Its hard to fit units into roles. How much should a ship cost? How much damage should it deal? Over which units is it good at now, what should it be good at the end? Are all hard question pretending to proper balanced RTS game.
Its hard to reach this balance. One needs many players and many play tests to get close and very impartially record keeping. Its really hard to do in a complex strategy game. Yes we generate charts. From charts try to guess on how ships behave in different situations. The guesses are not that accurate! We are still in the dark on how the ships will evolve to be played. Yes we play the game. But the amount of playing we can do is very limited. We need real players but the game is not fully finished. We cant unleash this half backed game on the public yet. What do we do? We cheat.
I wrote some simple yet ok AI's to play the game (Artificial Intelligence). They test individual ships and combinations of ships to see if they are under power or cost too much. Those AI's can play the game very fast. Faster then any human. The game runs almost x100 time the normal speed. 15 seconds and entire 20 minute match has been played.
Its writing tools like this that takes time. But having such tools at our disposable helps us create and try out new theories for unit balance and new game play strategies. Hopefully the AI would grow and tactics it uses become ever more interesting.
But 2aw is designed to be played by people in that its still quite complex for AI of any complexity to play well. The AI choose sub standard tactics and in the end of the day is not a full sub-statute for players but at lest it will get the seat worm.
Maybe when the game is launched we can place bets on the AI vs AI battles :)
Comments:
Maybe when the game is launched we can place bets on the AI vs AI battles :)
if you've reached this point then you have written one of the best AI's, probably
testing balance with the AI only works for a very small part because players tend to completely use things (ships in this case) very differently.....
a better use for testing with the AI is to see if the AI needs more work