Distant Worlds AI

[Tormodino]

Asking for a highly qualified AI player is nice and all, but does anyone know why basic stuff like fleets/ships going off on crazy treks into the great unknown, constantly throwing themselves at superior numbers (seems to not remember the size of targets properly), resources being allocated incorrectly (spent, out of interest, 2 years waiting for a few units of Helium to be delivered to a home system mining station while it was happily building other ships as soon as the homeworld received the Helium) etc... are happening so frequently?

I enjoy the fact that it seems that the AI plays "dumber" when it plays the dumb races, but there are still many cringe inducing moments.

Btw, for reference, I spent many hours looking at how the AI plays with Rule in Absence activated. It is fascinating, but it also reveal some absolutely insane descisions by the AI. If what the player can see here represents how the AI empires actually play I would say that a slightly more conservative weighting of travel ranges and fuel requirements, designs with fewer obvious mistakes (always adding stealth (size 60!)/not adding damage control), as well as slightly tighter resource discipline could go a long way to make the AI a slightly more challenging opponent.

Also, I want to note the erratic behaviour of the AI in general. It seems to change its mind, or at the very least recheck its state(s), very often.

Case in point: I was at war with the Zenox. Their entire empire was located in the bottom left of the map. They had a mining base at the top right of the map. My biggest fleet, loaded with troops, instead of invading their closest worlds (cakewalk, they of course have no troops...) goes flying across the universe to kill the single gas mining station. Running out of fuel, stopping several times along the way etc.
To be honest, this stuff should not happen. Some limit to how many resources are devoted to a far off target could be implemented to prevent that type of behaviour. I don't want the game to be predictable and having race intelligence levels affect descisions is fine, but I draw the line at crazy stuff like that. It is totally detrimental to the AI, and makes the simple stomping of the AI via deathballinvaderfleet an excersise in tedium.

Observing the game has given me a tremendous respect for the model, but I would much rather that the current model be worked on than wait for DW2. Unless there are issues with the engine heavily limiting what is possible, I see no reason why DW in its current form could not be brought to be the new standard for 4x simulations through balancing and AI tweaks. I've outlined a few other things I would like to see elsewhere, but as it stands the inconsistent AI is a glaring flaw in each game I play.

[MartialDoctor]

Tormodino makes a very good point. I've noticed those sorts of things in the past as well.

[aaatoysandmore]

ORIGINAL: Tormodino

Asking for a highly qualified AI player is nice and all, but does anyone know why basic stuff like fleets/ships going off on crazy treks into the great unknown, constantly throwing themselves at superior numbers (seems to not remember the size of targets properly), resources being allocated incorrectly (spent, out of interest, 2 years waiting for a few units of Helium to be delivered to a home system mining station while it was happily building other ships as soon as the homeworld received the Helium) etc... are happening so frequently?

I enjoy the fact that it seems that the AI plays "dumber" when it plays the dumb races, but there are still many cringe inducing moments.

Btw, for reference, I spent many hours looking at how the AI plays with Rule in Absence activated. It is fascinating, but it also reveal some absolutely insane descisions by the AI. If what the player can see here represents how the AI empires actually play I would say that a slightly more conservative weighting of travel ranges and fuel requirements, designs with fewer obvious mistakes (always adding stealth (size 60!)/not adding damage control), as well as slightly tighter resource discipline could go a long way to make the AI a slightly more challenging opponent.

Also, I want to note the erratic behaviour of the AI in general. It seems to change its mind, or at the very least recheck its state(s), very often.

Case in point: I was at war with the Zenox. Their entire empire was located in the bottom left of the map. They had a mining base at the top right of the map. My biggest fleet, loaded with troops, instead of invading their closest worlds (cakewalk, they of course have no troops...) goes flying across the universe to kill the single gas mining station. Running out of fuel, stopping several times along the way etc.
To be honest, this stuff should not happen. Some limit to how many resources are devoted to a far off target could be implemented to prevent that type of behaviour. I don't want the game to be predictable and having race intelligence levels affect descisions is fine, but I draw the line at crazy stuff like that. It is totally detrimental to the AI, and makes the simple stomping of the AI via deathballinvaderfleet an excersise in tedium.

Observing the game has given me a tremendous respect for the model, but I would much rather that the current model be worked on than wait for DW2. Unless there are issues with the engine heavily limiting what is possible, I see no reason why DW in its current form could not be brought to be the new standard for 4x simulations through balancing and AI tweaks. I've outlined a few other things I would like to see elsewhere, but as it stands the inconsistent AI is a glaring flaw in each game I play.

If you have trade agreements and alliance's with other races then the AI will use them for figuring out the rations they need to get from point a to poing b. So, your alliances or trade agreements could make the trek shorter or more cost effective and it will use all those hubs for resupply. Everything is not 1-2-3 in this game or by the numbers some things will be erratic and unbelieveable just like things happen in real life. I wouldn't want a game that plays by the numbers all the time. I like randomness and whatif's and anything can happens in the games I play. This one does it quite well. It just needs to ramp up the stats for the ai to make it a little more challenging when it does things that are random or unheard of.

[Tormodino]

We have to assume that the people who play this game a lot use the military refueling rights and other options to make fleet movement and resource flow more reliable, but that it not my point in the slightest. I am talking about the downright strange and detrimental actions of the AI such as trekking across the map to kill a single mining base while the enemy's fleets and colonies are ripe targets just next door. An action like that would get an admiral fired or shot.

I am obviously fine with the randomness and strange actions that lead to interesting outcomes.

[thefinn12345]

Yeah, bascially - the craziness should go.

[Icemania]

ORIGINAL: Erik Rutins
RESEARCH AND SHIP DESIGN
- AI ship designer now always adds sufficient reactors to ship designs (possibly more than one) to enable hyperdrive to operate at full speed
- AI ship designer now more willing to shrink new designs to fit within current construction size limits, especially when first obtain hyperdrive technology
- NEW MODDING FEATURE:[/b] allow restricting any research project to specific races using the new ALLOWED RACES line in the research.txt file (updated header comments of research.txt explains)
- improved AI research pathing to better emphasize tech focuses from empire policy (ResearchDesignTechFocus1-6), while still researching other important techs
- increased research output for lab components by approximately 50% (components.txt and research.txt)

GAME BALANCE
- Shakturi now have faster construction and lower ship maintenance (updated Shakturi.txt race file)
- Shakturi ship design templates now use more weapons and other components (military ships, medium & large spaceports, defensive bases)
- Shakturi empire policy now builds more large military ships and less small military ships
- empires are now more willing to use excess cash on hand as factor when deciding whether to build new ships (e.g. low or negative cashflow, but large cash reserves on hand), especially when at war
- now much harder to perform intelligence missions against Ancient Guardians (steal maps, tech, etc)

Great work ... I can't wait to test this out!

The new research modding feature could be used to further improve AI research focus e.g. don't waste time of researching weapons technologies that will never get used.

[MartialDoctor]

Yeah, those changes are looking nice. However, I don't get the increased lab output. It's already too easy to reach maximum research capacity as is.

[Bingeling]

ORIGINAL: Icemania

The new research modding feature could be used to further improve AI research focus e.g. don't waste time of researching weapons technologies that will never get used.

As far as I know, the AI have not spent time on non-used weapons since Shadows or so. The research order can always be discussed, though.

What happens at end tech, I am not sure. I guess it will eventually research everything.

[Cauldyth]

ORIGINAL: MartialDoctor

Yeah, those changes are looking nice. However, I don't get the increased lab output. It's already too easy to reach maximum research capacity as is.

Maybe it'll help the AI, for when it's not building enough research capacity.

[Tcby]

ORIGINAL: Cauldyth

ORIGINAL: MartialDoctor

Yeah, those changes are looking nice. However, I don't get the increased lab output. It's already too easy to reach maximum research capacity as is.

Maybe it'll help the AI, for when it's not building enough research capacity.

Pretty much this. The issue is that the AI never reaches capacity. Icemania did some testing a couple weeks ago that showed they were only using a small percentage of their total capacity (around 50% I think?). The following patch increased the AI desire for research stations by 1, bringing them up to about 70%. Now increasing the output of each lab component by 50% should allow the AI to surpass their research capacity using those same labs.

The great thing about increasing lab output like this is that it simultaneously reduces the benefit to putting them on private sector stations, because the overall cost is less than before. So another natural human advantage over the AI has been reduced in a pretty simple fashion.

yay patches

[BlueTemplar]

But now it's even more tempting to micromanage your research by having 3 super research base designs (one focusing on each tech tree) and retrofitting between them when you want to research a tech from a different tree... something that the AI don't know how to do (though it could certainly be programmed to)...
but then it's another "exploit" resulting from the game mechanics that the game doesn't seem to have been initially designed with in mind, otherwise why having separate tech trees in the first place?

[Tcby]

a) The incentive it gives you is really very minor. The only reason its more viable now is because labs are overall slightly cheaper.
b) I have no issue with people using that tactic. As long as the AI can reach capacity, and can prioritize research targets properly, I don't care how people choose to use their capacity.

[Tehlongone]

I wouldn't say anything has changed in that regard, the cost of research centers never ever deterred anyone, this change only affects the AI.

It's a rather round-about way of fixing it though, I don't know why the AI can't just check it's research capacity and retrofit it's spaceports until it's filled.

[Tcby]

The talk of cost came from a discussion I had with Spidey, in which he mentioned that he puts his research labs on private sector bases to avoid the maintenance fee. I agree: it is so minor that its only a consideration for heavy powergaming. Not a balance issue.

[Icemania]

ORIGINAL: MartialDoctor
Yeah, those changes are looking nice. However, I don't get the increased lab output. It's already too easy to reach maximum research capacity as is.

My take was that it would help the AI more frequently research at or near full capacity.

[Icemania]

ORIGINAL: Bingeling

ORIGINAL: Icemania

The new research modding feature could be used to further improve AI research focus e.g. don't waste time of researching weapons technologies that will never get used.

As far as I know, the AI have not spent time on non-used weapons since Shadows or so. The research order can always be discussed, though.

What happens at end tech, I am not sure. I guess it will eventually research everything.

A mod that uses this and also improves Ship Design templates to focus on less weapon types (noting the current smorgasboard) could be made to vastly improve the AI's weapons research/use deficiencies that I've pointed out in this thread.

Modders ... ??? And even better a topic for a future patch ... ???

[Locarnus]

I like the Idea that the Research Projects themselves can be restricted to races, although i First have to Look at how it is implemented and can only do this monday.
I Hope it is only optional and the Old Way is still possible. With the current Way i can Mod the Research.txt with Restrictions, but it is compatible with all Mods using the Vanilla Research.txt. I ll have to Look at the New Way First.

Imho the only Way to have competitive Research from pre Warp Start for the ai is a totally different Research Points generating System. The current One is just massively exploitable.

[MartialDoctor]

ORIGINAL: Icemania

ORIGINAL: MartialDoctor
Yeah, those changes are looking nice. However, I don't get the increased lab output. It's already too easy to reach maximum research capacity as is.

My take was that it would help the AI more frequently research at or near full capacity.

Well, at least they addressed it. However, it would have been better if they had actually just adjusted the AI... I'm guessing it was not an easy fix to change the AI scripts.

Oh well. Beggars can't be choosers [:)]

[Icemania]

ORIGINAL: Locarnus

I like the Idea that the Research Projects themselves can be restricted to races, although i First have to Look at how it is implemented and can only do this monday.
I Hope it is only optional and the Old Way is still possible. With the current Way i can Mod the Research.txt with Restrictions, but it is compatible with all Mods using the Vanilla Research.txt. I ll have to Look at the New Way First.

Imho the only Way to have competitive Research from pre Warp Start for the ai is a totally different Research Points generating System. The current One is just massively exploitable.

If you look earlier in the thread I had a lot of difficulty getting Disallowed Components and Disallowed Research to work ... is that what you meant by the old way?

Either way, if we can get the AI to restrict weapon research to focus on one or two weapons, and also change the ship/base design template files to align with those restriction, we could see another good improvement in AI performance with weapons.

That said, I don't know what Matrix have cooking in coming patches. Erik/Elliot?

[Locarnus]

ORIGINAL: Icemania

ORIGINAL: Locarnus

I like the Idea that the Research Projects themselves can be restricted to races, although i First have to Look at how it is implemented and can only do this monday.
I Hope it is only optional and the Old Way is still possible. With the current Way i can Mod the Research.txt with Restrictions, but it is compatible with all Mods using the Vanilla Research.txt. I ll have to Look at the New Way First.

Imho the only Way to have competitive Research from pre Warp Start for the ai is a totally different Research Points generating System. The current One is just massively exploitable.

If you look earlier in the thread I had a lot of difficulty getting Disallowed Components and Disallowed Research to work ... is that what you meant by the old way?

Either way, if we can get the AI to restrict weapon research to focus on one or two weapons, and also change the ship/base design template files to align with those restriction, we could see another good improvement in AI performance with weapons.

That said, I don't know what Matrix have cooking in coming patches. Erik/Elliot?

Yes, basically I meant with the old way, that race specifics are stated in the race files and not spread out to other files.

From what I saw in the research.txt, using the new way will remove the modularity of modding in research. It further shifts modding towards full conversion and is thus incompatible with my research mod.

A better way would have been to merge the tech foci and allowed/disallowed components into race component/tech modifiers.
Eg
15, 0.0, 87, 0.2, 114, 2.0, 64, 9.9
would mean that component 15 is disallowed, component 87 is avoided (1.0 would be normal), component 114 is a tech focus and component 64 is a racial component.

I will have to wait and see if and how much Haree78 and other race modders make use of the new approach (non-full conversion race modders will be severely handicapped by this as well, since it will become much harder to just add their race(s) to other mods).

I like to use different themes/racial mods myself and I liked combining mods.
If my research mod is thus effectively restricted to vanilla or a support role for a specific mod, it is not worth my time and effort anymore, so I ll have to stop development of my mod once again (just after the energy balancing is resolved) and wait if/how this "feature" is adopted by the race modders.