Evolutionary training of a biologically realistic spino-neuromuscular system

This paper presents a biologically realistic model of the spino-neuromuscular system (SNMS). The model uses a pulse-coded recurrent neural network to control a simulated humanlike arm. We use a genetic algorithm to train the network based on a target behaviour for the arm. Our goal is to create a useful model for studying the function and behaviour of neural pathways in the SNMS. The genetic algorithm is able to train the network to actuate the arm to achieve controlled motion. Our experimental results demonstrate that certain types of feedback pathways are important for controlling certain movements.