AI vs stealth-as-a-strategy

Describe your experience with the latest version of FreeOrion to help us improve it.

Moderator: Oberlus

Forum rules
Always mention the exact version of FreeOrion you are testing.

When reporting an issue regarding the AI, if possible provide the relevant AI log file and a save game file that demonstrates the issue.
Post Reply
Message
Author
User avatar
MatGB
Creative Contributor
Posts: 3310
Joined: Fri Jun 28, 2013 11:45 pm

AI vs stealth-as-a-strategy

#1 Post by MatGB »

Wondering how much, if any, ability the AI has for keeping track of stealthed fleets, especially stealthed fleets that've very recently destroyed ships &c.

Because the more I play around with stealth in combat, the more I notice the AI effectively suiciding fleets. A system I'm blockading with ships it cant see will have small(ish) fleets effectively suicide themselves peacemeal and similar, so my moderately sized garrison can wait for the troop ships and pick of individual ships that're attackign that would never do so if they could see my ships.

Sometimes in a way that's slightly annoying, that suicide ship taking out a couple troop ships, etc.

(aside: BioAdaptive Hull, Absorption Field, no shields, 2 guns one armour. Cheap so you can field them in numbers and they kill everything they come across up until the AI gets Omni Scanner. At which point they're effectively useless but you can normally win by then...)
Mat Bowles

Any code or patches in anything posted here is released under the CC and GPL licences in use for the FO project.

User avatar
Dilvish
AI Lead and Programmer Emeritus
Posts: 4768
Joined: Sat Sep 22, 2012 6:25 pm

Re: AI vs stealth-as-a-strategy

#2 Post by Dilvish »

MatGB wrote:Wondering how much, if any, ability the AI has for keeping track of stealthed fleets, especially stealthed fleets that've very recently destroyed ships &c.
The AI does keep track of losses to stealthed enemies, and figures they must be at least as strong as the fleet lost. That imputed threat in a given system goes down slowly over time, since most stealthed enemies have a tendency to move around. I probably should make the initial threat assessment have a larger safety factor built into it; I'll give that a go. The AI might then get overcautious, but we can see how it goes.
If I provided any code, scripts or other content here, it's released under GPL 2.0 and CC-BY-SA 3.0

User avatar
MatGB
Creative Contributor
Posts: 3310
Joined: Fri Jun 28, 2013 11:45 pm

Re: AI vs stealth-as-a-strategy

#3 Post by MatGB »

From observation over-cautious would almost certainly be better for them, getting ahead with fast stealth ships currently means they're defeating themselves anyway, but OK that they wait a bit and assume you've moved some on makes sense and given it's what I'll tend to do is at least sensible.

Actually, keep forgetting to ask, are they set to use stealth granting parts in designs? I've never seen it except the shields and I'm looking at splitting the the planetary stealth and the ship part stealth techs up.
Mat Bowles

Any code or patches in anything posted here is released under the CC and GPL licences in use for the FO project.

User avatar
Dilvish
AI Lead and Programmer Emeritus
Posts: 4768
Joined: Sat Sep 22, 2012 6:25 pm

Re: AI vs stealth-as-a-strategy

#4 Post by Dilvish »

MatGB wrote:Actually, keep forgetting to ask, are they set to use stealth granting parts in designs?
Currently, (IIRC) the only ship design that's really geared to assign value to stealth is the nya-nya-krill spawner, but I think we haven't yet put in place the production code to actually build such things. As far as warships go, we haven't yet put together a rating system to give serious weight to stealth. If you care to make suggestions you might help speed it along.
If I provided any code, scripts or other content here, it's released under GPL 2.0 and CC-BY-SA 3.0

User avatar
MatGB
Creative Contributor
Posts: 3310
Joined: Fri Jun 28, 2013 11:45 pm

Re: AI vs stealth-as-a-strategy

#5 Post by MatGB »

Yeah, it's really hard now, let alone if we get the strategy to be viable, fun and balanced. And it's way more dependent on what other players are doing than any other design factor.

For example, the Symbiotic Hull is the best early game scout, vision boost and refueling giving it a nice edge. But, it's also, with no other part, invisible to other players if they haven't got Active Radar. Immediately the AIs get AR then mine tend to start dying, but as long as I stay ahead in stealth then I can get eyes on the whole map and basically not really face a threat as long as I keep a responsive reserve.

Against other AIs, the ability to both not have scouts die and/or have fleets get an entire round of shooting free is going to be powerful (since you fixed the stealth shot code it's really powerful but the designs that work best tend to be fragile), but against a player like me that puts very early RP into detection as a high priority it's going to be very difficult to make it work at all.

It's probably best rating it similarly to shields but they devalue quickly with enemy detection, similarly to early game guns being irrelevant once higher end shields come into play.

And of course it's likely to change when we do a more thorough revamp of the way stealth works which we've kinda agreed needs to happen even if not on the how. But I quite like the basics of the current mechanic, it just needs a lot more subtletly
Mat Bowles

Any code or patches in anything posted here is released under the CC and GPL licences in use for the FO project.

Morlic
AI Contributor
Posts: 296
Joined: Tue Feb 17, 2015 11:54 am

Re: AI vs stealth-as-a-strategy

#6 Post by Morlic »

Probably it is best to wait for the new ResearchAI to be working before we use any really complex decision-making as so much depends on the interaction between stealth and detection level techs.

However, I think it may be a good idea to add some random chance for each AI to go for a stealth-heavy game with (somewhat) early stealth techs and build ships with it. This should force a decision for the human player how long he may neglect detection techs. It also will probably give some good first ideas about suitable timeframes for the AI.

To be most efficient, we will kind of need an almost completely different AI for a stealthy game. For example, the stealthed fleet shouldn't attack too early but aim for a single devastating battle. After the enemy knows about the stealthed fleet, he will go for detection techs. So the AI needs to gain some deciding victory within that very short time window where it has the stealth advantage.
I expect, however, that this is way beyond the scope of anything to come within the next year(s).
If I provided any code, scripts or other content here, it's released under GPL 2.0 and CC-BY-SA 3.0

User avatar
Dilvish
AI Lead and Programmer Emeritus
Posts: 4768
Joined: Sat Sep 22, 2012 6:25 pm

Re: AI vs stealth-as-a-strategy

#7 Post by Dilvish »

Morlic wrote:Probably it is best to wait for the new ResearchAI to be working before we use any really complex decision-making as so much depends on the interaction between stealth and detection level techs.
I suppose your point is that since we have been somewhat prioritizing the new research AI, it might be wasted effort to adapt the current research AI to consider this strategy. I would tend to agree at least for the near term, but I wouldn't consider it a hard-fast decision.

Since we're having a hard time still just getting it to manage our current content as well as the current ResearchAI, and it seems likely to take a fair while longer, CJ and I were thinking we'd go ahead and bring it in to master, but as an alternate ResearchAI with the selection defaulting to the current ResearchAI but the AI config file could dictate the new one (or some likelihood of it). That way we'll be less likely to continue creating nuisancesome branch conflicts during the expected lengthy development time remaining for the new ResearchAI. Also as it starts to get better we could gradually phase it in as a possible choice for the AIs, without necessarily having to wait for it to be clearly better (or even clearly as good) than the existing one.

Also, the real strength of the new AI is intended to be automatic adaptability to new content; I don't see anything about it that would really make it any easier for to assess the stealth-detection situation with respect to enemies. So if you were interested in putting in the time to develop a stealth strategy for the current Research AI, it wouldn't go to waste (and would probably help guide the stealth strategy for the new ResearchAI as well).
If I provided any code, scripts or other content here, it's released under GPL 2.0 and CC-BY-SA 3.0

Post Reply