Starcraft AI blog

The story of AIST S2 is Velocirandom, the winner that struck by surprise. Locutus defeated 5 other opponents and looked strong, but it lost two matches to Velocirandom. How did Velocirandom do it?

Velocirandom plays random on SSCAIT, but chose protoss for this tournament where random is not allowed. Velocirandom is a cheese loving bot with tailored builds for given opponents. Here are the names of the enemy-specific strategies it was configured for:

Velocirandom was specifically prepared for everybody who was there, plus LetaBot which did not compete (and was set to face the strangely named Protoss_Zerglings build, nexus first into zealots and dark templar). The only opponents it actually faced were Steamhammer, Hao Pan, and Locutus. I have not checked whether Velocirandom’s preparation for the other opponents was as good as for these three. But I think it’s likely that each build is a timing attack or exploit to take advantage of a weakness of specific opponents. That’s certainly how I read the choice against Steamhammer.

The openings have short names, but they are long and detailed build orders. The ones with “zealot” in the name, for example, make a fixed number of zealots at fixed times, then a fixed number of dragoons, and so on. The lesson of the tournament is that even the strongest bots still have weaknesses that can be exploited by a well-chosen fixed build order.

The Velocirandom DLL is about 3.5KB bigger than the SSCAIT Velocirandom that was last updated on 18 February. That’s a reasonable chunk of code, but not huge. In the AIST vods, organizer Nathan Roth speculated that it may have some transplanted Locutus dragoon micro code. That seems plausible to me.

It was a near miss for Locutus, which had a hard road and made all opponents but one look weak. Hao Pan also performed well. I was impressed with McRave’s PvP play (and it still has those deadly scarabs).

The good news for Steamhammer is that it maintained its performance from last year even as the field grew stronger (always one of my goals): It lost 4 of its 5 games and was knocked out as early as possible. In a knockout tournament this strong, that’s within the expected range, so it’s only a little disappointing. Still, with different pairings Steamhammer could have won a few matches. Knockout tournaments are unpredictable.