DXb\1;GD7DIH
Walter was very much a showman who enjoyed
the attention he got for his robots.
stability). The robots were called “tortoises”
because of the shell-like, bump-sensitive bodies
that covered them. He also called them Machina
speculatrix, Latin for “machine that speculates.”
Built with postwar military surplus, each electromechanical tortoise was three-wheeled, with a front
drive wheel and two unpowered rear wheels. Each
side of the bot carried a vacuum tube “neuron” and
a relay. Power to the robot was provided by a 3-cell
lead-acid accumulator.
An ingeniously designed touch-switch at the top
of a mast toward the back made the entire shell
touch-sensitive. Any pressure against the shell
would close the circuit and trigger an avoidance
sequence. The main active sensor was a photoelectric cell housed on a rotating spindle at the front of
the robot, above (and connected to) the drive wheel.
Once a light source was detected, a linear gear
would disengage and the drive wheel would be fixed
in the direction of the light.
By using the switching power of two relays, and
the amplifying (and oscillating) power of the two
vacuum tubes, a series of “behaviors” could be
orchestrated within the circuit:
Pattern E: Exploration. Moving around in a room
without a light source to follow
Pattern P: Positive phototropism. Moving toward
a light source
Pattern N: Negative phototropism. Moving away
from a strong light source
Pattern O: Obstacle avoidance. Pushing, turning,
and recoiling to push away and/or steer clear
of obstacles
Another important aspect of the design was what
Walter called “internal stability.” Today we might
call it autonomy — the ability of the bots to “feed”
themselves. A hutch was built with a charging
mechanism inside. As the tortoise’s battery died,
the gain on its tube amps would be decreased,
making it harder to engage behavior N (normally,
the brightness of the light would trigger pattern N)
and the bot would be attracted to and drive toward
an ultra-bright light in the back of the hutch. Once
the battery was fully charged, pattern N would be
excited again, and the tortoise would react to the
light by backing out of the hutch.
From 1948 into the early 50s, Walter did studies
with Elmer and Elsie. Six other tortoises were built
by BNI technician W. J. “Bunny” Warren. In 1950,
Walter published his findings in an article in Scientific
American entitled “An imitation of life.” This was
followed by a second article in Scientific American,
entitled “A machine that learns,” in 1951, and by a
book, The Living Brain, in 1953.
In these publications, Walter reported that he’d
witnessed surprising behaviors in his interacting
tortoises. He claimed to have seen actions he was
unafraid to call “goal-seeking” (starting out in
the dark and making way toward a light source),
“recognition of self” (bots “performing” in front of a
mirror), “discernment” (making distinctions between
effective and ineffective behavior), and “free will”
(exhibiting unpredictable behavior and making
choices). He witnessed what we’d call “emergent
behavior” today. He saw things in the robots that
he didn’t design into them, and saw unanticipated
behaviors emerge from interactions between bots.
Ever the Showman
Most of the articles about Walter’s experiments,
and the information available online (much of it
inaccurate), take him at his word and don’t question
his findings or the effectiveness of his designs.
Professor Owen Holland, from the computer
sciences department of the University of Essex, built
two replicas of the tortoises (with Bunny Warren) and
has carefully studied Walter’s findings, photos of the
tortoise experiments, and such artifacts as the BBC
newsreel footage mentioned in this article’s opening
paragraph (see at makezine.com/19/greywalter).
Holland points out many inconsistencies in the
conclusions and the surviving evidence of the
experiments (such as the famous time-exposure
photographs of Elmer and Elsie taken in late 1949
or 1950), and obvious fudges and edits to the BBC
newsreel to, for instance, imply that Elsie has made
her own way into the charging hutch. There’s no
evidence (or even written claim) that the charging
hutch ever actually worked.
But even with some questionable conclusions,
the documented accomplishments of these experiments are impressive, and their basic observations
are borne out by today’s simple “behavior-based
robots” (BBR), such as BEAM robots (see MAKE,
70 Make: Volume 19
