robot has character
Eric Smalley and Susanna Space,
Technology Research News
Combine some of the most advanced human-computer
interaction technology with one of the oldest forms of entertainment --
puppetry -- and you get Horatio Beardsley.
Doc Beardsley is an animatronic robot, a descendant of the mechanical
humans and beasts that rang bells and performed other actions as parts
of the clocks of medieval European cathedrals. Modern science, however,
has carried Doc far beyond these ancient automata, endowing him with the
ability to see, understand spoken words and carry on a conversation.
Researchers at Carnegie Mellon University made the amusing, forgetful
inventor as a literal embodiment of a computer interface. Doc performs
for audiences, answering questions about himself. He claims to have been
born on a mountaintop in Austria to a family of goatherds, and to have
invented endless bread, the milkbed, the antisnooze and the foon.
In addition to paving the way for a future generation of theme park animatronic
characters, the technology could lead to embodied personal digital assistants
with personalities, interactive electronic pets, animated historical museum
figures, and robotic waiters and salespeople, said Ron Weaver, a graduate
student at Carnegie Mellon.
Several layers of software drive Doc's apparent wit. Synthetic interview
software, which includes speech recognition abilities, allows Doc to react
to spoken questions. The technology, developed at Carnegie Mellon for
use with video characters, gives a character sets of lines to deliver
on given topics. This allows Doc to give appropriate answers to questions
that match an anticipated query closely enough, said Weaver.
If the question hasn't been anticipated, another layer of software takes
over. A discussion engine tracks the questions and answers during a conversation
and allows Doc to make relevant comments by keying off individual words
even if he doesn't understand a specific question, said Weaver. And if
that doesn't work, the discussion engine tosses the conversation back
to the questioner, he said.
"The discussion engine will first try to deliver a comment that is still
relevant based on whatever individual keywords can be found in the text,"
said Weaver. "Failing that, the character gives a random comment that
either pretends to reflect what is being discussed to try to keep [the
conversation] going, or transfers the onus of the conversation back to
The researchers made Doc forgetful to help keep his lapses from breaking
the illusion of a realistic personality. Here's an actual exchange:
Questioner: "How are you feeling today?"
Doc: (pause) "My-- my dog just died. No, I'm just kidding. He just got
run over. He's still alive. Well, it's a she really. I just refer to her
as a boy to bust her chops, keep her honest, you know?"
Questioner: "What's your best invention?"
Doc: "Well, you know, there was this one invention I made accidentally
while working on genetic recombination. I crossed a spoon with a fork
and thus created the foon. It was used either as a spoon or a fork."
Microphones and cameras that track the sounds and movements of an audience
provide Doc's audiovisual perception, said Todd Camill, a research engineer
at Carnegie Mellon University's Robotics Institute. "Microphones listen
for sounds made by people in the room, and a camera subsystem tracks the
movement of people by finding areas of skin tone in the room. The audio
and vision systems generate position data that tell Doc where to turn
One aspect of making Doc Beardsley a believable character is keeping the
technology in a supporting role, according to Tim Eck, another Carnegie
Mellon graduate student. "Character and story are the most important aspects
to creating believable, entertaining characters," he said. "We are striving
to provide the illusion of life, to create an entertaining experience,
which is an important distinction. We are not trying to create artificially
intelligent agents. We are creating the illusion of intelligence with
time-tested show business techniques: drama, comedy, timing and the climactic
As with many creative endeavors, serendipity plays an important role.
"From time to time, we find ourselves caught off guard by conversations
that seem to make sense in ways we did not intend," said Camill. "For
example we've recently heard this exchange:
Guest: 'Doc, why are you wearing a Carnegie Mellon University sweatshirt?'
Doc: 'I've spent time at many universities. You'd be surprised at the
things they throw away.'"
In addition to using traditional storytelling and theatrical techniques,
the researchers are studying the human side of human-computer interaction.
"Since our goal is the illusion of human intelligence or intent in the
service of a story, a large part of our results concern the human audience
rather than the robot," said Camill. "We are exploring the social dynamics
between human and machine by exploiting the tendency of people to project
human qualities on the objects around them."
From the entertainment perspective, the ultimate goal is creating synthetic
characters that seem to possess dramatic human qualities, like a sense
of humor, comic timing, personal motivations and improvisation, said Camill.
"When an audience can get so engrossed in interacting with Doc's dialogue
and story that the technology is completely forgotten, then we know we
have accomplished our goal," he said.
The next steps in the project are improving the character by adding skin
and a costume, building a set and props, creating a show, building puppeteering
controls for the props, and writing software for producing other shows,
The technology is not yet ready for the entertainment industry, said Eck.
"The main reason [is] speech recognition technology. We believe once the
overall accuracy of speaker-independent speech recognition is 80 percent
or higher, applications such as ours will be seen in the entertainment
industry. This will be approximately 5 to 8 years from now," he said.
The research is funded by the Carnegie Mellon University.
Timeline: 5-8 years
TRN Categories: Robotics; Artificial Intelligence
Story Type: News
Related Elements: Project website: http://micheaux.etc.cmu.edu/~iai/web/newIAI/doc.html
Atomic cascade broadens
gain quantum afterburner
Nanotubes branch out
Research News Roundup
Research Watch blog
View from the High Ground Q&A
How It Works
News | Blog
Buy an ad link