Signal Processing - September 2017 - 8

SPECIAL REPORTS
John Edwards

Building the Foundation for "Generation Robot"
Signal processing is helping to make robots a part of everyday life

L

eaping off the pages of fiction and
movie screens, robots are poised to
create the most profoundly disruptive
technological shift since the Indus-
trial Revolution.
While robots have labored in nu-
merous industries for many years-
particularly in the automotive and
manufacturing sectors-experts predict
that a tipping point in robotic deploy-
ment is now at hand and that robots will
soon join the business and consumer
mainstream. To put it plainly, robots
are set to become an essential part of
everyday life, and life itself will never
be the same.
Market research firm Tractica pre-
dicts that the global robotics industry
will expand from US$34.1 billion in
2016 to US$226.2 billion by 2021, rep-
resenting a compound annual growth
rate (CAGR) of 46%. "The definition of
a robot is in flux and traditional robot
manufacturers that have been building
and supplying robots for decades have
seen this industry undergo a dramatic
transformation in the past few years,"
observes Tractica Research Director
Aditya Kaul.
As robots transition out of the fic-
tional world and into everyday life, sig-
nal processing is playing a major role in
their development. In key areas, such as
control, sensing, and modeling, signal
processing is helping robots become
our trusted new home, road, and work-
place partners.
Digital Object Identifier 10.1109/MSP.2017.2715599
Date of publication: 6 September 2017

8

Aiming for easier control
Most current-generation robots possess
only a limited amount of autonomy. In
many industrial, health care, military,
and even consumer applications, humans
still need a direct way of instructing
robots on how to move and act, as well
as to take full control of a robot when
things go awry. Yet existing robot con-
trol interfaces leave much to be desired.
Most require a computer and mouse to
independently control a range known
as the "six degrees of freedom," which
enables a rigid body to operate within a
three-dimensional space.
Yet, for most users, turning three
virtual rings and adjusting arrows on a
screen to move a robot into position to
grab items or perform a specific task
is as tricky as it sounds. For someone
who isn't an expert, the ring-and-arrow
system is cumbersome and error prone.
The approach certainly isn't ideal for
people who aren't technically skilled,
such as older people trying to control
assistive robots at home.
A new interface designed by Geor-
gia Institute of Technology researchers
promises simpler and more efficient
remote robot operation from almost any
location without requiring significant
training (Figure 1). The user just points
and clicks on an item and then chooses
an appropriate task grasp. "Instead
of a series of rotations, lowering and
raising arrows, adjusting the grip and
guessing the correct depth of field,
we've shortened the process to just two
clicks," says David Kent, the Georgia
Tech doctoral robotics student who
IEEE SIGNAL PROCESSING MAGAZINE

|

September 2017

|

led the project. Sonia Chernova, a Geor-
gia Tech assistant professor in robotics,
advised the research team.
Kent says that the new interface was
developed to allow nonexpert users to
effectively control a robot through their
web browser. "There are many potential
applications-basically anywhere you
would have a remote human operator
controlling a robot with a manipulator,
including controlling robots for space ex-
ploration, deploying robots in dangerous
environments, such as disaster response
and search-and-rescue, and controlling
assistive robots," he says. Kent notes that
besides being more intuitive for novice
operators, intelligent interfaces can po-
tentially reduce the workloads of experi-
enced operators.
"Signal processing, coupled with the
interface itself, is what makes our inter-
face more effective than other commonly
used robot teleoperation interfaces," Kent
states. "By analyzing and incorporating
information from a depth image, or a
point cloud, our interface can focus the
operator's interaction around relevant
commands for the robot."
The point-and-click control interface's
key element is an algorithm that calculates
and ranks grasps based on their effective-
ness for general manipulation. The pro-
cess consists of the following steps.
■ Based on an input point provided by
the user clicking on a camera feed, the
algorithm filters a point cloud of the
environment down to a small region
of interest.
■ The algorithm next fits a plane to the
region of interest using a sample
1053-5888/17©2017IEEE
Table of Contents for the Digital Edition of Signal Processing - September 2017
