13th World Congress in Mechanism and Machine Science, Guanajuato, Mexico, 19-25 June, 2011

IMD-123

Kinematic Analysis of the TSAI-3UPU Parallel Manipulator

using Algebraic Methods
D. R. Walter
University of Innsbruck
Innsbruck, Austria

M. L. Husty
University of Innsbruck
Innsbruck, Austria

Abstract This paper discusses in detail the TSAI 3UPU parallel manipulator. This very special parallel manipulator was presented by Lung-Wen Tsai in 1996. Up to
the knowledge of the authors practically all published work
about this manipulator deals with the property that it exhibits pure translational motion if it is properly assembled.
Quite evidently the question arises which type of motion occurs, when the manipulator is not properly assembled for
translational motion. Here it will be explained, how the
manipulator can be described by a set of algebraic equations. This set is used to analyze its motion capabilities exhaustively using methods from algebraic geometry. It turns
out that the manipulator has theoretically up to 78 solutions
of the direct kinematics (including complex ones), and that
is has in addition to the translational mode four other operation modes. It is shown that there exist poses where a
transition from one assembly mode to another is possible.
All necessary conditions on the leg lengths are determined
which lead to changes of the operation mode.

in which a change of operation modes is possible.

This paper is organized as follows: A description of the manipulators design is given in Section II. The deduction of
the constraint equations follows in Section III. These equations are used to discuss possible assembly modes and operation modes in Sections IV and V. Finally the possibility
of changing the operation mode is discussed exhaustively
in VI.
II. The manipulators design
First of all an exact description of the manipulators design is necessary. Due to the fact that the TSAI 3-UPU can
be obtained from the SNU 3-UPU by simply rotating each
limb by 90 degrees about the axis of the prismatic joint, the
following is almost identical to the description of the design
in [6].
B3

B2

Keywords: TSAI-3UPU manipulator, direct kinematics, assembly

mode, constraint equations, linear implicitization algorithm, operation mode, primary decomposition.

y
d3

h2
d2

I. Introduction
The TSAI 3-UPU parallel manipulator was presented by
Tsai in 1996 [1] as a new 3-dof manipulator to generate
pure translational motion. This mechanism and its generalizations were discussed with respect to kinematic properties
in e.g. [2], [3], [4], and [5]. It is interesting to note that almost all of these papers focus on the translational operation
mode. On the other hand it is well known that the manipulator also exhibits some other operation modes. Especially in
[2] it is pointed out nicely that the manipulator only shows
translational motion if it is properly assembled.
Due to the fact that in [6] it was possible to find all different
operation modes of the SNU 3-UPU manipulator, one could
ask if this can also be achieved for the TSAI version of this
manipulator, especially because both manipulators are very
similar.
Using an algebraic description of the manipulator and
methods from algebraic geometry a complete description
of the manipulators operation modes can be given. Additionally conditions are presented which lead to assemblies,

B1

A2

A3

d1
y

h1
1

A1
2

Fig. 1. The numbers at the limb from A1 to B1 describe the order of the
rotational axes of the U-joints.

In the base there are three points A1 , A2 and A3 forming an equilateral triangle with circumradius h1 (Fig.1).
The frame 0 is fixed in the base such that its origin lies
in the circumcenter of the triangle, its yz-plane coincides
with the plane formed by the triangle and its z-axis passes

dominic.walter@student.uibk.ac.at
manfred.husty@uibk.ac.at

13th World Congress in Mechanism and Machine Science, Guanajuato, Mexico, 19-25 June, 2011

through A3 . The same situation is established in the platform. There we have an equilateral triangle with vertices
B1 , B2 , B3 , circumradius h2 , and the platforms frame is
denoted by 1 . The parameters h1 and h2 are the two first
design parameters, which are always different, if not mentioned explicitly.
Now, each pair of corresponding points Ai , Bi is connected
by a limb with U-joints at each end. The length of each limb
is denoted by di and adjusted by a prismatic joint. The first
and the fourth axis are embedded in the base resp. platform
such that each of them is tangent to the corresponding circumcircle (see Fig. 1). The second and the third axes of this
link-combination are parallel to each other and perpendicular to the axis of the limb and its first and fourth axis. All
together we need five parameters to describe the design of
the TSAI 3-UPU mechanism: d1 , d2 , d3 , h1 and h2 , where
the first three of them are used to actuate the manipulator,
the latter are fixed design parameters.

correspondence between all spatial displacements and the

projective points [x0 : x1 : x2 : x3 : y0 : y1 : y2 : y3 ] of the
so called Study-quadric S P7 which is a semi-algebraic
set described by
x0 y0 + x1 y1 + x2 y2 + x3 y3
x20 + x21 + x22 + x23

0 2
x0 + x21 x22 x23
@ 2 (x1 x2 + x0 x3 )
2 (x1 x3 x0 x2 )

1
2i
3

2 (x1 x2 x0 x3 )
x20 x21 + x22 x23
2 (x2 x3 + x0 x1 )

1
2 (x1 x3 + x0 x2 )
2 (x2 x3 x0 x1 ) A .
x20 x21 x22 + x23

For the inverse operation, namely to obtain the Study parameters from a given displacement matrix, there exists a
quite nice method introduced by Study himself. Due to
space limitations this method, although used in the paper,
is not explained here, see e.g. [8] for further information.
Due to the fact that the limbs do not allow free motion
of the platform, there have to be some constraints on the
Study parameters describing the platforms possible poses,
i.e. only a subset of the Study-quadric describes all possible poses of the manipulators endeffector frame. In the following equations depending on x0 , x1 , x2 , x3 , y0 , y1 , y2 , y3
and describing this subset of S are deduced from all three
transformation matrices T1 , T2 , T3 .
First of all half-tangent substitutions for all uji are performed to get rid of the trigonometric functions. This results in displacement matrices Ti containing the new parameters tji . Then from each matrix Ti the Study parameters are computed using Studys method, leading to expressions

(1)

x0
d
0
0
0

(2)
(3)

where the expressions for MT and MR are as follows

2 (x0 y1 + x1 y0 x2 y3 + x3 y2 )
2 (x0 y2 + x1 y3 + x2 y0 x3 y1 ) ,
2 (x0 y3 x1 y2 + x2 y1 + x3 y0 )

where the matrices Fji are fixed transformations, matrices Mji are responsible for the rotations about the axes
of the U-joints, depending on rotation angles uji , and the
-matrices manage the transformations from one rotational
axis to the next one. Due to the special arrangement of the
axes in each limb no separate transformation matrix was introduced for the active prismatic joints, their parameter di
appears in the matrix 2i . A clear and brief introduction of
the Denavit-Hartenberg convention and the systematic deduction of the forward transformation matrices Ti can be
found e.g. in [7] and [8]. The DH parameters are nearly the
same for all three legs Li and read as follows:
a
0
di
0

= 0
6= 0

The relation between points of S and the corresponding displacements is established by the 4 4 matrix operator
 2

x0 + x21 + x22 + x23
0
,
(4)
MT
MR

III. Deduction of constraint equations

In the following equations are derived which describe the
position of 1 with respect to 0 , and with it the position
of the platform with respect to the fixed base.
First of all for each limb Li a chain of 4 4 transformation
matrices can be given, according to the Denavit-Hartenberg
convention, such that the position of 1 with respect to 0
via limb Li can be described by the matrix product
Ti = F1i M1i 1 M2i 2i M3i 3 M4i F2i

IMD-123

/2
0
/2

y3

= f1i (t1i , t2i , t3i , t4i )

..
.
= f8i (t1i , t2i , t3i , t4i ) i = 1, . . . , 3

(5)

From each matrix Ti we obtain a parametrization of a subset of S depending on four parameters. This subset essentially describes the motion capability of one limb. The final step is then to eliminate from each parametrization the
corresponding four parameters to obtain equations which
contain only the Study parameters and, of course, the design and motion parameters d1 , d2 , d3 , h1 and h2 . This can
be achieved easily by using the linear implicitization algorithm (LIA) published in [9]. This algorithm is a method to

TABLE I. DH parameters occurring in matrices 1 ,2i , and 3

It has to be noted explicitly that in Table I the d has nothing

to do with di .
Next, the so called Study parameters are introduced, which
are a very convenient way to parametrize spatial displacements, see e.g. [8] for a concise review of that concept.
The most important fact here is that there is a one-to-one
2

13th World Congress in Mechanism and Machine Science, Guanajuato, Mexico, 19-25 June, 2011

g7 : (h21 2 h1 h2 + h22 d23 ) x20 + 4 (h1 h2 ) x0 y3 +

find all equations of given degree which are fulfilled by a

parametrization like (5).
Because the kinematic chain of each limb has four degrees
of freedom, represented by the four parameters in (5), the
elimination will result in two constraint equations and the
constraint equation S. Each of the three eliminations yields
equation (2) and two other quadratic equations, i.e. all together there are seven equations in P7 whose solutions describe all possible poses of the manipulator, given that (3) is
fulfilled. This result was checked by performing the elimination with other methods and it can be said that definitely
not more than these seven equations are necessary to describe the motion capabilities of the
TSAI manipulator. After some simplifications to remove 3 from at least some
equations the following equations are obtained:
g1

x0 y0 + x1 y1 + x2 y2 + x3 y3

IMD-123

+ (h21 + 2 h1 h2 + h22 d23 ) x21 + 4 (h1 + h2 ) x1 y2 +

+ (h21 + 2 h1 h2 + h22 d23 ) x22 4 (h1 + h2 ) x2 y1 +
+ (h21 2 h1 h2 + h22 d23 ) x23 4 (h1 h2 ) x3 y0 +
+ 4 (y02 + y12 + y22 + y32 ) = 0

This system of algebraic equations describes the mechanism. Fixing the motion parameters di it can be asked now
for all projective points in P7 which fulfill all these seven
equations, under the condition that x20 + x21 + x22 + x23 6= 0.
These points represent then all possible poses of the platform which are the solution of the direct kinematics of the
TSAI 3-UPU manipulator. Because it is more convenient
to do all computations in affine space, without loss of generality, the following normalization equation is added:

(6)

g8 : x20 + x21 + x22 + x23 1 = 0

g2 : (h1 h2 ) x0 x2 + (h1 + h2 ) x1 x3
x2 y3 x3 y2 = 0 (7)

IV. Solving the system

Now the system of equations (6)-(13) is studied. In the
following this system of equations is always written as a
polynomial ideal. Therefore, the ideal that has to be dealt
with is

g4 : (h1 h2 ) x0 x3 (h1 + h2 ) x1 x2 +
+ 2 x1 y1 + 2 x3 y3 = 0 (9)

g5 : (h21 2 h1 h2 + h22 d21 ) x20 + 2 3 (h1 h2 ) x0 y2

I = hg1 , g2 , g3 , g4 , g5 , g6 , g7 , g8 i,

2 (h1 h2 ) x0 y3 + (h21 + 2 h1 h2 + h22 d21 ) x21

2 (h1 + h2 ) x1 y2 2 3 (h1 + h2 ) x1 y3 +

+ (h21 h1 h2 + h22 d21 ) x22 + 2 3 h1 h2 x2 x3

2 3 (h1 h2 ) x2 y0 + 2 (h1 + h2 ) x2 y1 +

where each gi here denotes the polynomial on the left hand

side of the corresponding constraint equation. First of all
the following ideal is inspected, which obviously is independent of the joint parameters d1 , d2 , and d3

+ (h21 + h1 h2 + h22 d21 ) x23 + 2 (h1 h2 ) x3 y0 +

+ 2 3 (h1 + h2 ) x3 y1 + 4 (y02 + y12 + y22 + y32 ) = 0

(10)

J = hg1 , g2 , g3 , g4 i
Allthough that ideal isnt that complicated, it is tried to
make it simpler. The computation of the primary decomposition of J shows that it indeed can be written in a very
simple way:

3 (h1 h2 ) x0 y2

+ 2 h1 h2 + h22 d22 ) x21

2 (h1 + h2 ) x1 y2 + 2 3 (h1 + h2 ) x1 y3 +

+ (h21 h1 h2 + h22 d22 ) x22 2 3 h1 h2 x2 x3 +

+ 2 3 (h1 h2 ) x2 y0 + 2 (h1 + h2 ) x2 y1 +

2 (h1 h2 ) x0 y3 +

(13)

Now it is guaranteed that no solution of this final system lies

in the forbidden variety described by x20 +x21 +x22 +x23 = 0.
The downside of the normalization is that for each projective solution point two affine representatives as solutions for
(6)-(13) are obtained. This has to be taken in account when
different solutions are counted.

g3 : (h1 h2 ) x0 x3 (h1 + h2 ) x1 x2
4 x1 y1 3 x2 y2 x3 y3 = 0 (8)

g6 : (h21 2 h1 h2 + h22 d22 ) x20 2

(12)

(h21

J =

6
\

Ji

i=1

with

+ (h21 + h1 h2 + h22 d22 ) x23 + 2 (h1 h2 ) x3 y0

2 3 (h1 + h2 ) x3 y1 + 4 (y02 + y12 + y22 + y32 ) = 0

(11)

J1 = hy0 , x1 , x2 , x3 i, J2 = hx0 , y1 , x2 , x3 i,
J3 = hy0 , y1 , x2 , x3 i, J4 = hx0 , x1 , y2 , y3 i,
3

13th World Congress in Mechanism and Machine Science, Guanajuato, Mexico, 19-25 June, 2011

J5 = h(h1 h2 ) x0 x2 +(h1 +h2 ) x1 x3 x2 y3 x3 y2 ,

(h1 h2 ) x0 x3 (h1 + h2 ) x1 x2 x2 y2 + x3 y3 ,
2 x1 y1 + x2 y2 + x3 y3 , x0 y0 x1 y1 ,

IMD-123

the fact that always two solutions of a system describe the

same position of the platform, each number has to be halved
(see paragraph below (13)). In the following only these essentially different solutions are considered. The number of
solutions for each system Ki in the generic case is

(h1 h2 )2 x20 + (h1 + h2 )2 x21 y22 y32 ,

(h1 + h2 ) x32 y0 3 (h1 h2 ) x22 x3 y1 2 x22 y0 y1

|V(K1 )| = |V(K2 )| = 2, |V(K3 )| = 4,

3 (h1 h2 ) x2 x23 y0 + (h1 h2 ) x33 y1 2 x23 y0 y1 i

J6 = hx0 , x1 , x2 , x3 i.

|V(K4 )| = 6, |V(K5 )| = 64

Actually all these ideals are prime ideals and there are
no embedded components. The primary decomposition
was computed over Q[x0 , x1 , x2 , x3 , y0 , y1 , y2 , y3 , h1 , h2 ]
to find possible changes of dimension, which could occur
for special values of h1 and h2 . Fortunately ideal J5 has
always the same dimension, i.e. dimension 3. For the zero
set or vanishing set V(J ) of J it follows that

So all together there are 78 essentially different solutions

for generically given parameters h1 ,h2 ,d1 ,d2 , and d3 , i.e.
78 possible poses of the platform, theoretically. It is clear
that for arbitrarily chosen parameters all these solutions will
be complex. Mechanically this means that the manipulator cannot be assembled because of e.g. too different limb
lengths. But of course, when design and joint parameters
are chosen thoughtfully, many of the solution poses also
can be real. In Section V design parameters will be given
such that for almost all Ki all solutions of the corresponding system are real. Using reasonably chosen parameters,
attempts were made to maximize the total number of real
solutions. Surprisingly this number never exceeded 28. A
strict proof for 28 to be an upper bound for real solutions is
missing.

V(J ) =

6
[

V(Ji ).

i=1

Now the remaining equations are added and by writing

Ki := Ji hg5 , g6 , g7 , g8 i the vanishing set of the whole
system I can be written as
V(I) =

6
[

V(Ki ).

B. Solutions for equal limb lengths

In the following subsection it is assumed that all limbs
are of equal length. Because of the structure of the equations this is a non-generic case and has to be treated separately.
d1 := d d2 := d d3 := d

i=1

So, instead of studying the system as a whole, each of the

smaller systems Ki can be inspected separately. Then the
solution of system I is the union of the solutions of the subsystems.
It can easily be seen that the last set V(K6 ) is empty because
K6 contains equations {x0 , x1 , x2 , x3 , x20 +x21 +x22 +x23 1}
which cannot vanish simultaneously. Therefore it is only
necessary to study systems K1 , . . . , K5 .
Furthermore, it has to be noted, that for each assembly
mode described by a solution, there exists another solution
where the manipulator can be assembled mirrored with respect to the base.

Now the same computations can be performed which were

done in the previous subsection to obtain the Hilbert dimension of each ideal. Due to the fact that there are less parameters all Groebner bases can be computed without specifying
any parameter. For the dimension the result is the same as
in the previous case.
dim(Ki ) = 0, i = 1, . . . , 5.

A. Solutions for arbitrary design parameters

In the following subsection all computations are performed under the assumption that the five design parameters are arbitrary i.e. generic. To find out the Hilbert dimension of each ideal Ki the necessary Groebner bases are
computed for general parameters, except the basis for K5 ,
where randomly chosen parameters h1 ,h2 are substituted
before the computation of the basis. So, for arbitrary design parameters the result is that

When the number of solutions is computed for each system

and halved afterwards the following results are obtained.
|V(K1 )| = |V(K2 )| = |V(K3 )| = 2,
|V(K4 )| = 6, |V(K5 )| = 60
Here there exist theoretically 72 solutions for the platforms
position, six less than before. Concerning the question
where they could have gone, one should not forget that we
have that forbidden subvariety on the Study-quadric. It is
possible that for special design parameters a solution lies
on this subvariety. In the previous section it was mentioned
that for special design parameters 28 real solutions can be

dim(Ki ) = 0, i = 1, . . . , 5
which means that all sub-systems have finitely many solutions. Reusing the computed bases from above the number
of solutions can be determined for each system Ki . Due to
4

13th World Congress in Mechanism and Machine Science, Guanajuato, Mexico, 19-25 June, 2011

obtained. Actually it was a set of parameters with equal

limb lengths. The exact values for the parameters were
h1 = 12, h2 = 7, d1 = d2 = d3 =

IMD-123

limb lengths which are now free parameters. Equation (13)

is used to simplify the matrix entries, if possible, i.e. if three
unknowns of (13) are 0, the remaining unknown is set to 1.
Furthermore the position of the platform is described by a
series of simpler transformations, starting from the planar
home position.

181
13.923
13

A very important difference to the SNU 3-UPU manipulator which was discussed in [6] is the fact that here in general
all 72 solutions have multiplicity 1, i.e. particularly the solution which corresponds to the so called home position
has multiplicity 1, and not 4, which is the case for the SNU
3-UPU. One could conclude that the TSAI should show a
better behavior in the home position, not being that shaky
like the SNU. The home position can be seen in Fig. 1 and
is described by the following values

System K1 : Translational Mode

{y0 = 0, x1 = 0, x2 = 0, x3 = 0}

1
2 y1

2 y2
2 y3

x0 = 1, x1 = 0, x2 = 0, x3 = 0
p
y0 = 0, y1 = d2 (h1 h2 )2 /2, y2 = 0, y3 = 0.

0 0 0
1 0 0

0 1 0
0 0 1

This is the operation mode which was discussed in almost

all articles about this manipulator. The transformation matrix was simplified by substituting x0 = 1. From the matrix it can easily be seen that it can be parameterized using
y1 , y2 , y3 . It is possible to compute the necessary condition
for singular positions by making use of the determinant of
the Jacobian matrix of the equations of K1 . The result is an
equation of degree 4 in d1 , d2 , d3 and looks as follows:

The following section will deal with the most interesting interpretation of the systems K1 , . . . , K5 . So far they
were simply seen as a decomposition of the original system
of equations, that made solving the system at least a little
easier.

d41 + d42 + d43 d21 d22 d21 d23 d22 d23

V. The manipulators operation modes

Until now d1 ,d2 and d3 were treated as fixed design parameters. In this section they will be treated as parameters
which are allowed to change, i.e. the behavior of this mechanism will be studied, when the prismatic joints are actuated. Computation of the Hilbert dimension of each ideal
Ki with d1 , d2 , d3 used as unknowns shows that

3 (h1 h2 )

(d21

d22

d23 )

(14)

+ 9 (h1 h2 ) = 0

Further inspection shows that for given values for h1 and

h2 describes a Kummer surface in the joint parameter space
d1 , d2 , d3 . The Kummer surface is a very famose algebraic
degree four surface, possessing the maximum number of
16 double points (see e.g. [11]). The part of the surface
in the first octant is the interesting part. If the parameters
d1 , d2 , d3 are chosen inside resp. outside the pipe (Fig.2),
both solutions of this system are real resp. complex. All
solutions are real for e.g. the parameters

dim(Ki ) = 3, i = 1, . . . , 5
where dim denotes the dimension over C[h1 , h2 ], in
contrast to dim which denotes the dimension over
C[h1 , h2 , d1 , d2 , d3 ] as in the previous sections. It follows
that in general the 3-UPU manipulator has 3 DOFs.

h1 = 12, h2 = 7, d1 =
As it was shown in [6] for the SNU 3-UPU manipulator
each subsystem Ki of a mechanisms set of equations corresponds to a specific operation mode of the manipulator. In
the following each system Ki will be discussed separately,
particularly with regard to the type of motion and possible
singular poses. It has to be mentioned explicitly that the
singularities, where a change of operation mode can occur,
are not the subject here. They will be discussed in the next
section.
The following algorithm is applied in the next paragraphs:
Each system Ji is solved and the solution is substituted into
the transformation matrix (4). From the obtained results
properties of the solutions of the sub systems Ki can be
deduced and from these follow properties of the motion of
the platform. It is absolutely not necessary to use equations
(10)-(12) for this inspection, because they describe only the

243
611
670
, d2 =
, d3 =
21
7
21

On the other hand, if the parameters of the limbs are chosen

such that the condition (14) is fulfilled, it is not difficult
to show that both solutions coincide and the corresponding
positions are described by

1 0 0 0
0 1 0 0

p 2 0 1 0
p3 0 0 1
i.e. the singular positions of the translational mode are
those, where base and platform lie in the same plane.
System K2 : Twisted translational Mode
{x0 = 0, y1 = 0, x2 = 0, x3 = 0}
5

13th World Congress in Mechanism and Machine Science, Guanajuato, Mexico, 19-25 June, 2011

IMD-123

Fig. 2. The singularity surface of the translational mode.

1
2 y0

2 y3
2 y2

0 0
1 0
0 1
0 0

0
0

0
1

Fig. 3. The singularity surface of the twisted translational mode.

0
0

1
0

2 (x0 y2 + x1 y3 ) 0 x20 x21

2 (x0 y3 x1 y2 ) 0 2 x0 x1

This is a mode which was already mentioned in [10]. Here

x1 = 1 was used to simplify. Each solution of system K2
corresponds to a rotation of the platform about its normal
axis N by 180 degrees and a subsequent translation. It follows that the described operation mode is basically a pure
translation. To parameterize it one could use y0 , y2 , y3 as
parameters. Once again it is possible to compute the necessary condition for singular positions. The result is again an
equation of degree 4 in d1 , d2 , d3 :
d41 + d42 + d43 d21 d22 d21 d23 d22 d23
2

3 (h1 + h2 )

(d21

d22

d23 )

2 x0 x1
x20 x21

Solutions of K3 correspond to poses of the platform where

it is coplanar to the base. To parameterize this planar operation mode one could use x0 , y2 , y3 in connection with
x20 + x21 = 1, where x0 is responsible for the rotation of the
platform about its normal axis N and y2 , y3 for the translation in the base-plane, i.e free planar motion is available in
this mode. Once more it is possible to compute the necessary condition for singular positions. The result is an equation of degree 12 which can be factorized into

(15)

+ 9 (h1 + h2 ) = 0

F1 F2 (d1 + d2 d3 ) (d1 + d3 d2 ) (d2 + d3 d1 ) F3 = 0

It it is again a Kummer surface with the difference that the

pipe (Fig.3) has a larger diameter. Again if the limb parameters are chosen inside resp. outside the pipe, both
solutions of this system are real resp. complex. All solutions are real for e.g. the parameters
h1 = 12, h2 = 7, d1 =

1
0

where F1 and F2 are exactly the singularity-conditions

from the previous modes and F3 is the factor d1 + d2 + d3
which does never vanish. The union of the corresponding
varieties of these five factors separates the joint parameter space into several zones where zero, two, or four real
solutions are obtained. All solutions are real for e.g. the
parameters

121
965
62
, d2 =
, d3 =
3
21
3

If the condition (15) fulfilled, both solutions coincide and

the corresponding positions are described by

1 0 0
0
0 1 0
0

p2 0 1 0
p3 0 0 1

h1 = 12, h2 = 7, d1 =

493
66
125
, d2 =
, d3 =
7
21
7

Due to the fact that the condition is a product of essentially

five factors there are more than one representatives for a
singular position. First of all both matrices from the previous modes, describing their singular positions. Furthermore
from the small factors one obtains

1 0
0
0
0 1
0
0

p2 0 v33 v32
p3 0 v32 v33

i.e. the singular positions of the twisted translational mode

are once more those, where base and platform lie in the
same plane.
System K3 : Planar Mode
{y0 = 0, y1 = 0, x2 = 0, x3 = 0}
6

13th World Congress in Mechanism and Machine Science, Guanajuato, Mexico, 19-25 June, 2011

with
2
2
v32
+ v33
= 1,

IMD-123

also self-intersections which again separate the parameter

space in different zones. It was possible to find parameters
where all six solutions are real, which are e.g.

p22 + p23 = h21 + h22 2 h1 h2 v33

i.e. amongst others a singular position is obtained when the

origin of the platform frame lies on a circle in the base plane
whose radius is determined by the rotation of the frame
about its x-axis.

h1 = 12, h2 = 7, d1 =

493
66
125
, d2 =
, d3 =
7
21
7

Up to now it was not possible to deduce all singular positions from the condition above due to its complexity.

Fig. 5. The singularity surface of the upside-down planar mode.

System K5 : General Mode

Fig. 4. The singularity surface of the planar mode. It contains the surfaces
from Fig. 2 and Fig. 3
.

This mode is the most difficult one, because of the complexity of the equations in ideal J5 . What definitely can
be said is that is has in general for given limb length 64
solutions and that the system has 24 real solutions if the
parameters are

System K4 : Upside-down planar mode

{x0 = 0, x1 = 0, y2 = 0, y3 = 0}

1
0

0
1

2 (x2 y0 x3 y1 ) 0
2 (x2 y1 + x3 y0 ) 0

0
0
x22 x23
2 x2 x3

2 x2 x3
x22 + x23

h1 = 12, h2 = 7, d1 = d2 = d3 =

181
13

Although the equations of J5 can be solved linearly for y1 ,

y2 ,y3 , and y4 , due to the complexity of the equations there is
no possibility to find a neat description of this mode, neither
the condition for singularities nor a description of them.

Solutions of K4 correspond to positions of the platform

where it is turned upside down and coplanar to the base.
This can be achieved by starting in the planar home position
and turning the platform about its y-axis, while the limbs
are always attached. To parameterize the upside-down planar operation mode one could use x3 , y0 , y1 in connection
with x22 + x23 = 1, where x3 is responsible for the rotation
of the platform about its normal axis N and y0 , y1 for the
translation in the base-plane.
Computation of the singularity condition was rather hard in
this case. The result is a very lengthy equation of degree
24 which cannot be factorized over Q. Due to space limitations it is not displayed here. Several plots of it were made
for given h1 , h2 . The variety is again tube-like but it has

All together there are five essential operation modes of

the SNU 3-UPU manipulator. It is quite obvious that this
mechanism is more complex than the SNU 3-UPU whose
essentially seven operation modes were quite simple.
VI. Changing operation modes
As already mentioned there exist poses where the mechanism can change from one mode into another mode. One
of them is e.g. the planar home position where the mechanism can bifurcate into the planar mode or in the translational mode. In the following an overview is presented
7

13th World Congress in Mechanism and Machine Science, Guanajuato, Mexico, 19-25 June, 2011

of those the poses where a mode-change is possible. This

is done by inspecting each pair of ideals {Ki , Kj } with
respect to common real solutions. For each pair the dimension of the intersection of the corresponding varieties
is computed and the following results are obtained.

K1

K2

K3

K4

K5

K1

K2

K3

K4

K5

IMD-123

i.e. platform and base are in the same orientation and the
origin of the platforms frame has to lie on a cylinder with
radius 2 (h1 h2 ). This is a result which verifies statements
about singularity loci from [10].

TABLE II. All values of dim(Ki Kj )

Fig. 6. The singularity surface for changes from translational to general
mode.

The numbers in Table VI correspond to the dimension

of the intersection variety. As it can easily be seen there
are four possible combinations of operation modes which
have no pose in common, to change from one to the other
a detour has to be made, which is always possible via the
general mode, corresponding to K5 . It has to be noted that
mode changing poses of the manipulator are singular, because each of them lies in the intersection of two varieties.
In the following all possible changes are discussed with respect to necessary conditions and a description of related
poses.

Twisted translational mode planar mode

The condition for this change is exactly the singularity condition from the twisted translational mode, i.e. all singularities of this mode coincide with the intersection singularities with the planar mode. Hence the possible mode change
poses have already been mentioned and the corresponding
singularity surface can be seen in Fig. 3.
Twisted translational mode general mode

Translational mode planar mode

The condition for this mode change is as follows:

The condition for this change is exactly the singularity condition from the translational mode, i.e. all singularities of
this mode coincide with the intersection singularities with
the planar mode. Hence the possible mode change poses
have already been given in Section V and the corresponding singularity surface can be seen in Fig. 2.

d41 + d42 + d43 d21 d22 d21 d23 d22 d23 36 (h1 + h2 )4 = 0

For this change a new condition appears, that is

and for the corresponding poses of the platform the following description can be deduced

1 0 0
0
p 1 1 0
0

p2 0 1 0
p3 0 0 1

d41 + d42 + d43 d21 d22 d21 d23 d22 d23 36 (h1 h2 )4 = 0

with

and for the corresponding positions of the platform the following description can be deduced

1 0 0 0
p 1 1 0 0

p 2 0 1 0
p3 0 0 1

i.e. once more platform nd base have the same orientation

and the origin of the platforms frame has to lie on a cylinder, but this time with radius 2 (h1 + h2 ). The same result
can again be found in [10], p. 598.
Planar mode general mode

Translational mode general mode

p22 + p23 = 4 (h1 + h2 )2

The condition which has to be fulfilled for this case reads

with
7 (d41 + d42 + d43 ) 11 (d21 d22 d21 d23 d22 d23 ) = 0

p22 + p23 = 4 (h1 h2 )2

8

13th World Congress in Mechanism and Machine Science, Guanajuato, Mexico, 19-25 June, 2011

IMD-123

Upside-down planar mode general mode

The condition for this case can be computed but it is rather
complicated. It is an equation of degree 24 and it is not
equal to the singularity condition of the upside-down planar
mode itself. As a result of the complexity up to now also no
description of the platforms poses could be found.

Fig. 7. The singularity surface for changes from twisted translational to

general mode.

The corresponding poses of the platform are given by

1 0
0
0
0 1
0
0

p2 0 v33 v32
p3 0 v32 v33

Fig. 9. The singularity surface for changes from upside-down planar to

general mode.

All other combinations which are {K1 , K2 }, {K1 , K4 },

{K2 , K4 }, and {K3 , K4 } do not allow any operation mode
swap.

with
2
2
v32
+ v33
= 1,

p22 + p23 = 4 (h21 + h22 2 h1 h2 v33 )

VII. Conclusions
Like in [6] methods from algebraic geometry have
proven to be very useful to analyze a mechanism like the
TSAI 3-UPU manipulator. In particular primary decompositions can be used to inspect a manipulator with respect to
possible different operation modes.
It could be shown that the direct kinematics of the TSAI 3UPU has up to 78 solutions. The maximum number of real
solutions is 28 so far. Furthermore the mechanism seems to
be less special than the SNU 3-UPU. Maybe a precisely
manufactured model of it would not be that unstable, at
least in a region around the home position. Nevertheless
several regions could be found where one has to expect singular positions, not to mention possible singularities of the
general mode, which could not be found until now.

It is noticeable that these positions are very similar to the

singular poses of the planar mode itself.

References
[1]

[2]
Fig. 8. The singularity surface for changes from planar to general mode.

Tsai L.-W., Kinematics of a Three-DOF Platform with Three Extensible Limbs, Recent Advances in Robot Kinematics, J. Lenarcic and
V. Parenti-Castelli (eds.), Kluwer Academic Publishers, 401410,
1996.
Di Gregorio R., Parenti-Castelli V., A Translational 3-DOF Parallel
Manipulator, Advances in Robot Kinematics: Analysis and Control,
J. Lenarcic and V. Parenti-Castelli (eds.), Kluwer Academic Publishers, 4958, 1998.

13th World Congress in Mechanism and Machine Science, Guanajuato, Mexico, 19-25 June, 2011

[3]

Di Gregorio R., Parenti-Castelli V., Mobility Analysis of the 3UPU Parallel Mechanism Assembled for a Pure Translational Motion, IEEE-ASME International Conference on Advanced Intelligent
Mechatronics, Atlanta, 520525, 1999.
[4] Parenti-Castelli V., Di Gregorio R., Bubani F., Workspace and Optimal Design of a Pure Translation Parallel Manipulator, Meccanica,
Vol. 35, No. 3, 203214, 2000.
[5] Tsai L.-W., Joshi S., Kinematics and Optimization of a Spatial 3UPU Parallel Manipulator, ASME Journal of Mechanical Design,
Vol. 122, 439446, 2000.
[6] Walter D. R., Husty M. L., Pfurner M., A Complete Kinematic Analysis of the SNU 3-UPU Parallel Robot, Interactions of Classical
and Numerical Algebraic Geometry, Vol. Contemporary Mathematics 496, 331346, 2009.
[7] Husty M. L., Karger A., Steinhilper W., Kinematik und Robotik,
Springer-Verlag, Berlin, Heidelberg, New York, 1997.
[8] Pfurner M., Analysis of spatial serial manipulators using kinematic
mapping, Doctoral Thesis, University of Innsbruck, 2006, (available
at http://repository.uibk.ac.at).
[9] Walter D. R., Husty M. L., On Implicitization of Kinematic Constraint Equations, Machine Design & Research (CCMMS 2010), Vol.
26, 218226, ISSN 1006-2343, Shanghai, 2010.
[10] Chebbi A. H., Parenti-Castelli V., Geometric and Manufacturing Issues of the 3-UPU Pure Translational Manipulator, New
Trends in Mechanism Science, D. L. Pisla et al. (eds.), Springer Science+Business Media B.V., 595603, 2010.
[11] http://en.wikipedia.org/wiki/Kummer surface

10

IMD-123