Grasping in robotics

Chapter 2
Stiffness Analysis for Grasping Tasks

Giuseppe Carbone

Abstract This section addresses key aspects that are related with stiffness prop-
erties when dealing with grasping tasks. Main theoretical aspects are formulated
for computing the Cartesian stiffness matrix via a proper stiffness analysis and
modeling. Basic concepts are given for the comparison of stiffness performance for
different robotic architectures and end-effectors by referring both to local and
global properties. Cases of study are described for clarifying the effectiveness and
engineering feasibility of the proposed formulation for stiffness analysis. Then, an
experimental set-up and tests are proposed for the experimental validation of
stiffness performance.

2.1 Introduction

A load applied on a body produces deformations of the body itself that are known
as compliant displacements. Stiffness can be deﬁned as the property of a
mechanical system in sustaining loads without too large compliant displacements.
Stiffness can be also deﬁned quantitatively as the amount of force that can be
applied in one direction per unit of compliant displacement of the body in the same
direction, or the ratio of a steady force acting on a body to the resulting compliant
displacement [1–8].
Stiffness plays a key role both in the design and control of any robotic system
for grasping tasks. Thus, stiffness is widely investigated for any robotic system

G. Carbone (&)
LARM: Laboratory of Robotics and Mechatronics, University of Cassino and South Latium,
Via G. Di Biasio, 43 03043 Cassino, FR, Italy
e-mail: carbone@unicas.it

G. Carbone (ed.), Grasping in Robotics, Mechanisms and Machine Science 10, 17
DOI: 10.1007/978-1-4471-4664-3_2, Ó Springer-Verlag London 2013

18 G. Carbone

(a) A yA (b) yA

End-Effector zA A XA
xA
zA

End-Effector

y0

y0

0 0
x0 x0
z0 z0

Fig. 2.1 Schemes of multibody robotic systems: a A 2R serial manipulator; b A parallel
manipulator with three RPR legs

spanning from conventional serial robots to non-conventional parallel manipula-
tors, such as those that are schematized in Fig. 2.1. Few examples in a wide
literature can be found in [6–28].
Given the peculiarities of grasping tasks, one should carefully study the stiff-
ness performance of the robotic system but also of the robot extremity, which is
generally denominated as end-effector. In fact, usually only the end-effector is
directly interacting with the environment and the objects that have to be manip-
ulated. There are many different types of end-effectors with different sizes, shapes,
operation, and actuations principles, as stated for example in [2, 29–31]. Stiffness
can be considered of particular significance for all those mechanically achieving
the grasp. They may range from dedicated mechanical grippers having two fingers
(widely used in industrial applications) up to versatile multi-fingered robotic hands
(widely investigated for mimicking the high multi-purpose operation of human
hands). Some examples from an extensive literature are reported in [32–40].
Several grasping devices have been also designed and built at LARM in Cas-
sino, [41–49]. For example, Fig. 2.2 shows two prototypes at LARM: a two-finger
gripper, Fig. 2.2a, and a robotic hand with three fingers, Fig. 2.2b, respectively.

Fig. 2.2 Example of grasping devices that have been designed and built at LARM in Cassino:
a A two-finger gripper; b The LARM Hand IV with three fingers

2 Stiffness Analysis for Grasping Tasks 19

Many researchers have investigated stiffness with different approaches and
focuses. Most of the published works on stiffness can be classified into three main
categories. The first category deals with stiffness analysis and determination of
overall stiffness. Given the stiffness of main components motors, joints, links, the
overall stiffness has to be determined as reported, for example, in [7–18]. Once a
proper stiffness model has been defined it can be used for controlling the grasp
stability, or in compliance control algorithms as proposed for example in [19–21].
Moreover, a proper stiffness model can be used also for design purposes, for
example, in order to find an optimum compromise between weight of links and
stiffness performance as proposed in [22–24]. A second category studies the inverse
decomposition of a stiffness matrix into constituent stiffness parameters that are
often assumed to be simple linear springs, as proposed for example in [25]. In a third
research line, mathematical properties of the stiffness matrix are investigated,
mainly with the aim of finding intrinsic properties that are independent from the
coordinate frame in which the stiffness matrix is expressed, [26–28].
Although stiffness is widely investigated there are still open problems.
For example, experimental determinations and evaluations of stiffness perfor-
mance are prescribed in standard codes for robotic manipulators, [50–52] that
should be extended also to non-conventional robotic systems, grippers, and hands.
Still an open issue can be considered also the formulation of computationally
efficient algorithms that can give direct engineering insight of the design parameter
influence and can be translated into experimental tests for experimental determi-
nations. Moreover, still few preliminary comparisons of numerical results with
experimental experience have been proposed, [53, 54].

2.2 Stiffness Modelling and Analysis

Usually, stiffness analysis of a robotic system is aiming to determine the stiffness
performance through the computation of a 6 9 6 Cartesian stiffness matrix K.
This stiffness matrix K expresses the relationship between the compliant dis-
placements DS occurring to a frame fixed at the end of the kinematic chain when a
static wrench W acts upon it and W itself. Considering Cartesian reference frames,
6 9 1 vectors can be defined for the compliant displacements DS and the external
wrench W as

DS ¼ ðDx; Dy; Dz; Da; Dc; DdÞt ;
ð2:1Þ
W ¼ ðFx; Fy; Fz; Tx; Ty; TzÞt

where Dx, Dy, Dz, Da, dc, and Dd are the linear and angular compliant dis-
placements on the robotic system extremity; FX, FY, and FZ are the force com-
ponents acting on the robotic system extremity along X, Y, and Z directions,
respectively; TX, TY , and TZ are the torque components acting upon the same
point on the robotic system extremity about X, Y, and Z directions, respectively.

20 G. Carbone

Compliant displacements have usually negative effects on a robotic device for
grasping tasks, since compliance detrimentally affect accuracy, repeatability, and
payload capability. Additionally, in dynamic conditions, the presence of large
compliant displacements can affect fatigue strength, can produce vibrations and
energy losses. However, in some cases, compliant displacements can even have a
positive effect if they are properly controlled, [55,56]. In fact, they can enable the
correction of misalignment errors encountered; for example, when parts are mated
during assembly operations [5] or in peg into hole tasks [21], or in deburring tasks,
[57], or in the operation of a prosthesis [58]. It is to note also that a stiffer behavior
is often obtained at cost of an higher own weigh of a robotic system that can rise
manufacturing costs and detrimentally affect dynamic performance and power
consumption. Thus, a proper stiffness modeling and analysis is of key significance
to identify optimal trade-off solutions both at design and control stage.
Provided that the assumptions of small compliant displacements hold, one can
write
KðqÞ : <r ! <r ; W ¼ K DS ð2:2Þ

where K is the so-called 6 9 6 Cartesian or spatial stiffness matrix.
It is worth noting that according to the definition in Eq. (2.2), the stiffness
matrix K is in general posture dependant. Moreover, the stiffness matrix K is
generally nonsymmetric and its entries depend on choice of reference frame, since
it is not reference frame invariant, as demonstrated for example in [2, 7, 16, 26–
28].
The computation of stiffness matrix K can be achieved with different approa-
ches such as finite element methods (FEM) or methods based on models with
lumped parameters (MLP). FEM methods can be used for a stiffness analysis of
multibody robotic systems, although with very difficult numerical implementation.
In fact, even if FEM methods could be more accurate than MLP methods they are
time consuming and they require a complete recalculation at each configuration/
loading condition under analysis. Therefore, the stiffness analysis of robotic sys-
tems is usually carried out by means of MLP methods that are based on using
lumped stiffness parameters for taking into account the stiffness properties of links
and joints with configuration dependant relationships. Therefore, main advantages
of MLP methods can be understood in reduced computational efforts and possi-
bility to use the same stiffness model for the analysis of several different config-
urations. These aspects give the possibility to investigate the stiffness performance
through the whole workspace of a robotic system in a reasonable amount of
computational time. Moreover, MLP methods can be conveniently used for
developing parametric models within optimal design procedures.
Equation (2.2) defines K as a 6 9 6 matrix whose components are the amount
of forces or torques that can be applied per unit of compliant displacements of the
end effector for a robotic system. However, the linear expression in Eq. (2.2) is
valid only for small magnitude of the compliant displacements DS. Moreover, Eq.
(2.2) is valid only in static (or quasi static) conditions. The entries of a 6 9 6


stiffness matrix can be obtained through the composition of suitable matrices. A
ﬁrst matrix CF gives all the wrenches WL, acting on the links when a wrench
W acts on the manipulator extremity as
W ¼ CF W L ð2:3Þ
with the matrix CF representing the force transmission capability of the manipu-
lator mechanism.
A second matrix Kp gives the possibility to compute the vector Dv of all the
deformations of the links when each wrench WLi on a ith link given by WL, acts
on the legs according to
W L ¼ Kp Dv ð2:4Þ

with the matrix Kp grouping the lumped parameters of the robotic system.
A third matrix CK gives the vector DS of compliant displacements of the
manipulator extremity due to the displacements of the manipulator links, as
expressed as
Dv ¼ CK DS ð2:5Þ
Therefore, the stiffness matrix K can be computed as
K ¼ CF Kp CK ð2:6Þ

with matrix CF giving the force transmission capability of the mechanism; Kp
grouping the spring coefﬁcients of the deformable components; CK considering the
variations of kinematic variables due to the deformations and compliant dis-
placements of each compliant component.
Matrices CK and CF can be computed, for example, as a Jacobian matrix J and
its transpose, respectively, as often proposed in the literature. Thus, one can
compute the Cartesian stiffness matrix as

K ¼ Jt K p J ð2:7Þ
Nevertheless, this is only an approximate approach as pointed out, for example,
in [59]. A more accurate computation of matrices CK and CF can be obtained as
reported, for example in [2, 7, 13]. The KP matrix can be computed as a diagonal
matrix whose components are the lumped stiffness parameters of links, joints, and
motors that compose a multibody robotic system. The lumped stiffness parameters
can be estimated by means of analytical and empirical expressions or by means of
experimental tests.
The lumped stiffness parameters can be graphically represented as linear or
torsion springs. Examples of stiffness models with lumped parameters are shown
in Fig. 2.3. In particular, Fig. 2.3a shows a stiffness model with lumped parameters
for the 2R serial manipulator in Fig. 2.1a. This model considers the stiffness of the
two motors and joints by means of two lumped parameters kT1, kT2.

22 G. Carbone

yA
(a) (b) A
k2
1

yA W
zA
A xA
K kT2 xA
k1
y0 1 k2
k1 1 k3
W zA y0
0
z0 kT1 x0
K

0
x0
z0

(c) (d) (e)
Y 1 k2 W
W Q
Q
kT2
W
K

X
kT1 K kT3
K

k1
1 k5
Fin X
G kT2
Y
K

kT4 k4
1

K

Q
kT5
K

G y0
kT1
kT3
K

Fin
K

0 x0
W
1 k3 z0

(f) (g) (h)
W W
k1 k2
1

k1
k
Y Slipping line k3
1

W W kB
kA B1
er tip

1

1

B
p

keq X
k6
er ti

Fing

A1 Z
A
K 1
Fing

Object

Contact line
k5
1

k4
1

Squeezing line
k1
k

Y Y
O X O X

Z Z

Fig. 2.3 Examples of stiffness models with lumped parameters: a For the 2R serial manipulator
in Fig. 2.1a; b For the parallel manipulator in Fig. 2.1b; c For the two finger gripper in Fig. 2.2a;
d For the symmetric part of the two finger gripper in Fig. 2.3c; e For one finger of the LARM
Hand IV in Fig. 2.2b; f Two linear springs and their equivalent as a single linear spring; g For the
the grasping of a generic object with a two finger gripper; h For representing a generic stiffness
matrix with a six linear springs model


Additionally, the axial stiffness of the links is considered by means of the
lumped parameters k1, k2. These two scalar lumped parameters can be replaced
with two 6 9 6 stiffness matrices K1, K2 that can be even obtained by means of
finite element softwares for taking into account the whole stiffness behavior of the
links. Similarly, Fig. 2.3b shows a stiffness model with lumped parameters for the
parallel manipulator in Fig. 2.1b. This model considers the stiffness of the two
prismatic actuators and links by means of two lumped parameters k1, k2. These two
scalar lumped parameters can be replaced with two 6 9 6 stiffness matrices K1, K2
that can be even obtained by means of finite element softwares for taking into
account the whole stiffness behavior of the links.
Figure (2.3c) shows a stiffness model for the two-finger gripper in Fig. 2.2a.
This model takes into account the stiffness of joints and links by means of the
lumped parameters kT1, kT2, kT3, kT4, and k1, k2, k3, k4, respectively. It is to note
that the two-finger gripper in Fig. 2.2a has a symmetric design. Thus, one can
study only half of the model. Additionally, the effect of the lumped parameters kT1,
kT2 and k1, k2, can be combined in the lumped parameters kT5 and k5, respectively,
as shown in the simplified scheme of Fig. 2.3d. Similarly, Fig. 2.3e shows a
simplified stiffness model for the robotic hand in Fig. 2.2b. In this simplified
model, the stiffness properties of the motor and the driving mechanism as well as
the flexional properties of the links have been combined in the lumped parameters
kT1, kT2, kT3. It is to note that it is advisable to keep the number of chosen lumped
stiffness parameters as equal to the desired rank of the stiffness matrix (that is
equal to six in the spatial case, to three in the planar case). In fact, choosing a
different number of lumped stiffness parameters yields to nonsquare matrices that
can lead to computation problems. For this purpose, if the superposition principle
holds, one can combine the effect of more compliance sources in a single lumped
parameter as described in the example of Fig. 2.3f where the lumped stiffness
parameters k1 and k2 have been combined in the equivalent lumped parameter keq,
whose value can be obtained as

ðkeq ÞÀ1 ¼ ðk1 ÞÀ1 þ ðk2 ÞÀ1 ð2:8Þ
The grasping of an object also should require proper stiffness models that need
to take into account the contact forces, the position of contact points/areas, the
curvature of surfaces in contact. For example, Fig. 2.3f shows a simplified stiffness
model with lumped parameters for the grasping of a generic object with a two-
finger gripper. The contact areas on the two fingertips are assumed to coincide with
the contact points A and B along the contact line. The stiffness properties of the
contact are lumped in the parameters k1, k2 that are shown in Fig. 2.3f as ideal
springs having no mass and length A-A1 and B-B1 equal to zero. The grasping
model can become more complex as the one in Fig. 2.3f if one considers contact
areas, multiple contact points, variable curvature of surfaces, and effect of friction
on stiffness. In these cases, the scalar lumped parameters k1, k2 can be replaced by
6 9 6 stiffness matrices KA and KB. It is to note that any 6 9 6 stiffness matrix

24 G. Carbone

can be decomposed in a model with linear springs having scalar lumped param-
eters such as the general model with six linear springs that is shown in Fig. 2.3g.
Stiffness properties usually have different expressions according to the chosen
reference frame. Thus, each stiffness model should clearly indicate the chosen
reference frame. For example, in two-finger grasping models as in Fig. 2.3f a
Cartesian reference frame can be chosen with axes coinciding with the contact
line, squeezing line, and slipping line.
The Cartesian stiffness matrix K is posture dependent. Thus, one should define
configuration(s) of a multibody robotic system where the stiffness matrix can be
computed. The configuration(s) should be carefully chosen in order to have sig-
nificant information on the stiffness performance of the system in its whole
workspace. Then, the kinematic model can be used for computing the vector h that
express input angles and strokes in the Joint Space for any posture. It is worth
noting that the accuracy in the estimation of model data such as geometrical
dimensions and values of lumped stiffness parameters can significantly affect the
accuracy of the computed stiffness matrix. Thus, experimental tests should be
carried out in order to validate model data and overall stiffness model.
Once the stiffness matrix has been computed, it is also necessary to give syn-
thetic evaluation of the stiffness performance both for analysis and design pur-
poses. Thus, an index of merit can be formulated by using properties of the
stiffness matrix, so that it represents numerically the stiffness performance of a
new multibody robotic system.
The current standard codes for stiffness evaluation of manipulators are given as
short parts of the norms ANSI/RIA15.01.11-1990 [51] and ISO9283-1995 [52],
which refer explicitly to serial chain industrial robots only. In particular, Sect. 8.6
in [51] and Sect. 10 in [52] are devoted to Static compliance with a very similar
approach but only referring to a performance evaluation through measures of
position compliant displacements. Then, a recommendation states to express the
results in term of millimeters per Newton for displacements that are referred to the
directions of a base coordinate system. Thus, the standard codes do not yet con-
sider the stiffness matrix as a performance index for the elasto-static response of
multibody robotic manipulators, but they still refer to a practical evaluation with a
direct natural interpretation that is related to the compliance response of the
stiffness of a manipulator structure. Of course, it is evident that compliant dis-
placements can be considered as a measure of the manipulator stiffness since the
fundamental relationship in Eq. (2.1). Nevertheless, the compliance response is
system posture and wrench direction dependant since one can find a 6 9 1 vector
of compliant displacements at any posture for any wrench. Therefore, one should
define a single local index of stiffness performance and then a global index
expressing the stiffness performance in the overall workspace of a multibody
robotic system.
A local stiffness index can be directly related with the Cartesian stiffness matrix
by means of different mathematical operators that can be applied to a matrix, as
proposed for example in [54].


Compliant displacements can also provide an insight on local stiffness perfor-
mance due to their simple physical interpretation, as indeed suggested by ISO and
ANSI codes [51, 52]. In fact, one can compute the compliant displacements for a
given configuration by multiplying the computed stiffness matrix for a given
external wrench WGiven. Reasonable choices for WGivencan be a unit vector or a
vector equal to the expected payload for a multibody robotic system as proposed
for example in [7]. The first choice gives a measure of the compliant displacements
per unit of external wrench. The second choice provides a measure of the maxi-
mum compliant displacements for the system in specific applications. Neverthe-
less, compliant displacements have usually six components. Thus, they cannot be
treated as a single merit index.
Eigenvalues and eigenvectors of a stiffness matrix are also very useful for their
physical interpretation with respect to local stiffness performance. In fact, the
eigenvectors are related with the maximum and minimum eigenvalue and they
provide the directions of maximum and minimum stiffness performance, respec-
tively. Moreover, a smaller difference among the eigenvalues stands for a smaller
anisotropic stiffness behavior at a given posture. Nevertheless, eigenvalues and
eigenvectors cannot be treated as a single merit index. But their values can be used
for drawing graphical local representations of the stiffness performance such as
compliance/stiffness ellipses and ellipsoids, as reported for example in [8]. These
graphical local representations also provide a graphical tool for the comparison of
stiffness performance along and about different directions. The graphical repre-
sentations can be very useful when specific design requirements arise. In partic-
ular, they are useful if there is a need of the best stiffness performance only in a
given direction or if equal stiffness is preferred in all directions.
Other graphical tools for a comparison of stiffness performance can be obtained
through the definition of the so-called center of stiffness or the center of com-
pliance and by means of stiffness or compliant axes that can be used for defining
directions and orientations in which a robotic system acts as a simple spring, as
mentioned for example in [22].
A local index of stiffness performance is neither suitable for an accurate design
analysis nor useful for a comparison of different designs. In fact, even if a robotic
system has suitable stiffness for a given system posture it can have inadequate
stiffness at other postures. Therefore, one should look at stiffness performance at
all points of workspace or define a single global stiffness index over the whole
workspace yet.
A global index of stiffness performance for a robotic system can be defined with
graphical methods that are based on plotting curves connecting postures having the
same value of the local stiffness index (iso-stiffness curves or surfaces), as pro-
posed for example in [6]. Nevertheless, the number of iso-stiffness curves or
surfaces that one can plot is graphically limited. Moreover, few curves or surfaces
usually do not provide sufficient insight of the overall stiffness behavior of a
robotic system. These aspects significantly reduce the effectiveness of iso-stiffness
curves or surfaces.

26 G. Carbone

Global stiffness indices can be defined also in a mathematical form by using
minimum, maximum, average, or statistic evaluations of a local stiffness index.
For example, one can compute a global index in the form
R ÈpffiffiffiffiffiÉ
max kÃ dV
i
i¼1;...6
GIMN ¼ ð2:9Þ
L3
ÈpffiffiffiffiffiÉ
where kÃ is the set of nonnegative eigenvalues of KKT V is the workspace
i
volume; L is a characteristic length that is used in order to obtain information that
is independent from the workspace volume. Alternatively to L3, the denominator
can be expressed as the volume V of workspace. Moreover, the dimensional
inconsistency can be solved by using a proper dimensionless value of the merit
index (that is indicated with a superscript *) that can be obtained by dividing the
length entries by a characteristic length L. This global index can be useful when a
design goal is to maximize the stiffness performance along or about one or more
specific direction(s). A similar global stiffness index can be defined by referring to
the minimum eigenvalue as
R ÈpffiffiffiffiffiÉ
min kÃ dV
i
i¼1;...6
GImN ¼ ð2:10Þ
L3
This global stiffness index can be useful to detect and avoid design with weak
stiffness performance along or about a specific direction. A global index can be
defined also as the difference between GIMN and GImN.
It is to note that the integration operator in Eqs. (2.9) and (2.10) is usually
ÈpffiffiffiffiffiÉ
numerically calculated, since the analytical expression of kÃ dV is usually not
i
available. Thus, for comparison purposes it is advisable to have the same number
of calculation configurations.

2.3 Numerical Computation of Stiffness Performance

The models and formulations in the previous section can be used to get a
numerical insight of the stiffness performance for a robotic system. In particular, a
numerical algorithm can be composed of a first part in which all the model data are
provided such as the numerical values of the geometrical dimensions, masses, and
lumped stiffness parameters. A second part defines the kinematic model, the force
transmission model, and the lumped parameter model through the matrices CF, CK,
and Kp, respectively. Then, a third part can compute an expression of the stiffness
matrix K by means of Eqs. (2.6) or (2.7), as shown in the flowchart of Fig. 2.4.
It is worth noting that the matrices CF, and CK as well as the Jacobian matrix
and its transpose are configuration dependant. Thus, they may have nonlinear
components in a complex formulation. In general, this makes very difficult to


Start

SetModelData
- Value of link lenghts
Set Force Transmission Model - Value and distribution of masses Set Kinematic Model
- Value of lumped parameters
- Operation ranges

Set the first configuration
(entries of vector θ)

Compute the stiffness matrix K
[Eq.(2.6) or (2.7)]

Compute local stiffness index Set a new feasible
- compliant displacements
- detertminant of K configuration (vector θ)
- eigen values of K
- other local stiffness index

All configurations No
investigated ?
Yes
Plot of local stiffness indices
as function of configuration

Compute global stiffness index
- Eq.(2.9) or (2.10)
-other global stiffness index

End

Fig. 2.4 A flowchart for the proposed numerical computation of stiffness performance

identify a close-form formulation of the stiffness matrix K. Thus, often the
matrices CF and CK or the Jacobian matrix are numerically computed at a given
configuration, and then combined into Eqs. (2.6) or (2.7) to calculate the stiffness
matrix K. However, one should carefully define the configuration(s) where the
stiffness matrix will be computed. The configuration(s) should be carefully chosen
in order to have significant information on the stiffness performance of the system
in its whole workspace. Careful attention should be addressed also to avoid
numerical singularities for the stiffness matrix K that might occur also due to
nonlinear terms. However, a linearization might be possible under the assumption
of small compliant displacements.
Once the stiffness matrix has been computed, it can compute local stiffness
indices at a given configuration. Different local stiffness indices can be chosen as
also mentioned in the previous section. Values of local stiffness indices allow to
compare the stiffness performance of a robotic system at different configurations.

28 G. Carbone

An iterative loop can be defined to investigate a significant (but finite) number of
configurations within the workspace of a robotic system. In some cases, a robotic
system can have few trajectories that are mostly used during its operation. In these
cases, the kinematic model can be used together with a proper path planning
strategy for properly computing the time evolution of all the entries in the vector h
that express input angles and strokes in the joint space as function of time for a
given trajectory. Thus, the vector h(t) can be used for computing the stiffness
matrix as function of time for a given end-effector trajectory.
After completing the calculation of the local stiffness indices one can plot their
values as function of the configuration. Additionally, one can plot other graphical
representations of stiffness properties such as the iso-stiffness curves. Finally, one
can compute the global stiffness parameter, for example, by using Eqs. (2.9) or
(2.10). These global values are useful for quantitative comparisons of different
robotic systems especially at design stage.
It is to note that a proper stiffness analysis of a robotic system for grasping tasks
should start from the identification of the main sources of compliance. In fact,
there is always a trade-off between accuracy of the model and computational costs.
For this reason, very often the stiffness analysis is limited to the robot architecture
while one should carefully consider also the end effector.

2.4 Cases of Study for Stiffness Modelling and Analysis

2.4.1 A 6R Serial Manipulator

A 6 DOFs PUMA-like manipulator has been considered as a case of study for the
above-mentioned formulation as specifically applied to a serial type robotic sys-
tem. Main design parameters for a PUMA-like manipulator are shown in Fig. 2.5a.

(a) (b)

kT2
kT3 kT4
Y X

kT1 kT5
kT6
Y0 Z

Z0 X0

Fig. 2.5 Models for a PUMA-like manipulator: a Main dimensional parameters; b Lumped
stiffness parameters


Table 2.1 Main design parameters and workspace ranges for a PUMA 562
a2 a3 d3 d4 x y z / w h
(mm) (mm) (mm) (mm) (mm) (mm) (mm) (°) (°) (°)
431.8 20.3 125.4 431.8 529.2 472.4 625.0 180 180 180

Values of the design parameters a2, a3, d3, and d4 are reported in Table 2.1. These
values have been defined by referring to a PUMA 562 design and the mobility
ranges for the joint angles have been assumed equal to 180° for the first three joints
(of the arm) and 90° for the last three joints (of the wrist).
A simplified stiffness model for a PUMA-like manipulator is shown in
Fig. 2.5b. In this model, the links have been considered as rigid bodies. In fact,
in this type of robots, the payloads are limited and the compliant displacements are
in general due to the flexibility of joints only. The link compliant displacements
are much smaller than the compliant displacements that are due to the compliance
of motors as pointed out for example in [44]. Thus, in the model of Fig. 2.5b the
lumped parameters kT1 to kT6 take into account the stiffness motors and joints
only.
Moreover, if the only contributions to the overall compliance are given by
motor compliances, the stiffness matrix K can be computed through Eq. (2.7)
where J is the well-known Jacobian matrix of the PUMA-like robot. The matrix
KP in Eq. (2.7) can be computed as a diagonal matrix with lumped stiffness
parameters of the motors that can be set as kT1 = kT2 = kT3 = 5 9 106 Nm/rad
and kT3 = kT4 = kT5 = 5 9 104 Nm/rad as reasonable values by referring to a
PUMA 562. The stiffness matrix of the PUMA-like robot that can be computed
through Eq. (2.7) as function of the input joint angles. The expressions of the input
joint angles can be computed as functions of the coordinates (x, y, z, /, w, h) for
the position and orientation of the end effector from the well-known inverse
Kinematics of PUMA-like robot.
It is worth noting that the accuracy in the estimation of model data such as
geometrical dimensions and values of lumped stiffness parameters can signifi-
cantly affect the accuracy of the stiffness matrix that is computed through Eq.
(2.7). Thus, experimental tests should be carried out in order to validate stiffness
model and model data.
Results of the proposed stiffness analysis as applied to the PUMA-like archi-
tecture are reported in Table 2.2 and Figs. 2.6 and 2.7.

Table 2.2 Maximum values of compliant displacements and values of global stiffness indices
within the feasible workspace of PUMA 562
Dx (mm) Dy (mm) Dz (mm) D/ (°) Dw (°) Dh (°) GIMn GImn
1.81 1.76 1.98 1.80 2.73 2.75 5.6202e ? 034 5.3184e ? 017

30 G. Carbone

(a) 2 (b) 2

Δy [mm]
1.5 1.5
Δx [mm]

1 1

0.5 0.5

0 0
0 20 40 60 80 100 0 20 40 60 80 100
θ2 [deg] θ 2 [deg]

(c) 2 (d) 2

1.5 1.5
Δz [mm]

Δφ [deg]
1 1

0.5 0.5

0
0
0 20 40 60 80 100 0 20 40 60 80 100

(e) 3 (f) 3
Δψ [deg]

Δψ [deg]

2 2

1 1

0 0
0 20 40 60 80 100 0 20 40 60 80 100


Fig. 2.6 Compliant displacements of Puma-562 as a function of the input angle h2: a Linear
compliant displacement along X-axis; b Linear compliant displacement along Y-axis; c Linear
compliant displacement along Z-axis; d Angular compliant displacement about X-axis; e Angular
compliant displacement about Y-axis; f Angular compliant displacement about Z-axis

40
x 10
1
det K

0

-1
0 20 40 60 80 100
θ2 [deg]

Fig. 2.7 Determinant of the matrix K for the case in Fig. 2.6


(a) (b)

Fig. 2.8 CaPaMan (Cassino Parallel Manipulator) design: a A kinematic diagram; b A built
prototype at LARM

2.4.2 A 3 DOF Parallel Manipulator

The Cassino parallel manipulator (CaPaMan) has been considered to test the
engineering feasibility of the above-mentioned formulation as specifically applied
to parallel architectures which can be different from a general Gough-Stewart
platform. CaPaMan architecture has been conceived at LARM in Cassino since
1996, where a prototype has been built for experimental activity. A schematic
representation of the CaPaMan manipulator is shown in Fig. 2.8a, and the pro-
totype is shown in Fig. 2.8b. Indeed, by using the existing prototype, simulations
have been carried out to compute its stiffness performance.
Kinematics of CaPaMan manipulator has been already investigated in previous
works at LARM. In particular, matrices A and B have been formulated in the form
2 3
ðD À FÞb1 ca1 ðD þ 2FÞb2 ca2 Àð2D þ FÞb3 ca3
A ¼ 4 ðD þ FÞ b1 ca1 ÀDb2 ca2 ÀF b3 ca3 5 ð2:11Þ
b1 ca1 b2 ca2 b3 ca3
2 À pffiffiffiÁ 3
6E 3 0 0
6 rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
7
B¼6 4 0 2
E 9rp À 4E 075 ð2:12Þ
0 0 3
in which

E ¼ z2 þ z2 þ z2 À z1 z2 À z2 z3 À z1 z3
1 2 3
ð2:13Þ
D ¼ 2z2 À z1 À z3 ; F ¼ 2z3 À z1 À z2

32 G. Carbone

Fig. 2.9 A scheme for stiffness evaluation of a CaPaMan leg

with
zk ¼ bk sin ak ; for k ¼ 1; 2; 3 ð2:14Þ
By modeling each leg of CaPaMan as in Fig. 2.9, the stiffness matrix of
CaPaMan can be computed by using Eq. (2.6) with

CF ¼ MFN ; CK ¼ CP AÀ1
À1
d ð2:15Þ

where MFN is a 6 9 6 transmission matrix for the static wrench applied on H and
transmitted to points H1 H2 and H3 of each leg; Kp is a 6 9 6 matrix with the
lumped stiffness parameters of the three legs; Cp is a 6 9 6 matrix giving the
displacements of the links of each leg as a function of the displacements of points
H1, H2, and H3; Ad is a 6 9 6 matrix that has been obtained by using the Direct
Kinematics of the CaPaMan to give the position of point H on the movable plate as
function of the position of points H1, H2, and H3 in the form
XH ¼ A d v ð2:16Þ

with v = [y1, z1, y2, z2, y3, Dz3]T and XH = [xH, yH, zH, u, h, w]T. The derivation
of matrices MFN, Kp, Ad, and Cp for CaPaMan can be found in [18].
The lumped stiffness parameters have been assumed as
kbk = kdk = 2.6 9 106 N/m and kTk = 58.4 9 103 Nm/rad; the couplers ck have
been assumed rigid bodies because of the massive design that has been imposed to
have a ﬁx position of the sliding joints. Further, details on the derivation of the
matrices in Eqs. (2.15) and (2.16) can be found in [18]. In the numerical example,
for evaluation and design purposes we have assumed rp = rf, ak = ck, bk = dk.
Results of numerical simulations are shown in Figs. 2.10 and 2.11, while main
numerical results for stiffness performance are summarized in Table 2.3.


Fig. 2.10 Plots of the determinant of KCaPaMan: a Versus a1 = a2 = a3; b Versus bk = dk
(k = 1,2,3) with a1 = a2 = a3 = 60 °

Fig. 2.11 Compliant displacements of CaPaMan as a function of Fx = Fy = Fz when
Nx = Ny = Nz = 0 for a1 = a2 = a3 = 60°: a Dx; b Dy; c Dz; d Du; e Dh; f Dw

within the feasible workspace of CaPaMan with W = (1.0;1.0;1.0; 0.0;0.0;0.0)t
Dx (mm) Dy (mm) Dz (mm) D/ (°) Dw (°) Dh (°) GIMn GImn
-0.0982 0.0309 0.0357 5.7932 0.0000 1.8620 4.8367e ? 019 1.2882e ? 017

34 G. Carbone

Table 2.4 Illustrative value of forces as a function of the dimension L of an iron grasped object
Law Type of force Force [N] Force [N]
(at L = 1 mm) (at L = 100 mm))
kv L Van der Waals 0.119 1.19 10–9
kv = 1.19 102 kv = 1.19 10–10
ke L2 Electro-static 0.014 1.36 10–6
ke = 1.40 104 ke = 1.36 10–12
kG L3 Gravity 3.00 10–5 30.0
kG = 3.00 104 kG = 3.00 104
kI L 4 Inertia 3.06 10–9 0.306
kI = 3.06 103 kI = 3.06 103
km L4 Magnetic force 1.00 10–6 1.00 10–1
km = 1.00 105 km = 1.00 105
kg L Grasping 1 300
force kg = 1 103 kg = 3 103

2.4.3 A Two-Finger Milli-Gripper

The modeling and analysis of any grasping device should take into account its
peculiarities and constraints. For example, the modeling and analysis of a milli-
gripper should take into account many aspects including expected accuracy,
dimensions, displacement ranges, acting forces, and loads. For example, the
required accuracy can be considered inversely proportional to the geometric
dimensions since smaller objects require greater positioning accuracy. The dis-
placement and force capability are directly proportional to the geometric dimen-
sions, since bigger objects require larger displacements of fingers and greater
grasping force. For these reasons, in manipulative tasks at small scale the needed
accuracy is high, while displacement and force capability are of small magnitude
in agreement with the dimensions of the handled objects, [47].
A stiffness model for a two-finger milli-gripper is quite similar to the one shown
in Fig. 2.3, even if important differences must be considered such as the scale of the
acting forces and the kind of contact between the object and the fingers. As regards
the scaling of the acting forces, Table 2.4 can be deduced by using dimensional
analysis, similitude laws and experimental measurements as shown for example in
[47]. Table 2.4 illustrates the main forces and their proportionality to the geometric
dimension L of a grasped object by using coefficients kv, ke, kG, kI, km, and kg that
summarize have been computed from theoretical and experimental results in the
literature. Moreover, Table 2.1 shows numerical examples, giving the intensity of
different types of forces that have been evaluated in the case of L having a size of 1
and 100 mm. These examples show that, when the grasped object is of millimeter
order, forces due to gravity and inertia can be neglected. The action of Van der Waals
forces and electrostatic forces is still negligible as compared with the required
grasping force at millimeter scale. A further scale reduction can make adhesive
forces more significant than grasping forces, so that the releasing of the object
becomes the most critical phase at micro- or even nano-scale [59].


(a) Y (b)

kγ
O
X

Z γ T

Fig. 2.12 A flexural joint: a Manufacturing scheme; b Kinematic model

The operation of a mechanical milli-gripper strongly depends on the design and
behavior of the driving mechanism, which transmits the motion and force to the
gripping fingers. Theoretically, a milli-gripper could have the same mechanism
type of a conventional gripper. However, these mechanisms are not always fea-
sible due to the small dimensions. In fact, for example, conventional joints cannot
be easily miniaturized. This problem can be solved, for example, by using flexural
joints. In fact, flexural joints can be manufactured from a single piece of material
by using milling machines to provide a monolithic mechanism, which eliminates
interface wear and allow very high miniaturization, as pointed out in [44, 46, 47].
Figure 2.12a and b shows a design scheme and a kinematic model for a flexural
joint obtained by manufacturing two notches on a single piece of material to have
rotations about the Z-axis related with the stiffness parameter kc. The stiffness
about the X-axis and the Y-axis is much higher than about the Z-axis Therefore,
the rotations about the X-axis and the Y-axis can be neglected and the flexural
joint allows only a rotation c of few degrees about the Z-axis when a torque T is
applied. Actual misalignment of the actuation force or any unexpected forces may
cause rather large parasitic deflections in other direction than the desired one.
However, generally, this is enough for the microworld applications.
Figure 2.13 shows a design solution that has been proposed at LARM in
Cassino. It uses flexural joints to obtain four-bar Chebichev type driving mecha-
nism that allows an approximately straight-line motion of the fingers as shown in
the kinematic chain of Fig. (2.1c). Considering this specific planar case, one can

(a) (b) (c)

Fig. 2.13 A milli-gripper design: a Kinematic chain with design parameters; b A design scheme
considering shape memory alloy (SMA) actuators; c A built prototype at LARM

36 G. Carbone

(a) (b)
Q Q
W W

p p

Γ
Γ
d
d
1

c
1

X
c kd
1

X
α
1

1

α θ β
θ β
Y
Y
b
1

kα kβ
kT kθ
K
K

b kb
K

K

1
1

a
1

a
1

Fig. 2.14 Stiffness models with lumped parameters of the milli-gripper shown in Fig. 2.13:
a With two linear and one torsional lumped parameters; b With three torsional lumped
parameters

define the stiffness matrix K as a 3 9 3. The vector of compliant displacements
can be defined as DS = [Dx, Dy, Dh]t. The external acting wrench acting on the
point Q can be defined as W = [Fx, Fy, 0]t when the external moment is assumed
to be equal to zero. Referring to Fig. 2.13a a denotes the length of the frame, b and
d are respectively the input and the output length links, and c is the length of the
coupler whose fingertip point Q is located by the length p and the angle C.
Stiffness models of the milli-gripper in Fig. 2.13 can be proposed as shown in
Fig. 2.14 by considering three lumped parameters. In particular, Fig. 2.14a shows
a stiffness model with two linear lumped parameters that express the linear stiff-
ness of the input and output links and one angular lumped parameter that express
the stiffness of the input actuator and input flexural joint. Figure 2.14b shows a
stiffness model with three angular lumped parameters that express the angular
stiffness of the corresponding three flexural joints.
The bending of links and the stiffness of the actuator(s) can be also taken into
account as additive components of these angular lumped parameters.
If OXY is a fixed reference and assuming that a is the input angle, positive
counterclockwise, b the output angle, and h the angle between the generic position
of c and X-axis, from the loop closure equations one can write

x ¼ b cos a þ c cos h À p cosðC À hÞ
y ¼ b sin a þ c sin h þ p sinðC À hÞ ð2:17Þ
sin
h ¼ 2 tanÀ1 À ðisin2 a þ B2 ÀD2 Þ1=2 B þ D
a


with x and y being the position of the point Q along the X- and Y-axis, respectively
and with

a2 þ b2 À c 2 þ d 2 a
a
B ¼ cosa À ; C¼ À cos a;
b
2bd d ð2:18Þ
2 2 2 2
a a þb þc Àd
D ¼ cos a À
c 2bc
Thus, the kinematic equations (2.17) and (2.18) can be used to obtain the matrix
Ck.
If one refers to the stiffness model in Fig. 2.13a, the static equilibrium can be
expressed by referring to the equilibrium of the coupler as

Fx kb cosa kd cosb À kb sina Db
T

Fy ¼ kb sina kd sinb kT Dd ð2:19Þ
b cosa
Tz k r Àkd rd kT Da
b b b rT

with
c c
rb ¼ sinða þ hÞ þ p cosða þ hÞ; rd ¼ À sinða þ bÞ þ p cosða þ bÞ;
2 2
c
rT ¼ cosða À hÞ þ psinða À hÞ ð2:20Þ
2
where the moment of the forces has been computed about point Q.
By using Eq. (2.19) one can compute the product of the matrices CF and Kp in
the form

kb cosa kd cosb À kb sina
T

CF KP ¼ kb sa kd sinb

kT
b cosa
ð2:21Þ
k r Àkd rd kT
rT
b b b

Thus, Eqs. (2.17–2.21) can be used to compute the stiffness matrix K of the
milli-gripper as described in Eq. (2.6).
It is to note that it is usually not simple to find a close form expression for the
stiffness matrix even for a rather simple mechanism. A simplified expression of the
stiffness matrix one can obtained by referring to the stiffness model in Fig. 2.6b
and by using the Jacobian matrix of the proposed mechanism. For example, if one
assumes C = p, a = 0.005, b = 0.01, c = 0.005, d = 0.01, p = 0.01 m, the
Jacobian matrix can be computed as
2 3
À0:01sinaþ0:02sinðpþasinðÀ2textsinbþ2sinaÞ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi À 0:02sin(pþasinðÀ2sinbþ2sinaÞcosb
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi À0:005 sin h
6 1À4ðsin bÀsin aÞ 2
1À4ðsin bÀsin aÞ 2
7
6 7
J=6
6 0:01 sin a À 0:02cosðpþasinðÀ2sinbþ2sinaÞcosb
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
0:005 cos h 7
7
4 1À4ðsin bÀsin aÞ 5
2sinðaÀbÞ
sinðbÀhÞ 0 1
ð2:22Þ

38 G. Carbone

0.07 0.07

0.06 0.06

0.05 0.05

0.04 0.04

dy [mm]
dx [mm]

0.03 0.03

0.02 0.02

0.01 0.01

0 0
0 2 4 6 8 10 0 2 4 6 8 10
α (deg) α (deg)
15
-3 x 10
x 10 3
8
2.5
6
2
dtheta [deg]

detK

4 1.5

1
2
0.5

0 0
0 2 4 6 8 10 0 2 4 6 8 10
α (deg) α (deg)

Fig. 2.15 Simulation results by referring to the model in Eqs. (2.15, 2.16) versus the value of the
input angle a when WGiven = [1,1,0]t: a Linear compliant displacement along X-axis; b Linear
compliant displacement along Y-axis; c Angular compliant displacement about Z-axis;
d Determinant of the stiffness matrix K

The matrix Kp can be defined according to the model in Fig. 2.14b as

ka 0 0

K p ¼ 0 kb
0 ð2:23Þ
0 0 kh
Then, the Cartesian stiffness matrix can be easily obtained as in Eq. (2.7).
The stiffness matrix and compliant displacements can be computed at each
configuration of the milli-gripper by means of Eqs. (2.17–2.21) or (2.22 and 2.23)
by defining the values of the input angle and the values of the lumped stiffness
parameters. For example, if one refers to the model in Fig. 2.14b, and one sets
ka = kb = kh = 106 Nm/rad as reasonable values for the proposed milli-gripper,
one can use Eqs. (2.22) and (2.23) to compute the stiffness matrix K at any
configuration. For example, the specific configuration when the input angle a is
equal to zero yields

4:1758 À0:1215 À1:9500

K ¼ 105 À0:1215 0:4565 À0:0979
ð2:24Þ
À1:9500 À0:0979 1:0250


within the feasible workspace of the milli-gripper in Fig. 2.13
Dx (mm) Dy (mm) Dz (mm) GIMn GImn
0.068 0.069 7.551 e–003 2.6422e ? 008 3.2462e ? 007

One can compute the stiffness matrix K at any other configuration by setting
different values of the input angle a. A proper ‘‘for’’ loop can be implemented in
any programming environment (such as Matlab) to span the whole operation range
of the milli-gripper. The same code can be used to compute any local and global
stiffness performance index as well as the compliant displacements at a given
configuration for a given acting external wrench WGiven.
For example, if one assumes WGiven = [1,1,0]t then the complaint displace-
ments versus the input angle a can be computed as shown in Fig. 2.15.
The compliant displacements in Fig. 2.15 can be seen as local indices of stiffness
performance. Similarly, the determinant of the stiffness performance can give an
useful graphical information of the local stiffness performance as it grows when
stiffness performance is improving. Global stiffness indices can be computed such
as proposed in Eqs. (2.9) and (2.10) for the above-mentioned case of study as
reported in Table 2.5.

2.4.4 LARM Hand IV

LARM Hand IV, Fig. 2.2b, is a robotic hand having three one-DOF human-like
fingers that has been developed and built at LARM in Cassino. Figure 2.16 shows
a scheme of a finger including a kinematic model of its driving mechanism. Each
finger is basically composed of two four-bar linkage mechanisms as shown in
Fig. 2.16. The first phalanx is the input bar of the first four-bar linkage mechanism.
It is also the base frame of the second four-bar linkage mechanism. The second
phalanx is the input bar of the second four-bar linkage mechanism and it is also the
coupler of the first four-bar linkage mechanism. Then, the third phalanx is the
coupler of the second four-linkage mechanism. Table 2.6 shows the main design
parameters of a finger.
Referring to the scheme in Fig. 2.16, the angular velocities of the second and

_ _
third phalanxes can be defined as hg ¼ dhg dt and hj ¼ dhj dt, respectively. Both
_ _
hg and hj can be computed as function of the input angular velocity of the first
_
phalanx hb ¼ dhb =dt in the form
Á Á b sinðhb À he Þ Á
hf ¼hg ¼ À Á hb ð2:25Þ
f sin he À hf

40 G. Carbone

Fig. 2.16 Scheme of the LARM hand driving mechanism: a Complete system; b First phalanx;
c Second phalanx

Table 2.6 Main design parameters of a finger of LARM Hand IV
Phalanx 1 Phalanx 2 Phalanx 3
Frame Phalanx body Rod Phalanx body Rod Phalanx body
Label in Fig. 2.2 a b c d e f g h k j
Length (mm) 8.4 40.9 4.8 38.4 50 5.4 25 28.7 26 5.4

À Á
Á Á g sin hg À hk Á Á
hj ¼hc þ À Á hf Àhc ð2:26Þ
j sin hk À hj

_ _ _ _
where one can also replace the angular velocity hc with hb , since both hc and hb are
angular velocities of the same rigid body (the first phalanx) with respect to the
fixed reference frame. The time derivatives of Eqs. (2.25) and (2.26) can be also
used for computing the angular accelerations on each phalanx. The above-men-
tioned kinematic equations can be used to derive the speed of each phalanx as
function of the input speed or vice versa. Therefore, one can use Eqs. (2.25) and
(2.26) for verifying that the speed of a phalanx provides a human-like behavior.
The LARM Hand IV is also equipped with three force sensors on each finger for
measuring the grasping force on each phalanx. The location of these force sensors
is shown in the scheme of Fig. 2.17a. It is worth noting that in the model of
Fig. 2.3 each sensor has been modeled as a prismatic joint with a spring. The
lumped stiffness parameter of each spring as been assumed as k = 10 N/mm by
referring to a piezoresistive low-cost force sensor. Thus, the grasping force that is
measured on each phalanx can be computed F = kDd where is Dd is the compliant
displacement on each spring.
Figure 2.17b shows a scheme of LARM hand grasping a cylindrical object. In
this scheme, the object is in contact with the second phalanx of LARM Hand at
point P. The same grasping conditions of Fig. 2.17 have been modeled in


Fig. 2.17 A finger of LARM Hand IV: a Model of the grasp and contact; b A scheme of the
kinematic chain for the numerical solution in MSC.ADAMS environment

MSC.ADAMS environment as shown in the scheme of Fig. 2.18. The proposed
model in Fig. 2.18 has been carefully designed and implemented in order to obtain
a suitable numerical solution of the proposed grasping conditions. It is worth
noting that a firm grasp is achieved when all forces are in equilibrium. Therefore,
the input torque has to change as function of several parameters including the
external force acting on the object, and position, size, shape of the grasped object.
Numerical simulations of LARM Hand operation have been computed in
MSC.ADAMS environment.

Fig. 2.18 A MSC.ADAMS mode of LARM Hand IV: a A finger with main constraints; b Detail
of the third phalanx

42 G. Carbone

Fig. 2.19 Control PID with saturation block for the contact force control of the finger

The contact of the elastic bar with the frame and rollers have been modeled
with the characteristic of the contact data as: stiffness: 1.0E11 Pa; damping:
10.0E3 Ns/m; force exponent: 2.2 N; maximum penetration: 0.1 mm. The friction
has been modeled as a Coulomb force with the followings parameters: Static
coefficient 0.6; Dynamic coefficient 0.5; Stiction Transition Velocity 100 mm/s;
Friction Transition Velocity 1000 mm/s.
The operation mode is based on a control of the torque of the motor aiming to
obtain a desired maximum value for the contact force. A close loop control is
achieved by using as feedback the data that are obtained from force sensors on the
finger. The force sensors are modeled as a linear spring as it is shown in Figs. 2.17a
and 2.18b. The value of the contact force is obtained by evaluating the displacement
of the prismatic joint of the force sensor. Figure 2.19 shows the scheme of the PID
control with the input data that are obtained from the force sensor.
A saturation block has been added, as it is shown in Fig. 2.19. This block
imposes upper and lower bounds on a signal. When the input signal is within the
range specified by the lower limit and upper limit parameters, the input signal
passes through unchanged. When the input signal is outside these bounds, the
signal is set to the upper or lower bound. The system uses the measures that are
provided by the force sensors to trigger the change from one control scheme to
another. This trigger has been modeled with an ‘‘if’’ statements.
If the absolute value of the deformation of the spring that models a force sensor
is lower than a threshold value the system remains in the speed control, but if the
deformation is greater than this value the control switches to the contact force
control. A hysteretic gap has been added to avoid an oscillation from one control to
another in the transient phase between the two operation modes.
Several simulations have been carried out in MSC.ADAMS environment by
implementing the proposed grasp model and control algorithm. Figures 2.20 and
2.21 show results of the simulation of LARM Hand IV while grasping a rigid
cylinder of 50 mm of diameter, under various grasping conditions. The control has
been designed to obtain a normal contact force of 1.5 N for the two parallel fingers
and of 3 N for the opposite finger. A friction coefficient has been assumed at the
contacts with low values, as in usual robotic devices.
The integrator is an algorithm that solves the differential equations of a dynamic
problem over an interval of time during a simulation. The used integrator for the


Fig. 2.20 Simulation results of the LARM Hand grasp in Fig. 2.3 when an external disturbing
force is applied to the object: a Input angle; b Motor torque; c Contact grasp forces; d Reaction
forces on finger 1; e Reaction forces on finger 2; f Reaction forces on finger 3

simulation has been gear stiff integrator (GSTIFF), [60]. The GSTIFF integrator is
the default in MSC.ADAMS environment since it provides good solutions for
simulation of stiff models. The GSTIFF integrator uses a backwards differentiation
formula to integrate differential and algebraic equations. In addition, it assumes a
fixed time step that results in fixed coefficients for predicting the errors. The time
interval 0.1 (10 intervals in 1 s) has been selected after several attempts for
developing the simulation of the LARM Hand in order to obtain suitable numerical
results with suitable computational efforts. The simulation time has been 20 s in a
standard Pentium II computer.
The speed control is used from the initial position until the spring suffer a
compliant displacement of Dd = 0.1 mm. After this event occurs, the system
switches to the contact force control. Results also show the transition from speed
control to the contact force control. The PID constant Kp = 1E4, Ki = 1E5 and
Kd = 100 has been chosen in order to prevent upper-oscillation around the value
of the maximum force value.
Results in Figs. 2.20 and 2.21 show that the control is suitable for keeping a firm
grasp. In particular, Fig. 2.20a shows the values for the angles of the input bars of the
finger. These angles show that the finger does not return to its original configuration
with input angle at -0.155 rad. However, the new input angle value of 0.153 rad is
still a stable configuration. Figure 2.20b shows the motor torques that grow to their

44 G. Carbone

Fig. 2.21 Module of the external force that is applied to the centre of mass of the object that is
grasped for the simulation, whose results are reported in Fig. 2.13

maximum value of 400 Nm in order to return to a new stable condition after about
0.15 s. The contact forces are shown in Fig. 2.20c. The control cannot avoid that the
contact forces goes below the desired values of 3 N due to a restriction in the
maximum torque that the input motor can provide. However, contact forces never go
over 1.8 N. This value can be considered suitable for keeping the grasp and avoiding
any damage to the objects to be grasped. Figures 2.20d–f show the reaction forces in
the frame joints of the finger. In particular, plots in Fig. 2.20d show reaction forces
ranging from 0 to about 12 N with sufficiently smooth time history, and Fig. 2.20e
and d show reaction forces in the range from 1.5 to 4 N.

2.5 Experimental Determination of Stiffness Performance

Experimental determination of stiffness performance of multibody robotic systems
can be performed for calibration purposes and operation characterizations by
identifying the entries of stiffness matrix of the proposed formulation according to
practical aims that can be related also to standard codes that are reported in [51,
52]. Nevertheless, the computation of the coefficients of the stiffness matrix
requires carrying out experimental tests in which compliant displacements and
wrenches will be measured contemporaneously.
One should note that the stiffness matrix K can be symmetric if and only if
some conditions are satisfied on the external wrench and choice of reference
frames for the representation of compliant displacements as demonstrated for
example in [26–28]. Therefore, the computation of the 6 9 6 Cartesian stiffness
matrix K in the most general case requires the identification of its overall 36 kij
entries. The identification of all these 36 entries can be achieved if wrenches and
compliant displacements are measured in at least six experiments for a given
manipulator in a given configuration. In fact, with six experiments Eq. (2.2) can be
used to give as many equations as the 36 unknown entries of K in the form


01 1 1 1 1 1
10 1 01 1 0 1
Dx Dy Dz Du Dw Dh ... 0 0 0 0 0 0 k 11 Fx 0
B 0
B 0 0 0 0 0 ... 0 0 0 0 0 0 CB k 12 C B 1 Fy C B 0 C
CB C B C B C
B 0 0 0 0 0 0 ... 0 0 0 0 0 0 CBCB k 13 C B 1 Fz C B 0 C
B C B C B C
B 0
B 0 0 0 0 0 ... 0 0 0 0 0 0 CB k 14 C B 1 Nx C B 0 C
CB C B C B C
B 0 0 0 0 0 0 ... 0 0 0 0 0 0 CBCB k 15 C B 1 Ny C B 0 C
B C B C B C
B 0 0 0 0 0 0 ... 1
Dx 1
Dy 1
Dz 1
Du 1
Dw 1 CB k 16 C B 1 Nz C B 0 C
D h CB
B C B C B C
B : : : : : : ... : : : : : : CBCB : C B : C B : C
B C B C B C
B : : : : : : ... : : : : : : CBCB : C À B : C ¼ B : C
B C B C B C
B : : : : : : ... : : : : : : CBCB : C B : C B : C
B C B C B C
B 6D x 6
Dy 6
Dz 6
Du 6
Dw 6
Dh ... 0 0 0 0 0 0 CBCB k 61 C B 6 Fx C B 0 C
B C B C B C
B 0 0 0 0 0 0 ... 0 0 0 0 0 0 CBCB k 62 C B 6 Fy C B 0 C
B C B C B C
B 0 0 0 0 0 0 ... 0 0 0 0 0 0 CBCB k 63 C B 6 Fz C B 0 C
B C B C B C
B 0 0 0 0 0 0 ... 0 0 0 0 0 0 CBCB k 64 C B 6 Nx C B 0 C
B C B C B C
@ 0 0 0 0 0 0 ... 0 0 0 0 0 0 A@ k 65 A @ 6 Ny A @ 0 A
6 6 6 6 6 6 6
0 0 0 0 0 0 ... Dx Dy Dz Du Dw Dh k 66 Nz 0

ð2:27Þ

where the kij coefficients refer to the stiffness matrix as
2 3
k11 k12 k13 k14 k15 k16
6 k21 k22 k23 k24 k25 k26 7
6 7
6k k32 k33 k34 k35 k36 7
K ¼ 6 31
6 k41 k42
7 ð2:28Þ
6 k43 k44 k45 k46 7
7
4 k51 k52 k53 k54 k55 k56 5
k61 k62 k63 k64 k65 k66
The numerical solution of Eq. (2.28) provides the required values of the 36
coefficients of the stiffness matrix in Eq. (2.14) once the wrenches (due to known
masses) and compliant displacements (due to those wrenches) that have been
measured in six experiments are available for a given configuration. These
experiments can be carried out by means of Milli-CaTraSys that has been con-
ceived and built at LARM as schematized in Fig. 2.22.
Milli-CaTraSys is a wire tracking system whose scheme is reported in
Fig. 2.23. It is composed of six LVDT sensors that can measure the distances ui
(i = 1,…,6). Then, the distances ui are used as radii of arcs from corresponding
center points Oi. The position of H on the robot end effector is defined as the
position of the point at which three arcs intersect. Similarly, it is possible to
measure the position of points F and Q. Then, the vectors connecting H, F, and Q
can be used to determine the end-effector orientation through the orientation
angles a, c and d, as shown in Fig. 2.22b. Thus, a trilateration technique can be
used with Milli-CaTraSys in order to measure both position and orientation of its
end effector that can be attached to a mobile body through reference points H, F
and Q on it, as shown in Fig. 2.23. Moreover, that known masses can be attached
on the free end of each wire as shown in Fig. 2.22a. Therefore, Milli-CaTraSys can
measure the changes in position and orientation of its end effector (compliant
displacements) while different known wrenches are applied. Several experimental
trials can be carried out for each configuration just by applying different masses to
the wires.
A virtual instrument in LabVIEW environment has been developed for acqui-
sition and processing the data from the LVDT sensors. This virtual instrument has

46 G. Carbone

(a) (b)

Fig. 2.22 A scheme of Milli-CaTraSys: a With the reference frame, LVDTs and masses mi as
applied to the wires ui (i = 1,…,6); b Orientation of the end-effector through the angles a, c and d

been used to measure the displacement of the cores inside the LVDT sensors.
In this configuration of Milli-CaTraSys, the H point can be measured with any
combination of three wires among the six available wires. In particular, the
combination of three wires 1, 3, 5, and the combination of three wires 2, 4, 6 have
been selected. In fact, these configurations are symmetric, have been proved to
reduce computational costs, and increase the accuracy of measurement, as reported
in previous experiences at LARM.

2.6 Cases of Study for Stiffness Experimental Tests

2.6.1 CaPaMan 2bis

CaPaMan 2bis is a parallel manipulator that has been designed and built at LARM
in Cassino. A kinematic scheme of CaPaMan 2bis is shown in Fig. 2.23, where the
fixed platform is FP and the moving platform is MP. MP is connected to FP
through three identical leg mechanisms and is driven by the corresponding artic-
ulation points. An articulated parallelogram AP, a revolute joint RJ and a con-
necting bar CB compose each leg mechanism. AP’s coupler carries the RJ and CB
transmits the motion from AP to MP through RJ; CB is connected to the MP by a
spherical joint BJ, which is installed on MP. Each plane, which contains AP, is


(a) (b)

Fig. 2.23 The CaPaMan 2bis: a A kinematic scheme; b A prototype with Milli-CaTraSys set up
at LARM in Cassino

rotated of p/3 with respect to the neighbour one. Design parameters of a k-th leg
are identified through: ak, which is the length of the frame link; bk, which is the
length of the input crank; ck, which is the length of the coupler link; dk, which is
the length of the follower crank; hk, which is the length of the connecting bar. The
kinematic input variables are the crank angles ak (k = 1,2,3). Sizes of MP and FP
are given by rp and rf, respectively. Table 2.7 reports the sizes of main design
parameters of CaPaMan 2bis.
Experimental tests have been carried out by applying six different wrenches for a
given configuration of CaPaMan 2bis. In particular, Fig. 2.24 shows the measured
compliant displacements when m1 = m2 = m3 = m4 = m5 = m6 = 0.03 kg and
CaPaMan 2bs is in its vertical configuration. Similar results have been obtained for
six different masses distributions. It is worth noting that the plots of Fig. 2.24 show
the measured compliant displacements versus time during an static experiment. This
is necessary in order to find the stationary values of the measured compliant dis-
placements after applying the external wrench. In particular, the stationary values
for the experimental test that is reported in Fig. 2.24 are Dx = -0.041, Dy = 0.035,
Dz = -0.155, Da = -1.964°, Dc = -1.667° and Dd = -0.277°.

Table 2.7 Sizes of main design parameters for CaPaMan 2bis
ak = ck (mm) bk = dk hk rP = rf (mm) ak
(mm) (mm) (°)
100 100 50 65 45:135

48 G. Carbone

(a) 0 (b)0.05

-0.01

-0.02
Δ x [mm]

Δ y [mm]
0
-0.03

-0.04

-0.05 -0.05
0 2 4 6 8 0 2 4 6 8
time [sec] time [sec]
(c) (d)
0.05 1

0
0
-0.05
Δ z [mm]

Δα [deg]

-1
-0.1

-2
-0.15

-0.2 -3
0 2 4 6 8 0 2 4 6 8

(e) 0.5 (f) 0

0
-0.2
-0.5
Δδ [deg]
Δγ [deg]

-1 -0.4

-1.5
-0.6
-2
-2.5 -0.8
0 2 4 6 8 0 2 4 6 8

Fig. 2.24 Measured compliant displacements for a wrench given by
m1 = m2 = m3 = m4 = m5 = m6 = 0.033 kg when CaPaMan2bis is in its vertical conﬁgura-
tion: a Dx; b Dy; c Dz; d Da; e Dc; f Dd

The experimental analysis has given results, which conﬁrm the numerical
computations of the stiffness matrix. For example, when the three legs of CaP-
aMan 2bis are inclined of 45° the stiffness matrix is measured as given by


2 3
0:013 0:002 0:040 À10:13 0:031 3:058
6 0:003 0:001 0:010 À2:595 0:012 0:782 7
6 7
6
8 6 0:000 0:000 0:001 À0:093 À0:013 0:028 7
K ¼ 10 6 7 ð2:29Þ
6 0:000 0:000 0:000 À0:027 0:000 0:008 7
7
4 À0:000 À0:000 À0:001 0:195 À0:001 À0:059 5
0:000 0:000 0:000 À0:051 0:000 0:015

when wrenches are obtained by using additional masses of 50 grams on each wire
once in tension.
When two legs are inclined of 60° and one leg is in vertical configuration, the
stiffness matrix is measured as
2 3
0:014 À0:094 0:268 7:416 4:410 À2:051
6 0:009 À0:174 0:606 1:740 13:99 À5:299 7
6 7
6 0:008 À0:138 0:500
56 20:09 7:603 À5:923 7
K ¼ 10 6 7 ð2:30Þ
6 À0:000 0:000 À0:000 0:003 0:007 À0:000 7 7
4 À0:000 0:000 À0:001 0:169 À0:026 À0:050 5
À0:000 0:000 À0:000 À0:307 À0:126 0:092
The determinant of the stiffness matrix in Eq. (2.29) is equal to 5.987 9 1020
while the determinant of the stiffness matrix in Eq. (2.30) is equal to -
2.452 9 1020. By using this measure one can conclude that CaPaMan 2bis is stiffer
in the first configuration, as intuitively expected.

2.6.1.1 LARM Hand IV

The attached problem is to determine the performance of the LARM Hand in terms
of operation properties. This has been obtained by using the system Milli-CaTraSys.
A proper end effector has been installed on a finger of the LARM Hand as shown in
Fig. 2.25 and Fig. 2.26 in order to provide a proper location of points F, Q, and H and
suitable frame for attaching the wires. In particular, three wires have been attached to
point H, two wires to point F, and one wire to point Q. This setup refers to the Gough-
Stewart 3-2-1 parallel manipulator configuration. Experiments with two redundant
wires have been also carried out to have a 3-3-3 configuration and to validate the use
of the 3-2-1 configuration with redundant measurement data.
A calibration process has been carried out on the experimental setup, as pro-
posed for example in [24–27], in order to determine a suitable initial configuration
for the measuring system and to verify its accuracy.
As a result of the experimental calibration process, it has been determined an
average accuracy of Milli-CaTraSys of about 0.1 mm when LVDT sensors are
used with 100 mm range and an accuracy of Milli-CaTraSys of about 0.01 mm
when LVDT sensors are used with 2.5 mm range.
Experimental tests have been carried out by means of the setup that is shown in
Fig. 2.25. In the experimental tests, pretension of all wires has been obtained by

50 G. Carbone

(a) (b)

w3

w5

w7

Fig. 2.25 The experimental set-up for the test-bed operations with LARM Hand: a A scheme;
b The laboratory set-up

Fig. 2.26 End-effector for Milli-CaTraSys that has been attached to the fingertip of LARM
Hand; a A zoomed frontal view with connections for six wires in a 3-2-1 platform configuration;
b A zoomed lateral view with connections for six wires in a 3-3-3 platform configuration

means of known masses mi (i = 1,…,6) of 30 g. This pretension value has been set
up experimentally since it keeps all the wires pulling during the whole duration of
experimental tests. Moreover, this pretension produces negligible compliant dis-
placements of the twisted iron wires that have been used in Milli-CaTraSys.
Experimental tests have been carried out by operating the LARM Hand in an
open-close mode. Namely, the fingers of LARM Hand begin the tests in the fully
open configuration; they start moving after about 1.6 s; they move to the fully
closed configuration in about 1 s; they wait in the fully closed configuration about
2 s; they move back to the fully open configuration in about 1 s.
For example, Fig. 2.27 shows the plots of the measured lengths of the wires 1-
3-5 during an experimental test. Then, the measured lengths of wires are converted
through trilateration technique to give the position of the fingertip of LARM Hand.


(a) (b) (c)
Lenght of wire 1 [m]


Time [sec] Time [sec] Time [sec]

Fig. 2.27 Plots of the measured lengths of the wires of Milli-CaTraSys during an open-close
operation mode for LARM Hand: a Wire 1; b Wire 3; c Wire 5

The position coordinates of the fingertip of LARM Hand during the experimental
test in Fig. 2.27 are reported in Fig. 2.28 as function of time. In particular,
Fig. 2.28a shows a 3D view of the operation workspace of a fingertip motion of
LARM Hand in Cartesian coordinates. Figure 2.28b–d show the projection of the
operation workspace of a fingertip motion of LARM Hand onto XY, YZ, and XZ
planes, respectively. The measured motion ranges have been about 0.005 in X
direction, 0.032 in Y direction, 0.113 m in Z direction, respectively.

Fig. 2.28 Workspace of a fingertip motion of LARM Hand during an open-close operation
mode: a A 3D view; b Projection onto XY plane; c Projection onto YZ plane; d Projection onto
XZ plane

52 G. Carbone

One should note that a nonzero value of the motion range in X direction is an
evidence of a slightly nonplanar motion of the finger mechanism. This motion can
be although as given by a certain clearance in the joints. Moreover, due to the
presence of this clearance, the motion in the closing phase does not coincide with
the motion in the opening phase as shown in Fig. 2.28. However, the above-
mentioned motion ranges show a good match with those referring to a cylindrical
grasping mode by an average human hand.
Additional experimental tests with the same set up have been carried in static
conditions by applying known wrenches on the LARM Hand fingertip. Maximum
compliant displacement that have been measured on the fingertip of LARM Hand
has been of 0.1 mm for a maximum force of 3 N in the same direction. Thus, a
stiffness coefficient can be computed as at least 30,000 N/m. The above-mentioned
values of maximum compliant displacement and stiffness coefficient give a proof
of a quite good stiffness behavior for LARM Hand operation, whose average
grasping force is 3 N. A more accurate estimation of stiffness behavior can be
obtained by computing a 6 9 6 Cartesian stiffness matrices for LARM Hand
through a general procedure that has been proposed in [28].

2.7 Conclusions

This section has given fundamentals of stiffness modeling and analysis by refer-
ring to robotic systems for grasping tasks. A formulation has been detailed for the
computation of the Cartesian stiffness matrix K. Considerations on local and
global stiffness properties have been reported also within a numerical procedure.
Cases of study have been proposed for the stiffness analysis of manipulators
having serial or parallel architecture. Stiffness modeling and analysis has been
reported also for mechanical end-effectors such as two-finger grippers or robotic
hands, given their significance in grasping tasks. A procedure for experimental
validation has been outlined and two cases of study have been reported.

References

1. Rivin EI (1999) Stiffness and damping in mechanical design. Marcel Dekker Inc., New York
2. Ceccarelli M (2004) Fundamentals of mechanics of robotic manipulation. Kluwer, Dordrecht
3. Tsai LW (1999) Robot analysis: the mechanics of serial and parallel manipulators. Wiley,
New York, pp 260–297
4. Duffy J (1996) Statics and kinematics with applications to robotics. Cambridge University
Press, Cambridge, pp 153–169
5. Nof SY (ed) (1985) Handbook of industrial robotics. Wiley, New York
6. Merlet J-P (2006) Parallel robots. Springer, Dordrecht
7. Carbone G (2003) Stiffness evaluation of multibody robotic systems. Ph D Dissertation,
LARM, University of Cassino, Cassino
8. Gosselin C (1990) Stiffness mapping for parallel manipulators. IEEE Trans Robot Autom
6(3):377–382


9. Tahmasebi F, Tsai LW (1992) Jacobian and stiffness analysis of a novel class of six-dof
parallel minimanipulators. In: Proceedings of the ASME 22nd biennial mechanism
conference, Scottsdale, vol 47, pp 95–102
10. Gosselin CM, Zhang D (2002) Stiffness analysis of parallel mechanisms using a lumped
model. Int J Robot Autom 17(1):17–27
11. Simaan N, Shoham M (2003) Stiffness synthesis of a variable geometry six-degrees-of-
freedom double planar parallel robot. Int J Robot Res 22(9):757–775
12. Kim HY, Streit DA (1995) Configuration dependent stiffness of the Puma 560 manipulator:
analytical and experimental results. Mech Mach Theory 30(8):1269–1277
13. Ceccarelli M, Carbone G (2005) Numerical and experimental analysis of the stiffness
performance of parallel manipulators. In: 2nd international colloquium collaborative research
centre 562, Braunschweig, pp 21–35
14. Chakarov D (2001) Analysis and synthesis of the stiffness of a hybrid manipulator with
redundant actuation. In: Proceedings of the 5th Magdeburg days of mechanical engineering,
Magdeburg, pp 119–127
15. Ciblak N, Lipkin H (1999) Synthesis of cartesian stiffness for robotic applications. In:
Proceedings of the IEEE international conference on robotics and automation ICRA’99,
Detroit, vol 3, pp 2147–2152
16. Pigoski T, Griffis M, Duffy J (1998) Stiffness mappings employing different frames of
reference. Mech Mach Theory 33(6):825–8381998
17. Carbone G, Ceccarelli M (2004) A stiffness analysis for a hybrid parallel-serial manipulator.
Robot Int J 22:567–576
18. Ceccarelli M, Carbone G (2002) A stiffness analysis for CaPaMan (Cassino parallel
manipulator). Mech Mach Theory 37(5):427–439
19. Svinin MM, Hosoe S, Uchiyama M, Luo ZW (2002) On the stiffness and stiffness control of
redundant manipulators. In: IEEE international conference on robotics automation ICRA
2002, Washington, pp 2393–2399
20. Cutkosky MR, Kao I (1989) Computing and controlling the compliance of a robotic hand.
IEEE Trans Robot Autom 5(2):151–165
21. Tsumugiwa T, Yokogawa R, Hara K (2002) Variable impedance control with virtual stiffness
for human-robot cooperative peg-in-hole task. In: Proceedings of the IEEE/RSJ international
conference on intelligent robots ans systems IROS’02, Lausanne, pp 1075–1081
22. Chakarov D (1998) Optimization synthesis of parallel manipulators with desired stiffness.
J Theor Appl Mech 28(4). Avaliable. http://www.imbm.bas.bg/IMBM/LMS/Chakarov/
TPM98.pdf
23. Liu X-J, Jin Z-L, Gao F (2000) Optimum design of 3-dof spherical parallel manipulators with
respect to the conditioning and stiffness indices. Mech Mach Theory 35(9):1257–12672000
24. Carbone G, Ottaviano E, Ceccarelli M (2007) An optimum design procedure for both serial
and parallel manipulators. IMechE Part C J Mech Eng Sci 221(7):829–843
25. Huang S, Schimmels JM (2000) The bounds and realization of spatial compliances achieved
with simple springs connected in parallel. IEEE Trans Robot Autom 14(3):466–475
26. Zefran M, Kumar V (1997) Affine connections for the cartesian stiffness matrix. In:
Proceedings of the IEEE international conference on robotics and automation ICRA’97,
Albuquerque, vol 2, pp 1376–1381
27. Howard S, Zefran M, Kumar V (1998) On the 6 9 6 Cartesian stiffness matrix for three-
dimensional motions. Mech Mach Theory 33(4):389–408
28. Chen S-F, Kao I (2000) Geometrical method for modeling of asymmetric 6 9 6 cartesian
stiffness matrix. In: Proceedings of the IEEE/RSJ international conference on intelligent
robots and systems IROS2000, Takamatsu, pp 1217–1222
29. Pham DT, Heginbotham WB (1986) Robot grippers. IFS Publications Ltd, Bedford
30. Rosheim ME (1996) Robot surrogate: work in progress. In: International conference on
robotics and automation, Minnesota, pp 399–403
31. Cutkosky MR (1989) On grasp choice, grasp model, and the design of hands for
manufacturing tasks. IEEE Trans Robot Autom 5(3):269–279

54 G. Carbone

32. Iberal T (1997) Human prehension and dexterous robot hands. Int J Robot Res 6(3):285–299
33. Fukaya N, Toyama S, Asfour T, Dillmann R (2000) Design of the TUAT/Karlsruhe
humanoid hand. In: IEEE/RSJ international conference on intelligent robots and systems,
Takamatsu, vol 3, pp 1754–1759
34. Butterfass J, Grebenstein M, Liu H, Hirzinger G (2001) DLR-Hand II: next generation of a
dexterous robot hand. In: IEEE international conference on robotics and automation, Seoul
Korea, pp 109–114
35. Zhang Y, Han Z, Zhan H, Shang X, Wang T, Guo W (2001) Design and control of the BUAA
four-fingered hand. IEEE Trans Robot Autom 3:2517–2522
36. Gosselin CM, Mountambault S, Gosselin CJ (1993) Manus Colobi: preliminary results on the
design of a mechanical hand for industrial applications. In: 19th ASME design automation,
Albuquerque, vol 65–1, pp 585–592
37. Dechev N, Cleghorn WL, Nauman S (1999) Multiple finger, passive adaptive grasp proshetic
hand. Mech Mach Theory 36:1157–1173
38. Townsend WT (2000) The Barretthand grasper—programmably flexible parts handling and
assembly. Ind Robot Int J 27(3):181–188
39. Venkataraman ST, Iberall T (eds) (1989) Dexterous robot hands. Springer, New York
40. Boudreault E, Gosselin C (2006) Design of sub-centimetre underactuated compliant gripper.
In: Proceedings of IDETC/CIE 2006 ASME 2006, 30th mechanisms robotics conference,
Philadelphia. Paper DETC2006-99415
41. Ceccarelli M, Luyckx I, Vanaelten W (1996) Grasp forces in two-finger grippers: modeling
and measuring. In: 5th international workshop on robotics in Alpe-Andria Danube region
RAAD’96, Budapest, pp 321–326
42. Ceccarelli M, Nava Rodriguez NE, Carbone G (2006) Design and tests of a three-finger hand
with 1-dof articulated fingers. Robot Int J 24(2):183–196
43. Carbone G, González A (2011) Numerical simulation of the grasp operation by LARM hand
IV, A three finger robotic hand. Robot Comput Integr Manuf 27(2):450–459
44. Carbone G, Jeckel M, Havlík S, Ceccarelli M (2003) An optimum multi-objective design
procedure for microgripping mechanisms. In: 12th international workshop on robotics in
Alpe-Andria-Danube region RAAD 2003, Cassino, paper 055RAAD03
45. Penisi OH, Carbone G, Ceccarelli M (2002) Optimum design and testing of mechanisms for
two-finger grippers. Int J Mech Control 03(01):9–20
46. Ceccarelli M, Carbone G, Kerle H (2001) Designing mechanisms for two-finger
microgrippers. In: CD proceedings of the 10th workshop on robotics and Alpe-Adria-
Danube region RAAD’01, Wien, paper RD-021
47. Carbone G, Ceccarelli M, Kerle H, Raatz A (2001) Design and experimental validation of a
microgripper. Fuji Int J Robot Mechatron 13(3):319–325
48. Ceccarelli M, Carbone G (2010) Design and operation of fingered hands and two-finger
grippers for space applications as from experiences at LARM. In: IX international scientific-
technical conference. Control vibration technologies and machines VIBRATION 2010,
Kursk, pp 208–216 (in Russian) ISBN 978-5-7681-0561
49. Lanni C, Carbone G, Havlík Š, Ceccarelli M (2007) Experimental validation of a milli-
gripper based on Chebyshev mechanism. In: Proceedings of the 16th international workshop
on robotics in Alpe-Adria-Danube region, Ljubljana, CD Proceedings, paper n. FP_GG2,
pp 42–51
50. Vaishnav RN, Magrab EB (1987) A general procedure to evaluate robot positioning errors.
Int J Robot Res 6(1):59–74
51. ANSI, American National Standards Institute (1990) American national standard for
industrial robots and robot systems: point-to-point and static performance characteristics—
evaluation. ANSI/RIA 15.05-1-1990, New York
52. UNI, Italian National Institute for Standards (1995) Manipulating industrial robots:
performance criteria and related test methods. UNI EN 29283 (= ISO 9283), Milan


53. Carbone G, Ceccarelli M (2004) A procedure for experimental evaluation of Cartesian
stiffness matrix. In: 15th CISM-IFToMM symposium on robot design, dynamics and control,
paper Rom04-24, Montreal
54. Carbone G, Ceccarelli M (2010) A comparison of indices for stiffness performance
evaluation. Front Mech Eng 5(3):270–278. doi:10.1007/s11465-010-0023-z
55. Pratt GA, Williamson MM, Dillworth P, Pratt J, Ulland K, Wright A (1995) Stiffness isn’t
everything. In: Preprints of the 4th international symposium on experimental robotics
ISER’95, Stanford, available. http://www.ai.mit.edu/projects/leglab/publications/
sitffness_isnt_everything.pdf
56. Schimmels JM (2001) Multidirectional compliance and constraint for improved robotic
deburring. Part 1: improved positioning. Robot Comput Integr Manuf 17(4):277–286
57. English CE, Russell D (1999) Mechanics and stiffness limitations of a variable stiffness
actuator for use in prosthetic limbs. Mech Mach Theory 34(1):7–25
58. Alici G, Shirinzadeh B (2003) Exact stiffness analysis and mapping for a 3-SPS ? S parallel
manipulator. In: 7th international conference on automation technology AUTOMATION
2003, Taiwan, paper F120
59. Zhou Y, Nelson BJ (1998) Adhesion force modeling and measurement for
micromanipulation, microrobotics and micromanipulation. In: Sulzmann A, Nelson BJ
(eds) International society for optical engineering, vol 3519. SPIE, Boston, pp 169–180
60. MSC.ADAMS (2010) Documentation and Help. User CD-ROM

http://www.springer.com/978-1-4471-4663-6

Grasping in robotics

More Related Content

What's hot (19)

Similar to Grasping in robotics (20)

More from Springer (20)

Recently uploaded (20)

Grasping in robotics