4 ms·
Brood War AI may end up being a significantly easier problem (with this formulation) than DeepMind's StarCraft II research environment for several reasons: + A
by loser777 8y ago
Brood War AI may end up being a significantly easier problem (with this formulation) than DeepMind's StarCraft II research environment for several reasons:
+ As others have mentioned: Brood War is significantly cheaper to run (can be headless, runs well on ancient CPU), making it more amenable to the self-play with massive numbers of games approach
+ Brood War benefits significantly (arguably more than SC2) from skilled micromanagement, which is arguably easier to exploit as it doesn't require long-term/high level planning (but requires a generous actions per minute cap)
+ This API doesn't require the AI to also do simple computer vision/recognition of the game (vs. DeepMind's simplified graphical representation)
+ This API interfaces with game actions directly (BWAPI) vs. mouse/keyboard level actions
It'll be interesting to see which problem (Brood War vs. StarCraft II) gets solved to a compelling degree with learning first.
- misschresser 8y agoBrood War definitely requires a much higher APM, but the quirks of the game could make it much harder to decipher what is the correct action. It'd be interesting to see if/when the AI would discover things like muta clumping and if it would uncover new micro techniques...that would be very cool.
- ebetica0 8y agoThe other side of the coin is that the game engine for StarCraft II is signifcantly smoother, and so its dynamics are easier for models to learn. Brood war has all kinds of weird quirks and dynamics, and some "bugs" are part of the game system, so it's much harder for computers to reason about some of these kinds of things.