The control of motor behavior is essential for us to interact with our environment. We possess an almost infinite variety of motor behaviors and acquire new ones with great facility. How the brain can control such a great number of behaviors and easily learn new ones remains something of a mystery. A recent theory based on the concept of ‘internal models’ of motor behavior has been proposed as a unifying approach to the understanding of motor control and learning. In this theory, internal models are defined as models of the motor periphery of an organism and the surrounding contextual environment, such that they can predict the sensory consequences of an intended movement and the forces required to generate that movement.
While much of the data from behavioral experiments in human subjects can be interpreted in the context of internal models, as yet we do not have a clear understanding of the neural mechanisms that might support such models. In this dissertation my primary focus is on the neural mechanisms of the learning of internal models and on the neural processes that enable us to rapidly switch from one model to another. To study this, I trained monkey subjects to learn to adapt their arm movements in the presence of a perturbing force field and recorded neural signals (single cell and local field potentials) from the motor cortex and the dorsal premotor cortex during the learning. We then studied how the subjects adapted when a force field in the opposite direction was introduced and the neural basis of this adaptation. This last manipulation was meant to simulate how we can acquire multiple internal models of behavior and dynamically switch between them. Our main result was that internal models are learned through gradual changes in the relative weighting of the motor parameters (direction, position, velocity and force) in single cells. In addition switching between models was accompanied by rapid and dramatic shifts in the parameter weighting. Interestingly, changes in the direction tuning of single cells seemed to have little to do with internal models. We also found systematic changes in the local field potentials associated with the models, though these data are somewhat preliminary. Finally, we did not find large differences between the neural activity in motor cortex and dorsal premotor cortex during learning or model switching.
In addition to the main experiment outlined above we performed two ancillary studies: one on the acquisition of multiple internal models in human subjects and the other in which we applied a decoding algorithm to the neural data collected in the primary experiment to predict the end point forces exerted by the monkey subjects.