Any plans for improved naval AI?

scarts

Chieftain
Joined
Feb 4, 2016
Messages
22
There’s really no excuse to lose a naval engagement aside from getting landed on. The AI just simply can’t micromanage putting destroyers in front and back of carriers/battleships to protect against subs, putting one carrier on intercept and one on air sweeps, shore bombardment with battleships blocked by destroyers, or subs a few tiles out of the fleet formation.

There are essentially 3 or 4 ‘formations’ in the code. One of them is “amphib-landing” where the AI throws a few marines + aa guns next to a “task_force” or “carrier group” (task force is destroyers + battleships only / carrier group is carriers + battleships only). This is why we tend to see so much destroyer spam, because they’re coded to free roam and not serve as point-interceptors (even in Naval Expansion

My point is, unless you let the AI strike first, they can’t retaliate to naval combat. If you sink their lead battleship/destroyer with subs, they will fall apart as opposed to sending out 1-2 destroyers to hunt them down. You can easily disrupt formations but simply hitting them while they’re formed up, and then the AI’s code goes crazy trying to determine whether to continue onto target or abandon entirely to chase some naval units. It’s the main reason I play majority-land based maps, because you can take every coastal city from the AI even if they have a land-based army garrisoning nearby. The AI also sucks at prioritizing naval kills from land - your whole fleet can be providing shore bombardment at a damage ratio of 4:1 compared to your landed troops (from an amphibious assault), and the most logical thing to do is take out the carriers offshore providing air cover.

BUT THE AI IS INCAPABLE OF AIRSWEEPING UNKNOWN TILES - unless AI has sight of a unit to air sweep, if your fleet is out of view from their controlled territory, they will never be able to locate it with air assets. I can’t tell you the amount of times I send in marines and keep my fleet 2 tiles away from AI border, with destroyers screening for subs 2 tiles away from my main fleet. By keeping battleships and carriers in dark tiles to AI -

  1. the AI simply cannot send a fighter to uncover that tile, then 2) use its remaining air assets to attack that “newly discovered” tile with a unit on or adjacent. This is a massive exploit, because all experienced human players know fighters are critical to scanning troops movements/locations.
Because the AI cannot do this, they are at a massive disadvantage. Humans can send a few fighters out in a search attempt for carriers/cities with nukes based in them, whereas the AI only uses fighters for air sweeps against cities, AA guns, or visible units in general.

This is one of the reasons the AI sucks at nuking. Aside from their spies revealing cities, their value calculated for choosing a nuke target is based off the cities they’ve seen and explored. If you had one city that was your industrial hub a few dozen/hundred turns ago, and the AI happened to have a spy there then, they typically lean towards nuking that (or your capital factoring in war score and leader aggression values). You can move industry around every 20-30 turns or so, and if the AI doesn’t move a spy around at the same rate (constantly updating your cities for their info), their nuke targets become outdated essentially.

I’m only discussing nukes because the AI is poor at using nuclear submarines too. In the event they actually place a missile on a sub 1) they send the sub unescorted (not in a Wolfpack or formation) very close to your territory (where you then see it and kill it with the missile inside still), or 2) they just never put a missile on a sub. The 3rd scenario is very rare and I seldom see it - where AI moves air units in response to city siege/under attack, and moves a missile to a sub, then strikes your closest frontier city to your advancing army (thus taking a few units out as well). Overall I love VP, but stick to land-based warfare as the formations and tactics of the AI are of course way better than their navals. Civ historically sucks at naval warfare I know and VP is MUCH better than any other version of civ.
 
@scarts, you seem to understand this late game naval warfare very well, perhaps you can collaborate with @ilteroi to improve the AI's understanding?

You can move industry around every 20-30 turns or so, and if the AI doesn’t move a spy around at the same rate (constantly updating your cities for their info), their nuke targets become outdated essentially.
Although the AI won't do this, it seems like bamboozling their spies (how do you know where their spies are? IGE?) by changing city focus is good play, no?

This discussion comes close to the nuclear option in AI planning. Namely: after some point give them cheat-vision of the whole map (above some difficulty level).
 
I don't see how you can exploit tactical flaws in naval battles tbh. I don't feel like theres much strategy involved
 
I don't see how you can exploit tactical flaws in naval battles tbh. I don't feel like theres much strategy involved
I would argue there is more strategy in naval than in land battles by a good bit tbh. Positioning and scouting are everything in naval battles, because they are very swingy. Often the fate of an entire enemy fleet is decided in the first round I engage that fleet, because of how much (or how little) damage I can do on that first round.

Naval units have lots of tricks and can hit like bricks when properly positioned, meaning one side can basically wipe the other out in a few turns. In contrast, land combats are much slower and methodical.
 
Last edited:
yeah I agree with this post
steam rolling enemy fleets with 1 or 2 submarines and an aircraft carrier gets old fast
 
Invisibility is a pain for the AI to handle, since they never remember a submarine attacked them.

Same with carriers that stay out of range. They can't deduce there's a carrier nearby if they're being attacked by planes without a possible city base.
 
Invisibility is a pain for the AI to handle, since they never remember a submarine attacked them.

Same with carriers that stay out of range. They can't deduce there's a carrier nearby if they're being attacked by planes without a possible city base.
can this be fixed?
 
No unless they cheat.
 
Invisibility is a pain for the AI to handle, since they never remember a submarine attacked them.

Same with carriers that stay out of range. They can't deduce there's a carrier nearby if they're being attacked by planes without a possible city base.

can this be fixed?

No unless they cheat.
Some speculative theory crafting here... Say these attacks applied a temporary promotion on the targeted unit... 1-2 turn duration. AI then doesn't have to remember it was attacked by a sub or plane etc, it will be able to read it from the unit itself when its turn rolls around. Pseudo code would be something like:
If unit:haspromotion(recentSubAttack) then
Do subhuntAI
Else
Do normal AI

This would also smooth out the human experience in hotseat somewhat

The way the current movement and combat regime works, I think it's the system itself that does not support much tactical play. That said the very decisive nature of these battles is somewhat realistic

In the past I've played with varied in-combat and out-of-combat movement rules, counter attacks and retreat mechanisms on naval units... Some of these things work well to add tactical elements but the AI is blind to them, so it doesn't make that side of things any better as modmod. If community wanted to look at significant changes to naval game, these are the directions I would consider
 
Last edited:
They already can "cheat" (having bonuses on higher difficulties) by desing, so that's fine.
I believe the easiest way of having ai cheat requires them being given extra vision in some way. I do remember talks of giving ai full knowledge of the nap at certain difficulties after a certain point in the game, done at the right time i think that should be fine
 
I believe the easiest way of having ai cheat requires them being given extra vision in some way. I do remember talks of giving ai full knowledge of the nap at certain difficulties after a certain point in the game, done at the right time i think that should be fine

Full knowledge of the map is not desirable imo, it's too core a mechanism to have such an imbalance in -- one of VP's focus early on was removing the asymmetries; I'd be very reluctant to start adding them in. I also wonder how full knowledge of the map might affect turn times... AI gonna have a lot more variables to consider

I wonder if we can't find a more creative and symmetrical solution re the vision issue. For example as a mod I've used a post-combat promo to increase vision ie unit that gets hit by sub can see much farther for a few turns. It's a little gamey, but even as symmetrical ability it seems to tighten things up a little between AI human. Much less gamey then full knowledge of map
 
How about giving AI memory of units from last turn? It doesn't give additional info, because the info already is available to both player and AI.
If irc, that requires new code cause ai atm can't remember stuff in the past, it was a major reason vision got changed to lasting for whole turn rather then dynamically changing as troops moved
 
How about giving AI memory of units from last turn? It doesn't give additional info, because the info already is available to both player and AI.
Then it's entirely guesswork whether the units are still there on this turn, which is more computation time when AI considers all cases.

I wonder if it's viable to make ships slow down (half speed) once an enemy is within a certain range. Some sort of wider zone of control. Or any other way to prevent alpha strikes taking out half the AI fleet before they can react.

Running away from AI navy is also easy. They don't give chase if they can't see you.
 
Last edited:
Top Bottom