WP-2003-Virtual World chapter#3 of book.pdfVIP

Chapter 3 The Virtual World This chapter outlines the virtual world used for conducting the experiments to be presented in this thesis. It is divided into five main sections. First, we briefly discuss the physics engine used to simulate the creature and its environment. Next, we explain in detail the setup of the creature’s morphology. Then we describe the basic workings of an artificial neural network (ANN) that act as the creature’s locomotion controller, as well as evolutionary methods of learning for ANNs. The next section then explains the genotype used to represent the artificial creature’s locomotion controller including an outline of the process which converts the genotype into the operational ANN used to control the movement of the artificial creature. Finally we discuss the choice of our common evolutionary, simulation parameters and statistical test of significance used for the experiments conducted in this thesis. 3.1 Physics-Based Simulation The accurate modelling of the simulation environment plays a crucial part in producing artificial creatures that move and behave realistically in 3D (Taylor and Massey 2001). A dynamic rather than kinematic approach is paramount in allowing for effective artificial evolution to occur. Physical properties such as forces, torques, inertia, friction, restitution and damping need to be incorporated into the artificial evolutionary system. To this end, the Vortex physics engine (CM Labs 2002) was 39 CHAPTER 3. THE VIRTUAL WORLD 40 employed to generate the physically realistic artificial creature and its simulation environment. By virtue of conducting our artificial evolution within a physically accurate virtual world, rich dynamical interactions are able to occur between the simulated creature and its environment. This in turn enables complex walking behaviors to emerge as the creature evolves the use of its sensors to control the actuators in its limbs through dynamical interactions with the environment. Furthermore, th