For Backers Only: AI building stuff

Sorian, is there a way to program the AI to micro large armies such that they can avoid defensive nukes? Maybe a parameter that indicates if the enemy player being attacked has nukes or not, then doing the micro part after.

Easiest way to do this is to split the army into 3-4 pieces once they get within a certain distance from the target, then converges on the defenses once they get within range. (at least, it is as a player).

Would definitely make it much harder to stop the AI late game.

just had the AI walk his commander into range of a few bombarding ships (4 narwhals and an orca) and died.

self-preservation could use a little more tuning.

AI seems much better about producing T1 factories in this build, it got its first three up in what I thought was a reasonable timeframe. Still too slow to T2 tech, but I'm guessing that until balance settles down that isn't a priority.

Still much too turret happy early on though.

Would you please develop on this idea? Does it invest too many resources into turrets relative to mobile units, fabbers and factories?

You don't need a lot of turrets around a base early on, that's what the Commander is for, so putting resources into this instead of expansion, energy, further production, etc. is putting you behind.

The fabber to mobile unit balance is going to vary from patch to patch. It definitely doesn't assist enough on T2 factories and is far too air fabber happy. Air fabbers should make use of their speed to do remote work, keeping them in your base is both an inefficient use of resources (use a land fabber) and also begging for a couple of fighters to do a flyby and blow away your build power.

Same for me. Three games against the AI and all won because of commander walking into my towers or attacking armies. First it made sense to stop some T1 but the com kept doing it till late game when he had to face T2 armies. So he slowly lost health and died at some point.

Soon?

Sorian, neural nets mean that the AI 'learns' right? It figures out how to deal with situations sensibly?

http://www.gdcvault.com/play/1015667/Off-the-Beaten-Path-Non

http://soriandev.blogspot.co.uk/search/label/neural network

The AI uses Halleys. I just found out the bad way.
Literally the last second before his commander got satellite-sniped by me, he sent his homeworld to mine.
He just suicided on me.
My only thoughts were "AI did WHAT?!"

Fun Fact: If you destroy a large part of the (easy) AI's base, without killing it, it will literally give up.
It doesn't blow up its commander or anything, its commander just stops moving.
I haven't tested it on normal or hard yet, but it's kind of funny to see happen.

Just noticed the AI doing something odd.

in the current PTE build, I attempted to invade the AI's planet, though was eventually pushed off. This left a whole bunch of anchors in orbit however. The AI kept on sending hoards of ground and air units to the location, being completely devastated by the anchors with no chance whatsoever of succeeding in killing the anchors. It appear that the anchors were being calculated as a threat, however it was sending the completely wrong units to kill them.

As funny as it was to watch hoards of units get ripped apart by a dozen anchors, it probably shouldn't be like that.

It does it with anything in orbit. It will send a horde of bombers to deal with your radar satellites.

In 64498, AI builds structures on metal spots.

They're not going to put down their guns and start a mining enterprise, are they?

When can we expect to see control systems in place for the AI/pathfinding

(E.g. I played a very cracky desert planet. The AI sent all of its units to get stuck on a crack in the shortest distance between our bases. They accumulated there for the entire game.)