Seminar: Automated Agent Design for Playable Scenarios in Computer Games

Scott Watson
Ph.D. Oral Comprehensive
Supervisory Committee: Dr. Wolfgang Banzhaf, Dr. Andrew Vardy
and Dr. Oscar Meruvia-Pastor

Automated Agent Design for Playable Scenarios in Computer Games

Department of Computer Science
Tuesday, Oct. 22, 2013, 2:00 p.m., Room EN 2022


Computer games are complex environments which often feature autonomous agents that must carry out challenging roles. Consequently, trying to find ways to create better agents has been the focus of much research. Despite these research efforts, very little advanced artificial intelligence has been utilized in commercial video games.

This proposal first reviews relevant background material and related work to establish the necessary information to formulate a novel approach to improving artificial intelligence in computer games. It is suggested that formulating a novel approach can be done by automating parts of the process of producing computer game agent controllers. The typical approach of trying to improve artificial intelligence in video games is to create new agent architectures. To illustrate that the principle underpinning this novel approach is worthy of investigation, a proof of concept experiment is described. To do this an evolutionary algorithm was used to generate finite state machines to control agents in a primitive computer game.

In this research project a system will be developed to automate the creation of agent controllers for certain kinds of computer game agents. Representation schemes for the input of game-specific information and the specification of desired output will be developed and tested. Extensive user consultation and testing will be carried out to validate the results of the project.