Ow.

[TechSY730]

But when CPA ships were changed to get dunked directly into threatfleet...  Well, I'm making a change now to not try to TF any ship that isn't on the AI's "side" of the galaxy.  Better that they stick to the "local" logic if they're cut off in some corner of the galaxy on the wrong side of one mean chokepoint.

That will sort of break down if there are two or three "parts" of roughly or exactly the same size. Especially if the to AI homeworlds do not have a fully AI space path to each other, which is pretty trivially doable on some map types like tree, snake, and X.


Still, I am glad that "should I go in" logic is getting a lookover, especially in terms of actual how to measure "threat levels", target selection, how do deal with enemy planets on the path to the target, other "groups" of stuff also going to that target, etc...

The...oddities with this logic have been a major thing holding the AI back. Good to see you addressing it.



And I agree with treating "support" units like forcefields, tractors, and grav turrets as force multipliers. That would seem to be a decent approximation. Finding good multipliers of course will be tricky.

[keith.lamothe]

I've got it doing the waiting well away from the wormholes, but I'm wondering if this isn't going to replace one vulnerability (getting tractored) with another (getting defeated in detail on the other side of the wormhole when going through due to spreading out from speed differences).

And having some camp the wormhole and others hang back would actually be the worst of both worlds: the campers would be even easier to tractor, and the hangers-back would have an even higher chance of coming through after a huge chunk has already died.

[keith.lamothe]

Ok, the labors of the day are in for the next version now:

* Replaced the "Firepower" metric (which was basically just BaseDPS, not considering other stats) with the "Strength" metric (which, while not perfect, has been used with success elsewhere and seems to be a much better measure for comparing groups of units to see which one is stronger; something that the game does rather a lot, particularly in AI logic).
** You're likely to see situations where the AI is undervaluing/overvaluing some things and making... "questionable" decisions about when to commit to an attack, etc.  Your reports (and save games) will be helpful in fixing these cases.

* Substantially improved the AI's "intelligence" in handling its threat ships:
** Previously when asking "is it safe to move this threat to its next destination planet?" it would only consider the strength of the threat on that group's current planet (and anything actually on the target planet), but now it also considers any other threat (on other planets adjacent to the target) waiting to go through.
** It now asks that question (is it safe, blah blah) even when moving on to other AI-held planets.  Unless it's a special forces ship, which is just supposed to charge in if it's an AI planet.
** If it decides a ship needs to wait, the ship now waits somewhere not directly on top of the wormhole leading to the target planet, because that position is far too vulnerable to tractor shenanigans and such.
** If a ship is in a part of the galaxy that the AI cannot reasonably easily gather forces from (meaning there's some nasty human chokepoint cutting it off from the AI homeworlds), the AI will no longer try to enlist that ship into the Threatfleet.  The "local threat" logic is better in those cases.
*** Note that ships already in threatfleet behavior (from previous saves, or before you cut it off from the AI HWs, etc) will stay that way.

* Added a debug chat command: "cmd:log ai wait"
** If used in MP this only affects the host, since only the host runs the AI thread.
** If you're in planet view when you give the command, this tells the AI thread to generate a "LogicLog_AICheckForClearingWaitPointsAgainstSpecifiedPlanet_AIThread.txt" log file in your RuntimeData directory whenever evaluating whether to clear wait points (i.e. charge!) against that planet.
** If you're in galaxy view when you give the command, this tells the AI thread to stop said logging.
** The contents of this log are fairly readable if you understand the terms, but the main benefit is so you can send it to us if you have questions about why the AI was (or was not) waiting to attack in a particular instance.

Not sure when the next release will be, but I'll definitely want to make this available for testing sooner than later.  Gotta break things to improve them when it comes to this stuff, but I'm hoping they won't be broken or at least won't stay there long.  Things seemed fine in my tests but that often means little.

[keith.lamothe]

Just got the "clear wait points" logic to group-move each planet's chunk of threat towards its target wormhole, so that deals with one of the downsides of "don't camp right on the wormhole, you moron".

The other is that right now I've got them waiting either at the wait-planet's command station or at a point selected in a way similar to adv-fact placement: near the inner grav ring, as far from wormholes as possible.

This does reduce the threatball's defensive utility (they'll FRD into attackers from there but it can take a while to get in-system) but I don't mind that so much: the AI's priority for these should probably be "conserve until I can use them offensively", and leave defenses to the guards and the SF and such.

The actual problem is that it can take quite a while for the threatball to actually reach the wormhole once the "stop waiting" order comes down.  It can also take a pretty variable length of time for one planet's threatball to reach the wormhole compared to another planet (that was waiting against the same target, for instance).

So I'm thinking that they need to pick a closer spot to the wormhole.

My question to y'all is: what range? 

In theory it could be variable (based on "does the human have a black widow golem?", probably ) but I think a single global range (or range of ranges) would be better for a first try.

[keith.lamothe]

Btw, in the test I just ran the AI did a good job of wiping out most of the satellite planets.  Basically only High Lab, Tripwire, Midnight, and some of the "satellites" around Box of Mazes stood; the friendly enclaves stopped the enemy from getting further in than Milkman's Lair and killed the stuff on Box that would have cleaned out its satellites.

Once the human went into brownout the AI attacked Midnight (not entirely sure why, I don't think it knows to specifically recognize that situation) and actually almost took it down.  But the energy balance came up enough from losses that the FF net came back on line.

Naturally any player-attention would have fixed the brownout much faster, but I think things are improving

[Faulty Logic]

So I'm thinking that they need to pick a closer spot to the wormhole.

My question to y'all is: what range? 

An AAR is probably not a good spot to get a lot of responses.

I'd say a flat 20,000 from the target wormhole.

In theory it could be variable (based on "does the human have a black widow golem?", probably   

No. Riots don't work as well as widows for this specific thing, but they get the job done.

Btw, in the test I just ran the AI did a good job of wiping out most of the satellite planets.  Basically only High Lab, Tripwire, Midnight, and some of the "satellites" around Box of Mazes stood; the friendly enclaves stopped the enemy from getting further in than Milkman's Lair and killed the stuff on Box that would have cleaned out its satellites.

I've got a bad feeling about this change. Of course, this result was without player intervention.

[keith.lamothe]

An AAR is probably not a good spot to get a lot of responses.

Yea, I was mainly asking the two of you actually tracking with me here.  20k sounds like a good first shot.

[TechSY730]

Sorry, I haven't quite internalized the unit used for distances and "practical" data points for them.

What the range for a plasma siege starship again? Zombard? Missile frigate? Ether jets? Tractor platforms? Martyr? Widow golem? *contemplates trip to the wiki*

[Faulty Logic]

20000 is in zombard range, but none of the others. Widow tractors operate at 12000 range.

What the range for a plasma siege starship again? Zombard? Missile frigate? Ether jets? Tractor platforms? Martyr? Widow golem? *contemplates trip to the wiki*

10000, 30000, 10000, 5000, 5000 (~1/2 for tractors), 3500, 19,000 (tractors at 12000). Note that some ranges increase with mark.

[keith.lamothe]

Ok, retested with the wait-20k-out rule and with the exos syncing (via pre-100% launch) with the CPA.  The CPA still isn't jumping in after the exo as thoroughly as it should but it did do partially.  The combined impact took down the Midnight command station but there was still just too much left on Midnight for the AI to take out.  It overran everything outside the Midnight chokepoint, though, except for a couple isolated planets it just wouldn't attack (debugging that now).

[Faulty Logic]

Ok, retested with the wait-20k-out rule and with the exos syncing (via pre-100% launch) with the CPA.

I'd be very careful with that rule. CPAs happen at close to twice the rate of exos, so having them try to sync could make it so I never see a full exo.

The combined impact took down the Midnight command station

What?! How? Infiltrators, or brute force?

[keith.lamothe]

I'd be very careful with that rule. CPAs happen at close to twice the rate of exos, so having them try to sync could make it so I never see a full exo.

True.  The alternative is to try to sync CPAs up to the exos (so you'd have something like CPA -> CPA/exo -> CPA -> CPA/exo, etc).

Anyway, it does seem that the AI's responses to your normal stuff and its responses to your superweapons need to cooperate, or there's no way through that door

The combined impact took down the Midnight command station

What?! How? Infiltrators, or brute force?

I wasn't watching closely, but I suspect the H/Ks.  That part of Midnight was hammered flat; nothing left.

[Faulty Logic]

Anyway, it does seem that the AI's responses to your normal stuff and its responses to your superweapons need to cooperate, or there's no way through that door

This principle applies to all the significant asynchronous responses (FS exos, Golem/Botnet/Spirecraft exos, Nemeses, CPAs). They should all sync up to some degree, or the player gets an advantage for each new, reasonably fair "deal."

True.  The alternative is to try to sync CPAs up to the exos (so you'd have something like CPA -> CPA/exo -> CPA -> CPA/exo, etc).

I think that would work out pretty well (for the AI).

[Hearteater]

I think the AI does need to combine its efforts to a degree, otherwise it is too easy to defend, especially with Superweapons.  You could make the degree to which it bends the rules for sync'ing be related to game time.  So early game syncs are just random.  But the longer the game drags on, the more "effort" is put into syncing.  For example, for every one hour of game time, a CPA or Exo can be bumped two minutes later to coincide with the other.

[Faulty Logic]

You could make the degree to which it bends the rules for sync'ing be related to game time.  So early game syncs are just random.  But the longer the game drags on, the more "effort" is put into syncing.  For example, for every one hour of game time, a CPA or Exo can be bumped two minutes later to coincide with the other.

That is an excellent idea.