An AI Competition to improve RTS AI

[Conzar]

Here is an interesting article on SC1 AI Competition.

From the article:

These openings are well optimized, and it’s tempting to simply choose one or several and hard-code the agent’s macro planning as a set of scripts or finite state machines. This is how StarCraft’s built-in AI and many other game AIs work: human expert knowledge is encoded as an explicit sequence of actions, with pre-defined transitions. Build tanks unless the enemy builds air units, then build anti-air units. If the enemy builds cloaked units, build detectors, and so on. ... Perhaps the first thing we learned was that this is not the best approach.

How does the AI work in Glest?  And could Glest utilize the research done by these students to better the ai?

[Yggdrasil]

I read this earlier today. It's a nice article.

Glest's AI is just rule-based, a finite state machine. There's no machine learning or plan optimization involved. Keep in mind that such concepts a quite resource consuming and need much expertise to get right. And much time...

[hailstone]

I completed a game AI unit at uni. I think GAE could implement these techniques eventually.

I think what they're doing is interesting but the overall strategy is really a fancy (although more believable) way of doing resourceMultiplier * n since they use techniques that a human player can't do well such as making better pathfinding and micromanaging Zerg Mutalisks so it's exploiting the game mechanics like we do. It was the same for the runner up:

There were some surprises as well. Krasi0, the runner-up, was another very capable agent, and it produced a behavior we never expected because it’s not seen in human matches. Terran workers have the ability to repair mechanical units and buildings, but this ability is rarely used in combat because workers are fragile and having enough to make a difference in combat requires too much micromanagement from the player.

In the end, as game developers, we shouldn't be concentrating on creating an AI that can win but rather creating one that can lose convicingly.