Hi chaps. Can I beg a little help here? I'm not a newb to general AI programming stuff, but I just can't my head around where to start with putting a decent AI together for my current project.

It's a turn-based strategy game using a hex board and all movment is per tile. To visualise, the closest game I can think of is DS/Advance wars and actually the AI in that is pretty much exactly what I'm trying to do.

Here are my key problems:

1) Each AI team needs to be able to make basic tactical decisions based on a triangle of desire between attack, defend, expand. I can handle this bit as far as getting to the point where I give individual orders to the units

2) Each unit needs to have simple goals such as "move to that hex", "operate the building (to take control of it)" etc.

3) This is where I'm tripping up. A lot of the required behaviour needs to be scripted, and then given over to the AI to fulfil the scripted behaviours.



Here is a simple example of my problem.



Level X. A load of enemy tanks get spawned by a script, popping onto the beach from a carrier. The gameplay in this level is that all the enemies then move through a trench to a target spot where they get evacced by helicopter (another script with a trigger box). The human player has to mullha most of the tanks to "win" that level.

The actual movement through the valley needs to be done by the AI as I would like it to react to threats (if a tank on the ridge gets to the edge it gets within firing distance, so shoot at it), but basically keep on making progress to the extraction point.


What I'm looking for is a method that allows some goals to be given (like my move to helipad) by hand, whilst other behaviour can happen based on circumstance.

Are there any resources I should be looking at here? I don't expect someone to tell me how to code this, but it would be nice to find the right book or article that sheds some light on this specific kind of AI behaviour.

Sorry for the long post and thanks in advance...

I don't think the previous poster addressed what it is you are trying to accomplish. If I understand, you are trying to make it so that the AI alternates between being self-sufficient and scripted. That is, have an overall goal, but be willing to suspend the execution of that goal while you do something that is more pressing.

I would consider doing things like hierarchical finite state machines (HFSM). The other thing you can do is simply a FSM with hysteresis. That is, you keep a note of what your prior state was so that you can return to it when you are done. The state change would be from "travel to my goal" to "defend myself". In "defend..." you can do whatever you need to do to fight. Upon conclusion of that action, you return to "travel...". That's the quick way.

Another way to so push actions onto a stack with the most recent being the top of the stack. When that action is solved, it gets popped back off the stack and you see what action is under it (now at the top of the stack). The benefit of this is that you can have many layers of goals that keep interrupting higher ones.

Thanks. I went for a fairly simple approach in the end.

I now do my strategic goals stuff in an "team" AI manager and they issue various orders to the units under their command (or they will when I get to it).

But also for scripted units they do use a small list of goals that are given by the script and marked as scripted. These units get no help from the team AI at all now, they just pick a best thing to do from the hand given goals.

I got this working over the weekend and its fairly convincing given that the level in question is more about beating a set environment. Now its on to the wider problems of making the team AI give sensible orders...

Thanks again for the help, was most useful..

hey Rubicon,

a useful site for you might be www.aigamedev.com loads of begginner video tutorials, interviews, master classes, case studies they even have their own sandbox going on. Hope you find it helpful.

Mark

Quote:Original post by AeroAstroArtsLtd
hey Rubicon,

a useful site for you might be www.aigamedev.com loads of begginner video tutorials, interviews, master classes, case studies they even have their own sandbox going on. Hope you find it helpful.

This.

While it may seem awkward for me/us/anyone to pimp AIGameDev.com here, this has been discussed with the powers that be here at GameDev.net a long time ago. They are cool with it since the target audience differs quite a bit. That being said, AIGameDev.com is the place to go for a lot of what AAA mentioned.

Thanks guys, I'll give that a bash.