Content uploaded by Muhammad E. Abdallah
Author content
All content in this area was uploaded by Muhammad E. Abdallah on Sep 03, 2014
Content may be subject to copyright.
Abstract— The second generation Robonaut hand has many
advantages over its predecessor. This mechatronic device is
more dexterous and has improved force control and sensing
giving it the capability to grasp and actuate a wider range of
tools. It can achieve higher peak forces at higher speeds than
the original. Developed as part of a partnership between
General Motors and NASA, the hand is designed to more
closely approximate a human hand. Having a more
anthropomorphic design allows the hand to attain a larger set
of useful grasps for working with human interfaces. Key to the
hand’s improved performance is the use of lower friction drive
elements and a redistribution of components from the hand to
the forearm, permitting more sensing in the fingers and palm
where it is most important. The following describes the design,
mechanical/electrical integration, and control features of the
hand. Lessons learned during the development and initial
operations along with planned refinements to make it more
effective are presented.
I. INTRODUCTION
ASA, working with its partner General Motors,
developed the Robonaut 2 (R2) hand as a device that
could work with a wide range of human interfaces, going
beyond the original mandate of using tools designed for
Space Walking Astronauts [1]. GM and NASA share the
vision of humans and robots working together, with the
safety of their people being the overriding priority. In
teaming with NASA, GM continued a long standing effort to
adapt robotics technology to help their employees perform
tasks which currently are ergonomically difficult. From
NASA’s perspective the GM collaboration was a way to
combine resources and expertise to produce a better
Robonaut and a much more capable robot hand.
Many advances in dexterous hand technology have been
made since the original Robonaut hand was developed in the
late 1990s. These are evident in several of the hands
developed over roughly the last decade. The three fingered
HRP-3 hand provides a significant grasping capability to an
impressive walking humanoid [2]. The SSL hand pursues
compatibility with a large number of tools through a balance
between simplicity and dexterity [3]. The most recent DLR
design has moved from an intrinsic to an extrinsic design [4]
adding to a series of impressive and capable designs [5],[6].
Like any attempt to recreate the subtleties of human
capabilities, in this case, a human hand, this is a work in
progress. Important improvements for the R2 hand vs. the
Robonaut 1 hand include increasing the thumb Degrees of
Freedom (DoF), overall joint travel, wire reduction and
durability. The following describes: design philosophy,
mechanism design, sensing, avionics, control, application,
lessons learned and future development.
II. DESIGN PHILOSOPHY
The R2 hand and forearm, shown in Figure 1, are
designed to improve upon the approximation of human hand
capabilities achieved by its predecessor, Robonaut 1 [1].
The five fingered dexterous hand and forearm is a
completely self-contained unit, all motors and avionics are
packaged inside the forearm, and only 6 conductors for
power and communication from the upper-arm is needed, a
huge reduction from the more than 80 conductors of
Robonaut 1. This makes the R2 hand an important
subsystem of the full humanoid robot and a modular,
extremely dexterous, stand-alone end-effector in its own
right.
Fig 1. The Robonaut 2 hand and forearm with all avionics
The modularity of the hand continues to the fingers as
each of the fingers is treated as an individual sub-assembly.
The actuator location and design is also modular allowing
for rapid replacement of components and sub-assemblies.
Further, the forearm is designed with a quick disconnect
from the upper arm allowing for ease of maintenance and
assembly.
Simulation is an important tool in evaluating candidate
designs. The Cutkosky's grasp taxonomy [7] is an excellent
benchmark for evaluating dexterity. While the Robonaut 1
hand could only emulate about 50% of these grasps, the R2
hand design is successful over 90% of the taxonomy [8].
Fig. 2 shows the actual R2 hand emulating 15 of the 16
Cutkosky grasps.
The Robonaut 2 Hand – Designed To Do Work With Tools
L. B. Bridgwater*, C. A. Ihrke**, M. A. Diftler*, M. E. Abdallah**, N. A. Radford*, J. M. Rogers*,
S. Yayathi*, R .S. Askew*, D. M. Linn **
*NASA/JSC, Houston, Texas
**General Motors, Warren, Michigan,
N
Fig. 2. R2 hand demonstrating the Cutkosky grasp taxonomy
III. MECHANISM DETAILS
A. Overview
The R2 hand and forearm is 127 mm in diameter at its
largest and 304 mm long from the base to center of the palm,
and has a payload of more than 9 kg. The hand has 12 DoF
with a two DoF wrist. Sixteen finger actuators and two wrist
actuators in the forearm control the 14 DoF. The fingers can
exert a tip force of 2.25 kg while fully extended and a tip
speed of more than 200 mm/sec.
The R2 Hand is designed to be comparable in size to a
human hand. Hand dimensions are within 60th to 85th
percentile human male [9]. Many design features of the
fingers and finger actuation system enable the compact
human size of R2’s hand while retaining the strength
required to work with human tools.
B. Finger design
A human finger is generally considered to have four DoF.
R2’s primary fingers contain four joints, only three of which
are independently controllable DoF, configured in such a
way as to effectively approximate the poses achievable by
the human finger (Fig. 3). The fourth human DoF that is not
controllable in this approximation is the distal
interphalangeal joint (J4). Rather than independent control,
the angle of J4 is a fixed relationship to the proximal
interphalangeal joint (J3). The angle is determined by the
output of a four-bar linkage consisting of the proximal,
medial, and distal phalanges, and a distal linkage is
illustrated in Fig. 4.
Fig. 3. R2 primary finger with labels and location for each axis
Fig. 4. The angle of J4 dictated by the four bar linkage and the angle at J3
R2’s secondary fingers are mechanically similar to its
primary fingers except the adduction/ abduction of the finger
(J1) has been eliminated. Of the remaining three joints, two
are loosely coupled, while the third remains linked as
described above. The tendons that control the secondary
fingers are grounded to the finger at the medial phalange;
therefore they produce torque at both the metacarpal
interphalangeal joint (J2) and J3.
Application of the “N+1” Rule – Minimizing the number
of tendons minimizes the space required to accommodate
actuation [10]. An N DoF tendon actuated manipulator can
be fully controlled by a minimum of N+ 1 tendons. For the
primary fingers, i.e. the index and middle fingers, three DoF
are fully controlled by four tendons. The secondary fingers,
which contain two DoF, are controlled by only two tendons.
The secondary fingers are therefore under-actuated and used
primarily for grasping, as opposed to dexterous
manipulation.
Individual joint travel limits of the fingers are designed to
approximate those of a human finger, such that a wide range
of comparable poses are achievable. Table I shows the joint
ranges of the fingers. Table I
Joint range of travel for the R2 fingers
Finger DoF
Range of travel
J1 (Only on primary fingers)
-20° to +20°
J2
-10° to +95°
J3
0° to +120°
J4
0° to +70°
C. Thumb design
A human thumb is accurately approximated with five
independently controllable degrees of freedom [11]. The R2
thumb contains 4 phalanges and 4 independently
controllable DoF which effectively approximate the posses
achievable by the human thumb.
Fig. 5. R2 thumb with labels and location of each axis
The fifth DoF of the human model was the angular twist
between the axes of the second and third joints, shown as a
dashed arrow in Fig. 5. This DoF has been replaced with a
ridgid link with a fixed angular twist between J2 and J3.
The angle selected was a design trade to most accurately
approximate the human thumb opposing to the fingers and
accommodate tendon routing. The ranges of motion in the
thumb are shown in Table II.
Table II
Joint range of travel for the R2 thumb
Thumb DoF
Range of travel
J1
0° to +74°
J2
0° to +85°
J3
0° to +90°
J4
-10° to +70°
The thumb also implements an N+1 configuration for
control of its four DoF, the selection and routing of the five
tendons is chosen to maximize the thumbs ability to oppose
the grip-force generated by the four opposing fingers of the
hand. Four of the five tendons cross the first two DoF on the
flexor side while a single tendon is antagonistic and is
shown in Fig. 6.
Fig. 6: Tendon routing through the R2 thumb
D. Finger Actuators
R2’s finger actuation system is a key enabler for compact
human size of the fingers of the robotic hand. The
components of the actuation system include the actuator
(consisting of the motor, gear head and ball screw
assembly), tendon, conduit, tension sensor (described
below), and terminator (Fig. 7).
Fig. 7. Actuator system of the R2 hand
The actuator (Fig. 8) is designed for a linear travel of 35
mm and to provide a pulling force of 23 kg. These design
values are based on the desired joint ranges and torques
required for R2’s fingers and thumb.
Fig. 8: R2 finger actuator
The tendons used for finger actuation in R2’s hand must
be mechanically anchored to the bone upon which they exert
forces. The space used for this termination is an important
consideration. In a new termination method [12], an
incompressible ball is placed inside the weave of a hollow-
weave braided polymer tendon. The tendon and ball are then
placed inside a small cylinder, which captures the ball and
lets the tendon pass through the ends of the cylinder. The
cylinder contains a feature through which the ball is too
large to pass. The assembly is then potted with adhesive to
prevent slippage of the ball inside the tendon within the
range of design tension loads. This design (Fig. 9) allows for
a single continuous tendon to be anchored in the middle,
where either end may exert forces on the bone as an
antagonistic pair of tendons and reduces the number of
terminators needed for each finger. Pairs of tendons in R2’s
fingers are terminated in a space of 3.1 mm diameter by 5
mm length.
(a)
(b)
Fig. 9: Tendon terminator (a) assembled and (b)in cros section
Tendon material – Through extensive testing Vectran™
was selected for the base tendon material for its strength and
resistance to stretch and creep. Its abrasion resistance
properties were improved by creating a hybrid weave of
Teflon™ with Vectran™. Tendon material of nominal
diameter 1.2 mm with break strength of 181 kg, was selected
as an optimal balance between size, strength, and abrasion
life for the R2 hand.
E. Wrist Design
The R2 hand is mounted to the forearm by a universal
joint that approximates the human wrist in range of travel.
The axes are orthogonal to the forearm roll axis, with the
pitch joint proximal and yaw joint distal to the forearm. The
axes are separated by 6 mm. The u-joint is attached to the
forearm with two shock mounts to absorb impacts as the
hand interacts with its environment.
The center of the universal joint is open such that the 16
finger actuator conduits may pass through, in addition to one
conduit used to protect wires that must cross the wrist. This
routing differs from the human anatomy, but was selected to
minimize coupling between finger and wrist motions.
Additionally, conduits are layered in a manner that avoids
kinking and entanglement during operation.
F. Wrist Actuator
The R2 wrist is a closed kinematic chain, differentially
actuated by two compact linear actuators (Fig. 10) located
on the dorsal side of the forearm. These actuators mount to
the main forearm bone and follow the tapered shape of the
forearm. Each actuator consists of a custom brushless DC
motor with a hollow shaft which drives the nut of a ball
screw. The screw extends and retracts a slider which drives
a ball and socket linkage connected to the palm. Located at
the rear, a magnetic incremental encoder tracks the position
of the actuator. The wrist actuators can exert a force of 27
kg and have a travel of 100 mm.
Fig. 10. Wrist actuator assembly
The wrist actuators are attached to the palm through two
ball and socket joints on either end of a stainless steel two-
force member. The sockets are equipped with bronze cups
backed by rubber inserts to allow for favorable wear and
shock absorption. The same ball and socket is used on both
the actuator and palm sides of the link to ensure
commonality of parts and simplicity of manufacturing.
IV. SENSING
There are three types of sensors in the fingers, thumb and
wrist:
A. Absolute joint position sensing
The absolute angular position of each joint is measured by
means of a hall-effect sensor along with a novel “ellipsoidal
shaped” magnet [13]. The magnet shape was designed to
achieve a linear relationship between angular position and
the change in the magnetic field that is measured by the hall-
effect read sensor. A circular magnet generates a sinusoidal
signal, whereas the novel shape used in R2’s fingers
generates an approximately linear signal over a 150° usable
range. Fig. 11 shows the magnetic field for the ellipsoidal
shaped magnet as compared with a ring magnet.
Fig. 11. Comparative magnetic field of the ellipsoidal magnet Vs ring
magnet
B. Tactile load sensing
The proximal, medial, and distal phalanges of the fingers
and the medial and distal phalanges of the thumb are each
designed to accommodate a phalange tactile load cell, which
is a novel six degree of freedom force torque sensor [14].
This load cell evolved from a low profile design (Fig. 12)
designed for the base phalange of each finger into an arch
that is compatible with all the phalange locations. Each load
cell utilizes eight pairs of semiconductor strain gages
mounted to an aluminum elastic element. The aluminum
strain element is designed to maximize measurable bending
strain within the range of design loads, and within the
limited space available on the phalanges of the R2 hand.
Hard stops redirect the load after approximately 2.2 kg of
force or 113 mNm of torque to ensure the strain element
does exceed its elastic limit.
(a)
(b)
Fig. 12. Evolution of the phalange sensor (a) initial concept, (b) as
implemented
C. Tendon tension sensors
R2’s finger actuation system accommodates sensing of the
tension in the tendons by means of a sensor that measures
compressive forces in the conduits in which the tendons are
routed [15]. It has been found that the measured
compressive force in the conduit is equal to the tensile load
on the tendon within an error of 5-10%. The tendon tension
sensors (Fig. 13) are protected from incidental contact from
conduits and objects in the hand by their installation into the
structure of the palm.
Fig. 13. Anatomy of the tendon tension sensors
V. AVIONICS AND CONTROLLERS
The forearm contains all the avionics and drive electronics
needed for hand operation.
A. Motor Drivers
Each motor driver “Trident” board (Fig. 14) consists of a
custom 3-axis hybrid motor driver, an Actel ProAsic3
FPGA, hall sensor feedback for motor commutation and
encoder accumulation, phase current sensing for each motor,
and digital temperature channels for the motors and hybrid
drivers. The Actel FPGA on each Trident communicates
with the main controller “Medusa” via a point-point serial
link. The Trident-Medusa comm. is fully managed in the
FPGA fabric and transfers information in parallel with the
higher-level Multi Drop Low Voltage Differential Signal
(MLVDS) communication. The Trident receives motor
PWM commands and provides feedback for the motor phase
currents, accumulated encoder position and temperature data
back to the Medusa. Current limit thresholds are internally
set to provide a level of protection for the motors, actuators,
and tendons.
Fig. 14. Location of Medusa and several of the tridents
B. Main controller
Consistent with the R2 communication system, the
MLVDS communication engine in the forearm allows the
R2 control system to command and sense all 18 forearm
actuators and 50 sensors via a two-wire communication link.
The MLVDS and distributed control strategy heavily
reduces the conductor count and processor load of the
central computer.
The Medusa in the forearm is responsible for
communicating with both the R2 brainstem as well as the six
Trident boards. The Medusa is primarily comprised of a
Xilinx Virtex 4-FX FPGA and a 16 channel analog to digital
converter. The MLVDS communication engine that
interfaces with the R2 control system is implemented on the
FPGA. In initial versions the Virtex 4 FPGA mainly served
as a communication distribution hub for the Trident motor
drivers, but more recently the onboard PowerPC has been
employed to perform all closed loop control for the hand
motion. The Medusa reads all finger position and tension
sensor data sent through the MLVDS from the hand control
processor. In addition the FPGA fabric communicates with
the six Trident motor drivers in parallel, and the on board
Analog to Digital Converter (ADC), which measures wrist
position hall sensors and internal voltages. The Medusa has
an SPI Flash memory for storing non-volatile parameters
such as control gains or sensor calibration. Available for
future use is an IO port for reading one-wire digital
temperature sensors.
C. Sensor Serialization
The FPGA device “Mini-Medusa” located on the dorsal
side of the palm controls the serialization of the hand sensor
data (Fig. 14). In order to maintain a low conductor count
and still process data from 50 different sensors comprised of
a total of 164 analog signals, serialization of the sensor chain
is necessary.
All of the phalange and angle sensor signals are
multiplexed and read by ADCs in the fingers communicate
back to the R2 control system through the Mini-Medusa.
Sixteen separate tendon tension sensors are mounted in the
palm and each one consists of two half bridges. These
signals are routed through a flexible circuit board on the
inside of the palm to a flat flex connector on the Palm ADC
circuit board. The Palm ADC then communicates its data via
a serial link to the Mini Medusa.
Packaging the large number of sensors found in the R2
hands was accomplished by utilizing a network of
distributed ADCs which digitize and serialize the data being
sent upstream. This in turn enables the use of small
connectors and localized wiring harnesses that make the
system extremely modular and serviceable
VI. CONTROL STRATEGY
A full discussion of this control law is available in [16].
The final control law is presented here, where the actuators
implement an inner position control loop.
t
K
W
R
pkpp p
T
dc
(1)
p and pc represent the actual and commanded positions for
the actuators, R (n × n+1) represents the kinematic mapping
from tendon tensions (f) to joint torques, W is a row matrix
selected orthogonal to R, t is a scalar measure of the internal
tension on the tendon network (t = Wf), and Kp and kd are
constant gains.
VII. APPLICATIONS
The hand’s capability to achieve a wide range of grasps
opens up a plethora of possible applications normally only
performed by humans. These range from very dexterous
tasks such as rotating a knob through active finger motion to
subtle grasps used to grip flexible materials (Fig. 15). As
learned with Robonaut 1[17], soft goods in the form of
compliant, low and high friction gloves are critical to
successfully achieve and maintain stable contact during
manipulation (Fig. 15).
(a)
(b)
Fig. 15. R2 hand interacting with (a) basic knob and (b) space blanket
The R2 hand can work with a variety of both terrestrial
and “space” tools. Fig. 16 shows the hand manipulating a
standard, but by no means lightweight, commercial drill.
The index finger has sufficient control to modulate drill
speed during bolt tightening. Fig. 16 shows two R2 hands
working together with a portable x-ray device used by
NASA geologists.
Fig. 16. R2 hand working with various devices, including (a) hand drill tool
(b) portable x-ray device.
VIII. LESSONS LEARNED
A key engineering development in the design of the R2
hand was the selection of the makeup of the braided polymer
tendons. SpectraTM was initially chosen as the base tendon
material for its high strength and favorable abrasion
resistance characteristics. Unfortunately, it was determined
early on that the creep inherent in SpectraTM cable is
incompatible with the R2 finger actuation system, which has
a finite travel of the ball screw actuator. Other high strength
braided polymer cables were investigated, and VectranTM
was identified as a promising candidate. However, the lower
abrasion resistance of VectranTM cable posed a significant
challenge.
To increase the abrasion resistance of the Vectran cable,
PTFE strands were added to the braid. Exhaustive testing
was performed to identify a combination of sliding surface
material and tendon lubricant that would maximize the
sliding life of the hybrid VectranTM PTFE braid. It was
concluded that leaded phosphor bronze along with Dupont
KrytoxTM lubricant significantly extended the abrasion life
of the cable vs. other combinations tested. Bronze was
subsequently incorporated into the sliding surfaces of the
finger design, and tendons are pre-lubricated with KrytoxTM
prior to installation. Finally, the diameter of the tendon was
increased by 50%, which increases the break strength to well
above the required level, but also contributes substantially to
abrasion life. The result is abrasion resistance of over 1
million cycles before failure in the experimental setup, an
improvement of approximately five fold over the initial
results.
IX. CONCLUSIONS/FUTURE DEVELOPMENT
Research and engineering development on elements of the
R2 hand are continuing in various phases of planning and
design. Future work includes adaptation of Series Elastic
Actuation, employed throughout Robonaut 2’s upper arms
and torso, for use in the lower arm wrist mechanism.
Refinements to the robots "skin" for better grasping are
being planned - improved "flesh," as well as tighter
integration of the robot's soft goods with its mechanisms are
design goals. Design concepts for additional passive degrees
of freedom in the fingers are close to realizing initial
prototypes. The need for smaller actuation, and for actuation
not limited by finite travel of the motor ball screw
arrangement, provide a wealth of opportunity for further
innovation.
In conclusion, the Robonaut 2 hand greatly improves upon
the capabilities of its predecessor in the areas of strength,
speed, sensing, and the ability to approximate human
grasps. NASA and GM are both actively identifying
applications where these capabilities can be exploited to the
benefit of their human workforce. Together the
organizations are also developing potential spinoffs that will
make use of the technologies advanced in the R2 hand for
applications outside the realm of traditional robotics.
ACKNOWLEDGMENT
This work represents the effort of a community of
researchers, and the authors would like to thank them for
their technical excellence which produced the R2 hand.
REFERENCES
[1] Lovchik, C.S. and Diftler, M.A. “The Robonaut Hand: a Dexterous
Robot Hand for Space.” Proceedings of the IEEE International
Conference on Robotics and Automation, Detroit, MI. pp. 907-912,
1999.
[2] Kaneko, K, et al. “Humanoid Robot HRP-3.” Proceedings of the
IEEE/RSJ International Conference on Intelligent Robots and
Systems, Nice, France, pp. 2471-2478, 2008
[3] Akin, D.L., et al. “Development of a Four-Fingered Dexterous
Robot End Effector For Space Operations.” Proceedings of the IEEE
International Conference on Robotics and Automation, Washington,
DC, pp. 2302-2308, 2002
[4] Grebenstein, M. et al, “Antagonistically Driven Finger Design for the
Anthropomorphic DLR Hand Arm System.” Proceedings of the IEEE-
RAS International Conference on Humanoid Robots, Nashville, TN,
USA, D. pp. 609-616, 2010.
[5] Haidacher, S, et al. “DLR Hand II: Hard- and Software Architecture
for Information Processing.” Proceedings of the IEEE International
Conference on Robotics and Automation, Taipei, Taiwan, pp. 684-
689, 2003
[6] Chen, Z., et al. “Experimental Study on Impedance Control for the
Five-Finger Dexterous Robot Hand DLR-HIT II.” Proceedings of the
IEEE/RSJ International Conference on Intelligent Robots and
Systems, Taipei, Taiwan, pp. 5867-5874, 2010
[7] Cutkosky, M. “On Grasp Choice, Grasp Models, and the Design
of Hands for Manufacturing Tasks.” Proc. of the IEEE Transactions
on Robotics and Automation, vol. 5, iss. 3, pp. 269–279. June 1989.
[8] Diftler, M.A., et al. “Robonaut 2- The First Humanoid Robot in
Space” Proceedings of the IEEE International Confrence of Robotics
and Automation, Shanghai China. pp 2178 - 2183, 2011.
[9] 1988 Anthropometric Survey of U.S. Army Personnel: Methods and
Summary Statistics,United States Army Natick Research,
Developmenet and Engineering Center, Natick Mass, September 1989
[10] Salisbury, J. K., & Mason, M. T., Robot Hands and the Mechanics of
Manipulation. MIT Press,Cambridge, MA, 1985.
[11] Chang, L. and Matsuoka, Y, “A Kinematic Thumb Model for the ACT
Hand.” Proceedings of the IEEE International Conference on
Robotics and Automation, Orlando, Florida, pp. 1000-1005, 2006
[12] Ihrke, C., et al. “Bidirectional Tendon Terminator.” US Patent
Application 12/269,579, May 2010.
[13] Ihrke, C., et al. “Robotic Thumb Assembly.” US Patent Application
12/564,085, Mar 2011.
[14] Ihrke, C., et al. “Phalange Tactile Load Cell.” US Patent 7,784,363,
August 2010.
[15] Abdallah, M., et al. “Sensing the Tendon Tension Through the
Conduit Reaction Forces.” US Patent Application 12/269,552, May
2010.
[16] Abdallah, M. Platt, R. Wampler, C. Hargrave, B.“Applied Joint-Space
Torque and Stiffness Control of Tendon-Driven Fingers,” IEEE-RAS
Intl. Conf. on Humanoid Robots, Nashville, TN 2010.
[17] Martin, T., et al “Tactile Gloves for Autonomous Grasping with the
NASA/ DARPA Robonaut”. Proceedings of the IEE International
Conference on Robotics and Automation, New Orleans, LA. pp.1713
– 1718 Vol.2 2004
