Motor Skill Development and Neural Networks for Position Code Invariance under Speed and Compliance Rescaling

Abstract

This chapter presents two neural network modules capable of providing a secure foundation for safe self-organization of readily generalized movement skills. Called VITE and FLETE, these networks ensure position-code invariance under speed and compliance rescaling, respectively. This invariance property enables use of a simple strategy for skill development: For safety, we begin skill learning while performing at relatively low speed with relatively low limb compliance. Once learning guided by error feedback has reduced positioning errors, we increase speed and compliance. The invariance properties ensure that the shift to new values of the speed and compliance control signals will not require relearning. Both neural network models and the developmental strategy are compatible with, and help organize, large bodies of existing data. The FLETE network constitutes a comprehensive new model of the mammalian spino-muscular system.