机器人逆标定方法比较研究

对两种基于神经元网络的机器人逆标定方法：把机器人位姿和关节角误差作为前馈神经元网络的输入输出,以及把机器人关节角和对应的关节角偏差作为神经元网络输入输出,进行了比较研究。仿真结果表明前者标定效果更好,且更方便简单。把前一种方法的标定结果与传统的几何参数标定方法进行了比较,证明了该方法的有效性,避免了其他传统标定方法繁琐的建模及参数辨识过程。     

机器人实际几何参数识别与仿真

机器人末端位姿的精度依赖于各连杆几何参数的精度,为了提高机器人的位姿精度,需要对机器人进行标定。讨论了运动学模型的建立方法,基于修正的ＤＨ模型和微分变换关系推导出机器人实际几何参数识别的全部公式。该方法要求测量各关节角的码盘数值和机器人的基坐标系下的位姿。     

利用粒子群优化方法标定机械臂关节参数

机械臂的运动精度对于完成精准的操作任务非常重要,而机械臂在使用过程中会由于磨损或维修等原因发生关节参数的变化而失去精度,因此需要再标定.传统的标定方法较为繁琐,依赖于对机械臂模型的建立和解算,且有可能因为关节扭角接近零度而无法收敛.对此对于常见的六自由度机械臂提出了基于粒子群智能优化算法的标定方法,通过测量机械臂多组关节位形及其对应的末端位姿,选取收敛效果最好的理论位姿误差作为适应度函数,编程实现了ABB IRB 120工业机器人关节参数的标定.标定结果表明,该方法虽然需要更长的运算时间,但简单有效、稳定性高,易于实现对长度参数和转角参数的分步标定,且对扭角为零不敏感,精度上也略优于传统方法,可用于不同构型工业机械臂的标定,具有良好的普适性.     

仿爬行类四足步行机器人运动学标定模型研究

针对步行机器人在实际运动中存在的运动误差问题,在介绍仿爬行类机器人Mini Quad-Ⅰ结构的基础上,建立了该类机器人的几何参数模型。讨论了该类四足步行机器人参数误差与结构尺寸及运动轨迹误差的关系,对该类机器人运动学的标定模型进行了详细研究。通过测量机器人在给定轨迹下运动时,机体位姿的变化及各关节运动角的实际运动角,对机器人的结构参数进行标定。开发了标定参数的自动计算系统,并以实例进行了说明。为建立机器人的相对精确控制模型奠定了基础。     

基于机器人的对接位姿视觉测量系统标定研究

针对传统的相机内参矩阵和相机位置参数标定方法较为繁琐复杂且与相机内参标定互相割裂的痛点,本文设计了一种基于机器人的对接位姿视觉测量系统标定方法,利用采集得到的不同位姿关系下标定板图像以及对应的末端机器人末端执行器位姿数据,解算视觉位姿测量过程中涉及到的不同坐标系间位姿转换关系,一次性完成相机内参标定和相机位置参数标定....     

基于最优位姿集的机器人标定及不确定度评定

为解决因标定位姿点随机选择导致机器人标定结果不稳定、可靠性低问题,研究了基于雅克比矩阵奇异值计算可观测 指标的最优位姿点数目及最优位姿集选择算法,建立了机器人 MDH 模型,采用 LM 算法对几何参数进行辨识,使用 LeicaAT960 激光跟踪仪分别在最优位姿集和随机位姿集下对 Staubli TX60 机器人末端位姿大量实测;在分析研究机器人标定不确定度来 源基础上,采用测量不确定指南(GUM)计算几何参数标定的不确定度及蒙特卡洛模拟法对机器人末端位置不确定度进行评 估,结果表明,经最优位姿集标定后的机器人不仅在测试点精度有大幅提升,而且几何参数及末端位置平均不确定度约为随机 位姿集标定的 0. 11 倍,标定结果稳定可靠,泛化能力强,适于在高精度、大范围作业场合推广应用。     

机器人神经网络逆标定法研究与仿真

基于运动学模型的机器人几何参数标定法需要建立复杂的误差模型,在不断变化的环境中缺乏柔性。基于神经网络的机器人逆运动学标定法,是通过一定的算法得到各关节角所对应的误差值,以关节角值为输入,关节角所对应的误差值为输出来训练神经网络,把所有的误差都归结为关节角误差,通过对关节角补偿来驱动机器人。利用遗传算法优化神经网络的权值与阈值,使优化后的BP神经网络能够更好地预测函数输出,利用MATLAB神经网络工具箱进行仿真,结果证明经遗传算法优化的神经网络标定法可以进一步提高标定精度。     

一种基于工具坐标系的机器人运动学参数标定方法

机器人末端执行器位姿误差在基础坐标系中表示时,误差模型中包含姿态误差与位置矢量的乘积项,影响了参数标定识别精度。以工具坐标系为参考系,给出一种基于指数积公式包含关节约束条件的机器人位姿误差标定模型,避免了姿态误差与位置矢量的乘积项对参数标定识别精度的影响。以UR5机器人为标定对象,采用LeciaAT960-MR激光跟踪仪为测量设备,进行参数标定试验。试验结果表明,经参数标定后UR5机器人位置误差模和姿态误差模的平均值分别减小了91.07%和89.16%。     

旋量理论与CPA技术结合的机器人位置误差标定方法

采用圆点分析法(Circle Point Analysis,CPA)进行机器人标定,其测量过程简单,但由于标定结果不够精确,影响了该方法的实用价值。基于旋量理论采用POE法建立机器人的运动学模型,结合圆点分析法获得实际关节的轴线位置,从而得到机器人实际末端位姿。采用最小二乘法标定机器人的末端位置精度。该方法应用于固体火箭发动机绝热层打磨机器人的精度标定过程中,得到了较高的机器人末端执行器的精度,并使CPA法的应用得到了拓展;在有效提高精度的同时,可获得各关节轴线的实际位置信息,为机器人局部关节结构的完善与优化提供了数据参考。     

基于镜面反射的手眼标定方法

针对Eye-on-Hand机器人手眼系统的标定问题,提出一种基于镜面反射的手眼标定方法。通过镜面反射获取图像,根据镜像原理,通过线性求解算法确定机器人末端执行器和相机坐标系的位姿关系。与以往的标定方法不同,该方法只需要3次改变平面镜的姿态,就能同时获得相机的内参数和手眼视觉系统的位姿关系,不会引入传统标定方法中利用机器人运动参数而产生的机器人运动误差。最后将其与Halcon手眼标定算法进行对比,验证了所提方法的正确性和可行性。     

Local POE model for robot kinematic calibration

A robot kinematic calibration method based on the local frame representation of the product-of-exponentials (Local POE) formula is introduced. In this method, the twist coordinates of the joint axes are expressed in their respective local (body) frames. The advantages of this new approach are threefolds: (1) revolute and prismatic joints can be uniformly expressed in the twist coordinates based on the line geometry; (2) the twist coordinates of the joint axes can be set up with simple values because the local frames can be arbitrarily defined on the links; (3) the kinematic parameters described by the twist coordinates vary smoothly that makes the method robust and singularity-free. By assuming that the kinematic errors exist only in the relative initial poses of the consecutive link frames, the kinematic calibration models can be formulated in a simple and elegant way. The calibration process then becomes to re-define a set of new local link frames that are able to reflect the actual kinematics of the robot. This method can be applied to robot manipulators with generic open chain structures (serial or tree-typed). The simulation and experiment results on a 4-DOF SCARA type robot and a 5-DOF tree-typed modular robot have shown that the average positioning accuracy of the end-effector increases significantly after calibration.     

虚拟封闭运动链法提高机器人运动学标定精度

介绍了一种基于激光器构造虚拟封闭运动链的标定方法,该方法通过末端操作器上安装的激光器向观测平板发射激光束,并保持激光投影点处于观测平板上一个固定位置,构造出一个虚拟封闭运动链。在此基础上对IRB140机器人的几何参数进行了标定,经试验验证表明,本标定方法有效地实现了机器人绝对精度的大幅度提高。     

A multilevel calibration technique for an industrial robot with parallelogram mechanism

This paper presents a multilevel calibration technique for improving the absolute accuracy of an industrial robot with a parallelogram mechanism (ABB IRB2400). The parallelogram structural error is firstly modeled based on the partial differential of the position function of a general four-bar linkage and the linearization of the position constraints of the parallelogram mechanism, the model coefficients are fitted from experimental data. Secondly, an absolute kinematic calibration model is established and resolved as a linear function of all the kinematic parameters, as well as the base frame parameters and tool parameters. Finally, contrary to most other similar works, the robot joint space (rather than Cartesian space) is divided into a sequence of fan-shaped cells in order to compensate the non-geometric errors, the positioning errors on the grid points are measured and stored for the error compensation on the target points. After the multilevel calibration, the maximum/mean point positioning errors on 284 tested configurations (evenly distributed in the robot common workspace) are reduced from 1.583/0.420 mm to 0.172/0.066 mm respectively, which is almost the same level as the robot bidirectional repeatability.     

Visual feedback control of a robot in an unknown environment (learning control using neural networks)

In this paper, a visual feedback control approach based on neural networks is presented for a robot with a camera installed on its end-effector to trace an object in an unknown environment. First, the one-to-one mapping relations between the image feature domain of the object to the joint angle domain of the robot are derived. Second, a method is proposed to generate a desired trajectory of the robot by measuring the image feature parameters of the object. Third, a multilayer neural network is used for off-line learning of the mapping relations so as to produce on-line the reference inputs for the robot. Fourth, a learning controller based on a multilayer neural network is designed for realizing the visual feedback control of the robot. Last, the effectiveness of the present approach is verified by tracing a curved line using a 6-degrees-of-freedom robot with a CCD camera installed on its end-effector. The present approach does not necessitate the tedious calibration of the CCD camera and the complicated coordinate transformations. This revised version was published online in October 2004 with a correction to the issue number.     

Active stereovision-based robot learning control for object tracking, fixating and grasping

In this paper, an active stereovision-based control approach is proposed for a robot to track, fixate, and grasp an object in an unknown environment. First, the functional mapping relationships between those joint angles of the active stereovision system and the three-dimensional (3-D) coordinates of the object are derived and expressed in the workspace frame. Second, two feed-forward neural networks are used to learn those functional mapping relationships, which are used for the robot tracking, fixating, and grasping control. Third, the present approach is verified by experiments based on the active stereovision system which is installed in the end-effecter of the robot. Last, the experimental results confirm the effectiveness of the present approach . Significance: The present approach does not necessitate the tedious CCD camera calibration and the complicated coordinate transformations between the visual frames and the joint space frames, and the experimental results show the effectiveness of the present approach.     

胶囊内窥镜无线遥测定位的校正

为了进一步提高采用交流励磁定位无线跟踪胶囊内窥镜的定位精度,减小系统误差,提出了改进的神经网络定位校正方法。首先,设计了适应于胶囊内窥镜定位校正的神经网络结构;然后,采用Levenberg-Marquart算法结合贝叶斯正则化方法改进校正网络,抑制校正网络的过拟合。通过定位实验平台,建立了定位目标的跟踪位置与实际位置的样本对照数据表,并应用校正网络对定位数据进行校正。定位校正实验表明,改进的神经网络校正法可进一步减小定位误差,校正后的X,Y,Z,α,β分量的平均误差分别减小至8.7 mm,10.1 mm,7.3 mm,0.086 rad和0.081 rad。与基本BP算法相比,采用Levenberg-Marquart贝叶斯正则化的改进算法有效提高了定位校正网络的泛化能力和收敛精度。     

一种基于手眼视觉的并联机器人标定方法

提出了一种基于手眼视觉的并联机器人标定方法。基于环路增量法，建立了平面2-DOF冗余驱动并联机器人运动学误差与标定模型；设计了一种标定实验靶板，利用相机采集靶板图像并对其进行分割、识别、旋转补偿的处理，获取机构末端目标位置和实际位置的像素误差值；针对机构自身结构的限制，利用边界曲线识别特征角点，提出了一种基于特征角点确定检测点旋转角度的方法，在补偿相机旋转角度的基础上，再利用简化后的相机针孔模型，将像素误差值通过转换得到机构末端执行器的真实位置误差值；最后利用标定模型和通过视觉系统获取的误差值进行运动学标定。经过4次迭代，机构误差减小为原来的1/3，验证了该方法的可行性。同时该方法具有标定过程用时短、数据量小、实验成本低等优点。     

基于单目视觉的并联机器人末端位姿检测

高效、准确地检测机器人末端位姿误差是实现运动学标定的关键环节。提出一种基于单目摄像机拍摄立体靶标序列图像信息的末端执行器6维位姿误差辨识方法,构造具有平行四边形几何约束的四个空间特征点,并以平行四边形的两个消隐点为约束,建立空间刚体位姿与其二维图像映射关系模型,实现末端位姿的精确定位,然后以Delta高速并联机器人为对象,进行了运动学标定试验,验证该方法的有效性,为这类机器人低成本、快速、在线运动学标定提供重要的理论与技术基础。     

Adaptive neural controller for visual servoing of robot manipulators with camera-in-hand configuration

In this paper, an adaptive neural controller is proposed for visual servoing of robot manipulators with camera-in-hand configuration. The controller is designed as a combination of a PI kinematic controller and feedforward neural network controller that computes the required torque signals to achieve the tracking. The visual information is provided using the camera mounted on the end-effector and the defined error between the actual image and desired image positions is fed to the PI controller that computes the joint velocity inputs needed to drive errors in the image plane to zero. Then the feedforward neural network controller is designed such that the robot??s joint velocities converges to the given velocity inputs. The stability of combined PI kinematic and feedforward neural network computed torque is proved by Lyapunov theory. It is shown that the neural network can cope with the unknown nonlinearities through the adaptive learning process and requires no preliminary off learning. Simulation results are carried out for a three degrees of freedom microbot robot manipulator to evaluate the controller performance.