A highly successful
collaborative
effort in robotic technology between NRL and university
and
industry partners won accolades from fellow researchers,
garnered technical achievement awards and brought media and public
attention to advances in the field at the annual conference of
the American Association for Artificial Intelligence (AAAI),
held recently in Edmonton, Canada. NRL, Carnegie Mellon University
(CMU), Northwestern University, Swarthmore College and Metrica
TRACLabs demonstrated how a collection of hardware, software
and sensors could work together to exhibit human social
skills.
AAAI annually holds
robot competitions in several
categories, allowing teams to demonstrate
best attempts at
solving common tasks in a competitive environment.
In a
competition called the Mobile Robot Challenge the NRL-CMU
team
entered "Grace" (short for Graduate Robot Attending
ConferencE), a six-foot drum-shaped robot with a digitally animated
face and a female persona.
Grace's assignment
was to autonomously
navigate her way from the lobby of the conference
hotel to the
AAAI conference registration counter, sign in, find
her way to
the elevators leading to the conference rooms, go
to the podium
and deliver a talk about herself. Along the way,
Grace would
engage in casual conversation with conference attendees,
saying
things like, "Could you please tell me how to get
to the
registration desk." Grace could then take directions
and
then reach intermediate goals such as needing to get to the
elevators by following those directions.
The contest objective,
first
announced at AAAI's 1999 Mobile Robot Challenge, was to
raise
the bar for intelligent robot behavior by creating "social
robots." In what conference organizers envisioned as a 10-year
effort, such machines would be capable of interacting with people
on human terms, not only engaging in conversations but also knowing
how to conform to social etiquette, like standing patiently in
a line at a registration desk.
The difficult issue
this year was the
integration. "A lot of research issues
come out of trying
to get very different systems to work together,"
says Alan
Schultz, the head of NRL's Intelligent Systems section,
and
leader of the NRL team.
Grace completed her
task in little
under an hour without assistance from her human
team.
Periodically, she did have difficulty understanding people,
which the researchers attributed to the noisy environment at
the conference. But throughout, Grace remained patient and
persistent,
and polite, with only one small glitch. When
getting in line
to register for the conference, Grace
interpreted a small space
between people as the end of the line
and inadvertently cut her
way into the line.
Built without
"arms,"
which would have required the addition of
mechanical engineers
to the team, Grace had to depend on the
kindness of strangers
to accomplish tasks such as opening
doors, pushing elevator buttons
or pinning on her conference
badge. "Could you pin the badge
to me?" Grace asked
the person behind the registration desk.
"I'm afraid I'm
all out of hands."
After negotiating
her way to the
conference room, Grace gave a talk entitled, "How
I Spent
My Summer Vacation," where she "spoke,"
using a
generated voice, about her software and the integration
effort.
At the close of her talk, Grace thanked the audience
and
quipped, "I'm going on vacation. See you in Acapulco."
Acapulco is the site of the next AAAI conference, where the research
team hopes to address more of the research issues, including
giving Grace the ability to answer questions posed by the
audience.
Carnegie Mellon developed
the software for mobility,
getting on and off elevators, standing
in lines. NRL's Navy
Center for Applied Research in Artificial
Intelligence
developed the algorithms for natural language understanding,
integration of speech and gestures, and for dialogue generation
and intelligence for interacting with people to find its way
to
the registration desk. Swarthmore developed the ability for
the
robot to find the registration sign, and Northwestern University
developed the system that allowed Grace to present her lecture.
TRACLabs developed a vision system for the NRL to use for
recognizing
gestures and for tracking the presence of
humans.
Schultz notes, "Although
attending a conference
does not sound like a hard task to a human,
it has many hard
research issues of importance to the Navy. Most
everyone has
heard about unmanned vehicles in the news. The problem
is that
currently, it can take anywhere from three to ten humans
to
control a single unmanned air vehicle, for example. We are
trying to make the vehicles more intelligent, with more autonomy,
and make the interfaces more natural so that a single soldier
can work with many vehicles."
The NRL contribution
is the
result of research funded by the Office of Naval Research
and
DARPA IPTO office.
The U.S. Naval Research Laboratory is the Navy's full-spectrum corporate laboratory, conducting a broadly based multidisciplinary program of scientific research and advanced technological development. The Laboratory, with a total complement of nearly 2,500 personnel, is located in southwest Washington, D.C., with other major sites at the Stennis Space Center, Miss., and Monterey, Calif. NRL has served the Navy and the nation for over 85 years and continues to meet the complex technological challenges of today's world. For more information, visit the NRL homepage or join the conversation on Twitter, Facebook, and YouTube.
Comment policy: We hope to receive submissions from all viewpoints, but we ask that all participants agree to the Department of Defense Social Media User Agreement. All comments are reviewed before being posted.