Journal of

Mechanical
Science and
Technology
Journal of Mechanical Science and Technology 22 (2008) 1084~1090
www.springerlink.com/content/1738-494x

Kinematic parameter calibration method for industrial robot

manipulator using the relative position
In-Chul Ha*
Robot System Research at Mechatronics Center, Samsung Heavy Industries Co., Ltd.
Institute of Industrial Technology 103-28, Mounji-dong, Yuseong-gu, Daejeon 305-380, Korea

(Manuscript Received August 17, 2006; Revised March 3, 2008; Accepted March 10, 2008)
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Abstract

A new calibration method for industrial robot system calibration on a manufacturing floor is presented in this paper.
To calibrate the robot system, a laser sensor to measure the distance between robot tool and measurement surface is
attached to the robot end-effector and a grid is established in the floor. Given two position command pulses for a robot
manipulator and using the position difference between two command pulses, the relative position measurement calibra-
tion method will find the real robot kinematic parameters. The procedures developed have been applied to an industrial
robot. Finally, the effects of the models used to calibrate the robot are discussed. This calibration method represents an
effective, low cost and feasible technique for the industrial robot calibration in lab. projects and industrial environments.
Keywords: Robot calibration; Relative position; Laser sensor
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

motion in complex environments. Contemporary

1. Introduction
robot manipulators are capable of efficiently execut-
When robots were first introduced into the indus- ing a variety of tasks with excellent repeatability.
trial market, the tasks they performed were relatively Repeatability refers to the robot's ability to return its
simple. Most robots required a teaching phase in the end-effector to a position that has been previously
programming. Here, the user taught the various posi- “taught”. Absolute accuracy refers to a robot's ability
tions by moving the robot to the points and then re- to position its end-effector relative to a fixed refer-
cording them. As long as the robot could repeatedly ence frame.
move back to the taught points, the robot could suc- A robot normally has high repeatability, but its ac-
cessfully perform the task. curacy is much worse than its repeatability. The chal-
Robots today are much more sophisticated. The lenge, therefore, is how to improve and maintain the
paths to be tracked can be entered directly into the system's accuracy in varying manufacturing environ-
robot controller. However, since the robot is now ments.
required to trace a path that is mathematically de- Robot calibration is a cost-effective way to im-
scribed rather than taught, a repeatable robot is no prove robot accuracy, and many researchers have
longer enough. devoted efforts to this field. Different models, meas-
In recent years, robotic research has been focused urement systems, and algorithms for identification
on solving a variety of tasks requiring sophisticated and compensation have been developed as summa-
rized by Roth et al. [1], Hollerbach [2], and Mooring
*
[3].
Corresponding author. Tel.: +82 42 865 4773, Fax.: +82 55 630 7673
E-mail address: [email protected] One method of improving robot accuracy involves
DOI 10.1007/s12206-008-0305-0 identifying the real kinematic parameters by minimiz-
 I.-C. Ha / Journal of Mechanical Science and Technology 22 (2008) 1084~1090 1085

ing the error between the position and orientation

calculated by the robot controller and the actual
measured position and orientation. (Driels and
Swayze [4], Driels [5], Render et al. [6], Judd and
Kaminski [7], Wyatt S. Newman, Craig E. et al [8],
Gursel alici and Bijan Shirinzadeh [9])
Other researchers have proposed the use of laser-
based measurement systems and vision-based meas-
urement systems. Generally speaking, non-contact
systems employing lasers or theodolites are very ac-
curate; however, they are complicated and require Fig. 1. D-H model diagram.
highly trained personnel to operate them. (Driels and
Pathre [10], Zhuang et al [11], Patrick Rousseau et al Ai = Rot ( Zi −1,θi ) ∗ Trans ( Z i −1, di )
[12]) ∗ Trans ( X i −1, ai −1 ) ∗ Rot ( X i ,α i )
It is the goal of this paper to develop a new calibra- ⎛ Cθi − SθiCα i Sθi Sα i aiCθi ⎞
tion method for industrial robot system calibration on ⎜ ⎟
Sθ Cθi Cα i −Cθi Sα i ai Sθi ⎟
a manufacturing floor. To calibrate the robot system, =⎜ i (1)
⎜ 0 Sα i Cα i di ⎟
a laser sensor to measure the distance between robot ⎜⎜ ⎟
tool and measurement surface is attached to the robot ⎝ 0 0 0 1 ⎟⎠
end-effector and a grid plate (an interval of 0.1mm) is
established in the floor. Where Sθi and Cθi represent sin θi and cos θi ,
This method can calibrate the robot without cali- and Sα i and Cα i represent sin α i and cos α i
brating the transformation from the world coordinate respectively.
system to the robot base coordinate system. This As pointed out by Hayti [14], small errors in the
makes the robot calibration easy and convenient to end-effector position could not be modeled by small
implement on the manufacturing floor. Due to the errors in the DH link parameters in the case of two
reduction of measurement number, this method can consecutive parallel joints or nearly parallel joints.
decrease the time consumption. These devices for This causes numeric instability during the identifica-
calibrated method are easy to use and low cost less tion process. In order to avoid the singularity problem,
than others. Also, this makes the robot calibration a small rotation of β about the y-axis, Rot (y, β ),
easy and convenient to implement on the manufactur- is added. As for a robot with two parallel joints or
ing floor. nearly parallel consecutive joints, the homogenous
An experiment using a 6-DOF robot is conducted transformation Ai becomes Eq. (2).
to show the effectiveness of the proposed calibration
algorithm. ⎛ CθiCβi − Sθi Sαi Sβi −SθiCαi Cθi Sβi + Sθi SαiCβi aiCθi ⎞
⎜ ⎟
Sθ Cβ + Cθi Sαi Sβi CθiCαi Sθi Sβi − Cθi SαiCβi ai Sθi ⎟
Ai = ⎜ i i
2. System model ⎜ −Cαi Sβi Sαi CαiCβi di ⎟
⎜⎜ ⎟
⎝ 0 0 0 1 ⎟⎠
2.1 Robot model
(2)
Since the kinematic model of a robot is used as the
foundation for further development of robot error Where S βi and C βi represent sin βi and cos βi
synthesis model, it is briefly reviewed here. The ki- respectively.
nematic model is based on the Denavit Hartenberg For a robotic manipulator with N degrees of free-
(DH) convention [13]. The relative translation and dom, the position and orientation of the robot tool
rotation between link coordinate frames i-1 and i can frame with respect to robot base frame can be repre-
be described by a homogenous transformation matrix, sented by
which is a function of four kinematic parameters θi , ⎛R PN ⎞
di , α i and ai as shown in Fig. 1. TN = A1 ⋅ A2 " AN = ⎜ N ⎟ (3)
⎝ 0 1 ⎠
The homogenous transformation Ai is given in
Eq. (1). Where RN is the rotation matrix representing the
1086 I.-C. Ha / Journal of Mechanical Science and Technology 22 (2008) 1084~1090

orientation of the tool frame relative to the base frame 3. Calibration method
and PN is the displacement vector representing the
3.1 Position error model
Cartesian position of the tool frame relative to the
base frame. If the position of the end-effector with respect to
the base, assuming the position of the manipulator
2.2 Calibration model with nominal link parameters is given by Pt , and the
The differential change dAi is estimated as a lin- real position of the manipulator with kinematic errors
ear function of the four kinematic errors. Due to the is given by Ptc then the correct position can be ex-
modified kinematics resulting in the homogeneous pressed as:
transformation given by Eq. (2), the differential
change dAi should be estimated as a linear function Ptc = Pt + dPt (8)
of the five kinematic errors, assuming the errors are
small so that the higher order terms are negligible If small errors in the position of the end of the ma-
nipulator with respect to the base can be modeled as
∂Ai ∂A ∂A ∂A ∂A small variation in the link parameters, then the ma-
dAi = δθi + i δ di + i δ ai + i δαi + i δβi
∂θi ∂di ∂ai ∂αi ∂βi nipulator position error can be approximated as:
(4)
⎡ (T ⋅ B ⋅ iT ) ⎤
⎛ dPx ⎞ N ⎢ i −1 θ i N (1,4) ⎥
The final robot positional and orientational change
∑
⎜ ⎟ ⎢ i ⎥
dPt = ⎜ dPy ⎟ = {⎢ (Ti −1 ⋅ Bθ i ⋅ TN ) (2,4) ⎥ ⋅ δθ i
can be calculated through Eq. (5). ⎜⎜ dP ⎟⎟ i =1 ⎢ ⎥
⎝ z⎠ ⎢ (Ti −1 ⋅ Bθ i ⋅ iTN ) (3,4) ⎥
⎣ ⎦
∂TN ∂T ∂T ∂T ∂T ⎡ (T ⋅ B ⋅ iT ) ⎤ ⎡ (T ⋅ B ⋅ iT ) ⎤
dTN = δθ1 + N δ d1 + N δ a1 + N δα1 + N δβ1 ⎢ i −1 di N (1,4) ⎥ ⎢ i −1 ai N (1,4) ⎥
∂θ1 ∂d1 ∂a1 ∂α1 ∂β1 ⎢ i ⎥ ⎢ i ⎥
+ ⎢ (Ti −1 ⋅ Bdi ⋅ TN ) (2,4) ⎥ ⋅ δ d i + ⎢ (Ti −1 ⋅ Bai ⋅ TN ) (2,4) ⎥ ⋅ δ ai
# ⎢ ⎥ ⎢ ⎥
⎢ (Ti −1 ⋅ Bdi ⋅ iTN ) (3,4) ⎥ ⎢ (Ti −1 ⋅ Bai ⋅ iTN ) (3,4) ⎥
∂TN ∂TN ∂T ∂T ∂T ⎣ ⎦ ⎣ ⎦
+ δθN + δ dN + N δ aN + N δα N + N δβN ⎡ (T ⋅ B ⋅ iT ) ⎤ ⎡ (T ⋅ B ⋅ iT ) ⎤
∂θN ∂d N ∂aN ∂α N ∂βN ⎢ i −1 α i N (1,4) ⎥ ⎢ i −1 β i N (1,4) ⎥
⎢ i ⎥ ⎢ i ⎥
(5) + ⎢ (Ti −1 ⋅ Bα i ⋅ TN ) (2,4) ⎥ ⋅ δα i + ⎢ (Ti −1 ⋅ Bβ i ⋅ TN ) (2,4) ⎥ ⋅ δβ i }
⎢ ⎥ ⎢ ⎥
⎢ (Ti −1 ⋅ Bα i ⋅ iTN ) (3,4) ⎥ ⎢ (Ti −1 ⋅ Bβ i ⋅ iTN ) (3,4) ⎥
where, ⎣ ⎦ ⎣ ⎦
(9)
∂TN ∂A ∂Ai
= A1 ∗ A2 ∗ " ∗ i ∗ Ai +1 ∗ " * AN , = Bxi
∂γ i ∂γ i ∂γ i The above equation can be written in the following
γ = {θ , d , a ,α , β } compact form as
dTN = Bθ 1 ⋅ 1TN ⋅ δθ1 + Bd 1 ⋅ 1TN ⋅ δ d1 + "
dPt = [ M θ ]δθ + [ Md ]δ d + [ Ma ]δ a + [ M α ]δα + [ M β ]δβ
+ T1 ⋅ Bθ 2 ⋅ 2TN ⋅ δθ 2 + T1 ⋅ Bd 2 ⋅ 2TN ⋅ δ d 2 + " (10)
#
+ TN ⋅ Bθ N ⋅ δθ N + TN ⋅ BdN ⋅ δ d N + " where,
(6)
M γ = Ti −1 ⋅ Bγ i ⋅ iTN
Therefore, (i-th column, γ = θ or d or a or α or β )
δθ , δ d , δ a, δα , and δβ : time invariant constant
N
dTN = ∑{(T
i =1
i −1 ⋅ Bθ i ⋅ iTN ) ⋅ δθi + (Ti −1 ⋅ Bdi ⋅ iTN ) ⋅ δ di 3.2 Relative position measurement method

+ (Ti −1 ⋅ Bai ⋅ iTN ) ⋅ δ ai + (Ti −1 ⋅ Bα i ⋅ iTN ) ⋅ δα i

Given two position command pulses for robot ma-
nipulator and using the position difference between
+ (Ti −1 ⋅ Bθ i ⋅ iTN ) ⋅ δβi } two command pulses, the relative position measure-
(7) ment calibration method will find the real robot ki-
(*) T0 = N TN = I nematic parameters. Fig. 2 is the block diagram.
 I.-C. Ha / Journal of Mechanical Science and Technology 22 (2008) 1084~1090 1087

Fig. 2. Relative measurement method block diagram.

The equation of first position command for robot

manipulator can be written as:

Pt1c = Pt1 + dPt1 (11)

The equation of the second position command for

robot manipulator can be written as:
Fig. 3. Calibration process.
Ptc2 = Pt 2 + dPt 2 (12)
The measurement number ( K ) of relative position
The relative position difference between two posi- can be written as follows:
tion commands results in the following equations.
K ≥ N × 5/3 (18)
Pt1c − Ptc2 = ( Pt1 + dPt1 ) − ( Pt 2 + dPt 2 )
(13)
= ( Pt1 − Pt 2 ) + (dPt1 − dPt 2 ) where,
N : Robot joint D.O.F.
(dPt1 − dPt 2 ) = ( Pt1c − Ptc2 ) − ( Pt1 − Pt 2 ) (14)
5 : Number of modified D-H model parameter
where, 3 : Number of position factor (x, y, z)

dPt1 − dPt 2 : Relative position error model To find calibrated robot kinematic parameters, we
Pt1c − Ptc2 : Real robot relative position (Measure- use the least square method (L.S.M.).
ment)
Pt1 − Pt 2 : Nominal robot relative position (Known) Model equation: Y = Φ ∗ Χ (19)
∴ Χ = (ΦT ⋅ Φ ) −1 ⋅ ΦT ⋅ Y (20)
From Eq. (10), the relative position error model is
derived: where,
1 1 1 1 1
dPt1 = [Mθ ]δθ + [Md ]δ d + [Ma ]δ a + [Mα ]δα + [M β ]δβ
Y = dPt1 − dPt 2 = ( Pt1c − Ptc2 ) − ( Pt1 − Pt 2 )
(15)
Φ = [( M θ 1 − M θ 2 ) ( Md 1 − Md 2 ) ( Ma1 − Ma 2 )
dPt 2 = [Mθ 2 ]δθ + [Md 2 ]δ d + [Ma2 ]δ a + [Mα 2 ]δα + [M β 2 ]δβ
(16) ( M α 1 − M α 2 ) ( M β 1 − M β 2 )]
Χ = [δθ δ d δ a δα δβ ]
Therefore,

∴ (dPt1 − dPt 2 ) = [ M θ 1 − M θ 2 ]δθ + [ Md 1 − Md 2 ]δ d 3.3 Calibration process

+[ Ma1 − Ma 2 ]δ a + [ M α 1 − M α 2 ]δα + [ M β 1 − M β 2 ]δβ From the above equations, Fig. 3 is the calibration
(17) process.
1088 I.-C. Ha / Journal of Mechanical Science and Technology 22 (2008) 1084~1090

① Initialize nominal kinematic link parameter RS232C serial data communication is set up to in-
② Relative position measure (K numbers) tegrate the subsystems such as the robot, and the PC
③ Least square method into measurement and calibration system. After the
: Find optimal kinematic parameters robot moves to one position, its controller sends a
④ Update optimal kinematic parameter signal to request that the sensor start measurement.
⑤ Repeat 2-4 until some minimum value After the measurement is completed the robot moves
to the next position. This pattern continues until the
At this calibration process, using the nominal ki- whole measurement task (the number of K ) is done.
nematic parameters with zi −1 and zi parallel (Fig. Then the robot controller sends all joint pulse read-
1), the errors in di and di +1 are dependent since ings to the PC, and the sensor sends all the measure-
they have the same effect on the reference point posi- ments to the person. Based on all these data, the ki-
tion for any manipulator configuration. This depend- nematic identification algorithm is used to find the
ency will cause a singularity in the matrix of partial true parameter value.
derivative used in the calibration equations. Through The overall system figure is shown in Fig. 5.
the pivot check for the matrix of partial derivative, if
the pivot is smaller than any tolerance value then the 4.2 Calibration results
column factor is rejected and that parameter is set
zero. 4.2.1 Robot model
From the DH model (Fig. 1), the model of a six
4. Experiment DOF robot (MOTOMAN UP20) can be given as
follows:
4.1 System configuration
Then, nominal robot kinematic parameters are
The experimental system for calibration consists of given as in Table 1.
a six DOF robot (UP20) manufactured by
MOTOMAN comp., a laser height sensor manufac-
tured by OMRON, Inc., a grid plate (an interval of
0.1mm) by us, and a PC.
The laser height sensor can conduct height meas-
urement within its field of view and it has a meas-
urement range 80mm-140mm and its resolution is
0.01mm. Using the mechanical height gage (its accu-
racy is 0.01mm), the calibration for laser height sen-
sor is achieved in Fig. 4.
The grid plate made by CNC machine can conduct Fig. 5. Overall system for calibration method.
xy-axis position measurement and its grid interval is
0.1mm. Namely, the position accuracy of xy-axis is
a3
0.1mm and that of z-axis is 0.01mm. d4
X4
X3

Z5 Z4 , X5
Z3 d6

Z1
a2
X6

Z6
d1 X1 X2

a1

Z0
d2

Z2

X0

Fig. 4. Calibration for laser sensor. Fig. 6. Robot model.

 I.-C. Ha / Journal of Mechanical Science and Technology 22 (2008) 1084~1090 1089

Table 1. Nominal robot kinematic parameters. Table 3. Calibrated kinematic parameters.

Par. αi ai di θi βi Calibrated kinematic parameters

Joint (deg.) (mm) (mm) (deg.) (deg.)
1-th iteration 2nd iteration
1 0 150 535 0 0
0, -0.1, -0.195, 0, -0.1, -0.198,
2 90 730 0 90 0 ∆θ
0.067, -0.151, 0 -0.067, -0.149, 0
3 0 140 0 0 0 0, –2.064, 0, 0, -2.117, 0,
∆d
4 90 0 765 0 0 0.377, 0.150, -0.018 0.364, 0.147, 0.002
5 -90 0 0 -90 0 2, 1.331, 2.943, 2.008, 1.332, 2.984,
∆a
6 0 0 105 0 0 -0.006, 0, 0.001 0, 0, 0
0, -0.063, 0.004, 0, -0.063, 0,
∆α
-0.004, 0, 0 0, 0, 0
Table 2. Relative position data (Nominal parameters).
∆β 0, 0, 0, 0, 0, 0 0, 0, 0, 0, 0, 0
Mea Real robot Nominal robot model
Num relative position relative position
(K) ∆X ∆Y ∆Z ∆X ∆Y ∆Z Table 4. Relative position data (Calibrated parameters).

1 0.0 -70.0 -0.11 0.04 -69.85 0.03 Real robot Calibrated robot model
Mea.
2 70.1 70.0 0.18 70.00 69.86 0.02 Num relative position relative position
3 -65.0 -5.2 0.01 -64.95 -5.08 0.04 (K) ∆X ∆Y ∆Z ∆X ∆Y ∆Z
4 0.0 9.8 0.04 0.02 9.78 0.02 1 0.0 -70.0 -0.11 0.0 -70.0 -0.11
5 -10.3 0.0 -11.12 -10.23 -0.02 -11.10 2 70.1 70.0 0.18 70.1 70.0 0.18
6 0.0 -9.7 4.66 0.00 -9.66 4.68 3 -65.0 -5.2 0.01 -65.0 -5.2 0.01
7 -15.0 -20.5 12.26 -14.97 -20.44 12.29 4 0.0 9.8 0.04 0.0 9.8 0.04
8 -10.1 -34.6 -0.04 -10.12 -34.51 0.03 5 -10.3 0.0 -11.12 -10.3 0.0 -11.12
9 40.3 19.7 8.84 40.21 19.68 8.78 6 0.0 -9.7 4.66 0.0 -9.7 4.66
10 40.3 -14.9 -48.28 40.23 -15.02 -48.19 7 -15.0 -20.5 12.26 -15.0 -20.5 12.26
RMS Error: 0.128 8 -10.1 -34.6 -0.04 -10.1 -34.6 -0.04
9 40.3 19.7 8.84 40.3 19.7 8.84
4.2.2 Robot relative position measurement 10 40.3 -14.9 -48.28 40.3 -14.9 -48.28
The real robot relative position is the measurement RMS Error: 0.0
data between the laser height sensor and grid plate.
The nominal robot model relative position is the cal-
Table 5. Absolute position data from robot base.
culation data of the kinematic equation from the real
robot joint pulses. P1 P2 P3
Then the measurement number (K) is 10 because X 1084.10 947.31 736.89
K ≥ N (6) × 5 / 3 = 10 . Nominal
Y 430.19 -326.24 181.86
Data
The RMS error is the root mean square between the Z 524.41 563.93 -17.91
real robot relative position and the nominal robot X 1090.18 953.91 741.54
model relative position.
Calibration Data Y 433.70 -326.54 184.63
4.2.3 Calibrated kinematic parameter Z 524.21 564.10 -18.18
Based on all above data, the kinematic identifica- Measured X 1090.10 953.72 741.16
tion algorithm is used to find the true parameter value, Data Y 433.51 -326.77 184.75
and the calibrated kinematic parameters are results in (LTD500) Z 524.43 563.97 -18.10
Table 3. RMS Error
According to the calibration process, Fig. 3, in case (Nominal data vs. 7.02 6.61 5.42
of second iteration, the calibration accuracy of second Measured data)
iteration is finer than that of the first iteration. RMS Error
The number of parameters is the same as the joint (Calibration data vs. 0.30 0.32 0.40
number (a six DOF). Measured data)
1090 I.-C. Ha / Journal of Mechanical Science and Technology 22 (2008) 1084~1090

4.2.4 Confirm robot position tion, John Wiley and Sons, New York, (1991).
Using the calibrated kinematic parameters, we con- [4] M. R. Driels and W. E. Seayze et al., Automated
firmed the improvement of position accuracy. The partial pose measurement system for manipulator
real robot relative position data is the same as the calibration experiments, IEEE Trans. Robot. Auto-
calibrated robot model relative position data. mat., vol. 10, Aug. (1994) 430-440.
RMS error of the relative position data is reduced [5] M. R. Droels, Using passive end-point motion con-
from 0.128mm to 0.0mm. To find the real absolute straints to calibrate robot manipulators, J. Dynamic
error from robot base to tool position, we measured Syst., Meas. Contr., vol. 115 (1993) 560-566.
the real absolute error by using the laser tracker de- [6] J. M. Renders, E. Rossignol, M. Becquet, R. Hanus,
vice. (LTD500 by Leica Inc.) Then, the residual error and 1991 et al., Kinematic calibration and geomet-
was reduced. rical parameter identification for robots, IEEE
Improving the accuracy of the grid plate and the la- Trans. Robot. Automat., vol. 7 Dec. (1991) 721-731.
ser height sensor can improve the residual error of [7] R. P. Judd and A. B. Knasinski et al., A technique
robot position. to calibrate industrial robots with experimental veri-
fication, IEEE Trans. Robot. Automat., vol. 6 Feb.
(1990) 20-30.
5. Conclusion
[8] W. S. Newman, C. E. et al, Calibration of a moto-
A new low-cost position measurement setup for an man P8 Robot Based on Laser Tracking, IEEE Int.
industrial robot and the calibration method has been Cong. On Robotics * Automation, Apr. (2000)
developed. The theoretic derivation of this calibration 3597-3602.
method shows that the base calibration is not neces- [9] A. Gursel and B. Shirinzadeh, Laser interferometry
sary. From the proposed calibration method, the robot Based Robot Position Error Modelling for Kine-
kinematic parameters were estimated. Then the posi- matic calibration, IEEE/RSJ Int. Conf. Intelligent
tion error of robot end-effector was measured below Robots and Systems, Oct. (2003) 3588-3593.
0.3~0.4mm by Laser Tracker system (LTD500). This [10] M. R. Driels and U. S. Pathre, Vision-based auto-
unique feature not only makes this calibration method matic theodolite for robot calibration, IEEE Trans.
easier to implement, but also shortens the time con- Robot. Automat., vol. 7 Jun. (1991) 351-360.
sumption. Since it is easy and convenient to imple- [11] H. Zhuang, K. Wang and Z. S. Roth et al., Simul-
ment we expect the proposed calibration method will taneous calibration of a robot and hand-mounted
have wide application on the manufacturing floor for camera, IEEE Trans. Robot. Automat., vol. 11 Oct.
robot on-line accuracy enhancement and maintenance. (1995) 649-660.
[12] P. Rousseau, A. Desrochers and N. Krouglicof,
Machine Vision System for the Automatic Identifi-
References
cation of Robot Kinematic parameters, IEEE Trans.
[1] Z. S. Roth, Z. W. Mooring and B. Ravani, An over- Robot. Automat., 17 (6) Dec. (2001) 972-978.
view of Robot Calibration, IEEE J. Rob. Automa- [13] J. Denavit and R. S. Hartenberg, A kinematic nota-
tion, 3, No. 5 (1987) 377-385. tion for lower pair mechanisms based on matrices,
[2] J. M. Hollerbach, A survey of Kinematic Calibra- ASME J. Appl., Mech., 22 (1955) 215-221.
tion, The Robotics Revies, MIT Oress, Cambridge, [14] S. Hayti and Mirmirani, M., Improving the Abso-
MA., (1988). lute Positioning Accuracy of Robot Manipulators, J.
[3] B. Mooring, Fundamentals of Manipulator Calibra- Rob. Syst., 2, No. 4 (1985) 397-413.