A distance and diversity measure for improving the evolutionary process of modular robot organisms

In this paper, we present a novel diversity measure for a population of modular robot organisms. Evaluating this diversity we will get a prospect of the fitness of future generations of organisms in an evolutionary process. The degree of the diversity in a population is a measure for the adaptability to the environment. If the diversity is too low, it is necessary to change evolutionary parameters such as mutation probabilities or to start a new evolutionary run in order to get better adapted organisms. The novel measure helps to improve and speed-up the evolutionary process, which is mandatory as the evolution of robot organisms, specially the coevolution of controller and topology of organisms is a very tedious process. In four different experiments using the Robot3D simulator and the Symbricator modular robots, we will show the usefulness of this diversity measure. The goal of the first two experiments will be the evaluation of two different selection strategies: fitness proportionate and rank selection. In the last two experiments we will explore the differences between a restrained evolution, where the type of agent, which controls an element in an organism is predetermined by its parent element, and a free evolution.