Lord Kailen,

Sounds like you selected the adaptive AI (or maybe the tech AI?) and it choose a more tech based approach. That’s part of the reason I choose to mod the existing AI... It has many tactics built in.

Ajunta,

I completely agree about the expansion bases - it builds too many and also on 5km maps tries to start them in the opponents base! That’s one of the things Im going to look at next along with the other confirmed bugs above.

yes, i almost always choose adaptive ai, unless i'm testing something and i don't want the ai to bother me, then easy or normal.

i've also noticed the ai trying to build in someone else's base, but i didn't know which ai was doing it.

Hi, since you've been looking at the stock AI, a question.

There's been speculation that the horrible performance issues some players are experiencing are actually because they play vs. AI instead of multiplayer.
There's talk of slowdowns happening when the AI gets "stuck" during late game when it's used up the unit limit.

Have you seen any evidence of this in the code? If so, can you fix it?

As you probably know, the AI in FA has been significantly redone from what was in SC. This means it uses a large number of functions, threads and checks to run the AI which is of course CPU intensive.

So it depends what kind of issue you are talking about...

If you are referring to a problem where the game runs slow as soon as you start a game with an AI in then this is probably a bug and I may be able to fix it (although I havent seen anything like this accept in other AI mods).

If you are referring to the way the games gets slower and slower as more and more AI units are in the game, I dont think this is something that can be fixed - its just consequence of having a complex AI. Its worth noting I may have inadvertently improved the situation by changing the guardbase AI or changing to attack force platoons as these where doing some odd things.

Also if you have a reproducible situation I can take a look and see if it can be improved.

Well, I dont play the AI, so I was just curious. Apparently when the stock AI hits the unit limit it sometimes gets into a state where it cant do anything and performance plummets. I dunno, havent seen it myself.

In that case my fixes may help because part of the reason it hits the unit limit is because it bunches a whole lot of units at its base. I have fixed this. That and the attack force changes means it will do more attacking... again reducing the number of units it has and the chance of hitting the unit cap.

So maybe...

0.6 released.

Just a minor update to make sure some defenses are up before shields plus a couple of other minor tweaks to previous fixes.

Duncane, this is a continuation of feedback in Sorian's thread. The adaptive AI builds naval units aggressively but does not deploy them... with one exception. Cybran destroyers are immediately given attack move orders, with waypoints, that generally end up on land somewhere. I'm still testing Naval behavior on Concorde Lake.

In truth none of the AI (no matter what difficulty) use ships properly, or at least in a manner you would expect. They use them more as a shore line defence, so they never go to attack any enemy.

Version 0.7 released.

Okay ...this is hopefully quite an interesting release.

I have made some tweaks to the way the AI uses Navy forces. The AI should now send Navy AI attack forces sooner and show not over build sea factories.

A minor fix is that the AI will now produce Aeon attack boats. These were left out in the standard FA AI.

Good maps to try the naval changes on are Roanoke Abyss and Seraphim Glaciers.

Also I think I have fixed the Czar AI behavior. It now wont leave its beam on while it moves across the map to attack.

For land attacks I have made some tweaks to the early attack platoons and the engineer guard squads that I hope give the AI a boost in early map control.

Let me know how it goes.

Just gave .7 a spin on Concorde Lake. The Naval units for an adaptive AIX still stayed at home, but performance improved significantly on the map. The other thing I noticed is that the adaptive AIX, and the Sorian AIX it faced, never built any shield structures. Both sides were UEF.

The most important point here is that performance was great. I had game time set to fast, and it never slowed down (even when nukes were flying all over the place). I usually get slow down somewhere in the T2 phase on this map.

edit: completely different experience on roanoke. The AIX rush built and deployed naval units for the entire game. A few of the other AI teams sent out some naval units as well. Shields were built and performance bogged down in late T2. The AI teams had difficulty building because their bases were clogged up with units. Only the tech AI failed to build shields. I used every AIX type available with every side as UEF.

I played a few more games and the rush AI is by far the best at utilizing naval units. Naval behavior is much improved over the stock behavior. The AI still keeps a lot of units at home, but at least it is sending units into battle.

Thanks for the feedback and the tests mocafrost.

Im surprised about the performance gain. I didnt do anything specific for that but possible the fact it uses it Navy more means less naval units in play.

Concord lake is not really a navy map. I reckon if I was to play against a player that I would still heavily rely on land units and a lot of air units and transports. So the rush AI will still use a air based approach on that map.

Maps like Roanoke and Seraphim Glaciers are really what I call "naval control" maps. Glad to hear the AI are deploying navys on those maps.

Its still keeping to many navy units at is base by the sounds of things. I will take a look at that next.

AIs still create lots of useless land units even when they are on a small island (I tried yesterday). I think it's the cause of the sim slowdown with water maps.
The sim speed fell to -6 before we started to kill AIs.

Thanks for the info Youri....

What map were you on?

Does this mean the AI somehow starts to thrash when it's stuck and cant do what it wants? (like how a bad pathfinding algorithm would keep searching and not give up, and use lots of processing for no purpose)

AdmiralZeech, yes it does. Im pretty sure what you describe is what happens. Im going to take a look at it. For the naval AI it really only needs minimal land units or at least needs to build and use transports.

For everyone else:

I would be interested to know how people fine the cheat level in this mod for the AIx type AI's.

The standard without the mod is a 2x resources and build speed. In the mod its set at 1.4x. Im still finding this a bit hard and think maybe 1.3 would be better.

Anyway, let me know what you think.

Can you make your mode compatible with Sorian's "Cheating AI Adjustment Options"
It would be great..

sh6olin,

If you rename the cheatoptions.scd file to 000_cheatoptions.scd it will override the cheat settings I have made in this mod.

Having said that I would really like to know what everyone feels is a reasonable default to practice against. I dont think AI should have omni all the time and I think the 2x build/resources is too much, but I hope we can choose a good default....

1.4 to 1.5 seems about right on the sort of maps where the AI puts in a good innings.

Lurkah,

I find 1.4 to 1.5 AI bonus quite hard even on small maps like Williamsons bridge. What maps do you play it on?

Larger ones really.

Interestingly, I just played a 2 v 2 with an adaptive (your fix 0.7) on my team and with two sorian .9.1s on the other side. On Drake's Ravine.

Adaptive was quite aggressive in getting up front, capturing stuff etc. Then Sorian showed up in force and beat it all the way back to it's base and I had to send an experimental over to save it.

It then stayed there without expanding all game but it did finally start sending T3 platoons and fired a couple of nukes (again the standard nuke-cheat where it's fired at enemy commander by magic). Is there some sort of code that gives up resources and expansionism at some point? I noticed it wasn't upgraded Mexs again but then Sorian does that too.

I've seen this a lot with the FA AIs, if they have a chance to bounce back they don't take it. They continue to try build stuff as normal, despite a lack of resources. I can't help but feel it would play better a lot of the time if, when on a resource crunch, it thought harder about resources and pumped out engineers more often.

It's interesting though. Smaller maps, 20x20 and Sorian regularly gets killed by Adaptive. I put it down to early expansion and aggression while Sorian is slow to start and doesn't build defences very well.

It seems right now that's how things pan out:

1. Small(er) map, adaptive wins.
2. Large map, sorian wins.
3. Water map, sim speed -10 no one wins

Over time I would expect Sorians to beat the standard AI's. My goal was never to make the stock AI's "the best", just to make them usable in games and to fix any major bugs. You can always make them harder to play against by using the cheating versions.

That said, the water map bug is a real killer and I really should put a fix in for it. I will try to look into it when I have some free time in the next few days.

I tried to duplicate my performance gain using your AI. I just wanted to make sure that I wasn't crazy. I used Concorde lake again with a Seraphim Sorian AIX on the left, and a UEF AIX Adaptive on the right. I set game speed to "fast" and tried four runs alternating your mod in and out.

Sim speed without your mod:
0 at 10min, -1 at 12min, -2 at 16min, -3 at 22min, etc.

With your mod:
+5 at 10min, +3 at 13min (actual slow down from fast(+4) setting), +1/+2 at 20min, maintains 0/+1 through 40min (no games lasted longer).

I finally tried two tests without Sorian AI installed. I put another AIX Adaptive on the left.

Stock FA:
0 at 20min, -1 at 25min (game ended shortly after)

With your mod:
+7 at 10min, +5 at 13min, etc.

Your fixes do seem to have a positive effect on sim speed. Sorian AIs slow it down a little, but don't appear to be the real problem. Your fixes improve performance in every case. Of course, the issue is more profound on maps where the starting positions are separated by water. I wonder if heavier transport use (factory assist) would improve performance on those maps?

If the biggest sim-speed slow down occurs when the AI hits its population limit, would it be possible to program in some sort of trigger wherein the AI just sends all of it's units to attack once it's at its population limit? This seems like a two-for-one kind of deal; it would resolve the issue with sim speed slowdown and it would also have the AI send out major attacks on occassion.