Stanford Cart
By Les Earnest
August 2005
The Stanford Cart has had a 45 year career of ups and
downs. It was born as a research platform for studying the problem of
controlling a Moon rover from Earth. It then was reconfigured as a robot
vehicle for research in visual navigation, then went into show business for a few
years. It now resides in a home for retired robots at the
Stanford Cart with cable, 1961
1960-61 -
The Stanford Cart was originally constructed by Mechanical Engineering (ME)
graduate student James L. Adams to support his research on the problem of
controlling a remote vehicle using video information. He had been working at
the Jet Propulsion Laboratory on a NASA project called Project Prospector,
which was proceeding with the assumption that someone on earth could zoom
around the Moon using a TV camera on a vehicle and a radio control link However
Adams showed that assumption to be false.
The
Cart had four small bicycle wheels with electric motors powered by a car
battery and carried a television camera with a fixed view in the forward
direction. Tests were conducted using both 2-wheel steering, like a car, and 4-wheel
steering, in which the wheels and television camera swivel together. The cart
was connected by a very long cable to a control console with a television
display and controls for steering and speed. A magnetic tape loop made it
possible to vary the time delay of steering commands, to simulate communication
delays.
Stanford Cart with radio links, 1963
1962-63 -
Mechanical Engineering graduate student Paul W. Braisted devised a scheme to
improve the controllability of the vehicle by adding an analog computer that
functioned as a predictor that took into account preceding steering commands
and put a bright dot on the television screen at the predicted location of the
cart when a current steering command would begin to take effect. With this
addition the vehicle could be controlled at 5 mph (8 kph). Still there was a
fundamental limitation on teleoperation in that if the travel during the time
delay is greater than the distance from the vehicle to an unseen obstacle there
is no way to avoid hitting it. Braisted completed his dissertation in 1963.
However,
the immediate prospect of applying this technology was put off as a result of
President John F. Kennedy's announcement on
Stanford Cart at SAIL
1964-71 – The
cart evidently sat unused in an ME laboratory until 1966 when Les Earnest, a
senior research scientist who had recently joined the Stanford Artificial
Intelligence Lab (SAIL), came across it and talked its creator, James Adams,
into letting SAIL use it. Earnest wanted to try navigating on the road around
SAIL under computer control using the center line and other visual references.
However the radio links and other electronics that had existed earlier had
vanished, so he recruited Electrical Engineering PhD student Rodney Schmidt to
built a low power television transmitter and radio control link and undertake
the visual guidance project.
SAIL
was granted an experimental TV license by the Federal Communications Commission
for Channels 22 and 23 and experimental operation began with a human operator
controlling the cart via the computer based on television images. Prof. John
McCarthy became interested in the project at this time and, as Director of
SAIL, took over its supervision. Using the KA10 processor, which ran at about
0.65 MIPS, Schmidt was eventually able to get the cart to automatically follow
a high contrast white line under controlled lighting conditions at a speed of
about 0.8 mph (1.3 kph). Schmidt completed his dissertation in 1971.
Stanford Cart with slider, 1979
1971-80 -
PhD candidate Bruce Baumgart and a few other graduate students experimented
with the cart before moving on to other thesis topics. The cart was changed from
4-wheel to 2-wheel steering during this period. Hans Moravec, who had come to
Stanford specifically to work on visual navigation, stayed with it but suffered
a setback in October 1973 when the cart
toppled off an exit ramp while under manual control and ended up with battery
acid throughout its electronics.
Moravec
was able to enlist the aid of roboticist Victor Scheinman in 1977 to build a
“slider,” a mechanical swivel that moved the television camera from side to
side allowing multiple views to be obtained without moving the cart. Using the
KL10 processor then available, which ran at about 2.5 MIPS, Moravec was
eventually able to use binocular vision to navigate slowly around obstacles in
a controlled environment. The cart moved in one meter spurts punctuated be ten
to fifteen minute pauses for image processing and route planning. In 1979, the
cart successfully crossed a chair-filled room without human intervention in
about five hours. Moravec completed his dissertation in 1980. There is a longer
article about this research as well as a short
video of the cart in action.
1980-2000 – After
SAIL was shut down in 1980 the cart again went into storage until 1987 when, at
the request of the
2000-present
After the
Successor
The modern SAIL, under the direction of Sebastian
Thrun, developed a robot vehicle called Stanley, which in
2005 won the DARPA Challenge, a race across the Nevada desert.
Acknowledgement
Thanks to James Adams, Bruce
Baumgart, Hans Moravec, and Oliver Strimpel for providing information or
reviewing earlier drafts of this account.
References
The following Ph.D.
dissertations at
[1] Adams, James Lowell, Remote control with long transmission delays,
PhD in Mechanical Engineering, 1961.
[2] Braisted, Paul Wilder, Study of a predictor for remote control systems
operating with signal transmission delays, PhD in Mechanical Engineering,
1963.
[3] Schmidt, Rodney Albert, Jr., A study of the real-time control of a computer-driven vehicle, PhD in Elecetrical Engineering, 1971.
[4] Moravec, Hans Peter, Obstacle avoidance and navigation in the real
world by a seeing robot rover, PhD in Computer Science, 1980.