Spring RTS Engine

I don't think you have to get too detailed when building an AI. It is true that the AI must gather information about what the Human Player is doing in order to choose a battle plan.

I think the easiest way to build an AI would be for the AI to organize its battle plan into sets of "Operations" where each operation is essentially a queue of tasks. Each task comprises of 1 or more units comprising a "task force". Each "task force" should be able to call for re-inforcements and guard other members of its Operation, and optionally can be removed from that Operation and re-tasked into another Operation, if the Human Player is able to adapt. Obviously the AI can also perform multiple Operations at the same time which should give any human player a challenge :)

The AI would choose the appropriate Operation, given the dispostion of the Human Player's forces. This should provide a somewhat intelligent strategy.

Here are two Operations I just thought of. Anyone should be possible to create many different Operations just by using your imagination.

A sample operation:

Operation: Quick combined attack against ungarded enemy metal extractors
TaskForce1: 10 Level 1 wheeled fast attack units (can call for re-inforcements if 30% are lost)
Task Force2: 5 bombers (can call for re-inforcemnts if %30 are lost)


Another sample operation:

Operation: Build Guardian near enemy base to harass enemy structures
TaskForce1: 2 Constuctor units
TaskForce2: 10 Fast Attack Units (to guard the constructors)
TaskForce3: 5 Anti-Air units (to guard Task Force 2) (can call for re-inforcements if 30% are lost)
TaskForce4: 5 Heavy Units (to provide security) (can call for re-inforcements if 30% are lost)

Obviously a scripting language would have to be used to create the Operations/TaskForces, but this would really make development/testing of different AIs easy.


What do you think?

An interesting idea with a few major flaws. First of all, if it's reactionary, then it's going to lose. Once the player locates the flaw in the AI, he'll exploit it. For example. Show the opponent a force comprised of units that cannot attack aircraft. The correct reaction to this, is aircraft in large numbers. The response, to this is to have a second army, with significant numbers of AA in it, and it will roll over your opponents weakened ground forces.

In the first example, what it's going to do, is cause the opponent to locate his opponent, and be able to send a strike group of his own to attack the AI base. An ideal AI is unconcerned with unguarded mexxes, because it's kept up a swarming attack on the player's base.

The idea is a good one, but I don't know if it's possible to implement effectively.

I disagree...an AI should absolutely be striking at mexxes all that it can...what better way to cripple production o.O

Yes but it shouldnt rely on it implicitly to do the job.

As for what you propose, I am interested in how you would develop code to sort out the heuristics you proposed, such as determining the player disposition? Such thigns would quickly turn your AI into the sort of AI proposal that got me flamed for over Adaptive AI int he SSAI thread before I posted TAI threads.

Your AI is reactionary, and I fear that if I leave it alone it wont do anything, but I know you have more up your sleeve and that I'm not seeing everything.

Aside from that I have a 3rd path of thought going on at darkstars that seems similair to thsi however it is not a reactive AI in the sense that you've put it, and is a lot more abstract and thought out.

As for concern over putting all my stuff at darkstars out fo reach, anyone who can plainly be seen on these AI forums atm who isnt startign their own AI project in competition knows how to get there and shouldnt hesitate to ask if they want to. Just be aware of what I have said, and the no leaks rule.