AI: Attacking from multiple directions

Hey,

I just saw the latest lifestream on facebook with Sorian. One of the things Sorian did great in sup com 1 was deciding what to build. But one of the things he couldn't do due to limitations of the game engine I suppose was make the enemy smart enough to attack from a different direction.

Let met put this with an example. It was perfectly possible to build a forward base with a lot of static turret defenses a shield and stuff and nothing to the left, right or front of it. The ai would just send units and units and units to that one forward base and get annihilated. While all it had to do was go a little bit east or west bypass the forward base and blow my production facilities to pieces.

This question didn't seem to popup in the lifestream. So therefore I'm asking, if this is been taken into consideration in the neural networks. Cause they will probably learn to shoot the best optimal unit and learn to retreat, but what about ow wait ... let me change from direction ... maybe I can do it like that. I dunno some kind of integer count, try pathing a 3 times, if fails , calculate a new path away from that danger zone and try 3 times from there, rince and repeat ....

Just from the top of my head ofcourse, it's probably a lot more complex then this.

Thanks for that link Sorian, is really educational. Reminds me of a post of a guy I read one time on the Spring RTS engine.

He imagined to divide the map in several squares and then on regular intervals would let an ai engine determine which squares where high value targets and which where danger spots. He would then let the pathfinding find the best paths to the high value targets and avoid the danger spots. He seemed like a pretty smart guy. But now that I think of it , he seems to have been describing influence maps. I also think his system would only work with a cheating ai that knows everything at all times.

Anyhow great you are on this project I enjoyed your ai's a lot in sup com and tried fiddling with them myself at one part in time

Some other AI woes that come to mind. I've played quite a few of RTS vs AI in my time and there are some points that I do like to bring up just in case you might haven't thought of it.

- Earth 2150 series : They had a pretty decent and fun AI in the pregame, but for some reason after an hour or two of play, the AI stalled and stopped attacking. It was still building a moderate amount of units but they just sat there waiting in the base. The endgame vs AI was usually relax build enough forces and go smash. Same was true for 2160 game. Lets call this AI stalling

- Sins of a solar empire : Another example of a fun ai in the first stages of the game wehre the cpu is just better because well a human can only do so much things at one time. However at a certain point in the game when superstructures came into it, the ai stalled as well. It just stopped building big armies and was more focussed on building the superstructure static defenses. Like it cannot manage it's resources lets call it resource starvation

- C&C generals/ AOE 2 : Two games with finite resources that in the pregame leeched the entire map dry, all you had to do was build good enough defenses where it would rush itself to death on , and in the endgame when it's out of resources , just build a bunch of troops and go smash. I don't think this will be a problem in PA since we don't have finite resources right. But another prime example of resource starvation.

- Spellforce 2 : The ultimate cheating ai In all hounesty I don't think I ever beaten this ai ... It was able to build its troops so darn fast in the pregame, I never got very far except build a few barracks ... Maybe I just sucked at the game ofcourse ... but darn that ai was great (at cheating). So cheating ai's are good TA needed them for example. But well overdoing it is just frustrating

There these are a few examples of AI stuff that is pretty important imho Maybe other people can add to this list ?