Adaptive Impedance Control Method For Industrial Manipulator Writing Based On Kalman Filter.

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Proceedings of the 37th Chinese Control Conference

July 25-27, 2018, Wuhan, China

Adaptive Impedance Control Method for Industrial Manipulator Writing


Based on Kalman Filter
Qidan Zhu1, Xinru Xie1, Chao Li1, Guihua Xia1
1. College of Automation, Harbin Engineering University, Harbin 150001, China
E-mail: xiexinru@hrbeu.edu.cn

Abstract: This paper presents an adaptive impedance control method based on Kalman filter to write Chinese characters for
industrial manipulator. Kalman filter is used to estimate the actual contact force. For safety, the contact force is controlled to
maintain constant using the adaptive impedance control method. There is no need for the knowledge of the environment. A robot
Chinese characters writing system is designed to write characters on a stiff board. The writing experiment results verified the
performance of the proposed method, and the trajectory and force can be tracked simultaneously.
Key Words: characters writing, adaptive impedance control, Kalman filter, industrial manipulator

method, such as hybrid position/force control [7] or


1 Introduction impedance control [8], is needed to accomplish these tasks.
Manipulators have been used in many fields to accomplish The contact force between the EE of the robot and
repetitive or dangerous tasks. With the increase of environment is measured using a Force/Torque (F/T) sensor,
applications in unstructured environment, robots are which is attached to the robot wrist. Usually, the measured
required to interact with the environment autonomously. values of the F/T sensor are not the actual contact force and
Handwriting is a typical human motion [1], and it is a highly torque. Besides, in the industrial environment, the signals
complex and challenging task that need to contact with the sampled directly from the F/T sensor always contain strong
environment. Consequently, writing characters on a stiff noises, such as thermal noise, environment noise and so on,
board, especially for Chinese characters, is a challenging which degrade the performance of the force control. The
task for robots, because the safe contact force should be Kalman filter algorithm [7] could be used to estimate the
considered. actual contact force and torque. Liping YAN [9] present a
Recently, many studies have been focused on robot linear transformation to decouple the cross-correlated sensor
writing. Xiaobo Gong [2] realized drawing the Chinese noises into cross-independent and established new
characters on a given sphere based on continuous path measurement models.
control. Guohua Gao [3] developed a kind of writing Actually, the position and the stiffness of the environment
function of 4-DOF binocular vision manipulator with are usually unknown, and they have influence on the control
flexible algorithm, and it allows a common robot to write at performance. Many control method are studied to adapt the
high accuracy. Fei Chao [1] introduced a robot writing uncertain contact environment, such as neural network based
approach through human-robot interactions. Chinese force control [10], sliding mode hybrid impedance control
character’s strokes are extracted from the captured human [11], adaptive impedance control [12, 13], robust impedance
demonstrator’s arm trajectories. Chenguang Yang [4] control [14], adaptive neural impedance control [15, 16],
presented a control method on the Baxter robot to imitate observer-based impedance control [17].
human writing skills using electromyography(EMG) signals In this paper, we present an adaptive impedance control
successfully. Peidong Liang [5] used surface EMG (sEMG) method based on Kalman filter to write the Chinese
for writing skill, the joint stiffness extracted from sEMG characters using the 6-DOF industrial manipulator. The
signals is transferred to the robot to complete writing tasks. Kalman filter is applied to estimate the actual contract force.
Rongchuan YU [6] designed a writing robot control system The adaptive impedance control is robust to disturbances
based on the CoreXY structure, and the jitter problem was and impact forces. There is no need for the environment
solved using the improved genetic algorithm. knowledge, the presented method could track the outline of
All studies mentioned above made great contributions for the Chinese characters accurately and control the contact
robot writing. However, the contact force between the force safely. The experiments of writing the Chinese
end-effectror(EE), i.e. the pencil, and the environment has characters on a stiff board verified the performance of the
not been considered. If the robot writes on a stiff board, the proposed method. This method could be applied to
contact force should be controlled within a proper range to accomplish other interaction tasks that need tracking the
guarantee the safety. It is challenging for characters writing trajectory and controlling force simultaneously.
task to follow a desired trajectory while keeping a constant The paper is organized as follows: Section 2 estimates the
force with the environment. Therefore, compliant control actual contact force between the robot EE and the
environment using Kalman filter algorithm. Section 3
reviews the conventional impedance control model, and
*This work is supported by National Natural Science Foundation of
analyzes the steady state error. An adaptive impedance
China and the China Academy of Engineering Physics(NSAF) under Grant
conrol scheme based Kalman filter is developed. The
No.U1530119, and Discipline Construction Comprehensive Pilot Special Chinese characters writing process of robot is introduced in
Funding of College of Automation of Harbin Engineering University.

5610
Section 4. Robot writing experiments are designed in Section y(t ) H 0 x(t )  D0 gˈ (7)
5 to verify the proposed method. At last, conclusion is made where, A0 , D0 , H 0 are given by:
in Section 6.
ª012u12 I12u12 º ª 06u3 º ª  I 3u3 º
2 Contact Force Estimation Based on Kalman A0 «0 » , D0 « E » , E6u3 m « S (r ) » ,
¬ 12u12 012u12 ¼ ¬ 6u3 ¼ ¬ CE ¼
Filter
ª I 6u6 06u18 º ª I 3u3 03u3 º
H0 «C », C « S (r )  S (r ) I » .
2.1 Contact Force Modeling ¬ 6u6 06u18 ¼ ¬ CS CE 3u3 ¼

The typical way to measure the contact force between the 2.2 Kalman Filter
EE of the robot and the environment is using a F/T sensor
which is located at the wrist of robot. The ideal measured The forces and torques measured by the F/T sensor usually
force and torque can be written as: contain much noise, and these noises have a great influence
on the control accuracy. The Kalman filter algorithom is
ª FS º ª FE º ª FI º ª FG º
«M » «M »  «M »  «M » , (1) used in this paper to obtain the accurate contact forces.
¬ S¼ ¬ E¼ ¬ I¼ ¬ G¼ The noise is supposed to be white gaussian noises. The
where FS , M S are the measured values of the F/T sensor, system model and measurement model can be given
which are constituted by the contact force FE and torque according to (8), (9):
x(t ) A0 x(t )  wxˈ
x( (8)
M E , the inertial force FI and torque M I , and the gravity
y(t ) H 0 x(t )  D0 g  vy , (9)
FG and gravitational torque M G .
where, wx , v y are system noise and measurement noise
If the manipulator moves smoothly and slowly, the
influence of the inertial force on the measured values of the respectively.
F/T sensor can be neglected. In this case, the measured The state difference equation of the system and
values are only related to the gravity and contact force. As measurement difference equation can be obtained by
shown in Fig. 1, the mass center of the tool is set at Point C, discretizing the continuous system :
and the coordinate frame is established with the origin at C. E xk Axk 1  wk 1 , (10)
is the contact point between the EE and the environment. yk Hxk  Dg  vkˈ (11)
Similarly, the coordinate frame of E is set as 6 E . S is the
ª I12u12 I12u12 º
mass center of the F/T sensor. The frame of F/T sensor is 6 S . where, A «0 » , H H 0 , D D0 .
¬ 12u12 I12u12 ¼
Take 6C as the reference coordinate frame, the acceleration
The Kalman filter is used to estimate the optimization
and angular acceleration of EE are ignored. We can get the output of the system. The covariance of system noise and
Newton-Euler equation of the system: measurement noise are set as Q and R respectively. The state
FEC FSC  mg C , (2) xˆ k |k 1 and the covariance Pk|k 1 at time step k based on the
M EC C
S (rCS ) FSC  M SC  S (rCE
C
) FEC , (3) result at time step k  1 can be calculated firstly:
xˆ k |k 1 Axˆk 1 ,
C
whereˈ m is the mass of the tool, g is the gravitational
(12)
acceleration vector, S (a) is the skew-symmetry operator Pk|k 1 APk 1 AT  Q.
applied to vector a .
Then the Kalman gain K k and the optimal estamation at
time step k are calculate:
Kk Pk|k 1 H T ( HPk|k 1 H  R)1 , (13)
xˆk xˆk|k 1  Kk ( yk  Hxˆk|k 1 ). (14)
The covariance Pk is updated at last˖
Pk ( I  Kk H ) Pk|k 1. (15)
The actual contact force FE and torque M E between the
EE and the environment can be obtained from the estimated
state vector.
Fig. 1: Diagram of force and torque. 3 Position-based Adaptive Impedance Control

Define the state vector x  24


and measurement vector 3.1 Impedance Control Model
y  12 :
The spring-mass-damping model can be used in the
x [ FST M ST FET M ET FST M ST FET M ET ]T , (4) impedance system, and the ideal impedance control model
y [ FST M ST FET M ET ]T . (5) is:
The system model and measurement model can be d  Bd X dc
M d X dc d  K d X dc E, (16)
established based on (2), (3): where, M d , Bd and K d represent target inertia, damping
x(
x(t ) A0 x(t ), (6) and stiffness matrix of the impedance model, respectively;

5611
X dc X d  X c are the difference between the desired Thus, even though the position error 'xe is very small,
position and actual positon; X ddc and X ddc are the velocity the steady state error of the contact force ess could be large
and acceleration respectively cooresponding to due to the large environment stiffness.
X dc ; E =Fd  Fc are the difference between the desired 3.3 Adaptive Impedance Control
contact force and actual contact force.
For simplicity, one dimensional case is only considered. The typical industrial manipulator can be controlled using
The impedance control model is showed in Fig. 2. The position control mode and velocity control mode, while the
impedance control model (16) is simplified as: joint torque of robot can’t be controlled directly. Therefore,
md xdc  bd xdc position-based impedance control method is developed in
d  kd xdc d e. (17)
this paper to control the robot position and contact force.
The environment model can be approximated to The position-based impedance control method consists of
“spring-damping” system [18]: an inner position control loop and an outer indirect force
be ( x  xe )  ke ( x  xe ) f ˈ (18) control loop [19]. In the inner position control loop, the
where, be and ke are the target damping and stiffness of manipulator can be controlled to reach the desired position
environment respectively; x is the actual position of the EE through the joint controller. The position control accuracy of
and xe is the position of the contact surface; and f is the industrial robot are usually very high. Besides, the error of
the contact force is transformed into position correction
actual contact force.
though the impedance controller in the outer control loop.
In free space, the robot does not contact with the
environment, and the outer force control loop should not be
necessary. In this phase, Fd Fc 0 , X c X d . In
constrained space, the nonzero position output from
impedance controller modifies the desired position X d to
achieve force tracking and position tracking. And the robot is
controlled to track the modified desired trajectory through
the inner position control loop.
As described in 3.2, the unknown environment parameters
would have a great influence on the control accuracy.
Besides, the environmental is usually unstructured and
Fig. 2: The impedance model of the EE. unpredictable, the impedance control with constant
3.2 Analysis of Steady State Error parameters can’t be adjusted as the environment changes. If
the impedance control parameters are adjusted automatically
When the pencil, which attached to the EE, contact with according to the environemnt, the robustness could be
environment steadily and the system is in steady state, the guaranteed. We present an adaptive impedace control
impedance control model (17) can be simplified as: method to improve the control accuracy and robustness.
e f d  fc kd ( xd  xc ), (19) In the adaptive impedance control scheme, the parameters
where, xc is the actual position of the pencil. of impedance controller are corrected by adaptive control
law, see Fig. 3.
The environment model can be written as:
fc ke ( xc  xe ). (20)
According to (19) and (20), the force tracking error can be
given as:
kd
ess ( f d  ke xe  ke xd ). (21)
k d  ke
If the steady state error of the contact force is zero, the
desired position xd is: Fig. 3: Scheme of position based adaptive impedance control.
f
xd xe  d . (22) The main idea of this method is generating disired position
ke
xd according to force tracking error e and its derivative e .
It is observed that if the actual contact force is exepected
The adaptive law is:
to realize zero-error tracking, xd should be calculated based
xd (t ) f (t )  k p (t )e(t )  kd (t )e(t ), (24)
on environment parameters xe and ke . In reality, xe is
where, f (t ) is a time-varying parameter which is equivalent
unknown, and ke is varying. The errors of environment
to integral term; k p (t ) and kd (t ) are proportional and
parameters are defined as 'xe and 'ke , respectively. The
integral gain, respectively.
steady state error of contact force is:
In the adapative law, ee((t ) can be replaced by xx((t ) :
kd 'k
ess ( e f d  ke ˜ 'xe  'ke ˜ 'xe ) (23) xd (t ) f (t )  k p (t )e(t )  kd (t ) x(t ), (25)
kd  (ke  'ke ) ke
where x(
x(t ) is the actual velocity of the robot.

5612
Here, the adaptive control law is similar to the PI 5. Track the desired trajectory, and control the
controller. In order to ensure the stability of the whole manipulator using the joint position controller.
system, constants V are used to correct the control law. The
items in adaptive control law (25) are:
t t
f (t ) f (0)  D1 ³ q(t )dt  D 2 q(t )  V 1 ³ f (t )dtˈ (26)
0 0
t
k p (t ) k p (0)  E1 ³ q(t )e(t )dt  E 2 q(t )e(t )
0
t
(27)
 V 2 ³ k p (t )dt ,
0
t
kv (t ) kv (0)  O1 ³ q(t ) x(t )dt  O2 q(t ) x(t )
0
t
(28) Fig. 4: The flow chart of robot writing.
 V 3 ³ kv (t )dt , 5 Experiments
0

q(t ) Z p e(t )  Zv xm (t ), (29) Chinese character writing experiments are designed to


verify the control performance of the proposed method using
where, Z p , Zv are coefficient of position and
the Reinovo REbot-V-6R-6500 industrial manipulator. The
velocity,respectively; f (0) , k p (0) and kv (0) are initial positioning accuracy of the robot is r0.08mm . The
value; D1 , E1 and O1 are integration coefficient; D 2 , E 2 experimental setup is showed in Fig. 5. A six-dimensional
and O2 are proportionality coefficient; V 1 , V 2 and V 3 are F/T sensor Bioforcen is mounted on the wrist of the robot.
The sample frequency of the F/T informations is 5 kHz and
positive and small coefficient. And all of these coefficients the accuracy of this sensor is less than 1% F.S. A pencil is
are constant. tightly fixed on the robot EE, and the stiff paperboard is
Though regulation of adaptive control law, the tracking placed under the robot.
error of contact force could be reduced, and the stability and
robustness could be ensured.
4 Wrting Process Using Manipulator
True Type Font (TTF) vector fonts is a text file format
developed by Apple and Microsoft In this paper,the TTF
fonts are used to get the outlines information of the Chinese
characters that to be written. The Chinese characters writing
process using manipulater is shown in Fig. 4. It can be
divided into five steps:
1. Input the Chinese characters to be written. In general, it
is more difficult to write the complicated characters than the
simple ones.
2. Set the font type, height, width and angle of the
characters. The outline information mainly refer to the pixel Fig. 5: The experienmental setup of robot writing.
positions of the outline’s key points. These information can
be directly extracted from characters outline data’s memory As we all know, unlike the writing brush, the pencil is
structure. fragile. Therefore, when writing on the stiff board, the
3. The characters outlines in TTF fonts are divided into contact force should be controlled accurately to guarantee
straight line and quadratic spline curve. Interpolation method the safety. Moreover, to make the characters clear and
can be used to make the characters more smooth and beautiful, the contact force between the paperboard and
beautiful. The parabolic interpolation method is used to fit pencil should be appropriate. In our experimental scene, the
the quadratic spline curves. desired contact force is set as 6 N, and the desired positions
4. Express the pixel positions in the coordinate frame of in XY-plane are obtained as described in Section 4. The
the robot. The coordinate of EE in XY-plane can be position and stiffness of the paperboard are unknown. The
transformed from pixel coordinate of contour point. For each manipulator is expected to write characters “Zu Kang” on the
outline, the coordinate transformation formula is: paperboard while tracking the desired position and force
P( x, y) R ˜ >Q1 ( x, y)  Q( x, y)@  Max( x, y), (30) simultaneously. The characters are complicated enough to
where, R is a constant transform coefficient; P( x, y) is the verify effectiveness of the proposed method mentioned
above.
desired position of EE, and Q( x, y) is the corresponding The actual writing results using the adaptive impedance
position of pixel; Q1 ( x, y) is the pixel coordinate of the control method, are showed in Fig. 6, meetting the
initial point; Max( x, y) is the upper limit. requirements well. We can see that the lines of characters is
neither too light nor too heavy, and they are uniform. The

5613
actual trajectory tracking result is showed in Fig. 7. the beginning of the segments. The position-based adaptive
Obviously, the trajectory tracking accuracy of every contour impedance control method is robust, and there is no need for
line is high. The actual trajectory is almost coincide with the the knowledge of the environment, such as the surface
desired trajectory. position and the stiffness of the board. This method is proved
to write the Chinese characters on the stiff board succesfully.

Fig. 6: The actual writing result of robot.

Fig. 9: T Result of force tracking and trajectory tracking.

6 Conclusion
In this paper, an adaptive impedace control method based
on Kalman filter is proposed to write Chinese characters for
Fig. 7: Result of trajectory tracking. industrial robot. The Kalman filter is used to estimate the
actural contact force between the EE and the environment. A
The force tracking result during writing is showed in Fig. 8. constant contact force can be tracked using the adaptive
The green solid line is the desired value, the blue dotted line impedance control and there is no need for the environment
is the original signal sampled from the F/T sensor, and the parameters. The experimental results of writing on a stiff
red solid line is the contact force estimated by Kalman filter. paperboard prove that the proposed method could achieve
In the Kalman filter, the covariance of system noise is set as force control and position control steadily. This method
ª107 could be used to other complex interactive tasks in
0 º
Q « 7 »
, and the convairance of measurement unstructured environments. The future work will be focus on
¬ 0 10 ¼ the elimination of the impact force.
noise is set as R 0.5 . Q and R would influence the
estimation result by affecting the value of Kalman gain, and
References
they are adjusted according to the characteristics of system [1] F. Chao, F. Chen, Y. Shen, et al, Robotic free writing of
and sensor. Chinese characters via human–robot interactions,
Due to the fact that a character consists of multiple International journal of humanoid robotics, 11(01):
contour line, the force tracking path is divided into 414–5166, 2014.
[2] X. Gong, J. Wang, Drawing chinese character based on
multi-segments. At the initial stage of every segment, the
continuous path control of IRB140 industrial robot, Modern
impact force between the pencil and paperboard is slightly Manufacturing Engineering, 2010.
large due to the unknown position of the paperboard. The [3] G. Gao, Y. Dai, Research and implementation of writing
contact force can be adjusted quickly using the adaptive function of binocular vision manipulator, IEEE International
impadance control method, and the tracking error is within Conference on Multimedia Technology, 2011: 449-453.
the allowable range. [4] C. Yang, S. Chang, P. Liang, et al, Teleoperated robot writing
using emg signals, IEEE International Conference on
Information and Automation, 2015: 2264-2269.
[5] P. Liang, C. Yang, Z. Li, et al, Writing skills transfer from
human to robot using stiffness extracted from semg, IEEE
International Conference on Cyber Technology in
Automation, Control, and Intelligent Systems (CYBER),
2015: 19-24.
[6] R. Yu, J. Peng, Y. Qin. The Design and Implementation of
Simple CoreXY Structure Writing Robot, APCASE, 2017.
[7] G. Xia, C. Li, Q. Zhu, et al, Hybrid force/position control of
industrial robotic manipulator based on Kalman filter, IEEE
Fig. 8: Result of force tracking. International Conference on Mechatronics and Automation
Overall, the performance of writing is intuitively showed (ICMA), 2016: 2070-2075.
in Fig. 9. We can see from the results that the force is
effectively controlled, and the impact force is only appear at

5614
[8] N. Hogan , Impedance control: an approach to manipulators: [14] Fard, Mohammad Baradaran, and Saeed Khorashadizadeh,
Part illaplications. IEEE American Control Conference, Model free robust impedance control of robot manipulators
107(2), 2009. using fourier series expansion, IEEE AI & Robotics,
[9] L. Yan, Y. Xia, M. Fu, Optimal fusion estimation for 2015:1-7.
stochastic systems with cross-correlated sensor noises[J]. [15] W. Wang, C. Lee, Fuzzy neural network-based adaptive
Science China(Information Sciences), 60(12):120205, 2017. impedance force control design of robot manipulator under
[10] W. Xu, C. Cai, Y. Zou, Neural-network-based robot unknown environment, IEEE International Conference on
time-varying force control with uncertain Fuzzy Systems, 2014: 1442-1448.
manipulator–environment system, Transactions of the [16] H. Wei, Y. Dong, C. Sun, Adaptive neural impedance control
Institute of Measurement and Control, 36(8): 999-1009, of a robotic manipulator with input saturation, IEEE
2014. Transactions on Systems, Man, and Cybernetics Systems,
[11] Jafari, Aghil, et al, Sliding mode hybrid impedance control of 46(3): 334-344, 2016.
robot manipulators interacting with unknown environments [17] Homayounzade, Mohamadreza, and Mehdi Keshmiri,
using VSMRC method, IEEE International Symposium on Observer-based impedance control of robot manipulators,
Robotics, 2012: 1-6. IEEE International Conference on Robotics and
[12] Y. Ren, Y. Zhou, Y. Liu, et al, Adaptive object impedance Mechatronics, 2013.
control of dual-arm cooperative humanoid manipulators, [18] V. Mallapragada, D. Erol, and N. Sarkar, A new method of
IEEE Intelligent Control and Automation, 2015: 3333-3339. force control for unknown environments, International
[13] Sayyaadi, Hassan, and Mojtaba Sharifi, Adaptive impedance Journal of Advanced Robotic Systems, 4(3): 4509-4514,
control of UAVs interacting with environment using a robot 2008.
manipulator, IEEE International Conference on Robotics [19] K. Lee, M. Buss, Force tracking impedance control with
and Mechatronics, 2014: 636-641. variable target stiffness, IFAC Proceedings Volumes, 41(2) :
6751-6756, 2008.

5615

You might also like