Clase 13
Clase 13
Clase 13
A STEWART PLATFORM
FOR VEHICLE EMULATOR SYSTEMS
by
YOUHONG GONG
M.S., Materials Engineering
Shanghai Jiao Tong University
(1984)
MASTER OF SCIENCE
IN ENGINEERING
at the
Massachusetts Institute of Technology
January 1992
© 1992 Massachusetts Institute of Technology
All rights reserved
Signature of Author X
Department of Mechanical Engineering
January 18, 1992
Certified by
Harry West
Thesis Supervisor
Accepted by .
Ain A. Sonin
Chairman, Departmental Graduate Committee
ti' r'4'
< ,t;tr : I- STITUTE
F:B 0o 199
UtjRA~IE5
ARC·JI ,,;
DESIGN ANALYSIS OF A STEWART PLATFORM
FOR VEHICLE EMULATOR SYSTEMS
by
YOUHONG GONG
ABSTRACT
3
ACKNOWLEDGEMENTS
Acknowledgements 4~~~~~~~~~~~~~~
Acknowledgementts 4
TABLE OF CONTENTS
Abstract 2
Acknowledgements 4
Table of Contents 5
List of Figures 6
List of Tables 7
1. Introduction 8
1.1 Overview 8
1.2 Vehicle Emulator System 8
1.3 Contents and Organizations 10
2. Design Specifications 12
2.1 Design Considerations 12
2.2 Admittance Emulation Accuracy Analysis 13
2.3 Stiffness Analysis 23
3. Kinematic Analysis 35
3.1 Kinematic Models 35
3.2 Forward Kinematics 38
3.3 Kinematic Constraints 44
4. Graphical Simulation 48
4.1 Interactive Format 48
4.2 Design Tool 50
4.2.1 Workspace 51
4.2.2 Joint Angle 54
4.2.3 Flow Rate 56
4.3 Validation of Controller 57
5. Kinematics Error Correction 58
5.1 Error Calibration 59
5.2 Error Compensation 63
6. Conclusion and Recommendations 65
Appendix 68
References 98
Table of Contents 5
LIST OF FIGURES
ofFigures
List 6~~~~~~~~~~~
List of Figures 6
LIST OF TABLES
List of Tables 7
CHAPTER 1
INTRODUCTION
1.1 OVERVIEW
\~~~~~~
Chapter 1: Jntroduction 9
Chapter 1: Introduction 9
The force sensor measures the forces acting on the platform due to the
motion of the manipulator. The platform controller model the dynamic
response of the system to those forces, i.e. the trajectory which the
modeled system would follow if it were subjected to those forces, and
controls the six hydraulic actuators to achieve the leg lengths
corresponding to each desired platform motion and thus imposes the
trajectory of the modeled system on platform. Because VES is
programmable, the platform can emulate a wide range of different
dynamic system.
10
Introduction
Chapter J.: Introductiont 10
concludes the investigation of design of Stewart platform use as VES and
suggests areas where further work is needed. The appendices contain
derivations which are too lengthy to be included in the main body.
11
Chapter 1: Introduction
Chapter]: ntroduction 11
CHAPTER 2
DESIGN SPECIFICATIONS
Specfictios1
Chapter~~~~~~ 2: Deig
Chapter2 Design Specificatins 13
and some guidelines to the design process of Stewart platform.
tracking error
the range of motion Eqn. (2.1)
Relative error was used here, because absolute error is not a good measure
for simulation, e.g, a 2 inch error for simulation with a range of 2 feet
motion is large but maybe not so bad for a 20 feet motion.
2:
Design
Chapter Specifications 14~~~~~~~~~~
Chapter 2: Design Specifications 14
acting on it. The performance in achieving this goal is affected by the
accuracy of the trajectory generated by the admittance model and by the
performance of the platform control system. The control of the hydraulic
actuators is accomplished by analog servoamplifiers using proportional and
derivative feedback. The model of the electrohydraulic actuator and
controller design for VES were investigated by a lot of researchers. (West
et al [6], Dubowsky et al [7,9], Fresco [3], Stelman [5], and Ismail [4] )
Tracking error caused by PD controller, position sensor and servo actuator
dynamics were found small enough to be ignored for simulation within the
bandwidth of the system when controller gains were high. So, the major
error is caused by inaccuracy of the trajectory generated by admittance
model, particularly the accuracy of VES suffer from the error of the force
data obtained by the computer from the force sensor.
15
Chapter2:
Champter2: Design Spec4ications
DesignSpecifcationss 15
fo = constant = 1fsmax ]Eqn. (2.2a)
where and y are coefficients representing the quality of the force sensor,
andfs andfsmax are dynamic force the sensor measures and its maximum
value respectively.
-II
I I
I I
I I
I I
I I
I I
I.I I
I__________
I
I
I
Chpe 2: DeinSeifctos1
Chapter2: Design Specifications 16
of robot working in space. A sinusoidal motion was chosen as typical
robot motion type, i.e.,
17
Chapter 2
Chapter Specifications
Design Specifications
2. Design 17
*1~ ~ ~ ~ ~~~~I
- 4'
!l
Je
I
-
___?~~~~~~~
-- I
J = mr2
Figure 2.2 One-Degree-Of-Freedom Rotation Model of VES
Using the same initial conditions, and assuming small motion of the base,
we obtain very similar results. (see Appendix 1.2)
o= = Je 2t2 -
pme1 2 2,2
-lamax] J mr2g Eqn. (2.6a)
£ = J = _yJ2 -= me12
_a - J
~~~x
~~ mr2
Eqn. (2.6b)
18
Chapter2: Design Spec4ications
2: Design Specfications 18
The difference between these two models is that relative error for
rotation motion is proportional to the ratio of moment of inertia instead of
ratio of mass as in the case of translational motion.
T
Je I /
Z I
J= mr29 O I
__XI --I---~~~~~~~~~~~~~~~~~~~~~~
@ II
I I
o ZI me,2t2 1 1 + Me 2
IZmAu [ ri mr2 Eqn. (2.7a)
The results shows that the both offset error or gain error consists of
pure translational term, pure rotational term and a coupling term, which
reflects the difficulty of accuracy analysis when the degrees of freedom of
the system increases. However, this simple VES model relates the
simulation error with masses or inertias of the system, quality of force
sensor, frequency of robot motion and simulation time, and thus gives a
rough measure of the accuracy of the simulation for a given system
especially for the simulation of a robot working in space, and it provides
very helpful information for selecting a proper force sensor.
For VES to simulate more general base system, e.g., the suspension
system of a vehicle, the stiffness of the base system is a very important fact,
so, it should be considered in the base model in addition to the inertial
effect of the base, this two-degrees-of-freedom VES model is shown in
Figure 2.4. In the model the stiffness of the base system includes both
translational stiffness k and rotational stiffness kr. Since stiffness exists in
the base system, the base will produce a restoring force to balance the
error force caused by the offset error of the force sensor. Therefore, in
vehicle emulation case, the offset error is static and negligible and the gain
error will dominate.
Design
2.
Specifications
Chapter
Caper 2: Design Specifications
20~~~~
20
Je = mel2
I
J= mr2
- _rp_
k, kr
Using the same method and same conditions, the accuracy of VES
for the vehicle emulation could be obtained. (see Appendix 1.4)
2
M= (2
Eqn. (2.9a)
Mr=- )2
,R02_,i Eqn. (2.9b)
Specifications 21
Chapter2.
Chapter DesignSpecifications
2: Designz 21
-
= m+k
Eqn. (2.10a)
.O = = ikrkr
J+Je mrg+m1l 2 Eqn. (2.10b)
This model is consistent with previous models, e.g., let k-+O, and
krO, so, c0-O, M-1 and Mr+l, and the result will be the same as
o-O0,
that of simple two-degrees-of-freedom model. If let k->Oand kr-o, then
M---l and MrO, the result is exactly same as the result from simple one-
degree-of-freedom translation model. Like previous models, the error of
VES emulation due to the gain error of the force sensor consists of three
terms: pure translation term, pure rotation term and a coupling term. The
difference between space system and vehicle base system is that due to the
stiffness of the base system, the error now is also related to the ratio of
frequency of robot motion to the natural frequencies of the system. For a
practical case, when o << and co<< Onr, we have
M -->- and Mr
in~
02
Ct(4~~~~~ (2.11
Eqn. a)
1-M 1+2--1
02 and1 1+1
-Mr
02 -
Qo Or Eqn. (2.1 b)
£Az= - I 1 12 mel20) 2
Eg _zYmeCO2[k-
z k k+ kkr Eqn. (2.12)
2.Dsg pcfctos2
Chate
Chapter2: Design peci~fications 22
Therefore, for very low frequency robot motion, the error will be
dominated by quality of the force sensor, the mass or inertia of the robot,
the stiffness of the base system and the frequency of the robot motion.
Y
z
Y)
B1
(a) (b)
Figure 2.5 A Symmetric Stewart Platform
(a) Model of Platform (b) Top View
Chate 2
Chapter2: DesignSpecifcations 24
frame XYZ, whose origin locates at the center of the base, by the end-
effector position vector x = (x, y, z, ax, f3,y )T, where a, f3,y are roll,
pitch and yaw rotations. Let Bi = {XBi,YBi,ZBi}T, (i =1, 2, 6), be the
location of base joint center, defined as position vector with respect to
XYZ frame and let Ji = {xji,YJi,zji}Tand Pi = (XPi,YPi,ZPi}T, ( i = 1, 2,..
6) be the platform joint centers, defined as position vectors with respect to
xyz and XYZ frames, respectively. In matrix form, the transformation
from the xyz frame to XYZ frame is given by
X
R z(y)Ry(O)Rx(a) y D1 D 2 D 3 y
[D] = [D(x)]= z z
0 0 0I O O0 1 I Eqn. (2.14)
where
D 11
cospcosy
D1= D21 cos3sin-y
D31
i
I= -sin[3 Eqn. (2.15a)
I-
sinasinpcosy-cosasiny
Il
D1 2
D2= D22 sincxsin3siny+cosacosy 1
D32
sinacos3 I Eqn. (2.15b)
cosasin[cosy+sinasiny
D3= I 1
cosasinpsiny-sinacosy
I
- cosacos[B I Eqn. (2.15c)
Chpe 2:Dsg pcfctos2
Chapter2: DesignSpecifications 25
If we define leg i as a vector ij, ( i = 1, 2, .,6), we have
liy=Pi-Bi
i=
liz Eqn. (2.16a)
liy }
II Ifix -i Ii-D
li-D
1
Eqn. (2.16b)
[ liz
I
substitute Eqn. (2.13) - Eqn. (2.15) into Eqn. (2.16a), we obtain
= Dllxji+Dl2yJi+Dl3zJi+x-XBi
1i= liY = D 2 1 xi+D22YJi+D23zJi+Y-YBi
liz | D31 xi+D32YJi+D33ZJi+Z } Eqn. (2.17)
=
li =IPi- Bi 12 +2 +12z Eqn. (2.18)
11
12
! = 13 = (X)
15
16
Eqn. (2.19)
where
fx
f2
(F), = f; (F) = f3
m, f4
my f5
Im, f6 Eqn. (2.21)
are end-effector force vector and leg force vector respectively, as shown in
Figure 2.6. [j]Tis the transpose of the manipulator Jacobian matrix, which
is defined as
all all
ax ay
a12 a1 2
= ai = I ax ay
[axj
a16 a16 a16
ax ay Eqn. (2.22)
If we neglect gravity and friction, we can relate the, leg force to leg
deflection l = [ll, a12, , a16]T by the individual stiffness, which is
modeled as
where fi is the force produced by leg i and ali is the deflection of leg i.
is the spring constant. Eqn (2.6) could be rewritten in vector form
27
Chapter2:
2: Design Specifications
DesignSpecifications 27
Figure 2.6 Forces Applied on Platform
Chapter2: DesignSpecifications 28
kl
0
k2
[K] =
0 Eqn. (2.25)
where
[S] = [j]T [K] [J] Eqn. (2.29)
29
Chapter 2.
Chapter 2: Design Spec47cations
Design Specifications 29
Since [S] consists of individual leg stiffness and Jacobian matrix, it is
configuration dependent. Based on above analysis, stiffness at any point
throughout the workspace of the platform can be obtained.
30
Chapter2:
Chapter 2: Design Specflcations
Design Specifications 30
Yi = Y - YBi Eqn. (2.31b)
=
xj = i= =0
ill iml i=l Eqn. (2.32b)
ii=l
(X,)2
-I -I=
(y)2 = 3[r2+R2-2rRsin(Ci/6+(pl+p2)] = 3r*2
Eqn. (2.32c)
2
~ (XJi) (yji) 2 = 3r2
i=l i=l Eqn. (2.32d)
From Eqn. (2.17), (2.18) and (2.30), each term of Jacobian matrix
[J] can be given by (see Appendix 2.2),
ai liX DI qxj+Dl2yji+x-X3i
ax li li Eqn. (2.33a)
D2lxJ+D22YJi+Y- YBi
a i
ay
aij I:
1I ii Eqn. (2.33b)
ali yji(ll'D3) I ,
Spec4ications 32
2: Design
Chapter2: DesignSpecifications 32
(i) At home position, the stiffness in horizontal direction is the same,
i.e., it is independent of direction. And both the vertical or horizontal
stiffness of the platform depends only on its height zo, or 0, one of the
design parameters. Here zo = losinO, as shown in Figure 2.5. The vertical
stiffness increases with the increase of 0, but, horizontal stiffness decreases
when 0 is larger.
0
0 10 20 30 40 50 60 70 80 90
34
Chapter2:
2: Design Specifications
DesignSpecifications 34
CHAPTER 3
KINEMATIC ANALYSIS
Kieai
Chapter~ nlss3 3:
Chapter3: KinematicAnalysis 35
Z II
JI (Pi)/
Mi
l1i
Leg i i= zi
x 'Mi
IZ
Bie M iz
ISA B
Figure 3.1 Plucker Coordinates of Leg i
iy
Ui = iix1i (
Chpe3.Knmtc nlss3
Chapter3: KinematicAnalysis 36
According to skew theory, at every instant during the motion of a
body in space, there is an instantaneous screw axis (ISA) and the
translational velocity v and angular velocity o has a relation
where h is the pitch. From skew theory, given the displacement and
velocity of the platform, the velocities of the legs can be obtained. (see
Fichter [11])
(x\
YI
J1
(Y =X
Analysis 37
Chapter 3. Kinematic
Chapter3: KinematicAnalysis 37
frames XYZ and xyz, Cartesian reference frame xyzi, ( i = 1, 2, , ), is
denoted as the local coordinates system fixed to leg i, as shown in Figure
3.2.
The origin of xyzi is joint Bi and the axis xi points towards joint Ji.
The yi axis is parallel to the cross product of -Z and i, and the axis zi is
defined by the right hand rule. Thus the motion of the leg i could be
described by the reference frame xyzi with respect to XYZ.
These three models are essentially the same, because they represent
the same physical plant, just different in the mapping of the coordinates
from one vector space to another one. However, a different model is more
than just a varying representation of the platform, it can elucidate aspects
of the underlying theory and suggest results that might be otherwise go
unsolvable or unnoticed. Model 1 is an easy, straight-forward and efficient
model for calculation of inverse or forward kinematics and for real time
control of the platform. But, model 1 does not consider the rotation of the
legs, so generally it could not be extended to a dynamic, model. Model 2
takes consideration of the rotation of the legs and skew theory provides
qualitative and physical insight into underlying geometry of the platform
while quantitative calculation could be easier via coordinate map. Model 2
is used to calculate the rate change of the leg velocity, to determine the
singular positions of the platform and to do dynamic analysis based on
screw theory. However, the calculation using model 2 is complicated and
time consuming. Model 3 puts emphasis on each leg, so it is easy to
nlss3
Chpel.Knmai
Chapter3: KinematicAnalysis 38
analyze relative motion of the legs, such as the interference problem
between the legs, and a local coordinates system is more convenient to use
for each part of the platform as a free body, so that the equation of motion
of the system can be formulated for dynamic analysis.
x = [J]-ldl +xo
Eqn. (3.3)
Analysis 40
3 Kinematic
Chapter3: Knem~atic
Analysis 40
If 6(x)J 11 - Idl
f(x) i x) F16 l d26
Eqn. (3.4)
af a
f(x+Ax) f(x) + f Ax = f(x) + l x
ax axEqn. (3.5)
= f x)+[J] x Eqn. (3.5)
Analysis
41
Chapter 3.
Chapter KinematicAnalysis
3: Kinematic 41
could not improve the efficiency of the calculation without losing the
accuracy of the results.
42
Chapter3: KinematicAnalysis
3 Kinematic Analysis 42
state will cause the deflections of the virtual springs, which are the
differences between the current leg lengths and given leg lengths. The
platform driven by the corresponding spring forces will move towards the
equilibrium position until the disappearance of deflections of the springs.
Lyapunov stability theory is used here to derive the numerical forward
kinematics algorithm. For simplicity, we only discuss the model where the
effect of dampers are neglected. For a desired leg length Id, let x be the
current desired end-effector position estimate corresponding to the state off
the equilibrium position, and define the current error, i.e. the virtual
spring deflections as
~ [ K p] [
= 2iT
[Kp]i Eqn. (3.8)
so if we chose
Kieai nlss4
Chapter~~~~~~ 3.
Chapter3: KinematicAnalysis 43
then
t =4-[Kp] o0 Eqn. (3.11)
then, we obtain
V = -T[Kp][J][J]T[Kp]T'
T
=-AxAx <0 Eqn. (3.12)
45
Chapter 3.
Chapter 3: Kinematic Analysis
Kinematic Analysis 45
predict this phenomenon, where a cylinder contains all the geometry of the
actuator and at each time the shortest distance between two adjacent
cylinders is checked not exceeding the diameter of the cylinder. The model
using a single cylinder to represent the whole actuator, is very conservative
and will be fail to apply to an actuator whose lower portion may be thicker
than the upper portion due to the assembly of position sensor, hydraulic
hose, etc., and the locations of top joints are close. In order to overcome
this limitation, the actuator is modeled as a combination of two cylinders
with different diameters. The cylinder with a larger diameter and a fixed
length, represents the thicker geometry while cylinder with small diameter
d2 !
3:Kinematic Analysis
Chapter 47~~~~~~~~~~~~
Chapter 3: Kinemtic Analysis 47
CHAPTER 4
GRAPHICAL SIMULATION
49
Chapter 4.
Chapter 4: Graphical Simulation
Graphical Simulation 49
base. In addition to interactive control of platform position and
orientation, the program also provides sinusoidal motion and some other
types of motion. The program solve the inverse kinematics to get the leg
length, calculate joint angles and the clearance between the legs to check the
workspace violation, based on the analysis described in Chapter 2 and 3.
Also, the program calculates flow rate of the hydraulic pump according to
the amplitude and frequency of platform's motion, which is used to
establish the specification of the hydraulic pump and accumulator.
Simulation 50
Chapter4. GraphicalSimulation
4 Graphical 50
Table 4.1: Workspace Requirements of VES
Translation ± 12 in ± 12 in ±12 in
Rotation - _ ± 300 + 300 ± 300
4.2.1 WORKSPACE
Chapter
4: Graphical
Simulation51
Chpter 4: GraphicalSimulation 51
but in fact the workspace embedded in a six-dimensional space is not a
quantity. In recent years, several researchers have addressed the
workspace analysis, focus on generating planar graphical contour maps or
cross section of the workspace. (Yang and Lee [17], Fichter [1 1], Weng et
al [16], Cwiakala [23], Gosselin [24,25], Clearly and Arai [22])
52
Chapter Graphical Simulation
4: Graphical
Chapter4. Simulation 52
Table 4.2: Platform Geometric Parameters
__ _
-Platform
----------LGeometric
- ----------
I Parameters
- --- ---------
I 1 Ill|S r t
Simulation
4.Graphical 53~~~~~~~~~~~
Chapter
Chater 4 GraphicalSimulation 53
+30.
U 0.
a)
I--,
.N
.e
-30.
-100. 0. +100.
v, (deg.)
+90.
I
0 h(deg.) 0.
-90.
-30. 0. +30.
Horizontal axis (in)
C 4
Chapter4: GraphicalSimulation 54
____ 1
· a't a |.,'o
.1
eli.
...I'llml-a
... .% ' , gs,
*,nu mg
25 0°
00
!4
: "
E
,' :
· . U
: . I,.'.
Ist
'. *a
.,
M a
l
...
55
Chapter 4.
Chapter 4: Graphical Simulation
Graphical Simulation 55
4.2.3 FLOW RATE
10o
90
80
70
P" 60
v 50
50
p 40
30
LL 20
10
1
v
0.35
0 0.25
Flow In
/1
C..
EC 0.15
0.05
L··ll m w_ .-
Lan,~~~~~~
-' " -0.05
Ct -0.15
0
-
Flow Out
-0.25
I
-0.35 -- -
57
Chapter4.
Chapter 4: Graphical Simulation
Graphical Simulation 57
CHAPTER 5
= Cn + Ac Eqn. (5.1)
where ac is the parameter variations. For the real platform, leg length
vector is the function of geometric parameter c and the configuration of
the platform x, i.e.
Chapter5: KinematicsErrorCorrection 59
where I, represent the nominal leg length vector and it is exact Eqn. (2.19)
and corresponding Jacobian matrix is
'j-ca, Eqn.(5.4)
which is exact Eqn. (2.22). For the desired end-effector position vector
Xd, we have
therefore, even the VES controller is good enough to drive the legs to
reach exactly the desired lengths, there still exists an error between the
configuration of the platform and the desired configuration, i.e.
where
is the configuration deviation of the real platform when its leg length is Id.
If we expand Eqn. (5.8) in Taylor series and neglect the higher order
terms, we obtain
Id =
I(Cn, Xd) + lacx=x;C axe+ X =Xd Ax
C=Ca =CU Eqn. (5.10)
substitute Eqn. (5.4) and Eqn. (5.5) into the above equation,
Id = Id + [] Xd Ac + [ XnAx
L ax X =Xd
Eqn. (5.11)
AC Linac [J]Ax
Ic J[C x Eqn. (5.13)
difficulty of calculation of matrix [a] and its inverse matrix, and the
data not being "persistently exciting". Since we only deal with the
deviation of geometric parameters, how to delete the effects of non-
geometric factors such as backlash and joint compliance should be very
carefully considered in the experiments. In order to obtain an accurate and
stable solution, a parameter identification procedure including the method
of statistical approximation must also be applied.
where
A Id = Jax =-In
ac =
IdAc=[J]x Eqn.
(5.17)
From the analysis in previous section, it is shown that both approaches are
equivalent in terms of the compensation effect if the geometric parameter
variations are sufficiently small. However, the second method does not
require computation of the inverse Jacobian matrix and thus can be used in
the singular configurations of the platform. In addition, this method is
superior in terms of time efficiency.
Appendix 1.1
The follwing figure shows a one-degree-of-freedom translation
model of VES. The robot and platform move only along the vertical
direction. me is the mass of the robot and m is the mass of the base to be
simulated.
me
I I
I
I I
y I
I I
iZ
IMI f .
1-
I
I I
1 1
X I
I I
I I
I I
I I
---·I --
since fs = -f .. m= fs = -f
.'. mne(+Y)= -mi, or (me+m)/ = -Me
If we assume the motion of robot is a typical sinusoidal motion, y = Ysincot,
where Y is the magnitude of robot motion and w is the frequency of robot
motion, we have
y =Ycocoscot and y = -Yo2sincot = -o2y
AxiAr.
= jYme Y 2
AXIl
=m+me
Since VES uses admittance model, the Ax won't affect y, i.e., Aylaf.= 0
.. AzlIf
° = AXilf + AyIf = m+m Y2t2
lAzl me2 2
Eo = AZI1= - t2
(ii) Due to the gain error of the force sensor Afg= fs= -yf = yrnx = -mr,mne
and from the equation m(Axi)= Afg = ymni,
integration gives us
AxlIf = -m+m y=
therefore, the error of simulation due to gain error of the force sensor is
-Yn~
Eg= zi-
Z--
A=XIf=
Z
may+-
m
t
m+me y
Apedx116
Apendix 1 .1 69
Appendix 1.2
Je = me12
._ _ _ II YI
(r
_ _ ___(P
A
J= r2g
0 =0, =0 ,p = O = when t =
Je.0 2
= me1l
me12 +mr 2
7
1.
Appendix~~~~~~
Appendix 1.2 70
1-+3 Me12+mr t
the equation J(ai) if- f = PJsmax- 1 ,max, we integrate twice and get
1la&= =.2t2
J+Je
since BAfo = 0
0 = APIf =
.'. AXAf .- (e 2t2 = 12
1 . 8.oCt2
J+Je mri+m1el
(ii) Due to the gain error of the force sensor afg = yrs = -' = yJ = -i J"
and
fJ+J
te
= Afg = yJ*p,integration gives us
and from the equation J(4aq)
2
= me1 0
APlaf, = -=- jej 2
me1l2+mr
so, the error of simulation due to gain error of the force sensor is
Je0/(Je+J)
eg = A
a
=-_ = rg)2
J0/(Je+J) mr
71
Appltendix 1.2
Appendix 1.2 71
Appendix 1.3
I--- -r·--T --
Je
ZI
~~~~~~~~~~~I
= mr2
J J-,,29
XIf_ I~~~~~~~~~~~
I
Xi I
I I
= T 2
p z = x+la Jea = me12 a m.Ml1
(O+i) = I
72
Appendix 1.3
Appendix 1.3 72
Assuming zero initial conditions, integration gives us
_ 12 ,, o-
-J,+J (^1~4)
00 a 0+9 = J~e·f
-16+ 2
m1 +nm
and also
X=- mrrn, a
-(mel)(mrj) 0
(m+me)(me1 2 +m4)
Check the solution. Since initial conditions are all zero, the center of the
system must remain zero, i.e.,
mx = m( rnela)
= -M
. mez+mxO
(i)Assuming offset error of the force sensor is
Apedx137
Appenditx 1.3 73
A(lI rp J+J,
l I.42t2
since AOlar.= 0
...
male
- LAcpl _ JJO 22 = 12 4 2 t2
J+JI mrj+m,1 2
AxIlAo
= M1C . 2t2
(me+m) (J,+J)
so, we have
2 t2
= mC nl
)M J + 12]
(Je+J) (me+m)
zmax= + amax- + J
so, the error of simulation due to offset error of the force- sensor is
Az = t3o2t2mel
mem
[ J + 12]
= Izmi - - mlU Me+m
= me [ 1 +mZt2
m 12] = 2t2 -me r21 rsk
+ (1
[ +2 +
+ -e 2]
m mr2 m r8 mr 8
Afg =(Ax)
9 TgI x
fS
"S,
Apeni 1. 74
Appendix 1 .3 74
Integratingthe equation
J,&,. = Bs .. 0 Yj j
Alafs = 'yp=1 O
since aOlr = O
A=X
Assuming zero initialJ+je
conditions and integrating the following equation=-
. I
'Aia (lme+m) (Je+J)
so, we have
yJel
Azlir, = Axla + ILalif, = -
m
y Omel [ J +12]
=(Je+J) (me+m)
therefore, the error of simulation due to gain error of the force sensor is
Apedx137
Appendix 1.3 75
Appendix 1.4
For VES to simulate more general base system, e.g., the suspension
system of a vehicle, the stiffness of the base system is a very important fact,
so, it should be considered in the base model in addition to the inertial
effect of the base. On the basis of simple two-degrees-of-freedom model,
which is discussed in Appendix 1.3, we include stiffness of base system and
establish a two-degrees-of-freedom VES model. In the model, the stiffness
of the base system includes both translational stiffness k and rotational
stiffness kr. Since stiffness exists in the base system, the base will produce a
restoring force to balance the error force caused by the offset error of the
force sensor. Therefore, in vehicle emulation case, the offset error is static
and negligible and the gain error will dominate.
Je = me12
I ---- 4--
Z I
J= mr2
76
Appendix 1.4
Appendix 1.4 76
Z x+fla Ja 2
m1el r .. m-2(0 )
(o= °W2Je 0
kr- (Je+J)o2
Define
= kr
kr and Mr= 02 =
(Je+J) (ml1 2+mr2g) co2wZ- 1-_( _rw)2
co
.(p=-MrJe0
(Je+J)
aO
= +p = J+(1 - Mr)Je
Je+J
X
melo2
= = mel 2 [J+(1-Mr)Jele
k-(m+me) k-(m+nme) (Je+J)
Define
and M= C2 = 1
( (me+m)
2--l_ (1_(--)2
Apedx147
Appendix 1.4 77
'. x = - m (-M
ItJ+(l t)Je]
(mr+me) ($e+J)
Af
s =Af) fl
1!_[J -+(Mr)Je
"'. I Ax (nM+m) (Je+J)
= -ye
I A| MJe
(JO+J) }
Since AO =
Therefore, the error of simulation due to the gain error of force sensor is
78
ADpendi'x 1.4
Appendix 1.4 78
Let's discuss some special cases.
£g=AZ e+ _y2
M--- 2 and -M - 1+ 2 - 1
c= Az
= -r
_yymeO2 - _ 12
~+
] me12co2
Z k kr kkr
1.4 79~~~~~~~~~~
Appendix
Appendix 1.4 79
Appendix 2.1
1.1
y)
B1
80
Appendix 21
Appendix 21 80
since xJl+X2+XJ3 = rs(n/6+p2)+rs(l/6-qp2)-rc((p2)
= rs(7t/6)c((p2)+c(,r/6)s(p2)+s(gi/6)c(qp2)-c((X/6)s(p2)-c((p2)]
= r[O.Sc(qp2)+0.5c(p2)-c(P)2)]= 0
so e xJ=xl+xs2+xs3+xl4+xs5+xs6=xJl+Xl2+xJ3+(xJ3+xJ 2+XJI)
isl
=2(XJl+XJ2+x1 3) = 0
= 2 (XB1+XB2+XB3)= 0
e YJi=YJI+YJ2+YJ3+YJ4+YJ5+YJ6=(YJI+YJ2+YJ3)+(-YJ3-YJ2-YJI)= 0
i=l
6 6 6 6
I Xi=I (XJi-XBi)=E Xi- XBi = 0
i=l i=l i=l i=l
8
A pp di 21
Appendix 21 81
i Y= (Yli-Yi) = Yi- YBI=
iul ij l ji l iul
i xi = xy+2y2+x3y3x4y
+ 4 xsy5+x6y6
i=l
= XlYl+X2y2+x3Y3+X3(-Y3)+X2(-y2)+Xl(-Yl
) = 0
x?)2 *2 .2 *2 *2 *2 2 2 2 2
(x,2)= (X,) +(x2) +(3) +(X4) +(x5) +(x6) = 2[(xl) +(x2) +(X3) ]
i=l
= 2 [xJI+x2+x3+X Bi1++X2+X3-2J1X1--2XJ2XB2.-2XJ3XB3]
= 2 r 2 [s 2 (7r/6+ 92)+S2(7r/6-92)+2p2)+R2
2( 2 (pl )+s 2 0(/6-(p 1)
+s 2 (r/6+(p1 )]-2rR[s(7r/6+(p2)c( 1)-s(r/6-p2)s(r/6- 1)
+c((p2)s(r/6+p1)] }
= 2 r2[2(2)+2((q2 )+qs(q2)(2)+4cc(9
c2(p2)+s((p2)
-fs((p2)c(p2)+c2(p2)+R 2 [c2(p 1)+ 2(p)+ 3 s2 (p 1)
= 3[r2+R2-2rRs(g/6+pl+p2)] = 3[r2+R2-2rRc(7/3-qp-(q12)]
= 3r* 2
where r*2= r2+R2-2rRsin(r/6+p 1+(P2 ) = r2+R2 -2rRcos(0/3-p 1-(p2)
Apenix218
Appendix21 82
2 .2 .2 .,2 .,2 .2 2 ,2 .2
(Y) 2 (Y2) +(y,,)2 (Y+( 4) +(yS) +(Y6) = 2[(yl) +(Y2) +(y3) ]
iul
- 2rR[ Is((l)c((2)-4s((p1)s(92)+c((pl1)c((p2)
+sS(P1)c(2)+ c(( 1)S(2)+ lS((1)s(92)+ c(1 )S((P2)
-Is(l1 )s(9P2)] })
2
= 2{2r2+ 3 R2 -2rRi-[ I-C((p1+(p2)+ s(91+(p2)] }
= 3[r2+R2-2rRs(7r/6+ p1+ p2)] = 3[r2+R2L2rRc(r/3-qpl-(p2)]
= 3r*2
E (Xji )2 = +x 2 X
XJ12+J2 + XJ5 2+Xj62 = 2[Xj1 2 +XJ22 +XJ32 ] =
4422+
i=l
E (Y =
i)2 2
YJ1 +yJ 2 2 +YJ52+YJ62 = 2[yJ1 2 +yJ22 +yJ3 2 ] = 3r2
+YJ32+J4
i=l
83
Appendix 21
Appendix 21 83
j (yBI) 2 = 2] = 3R2
2[YB12+YB22+YB3
i-l
= -(3C
jXBi = 2(XJIXB1+XJ2XB2+xJ3XB3) -3r2-3R 2 )
im
2
= 3rRsin(n/6+p1+p2)
= - 1 (3C -3r2-3R
YJiYBi= 2(YJ1YB1Yy+YYB2+YJ3YB3) 2)
2
= 3rRsin(i/6+(pl+(p2)
= - 2 rR2s(n/6+(p2-21 )
2
E yj2iXsi= 2(yjlX B
+ yl X22 2B + 23
J X B3 ) =2r2R[c2/6+ 2))c(p1
l)
i=l-
-c2(/r6-_p2)s(//6- l )-s2((2)s(x/6+01)] = -2Rs(/6+q1-2q}2)
e XJiYJiYBi= 2(XJlYJ1YBI+XJ2YJ2YB2+XJ3YJ3YB3)
i=l
=2r2R[s( (/6+2)c(//6+2)s((pl)+s)c(/6- 6)
(l1
-c((p2)s(2)c(r/6+ ) = 2r2R3s ((pl1
)c(2(p2)---C((P1
)s(2(p2)
3-s((l)s(292)] = 2r2R[c(2(p2)s(ir/6+91)
+3c(l1)c(292)+
8 8 2
84
Aappendix21
Appendix 21 84
-s(2 qp2)c( /6+p)1)]I 2Rs(/6+p1- 2p2)
. yJiXBiYBi= 2(yjIXBIYB+YJ2XB2YB2+YJ3XB3YB
3)
is'
=2rR2[c(g/6+p2)c(p )s(pl )-c(x/6- 2)s(x/6- p1)c(X/6-VI)
) -=rR2s(/6+(p2-2pl)
-s(p2)s(x/6+pl )c(r/6+p1)1
Z XJiYJiXBiYBi=2(XJlYJ1XBIYBI+XJ2YJ2XB2YB2+XJ3yJ3XB3YB3)
i=l
=2r2R2 [s(7r/3+2p2)s(2pl1 )-2s(r/33-2p2)s(/3-2l)s(2p2)s(r/3+2p)]
4 4 4
Apeni 21 85
Appendix 21 85
=Ir2R 2( 3 s(2(p
l )s(2p2)- lc(2p1 )cs(2(p2) s(2(p l)c(2(p2)
2 4 44
3
+ fC(2p1)s(2(p2)J r2R2[.C(201 +2p2)- fs(2 p1+2(p2)J
= -Ar2R2s(n/6-2(p1-2(p2)
4
86
Appendix 21
Appendix 21 86
Appendix 2.2
Dl xji+D 1 2YJi+X-XBi
Eqn. (2.17) gives I = ity) D 2 1 Xi+D22YJi+Y-YBi
NZ1 D 3 1Xji+
D32YJi+z }
lI i= = IllD 2
I
iz I IID3
where the direction vectors DI, D2, and D3 are given by Eqn. (2.15)
coslcosy
Dli
D21 cossiny I
D31 I
-sin3 I
sinasinopcosy-cosasiny
D12
1
D2 = D22 )=f sinasinpsiny+cosacosy
D 32 I-1=L
sinacos3 I
cosasinpcosy+sinasiny
I D 13
It 1
D3= D23 cosasin3siny-sinacosy
D33 1- cosacosB Dr
li = 7 i 1~X++ 2y 12
Apeni 2.8
Appendix 2.2 87
and the Jacobian matrix will be
all all all
ax ay av
al2 a2 al2
[J]= a = ([ ali - ax ay
ai
al6 ai 6 a16
ax y . .
where
alix
- acV 1?+py+ Z ) ax =lix=DllxJi+Dl2YJi+x-XB
ax ax 2 I + I'+ ii Ii
ali
_ =
a(v/l+12
+ i )=
aX+2 + ) __21iyIay =liy =D21xii+D22YJi+y-YBi
ay - ay 2V 12 +12 + 12 li li
aliZ
a(V"_12-
,Xt 12 12 ) 2 iz-
iy i7l =z =D31xji+D32YJi+Z
Ii li
21 Ij2+1I.+ l;ZZ
since
aD I
aa
aDI _ i aD l,2
aD3,
I aD 12 I
aa cosasinf3cosy+sinasiny
aD2 aD 2 2 -J cosasinf3siny- sinacosy I=D3
aa aa (
aD 32 -II cosacoso
aa
88
Appendix 2.2
Appendix2.2 88
D3 8aD23
aD3 = aD23 -sinasinlsiny-cosacosy -D2
8d3 { | -sinacos I
Ia" I -sinopcosy
1-sinsin
aD-aD21 =-(sinaD 2 +cosaD 3)
ITai
aD -COS
aD2 aD3 (
a1I ap I1 D1
aD32 |.
aD I
aD
3 ap l cosacospcosy
IaD12
ay - sinasinsiny-cossa cosy
02
a n= sinasinocossin l-sinD3
cosacos
=--cosasin
89
Appendix 2.2
Appendix 2.2 89
aD
aD -cosasinsiny+sinacosy
aD3 = aD23 = cosasincosy+sinasiny =sinacosPDj+sin3D 2
aD33 l0
ay I
SO,
21i +aix
21 +·li
21·-az
al = a(! +12Y+i2) !iaa a aa
aa aa 2V 1 +1x+ 12
-i 1i .
li
DINali aliz
21iXaix +21ijYliY +21a
al-; ad iiiY+ lZ) ap a a1
2V I' y++1' 2
lix(-~XJi+-
ap ap ap X a Yi)+iy a YJi)
i)+ iz( al -xJi ap
li
avau~ ~ 2l~l~l
ai
ol = 0'1/2+1y+ 2_'"~-'r/'Y--'VPz'/'I
Apedx229
Appendix 2.2 90
,aD,1 aD ,.. aD a a. . . aD3 .aD3
II
r
xjcOsjicosp[csa(lD2 )-sina(Ir D 3)-yji[cosacos3(li. D )+sin1(Ir D3 )
li
= xjicos3(cosa-sina.i1
ii ii -yji(cosacosi+sin
i ii P,)ii
Apeni
2.9
Appendixc 2.2 91
Appendix 2.3
Eqn. (2.29) gives [S][(J]T[KI[J], so, for platform is at its rest
configuration, where x = (0, 0, zo, 0, 0, 0) and due to symmetry of the
configuration, each leg length will be same at this time, i = lo, ( i = 1, 2,
..., 6), if we assume that the individual stiffness of each leg is same, i.e.,
ki = k, ( i = 1, 2, ..., 6), the stiffness matrix [S] is
since x=0O,y=O, z=zo. oa=O, 3=0, y=0, the direction vectors are
liy=liy = D2 1 Xi+D22YJi+Y-YBi=Y,
=
liz=liz D 31 xi+D 3 2YJ i +Zo=zo
so,
ali _ lix Xi
ax ii lo
ali l _ yi
ay li lo
ali _ liz _ ZO
az li lo
92
Appendix 2.3
Appendixe2.3 92
Fi Yj 0
10
ali
= Xik,
ap i - j
1'i
..
a= Xj. -yJi
li X - y l i - x J IY s i
and
SI= k]t (ali.2=k: (x) 2 ) 3kr2
iml X i-l o I i=l1
where
r*2 = r2+R2 -2rRsin(1r/6+ pl+p2)= r2+R2 -2rRcos(r/3-(pl-(p2)
J1
B1I
X
93
Appendix 2.3
Appendix 2.3 93
so,
S tt - 3kCos20
2
6 a, 6 · 6
S224k ( ~ ()=kj(L)2k.(Y ,io(~~)2) 2O
(y2t )2)=3kc2 =3kcos
3kcos26
il ay i.l 1o 12il
&
i1l(~
Ss=kR 2 (42 2 4
2 [2c2=(-X+p2-c(2p1+22)+3s(l/6-2p
- 22) l
3r 2R2 [1+sr/6-2pl-2p2)]
Appendix 2.3 94
S 132 S 1 k 6 aliali
aXa) 6
X;I) o0
SS1143=
= S1 =k
S 4 1 =k x
ll)
(l axacx l (XJiyji)
xjij) =0=0
S42=
I =k i=1
6 "~~0
i=1
(6- iyi =
i=l
6
) [3rRs(i/6+p1+p2)]
S26 S61=kZi--1
(alali
" ( , = 2kzo
(l~ ) iYBi)]
[i(YJiXBi-O
S32 =k)
S23=S24
=S42
ki= ali= z =°=3kzr23rRs(t/6+ql+q2)
i=l lya i=l 12o
$26 = 62 =k (~ =k { [Yji(YJiXBi-(XJiYBi)]}
i=1 aly 120 i=l
·
Appendix 2.3 95
6
S34= S43 "ki allali - k ( Yi) =0
i-i lo il
6 6
aliali)
S 3 6= S63 =k2 = (YiXBi)-l (xJiYBi)] = 0
i=l azay iul i1
6 6
(al ali kzoI
S46= S64 =kX (YiXB)-1 (xJiYJiYBi)] = 0
i=1 caaay i= i=l
6
ali a li =-kzo [ (XJiYJiXBi-X (Xi YBi)
S56= S65 =kX
i=l a' 1o i=l i=l
= 0
Appendix 2.3 96
so, the stiffness in y direction is
Sy=Fy/Ay S 3kcos 2
Appen~ix 2.3
2.3
Appendix 97~~
97
REFERENCES
[1] Asada, H. and Slotine, J.E., 1986, Robot Analysis and Control John
Wiley and Sons
[2] Do, W.Q., 1985, Dynamic Analysis of Stewart Platform, M.S. Thesis,
University of California, Los Angeles, CA
[4] Ismail, A.N., 1988, The Design and Construction of a Six Degree of
Freedom Parallel Link Platform Type Manipulator, M.S. Thesis, MIT,
Cambridge, MA
[6] West, H., Hootsmans, N., Dubowsky, S., Stelman, N., 1989,
"Experimental Simulation of Manipulator Base Compliance" Proceedings
of the First InternationalSymposiumon ExperimentalRobotics, Montreal,
June 19-21
[7] Dubowsky, S., Tanner, A.B., 1987, "A Study of Dynamics and Control
98
References
References 98
of Mobile Manipulators Subjected to Vehicle Disturbances," Proceedings,
FourthInternationalSymposiumof RoboticsResearch, Santa Cruz, CA
[8] Dubowsky, S., Vafa, Z., 1987, "A Virtual Manipulator Model for Space
Robotic Systems," Proceedingsof NASA Workshopon Space Telerobotics,
Jet Propulsion Laboratory, California Institute of Technology, Pasadena,
CA
[9] Dubowsky, S., Paul, I., West, H., 1988, "An Analytical and
Experimental Program to Develop Control Algorithms for Mobile
Manipulators ," Proc., RoManSy, Udine, Italy, July
[12] Fichter, E.F., McDowell, E.D., 1980, "A Novel Design for a Robot
Arm," Advances in Computer Technology, an ASME Publication
[15] Vafa, Z., 1987, The Kinematics, Dynamics and Control of Space
Manipulators: The virtual Manipulator concept,Ph.D. Thesis, MIT,
Cambridge, MA
[16] Weng, T.C., Sandor, G.N., Xu, Y., Kohli, D., 1987, "On the
Workspace of Closed-Loop Manipulators with Ground-Mounted Rotary-
Linear Actuators and Finite Size Platform," Proceedings of the 1987
ASME Design Tech., Vol.II, Robotics,Mechanisms and Machine Systems,
Boston, MA, Sept. 27-30
[17] Yang, D.C.H., Lee, W.L., 1989, "Feasibility Study of a Platform Type
of Robotic Manipulators from a Kinematic Viewpoint," Transactions of
ASME Journal of Mechanisms, Transmission and Automationin Design,
Vol. 106
100
References
References 100
[20] Sugimoto, K., 1986, "Kinematic and Dynamic Analysis of' Parallel
Manipulators By Means of Motor Algebra" ASME Paper, .No. 86-DET-
139
101~~~~~
References
References 101
[26] Gosselin, C., 1990,, "Stiffness Mapping for Parallel Manipulators",
IEEE Transactions on Robotics and Automation,.Vol. 6, No. 3, June, pp.
377-388
[28] Stewart, D., 1965-66, "A Platform with Six Degrees of Freedom",
Proceedings of the Institution of Mechanical Engineers,.Vol. 180, Part 1,
No. 15
[29] Ahmad, S., 1988, "Analysis of robot drive train errors, their static
effects, and their compensation," IEEE Int. Conf. Rcbotics and
Automation,
References~
~ ~ ~ - -- - 102
References 102
[32] Payannet, D., Alton, M. J., and Liegeois, A., 1985, "Identification and
compensation of mechanical errors for industrial robots," Proc. 15th Int.
Symp. on Industrial Robots, Tokyo, pp. 857-864.
[33] Wu, C.-H., and Lee, C.C., 1984, "On an accuracy problem of robot
manipulators," Proc. 23rd IEEE Conf. decision and Control, Las Vegas,
Dec. 12-14, pp. 1636-1637.
[34] Wu, C.-H., and Lee, C.C., 1985, "Estimation of accuracy of a robot
manipulator," IEEE trans. Automatic Control, AC-30, pp. 304-306.
[35] Ziegert, J., and Datseris, P., 1988, "Basic considerations for robot
calibration," Proc. IEEE Int. Conf. Robotics and Automation, Philadelphia,
April 24-29, pp. 932-938.
103
References~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Referenzces 103