The human brain has 86 billion neurons. These individual cells act in concert to send electrical messages that result in our thoughts, memories and movements. No single neuron “knows” what it’s doing. Instead, neurons exhibit emergent intelligence, similar to millions of ants working together to construct and maintain a complex colony.
With this level of complexity, can we program a nervous system from scratch?
Modeling the nervous system
In the USC Viterbi School of Engineering’s Department of Biomedical Engineering, Francisco Valero-Cuevas collaborates with Terry Sanger and Gerald Loeb to build “brain on a chip” models of the nervous system, where computer programs simulate populations of neurons in the human spinal cord.
When running, the encoded neurons could control a robotic or prosthetic hand the same way we control our own bodies. This will be a practical test of our understanding of how complex function emerges in the nervous system from populations of relatively “simple” individual neurons, how they communicate with each other and ultimately how they control our muscles.
“If we really understand fundamental aspects of the nervous system,” Valero-Cuevas said, “shouldn’t we be able to reproduce fundamental functions like finger motions?”
The irony is not lost on us that we’re combining one of the oldest scientific disciplines, hand anatomy, with some of the newest elements of ultra-fast parallel computing.
His projects, funded by the National Institutes of Health, are tackling some of the National Academy of Engineering Grand Challenges about how our brains fundamentally control our bodies. Enhancing our understanding of how the nervous system works will at the same time shed light on how it breaks down and diseases emerge.
And this understanding will directly inform therapeutic interventions for neurological conditions as well as the designs for better, more functional and capable robots. “This work is a true interface between biology and engineering,” Valero-Cuevas said.
Why math is the language of nature
Translating a biological system like the spinal cord to computer code is no small feat, but all the information, complexity and beauty of a system like our brain traces back to mathematics.
“It turns out that we agree with something that Galileo said centuries ago: that nature is an open book if you only know the language in which it’s written, and the language is math,” Valero-Cuevas said.
To test his computer models of neural control, Valero-Cuevas is using a very faithful physical system: cadaver hands. Hand surgeons help him connect the hands’ tendons to strings driven by electric motors.
The activity of the motors is controlled by the neuron software, as if the motors were muscles themselves. This way the simulated neurons are confronted with the same problem the nervous system faces: controlling the hand as a marionette driven by complex muscles and tendons.
The goal is for the software and hardware to work in concert to control the cadaver hand the same way a healthy person can move his or her hand — complete with stretch reflexes, muscle tone and compliance.
“We are studying the very fundamental mechanisms of how muscles have tone and how you modify that to get function, and how their disruptions lead to the pathological characteristics of hypertonia, spasticity and dystonia, which are very common in cerebral palsy, stroke and spinal cord injury,” Valero Cuevas said. “But we don’t really know where they come from, and we’re trying to understand that. And a team that includes Sanger and Loeb is very powerful in this regard.”
The complexity in just one little finger
Each finger tendon is controlled by between six and 10 muscles, and in turn, each simulated muscle is controlled by a population of 256 independent neurons.
“The irony is not lost on us that we’re combining one of the oldest scientific disciplines, hand anatomy, with some of the newest elements of ultra-fast parallel computing,” Valero-Cuevas said. “We’re using this to answer central questions about evolution, health and disease, and how all these systems work.”
One application of this work is the design of better prosthetic hands, where there is still a major engineering challenge to make artificial hands that can be effective manipulators of objects. The most advanced current prosthetics are effective grippers, but the ultimate goal is truly dexterous manipulators.
“We see it as an impasse,” Valero-Cuevas said. “Over a century of trying to develop something that’s better than the split hook prosthesis. We now have modern robotic hands and prosthetic hands that are amazing grippers, but they’re not dexterous manipulators. They’re great at holding things, but is it the Luke Skywalker hand that would be able to pick something up, reorient and operate it? Think of all the operations that are needed to use your smartphone with one hand.”
The potential to answer age-old questions
But that’s just one application for this wide-ranging work. Another will be robots that are more compliant, meaning that their nervous systems will sense an obstacle and soften when contact is made. This will make robot-human interaction safer, especially for large, strong robots, so an accidental encounter could be more like bumping into a person than colliding with a brick wall.
And for Valero-Cuevas, this is just one way that his lab explores the interaction of the brain and body, which is why his research group is called the Brain Body Dynamics Lab. “Some people think I only work on hands,” he said, “but that’s just because hands are an example of a complex system I can have on my desktop.”
Embracing the ideals of Engineering+, Valero-Cuevas is also asking fundamental questions about evolutionary biology, such as why vertebrates appear to have “too many” muscles, what the evolutionary pressures were that led to the specialized body and brain in humans, and how the structure and function of the human nervous system defines the nature of dysfunction and rehabilitation.
“I’m very excited because it will be begin to narrow and define the conversations we should be having have about what is function, what is health, what is disease, and what a robot should be like,” Valero-Cuevas said, “and then open up new questions.”