Your feedback on AI - Endless Legend

Your feedback on AI

Copied to clipboard!

Mitroll

Newcomer

Mitroll

Newcomer

10 years ago

Jun 26, 2015, 12:49:57 AM

I only have two complaints of the AI at the moment:

1: AI feels dead. They just run around and do things and generally don't talk to me except to declare war or maybe after a couple dozen turns I'll get a message about how peaceful I am or how we have a common enemy or something. When playing as cultists they attack my converted villages constantly and then act suprised and angry when I kill the armies responsible (which makes being peaceful cultists practically impossible). AI should only attack the converted villages of cultists that they plan on going to war with. Or better yet, you should only be able to attack cultist converted villages that are in your own territory without declaring war.

2: AI doesn't understand the current combat metagame. Currently, stronger units are almost completely untouchable to weaker units, this is partly due to the fact that the current calculation for attack and defense has caps at about .66x and 1.5x, as seen in the following chart: http://hydra-media.cursecdn.com/endlesslegend.gamepedia.com/1/1a/Combat_graph.png

This means that if you have a unit that has 50% more attack and defense than an enemy's unit, your unit will be something like 4-6x as strong as the opposing unit, if all other things are equal (they will likely be in your favor if you're already beating them by this much on attack and defense).

What this means, especially if you use ranged units, is that essentially no number of units can fight against a good stack of well equipped units, if that's a design choice, that's fine, but the AI currently doesn't understand it, and just feeds me with dozens of weak armies that don't even damage my units no matter how much of a tech advantage they have and no matter what difficulty they're on.

Suggested fix: Either make the AI put more priority on properly equipping their armies, and/or elongate that combat graph so that the .4x mean damage occurs at .5x instead of .66x and the 1.1x damage occurs at 2x instead of 1.5x.

These mechanics currently make the vaulters and their special equipment and technolover traits unspeakably unstoppable (as far as I can tell), but I suppose that's for a different thread.

0badgesSend private message

gGeorg

Newcomer

gGeorg

Newcomer

10 years ago

Jun 26, 2015, 8:49:23 AM

Steam Roll tactic

*************

--->Once a player has ONE army with good equipment, starts steam roll other. There is no way to come back for loosing empire. Approval holds you back somehow, but its possible to raze city or sell them for technology and steam roll all day long.

>>>> 1. Make production army units cheaper about 50 % and upkeep higher. Especialy unit who is higher level then brand new unit, must be expensive upkeep. That way its possible quickly re-produce lost units in the back and repel/hold attacker.

>>>> 2. AI should be able to plan strategic defensive tactic. e.g. a survival plan for defensive action on the region and army level. For example raze own city with milita only which cant be guarded properly and Pump new units quickly in far safe city. <<<<

Well, strategic defensive tactic should be available in the first. Currently its not due to sky-high unit prices.

0badgesSend private message

Gilbari

Senior Officer

More friends is always good , have more enemies to have a decent challenge.

Gilbari

Senior Officer

10 years ago

Jun 27, 2015, 6:43:13 AM

Your idea 1) is bad cause it lead to build massed glass cannon and i will build then full firepower in race customisation in combat so i can wipe some ass before sieging with zerg tactics

Idea 2) is very hard to do see Heart of iron 3 for exemple and if AI get attacked in pincer attack he would be dead with this strategic plan due to he don't know how much the opponents has in relative power in more than 2 turn.

0badgesSend private message

Agryppa

Newcomer

Agryppa

Newcomer

10 years ago

Jun 27, 2015, 1:00:22 PM

The biggest problem with AI is simple to put words, but not likely simple to solve:

They simply do not take advantage of what's available to them. Be it resources, units, faction-specific aspects -- anything.

I just got done playing a game (hard difficulty, Endless, Huge, 8 players) where I was almost immediately at odds with another faction as we were the only two sharing a rather small continent and were starting quite close to one another. I was loving it. We were trying to work out our difference via diplomacy but we just wanted too much of the very little neutral ground between us and ended up butting heads. He had much more land than me toward mid game, though, due to some poor decisions on my part and I thought for sure I was going to lose -- and I was happy about it! I thought he had played a better game than me, and it was all good fun... I was wrong.

I decided that I needed to attack with what little I had before he built up too large of an army. I had decent income and a good amount of influence to set policies with.

I absolutely steam rolled his massive empire... little effort.

Turns out he had a stockpile of 10k influence and 5k dust by the time I killed him(?!?!?)

WHY on earth was he not using that to help him out while I was (very slowly) advancing through his empire and taking his land.

So there you have it. I don't know how to be more specific. Please ask questions of me if you want more details, but if I had to sum it all up it would be with my first sentence in this post. AI simply does not use what is available to them to their fullest potential (or even half).

0badgesSend private message

abmpicoli

Newcomer

abmpicoli

Newcomer

10 years ago

Jun 27, 2015, 2:10:53 PM

Just discovered this one... the AI Don't queue stuff... and this means that every time it finishes a research , a building, whatever, it simply throws all overflow away... And all production and research from the first turn is lost... All right, that's a minor one... unless you have a city with 1000 production output and you are building a PUBLIC LIBRARY...

How did I discovered this:

1)enter game in debug mode, do your empire moves / build city, assign research production and then switch to empire 2. You will see: empty research choices; empty queue in the main city. If you try to end turn from empire 2 it will even block you, mentioning that you haven't set production...

2) Switch to empire 1 and hit "end turn"... then go to empire 2. You will see that only now it had chosen a technology and something to build... And all are at 0% progress...

0badgesSend private message

BPrado

Newcomer

BPrado

Newcomer

10 years ago

Jun 27, 2015, 3:42:53 PM

Are you sure only queues make use of overflows? I´ve never seen any difference between leaving a queue and just choosing one building or tech at a time.

0badgesSend private message

abmpicoli

Newcomer

abmpicoli

Newcomer

10 years ago

Jun 27, 2015, 6:39:25 PM

This is kind of a minor issue, because you only lose part of that turn's production... But for the AI to don't even bother to choose a tech and production and lose a full turn worth of the 1st turn production... Well... yes, maybe I'm a little perfectionist...

smiley: biggrin

0badgesSend private message

synra

Newcomer

synra

Newcomer

10 years ago

Jun 28, 2015, 2:17:33 PM

@abmpicoli I'm pretty sure you get the overflow assigned to your next building whether it is queued or not, though I could be mistaken

---

An excellent thread

, I particularly agree with the diplomacy points. The AI needs to pursue broad diplomatic strategies that consider the global game state, formation of superpowers, as well as action and consequence (e.g. If I market ban someone way stronger than me, I might get completely obliterated in a war). There needs to be a sense of cohesiveness, such that your neighbour doesn't appear to be constantly blinking between neutral, mean, and lovely. And they also need to care about what you do -- currently however adorable or horrible you are, it feels like the AI will leave you alone until they desire your territory, and then they will attack you.

It should be possible for an AI neighbour to befriend the player, for example if the player is a small empire (thus not excessively competing for land) who does not actively offend the AI, and if larger factions exist elsewhere. Currently dealing with the AI is like giving a bully my lunch money. Sure, they might leave me alone for a few minutes, but I'm not really achieving any goodwill despite emptying my coffers.

0badgesSend private message

abmpicoli

Newcomer

abmpicoli

Newcomer

10 years ago

Jun 28, 2015, 7:52:33 PM

Regarding research and production queues, I think I stand corrected... I've just tested this and really there doesn't seem to exist a significant difference...

0badgesSend private message

VIPPropbuddha

Newcomer

http://ladderduel.com/league/elmultiplayer/

VIPPropbuddha

Newcomer

10 years ago

Jun 29, 2015, 7:27:12 PM

I haven't tested it lately, but leftover

smiley: industry

and

smiley: science

is supposed to carryover to the next turn and be applied to the item added to the queue. There should be no difference as long as you add something to the queue every turn.

0badgesSend private message

gGeorg

Newcomer

gGeorg

Newcomer

10 years ago

Jul 2, 2015, 11:37:56 PM

Gilbari wrote:
Your idea 1) is bad cause it lead to build massed glass cannon and i will build then full firepower in race customisation in combat so i can wipe some ass before sieging with zerg tactics

Idea 2) is very hard to do see Heart of iron 3 for exemple and if AI get attacked in pincer attack he would be dead with this strategic plan due to he don't know how much the opponents has in relative power in more than 2 turn.

1. your idea is bad because your sentence do not have a sense.

2. I am sorry, just don't get a meaning, if there is any.

Do you use some kind of a random word generator?

0badgesSend private message

DEVwilbefast

Lost Dev

Something is afoot at the end of my leg!

DEVwilbefast

Lost Dev

10 years ago

Jul 7, 2015, 4:13:28 PM

Hello everyone,

As mentioned briefly on the introduction thread I'm the new game designer on the AI team - I'll be taking over where Manu left off

To start I wanted to thank you all for the detailed feedback: it's extremely useful to us to have this all in one place. In keeping with tradition I've tried to sum up the past posts in a condensed form - feel free to correct me if I've misrepresented your suggestions in any way!

Tactics

Opt-out of using reinforcements if the battle seems to be doomed.
"Think" more holistically, consider overall unit formation rather than focusing on individual tiles or targets.
Move out of a reinforcement tile when reinforcements are still waiting during the battle.
Don't prioritise militia!

Strategy

Take into consideration the best angles to attack (reinforcements, deployment zone, etc.)
Focus more on strategic resources (on this point, the last update improved a lot how AI prioritizes strategic resources).
Gang up on winning empires
Have better army mixes.
Make better hero assignations based on hero specialisation.
Be able to equip accessories correctly based on hero specialisation and current assignation.
Retrofit its units faster/better.
Use faction abilities such as Force Truce or Vaulters’ teleportation
Be more aware of weak and strong points in their and other empires' borders.
Use the market more, especially Roving Clans.
Handle ships better, prioritise disembarking so that reinforcement is possible.
Handle settlers better: don't send them off alone or have them wait around in hostile territory when another empire colonises the territory first. Don't settle vulnerable areas without military support.
Use privateers more often, especially if playing as Roving Clans.
Recognise and respond adequately to "total war" type scenarios: concentrate on a single front rather than spreading forces evenly.
Respond negatively to "aggressive colonisation" (ie. "hey! I wanted that territory!").
Respond negatively to the conversion of minor factions on sovereign territory.
Queue production to avoid losing the overflow of resources.
Judge better when to abandon cities, pull out units and salt the earth.

Diplomacy

When trading with other empires take into account the market value of the resources being traded.
Certain personalities should not exist for specific factions (eg. peaceful for Necrophages).
Recognise trends in diplomatic relations: hold grudges against ancient enemies and don't betray old friends as readily and accept their infractions (if minor).

Thanks again for all the feedback - I'll see you around the forums

0badgesSend private message

Caotico09

Newcomer

Caotico09

Newcomer

10 years ago

Jul 7, 2015, 4:55:24 PM

Welcome wilbefast.

Your list looks pretty complete. If I think of anything else I will add to it.

I just wanted to reply to highlight two of your points, and stress why I think they are important.

wilbefast wrote:

Gang up on winning empires

It doesn't even have to be "ganging up". The idea here is that in EL, you can 'easily' manage the diplomatic relations that you want with the AI, although it might cost you at times. The diplomatic "yes/no" bar makes this transparent, so what may seem an automatic "no" in other games is actually seem in EL as "an almost yes". It almost seems to me that the AI needs to be more "stingy" because of the "yes/no" bar.

The main idea is that you can systematically destroy one empire at a time while not worrying too much about your borders with other civilizations. An easy comparison to draw is Civ 5. There, if you have systematically destroyed your 3 nearest neighbors and declare war on another one, then the Civ behind you normally takes the opportunity to attack you as well. Civ 5 is over the top in the respect that it is hard to ever gain trust back, but in my opinion a happy medium can be found.

wilbefast wrote:

"Think" more holistically, consider overall unit formation rather than focusing on individual tiles or targets.

Stress on the "holistic". EL's combat is a great idea, and the weighting seems like a great idea too. My main issue is there is not enough variation in the AI's plan to account for most scenarios. Focusing on certain units is great sometimes but horrible at other times. However, imo, the targeting wouldn't be a huge issue if unit movement was correct. A more holistic unit movement, where units don't try to chase down the "ideal" target that is behind enemy lines, and in turn segregate themselves from your own army would be fantastic. But of course it is 100x more complex then that I know. :/

0badgesSend private message

Andy06r

Newcomer

Andy06r

Newcomer

10 years ago

Jul 7, 2015, 8:15:43 PM

City Placement snowballs into the other economic issues. A popular civ4 mod improved the AI a full difficulty just by optimizing city placement.

1) Is the AI aware of rod and triangle floor plans (plus exploitation) for expansion? At a minimum the AI should consider the best 2x3 rod and 3-sided triangle and then pick the best individual tile within the floor plan for the city district.

2) Is the AI aware of the value of rivers and coastal tiles? Aquapulvasitics, the food on river tech, and cargo docks can make a city much better.

3) The cultist AI should consider any neutral land with converted villages as their land for the purpose of borders and should aggressively defend villages. A human cultist will frequently defend villages on common borders - an AI cultist just sends a wrongly worded message when you kill the undefended villages.

4) Are the broken lords, wild walkers, necrophages, and vaulters aware of their tile modifications? Implementing #1 would solve this.

Off topic) Since I know the devs are reading this and I'm invested in your franchise, are there plans to fix the power curve on Endless Space long range kinetics? Tier 1 LR kinetics have a better damage per ton than tier 1 LR missiles and the LR Kinetics tool tip describes a weakness that is at odds with the raw math.

0badgesSend private message

DEVwilbefast

Lost Dev

Something is afoot at the end of my leg!

DEVwilbefast

Lost Dev

10 years ago

Jul 8, 2015, 11:03:12 AM

Caotico09 wrote:

It doesn't even have to be "ganging up". The idea here is that in EL, you can 'easily' manage the diplomatic relations that you want with the AI, although it might cost you at times. The diplomatic "yes/no" bar makes this transparent, so what may seem an automatic "no" in other games is actually seem in EL as "an almost yes". It almost seems to me that the AI needs to be more "stingy" because of the "yes/no" bar.

The main idea is that you can systematically destroy one empire at a time while not worrying too much about your borders with other civilizations. An easy comparison to draw is Civ 5. There, if you have systematically destroyed your 3 nearest neighbors and declare war on another one, then the Civ behind you normally takes the opportunity to attack you as well. Civ 5 is over the top in the respect that it is hard to ever gain trust back, but in my opinion a happy medium can be found.

Ok, so it's more about having an overall understanding of the context of diplomatic relationships and how they've varied over time - again, "holistic". The way the AI is structured does make it more likely to be opportunistic and less likely to be "dogged" in pursing a goal, this may change in the future though it will require a lot of work.

Andy06r wrote:

Is the AI aware of rod and triangle floor plans (plus exploitation) for expansion? At a minimum the AI should consider the best 2x3 rod and 3-sided triangle and then pick the best individual tile within the floor plan for the city district.

The AI does have heuristics for choosing floor plans - the community has found better ones since they were implemented though (good job guys)

This is something we're thinking about, but of course implementing missing behaviour will always have the priority over making existing behaviour more optimal, and there are still a few situations the AI just react to.

Andy06r wrote:

Off topic) Since I know the devs are reading this and I'm invested in your franchise, are there plans to fix the power curve on Endless Space long range kinetics? Tier 1 LR kinetics have a better damage per ton than tier 1 LR missiles and the LR Kinetics tool tip describes a weakness that is at odds with the raw math.

I don't want to encourage off-topic, but please ask on the ES forums!

0badgesSend private message

Enlil

Newcomer

Enlil

Newcomer

10 years ago

Jul 9, 2015, 5:38:33 AM

wilbefast wrote:

Handle ships better, prioritise disembarking so that reinforcement is possible.
[/QUOTE]

Along with this, AI needs to build and utilize harbors as soon as they learn the technology.

I have traded it to them, or in the case of allies gifted it to them, yet they almost never build harbors. This is really incredible when their city has majority water tiles and they are allies and would greatly benefit from overseas trade.

0badgesSend private message

DEVwilbefast

Lost Dev

Something is afoot at the end of my leg!

DEVwilbefast

Lost Dev

10 years ago

Jul 9, 2015, 7:14:13 AM

Enlil wrote:

Along with this, AI needs to build and utilize harbors as soon as they learn the technology.

I have traded it to them, or in the case of allies gifted it to them, yet they almost never build harbors. This is really incredible when their city has majority water tiles and they are allies and would greatly benefit from overseas trade.

Trade-oriented AI is something we've discussed - currently the AI's "focus" is elsewhere ("direct" economy, military, etc). It's good to have the community take note of this too, so we know whether it's worth prioritising

0badgesSend private message

VIPPropbuddha

Newcomer

http://ladderduel.com/league/elmultiplayer/

VIPPropbuddha

Newcomer

10 years ago

Jul 9, 2015, 7:57:08 AM

wilbefast wrote:
Trade-oriented AI is something we've discussed - currently the AI's "focus" is elsewhere ("direct" economy, military, etc). It's good to have the community take note of this too, so we know whether it's worth prioritising

The problem with social/cooperative features (Diplomacy, Trade, War, the Marketplace, etc.) is that if the AI does not participate, they are essentially dead features.

It also doesn't help that foreign trade routes are difficult to set up:

You and/or trade partner must research Diplomat's Mance
You and/or trade partner must offer a Peace agreement and it must be accepted by the other party.
You and trade partner must research Right of Way -or- Cargo Docks
You and trade partner must build the Right of Way -or- Cargo Docks improvement in a region
There must be a land (continuous regions with Right of Way built) or sea-based trade connection between your and trade partner's cities
You and trade partner must have an available trade route slot in cities
The auto-trade route algorithm must determine that the foreign trade route is a better option than any of your domestic ones

In some games I've gone to the length of giving the AI empires the Right of Way tech. Even then it usually won't even take the step to build it.

0badgesSend private message

DEVwilbefast

Lost Dev

Something is afoot at the end of my leg!

DEVwilbefast

Lost Dev

10 years ago

Jul 9, 2015, 9:15:05 AM

Propbuddha wrote:
The problem with social/cooperative features (Diplomacy, Trade, War, the Marketplace, etc.) is that if the AI does not participate, they are essentially dead features.

It also doesn't help that foreign trade routes are difficult to set up:

You and/or trade partner must research Diplomat's Mance
You and/or trade partner must offer a Peace agreement and it must be accepted by the other party.
You and trade partner must research Right of Way -or- Cargo Docks
You and trade partner must build the Right of Way -or- Cargo Docks improvement in a region
There must be a land (continuous regions with Right of Way built) or sea-based trade connection between your and trade partner's cities
You and trade partner must have an available trade route slot in cities
The auto-trade route algorithm must determine that the foreign trade route is a better option than any of your domestic ones

In some games I've gone to the length of giving the AI empires the Right of Way tech. Even then it usually won't even take the step to build it.

The architecture we have in place makes the AI nice and flexible, but the draw-back is that it tends to be pretty opportunistic rather than doggedly pursuing its goals to the exclusion of all and any distractions that might make it stray off course. As the community has noted a few times in this thread this is better for certain factions than for others (RC and Drakken for example).

0badgesSend private message

Zenphys

Newcomer

Zenphys

Newcomer

10 years ago

Jul 9, 2015, 9:22:39 AM

It's great to get such dev interaction.

What is the game plan moving forward for the AI? Is the focus to primarily fix the issues listed in this thread (It would be fantastic if the whole of that list got delt with). But are there any other plans for the AI? Perhaps making them a little more dogged and focused or maybe introducing more long term thinking. Not being a programmer I’m not sure how easy/difficult these changes are and I imagine there is only so much time and the majority of the architecture will already be in place.

0badgesSend private message

Comment

Characters : 0

0badgesSend private message

PostCancel