Signal Processing - September 2017 - 9

GEORGIA TECH

FIGURE 1. The test version of a display interface developed by Georgia Tech researchers that promises to allow nonexpert users to easily and effectively
control a robot through a web browser.

consensus method. "We use the
plane to calculate an orientation
that's perpendicular to the region of
interest," Kent says.
■ The algorithm uses clustering, based
on Euclidean distance between
points, to identify individual compo-
nents in the region of interest. "We
use principal component analysis to
extract a principal direction, or
essentially an orientation, for each
cluster," Kent explains.
■ Information calculated from the pre-
vious point cloud processing-the
plane normal and the principal
direction-is used to inform heuris-
tics that ranks a set of grasp hypoth-
eses calculated from AGILE (http://
wiki.ros.org/agile_grasp), prioritiz-
ing grasps that align with the plane
normal and principal direction.
■ The best grasps are presented to the
operator, who can then select a grasp
for the robot to execute in a supervi-
sory manner.
A control interface designed for non-
expert users must be as foolproof as pos-
sible. "We have to make sure our system
can safely execute any commands given
by a user, which involves careful design
and thorough testing," he says.
Kent notes that the most significant
remaining technical challenge the team
faces involves the point clouds. "They're
constructed by observing a surface from

a certain viewing angle and, as such,
they are missing data due to occlusion,"
he says. "We're essentially trying to
extract useful information from a signal
that's missing a lot of data from the real-
world object it is trying to represent,
and, on top of that, the data is poten-
tially noisy, depending on the reflec-
tiveness and opacity of the objects in
the environment."
After testing college students on
both existing ring-and-arrow systems
and the new interface, the research-
ers discovered that the point-and-click
method created significantly fewer
errors, allowing participants to perform
tasks more quickly and reliably. Overall,
the point-and-click approach resulted in
approximately one mistake per task,
compared to nearly four per task for
the ring-and-arrow interface. The study
also found that point-and-click users
could perform a specific task about
2 min faster than when using the tradi-
tional interface.
The project's next step is deploying
an online version of the interface that
will allow authorized Internet users to
remotely control a robot and teach it
complex tasks.

Skin deep
The stereotypical robot is a rigid metal
machine capable of tolerating extreme
environments and external forces. Yet,
IEEE SIGNAL PROCESSING MAGAZINE

September 2017

in many situations, it's actually more
beneficial to build a robot with a flexible,
skin-like exterior that's embedded with
multiple sensors. Such a robot, like Co-
mau's recently introduced AURA (www
.comau.com/EN/media/news/2016/06/
aura), can act on its current environment
or adjust its activities to meet a spe-
cific physical command, such as being
tapped by a human to immediately stop
or nudged to move a certain way or per-
form a specific action.
Covering a robot with flexible sen-
sors requires a technology that's both
cost-effective and easy to manufac-
ture in bulk quantities. Researchers at
Massachusetts Institute of Technol-
ogy's (MIT's) Computer Science and
Artificial Intelligence Laboratory be-
lieve that three-dimensional (3-D) print-
ing promises the best approach to this
challenge. To demonstrate the feasibility
of a flexible robot that can interpret and
act on local conditions, the researchers
have designed and built a device that
responds to external mechanical stress-
es by changing the color of a spot on
its surface.
The device was inspired by the golden
tortoise beetle, or "goldbug." Native
to the Americas, the insect's exterior is
usually a golden tone. Yet the insect
quickly switches to a reddish orange hue
if poked or prodded or, technically speak-
ing, mechanically stressed.
9

http://www.comau.com/EN/media/news/2016/06/ http://http:// http://wiki.ros.org/agile_grasp

Table of Contents for the Digital Edition of Signal Processing - September 2017