[7.00 Beta 2][Feedback] NPC-controlled capital ship combat in high attention

[A5PECT]

Back to Test scenario 01 (https://drive.google.com/drive/folders/ ... WeJkoznYd5) for a trial of 7.00 Beta 4

Skipping the detailed breakdowns because a lot of details remain unchanged. I may or may not go back to these for full writeups.

Test run 1 - https://youtu.be/Lp6ByN_CcBo

Test run 2 - https://youtu.be/_hk2hMmum8c

Test run 3 - https://youtu.be/zee7YfTAm2c

There are some improvements to the NPC Behemoths responsiveness during the test, but the underlying problematic behaviors persist. Flying past moving targets - in and out of travel drive - is still an issue. When moving to attack a target traveling towards them, capital ships very frequently end up very close to or even past the target before switching from the "move to" script to "attack".

Destroyers that find themselves close to a target attempt to turn around and regain distance, but the time it takes for a destroyer to turn itself around and fly away is more than enough for it to be destroyed by even a single enemy capital ship. The time taken to turn away (and die, more often than not) would be better used simply attacking the target and contributing damage. Destroyers being designed around main batteries with their given movement speed and turn rate, combined with capital ship AI not being able to reverse thrust or strafe, means maneuvering is NOT a valid tactical option during combat for NPC-controlled destroyers. Just make them shoot.

Destroyers being unable to attack targets directly above or below them has never stopped being an issue. Test run 3 of this trial saw all 6 Behemoths destroyed in this exact manner, one after the other. Something has to give: destroyers can not function with the limitations currently imposed on them.

[Huib-Bloodstone]

it is awfull..
capitol ships never have been so weird as with this update.


Asguard ships don't use any of its weapons, L ships tent to just to in guns blazing but get to close..

i have 1 K destroying 2 L syn and 2 XL Asguards that were trying to get above the K first, just point the big guns towards the K ??

same for coordinated attack or normal, they want to get close, gets detroyed and its over.

[FatalKeks]

Just tested a little with your first savegame. I think the vertical target acquisition works better now. Overall with relatively minor intervention I could save all Behemots and destroy the Ks. Without intervention they still struggle. But the improvements are noticable.
Looking forward to your analysis!

[capitalduty]

I will like to give some feedback on capital ship ai combat behaviour on beta 5, for me is working very very well. Best version that we ever had IMO.

[abc0000]

I will like to give some feedback on capital ship ai combat behaviour on beta 5, for me is working very very well. Best version that we ever had IMO.

Destroyers still cannot properly aim their guns at the enemy, do not know how to strafe, and do not know how to use reverse . . .
But on 1 I agree with you, they are now a little better than they were =)

https://www.youtube.com/watch?v=bSCHYMBRto8
This is roughly how I use my destroyers if they have the task of attacking other capital ships. Rockets rule, although they require micromanagement.

But I have BIG questions about the implementation of missiles.

[SpaceCadet11864]

In this save: https://www.dropbox.com/scl/fi/z7hsrhel ... va5my&dl=0
Teladi are attacking a xenon station in low attention and doing well. However, if you switch to high attention by using live stream view, they will get too close to the station and die quickly if they decide to attack the defense drones. When they choose to attack the defense drones, they disregard the threat of the gravitons on the defense station, and try to use their main batteries on them. going full speed, turning, etc, chasing them even. I don't really want to give them an edge here since they already winning, but, I thought this might be helpful in regards to the conversation around high attention npc capital ship control.

[A5PECT]

7.00 Beta 5, Test scenario 01

Test run 1 - https://youtu.be/AYlCE4CkZzg

00:47 - The opening fly-by. Note that the K stops and begins turning immediately upon being hit by the passing Behemoth's L plasma turrets, but the Behemoth takes a few seconds to react to being hit by the K.

01:59 - Other Behemoths move towards the K. Several fly very close to the K, well into Graviton turret range. I write this off as movement scripts not updating fast enough, which is an entire issue in and of itself. Two Behemoths fly themselves into the K, begin taking Graviton fire at 02:24, and turn themselves around. The entire time they're turning, they continue to take Graviton fire. Seriously. Stop making them do this. I'd rather they just stop where they are and shoot; the time spent turning is time they could be dealing damage to the K.

03:04 - The Behemoth to the left of the screen is pitching up toward the K as the K closes in and flies above. The Behemoth is not firing its main batteries, only its L turrets. The K quickly passes the Behemoth's maximum pitch angle, with predictable results. The Behemoth is rapidly destroyed while flailing about, unable to do anything about the K directly above it. This is what I mean when the combination of the way Ks are designed and how AI destroyer behavior is programmed is a fundamental, intractable problem.

04:10 - Three of the Behemoths have "stalled": they're idling in place, not attacking or even moving into formation with the command Colossus. The two Behemoths engaging the K faff about trying to match elevation with the K as the K approaches. They find a firing position eventually and get a few main battery shots off, but by that time the K is entering Graviton range. One of the Behemoths begins taking Graviton fire at 05:21, does the turn thing, dies. The other Behemoth, not being directly targeted by the K, does its own turn and fly away attempt. Both of these Behemoths die without dealing any meaningful damage to the K. Same story as every loss in this test: the K firing from above while the Behemoths turn and fly around accomplishing nothing.

07:44 - The K targets and flies towards one of the three remaining Behemoths. The Behemoths remain in their "stalled" state until the K comes near (I'm guessing their radar range was the threshold), whereupon they target and move toward the K.

09:01 - More overshooting on approach. I'd expect this to not be an issue when travel drives aren't involved, but their scripts still don't update frequently enough even given the wider margin of error. Two Behemoths fly past the K at 09:30: one of the Behemoths gets close enough to hit the K with one of its M pulse turrets, but the K does not return fire with any of its turrets. So the "turrets not firing" bug may still be present.

09:50 - One of the Behemoths finally updates its script and begins turning around to face the K it passed. The K and the other Behemoth take a few more seconds to come to the same reaction.

10:54 - Two Behemoths facing the K, not firing their main batteries. There isn't even an elevation difference during this period. A third Behemoth enters, and two Behemoths fire their main batteries for a short time. The K closes in and begins firing on one of the Behemoths at 11:33. I'm honestly tired of writing out the following series of event by now: K flies overhead, Behemoths turn, pitch up, attempt to climb, die before accomplishing anything, none of the Behemoths fire main batteries or deal any damage to the K.

13:19 - A Behemoth flies directly into the K, facing directly at the K the entire time, barely firing its main batteries. It flies past the K... for a very long time? It flies straight up, dozens of km above the ecliptic. I have no explanation for this behavior.

13:44 - The K has destroyed another Behemoth in the meantime. Two Behemoths remain: one is continuously flying up for no discernible reason, another sits not attacking, moving, or even facing the K until it starts to take Graviton fire. This Behemoth responds by turning away, pitching up, and climbing while it repeatedly takes hits.

15:21 - The Behemoth manages to find a firing position, turns towards the K, and fires its main batteries. It misses most of its shots, but it's shooting, at least. Before long, the K is within Graviton range and above the Behemoth. Have I mentioned I'm tired of writing this part out?

17:50 - The Behemoth that flew off above the ecliptic has finally turned back around is attacking the K from above. This would be clever, if all the other Behemoths weren't already destroyed. At 19:07, the Behemoth encounters a new twist on the old problem: the K flies underneath the Behemoth, and the Behemoth is unable to pitch down enough to aim at it.

20:14 - Two NPC faction Behemoths have moved in and engaged the K. They exhibit many of the same behaviors and issues as the player-owned Behemoths in this and previous tests. The rest of the test run is rehashing those events with two extra Behemoths completely under AI control.

28:01 - The last player-owned Behemoth is destroyed. The test run ends with the target K still alive, not even having lost its shields or taken hull damage for the duration of the test.

[A5PECT]

Test conclusions

- A lot of stuff I've already gone over in previous versions and tests: destroyers can't hit stuff above or below them, destroyers don't use their main batteries when they can and should be using them, destroyers flying too close/past targets due to script update/polling rates, destroyers attempting to maintain distance in situations where it's useless or even actively detrimental to themselves and their objective.
- Destroyers need to spend less time matching elevation and more time aiming at and firing on targets.
- When NPC-controlled destroyers do use their main batteries, they tend to miss even large, slow-moving targets. This seems to be an issue with aiming targets: I think destroyers aim their main batteries at surface elements of large ships, making them very likely to miss moving targets. NPC-controlled destroyers should aim their main batteries at the main hull of their targets, not surface elements.

[A5PECT]

Sidebar: I'd really like some answers regarding certain movement limitations in capital ship AI behavior. Destroyers not being able to pitch closer to 90 degrees vertically and being so strictly required to match elevation with their targets before attacking are massive problems. I don't see destroyer AI improving in any meaningful way without these things changing.

[SaulX4]

Test conclusions

- A lot of stuff I've already gone over in previous versions and tests: destroyers can't hit stuff above or below them, destroyers don't use their main batteries when they can and should be using them, destroyers flying too close/past targets due to script update/polling rates, destroyers attempting to maintain distance in situations where it's useless or even actively detrimental to themselves and their objective.
- Destroyers need to spend less time matching elevation and more time aiming at and firing on targets.
- When NPC-controlled destroyers do use their main batteries, they tend to miss even large, slow-moving targets. This seems to be an issue with aiming targets: I think destroyers aim their main batteries at surface elements of large ships, making them very likely to miss moving targets. NPC-controlled destroyers should aim their main batteries at the main hull of their targets, not surface elements.

Fully support these statements. All this stuff needs to be resolved. L/XL class ships also really need to move where they're told to move, not just 'somewhere nearby the move order' or 'in the complete opposite direction' as so often happens. Some dev input on this topic would be very welcome.

[A5PECT]

Just tested a little with your first savegame. I think the vertical target acquisition works better now. Overall with relatively minor intervention I could save all Behemots and destroy the Ks. Without intervention they still struggle. But the improvements are noticable.

Test scenario 01 is designed with no player intervention to isolate AI behaviors and "decisionmaking" specifically. It also attempts to simulate the behavior of NPC faction destroyers in combat, ARG or ANT don't have the benefit of the player's control abilities when they fight against the Xenon.

Destroyers still cannot properly aim their guns at the enemy, do not know how to strafe, and do not know how to use reverse . . .

AI makes very minimal usage of lateral/strafe and reverse movement. I've just accepted it at this point; I assume the calculations to make the AI use those things effectively within the movement systems currently used by the game would be difficult to program and costly to run, game engine and user system resource-wise.

I'm only concerned with the AI using the tools it does have in ways that make sense.