IEEE Robotics & Automation Magazine - December 2015 - 128

Table 2. The proposed system level tests
to support testing the functionality.
System Test
Method

Description

Sensor-based
grasp efficiency

A measure of the ability to maintain
an efficient grasp on an object while
adapting to external forces.

In-hand
manipulation

The positioning accuracy and range of
motion when manipulating an object
within a grasp.

Hand stiffness

The stiffness properties of a grasp
inducing external forces and
measuring associated displacements
on a grasped object.

Finger force
tracking

The performance of fingertip force
tracking (see the "Finger Force Tracking" section).

Force calibration

The accuracy of calibrated sensors
in determining contact force
magnitude and directions (see the
"Force Calibration" section).

of palms, fingers, and parts under grasp as well as locations of
points of contact. Building test methods from this fundamental point of view will lead to relevant performance capture and
will span from lower-level capabilities, including primitive
sensing and control, to higher-level capabilities, including
touch-based manipulation and perception.
Presented below are descriptions for a subset of the metric
and test method pairs listed in Tables 1 and 2 with the accompanying experimental implementations and results. Those not
covered are still under development. Experiments were conducted for touch sensitivity, finger strength, grasp strength, slip
resistance, force tracking, and sensor calibration test methods
using three robotic hand configurations with different mechanical design, sensing, and control paradigms. More
specifically, two hands were used called Hand 1 and Hand 2,

Fcontact

Vjoint

Force Sensor
Figure 3. The robotic finger is commanded to close on an object
that is attached to a reference force sensor.

128

IEEE ROBOTICS & AUTOMATION MAGAZINE

DECEMBER 2015

where Hand 1 has the capability to support two exchangeable
touch sensory suites (impedance or resistance-based touch sensing), and Hand 2 incorporated current-based contact sensing at
the motor drives. Furthermore, Hand 1 has three 13.54-cm long
fingers and 7 DOF, whereas Hand 2 has three 13.5-cm long fingers and 4 DOF. Certain commercial equipment, instruments,
or materials are identified in this article to foster understanding.
Such identification does not imply recommendation or endorsement by NIST, nor does it imply that the materials or equipment
identified are necessarily the best available for the purpose.
Some of these experiments show a stationary robot supporting
the hand under test, where in all cases the hand could have been
supported using a static fixture instead. The testing of additional
robotic hand technologies will be included in these efforts to
ensure that the benchmarks under development support the full
spectrum of evolving robotic hand designs. In addition, some of
the benchmarks will also be applicable to fingerless gripping
technologies. Six tests were selected to provide concrete examples for both component-level and system-level tests, and are
meant to serve as an example for the development of further test
methods in either category. There is no particular reason for the
reported tests to be characteristically kinetic other than being a
direct result of the natural progression of our investigation.
Touch Sensitivity
Metrics and Test Methods
Touch sensitivity is a kinetic measure of the smallest self-registered contact force exerted by a robotic finger on an object.
The significance of this trait revolves around the hand's ability
to delicately interact with minimal disturbance to the immediate environment as well as detect small force perturbations.
Direct applications would include part acquisition with object
location or shape uncertainties as well as touch-based grasp
planning. This characteristic is a function of the hand's sensor
capabilities, motion controllers, bandwidth, joint speed, finger
size, and finger-object configuration.
To accurately capture the performance of a hand in this category, a dynamic test is needed. Of the previously listed dependence, only the joint speed and finger-object configuration are
assumed controllable. In particular, the robotic finger is commanded to close on an object that is attached to a reference
force sensor such that forces are measured before, during, and
after finger-object contact (see Figure 3). A reference force sensor capable of resolving forces in three dimensions is suggested
to make a more accurate capture of the full contact force. Once
contact is detected by the hand, the active finger is commanded
to stop its motion. To reduce the performance search space,
only the worst-case finger-object configuration was investigated. Specifically, the finger is commanded to close at a specified base-joint speed, V joint, while any remaining joints are controlled such that finger extension is preserved. The base-joint is
the first joint in the finger kinematic linkage. Thus, fingertip-
object collision occurs with full-finger extension, which maximizes fingertip Cartesian velocity and yields more aggressive
finger-object impacts. Meanwhile, by commanding different

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2015