WO2020125195A1 - 一种二自由度高速并联机器人零点标定方法 - Google Patents
一种二自由度高速并联机器人零点标定方法 Download PDFInfo
- Publication number
- WO2020125195A1 WO2020125195A1 PCT/CN2019/112852 CN2019112852W WO2020125195A1 WO 2020125195 A1 WO2020125195 A1 WO 2020125195A1 CN 2019112852 W CN2019112852 W CN 2019112852W WO 2020125195 A1 WO2020125195 A1 WO 2020125195A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- error
- zero
- robot
- improved
- identification
- Prior art date
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/16—Programme controls
- B25J9/1602—Programme controls characterised by the control system, structure, architecture
- B25J9/1605—Simulation of manipulator lay-out, design, modelling of manipulator
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
- B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
- B25J9/00—Programme-controlled manipulators
- B25J9/003—Programme-controlled manipulators having parallel kinematics
Definitions
- the invention relates to a zero-point calibration method for a two-degree-of-freedom high-speed parallel robot.
- the zero point of the high-speed parallel robot determines the starting position and posture of the system, and the accuracy of the zero point directly determines the accuracy of the end moving platform during the movement process.
- high-speed parallel robots will perform zero calibration before leaving the factory.
- high-speed parallel robots often perform operations such as rapid sorting or boxing. Robots are prone to motion collisions, active joint looseness, or other control failures, resulting in zero points. Lost. Therefore, for high-speed parallel robots, the zero error of the robot needs to be calibrated periodically.
- the zero-point error calibration of high-speed parallel robot usually needs to go through four steps, namely error modeling, measurement, identification and compensation.
- the choice of identification method directly determines the robustness and accuracy of the zero-point error identification process.
- the most commonly used identification methods are least squares and Kalman filtering. Among them, the least square method has poor robustness and accuracy.
- the Kalman filtering method has a relatively fast convergence speed when the prior parameters of the system process noise and measurement noise have been accurately known, so it is widely used.
- the purpose of the present invention is to overcome the shortcomings in the prior art, and a zero-point calibration method for high-speed parallel robots is proposed.
- the zero-point calibration model based on the Kalman filter method is improved to Reduce its dependence on a priori parameters such as process noise and measurement noise, thereby ensuring the robustness and accuracy of the zero identification process.
- an adaptive optimization method of correction parameters in the improved method is established to improve the robustness and accuracy of the machine zero error identification results and improve the efficiency of the robot zero calibration process.
- the technical solution adopted by the present invention to solve the technical problems in the known technology is: a zero-point calibration method for a two-degree-of-freedom high-speed parallel robot, and the steps are as follows:
- Step 1 Establish a mapping model between robot end motion error and zero point error.
- the high-speed parallel robot is composed of a static platform, a moving platform, a first kinematic branch chain and a second kinematic branch chain.
- a telescopic ruler is installed between the static platform and the moving platform, and the distance between the center point P of the moving platform and the origin O of the reference coordinate system O-xyz is measured indirectly through the telescopic ruler.
- the space vector method and the first-order perturbation principle are used to obtain the mapping model between the distance error of the robot end motion and the robot zero error, that is
- ⁇ C is the distance error matrix of the end motion
- ⁇ p is the robot zero error vector
- G is the error transfer matrix
- Step 2 Establish an improved Kalman filtering method for robot zero identification model.
- Equation of state: ⁇ p m ⁇ p m-1 +a m-1
- ⁇ p m and ⁇ p m-1 are the state vectors (that is, zero error vectors) of the system during the mth and m-1th identification processes;
- ⁇ C m is the robot end obtained by measurement during the mth identification process Motion error matrix;
- a m-1 is the process noise vector during the m-1th identification process, the vector is assumed to follow a multivariate Gaussian distribution with mean 0 vector and covariance matrix Q;
- b m is the mth measurement process The measured noise vector is assumed to follow a multivariate Gaussian distribution with mean 0 and covariance matrix R.
- the accuracy of the estimation of the covariance matrix Q and R directly affects the effectiveness of the Kalman filter algorithm.
- the covariance matrix R of measurement noise can be roughly given by the measurement error parameters of the measuring instrument itself under ideal conditions, in practice, it is difficult to be accurate due to the influence of changes in the working environment or the instability of the instrument itself. obtain.
- due to the servo system adjustment process and the reducer error it is more difficult to accurately obtain the a priori information of the process noise covariance matrix Q, so it can be simply set to the form of a diagonal matrix here, namely
- q and r are constants greater than 0.
- q and r can take smaller orders of magnitude, such as 0 ⁇ 1 ⁇ 10 2 ; I 1 and I 2 are the same order as the covariance matrix Q and R, respectively Identity matrix.
- Step 2.1 Estimate the state vector of the identification process
- Step 2.2 Estimate the covariance matrix Q of the estimation error
- m-1 is the covariance matrix of the estimated error estimated during the mth identification process
- m-1 is the covariance of the estimated error obtained after the m-1 identification process Variance matrix.
- Step 2.3 Establish an improved Kalman gain expression
- K m ( ⁇ m ) P m
- K m ( ⁇ m ) is the improved Kalman gain in the m-th identification process; ⁇ m is the correction parameter used in the m-th identification process, and ⁇ m >0.
- the correction term ⁇ m I 2 is introduced to avoid inaccurate estimation of the covariance matrix Q and R of process noise and measurement noise, resulting in an inverse matrix (GP m
- Step 2.4 Update state vector with improved Kalman gain
- Step 2.5 Update the covariance matrix of the estimated error
- m is the covariance matrix of the estimated error updated after the m-th identification process.
- Step 2.6 Repeat the recursive process from Step 2.1 to Step 2.5 until the estimation results of the two adjacent state vectors meet
- ⁇ is a threshold vector for the identification accuracy given artificially.
- Step 3 The modified L-curve method is used to optimize the modified parameters in the improved Kalman filtering method.
- the correction parameter ⁇ m in each recursive process needs to be optimized.
- the present invention establishes an improved L curve method to optimize the correction parameter ⁇ m.
- the source analysis is as follows: the updated state vector Is rewritten as
- the modified L-curve method is used to optimize the modified parameters in the improved Kalman filtering method.
- the process is as follows:
- Step 3.1 ⁇ m intervals in the section of ⁇ m sampling [0,1 ⁇ 10 2], thereby to obtain a series of discrete points, horizontal and vertical coordinates of discrete points are
- Step 3.2 Use cubic spline interpolation to fit discrete points to obtain an improved L curve
- f CSI represents the improved L curve equation obtained after cubic spline interpolation.
- the correction parameter ⁇ mi corresponding to the maximum curvature point is the optimal correction parameter.
- Step 4 Adaptive optimization of correction parameters.
- an adaptive optimization method for the correction parameters is established. Since the covariance matrices Q and R are fixed matrices, the value of the improved Kalman gain K m ( ⁇ m ) during the recursive process is determined by the covariance matrix P m-1
- ⁇ (m) represents the penalty factor in the m-th identification process.
- Step 5 Compensate the zero-point error identification results of the high-speed parallel robot to the robot kinematics model.
- the zero-point calibration method of the high-speed parallel robot proposed by the present invention can solve the problem of inaccurate estimation of the covariance matrix of the process noise and measurement noise of the zero-point identification model by introducing correction items into the zero-point identification model based on the traditional Kalman filter method in the past
- the resulting divergence of identification results effectively guarantees the robustness and accuracy of the zero identification process.
- the zero-point calibration method of the two-degree-of-freedom high-speed parallel robot proposed by the present invention optimizes the correction parameters in the recursive process of the zero identification model of the improved Kalman filter method through the improved L curve method, which improves the robustness of the identification results and accuracy.
- an adaptive optimization method for correction parameters is established, thereby effectively improving the efficiency of the robot zero identification process.
- Figure 1 is the zero-point calibration system of a two-degree-of-freedom high-speed parallel robot
- Fig. 2 is a simplified schematic diagram of the zero calibration system of a two-degree-of-freedom high-speed parallel robot
- Figure 3 is a schematic diagram of the improved L curve method
- Figure 4 is a comparison chart of the zero-point error identification effect in the experiment.
- a two-degree-of-freedom high-speed parallel robot is composed of a static platform 1, a moving platform 2, a first kinematic branch 3, and a second kinematic branch 4.
- a telescopic ruler 7 is installed between the static platform 1 and the moving platform 2, and both ends of the telescopic ruler 7 are respectively connected to the static platform 1 and the moving platform 2 through a rotating pair.
- the distance between the center point P of the moving platform and the origin O of the reference coordinate system O-xyz can be measured indirectly through the telescopic scale 7.
- the space vector method and the first-order perturbation principle can be used to obtain a mapping model between the distance error of the robot's end motion and the robot zero error, namely
- ⁇ C is the distance error matrix of the end motion
- ⁇ p is the robot zero error vector
- G is the error transfer matrix
- Equation of state: ⁇ p m ⁇ p m-1 +a m-1
- ⁇ p m and ⁇ p m-1 are the state vectors (that is, zero error vectors) of the system during the mth and m-1th identification processes;
- ⁇ C m is the robot end obtained by measurement during the mth identification process Motion error matrix;
- a m-1 is the process noise vector during the m-1th identification process, the vector is assumed to follow a multivariate Gaussian distribution with mean 0 vector and covariance matrix Q;
- b m is the mth measurement process The measured noise vector is assumed to follow a multivariate Gaussian distribution with mean 0 and covariance matrix R.
- the accuracy of the estimation of the covariance matrix Q and R directly affects the effectiveness of the Kalman filter algorithm.
- the covariance matrix R of measurement noise can be roughly given by the measurement error parameters of the measuring instrument itself under ideal conditions, in practice, it is difficult to be accurate due to the influence of changes in the working environment or the instability of the instrument itself. obtain.
- due to the servo system adjustment process and the reducer error it is more difficult to accurately obtain the a priori information of the process noise covariance matrix Q, so it can be simply set to the form of a diagonal matrix here, namely
- q and r are constants greater than 0.
- q and r can take smaller orders of magnitude, such as 0 ⁇ 1 ⁇ 10 2 ; I 1 and I 2 are the same order as the covariance matrix Q and R, respectively Identity matrix.
- m-1 is the covariance matrix of the estimated error estimated during the mth identification process
- m-1 is the covariance of the estimated error obtained after the m-1 identification process Variance matrix.
- K m ( ⁇ m ) P m
- K m ( ⁇ m ) is the improved Kalman gain in the m-th identification process; ⁇ m is the correction parameter used in the m-th identification process, and ⁇ m >0.
- the correction term ⁇ m I 2 is introduced to avoid inaccurate estimation of the covariance matrix Q and R of process noise and measurement noise, resulting in the inverse matrix (G P m
- m is the covariance matrix of the estimated error updated after the m-th identification process.
- ⁇ is a threshold vector for the identification accuracy given artificially.
- Updated state vector Is rewritten as
- f CSI represents the improved L curve equation obtained after cubic spline interpolation.
- the correction parameter ⁇ mi corresponding to the maximum curvature point is the best correction parameter, which needs to be brought into the improved Kalman filter method for calculation.
- the value of the improved Kalman gain K m ( ⁇ m ) during the recursive process is determined by the covariance matrix P m-1
- ⁇ (m) represents the penalty factor in the m-th identification process.
- the identification result of the zero-point error needs to be compensated into the kinematic model of the robot.
- the The theoretical rotation angle outputs corresponding to the first active arm 5 and the second active arm 6 are ⁇ 1 and ⁇ 2, respectively .
- the zero point errors ⁇ 1 and ⁇ 2 of the actual robot are known to be not zero, and the actual rotation angle output of the first active arm 5 and the second active arm 6 of the robot should be
- ⁇ ′ 1 ⁇ 1 - ⁇ 1
- ⁇ ′ 2 ⁇ 2 - ⁇ 2
- ⁇ ′ 1 and ⁇ ′ 2 represent the actual output rotation angles of the first active arm 5 and the second active arm 6, respectively.
- the traditional zero-point identification model based on the Kalman filter method may cause the zero-point identification results of the two-degree-of-freedom parallel robot shown in FIG. 1 to diverge, and the present invention can quickly obtain the convergent zero-point identification results.
Landscapes
- Engineering & Computer Science (AREA)
- Robotics (AREA)
- Mechanical Engineering (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
Abstract
一种二自由度高速并联机器人零点标定方法,在基于传统卡尔曼滤波方法的零点辨识模型中引入了修正项,并建立了一种改进的L曲线法用于对模型递归计算过程中的修正参数进行优选,此外结合传统卡尔曼滤波方法递归过程中估计误差的协方差矩阵特性建立了修正参数的自适应优选方法,从而有效提高了机器人零点辨识过程棒性、准确性和辨识效率。
Description
本发明涉及一种二自由度高速并联机器人零点标定方法。
高速并联机器人的零点决定了系统的起始位姿,并且零点的准确性直接决定了其末端动平台在运动过程的精度。通常高速并联机器人在出厂之前会进行零点标定,但在实际使用过程中由于高速并联机器人经常执行快速分拣或装箱等操作,机器人容易发生运动碰撞、主动关节松动或其它控制故障,进而造成零点的丢失。故对高速并联机器人而言,需周期性的对机器人的零点误差进行标定。
高速并联机器人的零点误差标定通常需要经历四个步骤,即误差建模、测量、辨识和补偿。其中辨识方法的选择直接决定了零点误差辨识过程的鲁棒性和准确性。目前,最常用的辨识方法有最小二乘法和卡尔曼滤波法等。其中,最小二乘法的鲁棒性和准确性较差,卡尔曼滤波法在已准确获知系统过程噪声和测量噪声等先验参数的情况下具有相对较快的收敛速度,故被广泛使用。
然而,在实际高速并联机器人的零点误差辨识过程中,系统的过程噪声和测量噪声特性受到工作环境变化、测量仪器特性波动、伺服调节系统波动以及减速器误差等因素的影响而难以准确获得,通常需要不断地对相关参数进行手动试凑,直至卡尔曼滤波方法所辨识出的零点误差结果收敛到一定阈值位置为止。这极大地制约了零点标定过程的效率,并降低了标定过程的自动化程度。
发明内容
本发明的目的是为了克服现有技术中的不足,提出了一种高速并联机器人零点标定方法,在高速并联机器人的零点标定过程中,通过对基于卡尔曼滤波方法的零点标定模型进行改进,以降低其对于过程噪声和测量噪声等先验参数的依赖性,进而保证零点辨识过程的鲁棒性和准确性。与此同时,建立改进方法中修正参数的自适应优选方法,从而提升机器零点误差辨识结果的鲁棒性和准确性,并提高机器人零点标定过程的效率。
本发明为解决公知技术中存在的技术问题所采取的技术方案是:一种二自由度高速并联机器人零点标定方法,其步骤如下:
步骤1.建立机器人末端运动误差与零点误差之间的映射模型。
高速并联机器人由静平台、动平台、第一运动支链和第二运动支链组成。静平台和动平台之间安装一根伸缩尺,通过伸缩尺间接测量出动平台中心点P到基准坐标系O-xyz原点O 之间的距离。
初始状态下,当该机器人的第一主动臂和第二主动臂与基准坐标系O-xyz的x轴平行时,即θ
1=θ
2=0°,此时机器人的零点误差Δθ
1=Δθ
2=0°。
利用空间矢量法和一阶摄动原理获得机器人末端运动的距离误差与机器人零点误差之间的映射模型,即
ΔC=GΔp
其中,ΔC为末端运动的距离误差矩阵;Δp为机器人零点误差向量;G为误差传递矩阵。
步骤2.建立改进的卡尔曼滤波方法的机器人零点辨识模型。
基于机器人零点误差和末端运动误差和之间的映射模型,得到辨识系统的状态方程和观测方程
状态方程:Δp
m=Δp
m-1+a
m-1
观测方程:ΔC
m=GΔp
m+b
m
其中,Δp
m和Δp
m-1分别为第m次和第m-1次辨识过程中系统的状态向量(即零点误差向量);ΔC
m为第m次辨识过程中通过测量所获得的机器人末端运动误差矩阵;a
m-1为第m-1次辨识过程中的过程噪声向量,该向量假设服从均值为0向量和协方差矩阵为Q的多元高斯分布;b
m为第m次测量过程中的测量噪声向量,该向量假设服从均值为0和协方差矩阵为R的多元高斯分布。
通常情况下,对协方差矩阵Q和R估计的准确性直接影响了卡尔曼滤波算法的效果。虽然测量噪声的协方差矩阵R在理想情况下可大致由测量仪器本身的测量误差参数给出,但实际中受工作环境变化或仪器本身不稳定性的影响,测量噪声的协方差矩阵R难以准确获得。此外,受伺服系统调节过程和减速器误差的影响,准确获得过程噪声协方差矩阵Q的先验信息将更加困难,故在此可将其简单设置为对角矩阵的形式,即
Q=qI
1,R=rI
2
其中,q和r为大于0的常数,在零点标定过程中q和r可取较小的数量级,如0~1×10
2;I
1和I
2分别为与协方差矩阵Q和R同阶的单位矩阵。
故,建立改进的卡尔曼滤波方法的机器人零点辨识模型的过程如下:
步骤2.1预估辨识过程的状态向量
步骤2.2预估估计误差的协方差矩阵Q
P
m|m-1=P
m-1|m-1+Q
其中,P
m|m-1为第m次辨识过程中所预估的估计误差的协方差矩阵;P
m-1|m-1为第m-1次辨识过程后所得到的估计误差的协方差矩阵。在初始状态下,令
P
1|0=P
0|0=pI
3
其中,I
3为单位矩阵;p为大于0的常数,如p=1。
随着改进的卡尔曼滤波方法递归过程的进行,估计误差的协方差矩阵将逐渐下降并收敛。
步骤2.3建立改进的卡尔曼增益表达式
K
m(β
m)=P
m|m-1G
T(GP
m|m-1G
T+(r+β
m)I
2)
-1
其中,K
m(β
m)为第m次辨识过程中改进的卡尔曼增益;β
m为第m次辨识过程中所采用的修正参数,且β
m>0。引入修正项β
mI
2是为了避免因为过程噪声和测量噪声的协方差矩阵Q和R估计不准确,造成逆矩阵(GP
m|m-1G
T+(r+β
m)I
2)
-1病态并产生接近于0的奇异值,进而造成辨识过程的发散。
步骤2.4利用改进的卡尔曼增益更新状态向量
步骤2.5更新估计误差的协方差矩阵
P
m|m=P
m|m-1-K
m(β
m)GP
m|m-1
其中,P
m|m为第m次辨识过程后所更新的估计误差的协方差矩阵。
步骤2.6重复步骤2.1-步骤2.5的递归过程,直至相邻两次状态向量的估计结果满足
其中,ε为人为给定的辨识精度的阈值向量。
步骤3.采用改进的L曲线法对改进的卡尔曼滤波方法中的修正参数进行优选。
W
X(β
m)=I
1-K
m(β
m)G,W
Y(β
m)=K
m(β
m)
可见,状态向量
是通过对其预估值
和实际观测信息ΔC
m之间按照一定的权重 进行分配后计算获得的,对应权重分别为W
X(β
m)和W
Y(β
m)。若修正参数β
m→0,不合理的过程噪声和测量噪声协方差矩阵Q和R将致使WY(βm)过高,从而状态向量
的更新过程更加信任于实际观测信息ΔC
m,而实际观测信息ΔC
m中的测量噪声将被改进的卡尔曼增益K
m(β
m)中病态的逆矩阵(G P
m|m-
1GT+(r+β
m)I
2)
-1过度放大,进而造成递归过程的发散。
结合正定矩阵、对称矩阵的性质、奇异值分解和谱分解等理论,可以证明随着修正参数β
m的增大,||K
m(β
m)ΔC
m||将逐渐减小,而||GK
m(β
m)ΔC
m-ΔC
m||将逐渐增大。前者表示当前递归过程在更新状态向量的估计值
时,改进的卡尔曼滤波算法对当前观测信息ΔCm的信任程度;后者为残差形式,表示没有被信任的观测信息量。曲线上的最大曲率点即为最佳平衡点,此时对应的修正参数β
m可以最小化算法的估计误差。
结合上述分析,采用改进的L曲线法对改进的卡尔曼滤波方法中的修正参数进行优选,过程如下:
步骤3.1按照间隔Δβ
m在区间[0,1×10
2]上对β
m进行采样,从而获得一系列的离散点,离散点的横纵坐标分别为
横坐标:γ(β
m)=||GK
m(β
m)ΔC
m-ΔC
m||,纵坐标:η(β
m)=||K
m(β
m)ΔC
m||。
步骤3.2利用三次样条插值法对离散点进行拟合,从而获得改进的L曲线
η(β
m)=f
CSI(γ(β
m))
其中,f
CSI表示三次样条插值后所获得改进的L曲线方程。
步骤3.3求解改进的L曲线上的最大曲率
其中,最大曲率点所对应的修正参数β
mi即为最佳修正参数。
步骤4.修正参数的自适应优选。
结合卡尔曼滤波方法递归过程中估计误差的协方差矩阵特性,建立修正参数的自适应优选方法。由于协方差矩阵Q和R为固定矩阵,故改进的卡尔曼增益K
m(β
m)的数值在递归过程中的变化是由估计误差的协方差矩阵P
m-1|m-1所决定的,当P
m-1|m-1收敛时β
m也应当逐渐收敛。由于估计误差的协方差矩阵的迹可用来衡量算法的收敛程度,故当m≥2时,采用自适应的方式获得最佳的修正参数β
m,即
其中,τ(m)表示第m次辨识过程中的惩罚因子。
步骤5.将高速并联机器人的零点误差辨识结果补偿至机器人的运动学模型中。
本发明所提出的一种高速并联机器人零点标定方法通过在以往基于传统卡尔曼滤波方法的零点辨识模型中引入修正项,进而能够解决零点辨识模型因系统过程噪声和测量噪声协方差矩阵估计不准确而造成的辨识结果发散问题,从而有效的保证了零点辨识过程的鲁棒性和准确性。
本发明所提出的二自由度高速并联机器人零点标定方法,通过改进的L曲线法对改进的卡尔曼滤波方法的零点辨识模型递归过程中的修正参数进行优选,提高了辨识结果的鲁棒性和准确性。结合传统卡尔曼滤波方法递归过程中估计误差的协方差矩阵特性,建立修正参数的自适应优选方法,从而有效提高机器人零点辨识过程的效率。
图1、是二自由度高速并联机器人的零点标定系统;
图2、是二自由度高速并联机器人的零点标定系统简化示意图;
图3、是改进的L曲线法示意图;
图4、是实验中的零点误差辨识效果对比图。
下面结合附图和具体实施案例对本发明加以详细说明。
以中国发明专利《一种两自由度高速并联机器人的零点标定方法》(ZL201410364282.3)中的标定系统为实施案例,本发明的具体实施方式如下:
1.建立机器人零点误差与末端运动误差之间的映射模型
如图1所示,二自由度高速并联机器人由静平台1、动平台2、第一运动支链3和第二运动支链4组成。静平台1和动平台2之间安装有一根伸缩尺7,伸缩尺7的两端分别与静平台1和动平台2通过转动副进行连接。如图2所示,通过伸缩尺7可以间接测量出动平台中心点P到基准坐标系O-xyz原点O之间的距离。
如图1和图2所示初始状态下,当该机器人的第一主动臂5和第二主动臂6与基准坐标系O-xyz的x轴平行时(即θ
1=θ
2=0°),此时机器人的零点误差Δθ
1=Δθ
2=0°;否则该机器人的零点误差Δθ
1和Δθ
2不为零。
利用空间矢量法和一阶摄动原理可以获得机器人末端运动的距离误差与机器人零点误差之间的映射模型,即
ΔC=GΔp
其中,ΔC为末端运动的距离误差矩阵;Δp为机器人零点误差向量;G为误差传递矩阵。
2.建立改进的卡尔曼滤波方法的机器人零点辨识模型
基于机器人零点误差和末端运动误差和之间的映射模型,可以得到辨识系统的状态方程和观测方程
状态方程:Δp
m=Δp
m-1+a
m-1
观测方程:ΔC
m=GΔp
m+b
m
其中,Δp
m和Δp
m-1分别为第m次和第m-1次辨识过程中系统的状态向量(即零点误差向量);ΔC
m为第m次辨识过程中通过测量所获得的机器人末端运动误差矩阵;a
m-1为第m-1次辨识过程中的过程噪声向量,该向量假设服从均值为0向量和协方差矩阵为Q的多元高斯分布;b
m为第m次测量过程中的测量噪声向量,该向量假设服从均值为0和协方差矩阵为R的多元高斯分布。
通常情况下,对协方差矩阵Q和R估计的准确性直接影响了卡尔曼滤波算法的效果。虽然测量噪声的协方差矩阵R在理想情况下可大致由测量仪器本身的测量误差参数给出,但实际中受工作环境变化或仪器本身不稳定性的影响,测量噪声的协方差矩阵R难以准确获得。此外,受伺服系统调节过程和减速器误差的影响,准确获得过程噪声协方差矩阵Q的先验信息将更加困难,故在此可将其简单设置为对角矩阵的形式,即
Q=qI
1,R=rI
2
其中,q和r为大于0的常数,在零点标定过程中q和r可取较小的数量级,如0~1×10
2;I
1和I
2分别为与协方差矩阵Q和R同阶的单位矩阵。
故,建立改进的卡尔曼滤波方法的零点辨识模型,其具体步骤如下:
2.1预估辨识过程的状态向量
2.2预估估计误差的协方差矩阵
P
m|m-1=P
m-1|m-1+Q
其中,P
m|m-1为第m次辨识过程中所预估的估计误差的协方差矩阵;P
m-1|m-1为第m-1次辨识过程后所得到的估计误差的协方差矩阵。在初始状态下,可令
P
1|0=P
0|0=pI
3
其中,I
3为单位矩阵;p为大于0的常数,可给定一个较大数值,如p=1。随着改进的卡尔曼滤波方法递归过程的进行,估计误差的协方差矩阵将逐渐下降并收敛。
2.3建立改进的卡尔曼增益表达式
K
m(β
m)=P
m|m-1G
T(GP
m|m-1G
T+(r+β
m)I
2)
-1
其中,K
m(β
m)为第m次辨识过程中改进的卡尔曼增益;β
m为第m次辨识过程中所采用的修正参数,且β
m>0。引入修正项β
mI
2是为了避免因为过程噪声和测量噪声的协方差矩阵Q和R估计不准确,造成逆矩阵(G P
m|m-1G
T+(r+β
m)I
2)
-1病态并产生接近于0的奇异值,进而造成辨识过程的发散。
2.4利用改进的卡尔曼增益更新状态向量
2.5更新估计误差的协方差矩阵
P
m|m=P
m|m-1-K
m(β
m)GP
m|m-1
其中,P
m|m为第m次辨识过程后所更新的估计误差的协方差矩阵。
2.6重复以上2.1~2.5的递归过程,直至相邻两次状态向量的估计结果满足
其中,ε为人为给定的辨识精度的阈值向量。
3.采用改进的L曲线法,对改进的卡尔曼滤波方法中的修正参数进行优选。
为提高零点辨识模型的鲁棒性和准确性,需对每一次递归过程中的修正参数β
m进行优选。本实施例建立了一种改进的L曲线,从而对修正参数βm进行优选:
W
X(β
m)=I
1-K
m(β
m)G,W
Y(β
m)=K
m(β
m)
可见,状态向量
是通过对其预估值
和实际观测信息ΔC
m之间按照一定的权重进行分配后计算获得的,对应权重分别为W
X(β
m)和W
Y(β
m)。若修正参数β
m→0,不合理的过程噪声和测量噪声协方差矩阵Q和R将致使W
Y(β
m)过高,从而状态向量
的更新过程更加信任于实际观测信息ΔC
m,而实际观测信息ΔC
m中的测量噪声将被改进的卡尔曼增益K
m(β
m)中病态的逆矩阵(G P
m|m-1G
T+(r+β
m)I
2)
-1过度放大,进而造成递归过程的发散。
如图3所示,结合正定矩阵、对称矩阵的性质、奇异值分解和谱分解等理论,可以证明随着修正参数β
m的增大,||K
m(β
m)ΔC
m||将逐渐减小,而||GK
m(β
m)ΔC
m-ΔC
m||将逐渐增 大。前者表示当前递归过程在更新状态向量的估计值
时,改进的卡尔曼滤波算法对当前观测信息ΔC
m的信任程度;后者为残差形式,表示没有被信任的观测信息量。曲线上的最大曲率点即为最佳平衡点,此时对应的修正参数β
m可以最小化算法的估计误差。
结合上述分析,改进的L曲线法的计算过程如下:
3.1按照一定的间隔Δβm在区间[0,1×10□2]上对βm进行采样,从而获得一系列的离散点,离散点的横纵坐标分别为
横坐标:γ(β
m)=||GK
m(β
m)ΔC
m-ΔC
m||,纵坐标:η(β
m)=||K
m(β
m)ΔC
m||。
3.2利用三次样条插值法对离散点进行拟合,从而获得改进的L曲线
η(β
m)=f
CSI(γ(β
m))
其中,f
CSI表示三次样条插值后所获得改进的L曲线方程。
3.3求解改进的L曲线上的最大曲率
其中,最大曲率点所对应的修正参数β
mi即为最佳的修正参数,需将其带入改进的卡尔曼滤波方法中进行计算。
4.结合卡尔曼滤波方法递归过程中估计误差的协方差矩阵特性,建立修正参数的自适应优选方法
由于协方差矩阵Q和R为固定矩阵,故改进的卡尔曼增益K
m(β
m)的数值在递归过程中的变化是由估计误差的协方差矩阵P
m-1|m-1所决定的,当P
m-1|m-1收敛时β
m也应当逐渐收敛。由于估计误差的协方差矩阵的迹可用来衡量算法的收敛程度,故当m≥2时,采用自适应的方式获得最佳的修正参数β
m,即
其中,τ(m)表示第m次辨识过程中的惩罚因子。
5.将机器人的零点误差辨识结果补偿至机器人的运动学模型中。
当通过辨识获得机器人的零点误差后,需将零点误差的辨识结果补偿至机器人的运动学模型中。如图1和图2所示,若二自由度并联机器人动平台2的中心点P位于工作空间中的任意处,假设机器人不存在零点误差(即Δθ
1=Δθ
2=0°),则第一主动臂5和第二主动臂6所对应的理论转角输出分别为θ
1和θ
2。然而,实际机器人的零点误差Δθ
1和Δθ
2通过辨 识得知并不为零,则此时机器人第一主动臂5和第二主动臂6的实际转角输出应当为
θ′
1=θ
1-△θ
1,θ′
2=θ
2-△θ
2
其中,θ′
1和θ′
2分别表示第一主动臂5和第二主动臂6的实际输出转角。
如图4所示,传统的基于卡尔曼滤波方法的零点辨识模型可能造成如图1所示的二自由度并联机器人零点辨识结果发散,而采用本发明则可以快速获得收敛的零点辨识结果。
Claims (1)
- 一种二自由度高速并联机器人零点标定方法,其步骤如下:步骤1.建立机器人末端运动误差与零点误差之间的映射模型二自由度高速并联机器人由静平台、动平台、第一运动支链和第二运动支链组成;静平台和动平台之间安装一根伸缩尺,通过伸缩尺间接测量出动平台中心点P到基准坐标系O-xyz原点O之间的距离;初始状态下,当该机器人的第一主动臂和第二主动臂与基准坐标系O-xyz的x轴平行时,即θ 1=θ 2=0°,此时机器人的零点误差Δθ 1=Δθ 2=0°;利用空间矢量法和一阶摄动原理获得机器人末端运动的距离误差与机器人零点误差之间的映射模型ΔC=GΔp其中,ΔC为末端运动的距离误差矩阵;Δp为机器人零点误差向量;G为误差传递矩阵;步骤2.建立改进的卡尔曼滤波方法的机器人零点辨识模型步骤2.1预估辨识过程的状态向量步骤2.2预估估计误差的协方差矩阵QP m|m-1=P m-1|m-1+Q其中,P m|m-1为第m次辨识过程中所预估的估计误差的协方差矩阵;P m-1|m-1为第m-1次辨识过程后所得到的估计误差的协方差矩阵;在初始状态下,令P 1|0=P 0|0=pI 3其中,I 3为单位矩阵;p为大于0的常数,如p=1;随着改进的卡尔曼滤波方法递归过程的进行,估计误差的协方差矩阵将逐渐下降并收敛;步骤2.3建立改进的卡尔曼增益表达式K m(β m)=P m|m-1G T(GP m|m-1G T+(r+β m)I 2) -1其中,K m(β m)为第m次辨识过程中改进的卡尔曼增益;β m为第m次辨识过程中所采用的修正参数,且β m>0;步骤2.4利用改进的卡尔曼增益更新状态向量步骤2.5更新估计误差的协方差矩阵P m|m=P m|m- 1-K m(β m)GP m|m-1其中,P m|m为第m次辨识过程后所更新的估计误差的协方差矩阵;步骤2.6重复步骤2.1-步骤2.5的递归过程,直至相邻两次状态向量的估计结果满足其中,ε为人为给定的辨识精度的阈值向量;步骤3.采用改进的L曲线法对改进的卡尔曼滤波方法中的修正参数进行优选步骤3.1按照间隔Δβ m在区间[0,1×10 2]上对β m进行采样,从而获得一系列的离散点,离散点的横纵坐标分别为横坐标:γ(β m)=||GK m(β m)ΔC m-ΔC m||,纵坐标:η(β m)=||K m(β m)ΔC m||;步骤3.2利用三次样条插值法对离散点进行拟合,从而获得改进的L曲线η(β m)=f CSI(γ(β m))其中,f CSI表示三次样条插值后所获得改进的L曲线方程;步骤3.3求解改进的L曲线上的最大曲率其中,最大曲率点所对应的修正参数β mi即为最佳修正参数;步骤4.修正参数的自适应优选采用自适应的方式获得最佳的修正参数β m其中,τ(m)表示第m次辨识过程中的惩罚因子;步骤5.将高速并联机器人的零点误差辨识结果补偿至机器人的运动学模型中。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811553685.7 | 2018-12-19 | ||
CN201811553685.7A CN109291056B (zh) | 2018-12-19 | 2018-12-19 | 一种二自由度高速并联机器人零点标定方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2020125195A1 true WO2020125195A1 (zh) | 2020-06-25 |
Family
ID=65142980
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2019/112852 WO2020125195A1 (zh) | 2018-12-19 | 2019-10-23 | 一种二自由度高速并联机器人零点标定方法 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN109291056B (zh) |
WO (1) | WO2020125195A1 (zh) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109291056B (zh) * | 2018-12-19 | 2021-08-17 | 南京埃斯顿自动化股份有限公司 | 一种二自由度高速并联机器人零点标定方法 |
CN114734440B (zh) * | 2022-04-15 | 2023-09-05 | 同济大学 | 一种混联双臂搬运机器人运动学参数精准标定方法 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2002091522A (ja) * | 2000-07-13 | 2002-03-29 | Okuma Corp | 作業機の誤差補正方法 |
CN104354166A (zh) * | 2014-07-28 | 2015-02-18 | 天津大学 | 一种三自由度并联机器人的零点标定方法 |
CN106239510A (zh) * | 2016-08-25 | 2016-12-21 | 芜湖瑞思机器人有限公司 | 一种三自由度并联机器人零点标定方法 |
CN106272412A (zh) * | 2016-08-25 | 2017-01-04 | 芜湖瑞思机器人有限公司 | 一种齿轮齿条式四自由度并联机器人的零点标定方法 |
CN109291056A (zh) * | 2018-12-19 | 2019-02-01 | 南京埃斯顿自动化股份有限公司 | 一种二自由度高速并联机器人零点标定方法 |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
FR2638115A1 (fr) * | 1988-10-26 | 1990-04-27 | Quintel | Procede de calage d'un bras de robot et robots utilisant ce procede |
CN104385281B (zh) * | 2014-07-28 | 2016-07-06 | 天津大学 | 一种两自由度高速并联机器人的零点标定方法 |
-
2018
- 2018-12-19 CN CN201811553685.7A patent/CN109291056B/zh active Active
-
2019
- 2019-10-23 WO PCT/CN2019/112852 patent/WO2020125195A1/zh active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2002091522A (ja) * | 2000-07-13 | 2002-03-29 | Okuma Corp | 作業機の誤差補正方法 |
CN104354166A (zh) * | 2014-07-28 | 2015-02-18 | 天津大学 | 一种三自由度并联机器人的零点标定方法 |
CN106239510A (zh) * | 2016-08-25 | 2016-12-21 | 芜湖瑞思机器人有限公司 | 一种三自由度并联机器人零点标定方法 |
CN106272412A (zh) * | 2016-08-25 | 2017-01-04 | 芜湖瑞思机器人有限公司 | 一种齿轮齿条式四自由度并联机器人的零点标定方法 |
CN109291056A (zh) * | 2018-12-19 | 2019-02-01 | 南京埃斯顿自动化股份有限公司 | 一种二自由度高速并联机器人零点标定方法 |
Also Published As
Publication number | Publication date |
---|---|
CN109291056B (zh) | 2021-08-17 |
CN109291056A (zh) | 2019-02-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2018188276A1 (zh) | 一种六自由度机器人末端空间曲线轨迹的误差建模方法 | |
CN106777656B (zh) | 一种基于pmpsd的工业机器人绝对精度校准方法 | |
CN109579824B (zh) | 一种融入二维码信息的自适应蒙特卡诺定位方法 | |
CN103389506B (zh) | 一种用于捷联惯性/北斗卫星组合导航系统的自适应滤波方法 | |
WO2020125195A1 (zh) | 一种二自由度高速并联机器人零点标定方法 | |
CN108908327A (zh) | 一种机器人定位误差分级补偿方法 | |
CN113459094B (zh) | 一种工业机器人工具坐标系及零点自标定方法 | |
CN110900610B (zh) | 一种基于lm算法和粒子滤波算法优化的工业机器人标定方法 | |
CN109129482B (zh) | 一种动态补偿机器人直线导轨运动误差的方法 | |
CN114147726A (zh) | 一种几何误差与非几何误差相结合的机器人标定方法 | |
JP2774939B2 (ja) | ロボットのツールパラメータ導出方法及び較正方法 | |
CN115420305B (zh) | 基于采样点权重自适应分配的三轴磁传感器误差补偿方法 | |
CN112798021A (zh) | 基于激光多普勒测速仪的惯导系统行进间初始对准方法 | |
CN111949036B (zh) | 一种轨迹跟踪控制方法、系统及两轮差速移动机器人 | |
CN115502968A (zh) | 一种基于标定修复刚柔耦合模型的机械臂末端位置误差补偿方法 | |
CN113799137A (zh) | 一种基于神经网络的移动加工机器人快速视觉定位方法 | |
CN107421476A (zh) | 一种空间孔位测量基准误差补偿方法 | |
CN117331309A (zh) | 一种基于超扭曲滑模自适应导纳控制的机器人抛光方法 | |
Rigatos | Control and disturbances compensation in underactuated robotic systems using the derivative-free nonlinear Kalman filter | |
CN105425582A (zh) | 一种基于卡尔曼滤波的Stewart机构在线标定方法 | |
CN109397293B (zh) | 一种基于移动机器人的地面水平误差建模及补偿方法 | |
CN115070731B (zh) | 一种面向并联机构的几何误差标定方法、系统和电子设备 | |
Lin et al. | Pose estimation in industrial machine vision systems under sensing dynamics: A statistical learning approach | |
CN114034290B (zh) | 放样机器人系统的放样方法 | |
CN113043271B (zh) | 一种基于天牛须算法的工业机器人标定补偿方法 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19898751 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 19898751 Country of ref document: EP Kind code of ref document: A1 |