Developing learnability — The case for reduced dimensionality

In this work, the notion of reduced dimensionality and its relevance for systems undergoing development is examined. The various motor control theories of degree of freedom change, optimal control, and motor primitives are related using the framework of control dimensionality reduction. Based on their relationship, we propose a developmental approach based on progressively utilising increasingly higher dimension representations of the system. A simulated planar 2 link arm model is then used to demonstrate the effect of utilising reduced dimensional models for control; comparisons on step and sinusoidal tasks are presented showing a progressive decrease in error that is task dependent quantitatively. Arguments are presented for why such a strategy might be essential from an evolutionary perspective for the developmental acquisition motor control in a tractable manner.