Professional Documents
Culture Documents
x
1i
x
1i1
: 7
Fig. 5a presents
1
an example of the stiffness being calculated dur-
ing one grasp with the nonlinear manipulandum stiffness setting
K
1M
. In this example, the grasping cycle resulted in 15 displacement
intervals (P
1
through P
15
); the red curve represents the linear inter-
polation between points P
i
and P
i+1
. The green signal is the approx-
imated stiffness (right y-axis) for the same period. A rst order
low pass lter was next used to smooth the instantaneous estimate
of the system stiffness:
K
30K
i
s 30
: 8
The time constant of the pole in(8) was determinedby examiningthe
frequency response of the system. Acut off frequency of 30 rad/s was
chosen because it is slightly higher than the bandwidth of the pros-
thetic hand, so it does not appreciably slowor destabilize the dynam-
ics of the system. This also reduces the impact of position and force
sensor noise in (7) to efciently lter the stiffness estimate in (8).
The comparison between the measured stiffness (6) and the cal-
culated stiffness after signal ltering (8) for the nonlinear stiffness
condition K
1M
showed a close correlation between the two signals
(Fig. 5b). The calculated stiffness was obtained via (8) whereas the
measured system stiffness was calculated using (6): data for grasp-
Fig. 3. Control diagram for the hybrid force-position sliding mode controller. The error from an outer force control loop is used to specify the desired position of the inner
sliding mode position controller.
Fig. 4. Control diagram for the adaptive hybrid force-position sliding mode controller. The slope of the sliding manifold is adaptively dependent upon the detected stiffness of
the system which changes when different objects are grasped.
1
For interpretation of color in Figs. 5 and 10, the reader is referred to the web
version of this article.
R. Andrecioli, E.D. Engeberg / Mechatronics 23 (2013) 11711179 1173
ing cycles were collected and the force-displacement curve was
approximated with a high order polynomial ofine, which was
subsequently differentiated with respect to displacement.
It is important to mention that people cannot detect the stiff-
ness of the environment while in a static posture [33]; it is neces-
sary to rst apply a force to cause deformation before ascertaining
the stiffness. The same is true with prosthetic hands: as the
grasped object is squeezed, the approximation of the system stiff-
ness improves. This is clear from Fig. 5b, where the calculated sys-
tem stiffness (8) lags behind the measured system stiffness (6)
slightly because it is rst necessary to displace the grasped object
before a stiffness estimate can be obtained. In this paper, a 1 mm
sliding window was used in which to estimate the system stiffness
to reduce the likelihood that sensor noise or vibrations of the hand
would produce noise in the stiffness estimate.
It is also important to emphasize that the stiffness detection
algorithm (8) provides an estimate of the stiffness of the overall
prosthetic hand-grasped object system, not of the grasped object
itself. However, unlike human hands, prosthetic hands like the Mo-
tion Control Hand have a constant stiffness. This means that vari-
ations in the overall system stiffness are purely dependent upon
the objects that are grasped as well as the slope of the sliding man-
ifold, as will be subsequently discussed.
4.2. Stiffness-dependent sliding manifold slope
The relationship between sliding manifold slope and system
stiffness in (9) will be used for this study. For cohesion of notation,
G
n
will be used to refer to the gains of the system where the sub-
script n distinguishes between the integral (G
I
), derivative (G
D
) and
proportional (G
P
) gains:
G
n
K
G
nmax
; if K > K
max
a
n
K b
n
; if K
min
< K < K
max
G
nmin
; if K < K
min
_
_
: 9
Here, the values for a
n
and b
n
are determined by dening the desired
manifold slope for a stiff (K
max
) and a compliant (K
min
) object and
then solving the set of equations for a
n
and b
n
. G
n
is initially low
and is increased as needed according to the system stiffness. Thus,
it is clear that the adaptive sliding manifold (5) has a slope that is
piece-wise linearly dependent upon the stiffness of the system. The
slope of the manifold saturates at the highest and lowest values cor-
responding to the maximumand minimumstiffnesses of the system.
Hence, the gain terms in (9) are always positive and bounded even
though the system stiffness could be negative (Figs. 1c, and 5).
The resulting process for choosing the upper andlower bounds of
sliding manifold slopes based on the stiffest and most compliant ob-
jects that are to be grasped is an intuitive method that could be clin-
icallyimplemented. The lackof robust andintuitive control methods
has traditionallybeenanimpediment tothe eldof upper limbpros-
thetics, so this technique is clinically relevant and could be readily
implemented on commercially available prosthetic hands.
4.3. Reaching phase dynamics of the adaptive sliding mode controller
In general, the system states (1) will not initially reside on the
sliding manifold (5), which means that the system error, e, will ini-
tially be nonzero. Thus, the initial goal of sliding mode controllers
is to drive the error states to the sliding manifold so that the error
can eventually be eliminated; the same is true of the ASMC. Thus,
the reaching phase of the sliding mode controller occurs as the sys-
tem states are driven to the sliding manifold. To that end, the
ASMC must enforce the constraint
SK
_
SK 6 bjSKj; 10
where b is a positive constant. To that end, the rate of change of S
must be driven to zero:
_
SK G
P
K_ e G
D
Ke G
I
Ke 0: 11
Hence, from (11) and (1), it is clear that
_
SK G
P
K_ e G
D
K x
D
K
I
x
1
B
I
x
2
n
I
E
D
I
_ _
G
I
Ke 0
12
Fig. 5. (a) The nonlinear forcedisplacement curve while grasping K
1M
is shown. Here, the instantaneous estimate of the system stiffness is calculated using (7) during 1 mm
sliding intervals of displacement. The force units of Newtons are on the vertical axis on the left side of this gure while the scale for the stiffness (N/mm) is measured on the
right side. (b) The instantaneous estimate of the system stiffness is ltered (8) and compared to the measured system stiffness from (6). There is a close correlation between
the two signals. However, the estimate of the system stiffness (8) lags behind the measured stiffness (6) because it is impossible to estimate the stiffness of the system in real
time without rst displacing the grasped object.
1174 R. Andrecioli, E.D. Engeberg / Mechatronics 23 (2013) 11711179
Therefore, if the magnitude of the input, E, is designed to be
greater than
C > n
1
D I
G
P
K_ e G
I
Ke
G
D
K
Ix
D
Kx
1
Bx
2
_ _
; 13
the system states can be driven to the sliding manifold. From this
expression, it is clear that C will be the same for the SMC and ASMC
controllers unless an object is in grasp. Once an object is grasped, C
will change with the ASMC due to the stiffness-dependent manifold
slope. However, it is clear that the second derivative of the desired
position, x
D
, in (12) is likely to be very noisy since it would contain
the second derivative of the EMG input from the amputee operator
[24]. Hence, it is good practice for sliding mode control of prosthetic
hands to use an upper bound estimate on C that is based on the sys-
tem model. In practical applications, this is done by equating C to
the maximum possible voltage that the motor of the prosthetic
hand can allow. This assures maximal robustness to externally ap-
plied disturbance torques.
4.4. Sliding phase dynamics of the adaptive sliding mode controller
The sliding phase occurs once the system error states are driven
to the sliding manifold. Once on the sliding manifold, they can be
maintained there and driven to zero through the control law
E CsatG
P
Ke G
I
K
_
e dt G
D
K_ e: 14
The saturation function is again used to reduce the likelihood of
chatter that can frequently occur with mechanical systems that use
the signum function. Since the stiffness-dependent terms within
the sliding manifold are all positive and bounded (9), stability of
this control law can be readily demonstrated in the same manner
as the SMC [32,34].
However, examination of the control law of the ASMC (14) re-
veals interesting system dynamics depending upon the object that
is grasped. For example, the system error states could follow a non-
linear trajectory while on the sliding manifold as they are driven to
zeroif the object ingrasphas a nonlinear or variable stiffness (Fig. 6).
Additionally, because the saturation function is used in the control
law to avoid chatter, the error states are not maintained perfectly
upon the sliding manifold. Therefore, rapid changes in the stiffness
of the system could change the slope of the sliding manifold suf-
ciently so that the error states will have to reenter a reaching phase
once again before being driven to the sliding manifold.
In a practical setting, this could be very useful to improve the
system response over a broad range of grasped objects. With high
stiffness objects, the gains would increase to improve the tracking
and with low stiffness objects, the gain decreases to prevent exces-
sive overshoot and oscillations. Furthermore, if a brittle object like
an egg was grasped and then slightly cracked, the stiffness of the
object would initially be very high and then drop rapidly. In this
scenario, the slope of the manifold would initially be very high
and then decrease rapidly, which would have the effect of reducing
the grip force applied on the object after it cracked. There is a clin-
ical need for an improved capacity to delicately handle different
objects [28] and to adapt to different object parameters as is evi-
denced by several recent surveys of upper limb amputees
[6,9,35], so this is an important design consideration.
5. Experimental methods
Both the SMC (4) and ASMC (14) prosthetic hand controllers
were implemented in Simulink using the realtime Windows target
kernel and a control loop frequency of 1 kHz. Extensive step, ramp,
and sine wave tracking bench top experiments were performed
with a wide range of linear and nonlinear manipulandum stiffness
to ensure that the ASMC functions appropriately in a well regu-
lated environment [30]. This is very important before conducting
experiments with amputees.
To that end, bench top experiments were performed with eight
different stiffnesses in total. Six different manipulandum stiffness
settings were used: K
1
, K
2
, K
3
, K
1M
, K
2M
, K
3M
. Additionally, a steel
bar was grasped and the situation where there was no object in
grasp was also explored. To perform these experiments consis-
tently, the Motion Control Hand was mounted to an aluminum x-
ture and the manipulandum was grasped with both controllers and
the aforementioned six different stiffness settings (Fig. 7). In the
same manner, the steel bar was grasped and the situation without
an object in grasp was also evaluated. The ngertips from a com-
mercially available prosthetic hand cosmetic glove were placed
on the Motion Control Hand to ensure a comparable object interac-
tion situation that is encountered by amputees daily. However, the
other portions of the cosmetic glove were removed to accommo-
date the Hall effect sensor that was mounted to the hand to pro-
vide the necessary position feedback for the sliding mode
controllers and the stiffness detection algorithm.
The constant manifold slope for the SMC was chosen to be the
highest slope possible to enable good tracking and to avoid over-
shoot and oscillations over a broad range of grasping scenarios
ranging from an empty hand to a steel bar. Further increases of
the manifold slope resulted in oscillations and overshoot with
low system stiffness whereas decreasing the slope yielded deterio-
rating steady state error. Thus, the manifold slope used with the
SMC corresponded to the maximum constant slope capable of
being used without oscillations over the full range of grasped
objects explored in this paper. This same slope of the SMC manifold
(with corresponding gains G
nmin
) was also the default slope for the
ASMC until an object was grasped, at which time the slope in-
creased in accordance with the detected system stiffness (9). The
maximum slope for the ASMC with a stiff object was determined
in a similar manner: gains G
P
, G
I
, and G
D
were systematically
increased to the maximum values (G
nmax
) before oscillations and
signicant overshoot occurred while grasping the steel bar, which
corresponded to the maximum system stiffness (K
max
). K
min
was
chosen to be zero. Thus, the minimum and maximum ASMC man-
ifold slopes and corresponding system stiffnesses were specied,
which enabled the calculation of the stiffness-dependent gain
functions listed in (9). The gain on the outer force control loop
(G
F
) and C were maintained the same for both controllers.
Fig. 6. The slope of the manifold of the adaptive sliding mode controller varies
depending upon the stiffness of the object that is grasped. The slope is initially low
and then increased with increasing stiffness. If the grasped object has a variable or
nonlinear stiffness, the error states will follow a nonlinear trajectory to the origin as
the slope of the sliding manifold will change with the different stiffness.
R. Andrecioli, E.D. Engeberg / Mechatronics 23 (2013) 11711179 1175
With each of the eight stiffness settings, both controllers
tracked a desired trajectory comprised of steps, a ramp, and
sinusoids with two different frequencies. Sixteen repetitions were
performed with both controllers at each of the eight stiffness
settings. This number was chosen based on the estimated standard
deviation from prior experiments [30]: a sample size of sixteen
repetitions yielded a power of 0.95 considering a difference of
0.5% or greater in the responses of each controller.
A two-way analysis of variance (ANOVA) was performed on the
controller performance metrics: percent overshoot (%OS) and the
percentage of steady state error (%E
SS
) during the step portions of
the trajectory. The absolute error (%E
ABS
) throughout the duration
of the experiments was also calculated. The two factors used in
the ANOVA are the eight aforementioned levels of stiffness and
the two prosthetic hand controllers used.
6. Results
Illustrative force and position tracking data from the SMC with
no object in grasp showed excellent tracking performance during
the step portions of the trajectory (Fig. 8). The percent overshoot
during the step portions was also minimal. The gear train applied
a torque on the linkage system that caused a small amount of
strain, which is why approximately 1 N of force was detected by
the strain gauges even though there was no object in grasp. The
amplitude of oscillation during the two different frequencies of
sinusoids remained fairly constant. The ASMC also had excellent
position tracking with no object in grasp (Fig. 9). Since the detected
system stiffness was at the minimal case with no object in grasp,
the gains remained nearly constant.
However, when an object was in grasp with the SMC, the ampli-
tude of oscillation during the higher frequency sinusoidal portion
of the input was attenuated (Fig. 10, blue lines). With stiffness
K
3
, the position was affected by the increased force from the
grasped object as one would expect with a hybrid force-position
control law (2). But, the inclusion of force feedback is very impor-
tant with high stiffness objects to prevent excessive force and chat-
ter that would occur with a position controller.
However, when the same object was grasped (K
3
) with the
ASMC, the slope of the manifold increased as the system gains in-
creased because of the stiffness of the manipulandum (Fig. 10, red
lines). The effect of the increased gains during the step inputs with
Fig. 7. The experimental set up where the Motion Control Hand was mounted to an
aluminum base. The manipulandum was placed in grasp for the six different
stiffness settings: K
1
, K
2
, K
3
, K
1M
, K
2M
, and K
3M
. Experiments were also performed in
this manner with no object in grasp as well as with a steel bar. Fingertips from a
commercially available prosthetic hand cosmetic glove were placed on the hand to
simulate a similar frictional situation encountered daily by amputees. However, the
remainder of the glove was removed to facilitate the placement of the Hall effect
position sensor needed for the sliding mode controllers and the stiffness detection
algorithm.
Fig. 8. Illustrative tracking data of the SMC with no object in grasp. This is the data
from all 16 trials averaged together.
Fig. 9. Sample data showing the ASMC with no object in grasp. The feedback gains
and the manifold slope remain nearly constant at the default level.
1176 R. Andrecioli, E.D. Engeberg / Mechatronics 23 (2013) 11711179
the ASMC noticeably improved the steady state error. The
improvement with the ASMC during the sinusoidal portions was
also substantial because the amplitude of oscillation was not sig-
nicantly attenuated during the higher frequency portion as oc-
curred with the SMC. Additionally, the applied force with the
SMC was roughly half the amplitude of the force applied while
using the ASMC during the sinusoidal portion of the trajectory.
Data from the ASMC while grasping K
3
also showed the effect of
the increasing system stiffness on the slope of the manifold: G
P
,
G
I
, and G
D
all increase with the increasing system stiffness (9)
which has the effect of improving the tracking capabilities of the
hand, particularly during the sinusoidal portions of the experi-
ments (Fig. 10). The overshoot with the ASMC during the step por-
tions of the trajectory was also very low.
Data from the nonlinear stiffness condition K
2M
with the ASMC
showed that the system gains rapidly increased and then de-
creased as the hand was displaced through the nonlinear force eld
due to the magnet on the manipulandum (Fig. 11). This had the ef-
fect of improving the tracking of the system when the stiffness was
high without excessive overshoot when the stiffness was low. Note
that only positive stiffness values are displayed, which had no im-
pact on the system gains since they saturated at the minimum
allowable values before the system stiffness became negative in
accordance with (9). The motor control law (E) fully saturated at
the maximum and minimum allowable values according to (14).
Both controllers were satisfactory in regard to the %OS during
the step portions of the trajectory; each controller had less than
1% overshoot at each stiffness setting (Fig. 12a). The two-way
ANOVA showed a statistically signicant difference between the
performance of the two controllers (p < 0.01) and a difference in
regard to the stiffness setting of the manipulandum (p < 0.05).
The interaction between these two factors was also signicant
(p < 0.01). There was a modest improvement of the steady state
error (%E
SS
) during the step portions of the experiments with the
ASMC (Fig. 12b) which was statistically signicant between
the controllers and also the different levels of stiffness (p < 0.01).
The interaction between these factors was also signicant since
there was less differences between the two controllers steady
state errors with low levels of stiffness in comparison to higher
levels of stiffness (Fig. 12b).
With respect to the %E
ABS
, the ASMC had less error with each of
the eight stiffness settings (Fig. 12c). This corroborates the previ-
ous observation that the impact of the ASMC during the sinusoidal
portions of the trajectory was more substantial than during the
steady state portions of the step inputs. The difference between
the two controllers was statistically signicant (p < 0.01) and the
effect of the system stiffness was also signicant (p = 0.01).
Fig. 10. Illustrative tracking data of the SMC with the spring stiffness K
3
(blue
lines). Note that the amplitude of oscillation during the higher frequency sinusoidal
portion of the experiment is attenuated due to the increased system stiffness with
the manipulandum in grasp. Sample data showing the ASMC tracking the desired
trajectory with K
3
is also presented (red lines). The feedback gains vary relative to
the detected system stiffness. As the system stiffness increases, the gains increase to
improve the tracking capabilities of the ASMC relative to the SMC. Note that the
system response with the ASMC is not as signicantly attenuated with the higher
frequency of sinusoidal oscillation, as occurred with the SMC. This is the data from
all 16 trials averaged together.
Fig. 11. Sample data showing the relevant signals from the ASMC while tracking
the desired position trajectory with nonlinear stiffness K
2M
. The abrupt changes in
the force signal in comparison to those shown in Figs. 810 are the result of the
nonlinear stiffness. Because of this, the gains change rapidly to improve the
tracking. The voltage control law (E) fully saturates at the maximum and minimum
permissible values of 8 V.
R. Andrecioli, E.D. Engeberg / Mechatronics 23 (2013) 11711179 1177
However, the interaction between these two factors was not signif-
icant (p > 0.05). Data in Fig. 12 were plotted with respect to the
average detected system stiffness for each of the eight stiffness set-
tings that were experimentally evaluated. The system stiffness of
the steel bar is at the far right of each plot.
7. Discussion
A biomimetic comparison can be made between the stiffness-
dependent adaptive sliding mode controller and the human
hand, which also exhibits stiffness-dependent gains during typi-
cal manipulation tasks [29]. The adaptive sliding mode controller
would be useful when handling different objects in tasks of daily
living. For example, if a stiff object was grasped, the gains would
increase to improve the tracking. If a low stiffness object was
grasped, the gains would decrease to prevent oscillations and
excessive overshoot. Additionally, if a brittle object was inadver-
tently cracked, the stiffness would decrease, triggering a de-
crease in the system control gains. This would have the effect
of lowering the applied force on the object, helping to prevent
additional damage to the object. Improving the grasp force sen-
sitivity for prosthetic hands is also important for amputees [28].
In addition to these benets, the stiffness detection algorithm
presented herein can lead to improved control algorithms for
dexterous articial hands [36,37] based on identication of the
grasped object.
The stiffness detection algorithm (7) that is implemented over a
1 mmsliding window to enable detection of nonlinear systemstiff-
ness conditions requires only two subtraction operations, a quo-
tient, and a low-pass lter. The adaptive system gains are piece-
wise linearly scaled by the detected system stiffness. Thus, the
relationship between the manifold slope and the detected stiffness
is intuitive and could be implemented in a clinical setting with
only a minor increase in computational expense. The existing force
and position sensors used to implement the hybrid force-position
sliding mode control algorithm are the only sensors needed to de-
tect the system stiffness, which yields a practical and economical
implementation in a clinical setting.
In this paper, the ASMC was implemented for a one degree of
freedom prosthetic hand. Extension of this variable slope ASMC
algorithm to a more complex robotic hand such as the Dexterous
Shadow Hand [36], could be readily accomplished with appropri-
ate joint angle feedback and ngertip tactile sensors such as the
BioTac [38]. Knowledge of the forward kinematics and the joint an-
gle feedback could be used to calculate the nger and thumb tip
Cartesian positions and thus the change in end effector position
during manipulation. Tactile sensors in the ngertips could be con-
currently used to detect the change in applied force during manip-
ulation. Fusion of these feedback signals could be used to estimate
the stiffness of the grasped object and subsequently to adapt the
sliding manifold slope as described in this paper. For any type of
grasp (precision, power, three nger, lateral pinch, etc. [39]), the
stiffness between any two contact points between the manipulator
and object could be calculated. As in this paper, the manifold slope
could be increased with increasing system stiffness to reduce
tracking error during manipulation. This could also be done digit
by digit to account for objects that have nonhomogeneous
stiffness.
8. Conclusion
In response to the desire expressed by amputees for prosthetic
hands to better adapt to the parameters of different grasped ob-
jects [5,9], an adaptive sliding mode controller was developed with
a stiffness-dependent variable-slope manifold. This controller
adapts the gains and manifold slope based on the detected stiffness
of the grasped object. Bench top experiments were performed to
compare the adaptive sliding mode controller to a traditional slid-
ing mode controller. Results indicated that both controllers had
satisfactory percent overshoot to the step inputs (less than 1%).
However, the adaptive sliding mode controller exhibited a statisti-
cally signicant reduction in average absolute tracking error over a
broad range of grasped object stiffness.
Acknowledgments
The authors would like to thank Motion Control, Inc. for their
generosity and J. Billock for helpful discussions during these
experiments.
Fig. 12. (a) The average percent overshoot of each controller with each of the eight
system stiffness settings. In each case, the average overshoot is less than 1%. (b) The
average absolute steady state error for each controller with the eight system
stiffness settings calculated during the step portions of the input. The adaptive
sliding mode controller has signicantly less error than the sliding mode controller
with higher levels of stiffness but the error is nearly the same with low levels of
stiffness. (c) The average absolute error for each controller with the eight system
stiffness settings. The adaptive sliding mode controller has signicantly less error
than the sliding mode controller.
1178 R. Andrecioli, E.D. Engeberg / Mechatronics 23 (2013) 11711179
References
[1] Dhillon G, Horch K. Direct neural sensory feedback and control of a prosthetic
arm. IEEE Trans Neural Syst Rehab Eng 2005;13:46872.
[2] Horch K, Meek S, Taylor T, Hutchinson D. Object discrimination with an
articial hand using electrical stimulation of peripheral tactile and
proprioceptive pathways with intrafascicular electrodes. IEEE Trans Neural
Syst Rehab Eng 2011;19:4839.
[3] Rossini P, Micera S, Benvenuto A, Carpaneto J, Cavallo G, Citi L, et al. Double
nerve intraneural interface implant on a human amputee for robotic hand
control. Clin Neurophysiol 2010;121:77783.
[4] Atkins D, Heard D, Donovan W. Epidemiologic overview of individuals with
upper-limb loss and their reported research priorities. J Prosthet Orthot
1996;8:211.
[5] Biddiss E, Beaton D, Chau T. Consumer design priorities for upper limb
prosthetics. Disab Rehab: Assist Technol 2007;2:34657.
[6] Biddiss E, Chau T. Upper limb prosthesis use and abandonment: a survey of the
last 25 years. Prosthet Orthot Int 2007;31:23657.
[7] Davidson J. A survey of the satisfaction of upper limb amputees with their
prostheses, their lifestlyes, and their abilities. J Hand Ther 2002;15:6270.
[8] Fougner A, Stavdahl O, Kyberd P, Losier Y, Parker P. Control of upper limb
prostheses: terminology and proportional myoelectric control a review. IEEE
Trans Neural Syst Rehab Eng 2012:66377. http://dx.doi.org/10.1109/
TNSRE.2012.2196711.
[9] Kyberd P, Wartenberg C, Sandsjo L, Jonsson S, Gow D, Frid J, et al. Survey of
upper-extremity prosthesis users in sweden and the united kingdom. J
Prosthet Orthot 2007;19:5562.
[10] Muzumdar A. Powered upper limb prostheses. Berlin: Springer; 2004.
[11] Engeberg E, Meek S. Adaptive sliding mode control for prosthetic hands to
simultaneously prevent slip and minimize deformation of grasped objects.
IEEE/ASME Trans Mechatron 2013;18:37685. http://dx.doi.org/10.1109/
TMECH.2011.2179061.
[12] Kyberd P, Chappell P. Object slip prevention using a derived force vector.
Mechatronics 1992;2:113.
[13] Mingrino A, Bucci A, Magni R, Dario P. Slippage control in hand prostheses by
sensing grasping forces and sliding motion. IEEE Int Conf Intell Robots Syst
1994;3:18039.
[14] Pasluosta C, Tims H, Chiu A. Slippage sensory feedback and nonlinear force
control system for a low-cost prosthetic hand. Am J Biomed Sci
2009;1:295302.
[15] Puchhammer G. The tactile slip sensor: integration of a miniaturized sensory
device on an myoelectric hand. Orthopadie-Technik Quarterly 2000;I:712.
[16] Connolly C. Prosthetic hands from touch bionics. Ind Robot 2008;35:2903.
[17] Engeberg E, Meek S, Minor M. Hybrid force-velocity sliding mode control of a
prosthetic hand. IEEE Trans Biomed Eng 2008;55:157281. http://dx.doi.org/
10.1109/TBME.2007.914672.
[18] Sears H, Shaperman J. Proportional myoelectric hand control: an evaluation.
Am J Phys Med Rehab 1991;70:208.
[19] Wettels N, Parnandi A, Moon J, Loeb G, Sukhatme G. Grip control using
biomimetic tactile sensing systems. IEEE/ASME Trans Mechatron
2009;14:71823.
[20] Andrecioli R, Engeberg E. Grasped object stiffness detection for adaptive force
control of a prosthetic hand. presented at the proceedings of the 2010 IEEE
BIOROB conference. Tokyo, Japan; 2010. http://dx.doi.org/10.1109/
BIOROB.2010.5626816.
[21] Engeberg E, Meek S. Backstepping and sliding mode control hybridized for a
prosthetic hand. IEEE Trans Neural Syst Rehab Eng 2009;16:709. http://
dx.doi.org/10.1109/TNSRE.2008.2006212.
[22] Scherillo P, Siciliano B, Zollo L, Carrozza M, Guglielmelli M, Dario P. Parallel
force/position control of a novel biomechatronic hand prosthesis. In: IEEE/
ASME international conference on advanced intelligent, mechatronics; 2003,
p. 9205.
[23] Miller L, Stubbleeld K, Lipschutz R, Lock B, Kuiken T. Improved myoelectric
prosthesis control using targeted reinnervation surgery: a case series. IEEE
Trans Neural Syst Rehab Eng 2008;16:4650.
[24] Oskoei M, Hu H. Myoelectric control systems a survey. Biomed Signal Process
Control 2007;2:27594.
[25] Carter R, Crago P, Keith M. Stiffness regulation by reex action in the normal
human hand. J Physiol 1990;64:10518.
[26] Hogan N. Adaptive control of mechanical impedance by coactivation of
antagonist muscles. IEEE Trans Autom Control 1984;29:68190.
[27] Kajikawa S, Abe K. Robot nger module with multidirectional adjustable joint
stiffness. IEEE/ASME Trans Mechatron 2012;17:12835.
[28] Engeberg E, Meek S. Improved grasp force sensitivity for prosthetic hands
through force derivative feedback. IEEE Trans Biomed Eng 2008;55:81721.
http://dx.doi.org/10.1109/TBME.2007.912675.
[29] Mugge W, Schuurmans J, Schouten A, Helm Fvd. Sensory weighting of force
and position feedback in human motor control tasks. J Neurosci
2009;29:547682.
[30] Andrecioli R, Engeberg E. Grasped object stiffness detection for adaptive
pid sliding mode position control of a prosthetic hand. presented at the
IEEE BIOROB conference, Roma, Italy; 2012. http://dx.doi.org/10.1109/
BioRob.2012.6290752.
[31] Sciavicco L, Siciliano B. Modelling and control of robot manipulators. 2nd
ed. Springer; 2000.
[32] Khalil H. Nonlinear systems. 3rd ed. Upper Saddle River: Prentice Hall; 2002.
[33] Chew J, Gandevia S, Fitzpatrick R. Postural control at the human wrist. J Physiol
2008;586:126575.
[34] Slotine J, Li W. Applied nonlinear control. Upper Saddle River: Prentice Hall;
2002.
[35] Kyberd P, Beard D, Davey J, Morrison J. A survey of upper-limb prosthesis users
in oxfordshire. J Prosthet Orthot 1998;10:8591.
[36] Karnati N, Kent B, Engeberg E. Bioinspired sinusoidal nger joint synergies for
a dexterous robotic hand to screw and unscrew objects with different
diameters. IEEE/ASME Trans Mechatron 2013;18:61223. http://dx.doi.org/
10.1109/TMECH.2012.2222907.
[37] Michalec R, Micaelli A. Stiffness modeling for multi-ngered grasping with
rolling contacts. presented at the IEEE-RAS international conference on
humanoid robotics, Nashville, TN, USA; 2010.
[38] Fishel J, Loeb G. Bayesian exploration for intelligent identication of textures.
Front Neurorobot 2012;6:120.
[39] Cutkosky M. On grasp choice, grasp models, and the design of hands for
manufacturing tasks. IEEE Trans Rob Autom 1989;5:26979.
R. Andrecioli, E.D. Engeberg / Mechatronics 23 (2013) 11711179 1179