The goal of this research project is to develop and refine a model of the human spino-neuromuscular system (SNMS). The SNMS is an extremely complex non-linear control system that helps to modulate movement, defines reflex actions, responds to short term disruptions to movements, and has a significant role in compensating for and recovering from injuries. Understanding the SNMS will result in improved therapies for injury rehabilitation and recovery based on the natural function of the SNMS.
In addition, understanding of the SNMS can be exported to the wider application field of non-linear control. The SNMS has been developed and fine tuned by hundreds of millions of years of evolution into a superb non-linear control system. Being able to apply the results of that evolutionary process to manmade non-linear control problems will significantly advance the field of non-linear control.
Our current model is a single degree of freedom joint (the elbow) that is actuated with a multi-fiber muscle (the biceps). Each fiber in the simulated biceps is controlled by a seperate, but interconnected, neural network. Each neural network consists of several different neurons. Weighted connections between the neurons determine the strength of the signals passed from neuron to neuron.
In the current model there are 114 neurons that make up 6 different neural networks to control the arm movement, resulting over 600 weights that need to be adjusted to produce the target motions. If these weights were balanced manually it would take days or even weeks to produce a properly functioning model. By using a Genetic Algorithm the weights are automatically adjusted in a few hours.
The initial focus of this program was to demonstrate that a suitable model for hypothesis testing could be devleoped and trained with a GA. Now that a suitable model has been developed we are concentrating on the next phases of the research program: determing the role of the SNMS and various segmental sensory and motor components asscociated with the SNMS in controlling movement and compensating for injury and disease. Using the model it is simple to add, remove or alter components of the SNMS and thus determine their role. For example, preliminary results suggest that the Renshaw cells may have a particular role in the control of eccentric movements. By modifying parameters of the uderlying model it is easy to simulate injury or disease and examine how the SNMS responds.
In addition to the above research, we are interested in refining the model to serve as an aid for undergraduate neuroscience study and improving the underlying genetic algorithm used to train the model and a physical version of the muscle and joint model is currently being developed as a capstone design project in the College of Engineering.
The University of Idaho (funded by the National Science Foundation) is taking part in an
undergraduate research program lasting from December 14, 2003 until February 28, 2007. The
focus of this research program is advancing knowledge throughout the areas of
neuroscience, including but not limited to: computational neuroscience involving genetic
algorithms and particle swarm optimization, neurogenesis in cell survival and regeneration,
analysis of current neural network models, and psycho-physical vision studies. Each of these
fields is the focus of one or more research teams comprised of undergraduate students from
around the country. The focus of this website is to present the research findings focusing on
the evolutionary training of a biologically realistic spino-neuromuscular system, particularly
a model of the human arm with the bicep being the main contracting element.
University of Idaho REU Project Web Site