For Backers Only: AI building stuff

heh, maybe depends on the system? because i tested 5 games with absurd vs hard, hard won 4 of them, probably should do more games to really see for me but right now hard is the much more dominant difficulty

Thank you Sorian!

The AI should really really try to send their comms through teleporters when they see a planet incoming

I just send an asteroid to the most AI orbital crowded planet and won the game because all commanders were on that planet and they didn't even try to flee, even though they both had a secondary base on another planet.

I have played A few Games against the Ai the last days, and I find it Odd that it never build anti nukes, but spam nukes late game. And I never ser A orbital fabber trying to build teleporters on enemy planets, it's allways astreaus with fabbers

Odd, im always seeing anti-nukes, although ai will usually build a nuke first before going anti-nuke, and AI still doesnt really use teleporters when assaulting a planet quickly/well

This is a Absurd AI with a 1.3 eco buff, no Anti nukes after 53 min.

Very very strange AIs who perfer to fight against each other and not attempting a single orbital attack against me :
Was playing a Team game with 3AIs on my side against 4 other AIs.
3AIs vs 4AIs spawned on the same planet. The three AIs playing with me fought against the 4AIs and lost.
The my 4 opponents kept turtling on the same planet without attempting a single attack against my planet while i basically had no orbital units but orbital canons. I finally decided to blast them with nuclear missiles. Got them within a few minutes. As far as i could see the last one attempted to send an orbital unit in order to discover what was happening on my side. But too late, i took it and game was over.

Game ID : 18351925577830461862 (doud + 7AI)

Have also noticed all AI commanders do the same mistake : Commander is loaded into the egg and the egg does not move. Any Air unit can destroy the egg and that's what happened to my three AI mates.

Yeah something is completely broken with the AI right now.
- They won't attempt to move to any other planets whatsoever unless you have done so already. They won't even use any orbital units at all
- They build a tiny base with 4-5 factories and simply stop expanding afterwards no matter the difficulty.

Anti nukes are only ever built if the AI has scouted any nukes. Interplanetary scouting is lacking though, and issues with the radar system (lack of ghosts!!!) absolutely kills it. Even if it saw the nukes, it forgets about them as soon as the satellite goes down.

That is of course if it ever saw them and it didn't crash it's energy eco long before.

I found one! quick find me a poke-ball, we gotta catch it!

As of the last patch, the AI is much more functional and can move from planet to planet, AI now has to focus on expanding their econ quickly, building large armies, and timing their T2 switch better. Oh and assaulting planets does need some work, although the AI builds the teleporter first (most of the time) they dont actually use it...in fact its quite rare to see the teleporters used at all

The AI needs to build larger armies. This patch definitely shows some big improvements though

Ha ha. Just listened to another idle thumbs(#111) where one of the guys played PA in alpha and was really impressed by the game and the community.

Nice hearing about peoples enthusiasm.

The ai should build at least one builder, that reclaims stuff in and around the base.

I would like to make some comments about the AI and figured I'd post here instead of start a new thread. I'd really like to hear from @sorian about the AI and what he plans for it.

First of all, the latest update does seem to have improved the AI slightly, particularly in scouting early, so that's great. However my key observations are that the Absurd AI:

1) does not expand much. Doesn't seem to send fabbers out to build nearby mex, for instance.
2) Seems to set rally points towards the enemy base once scouted, but this only results in a trickle of units. It does not seem to want to group its units into larger armies/squads, and attack with a cluster.

I largely play single planet 1v1s (everytime i try and play the AI on a larger system, we end up spawning on the same planet anyway - is there a way to force spawn on different planets? I'd really like that ability), and it all tends to be over in about 10 minutes. I'd really like to play a more challenging AI, and I am confident that Sorian is working hard to that end. I'd just really like to hear about what he's working on. I'm sure he's aware of the AI's weakpoints, so maybe this whole post is a bit pointless, but hey

I wanted to highlight some of the key issues with the AI, so made a video of me beating it at 1.5x eco. It goes through the battle twice, once with the eco display and once with the unit display.

Sorry for the weird *** aspect ratio thing going on.

Link to game and replay



All times are in-game

The AI uses its air fabbers in its base, the first one building a radar. This will significantly delay its expansion, not to mention leaving the air fabber exposed by having it in the most easily scouted location. The AI should be using its air fabbers away from its base to minimise the chance of discovery and maximise the space it owns.

At 1:56 the AI has a radar, but it fails to react to the incoming Dox and allows its fabber to die. Given the units it had it should have repositioned the Commander to defend the fabber, or retreated it. However, what it should really have done is get out offensive units earlier. There's no point to that radar because it has nothing it can really react with. Meanwhile I have multiple Dox spreading out, scouting but also acting a fabber snipers. You'll see in the PiP that there's some great expansion sites on the other side of the planet the AI could totally have exploited with its air fabbers.

The AI has gone air first, but doesn't build any bombers early, which means it has no ability to project that air power. If you go air first you must have a bomber really early otherwise why are you going air? Likewise, you need to hide that you went air, don't fly your scouts right over the top of enemy units (2:14) because the enemy will react and counter your advantage.

At 2:01 the Firefly scouts the Dox, but the AI still doesn't react to it.

Again, at 1:56 the AI is constructing a fourth fabber AND a spinner. It doesn't need a spinner, it has an air factory which can produce fighters, what it needs is ground defence that can protect its ground fabber expansion.

The AI need to learn the Dox dance. Look at 2:22, my Dox eliminates all four opposing units (though only two were armed) because it does the dance. Just spam to the left, spam to the right, and your Dox is going to dodge a lot of incoming fire. My Dox will win numerous encounters this way.

2:26 the Commander walks way too far to build. It should be building next to itself where possible. I imagine this is tricky because you don't want to block him in, but walking time is time wasted. Likewise, the fabber behind it does the same thing, walking half a mile to build a factory.

At 2:54 I move out with a sizable Dox force, something the AI never does. What we'll see a lot during this replay is the AI never really groups up. It moves units around in fours, and it uses unsuitable groupings. An Inferno and a Dox should never be grouped, the Dox have speed and should be in a position to use it. Dox are raiders, the only unit which has any business being in their group are Slammers.

3:34 AI doesn't react to incoming Dox with its Hummingbirds, despite them being hugely vulnerable to Dox while still.

4:04 Where are those Hummingbirds going and why are they leaving the Bumblebee exposed? Also, the AI doesn't react to the raid at the rear MEX despite having a Bumblebee.

4:11 Bumblebee bombs MEX and not the fabber. Fabbers are always the priority target for a bomber.

5:00 I fucked up and misqueued a lot of fabbers. Ignore this

5:49 is a good example of the AI not grouping. Its units are scattered all over.

7:35 AI goes T2 despite being on the back-foot, a huge mistake, especially T2 air of all things. Right now, in a single-planet 1v1 you almost never go T2.

8:37 it attacks with a mix of a bit of everything. The AI needs some coherence in its groupings. Dox for raids, Bolos for pushes. A few Infernos for structure attacks (but not until it feel that it's winning). It's still building Spinners but is doing fine in the air game, so that's just a wasted unit. It has a Skitter in there, so why send the Dox?

10:05 it scouts two air fabbers on the backside of the moon but doesn't send anything to take them out. A single fighter would get the job done.

10:27 more scattered units

10:49 another T2 factory but it hasn't finished the first one yet

Now the balance has settled down it's probably time for the AI to learn the meta. There's a really good video from elodea on a strong opening build order. More than that though the AI needs to work on its unit compositions. It looks like it builds everything but scouts relatively evenly, but it should be weighting certain units over others, and favouring keeping those units in specific groups. You can see that really early in the game I have far more combat units than the AI, even though we have the same number of factories, and that's because it's building things like Grenadiers when it should be building Dox.

Dox should operate by themselves or with Slammers. They're fast, they can go underwater, they're the raiders. This is what the AI should roam with.

I wouldn't bother with Grenadiers at all.

The AI needs more Bolos and less Infernos. Somewhere between one Inferno to every ten Bolos, and really it should be all Bolos until the AI wants to make a push on a structure heavy area.

If the AI goes air first it needs to use an early bomber to shut down enemy expansion, and it should probably use Hummingbirds to scout to ensure its air dominance.

The AI needs more factories and it should be using the Commander to build them. Fabbers to expand, Commander for production.

Certain biomes favour certain units as well. If it's a moon you probably want more early Dox, while on lava you're probably going to be tighter for space and will go for Bolos much sooner.

Anyway, maybe this will all useless balther, but hopefully it was in some way helpful.

It seems like we need a new difficulty setting called 'Anti-Quitch'

Nice detail @Quitch!

Before Sorian, there was Quitch!
http://www.moddb.com/games/total-annihilation/downloads/queller-ai-ta-cc-v17