NEAT neural networks to control and simulate virtual creature´s locomotion

This paper presents the application of evolutionary computation techniques for evolving behaviorisms in virtual creatures existing within a realistic virtual environment that are subject of the constraints as defined by the Newtonian model of physics. Evolutionary computation technique uses NEAT´s network that is based on three fundamental principles (Genetic Encoding with Historical Markings, Speciation and Minimizing Dimensionality). The creatures´ morphology is completely predetermined and is designed to elicit a variety of locomotive behaviors and test the generalization abilities of our framework. Three different morphologies are introduced into the simulation; each morphology represents an entirely different species of virtual creatures. The experiments show that NEAT´s network is able to generate efficient locomotive behaviors.