The evolution of simulated robots with three different architectures is stu
died in this article. We compare a nonmodular feed-forward network, a hardw
ired modular, and a duplication-based modular motor control network. We con
clude that both modular architectures outperform the non-modular architectu
re, both in terms of rate of adaptation as well as the level of adaptation
achieved. The main difference between the hardwired and duplication-based m
odular architectures is that in the latter the modules reached a much highe
r degree of functional specialization of their motor control units with reg
ard to high-level behavioral Functions. The hardwired architectures reach t
he same level of performance, but have a more distributed assignment of Fun
ctional tasks to the motor control units. We conclude that the mechanism th
rough which functional specialization is achieved is similar to the mechani
sm proposed for the evolution of duplicated genes. It is found that the dup
lication of multifunctional modules first leads to a change in the regulati
on of the module, leading to a differentiation of the functional context in
which the module is used. Then the module adapts to the new functional con
text. After this second step the system is locked into a functionally speci
alized state. We suggest that functional specialization may be an evolution
ary absorption state.