SBIR Phase II: Evolving Object Neural Networks

Natural Selection, Incorporated, San Diego CA

Investigators

Abstract

This Small Business Innovation Research Phase II research project will investigate the problem of generating evolutionary object neural networks for controlling characters in classes of entertainment software, with consideration given to genres of massively multiplayer online games. The objective of the research is to identify and develop general self-adaptive routines and software tools that can be incorporated in a software developer's kit (SDK) that is suitable for licensing to third-party developers. A series of experiments conducted within a statistical framework will identify first- and second-order effects of parameter choices for the evolutionary control of game characters, which will be incorporated into the SDK. R&D will be aimed at generating the most rapid evolutionary learning for game characters while having the smallest code "footprint." Additional research will facilitate automatic play testing and optimization of artificial intelligence in games. The scientific and technical understanding of hybridizing evolutionary computation and neural networks will be enhanced by the careful study of the nonlinear effects of parameter choices in the studied settings If successful this product will ease the transition of video games from development to products. The development of an SDK that will help reduce the time and cost of segments of video game production by 50-80%. The software developed may serve as educational classroom aids in university courses. Furthermore, the strong correlation between video games and military simulations suggests important contributions to dynamic planning in combat simulations, as well as extensions to optimizing courses of action in business operations, such as supply-chain management.