You are on page 1of 6

IDENTIFICATION OF GEOMETRIC AND NON GEOMETRIC PARAMETERS

OF ROBOTS

J.L. Caenen J.C. Angue

Ecole des Mines de Douai Universite de Valenciennes


BP 838, 59508 Douai Cedex 59326 Valenciennes Cedex
FRANCE FRANCE

ABSTRACT errors. They are often time expensive,


independent from the geometric model and
This paper presents a general method to may be difficult t o implement in an industrial
identify the geometric and non geometric environment. Other parameters are introduced
parameters of serial robots. The limit of into the geometric model of rotary 'oint robot
absolute accuracy of the robot is defined and to define non geometric effects [6, 73.They are
taken into account in the procedure. employed to characterize the compliance and
Experimental results are applied on a six baklash in the joints due t o the gravitational
degree of freedom robot of cylindrical type. effect.
In this paper, a general calibration model
1 - INTRODUCTION is developed to represent both geometric and
non geometric errors in manipulator structure
Tasks performed by the robots require a including rotary and/or prismatic joints. The
certain positionning accuracy. When these aim of this model is to obtain the limit of
tasks are defined using analytical (or off-line) accurac of the robot. Often define as the
programming tools, the absolute positionning repeatagility [7],a new definition of this limit
accuracy must pay a special attention. The is given wich take into account effects such as
efficiency of this type of programming is the encoder resolution and roundoff error of
however verified only if there is a real the robot controller. The geometric model use
correspondence between the robot model and the modified Denavit-Hartenberg notation
the robot put in action. The absolute including the "modified" p parameter. Non
positionning capability is limited by geometric geometric parameters are defined into the
and non-geometric effects. geometric model to take into account joint
compliance, backlash and flexion. The
The geometric errors in the manipulator calibration equations are presented. They
structure may result from impricise define a linear differential model of
manufacturing of the robot links and joints or independent parameter errors solved by a least
from the deviation of the encoder offsets. A square method. Experimental results are
method to define the geometric model of the applied on a six degree of freedom robot. The
robot is the homogeneous transformation resulting improvement in acuracy due to this
matrix defined by the modified Denavit- calibration is experimentally measured.
Hartenberg notation [l] : four parameters
express the relationship between two
consecutive joints of the robot. Using this 2 - DEFINITION OF THE GEOMETRIC
notation, the model of the end effector PARAMETERS
positionning errors may be developed as
function of the errors in all four link 2.1 - DEFINITION
parameters [2].In the case of two consecutive The system t o be considered is an open
parallel or near parallel joints, small errors in loop mechanism of n joints. Link 0 is the base
the link parameters do not correspond to small while link n is the terminal link. The coordinate
errors in the end effector position. So an other frame j is assigned f x e d with respect to link j.
parameter p must be introduced [3, 41 into the The definition of the link frames will be carried
homogenous transformation. out by the modified Denavit and Hartenberg
notation 111. The 'z. axis is along the axis of
The non geometric errors may result from joint j, the x. adis is along the common
friction, gravity, compliance, gear transmission perpendicular dk z. and z. Frame j is defined
and backlash. Non geometric models are with respect to dame jjl'by the matrix J-lT.
defined to predict and compensate [ 5 ] these which is function of the four parameters (a,:

CH2876-1/90/0000/1032$01.00 Q 1990 IEEE 1032


dj , ej, rj),figure 1, such that :
2.2 - NEAR PARALLEL JOINTS
j-1T -Rot(x, a.)Trans(x,dj)Rot(z,ej)Trans(z,rj) Using the four Denavit-Hartenberg
j- J notation, small variations in the position and
wich give : orientation of two consecutive links can not be
define as function of small variations in the
manipulator link parameters. So, a additional
term Rot (y, 8 ) must be added. In the case of
the modified Denavit-Hartenberg notation, if
z. and z. are parallel, this parameter will
&present rotation around the axis Y.-,.The
frame j is defiofd with respect to the f d m e j-1
cej -sej 0 by the matrix J- Tj :
dj
CajSB, CajCBj -Saj -rj Sa j j-lTj=Rot(y,pj).Rot(x , .).Trans(x,d-)
.Rot(z:ej).Trans(z$) (1)
J - ~ T ~ = sajsej sajcej caj rjcaj
0 0 0 1 3 - DEFINITION OF THE NON GEOMETRIC
ERRORS
where :
In our model of calibration we don't take
j -Is. j -In. , j - l a . define the orientation into account the dynamic effects wich may
of frdke j wizh respezt to frame j-1 (unit influence the robot behaviour (we can find i t in
ye tors). the repeatability of the robot). So the effects
3 -?Pa defines the position of the origin of of the gravity on the manipulator are studied
framd j with respect to frame j-1. here to define the non geometric parameters.
According to the Denavit-Hartenberg
notation the gravity can produce on the robot
three types of "compliance" or 'flexion'. They
are deformations around the X, Y and Z axis
due t8 the different gravity torques. Assuming
that is the position of the mass center of
the jt 'link defined with respect to the Ot
r f,ference frame, the mass mj produce on the
it link three torques :

In the general case, the resulting torques


on the ith axis are given as :

Figure 1
The joint variable j will be given as :
qj = U
j j
e +U
jrj n
where : u.=O for j rotational, uj=1 for j ri = k=i+((OMk-*Pi)x mk.0ao).0ai.g
translational, EChd Uj = ( 1-U ) . 1
We define also : x denotes the vector product.
q.=a-f3.+a.r.
3 3 3 3 3
It is to be noted that frame 0 can be
defined such that zo is along z1, and XQ is along
x1 when 1'0, so al=0, d -0 q 0, while frame
n can be 2efined such tha$-qL=d=[l].

1033
2- Since the parameter dj represents a
This torques modify the correspondent translation along xa-1. The column
geometric parameters such as : corresponding t o dj in th&Jacobian matrix Will
be given as :
ai = ai + K,i.Ti
"j-I
pi = pi + Kpi.Ci (3)
Jdj= I o
ei = ei si.ri
4-

and so introduce the new parameters K ~ K~~


~ ,
-
3- Since the parameter e represents a rotation
around the axis. The 'column corresponding
and si. to ej in the jjcobian matrix will be given as :

4 - DEFINITION OF THE IDENTIFICATION


MODEL
4.1 - GENERAL DEFINITION 4- Similarly the column corresponding to rj is :
Uspg the homogenous transformation
matrix J- T. (l), the location of the tool frame n
with respek t o the base can be represented as

~ OT~.~T~...~-~T,
O T = 5- Since the parameter p . represents a rotation
around the yj axis, we gdt :
Assuming the errors of first order,
location errors of the robot tool frame dues to
the errors in the geometric parameters can be
approximated as : I "j-1 I
All the vectors appearing in equations ( 5 ,...,9)
must be referred to the measuring fixed frame.
So we need to calculate the vectors s a , n . ,
referred to the fixed frame? Th Ae
where : %t$j '?an be obtained from the matrices %'.
o=O, ... n). The vector L. will be obtained ad
- D is the (3x1) translational differential pn-Ob. . These matrickSn can be calculated
vector ofnthe tool frame. symbol$cally or numerically. The numerical
- 6 is the (3x1) rotational differential calculation is general and preferred for the off
vector or the tool frame. line identification especially if B cannot be
- B defines the errors in the robot considered as first order and iterative
parameters. calculation is investigated to identify it. The
- J, is the extended jacobian matrix. symbolic calculation will be more efficient,
from the computation cost point of view, for
4.2 - DEFINITION OF THE JACOBIAN the correction problem.
MATRIX
Only the equations corresponding to the
4.2.1. GEOMETRIC PARAMETERS. The position error D,, on sufficient number of
calculation of the columns of extended jacobian points, are used in the identification process.
matrix can be calculated directly as follows [8] The solution of the corresponding system of
equation, represented by the first three
equations of relation (4), is obtained by a
1- Since the parameter a . represents a rotation classical least squares method.
around the x. 1 axis (of3unit vector s--1). The
column corr&ponding to a j in the jacobian 4.2.2. GEOMETRIC AND NON GEOMETRIC
matrix will be given as : PARAMETERSDue to the equation (2,3) using
the equations (5,..,9), the columns
corresponding to the non geometric parameters
are given by :

x denotes the vector product,


L. is the (3x1) vector between the
origin ofirame i and the origin of frame n.

1034
and the columns corresponding to the steady state errors, friction and the dynamic
geometric parameters become : effects on the robot links.
5.2 PRECISION OF THE GEOMETRIC
Jej= Jej(1 + Sj-Jrj) CONTROL SYSTEM
When the robot is under a geometric
Jaj= J , j ( l + Kaj J T j ) control, the positionning software must
transform the analytic orders (positions and
Jpj' Jpj(1 + Kpj-JCj) orientations) into joint orders. Even if the
geometric parameters of the robot are well
known, roundoff errors and errors due to the
Where Jrj - 6ri encoder resolutions may occur. So these errors
68j constitute a new cause of the robot
positionning uncertainty. The value of this
JTj = uncertainty is obtained using a simulation
6aj program.
Jcj = 5.3 LIMIT OF ABSOLUTE ACCURACY
68j The sum of the tool location errors
defined in 5.1 and 5.2 is used to express the
limit of absolute accuracy of the robot.
4.2.3. IDENTIFIABLE PARAMETERSK a
column of the jacobian matrix is a linear This limit includes the real behaviour of
combination of some other columns, the the robot (with friction, dynamic...) and the
corresponding error parameter can not be influence of geometric uncertainties wich can
identified. In this case, this parameter error is not be defined in the identification model of
forced to zero and his identification is carried the calibration.
out through the errors on the other
parameters. The combinations are obtained on This limit constitutes the purpose to
a case by case study. attain during the calibration process.
From section 2, when z . - ~and z. are 6 - EXPERIMENTATION AND RESULTS
parallel, the additional parameder p . muht be
introduced. In this case, we remard that the Experimental results are carried out on a
errors on r. will not be calculated because it is TH8-ACMA robot, wich is a six degree of
in the samb direction of ri-l. So the maximum freedom robot of cylindrical type (figure 2).
nomber of geometric errors is 4 and the
maximum number of non geometric errors of 6.1 INSTRUMENTATION
each frame is 3. The instrumentation is based on the use
of two theodolites and a cylindrical tool fixed
As a conclusion the definition of the on the robot terminal link. The coordinates of
geometric and non geometric parameters of an the robot point tool is determined with an
n link robot will need a maximum number of accuracy about &0,02mm/m. The point tool is
7.n independent parameters. materialized by a 2 mm diameter sphere. The
dimensions of this tool are measured with
precision.
5 - THE LIMIT OF ABSOLUTE ACCURACY
6.2 MEASUREMENTS
To perform an analytical programming, The measurements are expressed in the
the robot needs to use two types of elements : robot fixed frame to satisfy the definition of
the identification model (4). The location of the
- the positionning control system of robot fixed frame with respect to the
the joints, theodolites frame is obtained by using a lot of
- the geometric control system wich specific procedures [9]. In the case of the TH8
contain the robot model. robot, this location is defined as follow :
The limit of the robot accuracy depends - axis Z is the perpendicular on the
on the limits of these two elements. circular plane h t a i n e d by rotating joint 1.
- axis YO is obtained by identifying the
5.1 PRECISION OF THE POSITIONNING translation of joint 3.
CONTROL SYSTEM - the origin of the frame is the center of
This Drecision is defined bv the the circle during the rotation of joint 1 (other
repeatability-of the robot wich charactehze the robot joints on their position of geometric
behaviour of the robot mecanics under control. initialisation).
Measured using the I.S.O. TC 184/SC2
specifications, the repeatability includes The measurement points are divided in

1035
the complete volume of work of the robot. A
total of 48 configurations are obtained from Two identifications are carried out.
significant displacements of the robot. For each Including only the geometrical parameters, the
of them, the values of the joint encoders and first calibration permit to obtain an accuracy
the corresponding coordinates of the tool point about 0,69 mm with a standard deviation about
are recorded. The measured absolute accuracy 0,34 mm. The second calibration includes the
before calibration is 2,82 mm in average with a geometric and non geometric parameters. The
standard deviation about 0,88 mm. obtained accuracy is about 0,58 with a
standard deviation about 0,21 mm. The values
6.3 LIMIT OF ABSOLUTE ACCURACY of the correspondent parameters errors are in
The repeatability of the tested robot is the following tables :
about 0,3mm. Identification of geometrical parameters:
The resolution of the encoders is about frame P a d e r
0,l mm and 0,12 mm for the translation joints
and between 1.10-2 and 2.10-3 degrees for the 1 0 0 0 0,015 0
revolute joints. Dimensional parameters are I I I I I I I

taken into account in the robot software with


a roundoff about 0,l mm. The simulated
precision of the geometric control system is 3 0 0,099 0,874 0 2,087
found about 0,25 mm.
4 -0,062 0,031 0 -0,133 0
So the limit of absolute accuracy of the
robot is about 0,55 mm. 5 0 0,014 -0,041 -0,114 0

6.4 IDENTIFICATION RESULTS 6 0 0,063 0 0 -3,703


The identifiable and eliminated
parameters of the TH8 robot are as follow : angles in degrees and dimensions in mm
Identification of geometrical and non
frame /3 a d 8 r Ke K, K8 geometrical parameters:
I I

1 E E I I E E E E frame P a d e r

2 I I E E I E I I 1 0 0 0 0,014 0

2 0,049 -0,049 0 0 0,940

0,076 3 I 827

and
frame
Ke Ka Ks
1 0 0 0

2 0 0 0,021

3 0 0,076 0

4 I 0,011 I 0,043 I -0,032 I

9 Figure 2, the TH8 robot.


5

6
0,'006

angles in degrees and dimensions in mm


For this robot, we see that the accuracy
0 009

0
0

1036
by calibrating the geometrical parameters is
near to the limit of the absolute accuracy of
;the robot. The accuracy is improved by the
complete" calibration wich permit to obtain a
result comparable to the limit of absolute
precision of the robot. All these identified
values ma be used in an compensation
algorithm [& to correct the robot.
7.CONCLUSION
This paper presents a general calibration
procedure wich take into account geometric
and non geometric parameters of robots. The
limit of absolute accuracy is defined as the
limit to attain by the identification.
Experimental results are applied on a six
degree of freedom robot including revolute and
prismatic joints. The improvements permit to
obtain an accuracy value comparable to the
limit of the robot.
REFERENCES
[ I ] W.KHALIL, J.F KLEINFINGER, "A new geometric
notation for open and closed boop robots", Proc. IEEE
Robotics and Automation Conf., San Francisco, 1986,
pp.1174-1180.

[Z] J.L CAENEN, J.C ANGUE, "Robot Calibration", Proc.


International Symposium o n Industrial Robot, Tokyo,
1989.

[3] S.A.HAYAT1, "Robot arm geometric link calibration"


link calibration", Proc. IEEE, Decision and Control conf.,
pp 798-800, 1988.

[4] C.H.Wa, J.Ho, K.Y.YOUNG, "Design of robot accuracy


compensator after calibration'l, Proc. IEEE Robotics and
Automation Conf., pp 780-785, 1988.

IS] S.AHMAD, "Analysis of robot drive train errors, their


static effects, and their compensations", IEEE Journal of
robotics and automation, vol. 4, n)2, April 1988, pp117-
128.

IS] JSSHAMMA, D.E. WHITNAY, "A method for inverse


robot calibration", Transactions of the ASME, ~01.109,
March 1987.

I71 J.CHEN-L.M CHAO, "Positionning error analysis for


robot manipulators with all rotary joints", IEEE Journal
of robotics and automation vol RA-3, n"6, december 1987
pp 539-545.

[SI W.KHALIL, J.L.CAENEN, C.H.ENGUEHARD,


"Identification and calibration of t h e geometric
parameters of robots, First International Symposium o n
Experimental Robotics, Montreal 1989.

191 J.L. CAENEN, "Programmation hors-ligne, Etalonnage


d e robots : identification des parametres geometriques",
DEA- Valenciennes. 1988.

1037

You might also like