RHex marches up a dune in Pismo Beach, California, carrying a mechanical sensor that scrapes the dune’s surface to detect how much force is needed for the wind to pick up grains of sand.

In this photo from 2016, RHex marches up a dune in Pismo Beach, California, carrying a mechanical sensor that scrapes the dune’s surface to detect how much force is needed for the wind to pick up grains of sand.

Feifei Qian

Thinking on Your Feet—or with Them

Would you be able to walk, think, or react without a nervous system? University of Pennsylvania engineering student Sonia Roberts explores this question while building robots inspired by real animals.

A thumb-sized beetle with a shiny black shell ran toward me along the base of the white sand dune, pausing occasionally to pitch its head down and investigate some tasty-looking plant before picking up its long legs three at a time to continue skittering along. It moved so gracefully across the sand—much more gracefully than the boxy, six-legged robot I’d been trying to coax up the steep sides of dunes all day with varying success. But the sun was going down, and the temperature with it, dropping quickly from the typical daytime 80⁰F in the shade of a New Mexico desert in the fall. In combination with the long shadows and increasingly chilly wind, the little nocturnal visitor was gently reminding me that the park, White Sands National Monument, would be closing soon and we would have to pack up our experiments for the day.

RHex (short for Robot HEXapod), the machine I was experimenting with, is just one of the robots I work with to help scientists study desertification and the movement and production of sand dunes. These robots are equipped with sensors that measure things such as wind flow, the size of sand grains blowing through the air, and the amount of force required for the wind to pick up those grains. My job is to redesign and reprogram those robots so they can walk to the places where scientists need to make measurements.

rhex_robot.jpg

RHex and a darkling beetle at White Sands National Monument in southern New Mexico, 2014.

RHex and a darkling beetle at White Sands National Monument in southern New Mexico, 2014.

Sonia Roberts

To get a robot such as RHex to go up a sand dune, I looked for inspiration in real animals that have evolved to live in the desert. Understanding what these desert critters do and the bodies they do it with could help me reprogram RHex’s computer “brain” and change the shape of its body to make it a little more graceful on the dunes.

Most people believe that deliberate behaviors, such as walking or throwing a ball, are produced by their brain sending commands to their compliant body, but that’s not always true. There is a thin line between what is controlled by the body and what is controlled by the brain.

The idea that there is a separation between an intelligent brain and an inanimate body goes back at least to the 15th century, when the French philosopher René Descartes theorized that bodies were hunks of matter animated by the brain. In his analogy the brain uses the nervous system, a network of nerve cells that spread like tendrils throughout the bodies of all animals, to signal when and where to move. But not everything needs a nervous system to move, and some organisms don’t even need a nervous system to respond to signals from the environment. For example, it’s possible for single-celled, brainless amoebas to hunt by activating a sort of molecular motor in response to certain smells. Similarly, some species of parasitic plants, such as dodder vines, grow toward their host plants without any nervous system at all. In time-lapse videos these plants look like boa constrictors extending toward and encircling their prey.

What would a brainless behavior look like in a robot? Just like the organisms that move without a nervous system, it’s possible to build a robot that moves around and responds to its environment without a computer providing commands. Such a robot would only need two sensors and two motorized wheels hooked up to wires. When the left sensor is triggered (say, by the presence of a specific chemical), the right motor would spin faster, turning the robot to the left (similar to the way a tank turns). The opposite would happen when the right sensor is triggered.

If the nervous system isn’t necessary to move an organism’s muscles, then perhaps it is necessary to coordinate those muscles. That is the opinion of the Swiss roboticist Auke Ijspeert, who creates models of animal locomotion using artificial central pattern generators.

In animals, central pattern generators are neural circuits that, when stimulated, produce rhythmic pulses of neural activation that coordinate groups of muscles to contract and release. Coordinated muscles allow animals to perform any number of common rhythmic activities, including walking, running, breathing, and chewing.

Ijspeert built a robot that both swims and walks using artificial central pattern generators modeled on a salamander’s nervous system. His research shows that the nervous system definitely plays a role in muscle coordination, further supported by the fact that there are no plants or amoebas with central pattern generators. The simplest critters that have them seem to be mollusks, which lack proper brains but use central pattern generators to swim. Similarly, jellyfish have a nerve net but not a brain, and they use central pattern generators to coordinate their mesmerizing bloop-bloop-bloop mode of swimming.

Although Ijspeert’s salamander robot can move around very nicely on its own, it’s missing something vital that most animals have: reflexes. The same system that causes your knee to move when your doctor taps it with a hammer is responsible for all kinds of behaviors and movements. In fact, it’s possible to create a fully functional model of biological (and robot) locomotion using only reflexes.

So how do reflexes work? Continuing with the example of a human leg, each part of the leg reacts independently to local events, such as stretching, bending, or contact with the ground, and the whole body naturally stabilizes into an energy-efficient gait. Other animals use reflexes in a similar way to produce local reactions to local events. Putting all of this together, maybe the role of the nervous system in locomotion is some combination of responding to physical stimuli using reflexes and approximately following motor patterns from a central pattern generator.

But if automatic reactions in a leg are enough to get a body moving, then—if we think about our simple robot with no computer—shouldn’t it be possible to create a body that “walks,” or moves rhythmically, without a nervous system at all? In the animal world all kinds of one-celled organisms move around using rhythmic wiggles of their “tails” (flagella) and “limbs” (cilia)—all without muscles. Instead, they use the shapes and material properties of their flagella and cilia, which, when given a little bit of energy, wiggle around to propel the animal forward.

Likewise, we can build robots that produce extremely energy-efficient, biological-looking locomotion with just a small push, be it from gravity or a motor. The first versions of these robots were toys called “passive dynamic walkers” that trotted down ramps without any motors. Mechanical “programming,” where physical structure creates automatic movement, also appears in more complex organisms that have brains and nervous systems, enabling the animal to react faster than its nervous system would allow. If you knock a cockroach sideways while it’s trying to run, its gait will start to recover within a single stride, which, given the speed at which the cockroach is moving its legs, is faster than the minimum amount of time a neurological reflex would take.

So we don’t need a nervous system to act, we don’t need one to sense, and we don’t even need one to coordinate movement—though of course a nervous system can be used for all those things. Is there any reason we really, truly need a nervous system? What’s the nervous system for?

One possibility is learning. Even critters with very simple nervous systems, such as sea slugs, can learn, as Eric Kandel showed in a series of experiments in the 1970s and 1980s. He habituated sea slugs to foot pokes by prodding them until they stopped bothering to retract their feet, and then he resensitized them by accompanying the pokes with an electric shock. In 2000 he won the Nobel Prize in Physiology or Medicine for identifying some of the neurotransmitters that were synthesized during the learning process (the sea slugs’, that is) and showing that the experiences of the animal could structurally change its nervous system. So far slime mold, which anticipates annoying rhythmic pulses of dry air by shrinking, is the critter with no nervous system that has come closest to learning, but scientists question whether this anticipation can be called “learning.”

So what about the intrepid little RHex, a robot on a mission to explore the desert? How can I improve its ability to walk around on sand dunes, knowing what I now know about the nervous system and how special it is—and isn’t?

I trust the physical properties of materials I use to build robots more than I trust a sensor not to malfunction when it gets too much sand in it. That’s why I try to design more of a robot’s behavior into its physical body than into its computer programming, using mechanisms like those in passive dynamic walkers. Unless the robot needs to do on-the-job learning, there’s no reason to get its computer “brain” involved.

To help RHex get up sand dunes more easily, I took inspiration from nature in two ways. Biologists have found that desert critters tend to have big feet relative to their body size, a characteristic that helps them run on sand. So the first thing I did was widen the robot’s feet. This adjustment increased the area of each footstep and decreased the pressure on the surface of the sand, which means RHex can go faster and take longer steps without sinking.

image.png

Feifei Qian

Second, I increased RHex’s gear ratio, which traded top speed for lower energy cost. Animals, including humans, that evolved under selection pressures requiring long-term exertion in hot environments have similar adaptations that reduce energy expenditure and dissipate heat more efficiently. In fact, humans are so well adapted to vent heat during long runs that some scientists think we may have developed this ability specifically to chase down less energy-efficient prey. Today RHex is able to march up sand dunes even with heavy sensors strapped to its back, thanks to many iterations and improvements over the years.

The biggest lesson I learned from studying animals and their nervous systems is how hard it is to separate the jobs of the brain and the body. When designing robots, it’s not enough to build a body and then figure out how to control it later, because behaviors are defined by how organisms interact with their environments. No matter what job you think the brain might have, there always seems to be an animal (or robot!) somewhere doing that job with its body alone. So maybe the nervous system isn’t so special after all. If we asked amoebas, dodder vines, and RHex about it, maybe they’d say the old adage should be flipped: brawn over brains.

Sonia Roberts

is an Electrical and Systems Engineering PhD student at the University of Pennsylvania's Kod*lab.