What were Grey Walter's robot tortoises, and why do they matter?

In the late 1940s, visitors to a brain research institute in Bristol could watch two odd little machines trundle across the floor. They looked like tortoises under clear plastic shells, they moved slowly, and they seemed to have minds of their own. They would wander, poke into corners, back away from obstacles, and when their batteries ran low, head home to recharge. Their builder, William Grey Walter, called them Elmer and Elsie.

They are among the first autonomous robots, meaning machines that sense their surroundings and decide how to move without a person steering them. That alone makes them worth knowing about. What makes them important is the point Walter used them to make: complicated-looking behavior can come from a remarkably small number of parts.

Who built them, and where

Grey Walter was a neurophysiologist, a scientist who studies how the nervous system works. He was known for his work with the electroencephalograph, or EEG, the machine that records electrical activity in the brain. He worked at the Burden Neurological Institute in Bristol, England, a center founded in 1939 that combined laboratory research with treating patients.

According to the Grey Walter archive kept by the University of the West of England, he built Elmer and Elsie between Easter 1948 and Christmas 1949. The two shared similar electronics but had different shells and motors. He showed them in public in 1949 and 1950, though they were temperamental and needed frequent repairs.

In 1951 his technician, W. J. "Bunny" Warren, built six sturdier tortoises. Three of them appeared at the Festival of Britain that year, in the science exhibition, where the robot was billed under the mock-Latin species name Machina speculatrix, roughly "the machine that explores."

How a tortoise worked

Each tortoise sat on three wheels, like a tricycle. Motors drove it forward and turned its steering. Its senses were minimal: a light-sensitive cell that could detect brightness, and a contact switch that registered when the shell bumped into something. A battery rode at the back, and a plastic shell covered everything.

The "brain" was an analog circuit built around just two vacuum tubes, the glass electronic valves used before transistors. Analog means the circuit worked with smoothly varying electrical signals rather than with the 0s and 1s of a digital computer. Out of that circuit came a few simple tendencies:

None of those rules says "explore the room" or "go home when tired." Yet watching the robot, that is exactly what it seemed to do. Walter built a small hutch with a lamp on top. A tortoise would roam the room for a while, then, as its battery weakened, turn toward the hutch, go in to recharge, and set off again in search of new lights.

He tried other experiments too. When he fixed a small lamp to each tortoise's shell and let two of them loose together, they circled and approached each other in a way that looked oddly social, a kind of dance of attraction and retreat. He also built a version that could learn a simple association, linking the sound of a whistle with light, much as Pavlov's dogs linked a bell with food. He called that one Machina docilis, the "tameable" machine.

Why the tortoises mattered

Walter's aim was scientific. He was convinced that even creatures with extremely simple nervous systems could behave in complex and surprising ways, and the tortoises were his demonstration. With the equivalent of two "nerve cells," a machine could behave in ways that never repeated exactly and that looked spontaneous and goal-directed.

That made the tortoises a landmark for cybernetics, the field that studies control and communication in animals and machines, which Norbert Wiener and Warren McCulloch were shaping in the United States in the same years. Walter is often counted among its founders. The tortoises also drew a great deal of press attention, and some people even wrote asking to adopt Elmer and Elsie as pets.

The original Elmer and Elsie did not survive; they were not seen after 1951 and are believed to have been scrapped. One of the 1951 tortoises was rediscovered in 1995 and restored, and replicas were built that year using original parts. The Science Museum in London has a tortoise in its collection, and another survives at the Smithsonian Institution.

What can a simple robot like a tortoise teach us about AI today?

The first lesson is that behavior is not the same as intelligence inside the machine. A tortoise had no map, no memory of the room and no plan, yet observers described it as curious, hungry or shy. When a modern chatbot or voice assistant seems thoughtful, it is worth remembering how readily people read minds into simple mechanisms.

The second lesson is that a body in the real world can do some of the thinking. Much of the tortoise's apparently clever behavior came from the interaction between its few reflexes and its surroundings: walls, lamps, slopes and the hutch. Robot designers still use this idea. Floor-cleaning robots and planetary rovers such as NASA's Sojourner rely heavily on sensors and simple reactive rules rather than on detailed reasoning about every move.

The third lesson is about where to start. Walter did not try to build a whole mind. He built the smallest system that could show something interesting, studied it closely, and then added one capability at a time, such as learning. That habit of building a tiny working version first remains good advice for anyone experimenting with AI in 2019.