UC San Diego

Movements produced in everyday life pursue a goal. Key to the success of such movements is the motor system's ability to adjust sensorimotor strategies in a flexible way according to the goal. On the high level, flexibility entails taking into account multiple task requirements and properties of the environment and preparing a sensorimotor strategy customized for the present task and circumstances to better achieve the goal. On the low level, a strategy is flexible if it makes on-line adjustments that exploit the multiple ways in which a redundant musculoskeletal plant can achieve the same behavioral goal. Both levels of flexibility, however, are mostly ignored by traditional theories. This thesis uses both psychophysical experiments and computational modeling to explain how biological movements arise from different goals they pursue. Our first focus is on how task goals shape motor planning. We show that the motor system customizes sensorimotor strategies for current task requirements, rather than generating a rigid motor trajectory regardless of the goal. We account for such customization of task goals in the optimal feedback control framework by using a composite cost function instead of a homogeneous cost with multiple hard constraints. We also address how a control strategy is adapted in changing environments. We show that motor learning involves not only the statistical formation of an internal model to predict external changes, but also the flexible use of such predictions to adjust motor commands for maximum performance. Such flexible dependence on predictions is accommodated by extending the optimal control framework to deal with complicated noise and cost formulations. Finally, rather than ignoring the musculoskeletal structures of human body, we apply a hierarchical control framework to a more realistic arm model with 7 degrees of freedom. This framework is inspired by the facts that optimal feedback controllers for redundant systems exhibit hierarchical organization, and that sensorimotor control occurs simultaneously on many levels. The basic idea is to have the high level solve the optimal control problem with reduced dimensionality, and the low level perform an instantaneous feedback transformation of plant dynamics according to the high-level commands. This work sheds light on understanding how the brain controls human body which is a complex redundant system