Robots and Mechanical Men

from Sins Forums

[Bug v1.041] Once an Advent AI has researched homing mines, its fleets no longer have the support of strikecraft from its carriers.  The Drone Hosts might need a limit on how long they will carry mines.

I had decided to test the strength of the AI against each other on a custom 1v1 multi, but as the first match has taken three hours with a final result unlikely, the full test might proceed slowly!

The first contest had a hard TEC finance AI against the people's champion, the hard research AI, also TEC.  The researcher got off to a good start as its opponent cleared planets without colonising them, an ice because of a lack of labs, and a desert for some unknown reason.  The researcher had an advantage of 3 planets and 2 roids to a planet and a roid, but couldn't exploit it, despite 'catching' the finance AI's first cap with a starbase.  At the three hour stage both AIs had begun to colonise the smaller second system, and the finance AI had a two planet advantage as well as a 4 cap to 1 advantage- including a level 8 and level 6 Marza.

I was locked to the finance AI from the observer spot, so I can't comment as much on the researcher.  However here are my notes on the financier:

It was slow to put up a second civic lab and started with 1+1... unusual for the extra credits path?

Thirty cobalts is perhaps too many before fleet diversification..?  Also, the fleet had four caps at the end, shouldn't a thrifty AI rely on frigates?  Both AI's were very keen on starbases.  The AI may be too concerned with providing an obstacle for a player, rather than attempting to achieve victory, for this test to work..?  However the financier still had a full outfit of siege of both types, it just failed to attack with them after the first encounter.  A defending AI doesn't need the siege ship type until far later.

The AI lost its siege frigates attacking a starbase... thats the regular siege frigates.  Regular siege need an AI that makes them target planets or head for the nearest repair bay...  Also, if the 'lobbers' had a range increase of 50% or even better 100% it would improve the AI greatly, as the researcher AI did regroup its 'lobbers' to attack out of range of the starbase it was attacking, but it had already lost too many.  And the TEC ones look silly, as their weapon has a range of about a dozen times its own length...

The AI littered its back areas with random minefields... phase inhibitors first might be an improvement, if they have to improve defences away from the front at all, shouldn't this be a 'fortifier'?

For a finance-based AI, it only had 6 trade ports on its eight planets, with a longest trade route of two...

The researcher AI lost its caps by retreating them far too late.. 

It's a pity that the same AI name can't be locked to a type of AI, it would add atmosphere.. perhaps this could be added when it becomes easier to customise AI names?

It would be good to have the ability to customise the AI, as it is it seems that this is difficult even in mods, as most of it is hardcoded.  A script-based AI would be ideal, but even with only the ability to adjust some variables there is room for improvement.  I suspect that a customised AI could defeat either the finance or researcher AI easily.  This must be preferable to having poor AI with heavier and heavier production advantages?

Please fix the Illuminators, they are far too important to the game to be left bugged for this long.  And they are still too strong, even when 'fixed'.....

18,899 views 10 replies
Reply #1 from Sins Forums Top

The AI spends inordinately large proportions of its income on static defense, often crippling its offensive fleet.  Combined with the fact that it's really bad at attacking to begin with and you have a big problem.

What bugs me about the Entrenchment AI is that its expenditure is custom-built to inconvenience but not challenge me.  It takes me ages to break its chain of starbases, but I'm assured to do so because it doesn't have sufficient fleet to back it up even with immense resource cheating.  Once we hit the 2-hour mark or so I'm just knocking off planets one by one because there isn't an AI fleet in existance that can challenge me and I'm now outfitted to break starbases with ease, so they have nothing that can stop me.

I wouldn't be surprised if you found a scenario where the AI's literally never are able to beat each other because they can't get past each other's starbases.

Reply #2 from Sins Forums Top

Id like to see a screenshot of your map.

How did you manage to get 2 ai to fight? i cant ever seem to not let myself play.

Ashes of the Singularity II
Reply #3 from Sins Forums Top

The method I used was to have the player occupy a planet that is entirely detached from the rest of the system.  Of course the enemy TEC researcher AI then went for Insurgency, but that only takes one cap to defend and shouldn't affect the rest of the battle... Pirates might be a problem if I had them on, as the non-allied AI would have to split its bounty.  For the second expansion a proper observer slot would help greatly with testing. 

Once detached from the rest of the system you might realise that the AI seems to spout "support is here! the cavalry has arrived!" without it being related to any combat activity on the map...

Here's the next test map, with the detached planet at the top-

I still wanted it to be a multi with all the gravwell types present, but this one is slightly smaller and simpler.

More notes on the AIs...

They seem very reluctant to start with 2 civics which hinders their early expansion.

The aggressor AI I watched placed its second starbase at its home... all the AIs are far too quick to starbase, but the aggressor was even faster than the fortifier.  The AIs seem to value roids, even dead roids, as much as planets when placing starbases. 

[Bug? v1.041] The AI seems to have a problem with Drone Hosts and Homing Mines, so that they don't use their strikecraft.. this hinders any Advent AI which researches mines somewhat, and caused one AI to begin losing a game where it had 12 planets to 6....  The Advent AI also sends single mined-up drone hosts to 'observe' enemy grav wells, very expensive.

The researcher AI produces siege far too early, so that it attacks militia planets with them.  The fortifier AI will even stockpile starbase constructors...

The AIs cannot seem to recognise when pirates are off.

Their fleet composition is awful, especially the Advent AI, as it might have an entire fleet of starfish, purge ships and non-functioning Drone Hosts.   The AIs will have to be scripted away from their addiction to siege to even start to be any good.  Entrenchment might need AIs that cheat more just to compensate for the handicap of adding a second category of siege ships!  However, though it has been claimed that the AI cannot attack starbases, from the games I've watched that is what it is best at, especially non-Advent AIs with high level caps and numbers of carriers as well as the assault siege...

A series of games between the factions next, though it seems odd for single players to complain about multiplayer affecting play balance when the AI is so broken as to be almost unplayable...

Reply #4 from Sins Forums Top

If at first you don't succeed, fail, fail and fail again.  Finally got an AI vs AI win on 1v1.  I had to abandon the multi and simplify further, and it still took 3 hours.  I used this map:

The winner was a TEC finance AI against an Advent researcher... somewhat of a cheat, because as I expected the Advent AI researched homing mines, and its drone hosts began to carry mines everywhere- making them useless in combat.

More AI notes...

The AI needs to start placing repair bays before starbases.. I haven't seen any AI place a single repair before it has at least three starbases.  The AI seems to stockpile starbases before it has any kind of fleet...

There has to be some improvement in how the AI looks after its caps..  obviously they don't search for repair as the AI doesn't build any, but they could at least retreat when down to 33% health..?

I noticed that the Revelation has a habit of stunning a low health cap- then refusing to finish it off...

Siege behaviour is awful.  The standard siege frigates avoid point defence when circling into the planet... but not starbases!  If the AI builds 5-6 caps it should use them for siege, not build 5-6 siege in addition...  The 'lobbers' seem to prioritise point defence..?

The AI builds single trade ports.. it will also research ships then only build 3-4 of that type...

The fleet composition of the victorious TEC AI was so outstandingly bad that at one point the AI's fleet was 3 Akkans, 3 colony ships and 5 siege frigates.. The Advent researcher managed to put a fleet of 9 heavies together with a Revelation, and several Drone Hosts and defence vessels to keep enemy bombers away.  A good attempt, fatally undermined by the Drone Host bug.

The simpler map helped the AI's.  They avoided the centre but still had problems sending units from one side to another.  Typically they would win a battle, then send the fleet to the other side, while the enemy rebuilt.

Making a 5x resource cheat is no achievement for the AI programmers.... are there any other RTS games that have sunk to that level?  However, scripting good AI takes time.  The AI should be opened up to contributions in order to improve it.  There also needs to be a more appropriate method of testing it... just as the multiplayer game needs balance reports from top players playing 1v1 games if it is to improve, rather than a thread of opinions (though this is far better than the almost complete absence of reports from the single player game)...

Reply #5 from Sins Forums Top

Making a 5x resource cheat is no achievement for the AI programmers.... are there any other RTS games that have sunk to that level?
End of quote

Battle for Middle Earth 2.  Get this:  x7 resource cheat.

 

I stand by my initial analysis: the AI needs to be a better attacker.

Reply #6 from Sins Forums Top

I venture to differ with you on that analysis sirrah, though it seems unlikely that we'll ever get the ability to produce custom AI's to settle the matter appropriately, in the style of gentle-androids!  Many of the AI problems caused by Entrenchment are simply due to the nerf to fleet composition, so the Vasari AI becomes superior, as it builds far less mining ships than the other factions build lobbers...

Again as I expected, the Vasari Fortifier was able to dispose of the TEC researcher, though this time it took four and a half hours.  The Vasari AI wasn't especially advantaged by the five neutral extractors (with one of the neutral grav wells entirely empty) as the TEC fought hard for them and held them most of the time up until the end.  This was a pleasant surprise, though it might say more about the Vasari failure to use scouts properly.

However what caused the fight to turn against the TEC was simple- once they had lobbers researched and began to build them as well as siege frigates, their fleets began to lose.  Vasari also have an advantage because the AI is focused on caps, but this isn't nearly as telling as when the TEC AI has about 100 supply more tied up in siege.  The Vasari didn't use their mobile starbases to attack starbases, but this proved no substantial disadvantage, as the TEC AI could not coordinate a defence and their starbases were just overwhelmed.  The Vasari starbases were involved when there was a battle in their gravwell, while the TEC ones were often not a factor- until it was their turn to be destroyed.

So for Entrenchment v1.041 my provisional finding is that any Vasari AI will defeat any TEC AI, and they are both better than the bugged Advent AI, which struggles without its strikecraft.  I'll test further, but this is what I expected and what I've found so far.  So if you must play singleplayer Entrenchment, Vasari are your opponent of choice. 

AI notes-

The AI won't put starbases in gravwells without planets or roids, though they are perhaps too keen to starbase dead roids.  Starbasing Gas Giants is not perhaps a great plan, but given that the AI can't normally put repair near them, neutrals with more restricted space might be suitable?  If you have one or two hanging around that is, as the AI tends to... though the AI places far too many starbases...

There is the potential for an amusing trick on the AI if point defence can be placed correctly.. in theory its siege should then just circle...?  Better at roids where 3 might work...?

AI for patches-

Even if we don't get custom AI, I know that the AI can be much better within the existing hardcoded system, simply because it was, before it was dumbed for Entrenchment.  All the AI's seem to have become variants of fortifier AIs, even the aggressor AIs.  If Advent AIs are debugged and the AI's fleet composition is changed singleplayer will become very much harder, without the need for extravagant and demoralising resource cheats. 

Aggressive and financial AIs should severely limit the number of starbases placed, and all AI's need to have less- the fortifier AI does not need to have a couple of spare starbases already bought!  AIs that have cap-heavy fleets don't need regular siege at all.  AI's especially financial ones should get value from their ship research... why don't the AIs spam?!!  5-6 caps is far too many for a small map like this, even for an aggressive AI, the guideline that caps should be one stage behind supply isn't bad, perhaps with aggressors it could be level?  The lobbers should be a much later luxury item, not an essential element of early fleet composition.  AI's need to prioritise repair bays!

Much can be achieved by production, without the far more difficult alterations to AI tactical behaviour.  However, the most glaring tactical change that could be made is that AI does need to send its fleet to support their starbases when they are under attack. 

Reply #7 from Sins Forums Top

Fleet composition and money management aside, the AI's biggest problem is that it's a poor attacker and defender.  It just throws its forces head-first at the nearest enemy, not regarding the tactical importance of position in the coming battle.  Given how they use starbases, it's no surprise the Vasari variety is a massive advantage.

The problem with assault cruisers and siege frigates is what you've already described; once they make up a big chunk of your fleet, the fighting power of your fleet is greatly reduced.  Sure you can kill starbases and planets fairly quickly, but this means little if the enemy fleet is strong enough to drive you on to the defensive.

I agree that all the AI behaviors right now are essentially fortifiers.  They put WAY too much emphasis on starbase and upgraded starbases.  I see them placed too early and too many, which takes away from their fleet and allows me to control the rest of the map with ease.  Sure, I can't take down that heavily upgraded starbase until later, but that doesn't matter because there are plenty of other targets, and because the AI crippled its ability to counter-attack I'm free to take attacks of opportunity.

Reply #8 from Sins Forums Top

I next tried a TEC aggressor against a Vasari researcher.  This nearly became a TEC victory, but after seven hours it was still too level, as the TEC had only taken the Vasari roid...  what made the difference was that the Vasari reseracher stayed a level of fleet lower for too long, and having almost double the supply meant that the lobbers were affordable for the TEC.. also while the Vasari AI had 9 caps at one point the TEC had 2 very well-placed starbases.

So if you want a TEC AI I'd select the Aggressor... the others have lobbers as well, but without the extra fleet points early.

I might produce a full commentary and replay from the next matchup when I run it, if there's sufficient interest?

Ai notes...

The AI might use guidance as to capital selection.. the Vasari AI had 3 Marauders at one point..

Having observed that the Vasari don't use up supply on lobbers, the researcher promptly produced a Ruiner for every grav well it owned, perhaps too many...?

The AI improved the health of its rear starbases and not the forward ones....

When the lobbers appeared in groups of 2-3 they were quite effective, as they don't seem to be priority targets for starbases.. though they didn't prioritise the starbases either...

Neither AI piled up resource heaps even after seven hours, which was another surprise... so still no light on why this might happen.

The AI almost pulled off my 3 point defence trick on a roid before I could attempt it myself.. it was beautiful, three siege frigates were sent almost 270 degrees around, but they just found a gap before they hit the minefield... very near to absolute class.     

Reply #9 from Sins Forums Top

Very interesting, guys. Post this to the Diplomacy section, as the devs have promised to do some serious AI improvements in the expansion. 

Reply #10 from Sins Forums Top

This was on the small map posted in reply #4.  Vasari on the left, Advent on the right.  Normal start, I played it on normal speed to allow the AI more time.  Both AIs are Aggressors. 

The extractors seem to produce at the home planets before they have been built.  The three neutral gravwells produce a grand total of one crystal extractor.  The Egg and the Rapture are the first caps- 5 Skirmishers also for Vasari, 5 Disciples for Advent.. 

2:30 Both AIs have colonised the dead roid first?  Perhaps because the direct route to the enemy home is through the dead roid? Two Hostility labs for Advent, 1/1 for Vasari.

8:00 The Advent go for their roid, the Vasari opt to take out a desert with 2 heavies in its militia..

9:40 First siege produced.. an Advent Purge.  So far there have only been dead roids colonised.

11:00 Two Advent Disciples attack the Vasari dead roid, and kill a Migrator which fails to take any evasive action.  The Advent Aggressor AI builds two hangars at its home...? 

13:00 Advent have their roid, and its Purge Ships join the Disciples and kill the Vasari dead roid by 20:00.  The Vasari manage to destroy the Purge Ships, and when Advent return home they reoccupy the dead roid. 

15:00 The desert cost the Vasari 2 Skirmishers and almost its Egg.  Their Imperial lab has done no research at all so far.  Advent have a much easier desert.

21:00 The Egg is badly damaged, but the Vasari have a battleship as well and are researching supply for a third cap...?  The remaining Advent Purge Ships hover around their own dead roid.

25:00  The Vasari siege engage the pirates at the gas giant, where there are no extractors, and a pirate flak is destroyed.  The Advent siege decide to go past the militia at the volcanic to get to the Vasari desert.

27:00 A minor battle at the Advent dead roid between frigates, Advent are distracted from the Vasari siege at their roid, it is flattened and has to be recaptured.

34:00 The truculent Vasari siege then help to clear the volcanic.  Advent acquire a Revelation.  Labs Vasari 3 military 2 Imperial, Advent 2-2, both sides only have the first stage of fleet supply.

37:00 Advent has the ice.  The siege of both sides appear unescorted at the enemy roid and are destroyed.  Vasari colony ships are evident in numbers.

40:00 Egg vs Revelation at the volcanic, Vasari adds a starbase, the Revelation fights to the death and is destroyed. 

45:00 A Vasari siege attack gets the ice planet.  Vasari have their second supply stage.  Advent builds another Revelation.

50:00 Advent researching shield stages to Guardian, though they are outnumbered and with 100 unused of their lower supply stage.  Vasari have carriers, and a cap working each side of the map.

55:00 Another Egg vs Revelation battle this time at the Advent roid, same outcome, Advent lose their roid.

1:02:00 The Rapture plus support vs the Kortul at the ice... the Kortul reached 1000 health, then boldly turned its back and fled.  Vasari are using 395 supply to 235 Advent..

1:07:00 Nine siege at the Advent home, with flak, but all the siege are destroyed.  Advent builds a Radiance. 

1:12:00 The Rapture returns to the dead roid with its fleet, Advent want to attack over the sun...  Vasari building more siege- from their home though.. and a Marauder, the AI loves its Marauders.. 

1:18:00 Advent returns home, but Vasari have 3 siege there and the flak has reduced the hangar based defence strikecraft.  Enter... the Egg, it destroys its 3rd cap of the battle, the Radiance, and the Marauder joins up to make it 3 caps to 1.  Siege are expensive meat.... still Advent has a 5-2 carrier advantage plus hangars.   However the bombers prove ineffective on the Kortul and the Rapture is destroyed- the Egg becomes level 6...   yet another Revelation appears.   Still no forward frigate factory for Vasari, they even build from home rather than the nearer desert... 

1:27:00 The Revelation is destroyed. only 2 Drone Hosts left, 4 Vasari siege appear..   

1:30:00 A Halcyon this time..  but there are 5 siege, the 2 Vasari carriers and 5 flak are still there and all the Drone Hosts have been destroyed... as is the Halcyon.  Advent really missed the money and supply they have tied up in the 5 inactive purge ships...

1:34:00 The home is gone... the end takes another hour but this is it really.  Vasari had a credit advantage only from 1:06 it was 17-17 then and 20-14 at the end.  Metal 3-2 and crystal 2-2 then though it ended 5-2 3-2.  About 70 frigates destroyed each.. however Vasari destroyed 6 capitals to none...  Neither side had more than 6 lf at any stage, max 7 illums 9 kanraks...  10 siege and 5 siege... no Guardians were built and Vasari had 9 flak max to 2... Advent had 5 carriers at one stage but they were all destroyed, whereas the 2 Transporters survived.  The homing mine bug didn't occur though. 

Research: combat, defence, civil Vasari 4-2-10(!) Advent  8-4-5... fleet research Vasari 6 Advent 4... at the start of the 1:07:00 battle Vasari had a 100 supply advantage, just enough to absorb the loss of their vast siege train...