Abstract
Recently, There has been much work in the area of co-evolving morphology and control of autonomous robots. However, most work in this area remains acquiring simple motion like walking because of using primitive components such as cylinders and servo motors. We consider optimization method for combining and controlling "functional components" like wheels and joints with motors. Our method is based on Real valued Flexibly Connected Neural Network (RFCN). This paper describes the motion of autonomous robots acquired in a virtual three-dimensional space simulating physical law. We also show the motion of a real robot based on the result of simulation.
