This project was designed to compare and contrast Evolving Behavior Trees (EBTs) with NeuroEvolution of Augmenting Topologies (NEAT), a Genetic Algorithm for the evolution of Artificial Neural Networks. We used Super Mario Bros. as a benchmark to compare these two techniques. The results showed that NEAT had a slightly higher maximum fitness while performing poorly in all other comparisons. EBTs performed strongly in rise time, evolution time, generalization, and complexity.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.