抄録
Swarm robot control is not an easy task for a human
programmer, because the robot group behavior is emerged as a
result of many and asynchronous unexpected local interactions
between autonomous robots. In this paper, we approach to this
problem of designing robot controllers by using evolving artificial
neural networks (EANNs). From our preliminary computer
simulations, it has been found that the topology of the hidden
layer plays an important role of the evolvability of an EANN.
Therefore, we conduct a series of computer simulations to show
that an EANN having the small world properties in the hidden
layer performs better than EANNs with the hidden layer of
regular topologies or other conventional feed forward neural
network controllers from the viewpoint of the generalization
capability.
