See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224213388

The Twisted String Actuation System: Modeling and Control

Conference Paper in IEEE/ASME Transactions on Mechatronics · August 2010

DOI: 10.1109/AIM.2010.5695720 · Source: IEEE Xplore

CITATIONS READS

101 2,324

6 authors, including:

Chris May Benedikt Holz

Universität des Saarlandes Universität des Saarlandes
21 PUBLICATIONS 360 CITATIONS 19 PUBLICATIONS 268 CITATIONS

SEE PROFILE SEE PROFILE

Ciro Natale Gianluca Palli

Università degli Studi della Campania "Luigi Vanvitelli" University of Bologna
172 PUBLICATIONS 2,948 CITATIONS 157 PUBLICATIONS 2,766 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

SAPHARI View project

LOCOMACHS View project

All content following this page was uploaded by Ciro Natale on 15 August 2016.

The user has requested enhancement of the downloaded file.

 The Twisted String Actuation System: Modeling and Control
Thomas Würtz1 , Chris May1 , Benedikt Holz1 , C. Natale2 , G. Palli3 and C. Melchiorri3
1Universität des Saarlandes, Campus A5 1, Saarbrücken, Germany
2DII - Seconda Università di Napoli, Via Roma 29, Aversa, Italy
3 DEIS - Università di Bologna, Viale Risorgimento 2, Bologna, Italy
Twist
Angle Actuation
Abstract— This paper describes a novel actuation system for [rad] Length [mm]
very compact and light-weight robotic devices, like artificial
hands. The actuation concept presented here allows the imple- 50.00
mentation of powerful tendon-based driving systems, using as 0
actuators small-size DC motors characterized by high speed and
48.40
low torque. After the presentation of the basic concept of this
novel actuation system, the constitutive equations of the system 4π
are given, validated by means of laboratory tests. Moreover, the 43.37
problem of tracking a desired actuation force profile is taken 8π
into account, considering as load a mass-spring-damper system. 33.10
A control algorithm based on a second-order sliding manifold
has been firstly evaluated by means of simulations, and then 12π
validated by experiments. This output-feedback controller has
Actuation
been chosen to guarantee a high level of robustness against Linear Motor Frame
disturbances, parameters variations and uncertainties while Slider Output
maintaining a low computational burden. Encoder
Load Cell Twisted String Shaft
Index Terms— Actuation, Tendons, Transmission System, θ
Robotic Hands.
p
I. I NTRODUCTION
Linear Motor Stator DC Motor
Axial
Many robotic hands have been developed in the past, and (with Integrated Linear Encoder)
Bearing
(Rotative)
several solutions concerning their actuation system have been
proposed, differing mainly in the types of motors (often Fig. 1. Basic concept (top) and schematic representation of the twisted
string actuation system (bottom).
electrical rotative motors, sometimes pneumatic [1], [2], [3]),
in their location (i.e. integrated in the hand [4] or remotely
placed in the forearm), and in the transmission system (gears,
paper, is a very compact and low cost linear actuation system.
tendons or flex-shaft [5]).
As a matter of fact, with an appropriate choice of the
The development of a tendon-driven robotic hand [6],
rotative electric motors and of some design parameters of the
like those reported in [7], [8], [9], [10], implies strong
strings (in particular the radius and length), it is possible to
requirements on the actuation system. In particular, linear
satisfy all of the tight requirements for the implementation
actuators are needed, able to apply at least 40 N with a
of miniaturized and highly-integrated mechatronic devices,
stroke of about 1 · 10−2 m and a maximum velocity of
paving the way for the next generation of multifingered
2 · 10−2 m s−1 . Within the DEXMART project† [11] a
robotic hands.
new robotic hand corresponding in size with the human’s
In this paper, a novel concept of an actuation system is
one is under development. In this device, 20 independent
illustrated, presenting its kinetostatic and dynamic models as
actuators are necessary to drive the whole hand, and the
well as an algorithm for position/force control of a finger’s
entire actuation system must be contained within the forearm,
joint. Experimental activity validating both the model and
with a limited overall weight (e.g. less than 1 kg). Also
the control strategies is discussed, and comments and plans
the power consumption is another important parameter to
for future activity are finally reported.
analyze for avoiding the use of thermal dissipators. As it
turns out, commercially available linear motors do not satisfy II. T HE BASIC C ONCEPT
the given specifications, mainly in terms of dimensions and
The basic idea of the actuation system is quite simple and
weight.
is illustrated schematically in Fig. 1: one or more strings are
The use of the Twisted String Actuation System, the
connected in parallel on one end to a rotative electrical motor
novel actuation concept developed at Saarland University’s
and on the other end to the load to be actuated. The rotation
Laboratory of Process Automation and presented in this
imposed to the strings by the electrical motor reduces their
† This research has been partially funded by the EC Seventh Framework length, generating a linear motion at the load side.
Programme (FP7) under grant agreement no. 216239 as part of the IP This actuation concept, because of its high (though non-
DEXMART (DEXterous and autonomous dual-arm/hand robotic manipu- linear) reduction ratio, permits the use of very small and
lation with sMART sensory-motor skills: A bridge from natural to artificial
cognition). lightweight electric motors and therefore is very interest-
‡ Patent pending. ing in applications where size and weight are of crucial
 Real-Time
Control System rs
String
rs rc

Linear Motor

Load Cell Rotative Motor Encoder

(a) Rope section with n = 6. (b) Two-string section with
null core radius (n = 2).
Fig. 2. Experimental setup.
Fig. 4. Twisted string transmission sections for different values of the
parameters n and rc .

this will greatly simplify the design of the force control

algorithm. Nevertheless, the controller will be shown to be
effective also with elastic strings both in simulation and in
Fig. 3. Finger prototype actuated by the twisted string actuation system. experiments.
A. Kinetostatic Model of the Transmission System
importance. This concept was firstly implemented in an The kinetostatic model takes into account the effects due
experimental setup for verifying its main properties, see to the finite string dimensions and elasticity, together with the
Fig. 2, and is currently adopted as the actuation system for a case of multiple fibers forming a single string. This analysis
novel robotic finger 1:1 with the human finger, see Fig. 3. In will provide some useful information for the design of the
this robotic device, four independent twisted string actuation actuation system and for the choice of the driving motors,
systems have been integrated in a very compact design. like the limits of the contraction and reduction ratio of the
The experimental setup used to test twisted string actuation transmission system.
is comprised of a small rotative DC motor (Faulhaber 2233) As simplifying hypothesis, it is assumed here that some
and a string (single or in a pair) aligned along the rotation strands do not contribute to the total axial force: these fibers
axis of the motor and connected at one end to the motor form the core of radius rc of the helix, see Fig. 4(a). The
output shaft (without any speed reducer) and at the other load force Fz is balanced by the n external strands, of radius
end to the load, which in the setup is emulated by a linear rs that form n coaxial helices of radius r = rs + rc . As limit
motor (Linmot P01-37×120) equipped with a load cell for case, Fig. 4(b) shows a string formed by a pair of twisted
measuring the actuation force. The rotative motor is equipped strands, for which rc = 0 and r = rs , considering the helices
with an optical encoder to measure the rotation angle, while formed by the strand axes.
the encoder integrated in the linear motor is used to measure In order to obtain the relationships describing the statics
the actuation elongation. A suitable controller is used to drive of the actuation system, assume that the strands constituting
the load (the linear motor) so that it behaves as a mass- the string form an ideal helix of constant radius r = rc + rs
spring-damper system with adjustable parameters. Note that along the whole range of the motor angular position θ.
the encoders are used only for monitoring purposes and to With reference to Fig. 5(a), the external torque τL is
let the linear motor emulate any desired load. balanced by the tangential force Fτ , i.e.
A detailed kinetostatic model of this actuation system τL
Fτ = . (1)
is presented and discussed in the following sections, re- r
porting both simulation and experimental results. Another Assuming that the load is equally distributed over the n
important aspect considered in this paper is the possibility strands that form the string, one obtains
of controlling the system using only force feedback. In
Fτ
fact, no measurement of the system state (motor angular Fi sin α = , (2)
displacement and velocity) neither accurate knowledge of n
the system parameters (e.g. string or actuation length, string where α is the helix slope and Fi is the axial force of
radius, motor or string preload angle) are required. each strand. The resulting total axial force Fz acting on the
transmission system is
III. M ODELING OF THE ACTUATION S YSTEM τL
Fz = n Fi cos α → Fz = . (3)
In this section, the kinetostatic model of the transmission r tan α
system will be derived taking into account the elasticity It is supposed here that the strands act as linear springs
of the string, while a simplified dynamic model will be with the capability of resisting tensile forces only and not
presented under the assumption that the string is rigid since compressive forces. The total length of a strand L can then
 Fτ
2r Fi
α

α
α
Fz
L n rc

Fi p
rs
τL
Fig. 6. Simplified geometry and force distribution of the twisted string
transmission (n = 2).
Fτ θr
(a) Lateral and top view. (b) Unwrapped helix.
The helix slope satisfies the relation
Fig. 5. Helix schematic representation.
2πr
tan α = (9)
q
be computed as the unloaded length L0 plus the elongation and, from (7), it is possible to note that the maximum helix
due to the fiber tension Fi and the stiffness K of the strand slope is
(normalized with respect to the length unit), i.e. πr
tan αmax = . (10)
L0 p n rs
L = Fi + L0 = θ2 r2 + p2 , (4) From Eqs. (9) and (6) and in view of (3), the reduction ratio
K
where the right-hand side can be easily obtained by “unwrap- of the transmission can be found as
ping” the helix of total length L and radius r and applying τL θr2
h(θ) ≡ = r tan α = (11)
the Pitagora’s theorem to the resulting triangle in Fig. 5(b). Fz p
It follows that the actuation elongation p (i.e. the load
and it goes form 0 when θ = 0 to h(θ) = πr n when θ = θmax
position) is given by
s  and rc = 0.
2 All the model described above can be simplified assuming
Fi
p = L20 1 + − θ2 r2 (5) the string to be rigid, namely the stiffness K in Eqs. (4)
K
and (5) tends to ∞. Under this assumption, the total length
whereas, from the helix geometry, the pitch of the helix q is of the string is always equal to L0 and the actuation stroke
related to the string length p and to the motor rotation angle ∆p is given by
θ by the relation q
2 π p = q θ. (6) ∆p = L0 − p = L0 − L20 − θ2 r2 = L0 (1 − cos α) ≡ k(θ).
(12)
Fig. 6 shows a limit condition for the case of two strings
In such a case, the reduction ratio in (11) is simply given by
(n = 2): due the finite string radius, a maximum value of
the angle α exists such that the strings form a closely wound θ r2
h(θ) = p . (13)
helix. With reference to Fig. 6 and with some simplifications L20 − θ2 r2
on the system geometry, in the general case of n strings this
condition is given by B. Experimental Validation
qmin = 2 n rs . (7) The experimental setup used to validate the twisted string
transmission models proposed in this paper is described in
Combining (6) and (5), the maximum motor angle and Sec. I and depicted in Fig. 2, while the system parameters
actuation contraction in case of no load (i.e. Fz = 0, Fi = 0) are reported in Tab. I. The string used during the experiments
can then be defined as is a commercial ∅ 0.2 mm Dyneema fishing line.
L0 L0 Fig. 7 reports the comparison between the theoretical
θmax = q , pmin = q 2 2
2 2 π r
r n model (12) and the experimental data obtained in very
r2 + s 2 π r 2 n2 + 1 s
low load (1 N) condition. Note that the error between the
and the maximum no-load relative contraction is theoretical model and the experimental data is very small.
L0 − pmin 1 The first plot of Fig. 8 shows the effect of the load force Fz
=1− q 2 2 (8)
L0 π r
+1 on the transmission contraction observed during experiments
rs2 n2
with respect to the theoretical model: it is evident how the
It is important to note that in case of null core radius (rc = transmission contraction varies with the load force denoting a
0, n = 2), the maximum actuation contraction is about 46% finite transmission stiffness, anyway the capability of moving
of the maximum string length L0 . the load by twisting the string is preserved. The second
 500 0.04
theoretical

Motor Position [rad]

400 5 N load
0.03 10 N load
300

Displacement [m]
20 N load
200 0.02 30 N load
40 N load
100 50 N load
0.01
0
0.03
Experimental 0
Displacement [m]

Theoretical
0.02
−0.01

6000
0.01 5 N load
5000 10 N load
0 20 N load

Stiffness [N m−1]
6 30 N load
4000
Model Error [10−4 m]

4
40 N load
50 N load
2 3000
0
2000
−2
−4 1000
−6
0 2 4 6 8 10 12 0
Time [s] 0 50 100 150 200 250 300
Angle [rad]

Fig. 7. Evaluation of the static relation (1 N constant load).

Fig. 8. Transmission contraction (top) and stiffness (bottom) vs motor
angle for different load conditions.
plot in Fig. 8 also shows the transmission stiffness variation
(normalized with respect to the string length) with respect to which is related to the load torque τL according to (13), i.e.
the motor angular position measured during the experiments: τL = h(θ)Fz = KL h(θ)k(θ) (16)
it can be clearly noted how the stiffness of the actuation
system varies with the motor angle and the transmission load. where the nonlinear function h(θ) is defined in (13). Sub-
Note that the non-twisted (motor angle = 0) string stiffness stituting (16) in (14), the dynamic model of the actuation
is quite constant: this value represents the stiffness K of the system can be written as the second-order nonlinear system
string itself and it is also reported in Tab. I. J θ̈ + B θ̇ + KL h(θ)k(θ) = Kt u (17)
With the aim of deriving a suitable control law for track-
C. Dynamic Model of the Transmission System
ing the load force Fz = y (system output), let x =
In this section, a simplified dynamic model to be used for [ x1 x2 ]T = [ θ θ̇ ]T be the system state. The dynamic
the controller design of the transmission system is derived model of the system can then be written as
neglecting the compliance of the string (rigid transmission).
ẋ1 = x2 (18)
The robustness of the developed controller with respect to
the finite actuation stiffness will be tested both in simulation ẋ2 = −a1 h(x1 )k(x1 ) − a2 x2 + bu (19)
and during experiments. y = ck(x1 ), (20)
The classical equations of a current-controlled DC motor
where a1 = KL /J, a2 = B/J, b = Kt /J, c = KL .
are
Note that the model does not depend on the motor resistance
J θ̈ + B θ̇ = Kt u − τL (14) and the back-EMF constant, therefore enhanced robustness
characteristics are expected for the closed-loop system with
where u, Kt , J and B are the input current, the torque respect to a voltage-controlled motor1 .
constant, the inertia and the viscous friction of the DC motor,
respectively. IV. C ONTROL A LGORITHM
Under the assumption of a purely elastic load KL fixed to The control problem is formulated as the output tracking of
the end of the rigid string, the load displacement is related a desired force reference yd (t), which is assumed sufficiently
to the motor angle by (12), and thus the load force is smooth so that ẏd (t) and ÿd (t) are well defined and limited.
This control problem can be solved e.g. by resorting to
Fz = KL k(θ) (15) the control strategy based on a second-order sliding manifold
approach reported in [12] for a general mechanical system.
The resulting linear controller possesses good robustness
Parameter Value Unit
L0 0.2 m properties and is easily implementable in real-time owing
r 0.1 · 10−3 m to its low computational complexity.
n 2
K ∼ 5000 N m−1 A. Controller Definition
J 2.2 · 10−7 kg m2 First of all, the tracking requirement is translated into the
B 10−8 N m s rad−1
Kt 0.0134 N m A−1 design of a suitable manifold where the evolution of the
KL 2000 N m−1 system state has to be confined:
m 0.6 kg
S = (t, x)|(t, x) ∈ R+ × R2 : σ(t, y) = 0, σ̇(t, y) = 0

Bm 100 N s m−1
TABLE I (21)
S YSTEM PARAMETERS . 1 The improved robustness is provided by the feedback loop inside the
power amplifier.
where σ is defined as: 1) Select the ν poles of the boundary layer system, say
δi , then compute the coefficients di of the following
σ(t, y(t)) = yd (t)−y(t)+η(t) = yd (t)−ck(x1 )+η(t) (22)
polynomial as
and the function η(t) is needed to avoid the well-known ν
Y
peaking phenomenon of the high-gain controllers due to a sν + dν−1 sν−1 + . . . + d1 s + d0 = (s − δi )
mismatch of the initial conditions, i.e. i=1

2) Compute the coefficient n2 as in (25) by selecting x̄1 to

 
η(t) = ewt (1− wt)(y(0)− yd (0))+ (ẏ(0)− yd (0))t (23)
ensure uniform stability of the polynomial in (26) using
being w a negative constant. Note that this definition implies a root-locus method with parameter x1 ∈ [x′1 , x′′1 ];
that σ(0, y(0)) = 0, σ̇(0, y(0)) = 0, hence the system state 3) Select the two poles of the reduced order system,
belongs to the manifold at the initial time instant. The second say λ1 and λ2 , then compute the coefficients of the
step is to design a control law according to the following following polynomial as
theorem.
s2 + n̄1 s + n̄0 = (s − λ1 )(s − λ2 )
Theorem 4.1: Consider the nonlinear plant (18)-(20), and
let the control law be defined by the differential equation 4) Compute the numerator of the controller as
ν (ν) ν−1 (ν−1)
ε u +ε dν−1 u + · · · + εd1 u̇ = n2 σ̈ + n1 σ̇ + n0 σ N (s) = n2 (s2 + n̄1 s + n̄0 )
(24)
where ε > 0 is a “small” real constant, and di , nj (i = 5) Choose ε as small as possible, compatible with sensor
1, . . . , ν − 1, j = 0, 1, 2), are real constants to be selected, noise and desired closed-loop bandwidth, and compute
with ν ≥ 2. the denominator of the controller as
Assume that:
D(s) = εν sν + εν−1 dν−1 sν−1 + . . . + εd1 s
1) The constant n2 is such that
C. Remarks
n2 = d0 /(c b h(x̄1 )) (25)
• The sliding manifold control law is practically equiva-
where d0 is a nonnull constant and x̄1 is a suitable lent to a high gain PID with the advantage of ensuring
state value to be selected so that the next condition is robustness against disturbances;
satisfied; • The computational load is very low and the controller
2) The polynomial does not require any feedforward action unless a mis-
match of initial conditions exists;
sν + dν−1 sν−1 + · · · + d1 s + d0 h(x1 )/h(x̄1 ) (26)
• It is an output feedback control algorithm and thus
is strictly Hurwitz ∀x1 ∈ [x′1 , x′′1 ]; state measurement is not required but only the output
3) The polynomial force. This means that no sensors are needed, neither to
measure motor angle nor load position and/or velocity.
n2 s2 + n1 s + n0 Only a load cell is necessary at one end of the twisted
is strictly Hurwitz; string;
4) There exists a real γ < 0 such that • As every high gain strategy, it requires a low-noise sen-
sor signal, even though the high frequency amplification
w≤γ<0 is mitigated by the possibility to have a strictly proper
Then, there exist ε0 > 0, δ > 0, λ < 0, with λ > γ, such controller by selecting ν ≥ 3;
that for any ε ∈ (0, ε0 ], the solution (x(t, ε), u(t, ε)) of (18), • It can be shown that for ε = 0 the control signal is
(20), (24), is such that exactly the same as the classical feedback linearization
algorithm;
|yd (t)−y(t, ε)|+|ẏd (t)− ẏ(t, ε)| ≤ δ +ξeλt , ∀t ∈ [0, +∞) • The poles λ1 , λ2 can be selected by taking into account
that they practically fix the closed-loop bandwidth;
where ξ is a positive constant depending on the plant initial
• The roots δi have to be chosen at low frequencies,
conditions.
otherwise the poles of the controller will be at very
Proof: The proof follows the same steps of the one of
high frequencies due to the small value of ε.
Theorem 1 in [12].
B. Design Procedure V. S IMULATION AND E XPERIMENTAL R ESULTS
From the above theorem, the transfer function of the To verify if the proposed control strategy designed on
controller is the basis of a simplified model is robust enough to model
uncertainties and if it is able to cope with measurement noise,
n2 s2 + n1 s + n0 N (s)
C(s) = ν ν = (27) an intensive simulation and experimental activity has been
ε s +ε ν−1 dν−1 sν−1 + . . . + εd1 s D(s) carried out. The dynamic parameters used in the simulations
Moreover, the same theorem suggests the following design are summarized in Tab. I. Some of these parameters, e.g. the
procedure: torque constant, the motor inertia and the load stiffness, used
 50
x̄1 = x′1 Setpoint
Experiment
x̄1 = x′′1 40
Simulation

Force [N]
30

20

Instability
10
Region
0
15
Experiment
Simulation
10

Tracking Error [N]

5

Fig. 9. Root locus of the polynomial (26) for different values of x̄1 . −5
0 2 4 6 8 10 12 14 16 18
Time [s]

to design the controller, have been assumed to be affected

Fig. 10. Comparison between simulative and experimental results.
by uncertainties ranging from 20% to 50% of their nominal
values to test the performance and the robustness of the simulations and experimental results validates the proposed
control system in a wide range of conditions. model and shows the effectiveness of the control strategy for
The zeroes of the sliding controller C(s) in (27) have the twisted string actuation system.
both been selected to be at −6, so as to obtain a closed-loop Future activities will be devoted to improve the knowledge
bandwidth of about 1 Hz, and the coefficient n2 has been of the twisted string actuation system model, considering in
defined as in (25) with x̄1 = x′′1 . The ν = 3 poles have been particular the transmission stiffness, and to the implemen-
selected at δi = −2. The uniform stability of the boundary tation of a complete actuator for a tendon driven robotic
layer system has been checked by means of the root locus of hand. Some implementation problems, like the definition of
the polynomial (26) for x1 ∈ [x′1 , x′′1 ], see Fig. 9. The control a reliable fixing for the twisted string and the characterization
law is completed by selecting w = −2 and ε = 0.0016 and of a suitable set of strings, will be also investigated. Since the
then computing the denominator as in step 5 of the design efficiency is an important performance index for an actuation
procedure. system, also this paramenter will be evaluated.
Experimental results of the twisted string actuation system
driven by the proposed controller are reported in Fig. 10. R EFERENCES
In these plots, it is possible to distinguish between three [1] S. Jacobsen, E. Iversen, D. Knutti, R. Johnson, and K. Biggers,
different phases. The peaking avoidance action, see (23), “Design of the Utah/Mit dexterous hand,” in Proc. IEEE Int. Conf.
on Robotics and Automation, 1986.
is active when the system starts since the initial reference [2] Shadow Robot Company, “Design of a dextrous hand for advanced
value differs from the initial conditions and the system output clawar applications,” in CLAWAR, 2003, pp. 691–698.
reaches the desired value very smoothly. Then, the response [3] F. Rothling, R. Haschke, J. J. Steil, and H. Ritter, “Platform portable
anthropomorphic grasping with the bielefeld 20-dof shadow and 9-
to a setpoint step variation can be appreciated and, finally, the dof tum hand,” in Intelligent Robots and Systems, 2007. IROS 2007.
system is evaluated with a 20 N amplitude sinusoidal setpoint IEEE/RSJ International Conference on, 2007, pp. 2951–2956.
at 0.5 Hz. The very good fitting between simulations and [4] J. Butterfass, M. Grebenstein, H. Liu, and G. Hirzinger, “DLR-Hand
II: Next generation of a dextrous robot hand,” in Proc. IEEE Int. Conf.
experiments can be seen clearly undo all considered working on Robotics and Automation, 2001.
conditions. [5] C. Lovchik and M. Diftler, “The robonaut hand: a dexterous robot
hand for space,” in Proc. IEEE Int. Conf. on Robotics and Automation,
VI. C ONCLUSIONS 1999.
[6] C. Melchiorri and M. Kaneko, Springer Handbook of Robotics.
The twisted string actuation system is a very simple, Springer, 2008, ch. Robot Hands.
cheap, small and lightweight actuation system, suitable for [7] C. Melchiorri and G. Vassura, “Mechanical and control features of
the UB Hand version II,” in Proc. IEEE/RSJ Int.Conf.on Intelligent
highly integrated robotic devices like robotic hands. In this Robots and Systems, IROS’92, 1992.
paper both theoretical and experimental characterization of [8] L. Biagiotti, F. Lotti, C. Melchiorri, G. Palli, P. Tiezzi, and G. Vassura,
the system have been given and an output feedback controller “Development of UB Hand 3: Early results,” in Proc. IEEE Int. Conf.
on Robotics and Automation, 2005.
for the twisted string actuation system has been proposed. [9] S. Y. Jung, S. K. Kang, M. Lee, and I. Moon, “Design of robotic hand
The system model equations give a useful tool for the with tendon-driven three fingers,” in Proc. of IEEE Int. Conf. Cont.
design of the actuation system starting from the basic re- Aut. and Sys., 2007, pp. 83–86.
[10] G. Berselli, G. Borghesan, M. Brandi, C. Melchiorri, C. Natale,
quirements, like the maximum stroke, velocity, force and G. Palli, S. Pirozzi, and G. Vassura, “Integrated mechatronic design for
dimensions. a new generation of robotic hands,” in Proc. of the IFAC Symposium
Due to the presence of several unmodeled effects, like the on Robot Control, Gifu, Japan, 2009.
[11] “DEXMART Project website,” http://www.dexmart.eu/.
string friction and elasticity, and to the objective difficulties [12] A. Cavallo and C. Natale, “High-order sliding control of mechani-
in having a complete knowledge of all the system parameters, cal systems: theory and experiments,” Control Engineering Practice,
particular attention has been posed in ensuring the robustness vol. 12, pp. 1139–1149, 2004.
of the control system with respect to measurement noises
and parameter uncertainties. Finally, the comparison between

View publication stats