Professional Documents
Culture Documents
the constituting mechanisms. One key advantage of human- Despite the desire for dexterity in prosthetics, the most
like mechanisms is that the robotic hand possesses nonlinear, commonly used prosthetic hand is a mechanically controlled
nonconstant relationships between the actuators (muscles), and hook prosthesis [34] which was designed over a century ago.
joints. We develop mathematical models for these relations Several researchers [22], [50], [16] and companies [63], [47],
based on the data collected from experiments with the ACT [52] have designed robotic hands specifically for prosthetic
Hand. We demonstrate that the moment arm variations match purposes with attention toward minimizing weight, simplifying
with the known relations in the human hand. Moreover, our controls, and aesthetics. Some of the current commercial
model is more comprehensive than the existing models for the prostheses are controlled by means of electromyographic
human hand thus leading to a better understanding of human (EMG) signals recorded using surface electrodes, which detect
hand biomechanics. The moment arm relations are key for electrical activity related to the patient’s arm muscles [63],
achieving reliable position control of the digits of the ACT [52]. Because of the difficulty in translating the user intent
Hand. We present results from position tracking of the index into useful controls signals, current prosthetic hands have
finger and thumb. only one or two DOFs; thus leading to limited functionality.
Some surveys reveal that 30 − 50% of the upper extremity
amputees do not use their prosthetic hand regularly because of
II. R ELATED WORK
reduced functionality, poor cosmetic appearance, and limited
Robotic hands have been developed for many decades for a controllability [3], [59], [16]. Ideas for controlling of robotic
variety of applications ranging from industrial manufacturing hands using neural signals have also been explored. Recent
to prosthetics to humanoid robot research. Although numerous studies enable monkeys to control the 3D movement of a
grasping and manipulation results have been demonstrated by robotic arm to achieve self-feeding tasks [49], [69]. Over thirty
existing anthropomorphic hands (e.g. fast manipulation by the human arm/hand amputees have received nerve reinnervation
hand developed by the University of Tokyo [57]), the capa- surgery to rewire the peripheral nerves that used to go into
bilities of the current robotic hands do not match the abilities the hand/arm to the chest muscle instead [40]. The signals
of the human hand. Design of a versatile and robust robotic amplified by the natural muscle can then be tapped into with
hand that demonstrates human-like grasping and manipulation surface EMG for prosthetic arm/hand control.
capacity is a challenging task. The design decisions include A number of researchers have developed computational
the number of fingers, number of joints, degrees of freedom models of hand biomechanics [25], [33]. While some of the
(DOFs), range of motion for the joints, speed of movements recent models do include human-like muscles, tendons and
and force generation capacity. The design choices have to bones [33], [62], [65], these models still face difficulties
be made under the tight space and weight constraints. Focus in simulating critical nonlinear relationships, for example,
has been on increasing the number of DOFs, adding sensors, tendons sliding over bones, and nonlinear joint movements.
and implementing novel controls, including machine learning For instance Sueda et al. present an automatic technique for
methods. Table I presents a list of representative robotic hands generating the motion of tendons and muscles [62], and Tsang
with important characteristics of these hands. Note that in et al. present an anatomically accurate inverse dynamics of
many of cases the robotic hands are still under development the hand [65]. However, these methods do not represent the
so the grasping and manipulation abilities are based on the nonlinear moment arm relationships for the hand muscle which
design specification not from actual demonstrations. are critical in defining biomechanics of the hand. Considering
Many robotic hands are driven directly or through gears, the challenges in modeling nonlinear relations and advan-
for example, the Stanford/JPL Hand [54], Barrett Hand [64], tages gained through experiments with physical prototypes,
the Southampton Hand [42], the Gifu Hand II [38] , the we believe that it is important to continue developing both
NASA/JPL Robonaut Hand [11], the NAIST hand [67], and computation and physical models of human hand.
the high speed hand from the University of Tokyo [57] (the To improve the performance and capabilities of the robotic
drive mechanisms for the recently developed DEKA arm-hand hands new ideas have to be introduced for both the robot
prosthesis [24] have not been published). Examples of tendon- hardware and controls. Robotics researchers can greatly ben-
driven hands include: the Okada Hand [51], the Utah/MIT efit from a better understanding of the biomechanics and
Hand [35], [29], [61], the DLR II Hand [43], the UB Hand neuromuscular controls of the human hand. Although many
3 [45], the Vanderbilt Hand [22], the CyberHand [15], the robotic hands are designed to be anthropomorphic, the intrinsic
Karlsruhe Hand [36], [64], the smart motor and air muscle mechanisms of actuation and controls, in most cases, have not
Shadow Hands [58], and the Keio Hand [73]. been anatomical. In this context, the ACT Hand is designed
The idea of designing smart mechanisms to simplify con- to be a tool to investigate human dexterity. By incorporating
trols has been explored in some of the existing hands. For the biomechanical features of the human hand the ACT Hand
example, in the under-actuated grippers such as the SDM allows for the identification of the critical factors that lead to
Hand [27] and SPRING Hand [17], joint compliance is dexterity in the human hand.
implemented to be able to conform to different shapes for
grasping with relatively simple controls. Some researchers
have developed soft skin to accommodate against errors and
to embed sensors [56], [8]. However, the ability of these hands 1 T, I, M, R and P denote thumb, index, middle, ring, and little finger,
to execute human-level manipulation tasks remains limited. respectively.
3
TABLE I
F EATURES OF EXEMPLARY ROBOTIC HANDS
Robotic hands # # joints / DOF Range of Speed of Activation / transmission Types of grasps / ma-
identical (Total DOFs)1 motion motion method nipulation
fingers
Hosmer hook [34] 2 split 1/1 (1 DOF) <human <human body-powered splitting hook pinch
hooks hand hand
Utah Arm / Liberating / 2 T-1/1, I-1/0, M-1/0 <human <human EMG signal driven, DC three-finger pinch
OttoBock [47], [44], [52] (1 DOF) hand hand motor, cable
USC/Belgrade [7] 4 T-3/2, I-3/0.5, <human <human DC motor, cable, linkage grasp: power & finger tip
M-3/0.5, R-3/0.5, hand hand
P-3/0.5 (4 DOFs)
Harvard SDM [28] 4 4×2/1 (1 DOF) >human <human DC motor, cable, elastic enveloping grasp
hand hand joints
Gatech Dusty [72] 1 2/1 (1 DOF) <human – DC motor, cable, spring nonprehensile grasp
hand hinge joints
Barrett [64] 3 T(Right)-2/1.5, >human ≈1.2×human DC motor, worm drives grasp: power & finger tip
T(Left)-2/1.5, I-2/1 hand hand integrated with cable drive
(4 DOFs) and breakaway clutch
i-Limb [63] 4 T-3/1, I-2/1, M-2/1, <human ≈human DC motors, belt transmission grip: key, hook, power &
R-2/1, P-2/1 (5 hand hand precision;
DOFs) grasp: spherical & pal-
mar
Southampton [42] 4 T-2/2, I-3/1, M-3/1, <human ≈0.22×human DC motor, worm-wheel, lead power grasp, lateral
R-3/1, P-3/1 (6 hand hand screw pinch
DOFs)
Cyber [15] 5 T-4/2, I-3/1, M-3/1, ≈0.22×human ≈0.38×human geared DC motor, lead lateral pinch;
R-3/1, P-3/1 (6 hand hand screw, cable, exensor spring grasp: cylindrical, spher-
DOFs) ical & tripod
Univ. of Tokyo Hand [57] 3 T(R)-3/3, I-2/2, >human =15×human DC motor, harmonic and grasp: power & finger
T(L)-3/3 (8 DOFs) hand hand bevel gear transmission tip;
dynamic manipulation
Stanford/JPL [54] 3 T-3/3, I-3/3, M-3/3 >human — DC motor, cable finger tip grasp
(9 DOFs) hand
DARPA hand [23] 4 T-3/3, I-3/2, M-3/2, <human <human DC motor, cable, gear grasp: hook & power
R-3/2, P-3/2 (11 hand hand transmission
DOFs)
Robonaut [46] 4 T-5/3, I-4/3, M-4/3, ≈human <human DC motor, flex shaft, lead grasp: power & finger
R-3/3, P-3/1 (11 hand hand screw, cable tip;
DOFs) lateral pinch
Naist [67] 4 T-4/3, I-4/3, M-4/3, ≈human ≈human geared DC motor, bevel gear power grasp
R-4/3 (12 DOFs) hand hand
DLR II [14] 4 T-4/4, I-4/3, M-4/3, >human =3×human DC motor, belt, harmonic grasp: power & finger
R-4/3 (13 DOFs) hand hand drive, bevel gears tip;
lateral pinch
Utah/MIT [35] 4 T-4/4, I-4/4, M-4/4, <human ≈1.82×human pneumatic actuator, cable finger tip grasp
R-4/4 (16 DOFs) hand hand /manipulation
Gifu III [48] 4 T-4/4, I-4/3, M-4/3, ≈human ≈1.35×human DC motor, gear transmission, power grasp
R-4/3, P-4/3 (16 hand hand linkage mechanism
DOFs)
UB III [45] 4 T-3/4, I-4/4, M-4/3, <human ≈0.51×human DC motor, cable, helical grasp: power & finger tip
R-4/2, P-4/3 (16 hand hand spring
DOFs)
Shadow [58] 4 T-5/5, I-4/3, M-4/3, ≈human ≈0.5×human air muscle, cable and spring grasp: finger tip & power
R-4/3, P-4/3 (17 hand hand
DOFs)
Keio [73] 4 T-4/4, I-4/4, M-4/4, ≈human ≈2×human ultrasonic motors, elastic grasp: power & finger
R-4/4, P-4/4 (20 hand hand elements, cable tip;
DOFs) lateral pinch
4
TABLE II
III. ACT H AND M ECHANISMS ACT HAND PHALANGE LENGTHS
1) Index and Middle Finger Joints: There are three joints in arm. The sweep of the joint cavity restricts the movement of
the index and middle finger, namely, the metacarpophalangeal the gimbal assembly to the appropriate MCP joint range of
(MCP), proximal interphalangeal (PIP) and distal interpha- motion.
langeal (DIP). These are modeled by a novel design involving The CMC joint involves two pin joints at the ends of a single
pin joints in the ACT Hand. The PIP joint is located at the link arm to realize the AA and FE degrees of freedom. Though
distal end of the proximal phalangeal bone, and the DIP joint the CMC and MCP joints are conceptually similar in that they
is located at the distal end of the middle phalangeal bone. The both have FE and AA degrees of freedom, a gimbal design is
MCP joint has two DOFs: one to achieve flexion-extension and not suitable for the CMC joint because its two rotational axes
another to realize abduction-adduction finger motion. These are located in separate bones. One pin joint coincides with
two DOFs are realized by a gimbal mechanism at the distal the CMC AA axis in the proximal end of the metacarpal,
end of the metacarpophalangeal bone. To match the anatomical while the other pin joint represents the CMC FE axis, which
joint properties of the human index finger, as described in [12], intersects the trapezium carpal bone. Joint range of motion for
the abduction-adduction joint axis is oriented at 60◦ with each of the two axes is constrained by narrow slot cuts in the
respect to the metacarpophalangeal bone as shown in Figure metacarpal and trapezium bones. The IP joint design consists
2. of single pin joint to represent the flexion-extension degree of
freedom between the two phalangeal bones. A link arm rigidly
attached to the distal phalange rotates about an axle coinciding
with the IP FE axis in the proximal phalange. The geometry of
the articulating bone ends was maintained except for a narrow
slot that allows the small diameter link arm to rotate around
the IP FE axis pin. The span of the cavity enforces the joint
range of motion.
3) Joint Range of Motion: Joint limits for the flexion joints
are imposed by creating internal beam features that interfere
with one another at the limits. The ranges of motion were
chosen to match those of the human finger, and are shown
in Table III. Because the gimbals used for the MCP joints
are free to rotate by ±180◦ , we designed the bone shells to
limit the range of motion by contacting each other at the joint
limits. For example, Figure 2 shows the CAD model of the
MCP joint of the ACT Hand’s index finger with the bone shell
around the gimbal joint to ensure the correct range of motion.
TABLE III
ACT HAND FINGER JOINT MOTION LIMITS
pass over the joints, terminating at the insertion points on the chanics and control of the digits [1], [12]. To closely match
finger bones. Muscle contractions lead to hand movement and the moment arm variations in the human hand, we developed
force generation. In the ACT Hand, we used the brushless DC a web of tendons that mimics the human structure with the
motors (described later in the paper) as muscles and we fabri- spectra strings. The current version of the extensor mechanism
cated our tendons with 0.46 mm Spectra(R) strings. The string is fabricated by crocheting nylon composite to emulate the
was chosen because of its strength (200 N breaking strength), geometry and functionality of the human counterpart as closely
stiffness (4800 N/strain), and ability to slide smoothly over as possible as shown in Figure 5. We prioritized our tendon
the bones. The tendons in the human hand have elastic structure design to achieve any posture that a typical human
properties [74] which play a critical role in the hand dexterity finger can achieve, mimic the overall geometry based on [30],
and controls. The choice of a string with high stiffness allows produce a smooth surface to facilitate a sliding motion, and
for reliable and quantifiable relationships between the motors match the stiffness of a real tendon. The critical features of
and joints. Mimicking the stiffness properties of the human the extensor tendon web are the insertion points, the lateral
tendons, either through hardware or software controls is part bands and the hood, as shown in Figure 5.
of our future work.
1) Tendon Routing over the Wrist Joint: The human wrist
position affects finger joint properties (for example, joint
stiffness) because the tendons connecting the extrinsic muscles
to the fingers travel past the wrist joint, and the joint position
changes their path lengths. While designing the wrist for the (a) Schematic (b) Schematic
ACT Hand, we faced a tradeoff of minimizing tendon friction
versus routing tendons to be accurately affected by wrist
position. We chose to reduce tendon friction and explicitly
designed a routing to minimize the effect of wrist position on
tendon length. As shown in Figure 4, each tendon crosses the (c) Actual (d) Actual
wrist joint via a pair of sheaves: a large central sheave near Fig. 5. Tendon hood structure for the index finger and thumb.
the flexion/extension axis and a smaller outer sheave. The size
and placement of the tendon sheaves have been chosen so that We have mimicked the exact locations of the bone insertion
during wrist flexion/extension, the tendons unwind from one points in our tendon arrangements. The lateral bands serve
sheave as they are being wound onto the other, minimizing the following functions: (1) assist lateral movements via the
any change in path length. Wrist abduction/adduction has interosseous muscles when these muscles are activated as
a greater effect on path length due to the line of central antagonists, (2) allow the extension of the distal phalanx and
sheaves rotating out of the plane of the outer sheaves, but the finger as a whole when the interosseous muscles are
the effect is small, particularly for tendons routed near the activated as agonists in concert with the EDC muscle, and
abduction/adduction pivot point. Since the tension in each (3) coordinate the two distal joints (DIP and PIP) in flexion
tendon is under independent software control, it is possible and extension [71]. The hood structure enables the flexion of
to undo the abduction/adduction effects and apply the desired the MCP joint independent of other joints and rotation at the
flexion/extension effects as a function of wrist position. MCP joint [71].
3) Flexor Side: The tendons on the flexion side are con-
nected directly from the bone insertion points to the actuators.
For example, notice the tendon routing in on the flexion side
of the thumb in Figure 3. The flexion tendons pass through
guiding ‘rockers’ that allow for smooth travel of the tendon
while holding it close to the bone forms in order to achieve
accurate moment arm lengths about the finger joints.
actuate by contraction. We investigated the possibility of using for any motor to be easily replaced and entire clusters to be
a direct linear muscle-like actuator, such as a Shape Memory added or removed to match the number of actuators required
Alloy [41], McKibben muscles [70], or other artificial mus- for a particular setup. Additionally, the clusters incorporate
cles [66]. However, none of these actuators can provide the guiding sheaves to route the tendon strings from their radial
fast response time (< 200ms), total excursion lengths (> 4cm) orientation after leaving the motor shaft to an axial orientation
and possibility of variable stiffness that a human muscle can to reach the wrist.
produce. At this stage of the ACT Hand’s design our focus is
on mimicking the static and dynamic force and movement E. Mounting the Hand on to an Arm
generation capacity, rather than the shape, size, or weight Since our focus is on understanding the movements of
of the real muscle. Hence, the DC motors were chosen as the hand and fingers, we mimicked the anatomy of the
actuators. To overcome the torque ripple in the DC motors, we fingers and palm. Our wrist design is not anatomical in that
custom designed the DC motors 2 based on the Kollmorgen it does not match the bone shapes and joint axes of the
motor [39] as described below. human wrist. However, we mimicked its flexion-extension and
abduction/adduction motions.
The wrist has two DoFs: a “yaw" joint (±15 degrees)
attached to the end of the arm, and a “pitch" joint (±30
degrees) connected to the hand. Three banks of pulleys are
mounted on the gimbal structure joining the two DoFs and
guide the tendons from the arm-mounted actuators to the finger
joints. These pulleys minimize friction in the tendon paths and
route the tendons near the pitch axis to mimic the kinematic
coupling present in a human hand. As explained above, all
ACT Hand actuators are located in the forearm. The forearm
and hand are mounted on a Whole Arm Manipulator (WAM)
developed by Barrett Technologies, Inc. [6]. The WAM is a
4-DOF, cable-driven back-drivable manipulator. Its DC motor
control allows position and force control modes. The ACT
Hand forearm was connected to the WAM at the elbow joint
Fig. 6. ACT Hand actuators arranged in the forearm. using the same physical connection that connects the WAM
forearm. Figure 6 shows the entire assembly.
1) Magnet Skewing: One of the known issues with the IV. M USCLE -J OINT M APPINGS : M OMENT A RMS
permanent magnet DC motors is the motor cogging, which
shows up as torque ripple at very low speeds [9]. The cogging An important characteristic of the human hand is the me-
effects can lead to errors in finger positions and jerky finger chanical advantage, called the moment arm, that each muscle-
motions, especially during low speed operations. To address tendon combination has on each joint. The muscle moment
this, we designed a rotor with permanent magnets cut into arms in the human hand are configuration dependent and
three pieces, longitudinally, which are staggered at 10◦ axial play a critical role in hand movement control; however, the
rotation with respect to each other. This method, called step exact properties of the moment arm variations are not known.
skewing of the magnet, is a known, low cost way to reduce Because the ACT Hand mimics hand biomechanics, through
the cogging torques [9]. The magnet skewing reduced the bone shapes and tendon hood structure, its muscle moment
cogging torque to less than 30% of the original value, while arms are also configuration dependent, and determining the
the peak torque reduced by only 5%. exact moment arm relationships is critical for ACT hand
2) Motor Controller: Each motor is connected to a brush- controls. Also determination of the moment arm relationships
less servo controller (called the Puck [6]). Each controller has in the ACT Hand can potentially lead to a better understanding
an embedded photo-sensor and an encoder wheel (with 114 of human hand biomechanics.
ticks/deg) that allows for a high precision position sensing of We developed a method to acquire the moment arm relation-
the motor rotation to mimic muscle spindles (see Figure 6). ships for the ACT Hand that is based on an analysis of motion
The controllers are connected to an RTAI Linux [53] machine, capture data for finger and muscle-motor movements. What
which provides motor position readings at high frequency follows are sample results from the moment arm analysis for
(> 500 Hz). the index finger of the ACT Hand. To determine the moment
3) Motor Housing in the Forearm: All motors are located in arms, we moved the index finger through its joint range of
the forearm, as shown in Figure 6. The motors are arranged in motion and recorded the joint angles and changes in muscle
clusters of six units, and six clusters are connected end-to-end. lengths. We then found a functional mapping, fi , among all
The modular design of the clusters and motor housings allows four joint angles and each muscle excursion using a Gaussian
process (GP) based regression model.
2 The brushless DC motor properties are: length 3 cm, outside radius 2.2
cm, peak torque 40 mN-m, stall torque 40 mN-m, torque constant 0.13 mN-
m/A and rise time 0.1 s. li = fi (θ) i = 1, ..., 6, (1)
8
18
EI excursion (mm)
(θ = [θ1 , θ2 , θ3 , θ4 ]T ). We define θ1 - MCP Ab-Ad, θ2 - MCP 10
Flex, θ3 - PIP Flex and θ4 - DIP Flex, with the abduction and 8
6
flexion as positive directions. The moment arm relationships
4
were determined by taking the partial derivatives of the muscle
2
excursion functional mappings with respect to the joint angles. 0
In the case of the ACT Hand index finger, the moment arm is −2
28 29 30 31 32 33 34 35 36 37 38
defined by a matrix R of dimension 6 × 4. time (sec)
l˙ = R(θ)θ˙ (2) Fig. 8. Actual vs estimated length changes in the EI muscle during a typical
finger motion.
where,
∂li ∂fi
Rij (θ) = = i = 1, ..., 6 and j = 1, ..., 4. (3) shows the mapped muscle length when it was projected to EI
∂θj ∂θj
muscle length estimation. Table IV shows the mean absolute
Figure 7(a) shows the variations in the excursions of EI error for all muscle length excursions when tested on a data
muscles as two finger angles vary. The dots in the figure are the set of over 200, 000 angles and muscle lengths combinations.
data points, and the surfaces show the fitted mapping functions. The data covers the physiological ranges of motion for all
Figure 7(b) shows the variations in the moment arm of the EI the finger joints. The actual excursion length was recorded
muscle wrt to the MCP flex angle, and Figure 7(c) shows the directly from the encoder values, and the other estimations
variations in the moment arm of the EI muscle wrt to the PIP were from the joint angle information from a motion capture
angle. system translated to muscle lengths. We used an optical motion
tracking system (Vicon 360 with six M2 cameras) to record
motions of the finger involving all four joint angles. Thirteen
markers, each 3 mm in diameter were placed on the ACT
finger and the distribution of the markers was as follows: five
20
EI 10
on the MCP bone, three on the proximal and middle phalange,
0 60 and two on the distal phalange. The XYZ positions of the
g)
(de
0 30 markers were recorded at 120 Hz, and finger joint angles were
x
fle
MC 30
Pf
determined by using an angle determination algorithm built
PIP
lex 60
(de 0
g ) 90
into the motion analysis software (Vicon iQ 2.5).
(a) EI excursions TABLE IV
M EAN ABSOLUTE ERRORS IN MUSCLE TO JOINT MAPPINGS ( MM )
(de
(de
0 30 0 30
x
x
Fle
Fle
MC 30 MC 30
P Flex 0 P Fle 0
P
60
PIP
(deg 60 x
PI
(de
g)
) 90 90 B. Comparison with Cadaver Data
(b) Variations in the EI moment arm (c) Variations in the EI moment arm We compare our results with index finger cadaver data
wrt. MCP flex wrt. PIP flex
from [1] which is one of the most comprehensive data sets
Fig. 7. Muscle excursions and moment arm variations for the EI muscle available. To make the comparison variable moment arm data
with respect to the MCP and PIP flexion-extension, as MCP flexion and PIP
angles change.
from [1], which is from one female specimen, we generated
slice plots from our data by keeping the other joint angles
The mean error across all data points with the GP-based constant as it was done for the cadaver data. The moment
mapping is 0.65 (sd 0.33) mm. This error is low when arm values are higher in our case by, on average 125% due
compared with the total excursions which are in the range to the fact that we modeled the ACT Hand size after a male
of 30 − 40 mm. Our results show that the excursion lengths subject. A comparison between the ACT Hand moment arms
and moment arms for all muscles of the index finger depend and scaled human moment arms (by 125%) gives an overall
significantly on all the joint positions of the finger. mean error 3 ± 1.5 mm. The sign of moment arms, indicat-
ing contribution to flexion/extension or abduction/adduction,
for all muscles, except for LUM in parts of finger flexion,
A. Validation match with the cadaver plots. Table V gives the correlation
We conducted experiments to test the quality of the angle to coefficients between individual muscle plots for MCP Ab-ad
muscle length mapping determined by our method. Figure 8 and flexion variations. The first row gives the moment arms
9
wrt MCP flexion angle and the second row gives the moment joint at a time. Also, our study analyzes moment arm variation
arm wrt MCP abduction-adduction angle. The differences in with respect all four joints of the index finger which is missing
variations in LUM and RI with flexion angle and PI with in the previous studies which focused only on the two MCP
adduction might have arisen due to the differences in the joints of the index finger. Thus using the ACT Hand we have
structure of LUM in the cadaver hands and the ACT Hand. determined previously unknown moment arm relationship in
The ACT lumbrical tendon and associated motor is attached the human index finger.
to the equivalent of a skeletal anchor point. In contrast, the Variable moment arms play a significant role in the move-
lumbrical tendon and associated muscle in human hands are ment control of the finger. For example, Figure 9 shows a
attached to another sliding tendon. plot of variations in muscle excursion velocity of RI for the
same angular velocities of MCP Ab-Ad (0.5 rad/sec), MCP
TABLE V
C ORRELATION COEFFICIENTS BETWEEN MOMENT ARM CURVES FROM THE
Flex(1.0 rad/sec), PIP Flex (1.0 rad/sec) and DIP Flex (1.0
CADAVER AND ACT H AND . rad/sec). As the MCP Ab-Ad and PIP Flex angles vary, while
MCP Flex and DIP Flex are kept constant, the RI velocity
changes from positive to negative. This means that the muscle
EI PI FDP LUM FDS RI
switched from being active and pulling to being passive and
MCP FE 0.94 0.47 0.83 -0.71 0.85 -0.95 stretching. The figure also shows that a constant moment
MCP AA 0.99 -0.66 0.99 0.67 0.93 0.99 arm model, assumed in all the previous hand biomechanics
studies, does not capture the variations in muscle velocities
as functions of the finger configuration. Thus to control the
finger joint velocities muscle velocities are generated based
C. Physical Interpretation of Moment Arm Variations on the finger configuration using the variable moment arms.
The extrinsic muscle excursions show significant depen- A constant moment arms assumption will lead to erroneous
dency on flexion but little variation with abduction-adduction. model of neuromuscular controls.
In contrast, the intrinsic muscle excursions show significant
dependency on both flexion and abduction. This means that V. G RASPING AND H AND M OVEMENT I LLUSTRATION
moment arms for all muscles of the index finger depend
Taking advantage of the DOFs and ranges of motion in
significantly on all the joint positions of the finger and, for
the ACT Hand, which are copied from the human hand, the
some of the muscles, the moment arms change sign. A negative
ACT Hand is able to grasp and manipulate a number of
value for the moment arm of a muscle with respect to a joint
objects, which are part of daily activities. Figure 10 shows
indicates that the muscle is contracting for positive change in
nine examples of object grasping using the ACT Hand. These
joint angle, meaning that the muscle actively contributes to
pictures demonstrate the DOFs and range of motion of the
joint movement. Conversely, a positive value for the moment
finger joints, and overall anthropomorphic grasping abilities.
arm means that the muscle is stretched for positive change in
These nine objects were chose to demonstrate the wide variety
joint angle. The moment arms for the muscle with respect to
of grasps that can be achieved with the ACT Hand, for
the primary angles show large variations and maintain the sign
example, power grasp in case of the water bottle and pinch
of the moment arms.
grasp in case of the spoon. Grasping performance is greatly
affected by the object interactions through the skin contact.
Currently we are developing tactile skin for the ACT Hand.
constant
10 moment arm These grasps were achieved through a direct muscle control
RI muscle velo (mm/sec)
contracted and extensors are stretched for both the thumb and tions of the thumb, index finger, middle finger, and wrist
index finger. Notice that the muscle lengths change smoothly of the ACT Hand. The ACT Hand is designed to further
over the maneuver leading to smooth motions of the fingers. our understanding of human hand mechanisms and control
The mean tracking error over all muscles is 1.67 (sd 0.34) and to provide guidelines for building versatile prosthetic
mm. and dexterous hands. We reviewed the state-of-the-art robotic
Considering that thumb movements account for more than hands and demonstrated that the unique design goals and
50% of hand function [20], we also implemented a thumb features distinguish the ACT Hand when compared to the
movement of rubbing against the index finger, which is part existing robotic hands. In Section III we have presented a
of many daily hand movements such as counting money number of novel mechanical elements which are designed
and opening a plastic bag. Because of the mimicking of and built to mimic human hand biomechanics. These include
human thumb biomechanics in the ACT Hand, including axes the bone structures, joints, tendon arrangements and actuators
locations, number of DOF, and muscle arrangements, we were in the fingers and thumb, and also the tendon routing and
able to generate an abduction-adduction motion at the MCP actuation of the wrist.
joint in the ACT Hand thumb. Figure 12 shows the snapshots With the anatomically correct mechanical elements the ACT
of the thumb motion and corresponding length changes of the Hand uniquely mimics the human hand biomechanics. We
eight muscles during this motion. have carried out a data driven kinematic analysis of the
relationships between the muscles and joints, defined by the
Index Finger
2
neutral position EI moment arms. Our analysis shows that the values for the
1.5
moment arms for the muscles in the ACT Hand vary with
Excursion (cm)
1 RI
0.5 PI the configuration of the hand, and these variations match with
0 key grasp
−0.5
the available data on human hand biomechanics. Our model
LUM
−1
FDS for moment arm variations is more comprehensive than the
−1.5
FDP existing models for the human hand, thus leading to a better
0 0.5 1 1.5 2 2.5 3 3.5 4
understanding of human hand biomechanics. Development of
Thumb tactile skin and passive joint properties in the ACT Hand are
neutral position EPL
part of ongoing research.
Excursion (cm)
1 EPB
0.5
APB Completion of the ACT Hand mechanisms and software
APL
key grasp
0 OP platform allows us to conduct many experiments, which were
ADP
−0.5
FPB
hitherto impossible, to study human hand properties and trans-
−1
FPL
lation of important properties into robotic forms. For example,
0 0.5 1 1.5 2 2.5 3 3.5 4 now we can implement novel control algorithms to develop
Time (sec)
a deeper understanding of human dexterity. Our group has
Fig. 11. Length changes in thumb and index finger muscles during key grasp investigated the existence and importance of muscle synergies
maneuver starting from neutral position (fingers open) during hand movement and force control [5], [4] and we are
implementing these mechanisms in the ACT Hand. We are also
investigating control strategies for achieving neuromuscular
control of ACT Hand muscles [26]. We are planning to
develop replicas of the ACT Hand and make them available
for other researchers. This will allow for the simultaneous
experimentation and growth in understanding of human hand
complexities.
APL
EPL
EPB
APB
R EFERENCES
ADP
FPB
[1] K. N. An, Y. Ueba, E. Y. Chao, W. P. Cooney, and R. I. Linscheid,
FPL
OP “Tendon excursion and moment arm of index finger muscles,” Journal
of Biomechanics, vol. 16, pp. 419–425, 1983.
[2] K.-N. An, “Tendon excursion and gliding: Clinical impacts from humble
concepts,” Journal of Biomechanics, vol. 40, pp. 713–718, 2007.
[3] D. J. Atkins, D. C. Heard, and W. Donovan, “Epidemiologic overview of
individuals with upper limb loss and their reported research priorities,”
Fig. 12. With the anatomical DOFs the thumb in the ACT Hand can be Internal Journal of Prosthetics and Orthotics, vol. 8, 1996.
moved to rub the side of the index finger which is a common and useful [4] R. Balasubramanian and Y. Matsuoka, “Biological stiffness control
human motion. The top row shows a sequence of photos taken at different strategies for the anatomically correct testbed (ACT) hand,” in Pro-
times (dotted lines) during the movement execution. And the bottom row ceedings of the 2008 IEEE International Conference on Robotics and
shows the plots of length changes (in mm) in the eight muscles of the thumb Automation., 2008.
during this movement. [5] ——, “The role of small redundant actuators in precise manipulation,”
in Proceedings of the 2009 IEEE international conference on Robotics
and Automation., 2009.
VI. C ONCLUSIONS AND F UTURE W ORK [6] Barrett Technology, Inc., “WAM specifications,” September 2009,
http://www.barrett.com/robot/products-arm-specifications.htm.
This paper presents the novel constituting mechanisms, [7] G. Bekey, R. R. Tomovic, and I. Zeljkovic, “Control architecture for the
unique muscle to joint relations, and movement demonstra- belgrade/USC hand,” Dextrous Robot Hands., pp. 136–149, 1990.
11
Fig. 10. The ACT Hand can grasp everyday objects with human-like finger postures. Figures show grasping of nine objects, namely, a coffee plate, key,
credit card, hand towel, glasses, water bottle, cordless phone, medicine bottle, and spoon, which are identified as important for ALS patients [19].
[8] L. Biagiotti, P. Tiezzi, Vassura, and C. G. Melchiorri, Modelling and by people with ALS,” in International Conference on Rehabilitation
controlling the compliance of a robotic hand with soft finger-pads, Robotics, 2009.
F. Barbagli, D. Prattichizzo, and K. Salisbury, Eds. Springer Tracts [20] J. C. Colditz, “Anatomic considerations for splinting the thumb,” Reha-
in Advanced Robotics, 2005. bilitation of the hand: surgery and therapy, 1990, m. E. J. Hunter J. M.,
[9] N. Bianchi and S. Bolognani, “Design techniques for reducing the Callahan A. D., Ed. Philadelphia: C. V. Mosby Company.
cogging torque in surface-mounted PM motors,” Industry Applications, [21] W. P. Cooney, M. J. E. Lucca, Y. Chao, and R. L. Linscheid, “The
IEEE Transactions on, vol. 38, no. 5, pp. 1259–1265, Sep/Oct 2002. kinesiology of the thumb trapeziometacarpal joint,” Journal of Bone
[10] S. Bidic, J. Imbriglia, and Y. Matsuoka, “An anatomical hand for Joint Surg Am, vol. 63, pp. 1371–1381, 1981.
instruction and simulation,” in American Society for Surgery of the Hand [22] S. A. Dalley, T. E. Wiste, T. J. Withrow, and M. Goldfarb, “Design
Meeting, Chicago, 2003. of a multifunctional anthropomorphic prosthetic hand with extrinsic
[11] W. Bluethmann, R. Ambrose, M. Diftler, S. Askew, E. Huber, M. Goza, actuation,” IEEE-ASME Transactions on Mechatronics, vol. 14, pp. 699–
F. Rehnmark, C. Lovchik, and D. Magruder, “Robonaut: A robot 706, 2009.
designed to work with humans in space,” Autonomous Robots, vol. 14, [23] DEKA, 2008, http://www.dekaresearch.com.
pp. 179–197, 2003. [24] DEKA-HAND, “Dean Kamen Luke arm prosthesis,”
[12] P. W. Brand and A. M. Hollister, Clinical Mechanics of the Hand, 3rd ed. http://www.dekaresearch.com/index.shtml, 2010.
Mosby, 1999. [25] S. Delp and J. Loan, “A graphics-based software system to develop and
analyze models of musculoskeletal structures,” Computers in biology
[13] P. W. Brand and M. H. Anne, Clinical Mechanics of the Hand. St.
and medicine, vol. 25, no. 1, pp. 21–34, 1995.
Louis: Mosby-Year Book, Inc., 1993.
[26] A. Deshpande, J. Ko, D. Fox, and Y. Matsuoka, “Anatomically correct
[14] J. Butterfass, M. Fischer, M. Grebenstein, S. Haidacher, and
testbed hand control: Muscle and joint control strategies,” in Proceedings
G. Hirzinger, “Design and experiences with DLR hand II,” in Automation
of the 2009 IEEE International Conference on Robotics and Automation,
Congress, 2004. Proceedings. World, vol. 15, 28 2004-July 1 2004, pp.
2009, pp. 4416–4422.
105–110.
[27] A. M. Dollar and R. D. Howe, “A robust compliant grasper via shape
[15] M. C. Carrozza, G. Cappiello, S. Micera, B. B. Edin, L. Beccai, and deposition manufacturing,” IEEE/ASME Transactions on Mechatronics,
C. Cipriani, “Design of a cybernetic hand for perception and action,” vol. 11, pp. 154–161, 2006.
Biological Cybenetics, vol. 95, no. 6, pp. 629–644, 2006. [28] A. Dollar and R. Howe, “Simple, robust autonomous grasping in
[16] M. C. Carrozza, B. Massa, S. Micera, R. Lazzarini, M. Zecca, and unstructured environments,” in 2007 IEEE International Conference on
P. Dario, “The development of a novel prosthetic hand- ongoing research Robotics and Automation, April 2007, pp. 4693–4700.
and preliminary results,” IEEE/ASME Transactions on Mechatronics, [29] K. A. Farry, I. D. Walker, and R. G. Baraniuk, “Myoelectric teleoper-
vol. 7, pp. 108–114, 2002. ation of a complex robotic hand,” IEEE Transactions on Robotics and
[17] M. C. Carrozza, C. Suppo, F. Sebastiani, B. Massa, F. Vecchi, R. Laz- Automation, vol. 12, pp. 775–788, 1996.
zarini, M. R. Cutkosky, and P. Dario, “The spring hand: Development [30] M. Garciaelias, K. AN, L. Berglund, R. Linscheid, W. Cooney, and
of a self-adaptive prosthesis for restoring natural grasping,” Autonomous E. Chao, “Extensor mechanism of the fingers. I: A quantitative geometric
Robots, vol. 16, no. 2, pp. 125–141, 2004. study,” Journal of Hand Surgery-American Volume, vol. 16A, no. 6, pp.
[18] L. Y. Chang and Y. Matsuoka, “A kinematic thumb model for the ACT 1130–1136, NOV 1991.
hand,” in Proceedings of the 2006 IEEE International Conference on [31] A. Hollister, W. L. Buford, L. M. Myers, D. J. Giurintano, and
Robotics and Automation., 2006. A. Novick, “The axes of rotation of the thumb carpometacarpal joint.”
[19] Y. S. Choi, T. Deyle, and C. C. Kemp, “Benchmarking assistive mobile Journal of Orthopaedic Research, vol. 10, no. 3, pp. 454–460, May
manipulation: A list of household objects for robotic retrieval prioritized 1992.
12
[32] A. Hollister, D. J. Giurintano, W. L. Buford, L. M. Myers, and [57] T. Senoo, Y. Yamakawa, S. Mizusawa, A. Namiki, M. Ishikawa, and
A. Novick, “The axes of rotation of the thumb interphalangeal and M. Shimojo, “Skillful manipulation based on high-speed sensory-motor
metacarpophalangeal joints.” Clin Orthop Relat Res, no. 320, pp. 188– fusion,” in Proceedings of the 2009 IEEE International Conference on
193, November 1995. Robotics and Automation., May 2009, pp. 1611–1612.
[33] K. Holzbaur, W. Murray, and S. Delp, “A model of the upper extremity [58] ShadowHand, “Shadow robot company,”
for simulating musculoskeletal surgery and analyzing neuromuscular http://www.shadowrobot.com/hand/, 2010.
control,” Annals of biomedical engineering, vol. 33, no. 6, pp. 829–840, [59] D. H. Silcox, M. D. Rooks, R. R. Vogel, and L. L. Fleming, “Myoelectric
2005. prostheses - a long-term follow-up and a study of the use of alternate
[34] Hosmer Dorrance Corporation, “Body-powered prosthetic hand,” prostheses,” Jounral of Bone and Joint Surgey - Americal Volume, vol.
September 2009, http://www.hosmer.com/products/hooks/index.html. 75A, pp. 1781–1789, 1993.
[35] S. Jacobsen, E. Iversen, D. Knutti, R. Johnson, and K. Biggers, “Design [60] W. P. Smutz, A. Kongsayreepong, R. E. Hughes, G. Niebur, W. P.
of the Utah/M.I.T. dextrous hand,” in Proceedings. 1986 IEEE Interna- Cooney, and K. N. An, “Mechanical advantage of the thumb muscles,”
tional Conference on Robotics and Automation., vol. 3, Apr 1986, pp. Jounral of Biomechanics, vol. 31, pp. 565–570, 1998.
1520–1532. [61] T. H. Speeter, “Primitive based control of the UTAH/MIT hand,” in
[36] A. Kargov, C. Pylatiuk, R. Oberle, H. Klosek, T. Werner, W. Roessler, International Conference on Robotics and Automation, 1991.
and S. Schulz, “Development of a multifunctional cosmetic prothetic [62] S. Sueda, A. Kaufman, and D. Pai, “Musculotendon simulation for hand
hand,” in IEEE International Conference on Rehabilitation Robotics, animation,” in ACM SIGGRAPH 2008 papers. ACM, 2008, pp. 1–8.
2007. [63] Touch Bionics, Inc., “Fitting and service manual„” May 2009,
[37] J. A. Katarincic, “Thumb kinematics and their relevance to function,” http://www.touchbionics.com/.
Hand Clinics, vol. 17, pp. 169–174, 2001. [64] W. Townsend, “The Barrett hand grasper - programmably flexible part
[38] H. Kawasaki, T. Komatsu, and K. Uchiyama, “Dexterous anthropo- handling and assemble,” The International Journal of Industrial Robot,
morphic robot hand with distributed tactile sensor: Gifu Hand II,” vol. 27, no. 3, pp. 181 – 188, 2000.
Mechatronics, IEEE/ASME Transactions on, vol. 7, no. 3, pp. 296–303, [65] W. Tsang, K. Singh, and E. Fiume, “Helping hand: an anatomically
Sep 2002. accurate inverse dynamics solution for unconstrained hand motion,” in
[39] Kollmorgen, Inc., “Kollmorgen motors,” 2010, Proceedings of the 2005 ACM SIGGRAPH/Eurographics symposium on
http://www.kollmorgen.com/. Computer animation. ACM, 2005, pp. 319–328.
[40] T. Kuiken, G. Dumanian, R. Lipschutz, L. Miller, and K. Stubblefield, [66] T. Tsuji, S. Miyata, T. Hashimoto, and H. Kobayashi, “Controller
“The use of targeted muscle reinnervation for improved myoelectric design for robot with pneumatic artificial muscles,” in International Joint
prosthesis control in a bilateral shoulder disarticulation amputee,” Pros- Conference SICE-ICASE, 2006.
thetics and Orthotics International, vol. 28, no. 3, pp. 245–253, DEC [67] J. Ueda, Y. Ishida, M. Kondo, and T. Ogasawara, “Development of the
2004. NAIST-Hand with vision-based tactile fingertip sensor,” in Proceedings
[41] M. Kumon, I. Mizumoto, Z. Iwai, and A. Indou, “Shape memory of the 2005 IEEE International Conference on Robotics and Automa-
alloy actuator with simple adaptive control,” in Second International tion., April 2005, pp. 2332–2337.
Conference on Innovative Computing, Informationand Control., 2007. [68] M. Vande Weghe, M. Rogers, M. Weissert, and Y. Matsuoka, “The ACT
[42] P. J. Kyberd, C. Light, P. H. Chappell, J. M. Nightingale, D. Whatley, hand: Design of the skeletal structure,” in Proceedings of the 2004 IEEE
and M. Evans, “The design of anthropomorphic prosthetic hands: A International Conference on Robotics and Automation., 2004.
study of the southampton hand,” Robotica, vol. 19, no. 6, pp. 593–600, [69] M. Velliste, S. Perel, M. C. Spalding, A. S. Whitford, and A. B.
2001. Schwartz, “Cortical control of a prosthetic arm for self-feeding,” Nature,
vol. 453, no. 7198, pp. 1098–1101, June 2008.
[43] T. Lan, Y. W. Liu, M. H. Jin, S. W. Fan, H. G. Fang, J. J. Xia,
[70] S. Wakimoto, K. Suzumori, and T. Kanda, “Development of intelligent
and H. Liu, “Dsp & fpga-based joint impedance controller for dlr/hit
mckibben actuator,” in IEEE/RSJ International Conference on Intelligent
ii dexterous robot hand,” in IEEE/ASME International Conference on
Robots and Systems., 2005.
Advanced Intelligent Mechatronics, 2009.
[71] D. Wilkinson, M. Vande Weghe, and Y. Matsuoka, “An extensor mecha-
[44] Liberating Technologies, Inc., “Upper extremity prosthetics„” September
nism for an anatomical robotic hand,” in Proceedings of the 2003 IEEE
2009, http://www.liberatingtech.com.
International Conference on Robotics and Automation., Sept. 2003, pp.
[45] F. Lotti, P. Tiezzi, G. Vassura, L. Biagiotti, G. Palli, and C. Melchiorri,
238–243.
“Development of UB Hand 3: Early results,” in Proceedings of the 2005
[72] Z. Xu, T. Deyle, and C. Kemp, “1000 Trials: An empirically validated
IEEE International Conference on Robotics and Automation, April 2005,
end effector that robustlygrasps objects from the floor,” in Proceedings of
pp. 4488–4493.
the 2009 IEEE International Conference on Robotics and Automation.,
[46] C. Lovchik and M. Diftler, “The Robonaut hand: a dexterous robot hand May 2009, pp. 2160–2167.
for space,” in Proceedings of the 1999 IEEE International Conference [73] I. Yamano and T. Maeno, “Five-fingered robot hand using ultrasonic mo-
on Robotics and Automation., vol. 2, 1999, pp. 907–912. tors and elastic elements,” in Proceedings of the 2005IEEE International
[47] Motion Control, Inc., “The motion control ETD„” September 2009, Conference on Robotics and Automation., April 2005, pp. 2673–2678.
http://www.utaharm.com/. [74] F. E. Zajac, “Muscle coordination of movement: A perspective,” Journal
[48] T. Mouri, H. Kawasaki, Y. Keisuke, J. Takai, and S. Ito, “Anthropo- of Biomechanics, vol. 26, no. Supplement 1, pp. 109 – 124, 1993.
morphic robot hand: Gifu hand III,” Proc. Int. Conf. ICCAS, p. 1288,
2002.
[49] M. Nicolelis, “Actions from thoughts,” NATURE, vol. 409, no. 6818,
pp. 403–407, JAN 18 2001.
[50] D. Nishikawa, Y. Ishikawa, W. Yu, M. Maruishi, I. Watanabe, H. Yokoi,
Y. Mano, and Y. Kakazu, “On-line learning based emg prosthetic hand,”
Electrophysiology and Kinesiology, 2000.
[51] T. Okada, “Object handling system for manual industry,” IEEE Trans-
actions on Systems, Man and Cybernetics, vol. 9, pp. 79–89, 1979.
[52] OttoBock HealthCare, Inc., “Cable-controlled arm prostheses„” Septem-
ber 2009, http://www.ottobock.com.
[53] RTAI, “http://www.rtai.org,” 2005.
[54] J. K. Salisbury and J. J. Craig, “Articulated hands: Force control and
kinematic issues,” International Journal of Robotics Research, vol. 1,
1982.
[55] V. J. Santos and F. J. Valero-Cuevas, “Anatomical variability naturally
leads to multimodal distributions of denavit-hartenberg parameters for
the human thumb,” Engineering in Medicine and Biology Society, vol. 2,
pp. 1823– 1826, 2003.
[56] K. Sasaki, Y. Fujikake, H. Takahashi, and M. R. Cutkosky, “Manipu-
lation of an object using slip sensing tactile sensor,” in Proceedings of
the ICMA’98 - Advanced Mechatronics: First-Time-Right, 1998.