A Computational Theory of Motor Development
Part 2: Development after Birth

Peter D. Neilson & Megan D. Neilson

Neuroengineering Laboratory,
School of Electrical Engineering and Telecommunications,
The University of New South Wales, Sydney, Australia

 

The moment of birth brings no discontinuity in the nature of a baby’s movement. The many coordinated movements observable in the neonate are essentially the pattern-generated movements observable in the foetus throughout the second half of pregnancy. The repertoire continues unchanged for approximately the first two months and we propose that the operative mechanisms remain those from Part 1. Thus post-Hebbian learning continues to fine-tune the already established sensory maps defining functional muscles and elemental movements, and slow neural modelling continues to develop the partly-formed motor maps defining hardwired elemental synergies for controlling flexion and extension forces for each elemental movement.

We begin Part 2 by contrasting these processes, Hebbian learning being unsupervised and algebraic, neural modelling being supervised and dynamic. We present the latter in detail as the basis for description of a further process, namely fast neural modelling. We propose this as the mechanism responsible for the remarkable transition that commences at around two months and results in the purposive, goal-oriented, multijoint movement typical of the 4-6 month-old. Fast neural modelling employs adaptive filtering and we propose this is implemented in cortico-basal ganglia-cortical and cortico-cerebellar-cortical circuits made available at this age via myelination of associated white matter.

Fast neural modelling allows piecemeal learning of existing pattern-generated movements, resulting in a "seed" controller for each. Parameters are adaptively established for each controller and stored in long-term memory, providing a motor lexicon that allows implementation of each movement in goal-oriented rather than pattern-generated mode. The infant can now undertake motivated exploration of the environment using variations on the "seed" synergies produced by the seed controllers. This activity generates sets of associated controllers involving the same combinations of functional muscles and elemental movements as the original seed controllers, but having the same or reduced degrees of freedom of control.

Ultimately comes development of synergies comprising new combinations of functional muscles and elemental movements. Novel trajectories of required perceptual consequences, generated by imitation or computation, must first be executed with existing controllers, whereafter feedback of execution errors progressively establishes a new controller. This iterative process allows indefinite extension of the repertoire.