How AI is balanced across different type of races in RTS games?

There are some real-time strategy games (like StarCraft 1 or 2) where AI needs to be balanced across different type of races (in SC it’s Protoss, Terran, Zerg), so despite they’ve different type of troops to play, no specific type of opponent have any advantage. Not to mention that they need to consider different type of maps, so in overall each AI playing different race have the same changes of win.

How in general this balance can be achieved by gaming companies? Is it just by playing massive amount of games headlessly with different setups?

This study and this one plus this dataset may help answering it, but it could be not related.


Common ways to balance RTS AIs:

  1. Have all races be the same but with a different “coat of paint” on each races.
  2. Have the races be different but make sure each unit has an exact counter in each other race (rock-paper-scissor)
  3. Have the AI cheat
  4. Don’t balance them (You get to beat some races AIs quickly while others take more practice.)

Some of the different way AIs are programmed to cheat to increase the difficulty:

  • No fog of war for them
  • They know what you are producing so they build the counters as soon as you start your own production
  • They start with more resources
  • They collect resources faster
  • All AI races ally against you
  • They can micro-manage better than any human can, adjusting the micro-management latency to adjust AI difficulty.

There usually will be the regular amount of play testing in which the races might be tweaked but overall it does not matter if the player always wins against one race: If the player wants a higher challenge then the player can pick a different race as it’s opponent.

It is also very difficult to know during testing if one race is weak or if the player playing style in the case study sample just happen to be good against that AI race.

In the case of E-Sports RTS games there is usually no AI in matches and companies do not divulge their tweaking methodologies but from my observations the races get tweaked to make the E-Sport broadcasts more interesting, keeping the ratings up. With less focus on the player-vs-AI experience since the big money is in the PvP experience.

Often using a combination of in-depth analytics to know what to adjust and trial-and-error for how much to adjust.

Source : Link , Question Author : kenorb , Answer Author : Stephane Hockenhull

Leave a Comment