The history of natural evolution displays an inseparable coupling of organic bodies and the nervous systems that control them. In contrast to this almost all research in Evolutionary Robotics to date begins with a robot body whose features are fixed and proceeds to evolve a control structure for this body. Our research program is focused on exploring the coupled evolution of both the body and the control structure in real robots. In this paper we take early steps toward this goal by exploring the space of sensor and effector selection and positioning coupled with a neural network linking them within a simulated environment. This space is explored using evolved grammars for generating both the body and neural network. Results from several problem worlds are presented and analyzed.
展开▼
