Evolutionary robotics is a research area that makes use of the various
forms of evolutionary computation (EC) to provide a means of learning
in robots. The use of EC can reduce development effort and allow the
system to be adaptive to changes. However, EC also carries costs. It
can be computationally expensive enough to preclude on-line learning,
especially in small robots with simple controllers. In addition, most
forms of EC require that a population of possible solutions be tested
over several iterations. Current ER research deals with to what extent
these tests should be done on simulations. Simulation is faster, but
there is currently no provision for how to link the simulation to the
actual robot except by painstakingly increasing its accuracy. Our
research introduces a new way of integrating the actual robot and its
simulation during evolutionary computation.