| Abstract [eng] |
The research area of this paper is artificial intelligence used in computer games. Specifically it is focused on methods for controlling a group of agents with a specific goal. Because of the uniqueness of individual game mechanics, those kinds of methods are usually closely related to that games environment and rules. The goal of this study is to design and test a method that could control a group of multiple agents in a virtual environment. Methods for evaluating and selecting individual agent actions in a local environment, for gathering a database of optimal solutions and for applying that knowledge in distributing agents across the environment are analyzed. A design for controlling the actions of multiple agents in a real time virtual environment is designed, based on the results. Dynamic procedural combat tactics is used to model individual agent actions in a local environment. A neural network is used to model the movement of multiple agents in an environment. It is trained using optimal solutions, generated by a genetic algorithm. Designed system is implemented and tested. Using data that the system generates, an experiment is conducted. It shows that this solution is capable of correctly reacting to situations, occurring in a real time virtual environment, and of modeling multiple agent actions in it. |
