Understanding Grasp Control Using the Geometry of Dynamics

Grasping is a fundamental motor behavior that requires the coordination of many degrees of freedom and continuous interaction with objects, yet the neural principles underlying its control remain poorly understood. While dynamical systems approaches have successfully explained population activity in motor cortex during simpler movements such as reaching, these frameworks have struggled to account for grasp-related activity. In this thesis, I investigate how neural population dynamics in sensorimotor cortex support grasping by analyzing multi-area recordings from non-human primates performing reach-to-grasp behaviors. I show that grasp-related neural activity is dominated by a strong condition-independent temporal component that is shared across different grasp types. Unlike in reaching, this component overlaps heavily with condition-specific tuning. Using a flexible dynamical systems framework, I demonstrate that grasp-related activity exhibits conserved temporal frequencies that are organized across conditions within high-dimensional neural state space. Across conditions, neural dynamics are better characterized by changes in the orientation of condition-specific subspaces rather than simple shifts in neural state space location. These geometry-dependent dynamics relate to hand kinematics, linking neural population structure to behavior. Together, these results suggest that grasp is supported by high-dimensional, geometry-dependent dynamics and extend existing theories of motor cortical function.