Starcraft AI blog

Today I wrote about a dozen new burrow openings (mostly easy variations of existing openings). If burrow has many uses, there should be many timings to get it at. I also made a first pass at resource tracking, so that Steamhammer can finally stop expanding to bases which are mined out—and also as an input for strategy decisions. The last seen mineral amounts are a starting point to estimate the enemy’s total minerals mined, which is valuable for figuring out what they’re doing. It worked on the first try for minerals; vespene geysers are more complicated. I figure that just after a tournament deadline is the ideal time to add new basic infrastructure.

blocking an expansion

What is the ideal place for one of your units—a spider mine or a burrowed zergling or a pylon or e-bay in blocking position—to block an expansion? Steamhammer uses the center of the resource depot location, and as far as I’ve seen so do all other bots. But play a game against the built-in AI and burrow a unit there: It cheats and knows about the burrowed unit, and simply offsets its resource depot to avoid it. Krasi0 also has the skill to offset its resource depot, it just has to scan first. Mining is less efficient, but the base does go up.

Maybe the ideal place is one that forces the opponent to offset its building as far as possible from the right location. If the minerals are in a line to (say) the north of the base location, then let your blocker overlap the southernmost row of tiles where the resource depot should go. That forces a displaced resource depot as far from the minerals as possible. Similarly for the geyser. In a circular base layout, as found in the center of some maps, one or two burrowed units could block the entire circle.

I’ve seen pros place blocking buildings off-center like that to dissuade the opponent from displacing their resource depot. A bot could do a search to find the placement that forces the depot as far as possible from minerals and gas. The answer is different for a spider mine or burrowed zergling than for a building.

There are other considerations. The blocking unit provides vision; maybe you can place it so that it sees everything that comes down the enemy ramp. For a pylon, maybe you want to put it as near the natural choke as possible so that you have a better chance to send units in to defend it, or to use it to power a gateway or shield battery that might be out of enemy sight. That kind of thing.

Eh, not a big issue for any bot. But it reminds me how complicated the game is. In the long run, we have to seek methods that can weigh many considerations—and nothing stops us from seeking the methods now.