Using Grammatical Evolution to Parameterise Interactive 3D Image Generation

Abstract

This paper describes an Interactive Evolutionary system for generating pleasing 3D images using a combination of Grammatical Evolution and Jenn3d, a freely available visualiser of Cayley graphs of finite Coxeter groups. Using interactive GE with some novel enhancements, the parameter space of the Jenn3d image-generating system is navigated by the user, permitting the creation of realistic, unique and award winning images in just a few generations. One of the evolved images has been selected to illustrate the proceedings of the EvoStar conference in 2011.

Size Does Not Matter: Evolving Parameters for a Cayley Graph Visualiser Using 64 Bits

“…This is an extension of previous work [7], which created some award-winning visualisations. One won the Evolutionary Art competition at the EvoStar 2010 conference [4], and was subsequently chosen to illustrate all the proceedings associated with EvoStar 2011 [13].…”

“…There are several reasons for using GE, both in the original work [7] and in the current approach. The use of grammars really simplifies the representation of the search space, allowing for an easy to use typed version of GP.…”

“…This technique is especially useful when domain knowledge allows the apriori identification of blocks of information: special non-terminal symbols are introduced in the grammar, and the GA is then limited to these crossover points. It has been used in the past to separate well-defined behaviours for video controllers [12], and also in the previous application of GE to Jenn3d [7]; previous work has even allowed these crossover markers to evolve their position [8].…”

“…The user can control the evolutionary process through the fitness score. A fitness score of 1.0, for example, guarantees that the current visualisation is replaced by a random one in the next generation, whereas a score of 5.0 guarantees that the visualisation is passed unchanged to the next generation, thus potentially being used as a seed for alternative, similar visualisations (as per the scoring process proposed previously [7]). Table 3 shows example projections, achieved through evolution, along with their hexadecimal, integer and mapped strings.…”

Evolutionary Behavior Tree Approaches for Navigating Platform Games

“…This technique [42] uses a special grammar symbol (<GEXOMarker>) to label crossover points; the search algorithm then only slices an individual according to these points. It has been used in previous iterations of this work [4], [5], as well as in the evolution of 3D projections [43].…”