A Computational Theory of Motor Development
Part 1: Development in Utero

Megan D. Neilson & Peter D. Neilson

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

 

This two-part presentation provides a computational account of the development of motor control from the earliest movements in utero to the acquisition of new skills from infancy to maturity. In Part 1 we consider the well-documented emergence of foetal movement and propose the role this spontaneous pattern-generated activity plays in automatic formation of cortical somatosensory and motor maps, so providing the substrate for subsequent motor development processes addressed in Part 2.

We propose that reafferent feedback signals from foetal movement gradually establish, via post-Hebbian mechanisms, sets of sensory mappings in the developing somatosensory cortex, based on most probable long-term correlations between those signals. We consider mappings that reflect inter-relationships among joint receptors and cutaneous receptors as defining "elemental movements" and mappings that reflect interrelationships among muscle length receptors and tension receptors as defining "functional muscles". An elemental movement is the simplest voluntary movement of a body part occurring one at a time, independently of all other movements. A functional muscle is that set of muscle fibres that always change length together in a perfectly correlated way regardless of the movement made. We argue that knowledge of the relationships between elemental movements and lengths of functional muscles defines "seed synergies," the basis for subsequent voluntary control of movement.

We propose that relationship between sensory maps determines the development of motor maps, slowly instated by neural modelling during the second half of pregnancy. The relationship to be parameterized is multivariable and nonlinear but, being determined solely by anatomy, it is only slowly time-varying and independent of external forces. Importantly, the relationship is inherently redundant, with multiple sets of parameters equally capable of generating the same elemental movements using different synergies of the functional muscles. Thus we propose an optimization criterion, namely the minimization of muscular energy. We demonstrate mathematically that the required parameters are given by the ratios of changes in muscle lengths to changes in elemental movement angles. Maintained adaptively by slow neural modelling (see Part 2), these motor map parameters provide modulation of descending pathways from upper to lower motoneurone pools to achieve minimum energy coordinative structures for elemental movements.