Metadata of the chapter that will be visualized in

SpringerLink
Book Title Intelligent Computing Methodologies
Series Title
Chapter Title A New Robotic Manipulator Calibration Method of Identification Kinematic and Compliance Errors
Copyright Year 2020
Copyright HolderName The Author(s), under exclusive license to Springer Nature Switzerland AG
Author Family Name Le
Particle
Given Name Phu-Nguyen
Prefix
Suffix
Role
Division Graduate School of Electrical Engineering
Organization University of Ulsan
Address Ulsan, 680-749, South Korea
Email
Corresponding Author Family Name Kang
Particle
Given Name Hee-Jung
Prefix
Suffix
Role
Division School of Electrical Engineering
Organization University of Ulsan
Address Ulsan, 680-749, South Korea
Email [email protected]

Abstract In this work, a new robotic calibration method is proposed for reducing the positional errors of the robot
manipulator. First, geometric errors of a robot are identified by using a conventional kinematic calibration
model of the robot. Then, a radial basis function is constructed for compensating the compliance errors
based on the effective torques for further increasing the positional precision of the robot. The enhanced
positional accuracy of the robot manipulator in experimental studies that are carried on a YS100 robot
illustrates the advantages of the suggested algorithm than the other techniques.
Keywords Robot accuracy - Radial basis function - Robot calibration
 A New Robotic Manipulator Calibration
Author Proof

Method of Identification Kinematic

and Compliance Errors

Phu-Nguyen Le1 and Hee-Jung Kang2(&)

1
Graduate School of Electrical Engineering, University of Ulsan,
Ulsan 680-749, South Korea
2
School of Electrical Engineering, University of Ulsan, Ulsan 680-749,
South Korea
[email protected]

Abstract. In this work, a new robotic calibration method is proposed for

reducing the positional errors of the robot manipulator. First, geometric errors of
a robot are identified by using a conventional kinematic calibration model of the
robot. Then, a radial basis function is constructed for compensating the com-
pliance errors based on the effective torques for further increasing the positional
precision of the robot. The enhanced positional accuracy of the robot manipu-
lator in experimental studies that are carried on a YS100 robot illustrates the
advantages of the suggested algorithm than the other techniques.

Keywords: Robot accuracy
Radial basis function
Robot calibration

1 Introduction

The robot manipulators are widely used in the industry. Although the robots are high
repeatability, they are well-known by their low accuracy [1, 2]. The errors of the robot
end-effector mostly come from geometric errors and non-geometric errors. The geo-
metric errors are the results of misalignments, incorrect in manufacturing, and assembly
robot. The non-geometric errors may come from many non-geometric sources, such as
joint and link compliance, temperature variation, gear transmission, etc. Among the
non-geometric errors, the compliance errors are dominant. These errors are caused by
the flexibility of joints and links under the link self-gravity and external payload.
Geometric calibration methods are widely examined and become mature. The most
famous kinematic calibration method, the D-H model is suggested by Denavit-
Hartenberg [3–5]. This method is widely used in kinematic calibration by many
researchers recently [6–8]. Moreover, the other geometric calibration methods are CPC
model [9, 10], POE model [11, 12] and the zero-reference position method [13, 14].
However, these calibration methods do not consider the non-geometric errors. On the
other hand, some studies used another approach to investigate joint compliance errors
[15, 16]. However, these methods neglected effect of the geometric errors.
Some works have been proposed to deal with kinematic and compliance calibra-
tion. For instance, a calibration method to calibrate the geometric errors and

© Springer Nature Switzerland AG 2020

D.-S. Huang and P. Premaratne (Eds.): ICIC 2020, LNAI 12465, pp. 1–12, 2020.
https://doi.org/10.1007/978-3-030-60796-8_2
 2 P.-N. Le and H.-J. Kang

compensate the joint by radial basis function (RBF) [17] is proposed by Jang et al.
Author Proof

However, the work [17] focused on calibrating the geometric parameters and com-
pensating the compliance errors by compensating the joint (“joint level” calibration
[1]). The work also needs to divide the robot working space into many subspaces and
required many measurements and consuming a lot of time. Meggiolaro et al. proposed
a method to approximate the compliance errors by a polynomial function of joint
parameters and wrench using torque sensors [18]. Zhou and Hee-Jung proposed a
method to simultaneously calibrate the geometric and joint stiffness parameters of the
robot [19]. However, this method linearized the relationship between effective torques
and joint compliance errors. Recently, some studies have been performed on joint
stiffness calibration [20–22] with the need of the torques sensors.
This study proposed a new calibration algorithm for robotic manipulators. The
method includes the kinematic calibration and non-geometric compensation with a
RBF compensator that compensates for compliance errors based on the effective tor-
ques. It is assumed that the gravity compensation torques are nonlinearity related to the
compliance errors. These relationships can be constructed by a RBF. The advantages of
the suggested method are easy for implementing, removing the need for torque sensors,
high ability to enhance the precision of the manipulator. These advantages are firmly
confirmed by the experimental studies in contrasting with 2 other methods such as the
conventional kinematic calibration and the method for simultaneously calibrate the
geometric and joint stiffness parameters of the robot.
Following the introduction. Section 2 presents the kinematic model of the YS 100
robot. In Sect. 3, the geometrical and the gravity compensator using a Radial basis
function that is based on the effective torques are presented. Sections 4 is devoted to
the experimental calibration result of the proposed method in contrasting with other
methods. Section 5 summarizes the abilities and advantages of the proposed method.

2 Kinematic Model of the YS100 Robot

YS100 is a 6 DOF serial robot [19]. The kinematic structure of it is briefly described in
Fig. 1 and Table 1.

Fig. 1. Kinematic structure of the YS 100 robot.

 A New Robotic Manipulator Calibration Method 3

The transformation that relates base frame {0} to tool frame{T}:

Author Proof

ET ¼ 01 T(h 1 )12 T(h2 )23 T(h3 )34 T(h4 )45 T(h5 )56 T(h6 )6E T ð1Þ
0

The end-effector transformation:

ð6T ÞT ¼ TrX ða6 ÞTrY ðb6 ÞTrZ ðdT Þ ð2Þ

Table 1. Nominal D-H parameters of the Hyundai robot YS100.

i ai−1(deg) ai−1(m) bi−1(deg) bi−1(m) di(deg) hi(deg)
1 0 0 0 0 0.48 h1
2 90 0.32 – – 0 h2
3 0 0.87 0 – 0 h3
4 90 0.2 – – 1.03 h4
5 −90 0 – – 0 h5
6 90 0 – – 0.185 h6
T – 0.2 – 0.05 0.5 –

3 Identification Kinematic Parameters and Compliance

Compensation Based on the Effective Torques Using
a Radial Basis Function

Assuming that the robot’s end-effector position Preal is calculated by the following
equation:

Preal ¼ Pkin þ DPkin þ DPc þ DPextra ð3Þ

where Pkin is the position of the end effector calculated by the kinematic parame-
ter,DPkin is the position error caused by the geometric error, DPc is the position error
due to the joint compliance, and DPextra is the positional residual error that is not
modeled. Assuming that geometric errors and joint deflection errors are the main parts
in causing the position errors Preal (DPextra ¼ 0). The error model can be expressed as:

DP ¼ DPkin þ DPc ¼ Preal  Pkin ð4Þ

The position errors caused by geometric errors DPkin in the Eq. 4 could be iden-
tified by the conventional kinematic calibration [3–5]. DPkin can be expressed as

DPkin ¼ Jkin D/ ð5Þ

where Jkin (3  n) is a kinematic Jacobian matrix [1, 19]. D/ is a n  1 kinematic

parameter error vector. n is the number of the calibrated kinematic parameters. The
total number of kinematic parameters is equal to 32. However, the 6 DOF revolute
 4 P.-N. Le and H.-J. Kang

robot has several dependencies between some parameters. These dependency param-
Author Proof

eters are {Dh1 ; Dh0 }, {Dd1 ; Dd0 }, {Dd3 ; Dd2 }, {DzT ; Dd6 }, {ðDxT ; DyT Þ; Dh6 }. In each
pair, the parameter errors cannot be identified together. Therefore, the dependency
parameters that are chosen to calibrate are {Dh1 , Dd1 , Dd3 , DxT , DyT , DzT } while the
other error parameter in each pair is set to the nominal parameter value. So, the number
of calibrated kinematic is reduced to 27.
The Eq. 5 can be solved by the least-square method to overcome the effect of noise
and uncertainty:

D/ ¼ ½ðJT J)1 JT DP ð6Þ

The positional error DP is calculated by

DP ¼ Pm  Pkin ð7Þ

where Pm is the measured position vector and Pkin is the computed position vector by
the recent kinematic parameters. The Eq. (6) is employed repetitive until the geometric
parameters converge. Through the kinematic calibration process, the Pkin converges to
the Pckin value. The position errors of the robot end-effector after kinematic calibration
process are calculated by:

DPres ¼ Pm  Pckin ð8Þ

Assuming that the position errors due to joint deflection errors are the main parts in
causing these residual position errors (DPres ¼ DPc ). The joint deflections under link
self-gravity and external payload are also assumed to be dominant in causing com-
pliance errors. Therefore, the joint deflection errors can be calculated from the related
effective torque of joints.
It should be noted that previous literatures [15, 19] constructed the compliance
errors by linearizing the relationship of the effective torques and the joint compliances.
However, there are some residual errors that could not be neglected caused by the
nonlinear relation between joint torques and joint deflections. For further enhanced the
robot precision, the relationship of the effective torque and the residual errors is con-
structed by a RBF in this paper. The RBF has 6 inputs that represent the total effective
torque in 6 robot joints, 40 nodes in the hidden layer, and 3 nodes in the output layer
that represent three elements of the position error vector.
The total effective torques in the robot jth joint under related gravity forces are given
as:

X
N þ1 X
N þ1
si ¼ si;j ¼ JhT Fj ð9Þ
i;j
j¼i j¼i

where N = 6 is the number of DOF of the robot and FN+1 is the gravity force due to
payload Here, the gravity force accompanying to jth link is calculated by
 A New Robotic Manipulator Calibration Method 5

Fj ¼ ½ 0 Mj g  ð10Þ
Author Proof

0
where Mj is the mass of the jth link and g is the gravity coefficient. The transpose of the
Jacobian matrix is used as a force transformation to find the effective joint torques si,j in
the ith joint due to the gravity force in the jth link. The Jacobian matrix is defined as

Jhi;j ¼ zi  li;j ð11Þ

where li;j is the 3  1 vector between the origin of the ith frame and the mass center of
the jth link.
The total effective torques are set to be the input of the RBF. Figure 2 shows the
structure of the RBF. The output of the hidden node i in the RBF layer is calculated as
follow:

Fig. 2. Structure of the RBF.

oj ¼ en
2
ð12Þ

where n is a transfer function that describes the vector distance between the weight
vector wi and the input vector p, multiplied by the bias bi.

n ¼ kwi  pkbi ð13Þ

The output layer is a linear function with 3 nodes in the output layer that represent
three elements of the position error vector.
The output of the RBF is used to compensate for the compliance error(which is
assumed to be the residual error DPres ¼ DPc ). Therefore, the residual error after
compensated by the RBF is calculated by:
 6 P.-N. Le and H.-J. Kang

e ¼ DPres  Pnn ð14Þ

Author Proof

In this work, the weights and bias of the RBF are trained by the MATLAB toolbox
that creates a two-layer network. The hidden layer is the RBF layer (Eq. 12, 13). The
output layer is a linear layer. At the beginning, there are no neurons in the hidden layer.
The learning process is carried out following the steps below:
• Run the network and find the input vector with the greatest error.
• A RBF neuron is added with weights equal to that vector.
• The linear layer weights are redesigned to minimize error.
• Repeat until convergence.
In order to keep the RBF layer from increasing too much, the number of nodes in
this layer are limited at 40 nodes. Overall, the suggested method could be described in
the following flowchart (Fig. 3). AQ1

Fig. 3. Flowchart of the proposed method.

4 Experiment and Results

The experimental system is shown in Fig. 4. The 6 DOF robot manipulator (YS100). In
this work, the mass of link jth Mj (Eq. 10) is provided by the robot’s manufacturer. The
external payload weight is 110 kg. Therefore, the weight matrix is descried as follow:

M ¼ ½ 196:7 79:25 170:27 10:58 22:33 2:0 110  ð15Þ

An API laser tracker (accuracy of 0.01 mm/m, repeatability of ±0.006 mm/m) and
an accompanying laser reflector are used to perform the calibration process. The
proposed method (RBF-TCM) is used to calibrate the YS100 robot to show the
advantage of the method in comparing with 2 others methods including the kinematic
calibration method (KM) [3–5], the simultaneous identification of joint compliance and
kinematic parameters methods (SKCM) [19] in the experimental study.
 A New Robotic Manipulator Calibration Method 7
Author Proof

Fig. 4. Experimental setup.

4.1 Experimental Calibration

The robot configuration data are randomly collected in the working space and classified
into 2 sets. Set Q1 including 50 robot configurations is employed in the calibration
process and the other set of 50 robot configurations (Q2) is used in the validation
process. By using the conventional calibration method (Eq. 6), 27 geometric param-
eters are identified. The results are demonstrated in Table 2. The residual errors and the
computed torques are used for training the RBF to determine the weights and bias of
the RBF. It should be noted here the reason why the RBF is used in this working rather
than the conventional fed forward neural network. In the conventional feedforward
neural network, the sigmoid neurons can have outputs over a large region of the input
space, while radial basis neurons only respond to relatively small regions of the input
space [23]. Therefore, the RBF could be said to be more stable in responding to noises
and uncertainties inputs. However, the drawback of this method is that the larger the
input space the more radial basis neurons are required [24]. The experimental cali-
bration processes are carried out by 3 different calibration methods such as conven-
tional kinematic calibration, SKCM, and RBF-TCM. The results of these calibration
methods are shown in Fig. 5 and Table 3.
 8 P.-N. Le and H.-J. Kang

Table 2. D-H parameters of the Hyundai robot YS100.

Author Proof

i ai−1(deg) ai−1(m) bi−1(deg) bi−1(m) di(deg) hi(deg)

1 −0.1646 −0.016 0.4748 0.0493 0.4851 −0.3316
2 90.0578 0.3199 – – 0 1.1517
3 0.0004 0.8704 0.0681 – -0.0036 −1.6686
4 89.9919 0.2001 – – 1.0272 −1.2079
5 −90.121 0.0003 – – −0.0017 −0.0017
6 89.9656 −0.0035 – – 0.185 −1.8226
T – −0.28 – 0.0469 0.4219 –

Fig. 5. Residual errors of the YS100 robot after calibration.

Table 3. The absolute position accuracy of the YS100 robot (Calibration).

Mean (mm) Maximum (mm) Std. (mm)
Nominal robot model 13.5527 30.5911 6.0528
KM 0.6894 1.9318 0.4015
SKCM 0.6065 1.6811 0.3488
Proposed method 0.2785 0.9332 0.2095

The calibration results show that the precision of the robot after calibrated by the
proposed method is dramatically reduced. By employing the RBF-TCM method, the
position errors are lower than the results by other methods. In comparing to the con-
ventional kinematic calibration method, the proposed method reduces the mean of
position errors from 0.6894 mm to 0.2785 mm (precise increasing by 59.6%). It also
increases the accuracy by 54.08% in comparison to the results generated by the SKCM
 A New Robotic Manipulator Calibration Method 9

method (from 0. 6065 mm to 0.2785 mm). The suggested algorithm also generates the
Author Proof

lowest maximum position error (0.9332 mm), and the lowest standard deviation
(0.2095 mm).

4.2 Experimental Validation Results

The proposed method should be validated by another robot configuration to demon-
strate the ability of it over the working space. The robot configuration set Q2 that is
totally different from Q1 is hired for the validation process with 3 different methods.

Fig. 6. Residual errors of the YS100 robot after validation.

Table 4. The absolute position accuracy of the YS100 robot (Validation).

Mean (mm) Maximum (mm) Std. (mm)
Nominal robot model 14.1106 32.3303 5.9835
KM 0.7245 1.7584 0.3814
SKCM 0.6398 1.7031 0.3214
Proposed method 0.2802 0.7846 0.2084

By employing the method, the position errors are lower than the results by other
methods in the validation process (Table 4 and Fig. 6). In comparing to the conven-
tional kinematic calibration method, the proposed method reduces the mean of position
errors from 0.7245 mm to 0.2802 mm (precise increasing by 61.33%). It also increases
the accuracy by 56.21% in comparison to the results generated by the SKCM method
(0.6398 mm to 0.2802 mm). The suggested algorithm also generates the lowest
maximum position error (0.7846 mm), and the lowest standard deviation (0.2084 mm).
 10 P.-N. Le and H.-J. Kang

4.3 Discussion and Future Studying

Author Proof

In previous literatures [15, 19], the relationship of joint deflections of robot and the
effective torques are linearized:

Dhc ¼ sC ð16Þ

where Dhc is the N  1 joint deflection vector, s is the diagonal effective torque matrix,
C is the N  1 joint compliance vector. Then, the Cartesian position errors due to the
joint compliances can be modeled as:

DPc ¼ Je Dhc ¼ Je sC ð17Þ

where Je is vector of joint compliance parameters that is computed by the following

published work [25]. Je s is the transformation matrix relating the joint compliance
parameters and the deflections of robot end-effector. It should be noted here that the
effective torques in the ith joint is not only due to the gravity force related to the ith link
but also due to the gravity forces related to both the link after and the external load. In
this work, the relationship of the effective torque s and the positional errors due to the
compliance errors DPc could be constructed by a RBF for higher increasing the pre-
cision of the robot.
The work will be expanded in the future by implementing the optimizing method to
select the calibration poses for better calibration results.

5 Conclusion

In this work, a new robotic calibration method is proposed for reducing the positional
errors of the robot manipulator. First, geometric errors of a robot are identified by using
a conventional kinematic calibration model of the robot. Then, a radial basis function is
constructed for compensating the compliance errors based on the effective torques for
further increasing the positional precision of the robot. By using a RBF, the relation-
ship of the effective torque and the compliance errors is constructed for higher
increasing the precision of the robot. The advantages of the suggested method are easy
for implementing, removing the need for torque sensors, high ability to enhance the
precision of the manipulator. These advantages are firmly confirmed by the experi-
mental studies on a YS100 robot in contrasting with 2 other methods such as the
conventional kinematic calibration and the method for simultaneously calibrate the
geometric and joint stiffness parameters of the robot.

Acknowledgment. This research was supported by 2020 Research Fund of University of Ulsan,
Ulsan, Korea.
 A New Robotic Manipulator Calibration Method 11

References
Author Proof

1. Mooring, B.W., Roth, Z.S., Driels, M.R.: Fundamentals of Manipulator Calibration. Wiley,
New York (1991)
2. Whitney, D.E., Lozinski, C.A., Rourke, J.M.: Industrial robot forward calibration method
and results. J. Dyn. Syst. Meas. Control 108, 1–8 (1986)
3. John, J.C., et al.: Introduction to Robotics: Mechanics and Control. Addison-Wesley, Read
(1989)
4. Hayati, S., Mirmirani, M.: Improving the absolute positioning accuracy of robot
manipulators. J. Robot. Syst. 2, 397–413 (1985)
5. Hayati, S., Tso, K., Roston, G.: Robot geometry calibration. In: Proceedings of 1988 IEEE
International Conference on Robotics and Automation, pp. 947–951 (1988)
6. Klug, C., Schmalstieg, D., Gloor, T., Arth, C.: A complete workflow for automatic forward
kinematics model extraction of robotic total stations using the Denavit-Hartenberg
convention. J. Intell. Robot. Syst. 95, 311–329 (2019)
7. Faria, C., Vilaça, J.L., Monteiro, S., Erlhagen, W., Bicho, E.: Automatic Denavit-Hartenberg
parameter identification for serial manipulators. In: IECON 2019-45th Annual Conference of
the IEEE Industrial Electronics Society, pp. 610–617 (2019)
8. Morar, C.A., Hăgan, M., Doroftei, I., Marinca, Ş.: Analog matrix multiplier dedicated to the
Denavit-Hartenberg algorithm. In: 2019 International Symposium on Signals, Circuits and
Systems (ISSCS), pp. 1–4 (2019)
9. Zhuang, H., Roth, Z.S., Hamano, F.: A complete and parametrically continuous kinematic
model for robot manipulators. In: Proceedings of the IEEE International Conference on
Robotics and Automation, pp. 92–97 (1990)
10. Zhuang, H., Wang, L.K., Roth, Z.S.: Error-model-based robot calibration using a modified
CPC model. Robot. Comput. Integr. Manuf. 10, 287–299 (1993)
11. Okamura, K., Park, F.C.: Kinematic calibration using the product of exponentials formula.
Robotica 14, 415–421 (1996)
12. Chen, G., Kong, L., Li, Q., Wang, H., Lin, Z.: Complete, minimal and continuous error
models for the kinematic calibration of parallel manipulators based on POE formula. Mech.
Mach. Theory 121, 844–856 (2018)
13. Gupta, K.C.: Kinematic analysis of manipulators using the zero reference position
description. Int. J. Rob. Res. 5, 5–13 (1986)
14. Cheng, L.-P., Kazerounian, K.: Study and enumeration of singular configurations for the
kinematic model of human arm. In: Proceedings of the IEEE 26th Annual Northeast
Bioengineering Conference (Cat. No. 00CH37114), pp. 3–4 (2000)
15. Dumas, C., Caro, S., Garnier, S., Furet, B.: Joint stiffness identification of six-revolute
industrial serial robots. Robot. Comput. Integr. Manuf. 27, 881–888 (2011)
16. Slavković, N.R., Milutinović, D.S., Kokotović, B.M., Glavonjić, M.M., Živanović, S.T.,
Ehmann, K.F.: Cartesian compliance identification and analysis of an articulated machining
robot. FME Trans. 41, 83–95 (2013)
17. Jang, J.H., Kim, S.H., Kwak, Y.K.: Calibration of geometric and non-geometric errors of an
industrial robot. Robotica 19, 311–321 (2001)
18. Meggiolaro, M.A., Dubowsky, S., Mavroidis, C.: Geometric and elastic error calibration of a
high accuracy patient positioning system. Mech. Mach. Theory 40, 415–427 (2005)
19. Zhou, J., Nguyen, H.-N., Kang, H.-J.: Simultaneous identification of joint compliance and
kinematic parameters of industrial robots. Int. J. Precis. Eng. Manuf. 15, 2257–2264 (2014)
 12 P.-N. Le and H.-J. Kang

20. Kamali, K., Joubair, A., Bonev, I.A., Bigras, P.: Elasto-geometrical calibration of an
Author Proof

industrial robot under multidirectional external loads using a laser tracker. In: 2016 IEEE
International Conference on Robotics and Automation (ICRA), pp. 4320–4327 (2016)
21. Müller, R., Scholer, M., Blum, A., Kanso, A.: Identification of the dynamic parameters of a
robotic tool based on integrated torque sensors. In: 2019 23rd International Conference on
Mechatronics Technology (ICMT), pp. 1–6 (2019)
22. Besset, P., Olabi, A., Gibaru, O.: Advanced calibration applied to a collaborative robot. In:
2016 IEEE International Power Electronics and Motion Control Conference (PEMC),
pp. 662–667 (2016)
23. Xia, C., Liu, Y., Lei, B., Xiang, X.: Research on a generalized regression neural network
model of thermocouple and it’s spread scope. In: 2008 Fourth International Conference on
Natural Computation, pp. 109–113 (2008)
24. Corino, V.D.A., Matteucci, M., Cravello, L., Ferrari, E., Ferrari, A.A., Mainardi, L.T.: Long-
term heart rate variability as a predictor of patient age. Comput. Methods Programs Biomed.
82, 248–257 (2006)
25. Nakamura, Y., Ghodoussi, M.: Dynamics computation of closed-link robot mechanisms with
nonredundant and redundant actuators. Int. Conf. Robot. Autom. 5, 294–302 (1989)
Author Proof

Author Query Form

Book ID : 505930_1_En
Chapter No : 2

Please ensure you fill out your response to the queries raised below
and return this form along with your corrections.

Dear Author,
During the process of typesetting your chapter, the following queries have
arisen. Please check your typeset proof carefully against the queries listed below
and mark the necessary changes either directly on the proof/online grid or in the
‘Author’s response’ area provided below

Query Refs. Details Required Author’s Response

AQ1 Please check and confirm if the inserted citation of Fig. 3 is correct. If not, please
suggest an alternate citation.