Subject
hierachical ai

Text
: : : For each state, we define a rating. : : : : What is a state? : : State (i.e. a situation) of the game. Does that clarify it? : : : I wrote a post about some of my ideas a while ago, : : : : http://c-evo.org/aiforum/2005/p-119-1.html : : I agree on the need for a hierarchical ai. However, in the example you give, the informations exclusively flows top-down. For instance, it would be the responsibility of the group manager to decide which city produces a unit. How would he decide that? This requires deciding between confliciting objectives, therefore, the relative importance of the requests needs to be specified. : : : I don't think you can get a good AI with static evaluation, there must also be some 'goal planning'. : : I agree. However, the system can be extended to respond to intermediary goals set by the upper levels of the ai (Some general strategy advisor fiddling with the weights that govern the low-level algorithms). I think many low-level decisions dealing with economic optimization (e.g. exploited tiles management, improvement construction) do not benefit enough from setting goals themselves to justify the effort such planning requires. However, this is just a gut feeling, so if anyone has a well-founded, different opinion, I'd like to hear it. : : I conclude that a hierarchical AI, where the individual levels try to optimize conflicting objectives weighted by their superiors, seems the best bet. : : I wonder what approach the more successful ai's use? : : Has anyone ever tried to name the managers and work out their interfaces? : : Regards : Adrian

Administration

Password