Creating Complex Building Blocks through Generative Representations

One of the main limitations for the functional scalability of computer automated design systems is the representation used for encoding designs. Using computer programs as an analogy, representations can be thought of as having the properties of combination, control-flow and abstraction. We define generative representations as those which have the ability to reuse elements in an encoding through either iteration or abstraction and argue that reuse improves functional scalability by allowing the representation to construct buildingblocks and capture design dependencies. Next we describe GENRE, an evolutionary design system for evolving a variety of different types of designs. Using this system we compare the generative representation against a non-generative representation on evolving tables and robots and show that designs evolved with the generative representation have higher fitness than designs created with the non-generative representation. Further, we show that designs evolved with the generative representation are constructed in a modular way through the reuse of discovered building blocks.