Evolving Multi-line Compilable C Programs

We describe a Genetic Algorithm called Grammatical Evolution (GE) that can evolve complete programs in an arbitrary language using a variable length linear genome. The binary genome determines which production rules in a Backus Naur Form grammar deenition are used in a genotype to phenotype mapping process to a program. Expressions and programs of arbitrary complexity may be evolved using this system. Since rst describing this system, GE has been applied to other problem domains, and during this time GE has undergone some evolution. This paper serves to report these changes, and also describes how we evolved multi-line C-code to solve a version of the Santa Fe Ant Trail. The results obtained are then compared to results produced by Genetic Programming, and it is found that GE outperforms GP on this problem.