Evolving stable behavior in a spino-neuromuscular system model

Abstract

This paper demonstrates the effectiveness of genetic algorithms in training stable behavior in a model of the spinoneuromuscular system (SNMS). In particular, we test the stability of trained instances of the model with respect to unfamiliar control signals and untrained forearm weights. The results show that small changes to the input frequency and forearm weight result in small changes in velocity, demonstrating that the system can reasonably accommodate unfamiliar circumstances. This type of stability is a critical feature for virtually any type of control system.