Learning dynamics of reaching movements

Reza Shadmehr

Johns Hopkins University

When one moves their hand from one point to another, the brain guides the arm by relying on neural structures that estimate physical dynamics of the task and transform the desired motion into motor commands. If our hand is holding an object, the subtle changes in the dynamics of the arm are taken into account by these neural structures and this is reflected in the altered motor commands. These observations have suggested that in generating motor commands, the brain relies on internal models that predict physical dynamics of the external world. Here, I will review the neural and computational data on how the brain learns internal models of reaching movements. Data suggests that internal models are sensorimotor transformations that map a sensory state of the arm into an estimate of forces. If we assume this neural computation is performed via a population code, one can infer properties of the tuning curves of the computational elements from the patterns of generalization and trial-to-trial changes in performance. A new theory is presented that allows for quantification of generalization from trial-to-trial changes in performance. The patterns of generalization appear consistent with computational elements that are bimodal in velocity space and the discharge is modulated linearly as a function of the static position of the hand. These gain-field properties are consistent with tuning curves of some cells in the primary motor cortex and the cerebellum.