Bill Tilden robots
Originally published in SMITHSONIAN (February 2000)
by Paul Trachtman
(Editor's note: An obscure philosopher once noted that "Science serves two purposes: to examine and explain the inexplicable; and to suggest the yet-to-be." At times these two directions converge, particularly when something is created that functions despite a lack of understanding how exactly it operates. Tunnel diodes are just one example.
In this excerpted story, an engineer at Los Alamos National Laboratories has created a new approach to robotics that has completely contravened the "accepted practice" in design. Despite naysayers, he continues building units that: a) are easily replicated and b) function. He is regularly castigated or, worse, ignored by academics in AI and robotics.
And yet these robots are functioning.... RM)
The sign taped to the door reads: "Please Don't Feed The Robots." Mark Tilden unlocks the door and leads me into his Los Alamos lab. He sets his laptop case on a table and walks straight to the barred windows where sunlight filters in on a large sandbox he calls his "Robot Jurassic Park." The white sand is full of tracks, and alive with several species of little solar-powered "bio-mechanical" insects. He watches as they sun themselves, wake up for short, random walks, and then doze off again to soak up more energy. A wire-legged bug, soldered together from tiny transistors, sensors and motors, with a solar cell stuck to its back, begins walking in one direction, then suddenly veers off in another, catching Tilden's eye. "When it makes a decision like that, what has just gone on inside its head?" he asks. "As the builder, I have to admit that even I don't know."
At Los Alamos National Laboratory, in a cluster of research warrens, scientists with their heads in clouds of mathematics, physics and space sciences can spend their days deep in thought with their digital counterparts, the supercomputers that Norbert Wiener, a pioneer of cybernetics, once called "the cold, hard medium of logic." Tilden's medium is more like magic. His lab looks like an electronic boneyard of dead Walkmans, VCRs and tape decks, all of them salvaged, dissected and given new life as enigmatic machines, robots that seem to possess small minds of their own. But what kind of minds?
Tilden calls himself a "robobiologist," and watching the ingenuity, independence and true grit with which his critters get around in the world, it's easy to see why he calls them "living machines." When you pick one up and feel it scratching at your arm, you almost forget the metal. It feels more like an angry bug than a mechanical device that may someday be colonizing Mars or the Moon, or finding and blowing up land mines here on Earth.
The first thing he wants you to know is that his robots have nothing to do with computers. "Ninety-eight percent of all species don't even have a brain to speak of," he says. "Neither do my creatures. They think the way the body thinks, and for most living things, the body is everything." When they're given a purpose, such as seeking light, he explains, they try to accomplish that purpose any way they can. To illustrate, he grabs an insectlike robot in mid-stride, yanks a couple of its four wiry legs completely out of shape, and sets it down again. The mangled robot keeps going, dragging itself forward on its bent stumps and two good legs -- looking painfully like a big grasshopper that some kid has tortured.
Solar Walkman 2.1, running since 1995
Tilden is pleased with this. A digital robot, animated by computer programs rather than transistors, would have immense trouble doing this. "Inside a computer is a perfect world," he says, "but the universe is a noisy place. We stay coherent in the noise, but computers can't think like that. They are always just a bit flip or a wire cut away from total failure." Instead, he is working with more chaotic, wavelike "analog" circuits. In another demonstration, a solar-powered walking robot, turned loose in a field behind the lab, gets through the grass and ditches, the stones taller than itself, and the weeds snagging its antennae, without any programming from Tilden. "The machine has a natural ability to that by itself," he says. "If it was a digital machine, I would have to program in every possible condition based upon the environment."
Elsewhere at Los Alamos, a mathematician, a physicist and their supercomputer are analyzing data gathered from Tilden's robots, trying to figure out how they think. Most roboticists, with an abiding faith in the power of computers to conquer new worlds, are skeptical of Tilden's approach. In the Church of Artificial Intelligence he is a heretic. But the Department of Defense is keenly interested. Says Bill Warren, who is the DARPA funding officer supporting the research at Los Alamos, "Most people want to dismiss this because: a) it's too simple, and b) they don't understand it. That's part of the fun of this thing; gee, it's so simple, and it does something useful -- like find land mines! It's a beautiful machine." Another Tilden backer, John Dockery, is a physicist who became a top military analyst for the Office of the Joint Chiefs of Staff. He comments, "I regard Tilden as a pioneer who is so fundamental that people don't relate to him. You have to judge this work by what it may develop into, versus what it's doing now."
Back in his lab, Tilden pulls out a steely black, 10-in. long, battery-powered "ant" with its eyes recessed in protruding metal cylinders, a unicorn antenna thrust forward, and four angular legs splayed out from its circuit-board underbelly. "This is Walkman 1.5," he says, "the creature that wound up proving so much." As he grips it in his hand and powers it on, Walkman's legs start flailing in the air. "Watch this," he says, putting it down on the desk. Sensing terra firma, Walkman stops flailing, takes a few erratic, uncoordinated steps, then starts walking with a curious, high-stepping gait, as if awakened in a world it doesn't quite trust or understand. It picks its way over the electronic debris on the table, but stops as its antenna pokes a big metal box. Alternating flashes of red from its eye sockets throw light on the problem. Walkman assesses the light bounced back to its sensors and decides to back off. Then it crabs sideways for a while and starts forward again. Success!
Tilden then grabs up and sets Walkman back down, this time with a hind leg in the center hole of a thick roll of tape. As the robot senses that it's caught, we can see it trying out different ways to free itself. But how does it know what to try? "It knows," Tilden says, "because the loads, or stresses, on its legs disturb the analog wave patterns in its circuit that allow it to move, producing new and different behaviors until it finds one using the least energy." Suddenly I see the trapped leg is lifting a lot higher, and taking a longer time with each lift, while the free hind leg is making lots of quick steps, scratching at the desktop. And then in one graceful high arc, the hind leg clears the rim of the tape without even touching. Walkman takes off without missing a step, but Tilden grabs it again and sets it back in the trap.
The smallest autonomous robot, Bitman, runs on two watch motors
This time the robot lifts its leg once, twice, and on the next try it's free. "When it finds a solution and I put it right back into the situation, it gets out faster two out of three times," Tilden says. "More often than not it remembers. Not always. Sometimes it forgets." Walkman's "nervous system" doesn't seem complicated enough to have a memory, but damned if it doesn't have one anyway. It can remember because electrical waves have inertia, a certain energy that fades away only gradually. So the waves hold onto their shape even after the stimulus ceases or changes, and that's the equivalent of a very short memory.
When Tilden tries to explain what makes Walkman tick, it's easy to sympathize with skeptics. But Tilden's analog world is an adventure populated by creatures that seem to reveal hidden truths about the nature of our own world. "If I show you something that walks around and I say, look, 12 transistors, I can draw you the circuit and you can do it yourself, then there's no magic happening here," Tilden says. He corrects himself. "I'm sorry, there is magic, but it's not a magic you can't understand."
At a blackboard, Tilden shows me a simple arrangement of transistors wired together in a loop that is connected to motors and sensors. This is Walkman's neuroanatomy. The elements may be few and simple, but each one is linked to others in an overlapping pattern of interconnections. There is a potential both for chaos and for regular periodic patterns in this circuit, Tilden points out, which allows Walkman to adapt to its environment. "The world completes the robot's architecture," he says. "Without the world, it wouldn't do anything very interesting."
Within the core of the circuit, Tilden explains, the transistors are oscillating. A robot's different walking behaviors will then result from different rhythms, as the oscillations fall into sync. At the heart of Walkman's abilities is a two-transistor neuron invented and patented by Tilden. This "nervous net" neuron is essentially a loop of oscillators that emphasize the chaotic elements of electrical signals. "Think of it as a disco with four bad drummers at the corners of the dance floor," says Tilden. "Each one's going bash, bash, bash, and ultimately they start listening to each other, because it's a small disco. And after a while they will find certain beats and form a song."
"Now the signals you're generating are not just on-off, on-off; they're bash-bash, bash-bash, and what happens is, after a while you find people discovering backbeats and harmonics and all kinds of stuff. That's the best analogy I can come up with. Imagine the different sorts of ways they fall into different sorts of patterns. Then you get a glimpse of what it means to be a nervous net."
"Each creature has its own intrinsic songs," he explains. "And when you take these creatures and put them under stress, guess what happens to the nervous nets? They fight to hold onto their song harder. If you push them too much, they'll go into a different song."
Tilden's disco is a very nonlinear place, and so are his nervous nets. "If you use neuron nets in a linear way, you get nothing of facility or ability," he says. "You get million-dollar robots that can barely walk around in a box. Every animal on this planet thinks in analog. Nothing in the universe thinks in digital except for the computers we've built. And computers can't live off the desktop. We use neurons in a chaotic way; the firing of these neurons matches the complexity of the world."
As for his robots, he continues, while their neurons are nonlinear elements, their sensor and motors are strictly linear devices and must be so, in order to keep the robots in a stable state. The sensors give a precise report of the world, measuring load on the motors, pressure, heat, light, or any other elements the robot's chaotic brain must know about to survive. "When you are a goal-seeking creature, you have to have a brain which is inherently chaotic, but quelled by linear motors and sensors that act as its regulators.
"And of course, this is the ultimate heresy," he goes on. "Because all our psychology, psychiatry, artificial intelligence, many studies of consciousness, are based on the concept that we are rational animals. No! We are a solid core of pure chaos bounded by linear systems keeping us regulated toward some level of cohesiveness with our world. We are chaotic creatures who are made rational by our environment."
The Next Level
At his home in Los Alamos, Tilden introduces me to Strider, an erect "eosapien" who walks toward me with amazing grace. Strider is only a foot tall, with four long wire legs carrying its cylindrical body high off the ground, a stiff copper neck and a wide, flat head shaped from interlaced hexagons bejeweled with analog chips and blinking LED lights. Strider pauses, swivels its head from side to side for a while, and then decides to approach me.
Eosapiens like Strider became possible when Tilden discovered what he calls his "unicore" structure. "It's the smallest structure of neurons that gives the greatest range of stable behaviors," he says. "It gives the robot a 'neural net' brain that holds onto patterns of stimuli from the environment -- the short-term memory. This brain isn't a digital computer; it's another analog layer that holds onto electrical charges representing patterns of behavior. So a creature 'remembers' what it has done, enhancing its competence. This brain is connected to a 'nervous net' body that can hold onto rhythms. Non-unicore Walkman dances around rhythms very effectively, but it's limited. It can't evolve much further. With the unicore, you can actually remember, and move, and close that loop, so you have a structure which is pretty close to being an artificial life-form."
The secret of the unicore is the neck. "It took years to find that," Tilden says. "It wasn't enough to be a brain and a body, it's a brain connected to a body. The body has its own competence. The brain has its own awareness. The connection determines how one is mapped onto the other, bidirectionally.
"Imagine you're walking with your wife in one hand and your young son in the other," he explains. "He says slow down, she says speed up. She wants you to go to the store, but he's tired of walking all day. What do you do? You act as a mediator who effects a compromise between the two. You're telling your wife to slow down, but the information is also being transmitted to your son. He thinks, 'Oh, Daddy is trying to help me!' This sort of thing is what happens in these analog systems. Everybody knows what everybody's doing."
I am introduced to an experimental head for Nito, which may be the first robot with what Tilden and Brosl Hasslacher, a physicist, nonlinear theorist and friend at Los Alamos, call an "ego." "Nito takes the unicore structure to its next logical level," says Tilden. "When Hasslacher saw the first unicore, he asked me, 'What happens if you expand these things?' One way to do that, Brosl proposed, was to stack one unicore into another, so the first is walking in the world, but the second one is 'walking' in the other's mind, in its world representation. So now you have a brain on top of a brain that is not connected to the real world. At that point, you should have something that is aware of itself. That's why we call it an 'I' machine."
Defense analyst Dockery sees Nito as a creature of both history and the future. When the cybernetics field started off, there were passageways you could go along," he says, "and the field took off down one with great success. That was artificial intelligence. When that happens, other passages get stubbed off. What Tilden has done is go all the way back to the beginning and explore another passageway. So people's reaction is, 'It's been done before,' or 'It doesn't fit in the stream we're used to,' and they don't know what to do with it."
"But Tilden isn't waiting for them. I'm reminded of Indiana Jones. He just tackles problems and what he does works, and that's the criteria for him, and it's kind of an adventure. Now he's taken one robot and put it inside another, so that the output of one becomes the input to another. Suddenly you've got a thing which is both in touch with the environment and buffered from the environment. And I don't know what that is," says Dockery. "It's some kind of abstraction. You have the beginnings of a purposeful machine the like of which has not been anticipated. I don't yet know what that is."
For more information on autonomous robots at Los Alamos, connect directly at http://www.OneRS.net/102df-430
Tilden, one year later
The voice on the other end of the phone line sounds amused, yet intense. "Conventional feedback systems always have a straight-through drive component and a very subtle feedback component." It's Mark Tilden, adding comment to the preceding article. "One is the mantissa, one is the ampersand, and you get attenuation. The problem is you get a massive amount of drive and a very fragile sensor, whose information is key to operation. The only way to protect the sensor is to encapsulate it in the chassis or in a 'clean' environment. An alternate solution would be to use the motor itself as the measure." Tilden found, in 1988, a circuit that took the feedback energy and used it to regulate the motor's energy, via the circuit's own hysteresis. "It allowed me, subsequently, to create robots that I could literally put into a dishwasher, without concern for fragility. Since then, I've researched how to apply this technology to robots that could work in the real world."
According to Tilden, these circuits won't be useful in accurate SCARA robots in pick-and-place assembly, but you could use them to create adaptive fingers that could pick up an egg, without having to go to the computational extremes that are normally required. These devices are also robust enough for space applications. "Robots we built six years ago are still bopping around the Robot Jurassic Park today," he says. "Simple, elegant solutions are not as hard as you think."
Tilden's research into design and applications continues, along with involvement in the BEAM international robot games, where robots: a) must be built from scratch, b) must be autonomous, and c) are permitted to cheat to succeed. He believes that competitions can reawaken interest in engineering, ultimately producing not better robots, but better roboticists. Much about the competition, along with circuits and components, can be seen at the Solarbotics website http://www.OneRS.net/102df-431. Other projects include showing concepts at the January 2001 New York Toy Fair. And this summer, Tilden will release into New Mexico White Sands Missile Range, as both research and publicity, the largest contiguous robot colony -- a thousand units 12-in. long, 3-in. high, moving about at 20 cm/sec.
"And," he says with a laugh, "we might not get all of them back."