Positioning the End Eector of an

8-DOF Redundant Manipulator

Xavier A. Waller
EECE/ME 371
Robotics Final Project
Vanderbilt University
December 4, 2010
Prepared in L
A
T
E
X
1 Introduction
The purpose of my nal project was to demonstrate the validity of utilizing the in-
verse kinematics of a manipulator to accurately position its end eector. An 8-degree-
of-freedom (DOF) redundant manipulator was created using ROBOSIM as a tool for
visualization and using MATLAB and Mathematica as computational tools to facilitate
the analysis.
2 Developing the Model
ROBOSIM is a robot simulation package developed by faculty at Vanderbilt University.
It allows one to build a 3D model of a robotic manipulator and simulate its dynamic
behavior by inputing the inverse kinematics for the manipulator in question. Such a
tool is invaluable when it comes to designing and testing novel robotic models before
fabrication. Using the software, an 8-DOF manipulator was built in ROBOSIM (Fig.
1) and is comprised of 8 revolute joints (4 truncated-conical sections and 4 spherical
sections). Each spherical section allows for rotation of its respective truncated-cone about
the spheres x-axis while the cones themselves are free to rotate about their respective
z-axes
1
.
Figure 1: 8-DOF Manipulator model in ROBOSIM at the home position.
1
The components are color-coded for clarity. The blue spherical section corresponds to the blue
conical section, and so on for each of the three remaining pairs of spheres and cones.
1
3 Background
In order to position an end eector generally in 3-space, a minimum of six joints are
needed due to the fact that 6 degrees of freedom are needed for such positioning, namely
x, y, z,
x
,
y
,
z
, where x, y, and z are linear displacements and
i
is the rotation about
the i
th
axis [1]. There are two areas of concern with kinematic manipulations of robotic
manipulators: forward and inverse kinematics. The forward kinematic problem is: given
the joint parameters of a manipulator, what are the position and orientation of the end
eector
2
? The inverse kinematic problem, conversely, is: given a desired position and
orientation of the end eector, what joint angles are needed to achieve this goal? As
one may suspect, the computation of the inverse kinematics are generally much more
involved than the computation of the corresponding forward kinematics. This problem
becomes much more complicated when one introduces redundancies in the manipulator
to be analyzed. This is due to the fact that extraneous parameters cause the number of
potential solutions to increase.
3.1 General Solution Strategy
The general solution to an inverse kinematic problem always begins with the calculation
of the forward kinematics of the manipulator-end eector assembly via the Denavit-
Hartenberg (D-H) Parameters in order to develop a link transformation matrix, T, which
transforms the frame of the base (usually the starting frame) to some new frame (usually
the end eector frame). Fig. 2(b) shows a sample table to demonstrate the calculation
of the Denavit-Hartenberg parameters for a simple 3 DOF planar manipulator. The i
refers to the i
th
link,
i1
and a
i1
refer to the previous links joint angle and displace-
ment, respectively, and d
i
and
i
refer to the current links displacement and joint angle,
respectively.
(a) A three-link, planar manipulator (b) D-H Parameters
Figure 2: Notice that the values of the
i
s in (a) as well as the values for the L
i
s correspond the the
D-H chart shown in (b)[1].
The transformation matrix for the i
th
link with respect to the previous link is given
by:
2
In general, any link between the base of the robot and the end eector is acceptable, although one
is generally concerned with the explicit positioning of the end eector itself.
2
i
i1
T =
_

_
cos(
i
) sin(
i
) 0 a
i1
sin(
i
) cos(
i1
) cos(
1
) cos(
i1
) sin(
i1
) sin(
i1
)d
i
sin(
i
) sin(
i1
) cos(
i
) sin(
i1
) cos(
i1
) cos(
i1
)d
i
0 0 0 1
_

_
(1)
To determine the transformation matrix,
0
n
T, relating the base frame to the end eector,
one simply multiplies all n of the transformation matrices together such that:
0
n
T =
0
1
T
1
2
T
2
3
T ...
n2
n1
T
n1
n
T (2)
=
_

_
r
11
r
12
r
13
p
x
r
21
r
22
r
23
p
y
r
31
r
32
r
33
p
z
0 0 0 1
_

_
(3)
where n is the total number of frames and p
x
, p
y
, and p
z
are the nal x-, y-, and z- positions
of the end eector, respectively
3
. These three values are then utilized to calculate the
Jacobian, which is paramount to deriving the inverse kinematics of the manipulator.
3.2 Jacobian
The Jacobian of a transformation is a time-varying, linear transformation which maps
the velocities from one frame to another having dimensions m x n such that the number
of rows, m, are the degrees of freedom of the workspace in question and the columns, n,
are the total number of joints. As stated earlier, since navigation in 3-space involves 6
DOF, a Jacobian model of a manipulator in 3-space would have to have m =6 in order
to meet the translation and position requirements demanded.
To calculate the Jacobian, one determines the partial derivatives of the position vector
[p
x
, p
y
, p
z
]
T
obtained from Eq. 3 and the partial derivatives of the angles of rotation about
the x-, y-, and z-axes of the end eector frame, [, , ]
T
, with respect to each of the
joint parameters {
1
,
2
, . . . ,
8
}, where
i
is either a linear parameter, d
i
, or an angular
parameter,
i
. This process is shown in Eq. 4.
J() =
_

_
p
x
q
1
p
x
q
2
p
x
q
3
p
x
q
4
p
x
q
5
p
x
q
6
p
x
q
7
p
x
q
8
p
y
q
1
p
y
q
2
p
y
q
3
p
y
q
4
p
y
q
5
p
y
q
6
p
y
q
7
p
y
q
8
p
z
q
1
p
z
q
2
p
z
q
3
p
z
q
4
p
z
q
5
p
z
q
6
p
z
q
7
p
z
q
8

q
1

q
2

q
3

q
4

q
5

q
6

q
7

q
8

q
1

q
2

q
3

q
4

q
5

q
6

q
7

q
8

q
1

q
2

q
3

q
4

q
5

q
6

q
7

q
8
_

_
(4)
3
Due to the nature of the calculation of the nal total transformation from the base frame to the end
eector, p
x
, p
y
, and p
z
are necessarily functions of the D-H Parameters.
3
The vector

is used to denote {
1
,
2
, . . . ,
8
} and the Jacobian can be related to the
linear and angular velocity vectors of the manipulator by:
_

_
v
x
v
y
v
z

z
_

_
= J(
1
,
2
, . . . ,
8
)
_

8
_

_
(5)
where the left-hand side of Eq. 5 is the concatenation of the vectors representing linear
and angular velocities. This relationship can be expressed more succinctly as:

total
= J()

(6)
where
total
is equivalent to the left-hand side of Eq. 5. and

is the time rate of change

of the joint parameters. Because this project was only concerned with positioning the
manipulator,
total
can be replaced with = [v
x
, v
y
, v
z
]
T
. As a consequence, only the
rst three rows of J() are necessary for the calculation of position, which simplies the
analysis. This new matrix will be referred to as J
s
() The new relationship becomes:
_

_
v
x
v
y
v
z
_

_
= J
s
(
1
,
2
, . . . ,
8
)
_

8
_

_
(7)
3.3 Inverse Kinematics
Once the Jacobian was determined, the inverse kinematics of the system could be used
to provide a user with the required joint values to bring the manipulator to a desired
position. By algebraically manipulating Eq. 6, it was possible to solve for

as follows:

= J
1
() (8)
However, because the Jacobian for a redundant manipulator is necessarily non-square,
J
s
() is not invertible. Several methods have been developed for the determination of
a pseudo-inverse of the Jacobian, J

s
. The method utilized for this project was Singu-
lar Value Decomposition (SVD), which is capable of producing unique solutions for the
4
pseudo-inverse
4
. Using SVD to generate J

s
, the relationship between the joint parameter
vector,

, and the velocity vector, , then becomes:

= J

s
(9)
Next, the desired linear velocity, , was determined. This was performed by taking the
dierence between the current and desired position and scaling them by a constant, ,
which was determined empirically. This relationship is given as:
=
_

_
v
x
v
y
v
z
_

_
=
_
_
_
_

_
x
d
y
d
z
d
_

_
x
c
y
c
z
c
_

_
_
_
_
(10)
where the subscripts d and c represent the desired and current Cartesian positions
of the end eector, respectively. In general, the manipulator could be asked to move
from one end of its workspace to the opposite end. The ramications of a manipulator
instantaneously moving between two distant points are jerky motion and the need for an
innite amount of power with the former being highly undesirable and the latter being
physically impossible to achieve. To resolve this issue, the trajectory between the desired
position and current position were subdivided into discrete steps. The smaller these steps
are, the smoother the motion becomes
5
. Once the velocity was determined, it was plugged
into Eq. 9 to solve for the desired rate of change of the joint parameters,

. By utilizing
the following relationship, it then became possible to solve for the nal joint parameters
such that:

k+1
=

(11)
where the subscripts k and k + 1 correspond to the current set of intermediate joint
parameter values and the next set of intermediate joint parameter values, respectively.
Recall that the path from the current position to the desired position was subdivided.
Thus, the index k ranged from 1 to the number of subdivisions which was 10,000 for this
project. Furthermore, the variable represents a scaling constant
6
and was determined
empirically .
4 Results and Discussion
The values of and were determined to be 0.1 and 1.0, respectively. These values were
determined based upon how close the nal position of the manipulator was to the desired
(input) position. In this case, the actual positions either deviated by less than 2% from
the desired positions or were vastly dierent from the desired positions as shown in Tab.
I.
As the data illustrates, the actual output position of the system (x
a
, y
a
, z
a
) closely
matched the input positions (x
d
, y
d
, z
d
) in most cases. However, it is evident that there
4
MATLAB has the capability to perform the SVD.
5
This was achieved by utilizing the trajpar command in MATLAB starting at the current position
and ending at the desired position in 10,000 steps. A trajpar was used for each of the x-, y-, and z-
directions.
6
The scaling constants and have units of [seconds] and [1/second], respectively.
5
Table I. Measured Values
Run Desired Position Output Position
No. x
d
y
d
z
d
x
a
y
a
z
a
1 40 13 0 35.30 37.20 1.86
2 20 20 20 19.99 19.20 20.19
3 -2 36 20 -1.98 35,94 20.02
4 -30 22 42 -29.95 22.11 41.99
5 -30 22 56 -24.00 31.30 44.57
6 -30 22 55 5.21 11.69 14.83
7 -50 22 5 -48.90 41.38 4.95
8 56 56 56 25.00 28.00 34.00
9 0 0 0 0.02 -0.01 0.04
10 0 0 56 0.02 0.09 55.98
11 0 56 56 16.15 -29.80 6.88
12 0 6 32 0.02 5.96 32.08
13 0 6 32 0.02 5.96 32.08
are some trials when the output position deviated signicantly from the desired position.
An extreme example of such deviation is Run No. 8 where the input position was (56,
56, 56) and the output position was (25.00, 28.00, 34.00). In this case, none of the values
were in an acceptable range. A less extreme example is Run No. 7 where the input
position was (-50, 22, 5) and the output was (-48.90, 31.38, and 4.95). In this case, the
x- and z- coordinates were in an acceptable range, however; the y-coordinate was o.
The common factor between both of these cases is the fact that the manipulator has
approached a workspace singularity at its outer boundary. Since the workspace of this
particular manipulator is the volume of a sphere with a radius of 56 units and due to the
fact that the approaching of the outermost radius causes the loss of one or more degrees
of freedom of the manipulator, which causes certain positions to be unattainable[1], the
closer one approaches the point (56, 56, 56), the more unpredictable the manipulator
behaves. Thus, the eective operating range of this manipulator should be well within
the maximum radius that can be reached by the manipulator. Run No. 1, however,
shows a deviation which is not expected according to the statement about workspace
singularities. It is likely that a workspace singularity somewhere in the interior occurred,
which presents itself when one or more axes of a manipulator align (or approach such
alignment). Finally, , it was stated earlier without proof that the SVD produces a unique
solution for the pseudo-inverse of the Jacobian, J

. Runs No. 12 and 13 are a validation

of the fact that the SVD algorithm consistently produces the same J

for a given set of

parameters.
5 Conclusion
This exercise has demonstrated the validity of utilizing the inverse kinematics of a manip-
ulator to control and position its end eector. Care must be avoided when approaching
singular congurations as these will cause the robot to behave in unexpected ways, pos-
sibly causing equipment damage or personal injury. While the SVD method produces
unique solutions to the Jacobian, the Jacobian method of positioning an end eector is
6
maximally eective when exact positioning is not required. Thus, in scenarios such as
welding, where high precision is necessary, other methodologies should be used to posi-
tion the end eector. Such methodologies are obviously possible due to the fact that the
human body, for instance, is capable of positioning its many redundant end eectors with
a remarkable degree of accuracy despite the fact that the control schemes used for such
positioning are high in noise and uncertainty.
References
[1] John J. Craig. Introduction to Robotics: Mechanics and Control, 2nd Ed. Addison-
Wesley, 1989.
7