Extraordinary claims. Ordinary investigations.

Do Roombas Dream of Electric Sheep?


Release a Roomba in a room, shut down the lights and set the camera for a half hour long exposure. The end result is the beautiful image captured by SignalTheorist.com recording the path of the little cleaning robot through all the floor.

The artificial intelligence behind the navigation of the vacuum cleaner is somewhat secret, but HowStuffWorks describes in general terms how it works, from the initial measuring of the room size to the spiral and the seemly random vacuuming.


Besides its simple beauty, the image is also spectacular because it relates to the early age of robotics and artificial intelligence. In particular to the “electronic turtles” created by William Grey Walter in 1948, more than six decades ago.

Those were his “Machina Speculatrix”, named Elmer (ELectroMEchanical Robot) and Elsie (Electromechanical Light-Sensitive robot with Internal and External stability), the former seen below with Walter:


Walter’s artificial turtles are among the first autonomous electronic robots in history. As such, they are also among the simplest. What looks like a small head was in fact a photoelectric sensor, but different from real turtle’s heads, it was constantly spinning. Something like ‘The Exorcist’, but with no puking. Keep reading to know more about these adorable turtles and how they were “possessed” by free will, consciousness and intelligence. As Walter described them, of course.

Self-awareness and free will

The turtle’s brains was ingenious in its simplicity. The spinning light sensor was connected to the motors, in a circuit that activated them to make the turtle go towards the light. Not always, though: when the light was too strong, the circuit reversed things, so that the robots wouldn’t just stop in front of lights. The end result was that they usually walked around lamps, in circles.

Besides the photoelectric sensor, the turtles had a mechanical switch fixed to the shell, which allowed them to sense when they hit something. In this case, they ignored the light command and ran in an oscillating fashion, which usually let them get over obstacles.

The turtles’ electronic brains could hardly be simpler. Below, a version of the circuitry:


Those are less than ten electronic components, around 20 if you count the resistors! Only two vacuum tubes, those ancient things equivalent to modern transistors. Any electronic device you may have around you has more components, from remote controls to a telephone. The mouse in your hand has hundreds, thousands times those components. The computer in front of you is million times more complex than Elsie or Elmer’s brains.

Nevertheless the behavior of the artificial creatures was surprisingly complex, even unpredictable. Here we finally get to the modern Roombas, because Grey Walter recorded the robots movements with long exposures.


Looks familiar? The luminous zigzag records the turtle’s path, starting on the left. Note how it couldn’t “see” the candle directly, going to the obstacle, where it spent some time jiggling around. After overcoming the obstacle, it saw the light and ran towards it, but didn’t get too close, circling it instead.

All with a few electronic components.

Walter was very found of his turtles, interpreting very complex features in them. Placed in front of a mirror, for instance, and with their own little pilot lamp than turned on when the motors ran, the robots started to shake as they detected their own light. “It began flickering, twittering, and jigging like a clumsy Narcissus“, he wrote. If it were seen in an animal it “might be accepted as evidence of some degree of self-awareness.”


Elsie and Elmer also exhibited social behavior, as their movement, pilot lamps and sensors interacted. They head straight to each other, and do their dance, recorded in the upper part of the image:

image006In the meantime a hutch was opened below, with its bright light and the precious energy to recharge the batteries. Another of the creature’s simple behavior, when energy ran low, the negative phototropism of the turtles was lifted and they ran straight to the light. When recharged, the repulsion to strong light went back to work and they returned to wandering around.

All these complex and unpredictable behaviors led Walter to suggest the artificial machines had some degree of free will, even conveying “the impression of having goals, independence and spontaneity”. Self-awareness, free will, independence, spontaneity… it’s clear he had great hopes for the future of his Machina Speculatrix. Because he was, in fact, attempting to reproduce our own brains: he was first a respected neurophysiologist already noted for his work with electroencephalograms (EEGs).

Rise of the Roombas

Fortunately or not, sixty years later we know that even the Roombas, venerable descendants of Grey Walter’s turtles, are far from conquering the world. In fact, despite having million times more electronic components and sophisticated sensors, the modern robotic vacuum cleaners are not that much more complex in their behavior than Elsie or Elmer.

In the early days of robotics and artificial intelligence the best minds in the field underestimated the complexity of their own intelligence, believing that “electronic brains” equivalent to ours were just in the next corner. As it turns out, that was mostly wishful thinking.


Nevertheless, asking if Roombas dream is not just a cheap pun with PK Dick’s famous novel. Did Elsie really have free will, self-awareness and independence? Does Google or WolframAlpha think? Are they “intelligent”? The question is literally philosophic, as it can be found in much discussion over almost three decades about the Chinese Room Argument. Presented in 1980 by John Searle, in the speculative scenario a person is locked inside a room and is given a Chinese script. Following a huge manual, in plain English, the person is capable of returning a batch of Chinese symbols to the outside world which are, by the way, a fine answer for the original script. All of this without at any moment understanding a single word of Chinese, oblivious to the question or the answer. Similarly, computers, Roombas and electronic turtles may give the impression of intelligence and free will, but those are at most just simulations of the real thing. Spiritualists like this idea.

Back to Grey Walter and his interest on the human intelligence. As a neurophysiologist with contributions to EEGs, he also explored our own free will and in 1963 he conducted experiments with amazing consequences which would only become more widely known years later through Benjamin Libet. Decades before, Walter had already shown that long before we think we decided something with our precious free will, electric signals of such decision are already detectable in our brains. Our free will is not that free, nor is it “willed” when and how we think it is, emerging instead from a complex pattern of unconscious processes. Perhaps we all have many little people who doesn’t understand English inside our brains, whilst our brain is still fully capable of “understanding” Shakespeare?

While the pioneers of artificial intelligence underestimated the complexity of human intelligence, and the failure of Roombas to enslave humanity is a good evidence for it, Grey Walter’s insights into our brains suggest we shouldn’t overestimate the workings of our grey matter either.

After his tortoises showing how complex behavior can emerge from just two vacuum tubes and a simple yet ingenious circuitry, Machina speculatrix, Walter went on to create robots capable of learning, Machina docilis, which could be trained like Pavlov’s dogs and had similarly simple and ingenious electronic brains. They “learned”.

His robots were not taken very seriously by fellow colleagues at the time, taken as novelty toys, with attention inversely proportional to the popular interest on such artificial intelligence. Curiously, this is a case where the public got it right. Walter’s robots and ideas were much more than toys, as today’s Roombas and sophisticated search engines may be much more than cool gadgets. May we speculate?

– – –

Further reading
Imitation of Life: A History of the First Robots
The Grey Walter Picture Archive
Robot Critters
Machina Speculatrix
The Timing Experiments of Libet and Grey Walter (PDF)
Libet’s Experiments and Free-Will: Implications (PDF)

Blog Widget by LinkWithin

Popularity: 2% [?]

Posted in People,Science | 3 comments

3 Comments so far

  1. Bodrix May 29th, 2009 9:56 pm

    What’s a ‘roomba’?

  2. Mori May 29th, 2009 11:24 pm

    A roomba is a robotic vacuum cleaner… click the link about its navigation on HowStuffWorks (beggining of the text) for more.

  3. Yerrt Rethguals June 6th, 2009 10:33 pm

    I think this shows just how early we are in the robotics development.

    When I look at the time exposed picture, what jumps out at me, is how much waste this technology produces.

    I would like to see a map showing how many times each square inch was vacuumed, as more than two times is a waste.

    To me, a better cleaning model would be to start in the middle…. then go forward until an obstruction is hit… turn left or right… and follow alonge the obstruction (wall or chair or ? )… mapping out a parimeter until the first obstruction point is revisited.

    Then the robot should adjust by 9/10ths the width of the cleaning surface, towards the center of that perimeter, so as to have a 10% overlap, and then follow that original course, adjusting on each revolution until it has run out of uncovered area.

    Further programming would be needed to have it recall any obstructions, and make an attempt from all sides to get at all areas within the perimeter.

    In the end, the robot would have a map in it’s memory, as to the shape and size of the room, and what obstructions were encountered… and where. It could refer to this memory in future attempts, and recognise where it is, or obstruction it encountered before.

Leave a reply

Live Comment Preview