Escolar Documentos
Profissional Documentos
Cultura Documentos
Regina S u n - K m Lee
uisitii and
8' bgraphie Sewicm
3
Acquisitions et
services bibliographiques
Abstract
'Lhwarm is a VariableGeometry-aims (VGT) manipidator, developed and
Th-
its acctuacy d m fiom its phpicai size and complexily. However, the accuracy may be improved with c i ~ - 1 o o ppoeition conttoL
Acknowledgments
I wodd Iike to express my sincere gratitude to many people who made this
work possible.
First of dl, 1 w d d like to t h d my supervisor, Prof. Peter C.Hughes,
for his academic and technical advice throughout my graduate studies. 1 also
would like to thank my doctoral cornmittee members, Prof. Gabriele MT.
D'Eleuterio and Dr. Rank Naccarato, for th& technical advice in all aspects
The collvematio~~
with my cdleagues in the Spsceaaft Dynamics and
Robotice gmup have been invaluable to me. 1 am eepecially indebted to
Thierry Cberpillod for his technical support on the design and constmction
of 'km-.'
1 also wish to thank rny parents, Sang-Wook and Jung-Ja Lee, and my
sister, Sophia, for theh love and support. Their constant encouragement and
patience have been and will always be appreciated. Lsstly, 1wodd like to take
this opportunity to th& my hiuband, Thierry*Without his love and support
and most of all, his faith in me, this work wodd not have been posgible.
July, 1999
Maple, Ontario
Contents
List of Figues
Cbapter 1 Introduction
............................
1.1.1 Hyper Redundant Manipulators . . . . . . . . . . . . .
1.1.2 Variable Gbmetry 'Ihies Manipulatom . . . . . . . . .
Thissarm . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Motivation and Prnpose . . . . . . . . . . . . . . . . . . . .
.
Literature Eteview . . . . . . . . . . . . . . . . . . . . . . . . .
1.1 Background
1.2
1.3
1.4
1
1
2
4
Manipulatom
14
OverviewofthisDocument
............
...................
15
1.4.2
1.5
.......................
Chapter 3 System D d p t i o n
17
2.1.1
lhsmnnActuator
....................
17
17
.................
Siipporting Stnictiue . . . . . . . . . . . . . . . . . . . . . . .
2.2.1 M
o .........................
2.2.2 Passive Members . . . . . . . . . . . . . . . . . . . . .
2.2.3 'Ihissarm Coordinate System . . . . . . . . . . . . . .
2.1.2
2.2
sexvomechanism
19
20
20
20
21
22
23
2.3.5
...................
Hyper-Redmdancy Redution . . . . . . . . . . . . . .
Kinematics . . . . . . . . . . . . . . . . . . . . . . . .
GALIL corxuna~ds . . . . . . . . . . . . . . . . . . . .
Motion Execution . . . . . . . . . . . . . . . . . . . . .
2.3.6
Cartesian Control
2.3.1
2.3.2
2.3.3
2.3.4
Tkajectory Planning
.....................
Chapter 3 Kinematics
3.2
3.3
..........
23
24
24
26
.......... .
.. . . .
Kini?maticsofa'lkussarmModule . . . . . . . . . . . . . . . .
3.2.1 InverseKinematics . . . . . . . . . . . . . . . . . . . .
3.2.2 Forward Kinematics . . . . . . . . . . . . . . . . . . .
Joint-Limit Avoidance Problems . . . . . . . . . . . . . . . . .
3.1.2
23
25
23
28
31
31
35
37
42
..................
43
..................
44
.........
Differential Inverse Kinematics . . . . . . . . . . . . . .
50
51
..................... ....
Workspace Measmement Methods . . . . . . . . . . . .
'hssarm Workspace Experiments . . . . . . . . . . . .
Controllable Workspace . . . . . . . . . . . . . . . . .
Reach H i e r d y Algorithm . . . . . . . . . . . . . . .
59
5.1.2
60
5.1.4
5.1.5
........................
5.2.1 Deanition . . . . . . . . . . . . . . . . . . . . . . . . .
5.2.2 Dexterity Measmement . . . . . . . . . . . . . . . . . .
RepeatabiityAnalysis . . . . . . . . . . . . . . . . . . . . . .
5.3.1 Definition . . . . . . . . * . . . . . ..
.. . . . . . . .
5.3.2 Repeatabii Measurement Method . . . . . . . . . . .
5.3.3 Ihisarm Repeatabiliw . . . . . . . . . . . . . . . . . .
SpeedAnalysis . . . . . . . . . . . . . . . . . . . . . . . . . .
5.4.1 Actuator S p d . . . . . . . . . . . . . . . . . . . . . .
5.4.2 TipSpeed . . . . . . . . . . . . . . . . . . . . . . . . .
5.4
59
5.1.1 Definition
5.1.3
5.3
58
........................
Definition . . . . . . . . . . . . . . . . . . . . . . . . .
Poeition Control . . . . . . . . . . . . . . . . . . . . . .
Kinematc ('Ii.acking) Control . . . . . . . . . . . . . .
On-Line Cartesian Contd . . . . . . . . . . . . . . . .
Velocity Tkacking Control . . . . . . . . . . . . . . . .
5.5.3
5.5.4
5.5.5
68
71
73
75
75
76
82
82
82
84
85
85
85
88
88
89
90
96
97
6.3 Implementation
..........................
..........................
..........................
Chapter 7 Conclusion
103
104
108
110
. . . . . . . . . . . . . . . . . . . . . . . . . . . 112
7.2 Cont~bl~tiom
7.3 FlltureWork . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
References
................................
115
List of Figures
................*............
3-DOF VGT at NASA . . . . . . . . . . . . . . . . . . . . . .
'Lhrssann Mark 1 at UTIAS . . . . . . . . . . . . . . . . . . .
Tkussarm
.
Plansr VGT,Caltech . . . . . . . . . . . . . . . . . . . . . . .
Serpentine 'Ihiee MdpuIator, NASA Kennedy Space Center
..............................
Payload Inspection end Processing System . . . . . . . . .
Tetrobot
.........................
Actuator Geometry . . . . . . . . . . . . . . . . . . . . . . . .
Hinge Assembly . . . . . . . . . . . . . . . . . . . . . . . . . .
ThControl System . . . . . . . . . . . . . . . . . . . .
-Mue
................
Kinematia of a discrete m d e i . . . . . . . . . . . . . . . . . .
Vectors in the kinematica of a Ihissarm module
........
66
5.8
.....
Workspace Boimdary Approximation . . . . . . . . . . . . . .
workspace of 1 Modiile . . . . . . . . . . . . . . . . . . . . . .
Workspace of 2 Mod~des . . . . . . . . . . . . . . . . . . . . .
workspaceof 3 Modules . . . . . . . . . . . . . . . . . . . . .
Woxkspace o f 4 Modules . . . . . . . . . . . . . . . . . . . . .
Tkussarm Workspace by Sweeping Method, sz-plane . . . . . .
5.9
67
.......................
68
......................
5.12 Contracted Configuration . . . . . . . . . . . . . . . . . . . .
5.13 Extended Configuration . . . . . . . . . . . . . . . . . . . . .
5.14 Pitive x Reach . . . . . . . . . . . . . . . . . . . . . . . . .
5.15 NegativexReach . . . . . . . . . . . . . . . . . . . . . . . . .
5.16 Wor-e
of PUMA761 mruiipulator . . . . . . . . . . . . . .
68
62
5.3
65
5.4
5.5
5.6
5.7
66
66
66
67
69
69
69
69
72
..........
..........
73
...
75
..........
5.21 Dexterity Measures dong a straight line (2-axb) . . . . . . . .
5.22Paetiuel . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.23 Ptiue2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.24Ptu.re3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
78
74
79
80
80
80
.............................
80
...........
5.27SetupforRepeatabilityAnalysis . . . . . . . . . . . . . . . . .
81
83
...................
84
5.25Poetiire4
5.26 DexterityMeamesatPOehne1,2,3and4
5.28 RepestabityTestResultl
. . . . . .. . . .. .. . . . . . .
5-30 Repeatability Test Result 3 . . . . . . . . . . . . . . . . . . .
5.31 Vibration During Acceleration and Deceleration . . . . . . . .
5.32 The 12ich Pick-aad-Place Maneuver. . . . . . . . . . . . . .
5.33 Square Tkajectory in Amiracy Analysis. . . . . . . . . . . . .
90
91
91
92
93
93
5-29 RepeatabityTestResult2
xz-plane
&q
84
86
87
...
...
98
99
99
.. . .. . . .. .
l2
...... . ... .
105
..
. . . .. . . . . . . . . . . .
106
. . . . . . . . . . . .. . . . . . . . . . . .
107
. .. . . . . ... ... . . . . . .. . .. .
107
Control
. . . . . . . . . . . . . . . . . .. . . ,
.. . . . . . .
10s
. . . . . . . . . . . . . . - . . . . . . . . . . . . . .. . .
109
Chapter 1
Introduction
1.1
1.1.1
Background
Hyper Redundant Manipulators
Hyper-teciundant manipulators are those that have very large degrmatic redundancy. These manipulators are often r
of kin*
and the
grasping of objects. As Immega and Antonelli (1995) pointed out, "the pomi-
by the imagination."
1.1.2
numemus applications for VGT structures have been exBrnineci, inciuding the
manipuiation of payloads.
tor of the same volume and mass, were realiaed early in the development:
stiffness, modularity, flexibiity, and dexkrity. buse of these advantages,
VGT rnanip~ilatorshave been considered for applications in space and undersea explorations and inspection. Examples of proposeci applications for VGT
More recentIy, VGT teEhaology has been ale0 amiderd for manufac-
idators, they have adwtages for tasks that require changes in configuration,
or tasks in a constrained environment.
In order to develop an efficient control scheme for h m ,the following research objectives have been p r o p d
a to efnciently model the kinematics of 'Ihuisarm
O
model with a vety simple mathematical presmtation. One of the contributions of this dimertation in the field of VGT kinematics is the development of
a kinematics model at the velocity laval. The model pnsented hem is computstionally simple and easy to imp1ement.
Many hyper-redundant VGT d p u l a t o t e remained latgdy at the experimental level, primarily due to the difEiculties in the implernentation of
dynamics, or other relevant featues of the eystem. Many papers have been
curve dopes to pass through the vertical, without numerical dScultiesn and
aiso "aows for more flexibiity in controlhg the curve shape."
the reference shape c m , several spa curves were mggestecl inciuding Bezier
Bezier akes to
the backbone curve by using four independent shape fimctions. Their method
was refmed to as the continuum approach and has been applied in numerou8
task." An &rinsk scheme was p r e f d and it was suggested to use splines for
parameterization of the backbone cuves. Regarding the reference shape curve
and quintic splines. One of the advantages of this method is that the user can
directly incorporate the end-effector pose constrainta into parametric equations
of the backbone c m ,
d e 1 3 of
3-DOF eztensible
Mayhew, 1997aj. More direct methods of solving the kinematics problem for
1.4.2
While meny papera have been written on the design and kinematical analysis
of VGT manipul8tora, only a handhil of VGT manipulatom have been actudy
constructeci and experimentdly andyzed.
'
Reinholtz, 19881
i,
Figure 1.3: 3-DOF VGT at NASA Figure 124: 'Ihmm Mark 1at UTIAS
0
1994,199?j
The 3-DOF VGT at NASA (Figure 1.3) and the IhuissrmMark 1 (Figure 1.4)
are both SDOF VGT manipulators. In fact, they may be utilized as building
blocks for a larger and psibly redmdant VGT mrrnipulator. The 3-DOF
NASA VGT bears a atrong resernblance to the TRissMn Mark 1. In hia study,
10
Figure 1.7: P
k VGT,Caltech
Chapter 3.
tapered twebrry VGT (Figure 1.5) seem to belong to the same family of m e
nipulators dedopeci by LRC, which has Leen uivolved with numerom expar-
Robertsbaw and Reinholtz's two-bay VGT is one of the fin&VGT manipulators developed by LRC to dernonstrate deplqrment concepts. This spi+
System
onto a ZDOF &ai joint. The dimensions of the f h t VGT module are iden-
tical to the 3-DOF VGT in Figure 1.3. The seamd module is d e r in size.
of 5.49
VGT manipulafor hae the edvantage of king moulair and extmnely easy to
assemble and disassemble; however, it share~with Thrssarm an accurq prob-
&al VGT manipulator developed at the NASA Kennedy Space Center. This
manipulator is a mbsystem of the Payload Inspection and Processing System
(Figure 1.9).
manipulatot. Based on the eXpenen leamed h m this exercise?s e v d recommendations were made by Williams and Mayhew including reducing the
actuator noise, and ensuring that the control system aiIowa all motors to reach
ports modularity in not only the hardware configuration but also the software
architecture. The system is reconfigtuable with different applications.
Three
and the 3-DOF NASA VGT. The main diffaence is that 12 of the 21 links
in the doub1wxtahedral TETROBOT are actuatom The d p u i a t o r is 1.5
me-
kilogram and the peak end4ector speed is 6 m / s . The tetrahedral manipulator TETROBOT is corn&
1.4.3
Two approachts have been taken ta the bernatics problem of Thssarm: the
link-length metho and the peta angle m a .
The petal angle method wm used by Hughes and Naacarato (1991), Sallmen (1993), end Naccarato (1994). In all three cases, a dosed-form solution
was achieved for the i n . kinematics, but not for the fimuard kinematics.
Oikawa (1995) briefly corisidered the diflerentid imerae kinematics of
(1991), and Naccarato (1994). Instead of the petal angles, the link-length
1.5
frame of rderence used in the following chapters is also definecl in this chapter
in terms of the physical components of a h a r m module.
kinematics of each module are solved in terms of joint variables (8, 4 and I )
related to the discrete modeL The joint limit avoidance problem is crleo briefly
considemi.
Tnmsarm. A Jacobian matrUt and Hessian are derived using a discrete model.
For n modules (a 3n-DOF system) the system Jacobian matrix is 372 x 3n. ELements of this rnatrix are expresseci in terms of (8, A 6). The Jacobian mat&
of the lhmsarm module is &O solved.
racy are describeci. Through the series of sirnulatian tests, these measu~ements
control acheme is to apply coame control to brbg the end-efktor clthe target pOBture, then to apply fine position mntrol baseci on the
to
Beneor
Chapter 2
System Description
In general, a mmipulator pcissesses three attribut-:
actuators, supporthg
In the
h attributes of Ttvsaarm-
Ihissarm Actuator
The Thmsarm actuators are the most basic substructures of the Thissarm control systern. They are independent linear actuators, each dnven by a DC motor
and a gear reduction assembly (pulleys and a drive belt). The motion of each
motor is monitored and controiled by an optical encoder and senromechaniism,
respectiveiy. The date of each actuator, Le., the distance that the Iead screw
has traveled h m the zem position, is reiated to the encoder counts (M) as
ratio, and L the lead screw travel ratio. The factor 4 is included because the
2,771, 48 = 211,199 counts. F'or actuator 12, the same length equds 0.0762 x
1,351,734 = lO3,2 counts.
v
Motors
Mluiiit8cturer
Encoder
Counts
Tkansnrission Ratio
Gear Ratio Lead Screw
n a d
Factor
The enornechankm of each actuator is controUed by GALa motion controllers. Two DMC1080 controlks have been installeci on the wntrol oomputer to provide PID control. DMCl080's operate in the biie stmcture.
tion Inc. The gains selected for Thssarm actuators are ahown in Table 2.2
and are btued on the step responae of each motor.
2.2
2.2.1
Supporting Structure
Modules
Each 'Ihisami module is a double-octahecai pardel mAnipulator and possesses three tctuators of the bear electric lead-mew type. The lead mew
assembly acts as a bearing th& converts rotary motion to linear motion. Each
actuator is fasteneci at one end directly to the hinge amembiy that connecta
the actuator to the other end of the next actuator. As a result, tbree actuators
form a triangular actuator plane (see Figure 2.2). When fUlly extendeci each
actuator m
2.2.2
Passive Members
See Figure 2.3. This aasembly provide8 the rotetional degrees of keedom
needed at the juncture between the passive structure and the actuator plane.
At the base of the lateral triangles, three pacsive members, made of 0.64
cm cidl rod, form a base or top trimgth plane. A top plane of one module
and a base plane of the next module are connected with a passive comector
to connect contiguous modules.
2.2.3
h s a r m Coordinate System
of the plane are denoted hi, L,b,in clockwiee direction. The geometric
center of the base triangle, 00, is denoted the amter of the base Bame, FO.
orthonormai to the x and z-axea The actuator above bot is denoted actuator
1 and actitators 2 and 3 are numbered in clockwiae direction.
The geometnc centei: of the top plane, 01,is the cent- of 4.The
describe the codguiation of module 1. Fieleo coincides with the baee kame
+ 1 kames in total.
In the actual manipulator, the base plane of module i + 1 and the top
physid planes.
2.3
2.3.1
Tkajectory Planning
In order to demonstrate the kinematic controI, a simple path generatioa h c tion has been added to TCE.A lin-,
2.3.2
Hyper-Redundancy Rasolution
2.3.3
Kinematics
For a given end-&ector path, two DMC filas are prepared by TCE. Each
2.3.5
Motion Execution
For motion control with 7 or more actuators, both boards need to be selected.
For the kinematic control, two DMC files must be externdy downlosded to
DMC1080 boards. Then, fkom the motion menu, the program on two boards
are executed simuitaneously.
2.3.6
Cartesian Control
Using the terminal mode, the current Carmian position of the end4ectOt
may be manitoreci and commandeci.
Chapter 3
Kinernatics
The motion of a manipulator is often epecified by the mtion ad/or orientation of its end-effector. If the manipulator ie redUILdllllt, soma functional
specification is additionally regnired to resolve the rediindruicy. HHamn and
mode1 approach, first developed by Zanganeh, the author and Hughes (1997),
is presented in Section 3.1. In Section 3.2, the forward end inverse kinematics
of a IhL9881plmodule are exBrnuleci. In Section 3.3, another issue in kinematic
--
T h e g e o ~ d b ~ t i c aao ~f d u l e i e p ~ b y t h r e a
independent vatiablcs, O,+, and 1. As shown in Figure 3.1 (i), a gimhl m d e
wneisCB of a iuiiversa jad with tno rotational DOF and one prismatic joint.
the actuator plane) of the module. Instead of one prismatic joint, two coupled
priamatic joints are used (Figure 3.lCi)). The 1 4 kame
is assigneci to
the ith gimbal module such that its center is at the base of the nrSt prismatic
joint
is described
&O
3.1.2
dLPmte modeL Expressxi in the base hune, FO,the position and orientation
of the end &tor
where t ami Qt represexst position and orientation of the end &ector with
respect to Fn,
and a h ,
r
as the dundancy ~
methode are often at the velocity level. They are baseci on the manipulator
is configuration control.
The method proposeci by Zangaaeh, Lee and Hughes (1997) is to minimize the cost fimction, 2,of the sy~tem,aven as
while sat-g
where x, ami n, are the desireci position and orientation of the endeffector.
The Lagrangian approech is adopted to h d the solution set, q where
where X is the vector of Lagranpian multipliem. For n gimbal modules @ = 3nDOF system) with m kinematics constraints ( m = dim(g)
(3.14) and (3.15) lead to p
+ dim(h)), eqs.
of q and A.
Next, consider
we have
may be derived:
where
=Q&.
By d e t i o n , $ is
for i = 1,. .. ,n. For each i, the fimctions can be differentiated with respect
Inverse Kinematics
In this section, the inverse kinematics probiem for a ' I h s a x m module is considered. The objective is to solve for the actilator lengths of the jth module,
qj aie specined.
d=I~+f2=l(~+~)=diil
(3.29)
where nI is the unit normal of the actuator plane. The unit direction vector
of the top triangle, fi2, may be cornputeci using 0 and 4:
where
By definition,
sint#
a
n,and 73.
plane fkom the plane formed by petal nodes, ri, r~and Q) lies in the direction
of iil.Aence, for i = 1,2,3, we have
where
is the ith column of the matrix e that tepresent the the base plane,
lm is the side hgth of an ieometric petai triangle and hl is the side length of
hl
= 29.14
(3.44)
cm
lm = 27.33
(3-45)
til
(3.46)
= 0.95
62 = 4.83
(3.47)
constants.
Substituting q(3.40) into eq.(3.39), we obtain the following equation
for the petal mgles:
[(CI
Q C Q B ~ ~ ) ~
- b&)lT
d =O
(3-48)
where
Once the petal angles are cdcttlated, nociai coordinates are found h m eq.(3.40).
LBPtly, the actuator leogths are computed from the nodal coordinates:
The remaining nodes on the base triangle and the top triangles are
determineci by simple trigonometrie relations. The base nodes are
(3.58)
The top nodes for i = 1,2,3, are cornputeci using symmetric propertias,
i.e.,
b, = Rolbo,+d
(3.59)
& = to,-6ih
(3.60)
and
3.2.2
Forward Kinematics
The fornard kinematics problem deals with solving for enhffector pmition
when the actuator lengths are given. ki this section the solution for q = (O, 0,l)
are considered when & = (kl,ly, 4)is piven.
Upon substitiiting eq(3.40) into eqs. (3.54), (3.55), and (3.56), we get
numericaliy. Then,
or, a = 2ni,.
fi2:
resolution
actuator lengtbs of the jth bay. Hence, k j is a function of Bi, #j and lj. h o ,
,ka%).
( q )
w-v
= O
5 O
(3-77)
i-1,-
,3n
(3.78)
where Z(q) is the original objective function, P(q)is an i m p d penalty function, and the scalar r, is a multiplier which determines the magnitude of the
"the interior penalty hinction method," and "the extendeci interior penalty
hinction method." The nret method is to penaze the objective fimction only
design algorith. The second method penaiizes the objective function as the
design oppmache~a constraint, and the last method is the combination of the
fArst two*
r(
W ~ Y 3)-
2. If the solution convesges - that is, if the 801ution q satisfies the kinematic conetraint, w(q) = O - end the procedure.
3. If w t , the penalty parameter r, ie in-
to "~r,
and the procedure is
repeated.
[m=[O.h]
--
m=[O,Q]
(3.82)
g,let us define
is the i n . e Jacobian of the jth 'IhLssarm module and its solution wiii
be discussed in Section 4.3.
(ii) V
a
c
m on the base p h e
*Y
Chapter 4
Differential Kinematics
In the previous chepter, it was shcnun how m c u l t it is to h d an explicit
cld-form solution to the kinematim problem of h m . In genesal, e
lutions to the kinematic problem require numerical approech. An alternative
approach to the kinematic control problem is to examine the differential relk
tionship between the end-dfctor coatiguration, u (z, and fi-), and the joint
If the Jacmbian matrix, $,ie invertible, Le., if the manipulator is not redundant and has no kinematic singtilarities, the joint velocity vector 4 can be
easily obtained by
when O is specified.
task coordinates."
k, is relete to the equivaient dirwete model's joint velocity vector, q, and its
inverse hction is presented.
4.1
- n) x rn, dim(zi)==mx
1, and
Let iis defme the la-hand-side of the above equation as N. The problem in
Compareci to eqs. (3.14) and (3.15), the total number of equations hm been
d e c r d &an m + p to p. Upon differentiating eq.(4.8), with respect to tirne,
we obtain
4.2
In this section, we concentrate on the solution for W.in eq.(4.11). Note that
is
the andytical solution for $ is given in Section 3.1.2. The solution for
derived here. By dennition,
(c)=,
$($f)Tand &
where
and
while
and
respectively,
for a, j E (1,
--- ,
n, i
#j,
1),
. Moremer,
In rnatrix form,
O,
in the fonn
(e)
and
& (h).
= 0. Fhthemore, we
a.
2 and
First consider 2.
We now derive
The second derivatives of the rotation matrice8 Qjmay be derived from the
aml
aar
in eqe(4.14),
which is part of RMRC equation, (4.12). We are still misahg the solution
for
Wl depends on the number of hematic constraints, rn, we can not find the
general solution for the problern. However, an example case with m = 3 is
Wlis obtained as
4.3.1
Differentiating the above equations with respect to qi, for i = 1,2, and 3,
yidds:
2,we obtain
geh
2,namdy,
m eq.(3.49), we have
In order to solve this equation, we now need the s01utioas for f80,
K
hi. First, let us conaider
&,
8
and
$.Rom q(3.50):
f3),
their
where
6q
= (Ro
an, + g(-p
8%
81
+ h)T(-)l
+
(-)
a!?
Bq
Bq
[iiT(ng + w]
ternia
of derivatives of
where
4.3.2
where
and
for
to
with respect
2as
Rom the definition of the unit normal of actuator plane, fii, the following
relation is derived.
where
Note that
aArl Brl
-=---
Bk
Brz
akd
(4.99)
eq. (4.97) may be expandeci using the properties of the cross prcxiuct,' Le.,
where
The Ja~obianis, in fact, the collection of eqs. (4.105), (4.104) and (4.106),
Chapter 5
Performance Measurements
The robotic applications have been proposeci in numerous areas where the use
of humane iii imprELCticai and undesirabie. The following meas are examples
of the application of industriai dpulators:
O
repeatab'ity
speed, and
the f0-d
plied for position contrd and tracking control, respectively. In Section 5.5.5,
5.1
5.1.1
Workspace Analysis
Definition
The wotkspace of a mmipulator is the total volume swept out by the endeffectoc as the mruiipulatorexecutes dl possible motions. Two types ofworkspace
are o h considd. a reachab1eworkspace and a dexteroua workepece. Spong
is the set of pointe that "the manipidator can reach with an arbitrary orien-
tation of the end &ectorn In this section, only the reachable workspace of
Thssarm is considered. The dexterous workspace of Ihiasarm wiU be considered in the following section.
5.1.2
that yield end-effector position^ on the xz-pIane. Such joint states can be
uniqiiely d&ed
ber of discrete configurations amall, three joint states are chosem km=, 0,
and
&.
for
for an open-link kinematic chah by Korein (1985). First, the k t joint in the
method is applied to the nth module to determiae the workspeca of the last
module. Then, joints in the module n- 1are VIUjed to detamine the workspace
8. = tan-'(
c
.
,
CO$
- Pi,
- Pi&
pasition
a Determine points pj
< Bi 5
( Bj =
W),j = 1,-.* , N oN
lie.
Figure 5.4 shows the workspace of one module generated by the sweep
ing method. The nine actuator displacements that were used in the e x t a i o n
math& are also marked on the aame gmph to permit cornparison between
the two methode. Figure 5.5, 5.6 and 5.7 show the rorkepece of 2, 3 and 4
modula respectiveiy. The boundary of workspace is marked with a dotted
iine. The ehaded region an each plot ie the workspce of chah 1, 2 and 3,
respectively. Figure 5.8 is the coilection of the results lrom Figure 5.5, 5.5,5.6
and 5.7. It can be e d y aeen how quickly the workspace of Tivssazm grows
method and the sweeping method. Rom this coniparison, it is noted that
the extension and 8weepi.q methods provide very emiiler redts. However,
because sweeping is only the first-order approximation, it does not yield all
possible points. Instead, only a good approximation can be obtained. On
the other hand, this method is more computatiomlly efficient and less tirne
consiuning*
La-
uu-
sa)
....
u+
u
U
La
5.1.3
with animation was devdoped to simulate the motion during the worirspace experiments. Figures 5.10 through 5.15 show some exampIes h m the computer
simulation. Table 3.1 Iists the achiator lengths in cm useci in the simulation
tests and also the resulting end-efllitor coordinates. Even though egch of the
twelve actuators can extend aad antract up to & 7.62 cm, a f 4.0 to
mi extension wse
I 6.0
Contracted
6.0
Extended
6.0
Positive X Reach
4.0
-4.0
Negative X Reach
J-Shape
-4.0/4.0
6.0
6.0
4.0
-4.0
-4.0/4.0
Demonstration
formed with ?hiasarm at a rdatively low speed. The rnotor coordinates were
computed from the joint coordinates in Table 5.1. A set of DMC programs
were used to perform the experiments. Using the high-1evel control software
a r m workspace
tests:
Note that throughout the program, the DMC commans DT and CD are re
peatedly d.
These commands ere used for contour mode motions. The
contour mode results in smoother motion of the rxtanipuiator. The contour is
characterized by the pmition increments (CD) of ach actuator and the time
on some mernbers c81lSed this problem. For example, in the Pmitive 2-Reach
configuration, actuator 3 was unda a very large lord. It was noted that at thie
point the ectiietor was, in fact, moving s l o w and in a jerky m m e r . When
the commandai joint length exceeded 6 cm, the motion of the actuator came
nearly to a stop.
factors on the stnictural limitation, the size of the fimctional workspace highly
tors m o t be modeleci quntitatidy at this tirne, it was clear that this uncertainty had a significant && on the performance of 'Lhtssarm. This result
auggests that 'Ihuiserm, or auy other 'Ihrsssrm-like VGT manipulatore, may
in Length.
The workspace of ' I h m m is comparable to that of conventiod induetrial manipulators. For a~csmple,PUMA761 is an example of a typical
5.1.4
Controllable Workepace
So far, the workspace has been considered as a set of ali possible points in
space that may be reached by solving the forwatd kurematics, given the joint
that every point in such eprce may be actudy reached when the invercur h
c
+
matics ie eolved. In fact, ody a suluet of points in the reacbable worirspace
.3 rn d u s
TheBl;edtba-lem-varbnrIthmeriyfactasmchse
redundancy ramiution ulpritam, joint limft and dqpkity avoichm md-
on the xt-plane (y = O). Note thst maqy pointa P the ICBBChBble workspace
are in contro11A)ileworbpsce. Haraves, there are pointa mmr the ege of the
In Figure 5.17,5.18, and 5.19, 'oswee marked when the inseme kinematics was
succmstully s
0
1
d with joint coordinates nithin the joint limits, and k' was
5.1.5
In order to utilize the entire workspace, a control algorithm that makes use
of the workspace information needs to be implemented. One approach is the
regch
5.2
5.2.1
Dexterity Analysis
Definition
Dexterity Measurement
5.2.2
F i ,Ymhikawa (1-a)
(5.3)
selin, 19901:
w h e .,
o
and a ~ ste
n the largest and s d m t singular d u e s of the system
Jacobian matrix. Thie is a memue of the eccuracy with which forces can be
Third, the minimum slgular value, u d was also wed as the measme of
nearness to the ainpuienty [Klein and Blaho,
by the niinim~un
singular value approaching zero.
where q: is the center of the range of each actuator. This method provides
are a c u v e on the XI-piane and a straight iine almg the z-mis. A simulation
test was petfiormeci dong these trajectories and fow d u e s (q, c, o-
and E )
si@c811t.
that "it is not pomible to condude that one is superior to the others since the
actuator Iengths are equai throughout d four modules. Poeture 4 is when the
iength change of actiiators in module 1 are maximized and the otha modides
remain unmodifieci. J h t e r i t y Me88u~ementsat four postures are shown in
Figure 5.26. These measurements suggest that the second poshire is the mo&
dexteriw-
Danlll-h=-glrn+
..
:::;:::;....;:;::::::;..::::::::;;.:;::::::.:::::::::
..........................................................................................
................................
*
.....................
. ...\.
: . . . . . . . . . . . . . . . .: . . . . . . . .i .................. :. ........ :. . . . . . . :..........
iiiii!:il~~ii!lfi!iti!i!fi!tfi2i!!i!ii!iii
:LI:
..........Li.........
a i r -
..S....,..........
...,,
: .: . . . :. . :. ;: :. . : .:i
......................
..........
. . . . . . . . ..........
. .- . . . .
.........
L...
......
\.
..........
....:
....................
. - . . .........
...............
......................
........ ,. . . . . . . . . . .
S . . . . . . . . ,
It should be noted that the measurements were made with the @vaient k e t e mode1 approach, d i s c d in Chapter 3 and 4. Thus the result refiects the micmecopic geometry of the gimbal modules, rather than the
5.3
5.3.1
Repeatability Analysis
Definition
5.3.2
In order to measure the repeatabiiiw, a very fine pen was meci as a tool.
A white board was set up such that pen tip wodd touch the surface only
when it cornes to a stop aRer the motion. Unf~rtmately~
due to the stnicturd
vibration, short trams of the tool path were marked on some of the teet results.
test points. The result wee scannai and the radius R of the c i d e aurrounding
the scattered points memureci. The test points used are shonn in Table 5.2.
w2
Test 3
'
io,i5,0
10.10.10
0.11 cm'
0.10 cm
In order to measure the repeatability of the manipulator more accurately, a better method to locate the end4ector tip position is desireci. For
hitute referen, three-dimewioonal position emor or a vision system are recornmended*
5.3.3
Trussarm Repeatability
simple &ta
b m th-
tests.
5.4
Speed Analysis
5.4.1
Actuator Speed
(traas-
sembly. The motors are rated at 6,500 RPM (machuni speed with no load).
However, amplifiers that supply the power to the motors Lmit the actuator
speed to 2,925 RPM. Since quadrature encoders output 4 counts per motor
rotation, 2,925 PRM conesponds to 4 x 2,925 = 37,440 counts/sec. This
speed has been used as a guidaline for the rmuchum ectuator speed. 37,400
counts/sec comesponds to apprahatdy 1.4, 2.6, 2.1 and 1.4 cm/= for ac-
5.4.2
Tip Speed
Once the m8Icimum actuator speeds are determineci, the maximum tip speed
can be either caidated using the forward kinematics mode1 or measured ex-
p --j-- 2.1 cm
2.0 cm
L.J
,
36,432137,440 r~ 1 sec. Thue, the maximum tip speed for horizontal motian
a 12-inch path dong the x-, y- and z-axes. The maneuver is as follows: starth
1)
M m d m hy 1 inch, (6,-1)
l h i z ~ speed,
~ d fi Vertid Speed, V. Total T i e , & ,
x-8Xi8
16.9 cm/s
3.2 an/s
6.8 seconds
y-axis
z-axk
16.9 cm/s
3.2 cm/s
6.8 seconds
13.5 cm/s
20.0 seconds
3.2
in conclusion, the maximum tip speed of'Ihissarmis 20 cm/s in horizo11ta1motion, and 5 cm/s in vertical motion. 'Ihuniarm is capable of perfo-g
5.5
5.5.1
Accuracy Analysis
Definition
5.5.2
Position Control
and the paeition of the end-efFector was recorded, using a pen as a tool. The
setup is shown in Figure 5.27. The forwad kinematics was solved to compute
the Cartesian amdinates of the end-effector that correspondeci to the desired
joint coordinates. Then, the results from forward kinematics and the a c t d
Sample4-, 9.4
O
Sample 5
19.1
Sample 6
--
-.
O
14.1
O
10.1
0.2
0.1
10.0
O
20.0
O 10.0
O
15.0
O
O
4.2
6.1
4.5
P
. = (xa,h,za) is the actuai pen paeition and Pd = (xd,M,Y) is the deeired
Cartesian coordinates. The position error, (bP,I is d&ed es:
It was found that the position error was dway < 6 %. Note that the paeition
error also includes the measurement error.
5.5.3
Startmg at the home pition, the Pen wae moved to the upper lefk corner of
the zz-plane, (-5.0, 0, -5.0) cm,then the wume was t r d clockwise. See
Figure 5.33. The speed of the pen was ch-
the test conducted with Modulee 3 and 4 only is shown m Figure 5.34. Figure
5.35 shows the configuration d the backbone c m whik tradring a square
with 2 modules. The pedonnance appears to be vexy g d in the zdirection,
whe theze ie a 2 mm diacrepancy in x.
ali
ffouf
trajtxtory of the pen. The tracking error increased from 2% to 2.7%. Similm
r d t s were obgerve in the cnilar trajectory, shcswn in Figure 5.39.
The tracking error inmemes when the square is drirwn on the zyplane.
is shown below:
PA i8 the abdute position commsnd, aid SP is the command to set the joint
speed. As shown in the ibuve DMC progam, both speed and poaition at eech
control point dong the trajecfory was specified to ensure the ~ m o o t hmotion
betwean points. A very simple appMWm8tion was utilized to cornpute the
speed of each joint.
for i = 1,2,.
a path. To msitain 1.25 cm/s tip speed, AT was set to be 4 seconds, when
ation and negative infinite deceleration in the joint motion. In ptwtice, ody
In addition tu the trapezoidal pronle, Salimen (1994) also considered kimguiar and sinmicial velocity profiles for velocity control of anissarm Mark 1.
However, since the d e r a t i o n of Ihisearmactuatorsis &ciently hi& (maximum d e r a t i o n hae been ratecl at 2.7 cm/$),
the d e c t ofpmt d e r a t i o n
5.5.4
The next
set of experhents invoived Certeeiaa control with on-line mmputation. In
other wotds, the same d d a t i o n e fot fo-d
and inverse kinematics were
pedorme in reaEtime uaing the IhiesarmControl Environment software.
commlrnds axe
undataking
a series of actions. For exampIe, when the command WR is iseued, the controuer first sen& TP command (a DMC cammand to retrieve the current motor
counts) and, b&
TW
the motor coordinates and thia new aet of motor coruitn is sent off to DMC
boards using a DMC cornniand, PA.
5.5.5
where
5 )5 27r1
p, is the position vector of the center of the chde, r is the radius of the circle,
and % denotes the orientation of the plane on wbich the circle lies. For
example, the &de used in tndchg mntrol on -plane
o ilT.
[rad/sec].
Several simulation te&s wem performed with various nt and some examples with r = 0.4rn, and p, = (O,0,1.75),are shown in Figures 5.40 and 5.41.
The solid line in the figures repreent the desired tmjectory, zee,obtained by
for & = zO.It was sssumed that the speed of the eed-eEector19, is constant
O',QI
fAa'
= 20
The dotted lines are the r d t of the simulateci resolved motion rate
control. The initial conQuration, cf, was caldateci by configuration optimization as diScutseci in Chapter 2. The joint velmities q' was calcdated as
with ne = 20, 50, 100, 150, 200, and 1000, and the nom of the errors are
plotted againet
In thb
docity,
gigivea
Ji is the Jacobian mat& of the jth module and 4 is the 3 x 3 identity matrix.
See eq.(4.1 1) for definitions of W
;'
and i>,.
Chapter 6
kinematics
h m Cartesian
sensors coUect feedback idormation regerding its actual position. The Mer-
6.2
Simulation Test
Control, Square
$2-plane
l h s s a m mode1 was based on the seriee of data collected hom the pwious
experiments.
The same trajectories (circular and square paths) use in accmacy analysis were usai for c i d l o o p control again. A eolid line in Figure 6.3 and 6.4
in m
computed.
i 6.5.
6.3 Implementation
At the time of writing, a global eenaor was not avsilable. Thus, the proposed
shows the architecture for the control method. Figure 6.l(iii) shows how the
fine pdtioning mntrol
aui
pedofmaince.
size to the radiw of r=5 cm. Figures 6.6 and 6.7 show the simple resuits from
Figure 6.6: Wts lrom Cloeed-bop Control, Square Ifajectory, n=2, xzplane
Chapter 7
Conclusion
7.1 Final &marks
In th* work, a number of bues regarding the contrd characteristics of 'Pnieeemi
the end-effector position is on the xz-plane. The positive and negative 4 (r*
tation about the y-axis) cames an end-effector motion in either the negative
or positive xdirection. On the other hand, the relation between the motion
of the end-eff'tor and the lengths of the actuators is nonlinear end c m not
be described simply.
m of freedom, or even
a different mnfiguration.
in order to hande joint Mt amidance; and, the madel can be also used for
other types of VGT manipuhtocs.
to 7 %
were observed when kieematic control waa applied. In order to improve the
tracking mors, a
on
The
and automation. The r d t s obtained from this work may be used to explore
the industrial applications of '2hiseann and other VGT manipulators.
7.2
Contributions
was defined.
d .
desi@
'Ihissarm technology, there are still many areas of research that may be purmed for hiture improvement in Ihiseannpesformance. A few suggestions are
made here.
local sense. Park and Bloclrett (1994) also examineci the mathematical representation of the dexterity of a d p d a t o r m global sense. The global dexterity, or overall dexterity, can be used as a design criterion. Howewr, to
the megeuranent shodd not depend on the choi of the coordinates, or the
ment,or that involve high-epeed motion. 'Ihissam dynamics have been considered by the author [Lee and Zsnganeh,l997].The next step would validate
and imprm the propoeed 'Ihuruum dynIuaic8 modd using the a c t d joint
and force sensor data fkom aiurieaipl.
References
Chirikjian, G.S. and Burdick, J.W. ''Kinematics of Eyper-redundant Manipuletors," Advances in Robot Kinemotics :P m of the &d International Worbhop,
wth empli&
Chirikjian, G.S. and Burdick, J.W. "Design and Experiments with a 30 DOF
Robot," in P m of the LEEE Intemational Confmna on Robotics and Automation. Atlanta?GA, pp. 113-119 (May,1993).
grated Manuf-
Pm.
of
(1995).
Klein, C.A. and Blaho, B.E. "Dexterity Meamne8 for the Design and Control of
Kinematicdy Redmdant Manipdators," The International Journal of Rokta
Reseanch.
m of the 7th
%W.
pp.842-868 (1977).
I)usses.
Phanabhan, B., h,
V., Reinholtz, CCP. "Closed-Form Inverse Kinematic
Anirlysis of Variable-Geometry Thss Manipdators." Zhwtactions of the ASME.
114,pp.438443 (1992).
W.H.,T
*
?
S.A., Vetterling, W.T. and Fannery,B.P. N ~ m e n d
Rmpes in C: The Art of Scientfic Computhg. Cambridge University Pr-,
Pr-,
Mu-
U.S. Amy Research Office Workshop on Smart Materiab, Stmctures, and Mathematicai issues, Bladrsburg, pp. 105-120 (September 1988).
on Synzbolic Cornpu-
Spong, M.W., and Vidyasagar, M*Robot Dynamics and ConkoL John Wiley &
(1994).
3-9. (1985).
Zanganeh, K.E., Lee, R.S.K. and Hughes, P.C. "A Disaete Mode1 for the Configuration Control of Hyper-Redtmdant Manipuiator," in Pmc. of the IEEE In-
Zaoganeh, K.E. and Angelee,J. UA FormeJYnn for the Analpis and Design of
Appendix A
An Example of a Contrained
Optimization Problem
Before we iscuss an example of the constrained opthhation problem, let
for = 1, ... ,m. Equatiom (A.1) and (A.2) consist of i +rn equations. In eq.
(A.2), ho-,
is active, &(x)
problem, let us examine a system with two variables, X I and 2 2 , and with -0
constraints, xi 5 bl and q 5 62. The objective hinction, is
Let us also define Lagrangian multipliers y1 and y2. Then, the problem is to
whese
There are four istinct cases: (1) bi > O,b2 > O, (2) bl > O,
c O, (3) q < 0,
c 0.
Case 1
-
(A.9)
(A. 10)
(A*11)
>O, m d Z * = b i .
Case 3
The nrst inequality constraint is active; the second is inactive. Therefore, y1 > O and a = O. In this caee, we have
(A.13)
(A.14)
Hence, XI = bl, q = O , y1 = -2bi > 0, and Z* = b;.
Case 4
Both consttaints are active. Thesefore, ,y1 > O, yz > O and
aL =
-
2 ~ ~ + ~ = 0
(A.15)
hl
dL
0x2
= 2x2+m=O
OL
-
XI-bi
aL
-
X2
hl
asri
-b
(A. 16)
(A.17)
(A.18)
without trying them ail. For large m, the p b l e m is still difficult to compute
in rd-time no matter han "good" the algorithm is. The joint-Et avoida n problem preeented in Section 3.3 incorporates 12 ineqdty constraints,