MP)通用的逆运动学算法,将逆运动学问题转化为Newton-MP法的迭代求解问题。其次,针对协作机器人的零力控制问题,通过同时考虑摩擦力形成完整动力学方程。同时,建立基于加速度3次摩擦力模型的完全动力学方程,采用遗传算法对摩擦力模型进行多参数辨识。再次,提出基于One-class卷积神经网络的碰撞检测方法,构建无碰撞数据集。One-class卷积神经网络在特征空间中引入伪负高斯数据,并使用2元交叉熵损失对网络进行了训练。One-class卷积神经网络碰撞检测方法成功地补偿了模型不确定的动态影响,解决了传统碰撞检测方法建模不准确的问题。最后,通过实验证明,提出的Newton-MP优化方法具有良好的性能,绝对误差达到0.000 13 mm。与理想摩擦力模型进行对比,采用基于速度的3次摩擦力模型拟合出的摩擦力能够更好适用于零力控制。将外力矩观测器与One-class卷积神经网络碰撞检测进行优缺点分析,可以证明,One-class卷积神经网络可以在不依靠模型的情况下,准确地检测机器人的异常碰撞。
Abstract:In the 3C (computer, communications, and consumer electronics) industry, there are strict requirements for robots' safety, interaction, accuracy, and flexibility. To solve the problem of compliant interactive control with cooperative robots, the zero-force control and collision detection methods are studied in this paper. Firstly, a general inverse kinematics (Newton–MP) algorithm is established to analyze the redundant cooperative robots, in which the inverse kinematics problem is transformed into an iterative solution of the Newton–MP method. Secondly, for the zero-force control problem of cooperative robots, the friction force is considered to formulate a complete dynamic equation. Meanwhile, a complete dynamic equation is constructed based on the acceleration cubic friction model, in which the genetic algorithm is applied to identify multi-parameters of friction models. Furthermore, a collision detection method is proposed based on a One-class convolution neural network and an un-collision dataset is built to achieve the detection task. The pseudo-negative Gaussian data is incorporated into the One-class convolutional neural networks to optimize the feature space, and the binary cross-entropy loss serves as the loss function to train the network. The One-class convolutional neural network-based collision detection method has the ability to compensate the dynamic influence of model uncertainty, which solves the problem of inaccurate modeling of traditional collision detection methods. Finally, the experimental results demonstrate that the proposed Newton–MP method achieves desired performance, i.e., 0.00013 mm absolute error. In addition, compared with the ideal friction model, the velocity-fitted cubic friction model is a more preferred solution for zero force control. By analyzing the collision detection method of the external moment observer and the One-class convolution neural network, it can be proved that the One-class convolution neural network can accurately detect the abnormal collision of the cooperative robots in a model-free manner.
文章编号:202201117     中图分类号:TP242    文献标志码:
基金项目:国家自然科学基金重点项目(U20A20197);辽宁省科技重大专项项目(2019JH1/10100005);辽宁省重点研发计划项目(2020JH2/10100040)
作者简介:第一作者:赵彬(1987-),男,高级工程师,博士.研究方向:机器人及自动化;深度学习.E-mail:tech_zhaobin@126.com
引用文本:
赵彬,吴成东,孙若怀,姜杨,吴兴茂.面向协作机器人的零力控制与碰撞检测方法研究[J].工程科学与技术,2024,56(1):1-10.
ZHAO Bin,WU Chengdong,SUN Ruohuai,JIANG Yang,WU Xingmao.Research on Zero-force Control and Collision Detection for Cooperative Robots[J].Advanced Engineering Sciences,2024,56(1):1-10.