Starcraft AI blog

I rather like Arrakhammer. It quickly differentiated itself from its parent Steamhammer and now plays quite differently. And its strategies are logical.

Look how Arrakhammer defeats Wuli: It holds off the raging zealots with mass sunkens and no units, objectively poor but perfect against Wuli. Then it techs up to hydra-lurker. Wuli is unable to cope with the lurkers and falls over.

XIMP by Tomas Vajda follows the same abstract plan of holding off the enemy with mass static defense and following up with tech units that the opponent cannot cope with. Some similar plan probably works against most non-adaptive bots (it might fail against bots which always make tanks or other siege units). The plan seems so general and effective that I’ve been thinking of coding it into Steamhammer as a use of opponent modeling. When Steamhammer realizes from experience that the enemy bot has a fixed unit mix, it may build mass static defense while it techs up to a countering unit mix.

overhatch versus overhatch

Arrakhammer is based on the previous, 1.2.2, version of Steamhammer, which did not have the overhatch opening yet. But the author has not been shy about looking into the newer 1.2.3 version. Here in Arrakhammer versus Steamhammer on Icarus, Arrakhammer borrows the overhatch opening and tries to tweak it for an advantage over Steamhammer’s variant of the opening.

The openings are identical up to supply 15. Steamhammer spent its first 100 gas on zergling speed, since it was playing a zergling opening. Arrakhammer instead started a lair and added a sunken to hold the zergling pressure. It’s not the natural way to play the opening, but it is a logical attempt to cross up an opponent that is committed to zerglings.

The plan commits Arrakhammer to defense until mutalisks come out, so it brought its zerglings home even before the sunken was started. Steamhammer had made 2 zerglings more (which didn’t matter because they were far away) and Arrakhammer lost 3 as they retreated up the ramp (which did matter). Unfortunately, Arrakhammer defended inaccurately and lost 2 drones, a severe loss. You can see the orange zerglings out of position in the picture. Good defense could have saved all drones.

Play was left, but in the next zergling attack Arrakhammer pulled drones too far from the mineral line (a classic Steamhammer blunder) and lost 5 more. After that it was irrecoverable. A picture of the drone blood:

The opening try was logical and it could have won if Steamhammer had made the mistakes, or had chosen an inappropriate build. Its weakness was that it depended on correct defense. Bots are strong in attack and weak in defense; that is why Steamhammer is aggressive whenever possible. As defense grows stronger, a wider range of strategies will become practical.

Steamhammer repeatedly threw away zerglings trying to approach the sunken and made other mistakes, but its advantage was unbreakable and it finally won by getting mutalisks first.