Jul 06 2012

Robot Legs and Central Pattern Generators

The most common response of my patients when I test their deep tendon reflexes is to giggle. I bang gently on the infrapatellar tendon, their leg kicks out involuntarily, and they giggle. While I am acutely interested in the reflex response and what that says about my patient’s nervous function, the giggle is perhaps the more interesting response. Many patients also comment something to the effect of, “That is so weird.”

What is weird is the experience of movement outside of our conscious control. We exist in a neurologically induced illusion that we own and control our limbs.  There are, in fact, specific circuits in the brain that generate the experience of ownership and control. We know this partly because of patients in whom those circuits are disrupted and lose their sense of either ownership or control. I guess it is also not surprising that we are conscious of the conscious level of control of movement, but not of the subconscious elements of control. We are occasionally reminded of them when a reflex supersedes are conscious control, an experience we find weird and often giggle-inducing.

The reality is that there are different levels of hierarchical control in our nervous system. More basic circuits provide automatic function and are literally (phylogenetically) more primitive than higher levels of more sophisticated control, all the way up to voluntary control from the cortex. We are not consciously aware of most of the processing that is needed to produce smooth and coordinated movement. If we desire to walk across the room, for example, we guide our movements to accomplish that desire, but we are not aware (thankfully) of the many components that go into the astonishing balancing act of walking.

To quickly review some of the components of neurological control of movement, there are various subsystems in the brain that are involved. There is the pre-motor cortex, which is responsible for planning and initiating movement. The motor cortex (also called the pyramidal system, after the pyramid-shaped neurons in the motor cortex) is the direct activating system from the brain to the muscles. This is a two-neuron system – the upper motor neuron is in the motor cortex and sends its axon all the way down to either the brainstem or the spinal cord and synapses on the lower motor neuron. The lower motor neuron then sends its axon out through a motor nerve to synapse on the muscle (with one motor neuron and its axon controlling one motor unit, which is a collection of muscle fibers).

If that was all there were to the motor system, however, we would be an uncoordinated mess of twitching muscles, not able to do much. There is also the extra-pyramidal system, located in the the basal ganglia deep within the brain (and therefore in a more primitive part of the brain). The extra-pyramidal system is a delicate feedback loop that essentially modulates the gain of the motor system, including the connection between the pre-motor and motor cortex – the desire to move and the the action of moving. If the gain is too high then we are constantly writhing and moving (such as in the disease Huntington’s chorea), and if the gain is too low then we are stiff and unable to initiate movement (such as in Parkinson’s disease).  We take for granted that the gain is just right, but that is a complex and entirely subconscious feat of neuronal processing.

There is also the cerebellar system, which is responsible for coordination. Roughly a third of the axons from the motor cortex do not go directly to control muscles but go instead to the cerebellum for processing. The cerebellum also gathers information from various sensory systems, including vision, the vestibular system (senses acceleration and gravity), and proprioception (which senses where our limbs are in three-dimensional space). The various sensory inputs are compared and processed, in addition to our planned muscle movements, and then output from the cerebellum modifies the descending motor activity (as well as provide feedback to the motor cortex) to generate balance and finely coordinated movement.

We then get to the brainstem which is more primitive still. There we find collections of neurons which are responsible for some basic motor functions. For example the red nucleus provides tone to anti-gravity muscles, so we can stand without unnecessarily taxing cortical motor control. Automatic actions like breathing also originate and are regulated in the brainstem.

Below the brain there are still more primitive and basic circuits involved in motor control. There are a number of reflexes that are modulated entirely between the muscles and the spinal cord, and never reach up to the brain. Deep tendon reflexes are such an example of a spinal cord reflex. There is also something called central pattern generators. These are collections of neurons in a basic circuit that generate some alternating pattern of activity. There are probably among the most primitive circuits to have evolved. Their function is to modulate the pattern of muscles that contract as part of a conscious action, such as the alternating movement of muscles when walking.

Understanding the nervous system in general and motor control specifically as a function of processing and reflexes that occur at different hierarchical levels has greatly advanced not only our understanding of neurology but also has given us a model for robotics. Some early attempts to program a robot to walk, for example, took a top down approach – creating a single piece of software that would perform all the complex processing necessary to control robotic limbs to generate the desired action, like walking. However, over the years roboticists have turned to more of a hierarchical approach – creating simple algorithms to provide elements of function. The idea is that a circuit should not be any more complicated than it has to be to perform one specific function, which can then be incorporated into higher levels of control, just like the vertebrate nervous system.

Recently engineers have exploited the principle of the central pattern generators to help robotic legs walk in a more human-like fashion. It is an excellent example of copying this principle of hierarchical control. The BBC reports:

Dr Theresa Klein, who worked on the study, said: “Interestingly, we were able to produce a walking gait, without balance, which mimicked human walking with only a simple half-centre controlling the hips and a set of reflex responses controlling the lower limb.

Their robot legs did not just mimic a human looking gait, but recreated the “underlying human control mechanisms” controlling gait. This is just one incremental advance, but I do think it shows the promise of this approach. Simple circuits have been tweaked by millions of years of evolution, so understanding and copying them is bound to be useful.

The utility of this technology also goes both ways, as I have discussed with other research involving robotics and neuroscience. First, it will help us build better robots (or artificial intelligence) by modeling biological circuits. Second, doing that modeling and testing it in computer or robotic systems will help us understand the neurological systems. The practical applications of this kind of collaboration among robotics, AI research, and neuroscience are potentially huge. It might result in the development of devices (prosthetic, orthotic, or brain-machine-brain interface – think the bionic man) to replace, fix, or enhance motor control. Imagine a robotic exoskeleton on the legs of a paraplegic with a spinal cord injury able to provide sensory feedback and with all the primitive reflexes to mimic natural leg function. Or imagine someone with severe ataxia (poor balance) so that they cannot walk who is essentially cured by a neuroprosthesis that takes over balance processing.

And of course if we ever want to have humanoid robots walking around (for whatever reason) this research is essential.

10 responses so far