AbstractThis paper introduces a novel exoskeleton device [618271]

333
Abstract—This paper introduces a novel exoskeleton device
(HAND EXOS) for the rehabilitation of the hand for post-stroke
patients.
The nature of the impaired hand can be summarized in a limited
extension, abduction and adduction leaving the fingers in a flexed
position, so the exoskeleton goal is to train a safe extension motion
from the typical closed position of the impaired hand.
The mechanical design of HAND EXOS offers the possibility to
overcome the exoskeleton limits often related to the general high
level of complexity of the structure, mechanism and actuation. We
describe the mechanical design of the index finger module, the
dynamic model and some preliminary experimental results.
I. INTRODU CTION
TROKE is the leading cause of morbidity and mortality
for both adult men and women in Europe Union countries
and medical and social care consume considerable healthcare
resources [1] in terms of both health care costs (hospital care,
nursing, and home assistance) and indirect costs due to
inactivity that increase the burden both for families and society
[2]. Therefore, in the recent past, potentialities of robot-
mediated therapy have been exploited in order to try to
partially solve such problems.
A study to evaluate the needs of chronic stroke patients was
performed recently [3] and its results show that the most
desired function to recover is the hand ability because of the
need to perform again the Activities of Daily Living (ADL).
The main impairments of an hemiparetic hand are: weakness of
specific muscles, abnormal muscle tone (spasticity), lack of
mobility, abnormal muscular synergies, loss of interjoint
coordination, reduced Range Of Movement (ROM), reduced
finger independency and closed position [4]. In order to
recover such impairments, a useful device for the rehabilitation
of the hand should independently assist the motion of each
finger through dedicated finger exercises, training a safe and
controlled extension of each joint in order to improve their

1This work was partly suppo rted by the EU within the NEUROBOTICS
Integrated Project (The fusion of NEUROscience and roBOTICS, IST-FET
Project #2003 -001917 ).
A. Chiri, F. Giovacchini, S. Roccella, E. Cattin, N. Vitiello, F. Vecchi,
M.C.Carrozza, are with ARTS Lab Scuola SuperioreSant’Anna, Pisa, Italy (e-
mail: {a.chiri, f.giovacchini, s.roccella, e.cattin, n.vitiello, f.vecchi
}@arts.sssup.it, {carrozza}@sssup.it).

Fig. 1. Overview of the HAND EXOS index finger module.
ROM. From this point of view exoskeletons better suite for
execution of the correct rehabilitative motor practice because
of their functional advantages: the human machine interface is
extended to the entire hand so that the trajectories of all the
exoskeleton’s joints are as much as possible coincident to that
of the natural limb in the operational space and in the joint
space allowing an accurate and repeatable finger motion joint
by joint. So we are developing a novel exoskeleton device for
the rehabilitation of the hand, HAND EXOS, with a first focus
on independently practising the 5 fingers in order to return not
only flexibility and coordination but also the ability to perform
more complex movement patterns related to ADL tasks. More
in detail, its design has been conceived in order to enable the
activation of all the degrees of freedom (DOFs) of the human
finger with a natural ROM and to achieve requirements as low
encumbrance, light weight, comfort and good wearability.
This paper is organized as follows. Section II describes the
mechanical design of HAND EXOS and the main features of
the first prototype. The finger dynamic model is then presented
in Section III, whereas some very preliminary experimental
tests are reported in Section IV.
II. METHOD S AND MECHAN ICAL DESIGN
A. Biomechanical modeling
A wearable robotic system is physically coupled with the HAND EXOS: towards an exoskeleton device for the rehabilitation
of the hand
A. Chiri, F. Giovacchini, N. Vitiello, E. Cattin, Student Member, IEEE, S. Roccella, F. Vecchi,
Member, IEEE, M.C. Carrozza, Member, IEEE
S The 2009 IEEE/RSJ International Conference on
Intelligent Robots and Systems
October 11-15, 2009 St. Louis, USA
978-1-4244-3804-4/09/$25.00 ©2009 IEEE 1106

human hand, so an exoskeleton design has to be based on the
human model in terms of biomechanics. To design a wearable
mechanism compliant to the human hand movement is a great
challenge because of the complexity of the hand’s structure.
Each finger allows 4 DOFs: from the distal phalanx there are 1
DOF per DIP (Distal Interphalangeal) and PIP (Proximal
Interphalangeal) joints allowing their flexion/extension and 2
DOFs per MP (Metacarpo-Phalangeal) joint allowing both its
flexion/extension and abduction/abduction. The thumb,
instead, allows 6 DOFs and the oppo sition motion that is
fundamental for human dexterous manipulation. So, in addition
to IP (Inter-Phalangeal) and MP joints that allow the
flexion/extension of the thumb, also the CM (Carpo-
Metacarpal) joint allows the flexion/extension, the
abduction/adduction and the thumb oppo sition motions
simultaneously. One of the main features of HAND EXOS is to
try to enable fully mobility of the hand with a natural ROM
and, for that, the number of DOFs is similar to that of the
natural hand skeleton. Moreover we tried to keep the design
criteria as general as possible in terms of size: average values
of 51mm, 26mm and 25mm have been chosen for the index
finger from the proximal to the distal phalanx, but
HAND EXOS has been designed in order to partially fit over
hands of different sizes through a passive and adjustable
mechanism on the intermediate phalanx (Fig. 1,3).

Fig. 2. HAND EXOS concept.

Fig.3. Finger mechanism, exploded view. B. Finger mechanism
HAND EXOS is characterized by 5-fingers independent
modules (Fig. 2), low encumbrance both on the lateral side of
the fingers and on the upper and lower side of the hand to
allow an easy wearability, light weight, comfort, low inertia,
adjustable size to be adaptable to different hands and an
extrinsic actuation system.
The entire mechanical design of HAND EXOS is patent-
pending [5]. More in detail, the exoskeleton is composed of an
external backing element applicable on the dorsum of the
wearer’s hand, and shell-like elements applicable on each
phalanx and connected each other by translational and
rotational joints (Fig. 3,4).
So, each finger is provided with three active rotational joints
(flexion/extension), one passive rotational joint (abduction
adduction) and one passive translational joint (kinematic
coupling of the human/exoskeleton MP axes).
Six pulleys, two for each joint, are placed on both sides of
HAND EXOS finger module in correspond ence with the
wearer’s rotational joints. Such active joints are used for
flexion/extension of DIP, PIP and MP joints; moreover the MP
joint has been provided with a rotational passive joint
obtained through elastic bushing for the abduction/adduction
(Fig. 3,4). Moreover a passive translational joint acting on the
proximal phalanx provides the needed kinematic compatibility
between human and exoskeleton’s MP rotational axes; as
shown in figure 5, such passive mechanism is fundamental,
indeed, to enable the MP joint to cover its entire ROM with no
constraints. For the same purpose, a compliant orthesic

Fig. 4. Kinematics of a finger module.

Fig.5. Auto-aligning translational joint in extended and flexed configuration. 1107

material that fits the human finger anatomy has been placed
inside each shell in order to ensure the kinematic compatibility
also for PIP and DIP joints. For these two joints, indeed, the
compliance of the inner material has been proved to be enough
to ensure the alignment between the hand and exoskeleton
rotational axes. Furthermore HAND EXOS has been designed
in order to keep the palm area and each fingertip free, in order
to enable the subject to interact with objects and to exploit
tactile feedback. Moreover we are designing a thumb module
to follow thumb oppo sability as it is required in dexterous
object manipulation. Thumb kinematics is particularly complex
because its complete motion can be described through five
rotational axes: IP joint has a flexion-extension axis, whereas
the MP and CM joints have a flexion-extension and an
adduction-abduction axis. More precisely CM joint has a third
degree of freedom that is the axial prono-supination that is not
independent from the flexion-extension and adduction-
abduction angles but all simultaneously operate to obtain the so
called thumb oppo sability [6]. So, in order to simplify such
kinematics, the MP adduction-abduction motion is removed,
whereas the flexion-extension of the IP and MP joints will be
provided. The CM joint oppo sability is achieved through an
additional slider-crank mechanism (Fig. 2) placed on the
dorsum of HAND EXOS (in order to preserve the palm area
free) directly actuated by an on-board DC motor powering the
thumb in order to approach the palm approximately following
the thumb oppo sition motion.
The first HAND EXOS finger module (Fig. 1) has been made
of Aluminium alloy (Ergal) and its weight is 114.9 g. It’s
however important to point out that more than half of such
weight (64.3 g) is concentrated in the proximal slider-crank
mechanism (Fig.1) and such weight will be totally discharged
on the palm module (under fabrication) where the exoskeleton
finger will be fixed. Finally the overall perceived weight on the
index finger is very low (50.6 g).

C. Actuation system
One of the main design goals of HAND EXOS is the
activation of each DOF of the human finger in order to enable
a natural ROM. An underactuated mechanism has been used to
match this requirement with low overall size and light weight.
Such solution, indeed, allows to have lower number of
actuators than DOFs. Another advantage coming from the
underactuation choice is the possibility to passively adapt each
finger to the generic shape of the grasped object
(selfadaptation) because the geometric configuration of each
phalanx is simply determined by the external constraints due to
the particular shape of the object without the necessity to
actively coordinate all the phalanges [7].
More in detail each finger module of HAND EXOS is actuated
through a Bowden cables transmission, so only one DC motor is used to extend the DIP, PIP and MP joints. Such cable
transmission choice is critical especially for its intrinsic friction
losses but it is necessary in order to develop a wearable system
with low inertia and a remote actuation. Each finger is actuated
by a cable running across idle pulleys placed in each finger
joints and fixed to the distal phalanx through a cable stop. The
cable is pulled through a linear slider by a DC motor placed
extrinsically. The flexion of the finger is passively obtained by
means of a set of three (one for each joint) antagonist cables
running across the pulleys placed on the other side of the
finger, connected to three extrinsic linear compression springs
whose elastic torques cause the finger to flex (Fig. 6).

Fig.6. Underactuation with linear springs.
The underactuation solution is not the only possible actuation
strategy: HAND EXOS, indeed, has been designed on purpose
in order to implement different actuation/transmission
solutions, from the independent joint actuation to the
underactuation. So a study [8] that is beyond the scope of this
paper, has been carried out in order to analyze and compare
two different actuation strategies both allowed by
HAND EXOS: independent joint actuation with series of non
linear springs and underactuation with series of linear springs.
Both the strategies have been tested through a dynamic
simulator that we have implemented in LabVIEW®
environment (National Instruments LabVIEW 8.2) including
the modelling of the biomechanics of the human finger, the
mechanics of HAND EXOS finger modules, the mechanics of
the human/exoskeleton interface and the specific
actuation/transmission system. The derived performances for
both the actuation solutions are similar, but for rehabilitation
purposes, the underactuation solution better suite for the low
encumbrance and weight requirements.
So it is the goal of this paper to present a preliminary study
on the underactuation strategy, from the dynamic modelling of
an underactuated HAND EXOS finger module to some
preliminary experimental results.
III. FINGER DYNA MIC MODEL
The development of the dynamic model of the HAND EXOS
finger module allowed the simulation of the extension motion
in the sagittal plane and the optimization of the mechanical 1108

design. More specifically, the dynamic behaviour of a standard
human finger inside the exoskeleton finger module has been
explored through the Lagrange model of a three-links planar
manipulator [9]. The direct dynamics problem has been solved
determining the joints accelerations (q&&) then the velocities (q&)
and positions (q) resulting from the given joint torques (τ ) and
the three external forces, applied to each phalanx, representing
the resistance forces due to the muscular spasticity, once the
initial positions and velocities are known.
In fact spasticity, defined as a heightened velocity-dependent
reflex response to stretch [10], causes a continuou s contraction
of the hand muscles of stroke patients that interferes with the
normal hand posture. It contributes as a resistance to the
extension of the fingers, so we have preliminary considered
such resistant effect as three constant forces applied at the
centre of mass of each phalanx with maximum values (from the
proximal to the distal phalanx): F1=10 N, F2=6 N, F3=3 N, as
suggested by clinicians.
Because of the underactuation solution, the joints torques are
coupled with each other by the same tension T through the
following relations:
TrTrq q lq h lTr
3 32 21 1 1 11 1 1
1) cos() cos()) sin( ) cos( (
==Θ−−Θ=1
τττ

11
1 1)) cos( arcsin(
lh q dq− −+=Θ (1)
where Θ1 (derived from the particular geometry), l1 (0.029
m), h (0.0124 m), d (0.0139 m) and q1 are reported in Fig. 7;
whereas ri (i=1;3 from MP to DIP joint) is the pulley radius and
T the cable tension whose variation respect to time has been
assumed to be of the fifth order (Fig. 8) with an initial value of
63.11 N and a final value of 147.95 N (evaluated through the
static equilibrium of the distal phalanx with an initial position
q3=1.2 rad and a final position of 0 rad).
The equations of motion of the finger module (considering the
effect of gravity and friction) can be written in a compact
matrix form which represents the joint-space dynamic model
as:

(2)
where:
• q,q&,q& & are the (3×1) joint position, velocity and
acceleration vectors, respectively;
• B(q) is the (3×3) joint inertia matrix;
• C(q,q&) is the (3×3) matrix of centrifugal and
Coriolis torques;
• Fv is the (3×3) matrix of viscous friction coefficients;

Fig.7. HAND EXOS finger scheme.
0 2 4 6 8 1060708090100110120130140150
Time [sec]Cable Tension [N]

Fig. 8. Cable tension.
• g is the (3×1) gravity vector;
• τ is the (3×1) vector of the actuation torques;
• K is the (3×1) vector of spring stiffness coefficients;
• r is the (3×1) vector of the pulley radii;
• q0 is the (3×1) vector of the spring rest positions;
• Ji is the (6×3) matrix of geometric Jacobian evaluated
in the resistant force application points;
• Hi is the (6×1) vector of forces and moments exerted
by the resistant forces on each link.
In equation 2, the contribution of spasticity is considered
through JT
i(q)Hi(q) derived from the virtual work principle
[9] that allows the determination of the relationship between
the generalized forces applied to the joints and the generalized
forces applied to the links.
Simulation analysis has been carried out to iteratively optimize
the mechanical design in order to best fit the behaviour of the
human finger with the desired trajectories deriving from an
healthy hand extension motion. So several simulation trials
with different mechanical parameters have been tested in order
to iteratively define an accurate set of parameters for the
prototype, finally resulted in the following values: q3 ∈ [0, 1.2]
rad is the range of variation of the distal joint; K = [9370 9270
13960 ]T N/m are the spring stiffness coefficients whose values
have been chosen from the catalogue in order to be close to the
values calculated with the simulation; r1=9×10-3m, r2=6×10-3m,
r3=5×10-3m are the pulley radii and q0 = [3.6 2 2]T rad are the
spring rest position. Preliminary simulations results for a slow
(10 seconds) extension task, together with the required motor
torque are below. () () () ( )
( ) ( ) ( ) ( ) ( ) ( ) qHqJqHqJqHqJq q Kr qgqFqqqCqqB
T T Tv
3 3 2 2 1 102,
+ ++− +−= + − + τ &&&& &1109

0 0.2 0.4 0.6 0.8 1-0.500.511.522.533.54
Time [sec]Joint Angular Position [rad]

MCP Trajectory
PIP Trajectory
DIP Trajectory
0 2 4 6 8 100.150.20.250.3
Time [sec]Motor Torque [Nm]
0 0.2 0.4 0.6 0.8 1-0.500.511.522.533.54
Time [sec]Joint Angular Position [rad]

MCP Trajectory
PIP Trajectory
DIP Trajectory
0 2 4 6 8 100.150.20.250.3
Time [sec]Motor Torque [Nm]
/c30/c33
0 0.2 0.4 0.6 0.8 1-0.500.511.522.533.54
Time [sec]Joint Angular Position [rad]

MCP Trajectory
PIP Trajectory
DIP Trajectory
0 2 4 6 8 100.150.20.250.3
Time [sec]Motor Torque [Nm]
0 0.2 0.4 0.6 0.8 1-0.500.511.522.533.54
Time [sec]Joint Angular Position [rad]

MCP Trajectory
PIP Trajectory
DIP Trajectory
0 2 4 6 8 100.150.20.250.3
Time [sec]Motor Torque [Nm]
/c30/c33MP
0 0.2 0.4 0.6 0.8 1-0.500.511.522.533.54
Time [sec]Joint Angular Position [rad]

MCP Trajectory
PIP Trajectory
DIP Trajectory
0 2 4 6 8 100.150.20.250.3
Time [sec]Motor Torque [Nm]
0 0.2 0.4 0.6 0.8 1-0.500.511.522.533.54
Time [sec]Joint Angular Position [rad]

MCP Trajectory
PIP Trajectory
DIP Trajectory
0 2 4 6 8 100.150.20.250.3
Time [sec]Motor Torque [Nm]
/c30/c33
0 0.2 0.4 0.6 0.8 1-0.500.511.522.533.54
Time [sec]Joint Angular Position [rad]

MCP Trajectory
PIP Trajectory
DIP Trajectory
0 2 4 6 8 100.150.20.250.3
Time [sec]Motor Torque [Nm]
0 0.2 0.4 0.6 0.8 1-0.500.511.522.533.54
Time [sec]Joint Angular Position [rad]

MCP Trajectory
PIP Trajectory
DIP Trajectory
0 2 4 6 8 100.150.20.250.3
Time [sec]Motor Torque [Nm]
/c30/c33MP

Fig. 9. Joints trajectories and motor torque.
IV. PRELIMINARY EXPERIMENTAL RESULTS
In this section we are going to describe the very preliminary
experiments that we have performed with the first prototype of
the exoskeleton index finger in order to firstly verify its
wearability and kinematic coupling with the wearer’s hand and
secondly to test the underactuation solution in terms of enabled
ROM.
Such preliminary experiments have been carried out
respectively in passive (the wearer actuates the passive
exoskeleton) and active modality while recording joints
trajectories. Due to the absence of sensors on the joints of the
first prototype, joint trajectories have been recorded by means
of the OPTOTRAK Certus system (Fig.10) which is an
infrared optical device for movement analysis.
Firstly, the joints trajectories of the index finger (without
wearing HAND EXOS) have been recorded from an healthy
subject while performing a natural extension task from a flexed
to an extended position. Such trajectories represent both the
reference for the evaluation of the performances of the device
and the ideal ROM for the rehabilitative practice. Four active
infrared miniaturized markers have been then placed on MP,
PIP, DIP joints and on the end of the distal phalanx (Fig. 10) in
order to record the angular position of each phalanx as
shown in figure 11. Moreover other three active markers have
been placed on a suppo rting base in order to refer the joint
motion to a unique reference frame (Fig. 10,11). The reference
joint angles over time have been then calculated from the
acquired marker coordinates with an acquisition rate of 30 Hz
as shown in figure 12.
Then, the same markers have been placed on the exoskeleton
index module in correspond ence with the MP, PIP and DIP
rotational axes of the hand and on the end of the distal orthotic
shell as shown in figure 13. The reference frame has been then
placed on a preliminary mechanical suppo rt to maintain
HAND EXOS fixed. Exploiting such set-up, the first
experiment has been performed in order to evaluate the
HAND EXOS wearability: an healthy subject was asked to
perform a 10 seconds natural extension motion from a closed
position of the hand. The acquired joints trajectories
(acquisition rate of 30 Hz), are reported in figure 14. As we
can see, the ROM enabled for each joint by the exoskeleton is
approximately the same with the reference (Fig. 12); this
means that HAND EXOS ensures the right kinematic compatibility with the wearer’s finger. The slight observable
differences between the trajectories reported in figure 12 and

Fig. 10. OPTOTRAK Certus system and the experimental set-up.

Fig. 11. Schematic drawing of the phalanges, joint angles, markers and
reference frame.
0 1 2 3 4 5 6 7 8 9 1000.511.522.533.54
Time [sec]Joint Angular Position [rad]

MP Trajectory
PIP Trajectory
DIP Trajectory

Fig. 12. OPTOTRAK recordings during an extension motion of an healthy
hand without wearing HAND EXOS.

Fig. 13. Experimental set-up for exoskeleton joints trajectories
recordings. 1110

0 1 2 3 4 5 6 7 8 9 1000.511.522.533.544.5
Time [sec]Joint Angular Position [rad]

MP Trajectory
PIP Trajectory
DIP Trajectory

Fig. 14. OPTOTRAK recordings during an extension motion of the
HAND EXOS index module in passive modality.

14, very likely depend on the natural human hand variability in
performing non externally controlled motor tasks as well as on
the absence of the HAND EXOS palm module (Fig. 13),
currently under fabrication, that will allow the human MP
rotational axis to be rightly aligned with the exoskeleton one.
Then the same experimental set-up has been exploited also to
perform the second experiment in order to test the
underactuation solution: an healthy subject was asked to be
completely passive allowing HAND EXOS to extend his finger.
In this preliminary test one DC motor (Faulhaber Minimotor
1727 U006C) activated the MP, PIP and DIP joints through an
extensor cable fixed on the distal orthotic shell of the
exoskeleton, while flexion was not provided by the device
because no flexion cables and springs were included in this
preliminary experimental set-up (Fig. 13). This is the main
reason why such recorded data can not be compared with those
ones obtained through the dynamic model presented in the
previous section. However, as we can notice in figure 15, the
ROM is very similar to the previous experiments but, as a
consequence of the underactuation solution and the absence of
the fixed palm suppo rt to which the hand can be constrained,
the proximal phalanx remained in the same position during the
first part of the task, while the proximal and distal phalanges
first initiate the motion. Such result, however, has no
consequences in terms of wrong or uncomfortable motion.
Then we can conclude that underactuation can be a good
solution both for low encumbrance requirement and to enable
the desired ROM. However the first finger module needs to be
tested together with the palm suppo rt in order to properly
evaluate its performances.
V. CONCLUSION
This paper presented a preliminary study on a finger module of
a novel exoskeleton device for the rehabilitation of the hand.
Because of its design, HAND EXOS will allow the independent
actuation of all 5 fingers, low overall size, light weight and a
proper kinematic coupling with the human fingers. Moreover 0 1 2 3 4 5 6 7 8 9 1000.511.522.533.544.5
Time [sec]Joint Angular Position [rad]

MP Trajectory
PIP Trajectory
DIP Trajectory

Fig. 15. OPTOTRAK recordings during an extension motion of the
HAND EXOS index module in active modality.
HAND EXOS preserves the palm area and each fingertip free
so that the patient can directly interact with ADL objects while
exploiting tactile feedback.
Next planned work counts to exploit the device as an interface
for biomechanical assessment of a post stroke hand, with the
final aim to develop a proper model of spasticity to be used to
refine and test the dynamic model presented in Section III.
VI. REFERENCES
[1] D. Epstein, A. Mason, A. Manca, “The hospital costs of care for stroke
in nine European countries“, Health Economics, vol. 17, pp S21-S31,
2008 .
[2] R. Colombo, F. Pisano, S. Micera at al. “Robotic techniques for upper
limb evaluation and rehabilitation of stroke patients”, IEEE Trans. on
neural systems and rehabilitation engineering, vol.13, no.3, September
2005 .
[3] L. Dovat, O. Lambercy, E. Burdet at al., “A haptic knob for
rehabilitation of stroke patients”, International conference on intelligent
robots and systems, October 2006 , China.
[4] O. Lambercy, L. Dovat, E. Burdet et al., “Development of a robot-
assisted rehabilitation therapy to train hand function for activities of daily
living”, Proc. 2007 IEEE International conference on rehabilitation
robotics, Noordwijk, The Netherlands.
[5] International patent: “Wearable mechatronic device”,
PCT/IB2008 /001990 – Italian patent: ‘Ortesi meccatronica per la mano’,
PI2007 A000088 .
[6] I. A. Kapandji, The Physiology of the Joints: Upper Limb, vol. 1, 5th
ed. New York: Elsevier, 1986 .
[7] B. Massa, S. Roccella, M.C. Carrozza, P. Dario, “Design and
development of an underactuated prosthetic hand ”, Proc. 2002 IEEE
International conference on robotics & automation, Washington , DC, pp.
.3374 -3379 .
[8] A. Chiri , F. Giovacchini , S. Roccella , N. Vitiello, E. Cattin , F.
Vecchi, M.C. Carrozza, “Handexos: towards a suppo rt device for hand
activities and telepresence”, 10th ESA Workshop on Advanced Space
Technologies for Robotics and Automation, November 11-13 2008 ,
ESTEC, Noordwijk, The Netherlands.
[9] L. Sciavicco, B. Siciliano, “Modeling and control of robot
manipulator”, 2nd Edition, Springer-Verlag, London , UK, 2000 .
[10] D.G. Kamper, W.. Zen Rymer, “Quantitative features of the stretch
response of extrinsic finger muscles in hemiparetic stroke”, Muscle &
Nerve, June 2000 , pp. 954-961. 1111