抽时间大概翻了下那几篇paper，这个问题挺大的，也挺棒的。首先要把游戏里已有的实现和学术圈分开讨论：
游戏里的实现一般只用有限状态机者Behavior Tree，条件触发罢了，开销很低，不会出现“按照PC跑7个AI来设计”的情况，也谈不上深蓝式的竞争。
那篇phd thesis列举了一下RTS游戏：Ancient Art of War, Herzog Zwei, Dune II, Warcraft, Command & Conquer, Warcraft II, C&C: Red Alert, Total Annihilation, Age of Empires, StarCraft, Age of Empires II, Tzar, Cossacks, Homeworld, Battle Realms, Ground Control, Spring Engine games, Warcraft III, Total War, Warhammer 40k, Sins of a Solar Empire, Supreme Commander, StarCraft II.
研究集中在SC1的原因是"it helped define the genre and most gameplay mechanics seen in other RTS games are present in StarCraft“，而且”based on strategy than tactics, by opposition to the Age of Empires and Total Annihilation series “， 还有不错的API。
主要挑战包括：
• Planning
• Learning
• Decision making under uncertainty
• Spatial and temporal reasoning
• Domain Knowledge Exploitation
• Task Decomposition into Strategy, Tactics, Reactive control, Terrain analysis and Intelligence gathering
不过11年的比赛是这样的，水平很低吧，然后还经常crash
Xelnaga (Protoss): is a modification of the AIUR bot
that chooses the Dark Templar Opening in order to destroy
the enemy base before defenses against invisible units are
available.
Protoss Beast Jelly (Protoss): always goes for a 5-gate
Zealot rush, supported by an effective harvesting strategy
named power-mining (2 probes are assigned to every mineral
patch, thereby needing 18 probes for 100% saturation in a
normal map, prior to expanding). Gas is not mined as it is not
needed for constructing Zealots.
EvoBot (Terran): employs an evolutionary algorithm
for obtaining rational unit combinations and influence map
techniques for deciding the strategic locations. Note that this
bot was submitted in a very early version, with many of its
designed features not yet fully ready.
而且人类建筑卡入口都能发一篇论文了，M. Certicky, “Implementing a wall-in building placement in starcraft with declarative programming,” 2013.
因此反驳一下楼上观点，现在rts ai有learning，也有人做（有比赛，研究也能发IJCAI，会议还有专属的CIG和AIIDE），但是目前state of art 算法的确打不过人类。