Madam Fathom

In my first post ever, I discussed how specialized circuits in the spinal cord (called "central pattern generators," or CPGs) coordinate the intricate motions and muscle patterns involved in running and walking, without significant input from the brain. The autonomy of these circuits allows animals to run and walk while their mental efforts are otherwise engaged; for example, we can talk on the phone while walking to dinner, and decapitated chickens can still run away.

One of the most important features of CPGs is their adaptability. Whether running through a forest, walking on an oily surface, or dribbling a soccer ball, we need to continuously modify our movements. Thus, as opposed to generating rigid action patterns, CPGs provide a flexible template for coordinating our muscles and various joints. This template interacts with sensory information, allowing us to elegantly adapt to our unpredictable world. Flexibility, however, poses a challenging computational problem; not only must we decipher how circuits of neurons coordinate hundreds of muscles, but also how their output can be refined by incoming sensory information.

Without understanding these fundamental issues, it is difficult to produce machines that can move as intelligently as we. Honda's ASIMO, "The World's Most Advanced Humanoid Robot," is capable of executing an astounding range of human-like movements (running, walking smoothly, reaching for objects), but has previously stumbled and fallen down stairs. A recent article in PLoS Computational Biology describes a new and improved droid named RunBot, which is capable of adapting to unfamiliar terrain in an animal-like way.

Although not nearly as cute as ASIMO, RunBot's motor circuitry is more "intelligent" (i.e. more human). As I mentioned in my earlier post, the motor system is arranged in a hierarchy: the "higher" control centers give the signal to initiate a movement, recruiting the "lower" CPGs to take care of the details. These lower circuits respond to the environment reflexively, incorporating localized feedback to generate intricate adjustments in muscle tone. This responsiveness allows us to immediately compensate for small perturbations, such as unnoticed rocks on a trail. When we need to significantly modify our gait, however, such as stepping over a baby, we must enlist the higher centers, which will generate more dramatic modifications to the CPGs.

ASIMO lacks this hierarchy, requiring it to continuously calculate the position and motion of every joint. RunBot, however, has been engineered with several levels of control, allowing it to adapt to changes in terrain in a more computationally efficient manner. RunBot interprets the environment with an infrared sensor, which communicates with the lower circuits to regulate their activity. Thus, when RunBot encounters an alteration to its terrain and becomes unbalanced, this sensor modifies the pattern of the lower circuits, allowing the bot to change its gait.

However, like humans, RunBot must learn how to modify its movements with respect to sensory input. When we learn how to walk, our brains "train" our CPGs until they can execute the movement relatively independently. These same mechanisms come into play when a runner learns to hurdle or a soccer player learns a new move; these behaviors initially require significant concentration, but with practice can be executed with little mental effort. To replicate this learning process, RunBot's circuitry includes, according to the authors, "online learning mechanisms based on simulated synaptic plasticity." Thus, when RunBot first attempts to climb a slope, it falls over like poor ASIMO. With trial and error, however, its circuitry learns to properly compensate for the relevant sensory input, shortening and slowing its steps just like a human.

The invasion of the land by animals was an astonishing evolutionary feat, necessitating a number of substantial changes to the body: limbs with digits, structures for obtaining oxygen from the air, a relatively waterproof covering to prevent dehydration, and sturdy structures to support the body in a medium much less buoyant than water, to name a few. When these pilgrims first bridged the immense gulf between land and water, almost every system in the vertebrate body underwent substantial modifications, but what about the nervous system?

The different optical and sound properties of water and air required significant adaptations for our visual and auditory systems, but these adaptations were largely peripheral: the lens changed shape to adapt to the different refractive indices, and the bones of the middle ear evolved from other bones of the face. Perhaps the most significant behavioral modification (which would thus require notable rewiring of the neural circuitry) was the transition from swimming to walking. Did animals need to invent completely new pathways to support a wider repertoire of locomotion?

First, it's necessary to have a general understanding of the neural basis of locomotion: central pattern generators (CPGs). I posted on CPGs a little while ago; the basic idea is they are networks of neurons, located in the spinal cord, which coordinate all of the muscles involved in locomotion without input from the brain. I focused on bipedal motion, but CPGs control rhythmic locomotory movements in all vertebrates, including those that swim and fly. Thus a more focused question is: when animals transitioned from swimming to walking, could the same CPGs that controlled aquatic locomotion handle the different coordination needed between a body and its limbs for walking?

An excellent paper was published today in Science that explored this question using a robotic salamander (named Salamander robotica because scientists are pretty bad at naming things). Salamanders are considered to be the most similar to the first terrestrial vertebrates, and are thus often used as a model system for studying the evolution of new anatomical structures for terrestrial (vs aquatic) locomotion.

Salamanders can rapidly switch from swimming (using undulations similar to those of primitive fish), to walking (using diagonally-opposed limbs that move together while the body forms an S-shape, like an alligator). Looking at the animal's movements from above, the body can be seen as either moving in a traveling wave or a standing wave, respectively, and neural activity along the spinal cord is likely to mirror this effect.

The group, led by physicist Auke Jan Ijspeert and neurobiologist Jean-Marie Cabelguen, designed Salamander robotica with an electronic "spinal cord" to determine whether the same spinal network could produce both swimming and stepping patterns, and how it might transition between the two.

Their spinal cord was controlled by an algorithm that incorporated essential known or speculated attributes of salamander locomotion. First, the group knew (from a study they did in 2003) that the transition between standing and traveling waveforms can be elicited simply by changing the strength of the excitatory drive from a specific region of the brainstem. In this experiment, a weak drive induced the slow, standing wave of the walking gait, while a stronger drive induced the traveling wave of the swimming motion. Second, the authors reasoned that there are two fundamental CPGs controlling salamander locomotion: the body CPG, located along the spinal cord, and the limb CPG, located at each of the limbs.

With these parameters in place, Salamander robotica set forth on her quest to traverse land and sea, to test whether her "primitive" swimming circuit (the body CPG) would be able to coordinate with the "newer" circuits of her phylogenically recent limbs to produce the waddling gait of her sentient inspiration. Watch the results:



So mechanistically, how does she do it? The group found that at low frequencies, both CPGs are active; the limbs then alternate appropriately, and are coordinated with the movements of the body. At higher frequencies, the limb CPGs are overwhelmed, and thus the limbs tuck in as the body CPGs take over.

One interpretation here is that the group is good at building robots, so Salamander robotica did exactly what they wanted it to do. Another interpretation, the one that got this study into Science and into my blog (quite selective, really), is that the spinal locomotor network controlling trunk movements has remained essentially unchanged during the evolutionary transition from aquatic to terrestrial locomotion.

I think this experiment was highly innovative. As you might imagine, it is quite difficult to study evolution in a controlled laboratory setting (global warming suffers from similar drawbacks, but that's a whole 'nother post), so using robotics as an experimental model is quite promising. The core finding, however, was not surprising to me.

The transition from water to land necessitated a daunting number of anatomical changes and, looking back 370 million years, it seems an unsurmountable divide. But the success of evolution hinges on the fact that it occurs gradually, and rarely involves unusual or extraordinary biological processes. It is thus logical that a common, underlying neural mechanism for propulsion can produce a variety of movements; we see this in modern humans, as well. As I pointed out in my earlier post, the same CPGs--in fact, the same neurons in the same CPG--are used for walking, running, hopping, and skipping. The fact that locomotion is largely independent of conscious control strengthens this rationale; it is much more straightforward to make small adjustments to the system by tinkering "downstream." Thus, when animals adapted to terrestrial locomotion, they used the most efficient (and thus most likely to be successful) strategy: recruiting the same neural circuits used for aquatic locomotion.

Which is not to say that this finding is any less wonderful. It is simply a powerful reminder that, as evolutionary biologist Neil Shubin wrote, "the ancient world was transformed by ordinary mechanisms of evolution, with genes and biological processes that are still at work, both around us and inside our bodies." This is, in his words, "something sublime."

For more information, The Neurophilosopher has an excellent, more detailed post on this paper.