When Interval Research was founded the original
24 members of the research staff were challenged
by David Liddle to "create something
as different from the personal computer as
it was from the mainframe".
I thought about this, in a structural manner,
and concluded that the right way to proceed
was by systematically inverting a number
of the media theoretic elements of what a
personal computer (or work-station) was (circa
1993).
A work-station:
- Communicates to people via a two and a half
dimensional grid of semantic relations (i.e.,
the desktop metaphor).
- Has a paucity of input senses - a keyboard
and mouse.
- Is fixed in space - it sits on a desk.
- Constrains the body of the user to a cramped
posture and limited set of gestures.
- There is one work-station per user.
Doing structural inversion on these elements,
with the added heuristic to design to human
qualities and abilities that had been "left
out" of the user experience of computer
use, leads to:
- Communicate emotionally - facial expressions,
affective sound, bodily gestures.
- Use multiple sensory modalities that are
shared by humans - including vision, touch,
and hearing.
- Make something that can move around independently.
- Have it move/experience/act through the same
physical and social spaces as people
do.
- Make lots of them - and have them exhibit
flocking behavior.
I properly noticed that this design exercise,
carried through to its logical extreme, results
in a crowd of unruly puppies. Since I didn't
think Paul Allen wanted Interval Research
to be a puppy breeding farm, I backed off
a little, and decided to explore building
mobile robots, that communicated emotionally
with people and each other, and which existed
in the same realm of the senses as people
did. This was my entry into what is now known
as affective computing.
I argued against simulation of such, because
people would inevitably react differently
to something on a screen as opposed to an
entity with a body in space. Therefore building
real physical robots was deemed essential
in order to create a joined system where
people were the environment for the robots,
and vice versa, and we could ask such questions
of people as what is the robot feeling?
When the project was started, it was
in uncharted
territory - no one else was doing anything
similar. The patent search for the
background
intellectual property uncovered only
a few
remotely relevant pieces of intellectual
property - a flocking patent, a game
patent,
and a few others.
My grand strategic goal for the project
was
to do a strong design statement, that
would
produce design heuristics that others
could
emulate. Such a strong design statement
could
be subsequently relaxed from its initial
purity of expression to find use in
other
areas, such as adding affective communication
into workstations - in other words,
one would
not have to always make a full up robot
to
take advantage of the understandings
produced
by first building an emotionally communicating
robot.
|
John Pinto in the Aquarium lab controlling
the Mark One "Severed Head" |
We started by exploring what were the
minimal
elements needed to make a face that
people
would experience as having expressive
emotions.
We decided not to go for realism, but
for
what amounted to a line drawing in
aluminum
and steel. Mark Scheeff built what
we called
the Mark One "Severed Head" - a cube with six degrees of freedom for
two lips, eyes, and eyelids - each driven
by computer controlled servos. The Interval
"Aquarium" staff (our experimental
psychologists) did user testing with naive
subjects who were directed to use a computer
keyboard to drive the face to achieve various
emotions - happy, sad, angry, afraid, etc.
We used the resulting settings for the servos
to build a database of emotional control
points.
Then we proceeded to build a full up robot.
The end result, four years later, was a prototype
robot that had an expressive face, a mobile
body with a articulated neck and head; gestures
that emulated the ballistic motions of animal
gestures; a sense of touch that differentiated
between a caress, a light contact, and a
blow (using sub-modalities of an accelerometer
and a capacitive sensor); stereo color vision
that located peoples bodies in space and
tracked them via 3D blob analysis; had male
or female gendered emotionally expressive
audio - and implemented an internal emotional
state machine that emulated the seven basic
expressive emotions described by Charles
Darwin in in The Expression of the Emotions in Man and
Animals.
By 1998 others had started to work in the
same general area. Some of these poured far
more resources into their engineering - notably
Sony with the Aibo robotic dog. In the annual
project review I argued that we should either
spin the project out of Interval as a development
effort aimed at real products in the toy
market with enough resources to succeed,
or we should kill the project. In the end,
I killed my own dog - the project was ended.
Mark Scheeff, the member of the research
staff who did the mechanical design for the
robot, followed on with a six month "cremator
project" He ripped all of the AI out
of the system, and turned the robot into
a tele-operated puppet, to explore what people
perceived about the robot, not knowing what
it was "under the hood" - i.e,
what they projected onto the media surface.
He exhibited it at the Tech Museum in San
Jose, and members of the project under his
leadership published a paper summarizing
the results from this exploration of projective
intelligence:
Proceedings of the Workshop on Interactive
Robotics and Entertainment (WIRE-2000): Experiences with Sparky, a Social Robot, Mark Scheeff, John Pinto, Kris Rahardja,
Scott Snibbe, Rob Tow .
The other tangible result from the project
was a very broad patent (and a subsequent
improvement patent with the same title) covering
emotional communication between real robots
with actual bodies with each other and with
people:
- Tow, R. Affect-based Robot Communication Methods
and Systems - U.S. Patent No 5,832,189, November, 1998.
- Tow, R. Affect-based Robot Communication Methods
and Systems - U.S. Patent No 6038493, March, 2000. This
was a continuation of the first robot patent
above, emphasizing claims related to actual
bodies in space as opposed to simulated bodies.
|