| Abstract |
The field of Evolutionary Robotics (ER) aims to understand and establish appropriate evolutionary processes, environmental conditions and physical representations for populations of robots to be able to evolve autonomously. The evolution should be effective and efficient, i.e., it should happen as rapidly as possible, and, ensure that the population will not become extinct. This scenario envisioned by ER is important when there is a demand for robots to solve problems in unknown environments, to which robots can not be designed beforehand. But for succeeding in establishing such evolutionary setup, there are several challenges and issues to be understood. Among them, there is the study and comprehension of morphological aspects of the evolved robots. Importantly, it should be possible to assess the morphological traits of evolved bodies and understand if these traits are related only to the environment and the task, or also to the encoding method and the system of components used to construct the body. In the literature, diverse methods have successfully been used for evolving morphologies of robots, but little emphasis was given on analyzing the morphological diversity of the utilized system. Using a direct encoding and a generative encoding, this paper proposes a framework to assess morphological features of robotic bodies to verify the questions: How can we measure the virtual morphological diversity of an encoding method within a system of robotic body components? Are there any morphological tendencies when using this encoding and this system? Our results showed that within the tested system of robotic components, it is possible to represent a very diverse set of morphologies, but there is a tendency for certain morphological traits or types of morphologies to happen in a population. |