This master thesis presents an approach for how to implement an Evolutionary Multi-Agent Potential Field (EMAPF) based AI in small scale combat (SSC) scenarios in real-time strategy (RTS) computer games. The thesis will show how evolutionary algorithms can be used to auto tune and optimize potential field parameters for making unit agents able to navigate and hunt down enemy squads/troops in an ecient way in an unknown environment.
The very popular top-selling RTS game StarCraft: Brood War has been chosen as test platform, mainly because the game is extremely well balanced and has a complexity in the game world, which makes the development of AI's very challenging. Since StarCraft is not open-source, the free open-source framework Brood War Application Programming Interface (BWAPI) has been used to communicate with the game.
In order to make the AI able to control unit agents moving around in the game world, they will be effected by different types of potential fields placed in both static and dynamic tactical places around in the game world. One of the hurtles when designing the potential fields has earlier been the tuning of them, which can be time consuming, if done manually. This thesis therefore presents an approach for how auto tuning with the use of evolutionary algorithms can be implemented into the Multi-Agent Potential Fields methodology. The work that has been done in this thesis provides a good basis for future work for designing EMAPF based AI's for full scale RTS games. The potential fields found in this thesis could be adapted by an AI playing full RTS games, StarCraft in particular. The AI will be able to turn the potential fields on and off, depending on what tactics are found to be best suited to different states.
Evolutionary Multi-Agent Potential Field based AI approach for SSC scenarios in RTS games. Master of Science Thesis. 2011