Abstract
Being inspired by the existence of the small world networks in the neural net, we designed new structures of an artificial neural network and compare them with ones of conventional neural networks from a viewpoint of learning. We adopt the evolutionary computation for assigning proper weights to synapses. A learning test problem is given that a small robot located at the corner in a square field reaches the light source as fast as it can drive. Results in simulation shows that a small robot with a structure of small world networks in the nerves system can reach the light source faster than any other structures.
