| Convergence and stagnation are two of the biggest obstacles to the development of open-ended, evolving systems. These issues have received much attention in both traditional biological research and in the field of Evolutionary Computation. Here I study their effects on evolution in 3D virtual worlds, with a focus on the evolving population's ability to continuously produce new, nontrivial survival strategies. I pay particular attention to measures of convergence, stagnation, and fitness, which provide an avenue for comparison among distinct evolving systems. Toward this goal, I designed and utilized an evolutionary system called WhirlingDervish that evolves 3D morphology and behavior to examine these issues.;Convergence and stagnation are clearly related qualities. I define convergence as the reduction in genetic diversity among individuals in a population, and stagnation as the population's lack of progress towards the evolutionary goal over a period of time. The process of measuring these qualities in a population draws upon methodologies commonly used in biology and evolutionary computation, including measures of genetic diversity and both genetic and phenotypic variation. I test how informative these methodologies are toward approximating the level of continued evolution in WhirlingDervish and make use of their results to compare potential fitness functions. Finally, I look for correlations between these measurements and the structural and behavioral qualities of individuals in the final population. |
