当前位置: X-MOL 学术Ind. Rob. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Estimation of gravity parameters for multi-DOF serial manipulator arms using model learning with sparsity inducing norms
Industrial Robot ( IF 1.8 ) Pub Date : 2021-08-04 , DOI: 10.1108/ir-03-2021-0052
Chenglong Yu 1 , Zhiqi Li 1 , Dapeng Yang 1 , Hong Liu 1 , Alan F. Lynch 2
Affiliation  

Purpose

This study aims to propose a novel method based on model learning with sparsity inducing norms for estimating dynamic gravity terms of the serial manipulators. This method is realized by operating the robot, acquiring data and filtering the features in signal acquisition to adapt to the dynamic gravity parameters.

Design/methodology/approach

The core principle of the method is to analyze the dictionary composition of the basis function of the model based on the dynamic equation and the Jacobian matrix of an arm. According to the structure of the basis function and the sparsity of the features, combined with joint-angle and driving-torque data acquisition, the effective features of dynamic gravity parameters are screened out using L1-norm optimization and learning algorithms.

Findings

The theoretical analysis revealed that training data obtained based on joint angles and driving torques could rapidly update dynamic gravity parameters. The simulation experiment was carried out by using the publicly available robot model and compared with the previous disassembly method to evaluate the feasibility and performance. The real 7-degree of freedom (DOF) industrial manipulator was used to further discuss the effects of the feature selection. The results show that this estimation method can be fully operational and efficient in industrial applications.

Research limitations/implications

This approach is applicable to most serial robots with multi-DOF and the dynamic gravity parameters of the robot are estimated through learning and optimization. The method does not require prior knowledge of the robot arm structure and only requires joint-angle and driving-torque data acquisition under low-speed motion. Furthermore, as it is a data-driven-based method, it can be applied to gravity parameters updating.

Originality/value

Different from previous general robot dynamic modelling methods, the sparsity of the analytical form of dynamic equations was exploited and model learning was formulated as a convex optimization problem to achieve effective gravity parameters screening. The novelty of this estimation approach is that the method does not only require any prior knowledge but also does not require a specifically designed trajectory. Thus, this method can avoid the laborious work of parameter calibration and the induced modelling errors. By using a data-driven learning approach, the new parameter updating process can be completed conveniently when the robot carries additional mass or the end-effector changes for different tasks.



中文翻译:

使用稀疏诱导范数的模型学习估计多自由度串行机械臂的重力参数

目的

本研究旨在提出一种基于模型学习的新方法,该方法具有稀疏诱导范数,用于估计串行机械手的动态重力项。该方法是通过操作机器人、采集数据和对信号采集中的特征进行滤波来适应动态重力参数来实现的。

设计/方法/方法

该方法的核心原理是基于动力学方程和臂的雅可比矩阵分析模型基函数的字典组成。根据基函数的结构和特征的稀疏性,结合关节角度和驱动力矩数据采集,采用L1范数优化和学习算法筛选出动态重力参数的有效特征。

发现

理论分析表明,基于关节角度和驱动扭矩获得的训练数据可以快速更新动态重力参数。利用公开的机器人模型进行仿真实验,并与以往的拆卸方法进行对比,评估其可行性和性能。使用真正的 7 自由度 (DOF) 工业机械手进一步讨论了特征选择的影响。结果表明,该估算方法在工业应用中可以充分运行且高效。

研究限制/影响

这种方法适用于大多数具有多自由度的串行机器人,并且通过学习和优化来估计机器人的动态重力参数。该方法不需要机器人手臂结构的先验知识,只需要获取低速运动下的关节角度和驱动扭矩数据。此外,由于它是一种基于数据驱动的方法,因此可以应用于重力参数更新。

原创性/价值

与以往一般机器人动力学建模方法不同,利用动力学方程解析形式的稀疏性,将模型学习表述为凸优化问题,实现有效的重力参数筛选。这种估计方法的新颖之处在于该方法不仅需要任何先验知识,而且不需要专门设计的轨迹。因此,该方法可以避免参数校准的繁重工作和引入的建模错误。通过使用数据驱动的学习方法,当机器人携带额外质量或末端执行器因不同任务而变化时,可以方便地完成新的参数更新过程。

更新日期:2021-08-04
down
wechat
bug