IEEE Robotics & Automation Magazine - June 2021 - 35

implementation constraints) to maximize the dexterous
workspace, considering kinematic feasibility, friction stability,
and singularities. This optimization predicted a large
range of motion in all six DoF, including the xy translation
space, where the physical prototype was most challenged.
The primary issue with the design optimization was that it
assessed static stability at each individual pose but did not
consider transitioning from one to another. So, while the
grasped object might be kinematically and frictionally stable
at poses A and B individually, the optimization does not
account for the forces required to push the object between A
and B. That is, the grasp would not be able to react the necessary
forces required to reconfigure the differential
between A and B even if the hand configurations are statically
stable at the two poses.
This gap in optimization is more pronounced in the xy
workspace because the differential is required to reconfigure
the most while executing xy motions. For z translations and
pure rotations, the angle between the palm and fingers
remains relatively constant, or the change in angle is similar
across all fingers, and thus the differential does not need to
reconfigure significantly, meaning that the hand can perform
manipulations in these DoF. One design choice for applying
the necessary manipulation forces might be to actuate each
finger with a separate grasp actuator instead of a differential
as implemented in this prototype. Unless these actuators are
purely torque controlled, such a design would surrender the
benefits of underactuated grasping and necessitate a priori
knowledge of the object geometry on top of adding to the
overall device complexity and cost.
The optimized hand's ranges of motion on YCB and foam
objects with the first design iteration in [13] for both the
translation and rotation axes are comparable on similarly
sized objects. However, the optimized hand is 42% as large as
the first design. Normalized by these dimensions, such as the
palm radius and the linear actuator stroke lengths, the design
parameter optimization yields a significant improvement in
the manipulation ranges of motion relative to size. This also
highlights the potential scalability of the device. Depending
on the application in which the Stewart Hand is deployed, the
hand dimensions can be proportioned to grasp and manipulate
larger or smaller objects.
The multi-DoF workspace exploration experiment
extended the single-axis characterization analysis to
motions more akin to real-world tasks. The dome-shaped,
kinematically feasible workspace of an equivalent Stewart-
Gough platform was simulated to sample points for the
exploration. Kinematic stability was chosen over the stricter
frictional stability requirements discussed in the " Theoretical
Workspace Modeling " section for sampling the workspace
to further evaluate the stability calculations employed
in the design optimization. Comparing the error and slip
metrics of the explored points within the predicted statically
stable workspace against those outside it, we can observe
that the points deemed frictionally unstable indeed have
higher error and slip values. The nonconvex kinematic
workspace was sampled for points approachable only by a
linear trajectory from the hand's home position. As a result,
a portion of the hand's workspace was not explored-particularly
points with significantly negative z coordinates and
high absolute xy values-without a more complex pathplanning
algorithm. Most of these points lie outside the predicted
frictionally stable workspace, and, by extension, we
can expect that these unexplored points would also have
high error and slip metrics. Future work will explore methods
for generating nonlinear trajectories with the hand for
manipulation sequences to travel through only statically feasible
points and more finely explore the entirety of the
hand's statically stable workspace.
The teleoperated experiment with the WAM robot demonstrated
that the hand is capable of performing dexterous
manipulation of unknown objects in a real-world application.
The test was human-operated and thus closed-loop, validating
the promise of improved hand performance with a
closed-loop controller that compensates for slippage and
errors between desired and actual poses. Future work on this
IEEE Robotics & Automation Magazine - June 2021

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2021
