biomechaytronic
TRANSCRIPT
-
8/2/2019 biomechaytronic
1/6
AbstractThe development of a biomechatronic hand with
multi-FSRs (Force Sensitive Resistor) interface is presented.
The cambered palm is specially designed to enhance the stabilitywhile grasping. The opposite thumb could grasp along a cone
surface, while maintaining its function. By taken the advantage
of coupling linkage mechanism, each finger with three
phalanges could fulfill flexion and extension independently.
Besides, each finger is equipped with torque and position
sensors. Thus, the cosmetics and dexterity are improved
remarkably compared to commercial prosthesis. The hardware
architecture is hierarchically organized as control system and
human-machine interaction system. To evaluate the
performances of the hand, the pattern reorganization
experiment based on support vector machine (SVM) was
conducted. The results show that FSR sensing technology is
suitable for muscle activity detection; Compared to nearest
neighborhood classifier, SVM method improves the predictsuccess rate. Moreover, it also demonstrates that the hand is
dexterous and manipulates like human.
I. INTRODUCTIONROSTHETIC hand as a branch of humanoid hand
research is widely addressed in the field of rehabilitation
technologies. Commercial prostheses usually have simple
structure and rarely no sensory system. These limitations
deeply affect the functionality of hand. Due to the drawbacks
of commercial prostheses, many experimental prostheses
with more degree of freedoms (DOFs) and sensors have been
developed, such as FZK Hand [1], RTRII Hand [2], IOWA
hand [3], and Smarthand [4]. But these hands are still faraway from being accepted as part of the body by the
amputees.
Despite of several research effort aimed at innovating
humanoid hand technology, surveys on the relevant
functionality of the hand prosthesis with regard to daily use
activities show that the main concerns of the amputees are
aesthetics (62%), excessive weight (58%), lack of functional
capabilities (50%); further more, the opinions from
rehabilitation professionals reveal that tactile feedback is
preferred largely [5].
Due to the current status, it is particularly urgent to develop
a highly integrated prosthetic device that meets therequirements listed above. Meanwhile, there is an increased
Manuscript received April 5, 2010. This work was supported by the
National Natural Science Foundation of China under Grant 50435040, and
the National High Technology Research and Development Program of China
under Grant 2008AA04Z203, 2009AA043803.
Xinqing Wang, Jingdong Zhao, Dapeng Yang, Nan Li, and Chao Sun are
with State Key Laboratory of Robotics and System, Harbin Institute of
Technology, Harbin 150001, China. (e-mail: [email protected])
Hong Liu is with the Institute of Robotics and Mechatronics, German
Aerospace Center, Munich 82230, Germany. (e-mail: Hong. [email protected])
demand from intelligent control; traditional sEMG control
methods can acquire a higher success rate over prosthetic
hand control [6], [7]. However the surface electrodes requirevery good contact with human skin to keep D/C voltage
potential low [8], and the price of the electrodes is expensive.
Considering that human activities such as grasping can be
detected by contractions of forearm muscles, we propose to
use simple, low cost FSR sensors to capture the pressure of
muscles as source of control, the intention of human could be
realized through pattern reorganization method [9]. In order
to improve prehensile posture and success rate, advanced
machine learning algorithm is required, such as SVM which
is a useful technique for data classification.
The present work of this project focuses on the
biomechatronic approach to develop a human-like prosthesis.The designed hand is a light weight and simple constructed
multi-DOF hand and can accomplish motions such as nip and
grasp. The proprioceptive sensory system and control unit are
embedded in mechanical structure. Besides, the hand can be
controlled by pattern reorganization method on the basis of
multi-FSR sensors in the real-time. Its function and
appearance are very close to the human hand.
II. MECHANICAL STRUCTUREThe comparison between the prosthetic hand and human
hand is shown in Fig. 1.
The hand is 79mm wide, 159mm long and 21mm thick
(when in full outspread), weighs about 420g including circuit
boards. It is about 90% to a human hand out of consideration
for inconspicuous. The hand is composed of five active
fingers. Each finger is actuated by a DC motor independently.
By taking the advantage of coupling mechanism, each fingeris capable of rotate around the metacarpophalangeal (MCP),
proximal interphalangeal (PIP), and distal interphalangeal
(DIP) joints for a total of 15 movable joints of the entire hand.
The specifications of each finger are shown in Table I.
Compared with those of fixed shape fingers; this kind of
design enhances the functionality of the hand and mimics the
motion of human finger.
The sphere grasp range is 50-90mm, the cylinder grasp
range is 15-90mm, and these were designed to fulfill daily
Biomechatronic Approach to a Multi-Fingered Hand Prosthesis
Xinqing Wang, Jingdong Zhao, Dapeng Yang, Nan Li, Chao Sun, Hong Liu
P
Fig. 1. Comparison between the prosthetic hand and human hand
Proceedings of the 2010 3rd IEEE RAS & EMBSInternational Conference on Biomedical Robotics and Biomechatronics,The University of Tokyo, Tokyo, Japan, September 26-29, 2010
978-1-4244-7709-8/10/$26.00 2010 IEEE 209
-
8/2/2019 biomechaytronic
2/6
manipulation requirements. The hand could also complete
other grasp modes, such as precision grasp and hook
prehension. The research ideology is described in detail in the
following sections.
A. Configuration of PalmAccording to traditional anatomical definitions, human
palm is comprised of three arches providing the necessary
stability and mobility in the hand [10], as shown in Fig. 2.
The cambered palm is designed which is far different from
that of plain ones. The placement of each finger is shown in
Fig. 3. The MCP joint of middle finger is the reference both in
vertical and horizontal position. The index, the ring, and the
little finger have rotary angles of 3, 3, and 6 around the
axis of middle finger to realize the distal transverse arch. Thesimulation experiments show that there is better envelope
space because of the arches.
The opposability of the thumb is very important for
dexterity and grasping stability. The oppose angle between
the thumb and the middle finger is defined as . It is designed
to ensure the minimum distance larger than 95mm. By means
of the anti-trigonometric function, the value of can be
calculated as 69.
A performance index [11] of the thumb opposability is
defined by (1)
31
1 k
i iiJ w vd =
=
(1)
Where iv denotes the volume of intersection between themobility space of the thumb and that of the i-th finger, each
iv is the function of , and can be obtained by using theMechanica module of Pro-E. kdenotes the number of the
finger except the thumb, dis the length of the thumb, andiw is a weighting coefficient, simply, each iw equals to 1.00.
The biggerJ becomes, the stronger the hand performs.
According to the simulation results, the value ofJ reaches
maximum, when is 28.6.
B. Multi-Sensory Integrated Design of FingersCommercial prosthetic hands usually use fixed shape
fingers, which may bring two limitations: Firstly, the motion
of the finger is unnatural; secondly, the sensory information is
poor. In order to overcome these disadvantages, the
physiology of human finger needs to be studied.
The multi-functionality of the prosthetic hand depends on
the actuators performance. We adopted a DC motor to drive
the finger because it is more balanced in the efficiency,
response time, and reliability. The actuation unit consists of a
DC motor, a planetary gearheads and a magnetic encoder.
A tendon drive mechanism was first being considered to
actuate the finger, but it was abandoned due to small grasping
forces. Thus, the coupling linkage mechanism is applied to
transmit the power. The linkage mechanism has 9 bars, as
shown in Fig. 4.
Where 1l is the driving bar, 2l is the seat, 3l is the drivenbar, 4l is the transmission bar. 4l is driven by 1l through
3l during grasp. Thus a 4-bar loop ADBC has beenestablished. The finger mechanism consists of two groups of
planar four bar linkage mechanism; D and G are the
intersection points.
Compared to the underactuated linkage mechanism [12], it
does not contain elastic parts which are a consumption of
battery energy. The motion is transmitted from the motor unit
to the coupling linkage mechanism by means of bevel gear. In
Fig. 2. Three arches of human hand
Fig. 4. Schematic structure of index finger
Fig. 3. Structure of the hand
Parameters Thumb Index finger
Middle finger
Ring finger
Little finger
Operating angle [-5 40] [0 87] [0 87]
Angular velocity 68 (/s) 68 (/s) 118 (/s)Output power 1.5 (W) 1.5 (W) 0.75 (W)
Fingertip force 10 (N) 10 (N) 4.3 (N)
TABLEI
SPECIFICATIONS OF EACH FINGER
210
-
8/2/2019 biomechaytronic
3/6
order to better imitate the shape of each finger in relax, an
initial inflexed angel of 15 was adopted between proximal
phalanx and medial phalanx.
The hand needs as a minimum a set of force and position
sensors to enable intelligent control schemes. The
configuration of sensors and disposal circuits are shown in
Fig. 5. As the finger is compact, the signal transfers among
the disposal boards through flex-PCB to avoid enlace.
The motor encoder is applied to measure the relative angle
of the motor. A contact-less joint position sensor is applied to
measure the position of MCP joint. It can provide more
accurate information of joint position to decrease the
influence of elasticity and hysteresis of the transmission
system. Its measurement principle can be seen in [13]. The
torque sensor located in base joint is embedded in the driving
bar and can precisely measure the external torque. The
measurement principle is based on the induced mechanical
strains on elastomers. Shapes and dimensions of elastomers
are determined by utilizing ANSYS FEM software. The
analog signals were converted to digital by Wheatstone
bridge; the processing circuit is shown in Fig. 6.
SG1 and SG2 are strain gauges; the signals obtained are
transformed into two channel voltage outputs though the half
bridge circuit. The difference of the outputs indicates the
strain of the elastomers. As the amplitude is small, the
precision amplifieris adopted to amplify the signal; the gain(G) of the amplifier is decided by (2). The output of the
amplifier is then sent to the A/D converter. The sampling time
of A/D is 500Hz.
5
6 7
2
(1/ 1 )/
RG
R R
+= (2)
C. Characteristics of ThumbThe humans thumb usually inclines to the palm when
relaxed and grasp along a cone surface, which is far different
from that of most previous prosthetic hand [14], [15].
Based on the position of thumb mentioned above, the
geometry is specially designed to mimic the motion of human
thumb when grasp. The metacarpal of the thumb is placed at
30 deflexion to the base joint of the middle finger. In
addition, there is an initial abduction of 30 to the palm, as
shown in Fig. 3. By doing so, the thumb of the hand can grasp
along a cone surface which is similar to human hand.
In order to realize such an effect, an RSRRR spatial linkage
mechanism was designed, as shown in Fig. 7. It evolves from
the RSSR mechanism. The RSSR mechanism has two DOFs
which means motion uncertainty with only one actuator. Thus,
one spherical joint is replaced by two revolute joints to
eliminate one extra DOF. By doing so, the planar finger
motion is extended to three dimensional spaces and only one
motor is needed to drive the thumb. The last two joints are
coupled as other four fingers. The thumb, 85mm long, 17mm
wide and 14mm thick, weighs about 80g.
D. Packaging of the HandThe skin-like glove fulfils the patients emotional needs
and provides protection from dust and corrosion. Besides, this
silicone polymer gives the necessary friction which is useful
in sliding contact [16].
Two simple gloves with different thickness have been
manufactured by utilizing fast prototype technology.
According to our experiment, the wrinkle is unnatural and
doesnt fit the hand well. Besides, there is interference
between the hand and the glove, as shown in Fig. 8. Due to
the impenetrable of the finger, there will be resisting force
which may lead to unstable grasp and cost additional energy.
Basically, with the 2mm thick silicone polymer glove, there isup to 20% motor energy wasted due to the tensile force of the
glove. So it is important to optimize the shape and dimension
of the glove.
We propose to make the wrinkle as a camber. Because of
the non-uniformity and the anisotropy of the material, it is
hard to calculate the radius of the camber. According to
bending theory, the position of neutrosphere is moving
according to the curly angle, the radius of the neutrosphere
is defined as (3)
Fig. 6. Torque sensor disposal circuit
Fig. 5. Sensory configurations of the index finger
Fig. 7. Simple view of the thumb
211
-
8/2/2019 biomechaytronic
4/6
r x = + (3)Where ris the radius of the wrinkle;x is the coefficient, and is
decided by rand , is the thickness of the glove. Thus the
length of the neutrosphere abl is given by (4).
180/abl = (4)Thus the shape and dimension of the camber is decided, as
shown in Fig. 8.
Based on the idea mentioned above, the improved glove
was made, as shown in Fig. 9. Compared with the previous
ones, the resistance has a decrease of 8%. Besides, an
additional advantage is that the noise of the hand could be
lowered by the glove to a certain extent.
III. HARDWARE ARCHITECTUREThe currently developed prosthetic hand is controlled by
multi-processor controller based on digital signal processor
(DSP). The architecture of the control system is shown in Fig.
10. The control system of the prosthetic hand is hierarchically
organized: the human-machine interface allows the prosthesis
to communicate with human in an alternate way. Namely:
record muscle contraction signals through multi-FSR sensors
and realize contact force feedback to human through
electrical stimulator; a high-level controller is in charge of
implementing FSR signal processing and pattern recognition.
The lower-level controller is designed to coordinate the
movements of the fingers and accomplish grasp.
A. Motion Control BoardThe motion control board is embedded in the prosthetic
hand, and its main function is to actuate and control finger
motors, acquire and process the signals from torque &
position sensors and motor encoders. The core of motor
control is realized by TMS320F2810. It has 6 channel
independent pulse width modulation (PWM) outputs. In
addition, it has 16-channel 12 bit A/D converters to realize
signal sampling from sensors and motor encoders. The
direction of motor is discriminated through complex
programming logic device (CPLD). The controller area
network (CAN) module is applied to communicate with FSR
signal processing board. The control cycle can reach to
800 s without any code optimization.
B. FSR Signal Processing BoardThe hand is controlled by muscle contraction signals in
real-time. In order to realize the real-time control scheme,
TMS320F2812 of TI is used in the signal processing board. It
is a new 32 bit DSP executing 150 MIPS (million instructions
per second). The muscle contraction signals obtained by
multi-FSR sensors are transferred through serial peripheral
interface (SPI) to the signal processing board. The signals are
then classified by SVM in real-time. The outputs of the
classifier then send to the motion control board as command.
IV. PERFORMANCE OF THE HANDTo evaluate the performances of the control system and
finger mechanism, the pattern recognition control was
conducted. Pattern reorganization is the most commonly used
bionic control method. Research on human han
ds grasp function [17] indicated that, the thumb, index and
middle finger play a relatively important role than the othersin daily life. For the convenience of classification, we
configure the thumb, the index and the rest fingers into one
DOF each. The state of finger was classified at first, that is,
1 means the finger motion to the extend position, -1
means the motion to flex position and 0 means motion to
middle position. Every hand gesture corresponds to one mode.
We take 9 basic modes for pattern reorganization, as shown in
Fig. 11.
Fig. 10. Hardware architecture of control system
Fig. 8. Wrinkle of the finger
Fig. 9. Simple view of the artificial skin
212
-
8/2/2019 biomechaytronic
5/6
A. Experiment SetupControl signals are recorded from 5 muscles of the
operators forearm through 10 FSR-149NS [18] sensors. The
muscles are musculus flexor carpi ulnaris, extensor carpi
ulnaris muscle, extensor digitorum muscle, flexible flexor
carpi, musculus palmaris longus. The force sensitivity of
FSR-149NS is optimized for use in human touch control ofelectronic devices. Typically, the single part repeatability
tolerance held during measuring ranges from 2% to 5% of
an established nominal resistance. The number and placement
of the FSR sensors seriously affects the recognition of gesture
patterns. So, the quantity and array of FSR sensors are
optimized combining correlation coefficient analysis and
principal component analysis.
Fig. 12(a) shows the displacement of the sensors before
optimization. The sensors with a crossing in the shape of an X
are disposed according to the optimization result, as shown in
Fig. 12(b).
The FSR signals were classified by SVM. SVM algorithms
separate the training data in feature space by a hyperplane
defined by the type of kernel function used. Thus, the
selection of an appropriate kernel function plays an important
role. Generally, the kernel function is chosen if it fitsparticular data better than any other kernel. The Radial Basis
Function (RBF) kernel is chosen at first [19], maybe it is not
the best one. The kernel function is given by (5)2
( , ) exp( )i j i jK x x x x= (5)
Where is the kernel parameter, 0 >
By taken the advantage of Lagrange optimal method, the
optimal hyperplane can be transformed into dual problem.
Hence, the indicator function is now given by (6)
1
( ) ( , )l
i i i
i
f x sgn y K x x b=
= +
(6)
Where i are Lagrange multipliers, the upper bound is C, forthose i >0 are called support vectors; , i arevectors,
iy are labels, might be -1 or 1; b is the thresholdWe use one-against-one method to solve multi-modes'
classification. For the regression operation, we useepsilon-SVR method. For the optimization of the penalty
parameterCand kernel function coif gamma, we will make a
grid research combined with cross validations. The
experiment devices are shown in Fig. 13.
We take 50 samples for every mode. The sampling time of
PCI card is 8s; the procedure is relax-contract-hold-relax by
following the instruction of the DSP board indicator. The
hold condition must take up 50% of the sampling time or
more. After the training progress, every mode must repeat 20
times to calculate the success rate.
B. Experiment ResultsCompared to nearest neighborhood classifier, SVM
method improves the predict success rate, the recognition of 9
gesture patterns can be achieved at 75%, as shown in Fig. 14.
The decision frequency of current hand gesture can reach
to 40Hz in 9 modes, which is adequate for real time control of
the hand. The success rate is a bit lower than traditional
sEMG control method; this may be improved by carefully
choosing the kernel function of SVM, or by taking the
quantum behaved particle swarm optimization (QPSO)
algorithm.
The multi-FSR control scheme can be adopted in the
control of multi-DOF prosthetic hands and other life-like
artificial appendages, such as limbs, legs, and foot.
Fig. 14. Pattern recognition experiment result
Fig. 11. Hand gesture
Fig. 13. Experiment devices
Fig. 12. Before and after the optimization of sensors
213
-
8/2/2019 biomechaytronic
6/6
Fig. 15 displays that the prosthetic hand performs pinch,
three jaw chawk, cylinder grasp, etc. respectively. It shows
that the hand is able to grasping objects stably.
V. CONCLUSIONS AND FUTURE WORKThe concept of the novel anthropomorphic prosthetic hand
and an intuitive control scheme via forearm muscle
contraction signals are presented. This approach can be
synthesized by the term biomechatronic design. The hand
has cambered palm and five active fingers. Each finger is
integrated with position and torque sensors. This feature
offers the hand more grasping patterns and complex control
methods. On the other hand, by carefully configuring the
human hand gestures, 9 patterns of grasp modes can be
recognized with 10 FSR sensors. There may be some
prosthetic hands that have more DOFs or more sensors, but
considering the small size, low weight, human-like cosmetics,
biomimetic controllability and perception in all; the hand that
we developed is more capable of being an intimate extension
of human body. In fact, prosthetic hands as substitute for
human hand cannot be achieved by existing artificial
technologies.
The prosthetic hand can be applied in two aspects. Firstly,
it can be treated as an end operator of a multi-DOF robot arm
to interact with the environment. Second, with a human likeglove, it can be utilized for the prosthetic uses.
Furthermore, more work should be done to study
intelligent force control methods which can be applied for
regulating the grasp force. Besides, the bio-feedback control
strategy is being researched, which will communicate a sense
of touch back to the user. These features enable the hand to
respond seamlessly to the intent of human user.
REFERENCES
[1] C. Pylatiuk, S. Mounier, A. Kargov, S. Schulz, G. Bretthauer, Progressin the development of a multi-functional hand prosthesis, inProc. 26th
IEEE Annu. Int. Conf. on Engineering in Medicine and Biology Society,
San Francisco, USA, 2004, pp. 4260-4263.[2] L. Zollo, S. Roccella, R. Tucci, et al. Biomechatronic design and
control of an anthropomorphic artificial hand for prosthetics hand
robotic applications,IEEE Trans. on Mechatronics, vol. 12, no. 4, pp.
418-429, Aug.2007.
[3] J. Z. Yang, E. P. Pitarch, K. Abdel-Malek, A. Patrick, L. Lindkvist, Amulti-fingered hand prosthesis,Mechanism and Machine Theory, vol.
39, no. 6, pp. 555-581, 2004.
[4] C. Cipriani, M. Controzzi, M. C. Carrozza, Progress towards thedevelopment of the smarthand transradial prosthesis, inProc. of 2009
IEEE Int. Conf. on Rehabilitation Robotics, Japan, 2009, pp. 82-687.
[5] J. L. Pons, E. Rocon, R. Ceres, et al. The MANUS-HAND dextrousrobotics upper limb prosthesis: mechanical and manipulation aspects,
Autonomous Robots, vol. 16, no. 2, pp.143-163, March 2004.
[6] S. Bitzer, P. Van Der Smagt, Learning EMG control of a robotic hand:towards active prostheses, in Proc. of the 2006 IEEE Int. Conf. on
Robotics and Automation, Orlando, 2006, pp. 2819-2823.
[7] C. Castellini, P. Van Der Smagt, Surface EMG in advanced handprosthetics,Biological Cybernetics, vol. 100, no. 1, pp. 35-47.
[8] O. Amft, H. Junker, P. Lukowicz, G. Troster, C. Schuster, Sensingmuscle activities with body-worn sensors, inProc. of Int. Workshop on
Wearable and Implantable Body Sensor Networks, vol. 2006, pp.
138-141.
[9] Y. Honda, S. Weber, T. C. Lueth, Intelligent recognition system forhand gesture, in Proc. of the 3rd Int. IEEE EMBS Conf. on Neural
Engineering, Hawall, USA, 2007, pp. 611-614.
[10] A. P. Sangole, M. F. Levin, Arches of the hand in reach to grasp,Journals of Biomechanics, vol. 41, pp. 829-837, 2008.
[11] M. Tetsuya, K. Haruhisa, Y. Keisuke, Anthropomorphic robot hand:Gifu hand III, ICCAS2002, pp. 1288-1293.
[12] B. Massa, S. Rosella, M. C. Carrizozo, P. Dario, Design anddevelopment of an underactuated prosthetic hand, inProc. of the 2002
IEEE Int. Conf. on Robotics and Automation, Washington. DC, 2002,
pp. 3374-3379.
[13] S. Schulz, C. Pylatiuk, G. Bretthauer, A new ultra lightanthropomorphic hand, in Proc. of the 2001 IEEE Int. Conf. on
Robotics and Automation, Seoul, Korea, May.2001, pp. 2437-2441.
[14] D. Rajiv, Y. Clement, L. Maurice, The design and development of agloveless endoskeleton prosthetic hand, Journal of Rehabilitation
Research and Development, vol. 35, no. 4, pp. 388-395, 1998.
[15] L. Biagiotti, F. Lotti, C. Melchiorri, G. Vassura, Mechatronic design ofinnovative fingers for anthropomorphic robot hands, in Proc. of the2003 IEEE Int. Conf. on Robotics and Automation , 2003, vol. 3, pp.
3187-3192.
[16] D. J. Montana, The condition for contact grasp stability, inProc. ofIEEE Int. Conf. on Robotics and Automation, 1991, vol. 1, pp. 412-417.
[17] S. Schimizu, M. Shimojo, S. Sato, et al. The relationship betweenhuman grip types and force distribution pattern in grasping, inProc. of
8th International Conf. on Advanced Robotics, 1997, pp. 299-304.
[18] Force sensing resistor integration guide and evaluation parts catalog.Interlink Electronics. [Online]. Available:
http://www.interlinkelectronics.com.
[19] S. S. Keerthi, C. J. Lin, Asymptotic behaviors of support vectormachines with Gaussian kernel, Neural Computation, vol. 15, no. 7,
pp. 1667-1689, 2003.
Fig. 15. Grasp pattern of the hand
214