Stochastic optimal control and Bayesian inference: an emerging theoretical framework for sensorimotor integration
Department of Cognitive Science, University of California San Diego
Last modified: April 28, 2007
Presentation date: 08/11/2007 2:05 PM in MCC
A growing body of evidence supports the view that both sensory and motor processing are optimal in a probabilistic sense. Optimal performance in an uncertain environment requires Bayesian inference on the sensory side combined with stochastic optimal control on the motor side. In this talk I will briefly summarize key pieces of evidence supporting this theory, and then focus on clarifying its relationship to other theoretical ideas in the field: equilibrium-point and impedance control, internal models and adaptation, dynamical systems and pattern generators, uncontrolled manifolds, coordinate frames, motor synergies. I will also discuss new applications of the theory: predicting neural population codes, understanding exploratory movements, inferring behavioral goals from movement data.