IT'S ALL IN YOUR SPINE, PART ONE

By Senior Associate Editor Samuel K. Moore

A Swiss engineer brings animal-inspired locomotion to robots and robot-inspired research to animals.

Samuel K. Moore

I knew I was on the right path when I passed a cramped two-seat car on the side of which was printed Asimov's first law of robotics: "A robot may not injure a human being or, through inaction, allow a human being to come to harm." The quotation was small comfort as the car began to follow me down the street, and I realized that neither of the two young men within, both comically too tall for the vehicle, were steering.

I was at the Ecole Polytechnique Federale du Lausanne (EPFL), in Switzerland, to see Auke Ijspeert, head of the Biologically Inspired Robotics Group. His research is of a different sort than the robotic car that was trailing me more closely than I would have liked. Instead, Ijspeert is devoted to reproducing in robots the kind of surprisingly robust locomotion control systems in animal spinal cords. These control systems, called central pattern generators (CPGs), govern the basic motion of everything from the lowly lamprey to the sophisticated feline, and they are a hot item in the non-humanoid robot world.

In animals, CPGs are either single neurons or interconnected clusters of neurons called oscillators that reside in the spinal cord of vertebrate animals and in what passes for the brains of more primitive animals, such as snails and lobsters. The output of a single oscillator is a sine wave of neural activity. But if you link oscillators together so that the output of one drives or inhibits the other, you can get all kinds of complicated behavior. Scientists have shown that such linked oscillators produce basic repetitive motions such as slithering in snakes and swimming in salamanders. It's likely that CPGs contribute to walking and other motions in humans, but the science at the moment is not so solid.

Photo: BIRG

ROBO-SALAMANDER: The Swiss robotic salamander demonstrates mechanisms of how simple neural signals control movement.

The CPGs, says Ijspeert, reduce the otherwise complicated task of getting around to a simple one. In a salamander, for instance, its brain needs only two channels of input to walk, swim, speed up, slow down, or turn. In work to be published Friday in Science, Ijspeert and his team used a robotic model of the central pattern generators in the salamander spinal cord to show how the critters might make the transition from swimming to walking. That's something evolutionary biologists are keen on knowing because it could point to how our ancestors first crawled out of the ocean, and it's something neuroscientists have yet to figure out. "For salamanders, we're actually ahead of the biology," he says. "Biologists don't know how the oscillators are coupled."

Ijspeert's robotic salamander is basically an 8-segment mechanized snake with a pair of rotating legs attached to the second and sixth segment. Each segment is governed by two oscillators. Essentially one is in charge of bending the segment to the left and the other in charge of bending it to the right. In addition, each leg has it's own oscillator. Each of the oscillators in the body segments can influence its nearest neighbors but not the leg oscillators. Each leg oscillator can influence the body oscillators on its side of the salamander, as well as those of the two nearest legs.

Salamander walking and swimming both involve its body making a sine wave, but there are important differences. For one, walking has a lower frequency—less than 1 Hz. Also, the walking wave is a standing wave. That is, the two segments with legs do not sway side to side, but the rest of the body segments do. Swimming, on the other hand, is a traveling wave—imagine an eel slithering through the water.

Ijspeert's salamander system shows that if you couple the oscillators together and weight their influence on each other in just the right way, the salamander-bot makes a smooth transition from walking to swimming. Amazingly, the transition happens just by ramping up the volume on the two inputs that represent what the salamder's brain is telling its spinal cord.

Photo: BIRG

MAKING WAVES: Salamandra robotica simulates the central pattern generators in animals that govern motion.

The salamnder-bot was set up so that the influence of the leg oscillators on the body oscillators is much stronger than the influence of the body oscillators on each other. Start ramping up the input signals and the leg oscillators force the robot into the standing wave walking gait. Increase the input and the salamander walks faster. But, noting that a salamander never wriggles as fast when walking as it does when swimming, the EPFL group rigged the leg oscillators to "saturate" at about 1 Hz. That is, after that frequency, they lose their ability to influence the body segments. At that point, the body oscillators are free to produce their own, faster, intrinsic rhythm, a traveling wave.

Ijspeert hypothesizes that when a salamander finds itself deep in the drink its brain hits the accelerator on the central pattern generator, automatically shifting the slithery little fellow into swimming mode. Back on dry land, it puts on the brakes, shifting into a swaggering stroll.

Besides the salamander, Ijspeert's group also produced a CPG-controlled snake and a robotic boxfish; but, tragically, the boxfish drowned during a power outage two days before my visit. Salamander-bot was being worked on when I met Ijspeert last September, so he let me drive snake-bot. It was not the easiest thing to control. Because its body moves in a wave, its head sways back and forth continuously, making it difficult to turn. After I had steered the snake into a lab bench for the fourth time, Ijspeert confided that he's not a very good snake-bot driver either, and most of the cool videos he'd shown me were taken when it was under the command of one of his graduate students.

Ijspeert is porting his central pattern generator research to other projects, including a robotic baby and robotic furniture that reconfigures itself, but I'll tell you about those another time.