Distant Worlds AI

[Icemania]

ORIGINAL: Bingeling

ORIGINAL: Icemania
<Snip>

Still no idea how to make material changes to Ship Designs with these policy files .. help!

Good suggestions, but buffing mines that much would have to be combined with a pirate update. There would no longer be any purpose for small pirate fleets, or single pirate ships. It is a good thing for an empire to be safe from pirates, but pirates should be a fitting threat.

What is the problem with policy files?

The design files are easy to update, just look at them, but updating all of them is probably a chore.

To Shark: Does your designs give some variety to different race's designs, at least in what weapons are used? There has also been an idea of better designs for more militaristic and more intelligent races, do you have this to some degree? Also, things like more speed/less weapons, less size for the gizureans?

Per previous post I was hoping to see conditional logic with construction size. Explorers were an example but there are many others.

The Pirates have a huge effect on AI empires, buffing mines might help slow that down a bit, and they are EXTREMELY weak designs ... even relatively weak Pirates can do a lot against them.

That said, I certainly wouldn't object to buffing the highest strength setting for Pirates as well.

Certainly agree to keeping variety between races, the intent above was to pick up on the more obvious issues.

[Bingeling]

I think pirates are meant to be annoying, but if the human player is very good against them it gives a huge head start.

My point is that the various pirate settings should give a certain level of challenge, and their behavior and policies must be updated if their targets are significantly buffed. It could be possible to modify some fleet settings in some empire policy files, I have never touched the modding side much at all.

It could also be that pirates use a hidden "power" setting of their possible targets, that does not take actual designs into account. If so it would require a program update, or pirates will continue to throw single ships on the new fortress version of mines. For instance, early in the game pirates harass a small colony with 3 ship fleets. Later in the game they use much more. They could have a "power" measure that depend on for instance population.

[Buio]

ORIGINAL: Shark7

To all of those complaining about the AI, yet designing over-powered ships, I offer this challenge. Play the game with the AI designing your ships and do not design your own.

You will fast find out that its not the AI that is the problem, it is ship design. Namely the fact that most players cannot resist the urge to make their own ships so powerful that nothing could stand against them.

I made a mod that brought up all AI ships to my typical designs. I get beat routinely...by my own designs. It really does come down to ship design and the undeniable urge to over-design for the player, while the AI is not capable of adapting.

The AI is quite capable, when its not competing against a player designed navy that is undeniably over powered.

Wasn't it Stardrive that let the AI use the players designs automatically? I think that can be a good idea, maybe made somewhat more realistic with a delay to simulate industrial spionage over time. Probably also the easiest method to implement in the game. Redoing the AI design process could be a more daunting task, perhaps not possible until an eventual DW2.

[Bingeling]

Buio, there is not much of an AI design process. There are templates, and the AI design after what is in them. The only exception is that the AI tries to fit the designs under the maximum build size by dropping some stuff when necessary.

[Icemania]

Some other suggestions after looking through the Empire Policy files and Modding guide:

1. A lot of the Ship Designs and Empire Policy files show a focus on two primary weapons technologies (or even more in some Ship Design Templates). For example Quameno mix up Beams and Torpedoes. The problem with this is that it doubles the time required to get to higher tier weapons. The Ship Design Templates could be rationalised to focus on one primary weapon per race and Empire Policy files changed to focus on one weapon as well.

2. In the Empire Policy files all races should have a ResearchTechFocus for Construction and Hyperdrives (at least for pre-warp starts until the first couple of technologies are done as a minimum). Very few Empire Policy files have this at present. There is also a Tech Focus field in the Race Files where this could be applied.

3. Something I've done already is change the proportion of small versus large ships in the Empire Policy Files. It's far too focused on small ships I would recommend this is changed substantially. Not sure about the knock-on effect though.

4. I was hoping we could set a Tech Focus for the Wonder Research lines e.g. through the economy line for the Teekan say. Unfortunately most are not on the list except Medicine and Recreation. This would help make the AI more competitive with Wonder Building.

5. The AI massively overbuilds facilities. For example all the Acadamies are built (Spy, Science, Naval and Military) by population 5B in the Empire Policy files. Recommend the population at which they are built is massively increased e.g. 10B+.

Elliot/Erik, if you read this thread, is there anything you plan to pick up in a patch beyond having another look at AI taxation?

If not, will any of our mod team help Shark? Rather than a Ship Design AI Mod let's do an overall AI improvement mod that does everything possible.

Any other good suggestions other there?

[Gregorovitch55]

ORIGINAL: Oberst_Klink

Im a not a modder, neither I am good about AI scripting, but there's an article I posted for TOAW III, regarding AI design, behaviour and possible scripting that might be ancient, but proved to be very efficient in a 90s wargame that was unfortunately scrapped. The AI script and philosophy was quite advanced for that time.

fb.asp?m=3535976

Klink, Oberst

I guess this is a bit of a side line to the main discussion which is focused on practical tweeks, but I have known that war games based largely on scenario maps use scripting like this to good effect (the good ones anyways, TOAW, Steel Panthers etc) to provide a decent challenge to human player against programmed opponent, especially played first time blind.

I know GalCiv2 also uses this technique, which is instructive in the context of a 4X. An interesting example is it's methodology preparing for an attack. What it does is to prepare a number of troop ships in advance of declaration of war which it hides conveniently near to its target systems but just out of scanner range of the victim's patrols (it knows the scanner grade of the opponent after meeting some of it's ships in open space) so the player can't see them. After declaration it sends in its attack fleets to clear system defenses than immediately invades with the troop ships that are typically only 2 turns away if that from target. Against less experienced players this tactic is very effective.

However there is one big flaw in this script: the player can build scout ships that have multiple scanners which are additive in range, so instead of seeing maybe four or five parsecs they can see everything in a 20-25 parsec circle around them. This renders the AI's preparation transparent giving the player ample time to prepare an ambush to destroy the entire troop ship armada on declaration and intercept and destroy the attack fleets easily.

So one small flaw or oversight in a scripted approach renders a previously deadly AI tactic laughably easy to counter once a player discovers it because a script cannot adapt. This doesn't really happen in war games (unless you replay the same scenario over and over) but it argues that generic algorithms that are less powerful are better in 4X games since they are at least to some extent adaptable.

[Spidey]

Ice, I generally agree with your change suggestions, but I do think we have really different ideas about how things should be designed. It's probably just a difference of temper and playstyle.

Small Space Port
• General. Plants: reduce from 2 to 1. The extra plants do not change construction speed.
• General. Labs: reduce from 3 (1 of each type) to 0. The AI should be keep them in more heavily fortified installations.
• General. Construction yards: reduce from 4 to 2. Important from Small Space Ports to build quickly to aid defence.
• Race Specific. Weapons: there are three weapons (e.g. torpedo, beam and fighters) instead focus on one primary weapon where the majority of research is done for that race.

We seem to have the polar opposite perception of a small space port. [:)] I don't consider small ports a defense at all. I build them as sort of a pre-cursor to a medium port. They'll have six yards and 12 docking bays, obviously a bit of shield and armor, and really not much else, creating an outpost where stuff can be repaired and built if necessary, but without the ridiculous maintenance cost of a base that actually stands a snowball's chance on Hawaii to defend itself. It's not like there's any chance of me building a star port in a system that doesn't have a designated defensive fleet anyway, and it the port goes down then it's easy and cheap to rebuild.

But the way I see it, if the port can't build or repair more than two things at a time then what's the point? May as well just build a star base or defensive base instead. And it isn't like a small port can actually defend itself against tech parity enemies without not actually being small at all.

Large Space Port
• General. Construction yards: increase from 12 to 36. The number of yards is far too small, the AI has backlogs far too often.
• General. Docking bays: increase from 24 to 48. The number of docking bays is far too small, the AI has backlogs far too often.
• General. Plants: reduce from 4 to 2 (or even 1?). The extra plants do not change construction speed.
• General. Empire Policy: Large Spaceports are built at 3B, change to 10B. 3B is far too early for such a massive structure. Smaller worlds do not merit a Large Spaceport.
• Race Specific. Weapons: there are three weapons instead focus on one primary weapon where the majority of research is done for that race.
• Note. Labs could be optimised but I'm wary of potential knock-on effects.

I think 36/48 yards/docks might be a bit of an overkill. I know I'm capable of teching up my yard build speeds while the AI isn't, but I'm getting by fine with 24/30 for most of the game and usually only upgrade to 30/45 for the heck of it. 1 of each plant, 30 yards, 45 docks. How often do you ever need more? And how often are you paying maintenance for yards that are standing totally still? That said, I haven't actually checked the difference in maintenance so this might be a totally insignificant objection. [:D]

[Oberst_Klink]

ORIGINAL: Gregorovitch55

ORIGINAL: Oberst_Klink

Im a not a modder, neither I am good about AI scripting, but there's an article I posted for TOAW III, regarding AI design, behaviour and possible scripting that might be ancient, but proved to be very efficient in a 90s wargame that was unfortunately scrapped. The AI script and philosophy was quite advanced for that time.

fb.asp?m=3535976

Klink, Oberst

I guess this is a bit of a side line to the main discussion which is focused on practical tweeks, but I have known that war games based largely on scenario maps use scripting like this to good effect (the good ones anyways, TOAW, Steel Panthers etc) to provide a decent challenge to human player against programmed opponent, especially played first time blind.

I know GalCiv2 also uses this technique, which is instructive in the context of a 4X. An interesting example is it's methodology preparing for an attack. What it does is to prepare a number of troop ships in advance of declaration of war which it hides conveniently near to its target systems but just out of scanner range of the victim's patrols (it knows the scanner grade of the opponent after meeting some of it's ships in open space) so the player can't see them. After declaration it sends in its attack fleets to clear system defenses than immediately invades with the troop ships that are typically only 2 turns away if that from target. Against less experienced players this tactic is very effective.

However there is one big flaw in this script: the player can build scout ships that have multiple scanners which are additive in range, so instead of seeing maybe four or five parsecs they can see everything in a 20-25 parsec circle around them. This renders the AI's preparation transparent giving the player ample time to prepare an ambush to destroy the entire troop ship armada on declaration and intercept and destroy the attack fleets easily.

So one small flaw or oversight in a scripted approach renders a previously deadly AI tactic laughably easy to counter once a player discovers it because a script cannot adapt. This doesn't really happen in war games (unless you replay the same scenario over and over) but it argues that generic algorithms that are less powerful are better in 4X games since they are at least to some extent adaptable.

GalCiv2

Thanks for the answer and insights! I guess it ain't workable at 4X games like in wargames or cosims; much appreciated your comment.

Klink, Oberst

[Icemania]

Thanks Spidey, was hoping this would lead to some good discussion.

I agree we all have different styles, it's hard to put our natural bias aside, the focus here should be what works best with the AI. Personally I don't use Small Spaceports at all anymore and my Large Spaceports are even bigger ... I find 48 Yards is sometimes too few mid-game ... but again this is for the AI.

For the AI designs a Small Spaceport is also a pre-cursor to a Medium as shown in the Empire Policy files. The population where a Small is upgraded to a Medium, and a Medium is upgraded to a Large, is configurable. The idea here was to get rid of components that are definitely not needed such as labs and excessive plants. I'm flexible on the number of Construction Yards but the thinking was it will be upgraded at some stage anyway and early colonies don't have many resources. Note the AI small spaceport designs do have a reasonable degree of weapons and shields which could be rationalised if we wanted to, but given the AI limitations, I decided not to comment on that.

For Large Spaceports the answer probably depends on the difficulty setting and race. That many probably aren't needed on Normal but the design templates do not change with difficulty. Some races spam ships that all need to be built and upgraded, it's not uncommon to see the AI with a massive backlog, refer image below. One way or another, 12 construction yards is far too few ... 17 ships waiting below. I would say at least 24 yards are needed! Maintenance cost delta from 12 to 24 is about 600 ...

[Buio]

ORIGINAL: Bingeling
Buio, there is not much of an AI design process. There are templates, and the AI design after what is in them. The only exception is that the AI tries to fit the designs under the maximum build size by dropping some stuff when necessary.

Ok, I didn't know how it works technically. So should be possible to let players design a bunch of templates and let the AI use them in the game. In that case, I don't understand why it hasn't already been done.

[Spidey]

@ Ice

No worries, I'm happy to discuss most things, but it's hard to discuss without disagreement in the first place. Playstyle does make quite a difference and yes, we obviously need to consider the AI's occasionally peculiar way of playing. To that extent, my experience with the AI is mainly from very hard, 240k tech, normal-many normal-strong pirates average distance, 15x15 galaxy, and normal-plenty independents.

That said, those Gizureans really are freaking crazy, even for bugs, and that fairly massive maintenance reduction they have isn't helping. I guess they probably do need more yards in their space port designs, but in general I think you have to push pretty damn hard to actually need that much capacity. You do raise another good point too. Design templates don't change between difficulties nor between early and late game. Late game it might make sense to have "crazy capability" whereas going for it with early game components leads to excessive waste. I don't think there's a whole lot to do about that, though. Not without being able to specify a minimum tech level for designs.

Anyway, just checking my default Quameno vanilla LSP design, it has a maintenance of 6130. It's the usual pile of floating garbage with cargo bays, 5 of each lab, 4 of each plant, a monstrous 17 space reactors, a ridiculous 45 (!!) pulse blasters, 21 small fuel cells, and 20 seeking missiles. Granted, if the weapons weren't so worthless and the reactors actually provided any power then it wouldn't be quite as bad. The template says 45 beams and 20 torps and I guess 45 titan beams and 20 thunderbolts would be rather unpleasant to tangle with.

Taking that to 1 of each plant, 75% fuel cells, 30 yards and 45 bays (leaving the other stuff unchanged so as to not confuse the AI too much), the maintenance cost is 6593. Adding a further 10 yards and 5 bays takes the cost to 7168. Covering this in energy collectors costs a further 4-500 in maintenance. All in all, it doesn't really look like it's going to make much of a difference, but it sure is a heck of a lot more than the 1200-1400 maintenance on my homeworld space port while I'm rushing growth during the first 5-10 years of the game when I'm rushing growth.

[Tanaka]

ORIGINAL: Buio

ORIGINAL: Shark7

To all of those complaining about the AI, yet designing over-powered ships, I offer this challenge. Play the game with the AI designing your ships and do not design your own.

You will fast find out that its not the AI that is the problem, it is ship design. Namely the fact that most players cannot resist the urge to make their own ships so powerful that nothing could stand against them.

I made a mod that brought up all AI ships to my typical designs. I get beat routinely...by my own designs. It really does come down to ship design and the undeniable urge to over-design for the player, while the AI is not capable of adapting.

The AI is quite capable, when its not competing against a player designed navy that is undeniably over powered.

Wasn't it Stardrive that let the AI use the players designs automatically? I think that can be a good idea, maybe made somewhat more realistic with a delay to simulate industrial spionage over time. Probably also the easiest method to implement in the game. Redoing the AI design process could be a more daunting task, perhaps not possible until an eventual DW2.

Actually the pirates in DW do copy your designs and use them. Why this power not given to the races I have no idea as it would improve the AI dramatically. And as you say it would be much easier to give the AI races this power than to redo the entire AI design process. If it is possible for modders to give the AI races the same copycat power the pirates have I have no idea. If not please Erik and Elliot give this ability to the AI races as well!

[Shark7]

ORIGINAL: Icemania

ORIGINAL: Shark7
OK, give me a couple of days to get it all compiled and I'll post it in the modding forum.

Shark ... just having a look again at the ship design policy files.

I was hoping to see conditional logic with construction size as this is the biggest factor in determining the number of components that are feasible. If this was in place the vast majority of my ship designs could be emulated.

A specific example would be Explorer Ships. My pre-warp size 160 Explorer has 16 Ion Trusters so it can get around my home system very quickly but only 4 fuel cells because no more are needed. However, my post-warp size 300 Explorers has 14 fuel cells because there is a massive galaxy to explore and it can't afford to be refuelling all the time.

I was hoping this could be implemented but alas it appears not. Hopefully I've missed something? When I then consider what could be done with the constraints of these policy files it's extremely limited.

Base designs we could probably improve a fair bit.

The Empire Policy files could also use rework.

I'm going back through all of mine, making sure they're balanced, so it may take longer than originally planned.

There are only one set of templates unfortunately, so you have to design with the end game in mind.

As far as the design templates, the AI will modify them (to an extent) to fit within the current construction size constraints. Basically I have found you will always have a functioning Escort, Frigate, and Transport design at game start. In my case, the frigate has reduced numbers of shields, engines and weapons (when compared to the template) to make it fit under the 230 limit. The destroyer design is not modified and I have to research up to large scale construction (size 500) to build any.

[Bingeling]

ORIGINAL: Icemania

Some other suggestions after looking through the Empire Policy files and Modding guide:

1. A lot of the Ship Designs and Empire Policy files show a focus on two primary weapons technologies (or even more in some Ship Design Templates). For example Quameno mix up Beams and Torpedoes. The problem with this is that it doubles the time required to get to higher tier weapons. The Ship Design Templates could be rationalised to focus on one primary weapon per race and Empire Policy files changed to focus on one weapon as well.

2. In the Empire Policy files all races should have a ResearchTechFocus for Construction and Hyperdrives (at least for pre-warp starts until the first couple of technologies are done as a minimum). Very few Empire Policy files have this at present. There is also a Tech Focus field in the Race Files where this could be applied.

3. Something I've done already is change the proportion of small versus large ships in the Empire Policy Files. It's far too focused on small ships I would recommend this is changed substantially. Not sure about the knock-on effect though.

4. I was hoping we could set a Tech Focus for the Wonder Research lines e.g. through the economy line for the Teekan say. Unfortunately most are not on the list except Medicine and Recreation. This would help make the AI more competitive with Wonder Building.

You make a lot of sense, some comments, though.

The "2 weapon" setups are a feature and a choice, your 1 weapon way is probably stronger and should be considered.

I believe AI research is pretty scripted in pre warp. It seems to always put warp precursors as the "next" project, and hyperdrives right after that. There is a set of things needed to really "roam the skies" in style, and quite a few are in construction (230 construction size, shields, energy collection). The AI also has it pretty scripted to build research stations at the capital to work on reaching research potential, always preferring an energy station first.

I have run a lot of automated games starting in pre warp, and issues I have reported from watching the AI fumble at the start have been adjusted. They still run off to die at any kaltor, though

Research has its issues after getting hyperdrives, though. It ignores the "irrelevant weapons", but it is very "all at the same level, tech focus one level deeper" or something like that. Energy research suffer the most with multiple engines, hyperdrives, shields, all being researched to level 1 before moving to level 2 on anything. Of course, there are different roles that benefit from different hyperdrives, but the AI is not that fancy, it is just confused

[pycco]

ORIGINAL: Icemania

•Note. Labs could be optimized but I'm wary of potential knock-on effects.

No major comments on Defense Bases

building the majority of research stations at the primary space port, would allow for the science stations to not need as much and could be ok with no defense and attacking features. this would save lots of $$ for other things while still giving the bonus.

defensive bases need more firepower and defenses, this would go along way in helping them hold ground, free up ships for attacking and not have as many unguarded planets.

[Hannable]

What it comes down to is that the human player learns.

The A.I. doesn't.

Which means it's only a matter of time before most players will stumble - even accidentally - across a ship design that will wipe the floor with anything the AI knows how to build. The problem with the AI is that it will generally build the same design all the time - not counting retrofits.

The game Fleet Command, for instance, had an AI problem whereby the AI would always use its most powerful weapons first, which is why nukes were never modded into the game. If you give the AI nukes, that's the first thing it will hit the human player with, pretty much ending the game immediately.

In the same way, the AI will usually build the most perfect design it knows how to build given its tech level. This is an age old problem of the AI being always imperfect or always too perfect.

The best way, IMO, to solve this problem is for each AI race to have multiple templates to choose from, and the AI will build from one of those templates chosen either at random (easier to code) or based on the kinds of ships the player is building (much harder to code).

What this does is it prevents the AI from always building the same thing - which is usually the best design it can. This simulates a human player making both good and bad choices in terms of design. The AI might built a lousy cruiser at one point in the game, but giving it a chance to pick a better design later on. This can really keep a player on his toes not ever fully knowing how good the AI's ships are going to be. Some templates might even be of those overpowered superships that can be built ... other templates will be considerably weaker. This lends variety to the gameplay both within a game and between multiple games.

Obviously I haven't pondered this for very long, so I'm throwing out very preliminary thoughts, but I do think this (and most) games could benefit from having a plethora of avenues the AI can pursue when designing ships - so that every design isn't perfect and every design isn't horribly imperfect. And the player will never quite know what to expect.

[Icemania]

ORIGINAL: Spidey
No worries, I'm happy to discuss most things, but it's hard to discuss without disagreement in the first place. Playstyle does make quite a difference ...

As mentioned let's forget our styles and focus simply on what would work best with the AI. Some of the changes I've posted are a lot different to my personal designs for example that reason.

[Icemania]

ORIGINAL: Shark7
I'm going back through all of mine, making sure they're balanced, so it may take longer than originally planned.

There are only one set of templates unfortunately, so you have to design with the end game in mind.

As far as the design templates, the AI will modify them (to an extent) to fit within the current construction size constraints. Basically I have found you will always have a functioning Escort, Frigate, and Transport design at game start. In my case, the frigate has reduced numbers of shields, engines and weapons (when compared to the template) to make it fit under the 230 limit. The destroyer design is not modified and I have to research up to large scale construction (size 500) to build any.

Take your time Shark but I strongly suggest the end product needs the Empire Policy and Base Design changes we are discussing as well.

[Icemania]

ORIGINAL: Bingeling
You make a lot of sense, some comments, though.

The "2 weapon" setups are a feature and a choice, your 1 weapon way is probably stronger and should be considered.

I believe AI research is pretty scripted in pre warp. It seems to always put warp precursors as the "next" project, and hyperdrives right after that. There is a set of things needed to really "roam the skies" in style, and quite a few are in construction (230 construction size, shields, energy collection). The AI also has it pretty scripted to build research stations at the capital to work on reaching research potential, always preferring an energy station first.

I have run a lot of automated games starting in pre warp, and issues I have reported from watching the AI fumble at the start have been adjusted. They still run off to die at any kaltor, though

Research has its issues after getting hyperdrives, though. It ignores the "irrelevant weapons", but it is very "all at the same level, tech focus one level deeper" or something like that. Energy research suffer the most with multiple engines, hyperdrives, shields, all being researched to level 1 before moving to level 2 on anything. Of course, there are different roles that benefit from different hyperdrives, but the AI is not that fancy, it is just confused

The 1 weapon approach will make a big difference given research times.

On AI research scripts in some cases you may well be right, I would appreciate if Elliot or Erik could comment on that, because it's not clear in the Modding Guide.

In my games so far with Universe some AI still seem to take far too long to develop Hyperdrive.

I really hope they implement the ability mod the Research Build Order for each race as that would help a lot.

[Icemania]

ORIGINAL: pycco

ORIGINAL: Icemania

•Note. Labs could be optimized but I'm wary of potential knock-on effects.

No major comments on Defense Bases

building the majority of research stations at the primary space port, would allow for the science stations to not need as much and could be ok with no defense and attacking features. this would save lots of $$ for other things while still giving the bonus.

defensive bases need more firepower and defenses, this would go along way in helping them hold ground, free up ships for attacking and not have as many unguarded planets.

I got rid of Labs on Small Spaceports only so the AI will still have lots of Labs via Medium (3 of each type) and Large Spaceports (5 of each type).

That said, pre-warp the AI will need to build Research Stations to get anywhere near Maximum Research rates (with existing designs). This is an unnecessary handicap early game. We could increase the Labs to 10 of each type on a Large Spaceport? Problem is the AI will overbuild Labs by an even greater margin but it will help speed up their early game research. Again it's a pity there is no conditional logic i.e. to stop them overbuilding labs in time. Or if we could specify a "Homeworld Only" Spaceport design?

The Defensive Bases don't seem that bad. 16-32 Shields and 27-48 Weapons ... depending on race?