In my thesis, the main goal was to create a strategy game. This was inspired by StratOnAut which is a competition between agents organized for years by AUT. The goal was to create a new game based on StratOnAut, to re-develop it, and to create a simulation environment and GUI for multi-difficulty agents. I programmed the environment in C# .NET and made some significant modifications compared to StratOnAut including but not limited to: agents need to do well in partially observed environment; fighting is not odds-based but deterministic; the game is turn-based, however, the agents can do only one act alternately therefore the initial players don’t get significant advantage. The http-protocol based communication was replaced between agents and the simulation environment. Now every agent is a dll, thus the program is much faster and the diagnostic is better for both the agents’ developers and in the simulation environment.
I created three agents for the game: one random acting agent, one hard-coded agent which follows a simple strategy, and one adaptive agent which conforms to the game stance and can learn. The agents were programmed in this particular order and I tested and optimized them using a lot of gameplays.
The game’s rules are relatively easy, but if one thinks about it one can discover complex strategic combinations. This is on purpose and an integrated part of the game. The complexity is not an expectation but a possibility for the agent-developers.