Embodied Decision Making: System interactions in movement execution and action selection
Two well-established literatures have provided elegant computational models of sensorimotor learning and decision making, with relative little connection between the two.Β I will discuss ways in which we can bring these two worlds together.Β Β In the first part of the talk, I will discuss work that has brought into focus the relevance of multiple learning mechanisms for sensorimotor learning, examining how strategic processes interact with a cerebellar error-based learning system in sensorimotor adaptation.Β The results here highlight a distinction between error signals that improve action execution or action selection.Β Β In the second part, I will turn to learning in decision making tasks, asking how competence in motor execution might be incorporated in models of reinforcement learning.Β Β Our work here addresses the question of how an agent determines if the absence of reward reflects a property of the environment or an error in motor execution.Β We suggest that sensorimotor errors provide a βgatingβ signal to regulate reinforcement learning, providing a simple solution to this fundamental credit assignment problem.
