Creating Artificial Intelligence Agents for Spelunky using an Artificial Neural Network and a Genetic Algorithm

The research in this paper will focus on the use of artificial neural networks and genetic algorithms in creating an artificial intelligence (AI) agent that is able to successfully handle the various problem domains that exist within an electronic game known as Spelunky. A problem domain being the specific area of challenge within Spelunky such as the domain of terrain navigation, the domain of combat, the domain of planning and others (Scales and Thompson, 2014).This paper begins by giving an introduction which consists of the hypothesis of the paper as well as why this hypothesis is possible and worth researching. The introduction then lays out the aim and objectives of this paper that need to be met so a valid assessment of the accuracy of the hypothesis can be made.A literature review is then conducted which details the concepts of artificial neural networks as well as genetic algorithms. It will also cover their implementation and an example of work they have been used in. Using games as a test bed for AI research will also be reviewed as well why Spelunky should specifically be used.A methodology follows the literature review detailing how an artificial neural network will be implemented into Spelunky as well as what experimentations will be ran and what data will be looked for in the experiments as well as how data is to be retrieved from the experiments.The results of testing are then displayed and an analysis of the results is conducted. A discussion is then conducted which details some of the issues that have been experienced in the research and what the testing data is telling us about the system that has been implemented. Conclusions are then made from all the gathered information about the performance of the AI agents created to tackle Spelunky, what limitations they have, how well met the aim and objectives were and finally the accuracy of the hypothesis of this research will be assessed