Sneak peek at the AI changes

I didn’t realize how much I missed working on AI until I started working on it again. So far I think I have been able to make a vast improvement in how the AI plays. I want to thank everyone that posted feedback, it was very helpful. I will have to check the list again, but I am fairly certain I have fixed or at least improved upon all of the relevant issues. The change list for the AI so far is longer than any version of the AI I released for SupCom 1 or FA. Before I post highlights there are some things you need to know.

First, you need to realize that some of these fixes may not make it in right away, if at all. They still have to be tested to make sure I am not epically breaking anything. Second, I still do not know when these fixes are going to be patched it, so don’t get upset if they aren’t in the next patch, or even the one after that. There are a lot of fixes spanning many files, so in the interest of keeping patch sizes down it may be delayed until we can include it without killing everyone’s bandwidth.

Probably the biggest bug was the fact that threat conditions were not being used properly. Threat conditions are what we use to tell the AI to do something in reaction to a threat, such as building AA when an enemy has air units. Not only is this fixed, but the AI is using the threat value of its currently targeted enemy to determine how big of a platoon to send and how many experimentals it needs to have built before sending them out.

The AI also takes into account build restrictions when choosing which archetype to use. It also recognizes when it has all research, so it won’t waste mass on research stations when it does not have to. Also, the AI should no longer get stuck if you rush it and destroy something while it was in the middle of its starting build. If you restrict a unit type in the game the AI now acts like that unit doesn’t even exist in the game. Don’t want nukes in the game? The AI won’t even have the option of giving one of its engineers the order to build one.

Most of Thursday, Friday, and Monday the 22nd was spent fixing little bugs like the above. Tuesday and Wednesday were spent balancing and getting the AI focused. The biggest challenge was the AI’s economy.

The AI uses a bucket system for its economy. When the AI gets resources it puts a set percentage in each of its buckets (defense, expansion, experimentals, land units, etc) and it puts any excess resources in a slush bucket. When the AI wants to build something, it checks the appropriate bucket to see if the resources are available. For immediate items (like emergency shields or economy related stuff) we tell the AI to use from all the buckets. The problem with that is the AI ends up taking from the same buckets over and over. Later in the game, when expansion is harder, the expansion buckets ends up getting filled and never used, so the AI ends up in a situation where it has a lot of mass, but can’t build.

My solution was to make the AI take resources from the slush bucket first when taking from all buckets. Then, I sort the buckets from fullest to emptiest and take an amount from each bucket based on how full it is until we have enough or run out of buckets. If the first pass doesn’t get enough we re-sort and take all the resources from each bucket until we have enough. This way, full buckets that aren’t being used much are the first to get resources taken away and the AI has a much smoother economy.

I also worked to get the AI to focus a lot more. The AI would constantly send units, that were doing hefty amounts of damage to an enemy base, across the map to kill and few units that had the audacity to get near of its expansion bases. It also didn’t focus its attacks on its targeted enemy very well either. It took a lot of trial and error, but I think I have it dialed in a lot better.

Another nice upgrade is that the AI will now use the Mass Convertor. Prior to this the AI would never build one. Now, the AI will build one (it only needs one) or, in the case of Cybran, use the ability of one of its power generators.

I am still working out the cheating situation. Some people complain that the Hard AI cheats too much and the Cheating AI cheats way too much. Currently, I have asked a few people in the office of different skill levels to try out the new changes and see what they think.

Now, as with the last post, I have a request. I would like some ideas on research paths. I plan on going over them again and trying to optimize them and wanted ideas on what paths players use. I don’t need an exact path, per se, because I have a function I use in the AI that finds the cheapest path to get to a particular research. What I need is what research item you go for first, second, third, etc. If you are going air, do you get gunships first? Shields? Or, do you go for the experimentals? Let me know.

For those of you who want a tl;dr list of the AI fixes, here you go. Bear in mind, this list is not exhaustive, so don’t get upset if you pet fix is not listed. This is also not a guarantee that they will show up in any patch, let alone a patch that is released soon.

  • Fixed a bug where threat build conditions were not being used.
  • The AI will hold back units until it thinks it has enough to make a push.
  • Fixed an issue where the AI would constantly pull units back to attack an expansion base instead of pushing forward.
  • Added AI response to make a land AI get land unit AA upgrades if the enemy has air units.
  • The AI will no longer get stuck in its startup build sequence if rushed or if there are build restrictions in place.
  • The AI will no longer give every structure blip AA and Land threat; instead it will wait until it has seen the unit first.
  • The AI will no longer give blips on the water Naval threat, it will wait until it sees an actual ship.
  • If Research is disabled the AI will not build research facilities.
  • If a unit type is restricted the AI will act as though the unit does not exist in the game.
  • Platoons will try to only attack areas where they think they can do enough damage before dying.
  • The AI should no longer attempt to send attack platoons to unpathable areas.
  • The AI will choose an appropriate archetype based on unit restrictions.
  • Re-balanced the way the AI chooses an archetype.
  • The AI will react to enemy TML, artillery, or Experimentals by building more shields and placing them closer together.
  • Enemy shields are now given AntiLand and AntiAir threat.
  • AI will build and use Mass Convertors.
  • If the closest enemy to the AI is a lot weaker than the other enemies the AI will focus on the weak enemy.
  • You can now select which AI type you would like to play against from the dropdown in the game lobby (random is still an option).
  • AI will build factory shield upgrades more often, not just when the enemy has air.
  • Factory upgrades now have a threat value.
  • AI will not build research stations if it already has everything researched.
  • AI will now group units based on their target enemy's threat.
  • AI will now group experimentals based on their target enemy's threat. 
The last few changes are in for this round, unless something pops up in testing.
  • The AI is now aware of the victory condition and will focus attacks on the ACU more in assassination mode.
  • When an AI unit dies it will add threat to the instigator's location to warn other units.
  • AI will respond when an enemy turtles.
  • AI will now recognize when it is being attacked by nukes or artillery and respond.


      Christian said... / March 25, 2010 at 2:46 PM  

      this is just great! Great news and a great feeling to know, that its still being worked on supcom2

      Christian said... / March 25, 2010 at 3:02 PM  

      I forgot to post my ideas about the research tree.

      1. ACU Upgrades are too strong, therefore make them more expensiv (add 2-3 points on the way to overcharge for example, and 3-4 points to the highest Training level)

      2. Transports: In order to not change the balance values of transports, make them available later. Add 4-5 research points for transports, and reduce the costs of following research steps (Like experimental transporters)
      (just saw was adressed in 1.4 also, but still i would do that change, they probably still are too strong in 1v1)

      3. Reduce anti missile research.

      4. Increase Mass fab Research costs (just saw that this was already adressed a little bit in 1.4)

      5. Experimentals are useless. less research points for all Mobile Experimentals (not the buildings, they are strong), but regarding the weakness of Cybranzilla the 15 points are ridonkulous... but maybe its not only research here, its just the units themselves (Fatboy, Urchinow, Darkenoid, SoulRipper, Flying Fortress are good imo, megalith is medium, but the huge exps like King Kryptor, UC and Zilla are useless crap :P)

      Kryo (so that you know who I am)

      torben said... / March 25, 2010 at 3:56 PM  

      I think you misunderstood, Sorian is asking for research strategies for his AI to use, not for balance changes :)

      Fingolfin said... / March 25, 2010 at 4:15 PM  

      Bug ticket #4562
      'Units need more "je ne sais quoi" and 5.2% less Mojo.'


      æsir said... / March 25, 2010 at 4:58 PM  

      "If a unit type is restricted the AI will act as though the unit does not exist in the game."

      Cybran silo has TML. In No-Nukes will a.i. build it and use it for TML (and assume the nuclear missle doesn't exist, and bypass that use of the silo) or will it just not build the silo at all, losing out on TML capability?

      The larger question is about any exclusion items that serve multiple purposes. Whether or not the entire unit/structure is out, or just the sub-component.

      Kreyson said... / March 25, 2010 at 5:01 PM  

      Great list of AI changes Sorian!

      Can you look into reducing the ACU death nuke on supremacy game mode?

      Currently, if the ACU dies, it usually destroys enough of the base to knock the AI out of the game which defeats the purpose of Supremacy game mode (turns the game into assassination instead of supremacy).

      Or at least, make the ACU death nuke a game option during set up?



      Sorian said... / March 25, 2010 at 5:19 PM  

      As Cybran, if you select No Nukes than no one can build

      So, no TMLs.

      George said... / March 25, 2010 at 5:50 PM  

      Fighters and Guns ships if the AI goes air.

      Tanks > AA > Artillery > Missile > Shields/Bodaboom > Anti-Missile

      If the enemy has a lot of stationary defences the missiles should increase in number.

      Cybran should rush for Anti-Air Bot and the Megalith.
      If there is water Cybran should basically go crazy with navy especially Battleships.

      UEF should rush for the Fat-Boy or the Gunship Experimental if going air.

      Illuminate should rush for the Urchinow and use a lot of assault bots if there is no hover capability on the map.

      It would be nice to see the AI actually use the Experimental Transports at full capacity.

      Bastilean out

      Nainara said... / March 25, 2010 at 6:00 PM  

      I'd pick an AI that knows how to rush and mass cybran battleships or gunships over a human ally any day!

      Unfortunately, I imagine that the tactics involved in a good team ACU rush would probably be a bit complex for an AI to handle.

      David said... / March 25, 2010 at 7:00 PM  

      Tech Path could work better if they were grouped by efficiency upgrades vs military upgrades. I always upgrade structure costs first to make things cheap, then research/mass/energy/converter bonuses. Once I have everything as efficient as possible I branch out to military. I totally don't understand logically why you would need a shield to have mass conversion other than to insert some arbitrary stall tactic in getting them. Same with radar being needed for nanite repair and power shunt. Basing things on previous research should have some logical reason, not just to prolong the time it takes to get them. Overcharge is the best example of this done right, rate of fire and tactical missiles can be rationalized as to why it is a prereq for overcharge. Hunker-> Speed increase?? makes no sense.

      Definitely need pause in multi player games.

      Peter said... / March 25, 2010 at 7:07 PM  

      One thing I would mention about the AI is the, somewhat puzzling, Transport spam. In many skirmishes, I've launched a late, very large attack on a Hard AI, to find a cloud of transports in formation over his base, doing nothing. Slightly reminiscent of a field of WW2 barrage balloons.

      As for research... I always play UEF at the moment. On start, I usually go for things like the structure cost reduction, and the ACU health and regen upgrades.

      If it's a close-quarters map, I may buff my ACU early on, increasing damage, range, and getting anti-air, but usually I'll leave that until later, going instead for the important land upgrades (125% regen, range, mobile artillery). Once I have those, shields are usually on the cards, as well as the economy upgrades. (I may get structural or mobile shield generators earlier when playing against Cheating AI, as those games usually start with some mega-turtling on my part.)

      Once I've got a few more research stations going, I'll get Land Training 1, maybe 2, and also Light Artillery with Hardening (for base defence), and some light experimentals, like the Fatboy or AC1000.

      Next up is usually Stacked Barrels, Shields and Anti-Air for Land, followed by the King Kryptor (I should have a couple of Land Gantries cycling Fatboys by then, so they've gained a couple of veterancy levels, making the KK cheaper to produce. Then Mass Converters and Engineering Towers.

      As for research I tend to avoid (until I have nothing else to spend the points on, anyway); Air Bomb upgrades (I rarely use bombers at all) ACU Jump Jets (if I need them to escape, my base defences obviously aren't doing their job), and ACU Overcharge (I don't really use my ACU offensively very often). I haven't mentioned any Naval research; I haven't really played many games with Naval involved, so haven't really used it all that much.

      Just realised how long this comment is. I'm hoping you wannted a detailed response! Looking forward to the AI upgrades!

      S E said... / March 25, 2010 at 8:15 PM  

      I guess if im going land and its a small map then i'll go "experimental rush": for cybran, i'll get the C. Rex through the megalith (more expensive that way but you get 2 experimental units to build and a minimum cost to the third), and build a lot of those.

      If im going cybran air, i'll rush to the soul ripper (seeing a pattern here?) through the gunships, building about 20 regular gunships and some F/Bs before the SRII.

      If i go for structure, i dive for mass conversion and the proto-brain thenthe heavy artillery and just fill out the rest of the tree.

      With any experimental rush item, as soon as i unlock the experimental(s) in question i usually then research Training I-V as those boost the experimentals power.

      DeadMG said... / March 26, 2010 at 4:39 AM  

      UEF ACU:
      ADPS Artillery, Overcharge, the +50% Health, JumpJets, then training.

      Cybran ACU:
      Nanogun, Overcharge, JJ, training.

      Overcharge, Teleport, Training.

      UEF Air:
      Broadsword, the whole buildcost/regen/buildtime path, health/shield, then training. Skip nearly all of the bomber tree.
      Cybran air:
      Head straight for gunships and the damage/rof upgrades, then shields, then training.
      Aeon air:
      Pray you don't lose.

      UEF land: Grab range first, then pers shield, then head to AA. Then, spam trainings.
      Cybran: Mostly the same as UEF - speed instead of range.
      Pray you don't lose.

      UEF/Cybran naval:
      Spam battleships with range/damage as appropriate.
      Pray you don't lose.

      Noahetrex said... / March 26, 2010 at 5:37 AM  

      "Enemy shields are now given AntiLand and AntiAir threat."

      Will the AI spam navel if shields get spammed? Don't you think "AntiNavel" is important for shields as well?

      sircrunch2000 said... / March 26, 2010 at 6:25 AM  

      "I am still working out the cheating situation. Some people complain that the Hard AI cheats too much and the Cheating AI cheats way too much"

      I hope you're referring to both PC and console players and not just console players.
      The Hard AI provides no comp. The cheating AI is enough to make u work a bit at the beginning, but after you're set, it's just a matter of cleaning up.

      The changes you've mentioned in your blog are a good step in maybing making the cheating AI a tough opponent and I look forward to the changes..
      As a previous poster mentioned, it's not uncommon to charge into the AIs base and see a huge flotilla of transports or a sea of air factories.

      Anyway, I only ask that you do not nerf the cheating AI and maybe even kick it up a notch.

      teeman said... / March 26, 2010 at 6:33 AM  

      Quick question, do you know if these changes will be making it to the Xbox version?

      Sorian said... / March 26, 2010 at 6:45 AM  

      My apologies for posting AntiLand, it is actually AntiSurface threat which covers Naval as well.

      The cheating situation is still being tested. So far the consensus appears to be: no change necessary for the Hard AI. We will see. We are planning some major testing days soon.

      I do not know if these changes will make it to the XBox360 version or not.

      teeman said... / March 26, 2010 at 7:02 AM  

      Thanks for the response, I really appreciate it. You wouldn't happen to know if there is going to be any patches in general for the 360? Thanks,

      Kreyson said... / March 26, 2010 at 9:10 AM  

      Not talking about ICBMs, referring to the explosion when the ACU dies. The damage is too great and defeats the purpose of Supremacy game setting.

      Keep up the great work sorian.

      Wind said... / March 26, 2010 at 9:31 AM  

      I wonder if you fixed the bug that the AI will upgrade factories with the same upgrade twice, ended up the extra upgrade glitch on the center of factory (obvious on air factory).

      For research path, on bigger maps I suggest researching structure upgrade first. For Cybran, get the structure time and structure cost reduction research first. For smaller maps, maybe let the AI focus more on upgrading ACU.

      Let the AI upgrade the unit type they focus (land, air, naval) before researching Experimentals. Major experimental units are too weak now, don't let the AI rush to research them and don't let the AI send them before having maybe 5 or more.

      I also really hope to see AI use naval units.

      Thanks for the hard work Soria, can't wait to play with the new AI.

      Sorian said... / March 26, 2010 at 9:37 AM  

      I still have not been able to reproduce the factory upgrade bug, but by the time I heard about the bug I had already made a large number of changes, so it may be fixed. We have QA for the AI coming up, so if it still exists it will show up there.

      df said... / March 26, 2010 at 3:00 PM  

      screw patch size, its worth it

      foxinabox said... / March 26, 2010 at 3:10 PM  

      Glad to see you working on SupCom 2's AI :) It's just too easy... me and my friend just beat 5 Cheating AI's without any problems. It was hell lots of fun to play against your AI in FA though, quite a challenge, one of your Cheating AI's was enough to keep us fighting for hours... and then get wasted :)

      Brian said... / March 26, 2010 at 4:38 PM  

      My complaint with cheat settings were that they were giving the AI build speed boost but nothing for resources. This ended up with the cheat AI being super hard early in the game, but late game AI can build so much so fast, yet is only having mass income of a normal build speed player.

      Adding in mass and energy cheats I noticed a huge improvement in my AI, even with a drastic reduction in build speed cheating. This was after I fixed the AI running itself out of energy, so the improvement was not just from that.

      One of the first things I saw needing tweaking was the research paths, so I'm super glad to see you are giving them love. Some general thoughts I have on these for the AI are..

      1 - Structure build cost and build time first always.

      2 - ACU health upgrades very early. Either training 1 or the health upgrade. Reason for this is that the AI ACU is very vulnerable to early game gunship or even bomber attacks. I ran several tests of Air AI vs land AI, the air AI nearly always would kill the land AI ACU very early. Forcing an early game acu health increase made a big difference. Of course so would building anti air earlier, but there are other benefits to early acu health boost.

      3 - For land/air AI, pick either experimental or non experimental path and stick with that. If going experimental, get to training 5 for it quickly. I would just create multiple research paths with them being experimental or non experimental focused. Of course once the experimental+training 5 is done, it can do other things...

      4 - Ignore most major experimentals, King Kriptor, Cybranasaurusrex etc, do not research, do not build, they are worthless for the AI to use. Maybe a good player can find value in them, the AI can't at this point.

      5 - Unless it is a Rush AI, it should be somewhat well rounded. For a land AI, it should get various structure/acu upgrades along the way, especially to prevent ACU snipe. If it is a rush AI, then we take the gamble that focusing most of our power into attack will pay off, so we don't need to be as well rounded and can put off much.

      6 - Specific to the land focus AIs. We should put some research into air units to improve transports. Not too early in the list, but getting air unit training, and the air unit shields can help the transports out a bit.

      7 - Maybe the biggest one...Since you got the mass converters working, the AI should make these a huge priority for research. Just make sure he has enough energy! This of course will change if the mass converters get nerfed like they probably need...

      8 - Add into the Events something like. When air experimental spotted or baseairthreat greater than like 40-60, research ACU health upgrades and anti air land units if we don't have those yet. AI already researches shields when it spots experimental, but maybe also research shields if air threat if above whatever.

      - Machater

      Sorian said... / March 26, 2010 at 8:42 PM  

      The Cheating AI gets a resource bonus, that is what the handicap does. It just needs a higher resource bonus than build bonus.

      Jerry said... / March 27, 2010 at 1:34 AM  

      Regarding difficulty levels:

      I still think having cheat options (like your old FA cheat options mod) for the cheating AI is the best and easiest solution to diversify the levels of difficulty in skirmish.

      Because no matter how you setup the difficulty in hard and cheating, there's bound to be people who think it's too easy or too hard (bear in mind it's not just about player strength, it also depends on the map and team setup and game rule), I think it's just impossible to make one or two 'fit-all' settings.

      Or is this beyond the scope of things you could officially change/add?

      MeDDish said... / March 27, 2010 at 7:39 AM  

      id suggest having a 'normal' and a cheating AI for each difficulty level
      SO a Easy is easy
      A cheating Easy is a bit harder but easier then normal etc etc

      Also i wouldn't put pressure on the AI to build EXP until its within 50 units of the unit cap as currently 1000 mass worth of tanks is better then a exp!

      structure EXP should also build a few, spaced out around/behind the base depending on what they are (mag around, proto brain behind, loyalty gun in front)

      as more balance patches come out im sure there will be changes to the AI and its 'preference' between one item and another i understand currently its all about the ground work :)

      Sorian said... / March 27, 2010 at 8:18 AM  

      The last of the AI changes are in for this round. I didn't get to do everything I wanted, but there is some definite improvement and some more things for the modders to play with.

      Updating the blog post with last few things.

      Kreyson said... / March 27, 2010 at 11:20 AM  

      Greatly appreciate your hard work on the AI. Looks like these changes will give Turtlers a great challenge. My friends and I love to Turtle co-op against the AI.

      Will be looking forward to it.

      HDionS said... / March 27, 2010 at 7:40 PM  

      didnt see anything about AI playing in a team.
      Me VS 2 AI.
      1st goes rush, keeping me at bay, 2nd goes for mass convertor and then spam a lot of regular units, or building 3-5 experimental factories. Think its pretty simple, but will make the game way harder(in 4v4 when the battle ends, I can count up to 35 air factories that I have perma spamming units) :)
      Big maps:
      AI should go for mass convertor + air defence and territory control
      AI team: should use different stretegies (like cybran walking boats etc)
      AI air: > gunship > dmg upgrades > health > 1-5 vets
      AI sea(cybran): > looking and radar range > shooting range > legs > battleship (mass conv. mb before all this if map is big, but cybran navy rush is really unexpected, and poweful)

      thnx for hard work

      mocafrost said... / March 27, 2010 at 9:57 PM  

      Btw, when the AI uses the escape pod it just stays put and eventually tries to rebuild in that spot. Whatever killed the body is still right there to finish it off. I'd like to see the head find a safe spot to rebuild, or the research points spent elsewhere.

      Jeff Shyluk said... / March 27, 2010 at 10:56 PM  

      To briefly answer the question:

      I focus on training first, usually air training. I normally build an air base first, for defense and for scouts.

      After I've built some research stations, then I will look at stuff like adding speed, radar and visual range, again to help the scouts.

      By that time, I have to deal with some kind of attack, so I will focus on training structures and build defenses. If the attack succeeds, game over, but if not, then I have a lot of successful scouts. Time to plot an attack, and I will then use research to buff whatever I think will make the best attack.

      If my attack succeeds, game over, if not, then reserach more defenses. Wash, rinse, repeat.

      I am not a great player, so this is the advice from someone who only plays casually. If you want to simulate Casual AI (Turtle), there you go.

      olivier said... / March 27, 2010 at 11:54 PM  

      Sorian, great for modders, but as far as i can see, modding is still disabled on the full game and modders seems to lose faith and patience ... any news regarding gpg plans to finally enable modding in the full game ?

      mattatsu said... / March 28, 2010 at 7:59 AM  

      and i wondered why my AI suddenly seemed to live up to it's name ... nice work.
      Btw. is it an act of desperation or why does the UEF AI goes beserk on my base with only it's ACU when nearly defeated? Did happen 3 times on Clarke Training Center.

      I also noticed that LandAI sometimes doesn't bother to build AA-addons. it's really easy to rush with just Air. Even after 5 mins there was nothing but tanks.

      And for those craving for a real challenge there could be an AI with halved(or more decreased) build-time or full exp-units/structures. How about that :)?

      Sorian said... / March 28, 2010 at 12:45 PM  

      Current PC AI buffs (subject to change):
      * denotes change

      Resource: -30%
      Build: -25%

      Radar: +75%

      Resource: +65%*
      Build: +50%
      Radar: +150%
      Vision: +100%
      Vet Level: +25%

      Resource: +100%
      Build: +90%*
      Vet Level: +100%

      MeDDish said... / March 29, 2010 at 4:48 AM  

      so this ready to roll? ready to patch?
      also we know when it might be out?

      Sorian said... / March 29, 2010 at 9:58 AM  

      Still needs a lot of testing.

      No idea when it will released.

      æsir said... / March 30, 2010 at 6:54 AM  

      As a devoted skirmisher, I see that every game the AI will out-produce me on mass/energy. I usually tend to get Research humming along better though.

      What does this mean? The suggestions for the AI to get structure cost/time bonuses at the start are great suggestions for humans, but not for the AI.

      AI is more efficient than I at expanding and claiming mass deposits. I feel the AI should, for this reason, depart from a commonly used human strategy (early game structure efficiencies) and move towards developing research based upgrades for Land/Air/Sea/ACU and structure upgrades more focused on defenses instead of build efficiency.

      I base my observations/opinions on playing against Normal, Normal (Turtle), Hard, and Hard (Turtle) AIs.

      Also, I don't like games where I am clicking to build new anti-nuke amunition 4 to 6 times per minute. I don't know what to do about this - I want to play without exclusion rules but I don't want to keep having to click a build command so frequently. I suppose if ammunition production were automated, I'd hate this less, but I think that would upset people for other reasons.

      Why do the AI spend so much on Nukes, when they could spend those resources putting together a combined-arms package that does better than the usual 21 unit single-unit type formation? Humans do this in part because it's easier to manage building spam of one unit. But AI? It doesn't suffer from carpal-tunnel syndrome so should be able to handle the extra clicks used to produce efficient attack packages besides nuke spam.

      Brian said... / March 30, 2010 at 3:14 PM  

      Is it just the turtle AI building the nuke spam? The turtle AI gets a special resource bucket increase for Nukes, 20% IIRC.

      Its quite tweakable but with exclusions, is it worth taking time to mess with? Maybe

      Sorian said... / March 30, 2010 at 11:29 PM  

      The AI is not so nuke happy anymore.

      Wojtek said... / March 31, 2010 at 1:31 AM  

      Good to see you working on the AI Sorian, any chance of the latest changes making it into patch 1.05?

      Also I'm quite surprised to see the AI on Hard and even Normal cheating? I'm pretty sure the AI in SC/FA did not cheat at all unless it was actually a cheat AI, how come the SC2 Hard AI has such big boosts?


      Sorian said... / March 31, 2010 at 2:06 AM  

      Normal AI only gets a Radar boost. Hard and Cheat are the only ones that get a resource/build bonus.

      Azzer said... / March 31, 2010 at 6:53 AM  

      Is the issue with their limited "build radius" for bases, and poor base-building-templates, on the cards for fixes?

      I posted a number of things up here, some you have already listed as fixed/to be fixed... but some you haven't;

      The main issue that I didn't spot you comment on at all is the issue with how far out of their start area they are allowed to build structures. It seems to be a severely limited radius, and they do it in a disorganised manner, so they limit themselves with what they can construct - as for the turtle AI, their turret building seems to be utterly redundant (a random turret here or there - how about giving them the ability to "compress" a group of turrets in blocks if they are a turtle AI, building huge waves of nasty defences as part of their turtle strategy).

      Also, like others have commented, I've found AI's often building large armies of transports that go unused (or only using 1 or 2 at a time even though they have a large amount).

      Brian said... / March 31, 2010 at 10:43 AM  

      transports have a limit of only 5 at any given time, the problem is it only counts transports that are finished building. What happens, usually later in a game, is the AI will queue up 40 transports before 5 are finished building, bypassing its limit. Transports have a higher priority than other air units for being built, so if the AI has the resources he will go crazy building transports until 5 are actually finished being built. This problem actually extends well beyond transports to any other unit with quantity limits, like engineers being spammed.

      I actually wonder if the reason the cheat AI gets such a huge build speed increase is in order to alleviate this problem to some degree. I have played with the cheat settings extensively, I really hate the AI having such a large build speed increase, but I do think it should have a nice resource cheat. The problem is that whenever I decrease the build speed, the AI really goes extra insane with engineers and transports.

      We need either a limit on the AI creating factory queues, or we need units in queue to count towards limits

      Brian said... / March 31, 2010 at 11:01 AM  

      As far as base building goes, the issue with a lot of items is that they all try to build near the centerpoint of the base. This is fine for most things, but not for defenses.

      The base radius isn't a huge issue, we could use a bigger radius on some of the bigger maps, the bigger issue is the ai and expansion bases. I suspect a huge issue here is with map markers, but GPG has made a ton of stuff unable to be modded so I have no way to test my theory there. Not sure why GPG has moved so much AI functionality out of the realm of what we can mod, but its horrible....
      The other issue with expansion is when an engineer creates a base and dies, but the AI doesn't send a new engineer to take over.

      The idea is to have the AI make many expansion bases, including ones that are very close to the main base, so close that most people would consider them part of the main base, but they are built as expansion bases in order to better utilize turret defenses and many other advantages. Normally a map has 4 mass spots close together and then another 2 that are also quite close. Well there should be the main base around the 4 mass points, and those other 2 spots should each be an expansion base.


      Sorian said... / March 31, 2010 at 9:11 PM  

      There are markers denoting where the ai can expand to, but it's resource issue were hindering it. The reason the ai exists more in code this time around is performance.

      The base radius thing was actually something quite simple and is already fixed. We had some multiplayer ai tests today that went quite well.

      I will look at the queueing issues, but I have not seen any of the issues you mentioned, but will look into it anyway.

      Jane said... / April 1, 2010 at 7:51 AM  

      On the topic of nukes, why just build and launch all the time? if 1 didnt work, get the AI to build 2, then dual launch 2 silo's?
      its a AI, if i can scout, and sees u have 2 anti nuke silo's why doesnt it spend those extra resources on more nuke silo's then u have anti nuke silo's?
      i know if a AI launches 5 nukes all together @ me, i know im in trouble!

      Sorian said... / April 1, 2010 at 9:41 AM  

      @Jane: Actually, it is supposed to, however, I think it runs into the same problem as the FA AI; it counts the number of anti nukes around the target area, not the number that are between the launch location and the target. As in FA, the anti-nukes in SupCom2 will target nukes that fly over it.

      It is on my list to fix.

      Sorian said... / April 1, 2010 at 2:23 PM  

      @Brian: I looked into the transport spam issue, and found that we do, in fact, check for units being built. However, I have a suspicion that the problem is build orders that are made on the same tick (like all the AIs air factories get the order to build a transport all at the same exact time). So far, I have not seen this issue crop up anymore.

      The issue with the engineer spam was different and is also fixed.

      MeDDish said... / April 2, 2010 at 10:49 AM  

      @Sorin MeDDish/Jane here, signed into the wifes account by mistake last time :(

      if a player puts up Anti-Nuke in a forward base, the AI shouldnt be trying to nuke my 'home base' it should be going for the closest base/anti nuke to it!

      unless u have fed the AI a proto brain or 50 and it is going for my pgen farm?!?!?

      Brian said... / April 2, 2010 at 12:50 PM  

      Seems like the changelog you have for the AI is actually double what it appears to be! Awesome work.

      The big problem I am having is that I can't seem to get the AI to do my laundry for me...

      On a more serious note, with build queues...I have run into it several times with the AI having huge amounts of mass in 1 factory queue(when i self destruct the factory his mass skyrockets). Is there something I can investigate on this issue? Any type of limit for how many items or resources can be queued into 1 factory(or i guess engineer possibly)??

      Are you able to elaborate on the what the engineer problem was and how you fixed it?

      When I test the AI i run the game on +10 sim speed, is this safe to do? Are there any problems that could arise from running at that sim speed that would not happen at normal game speed?

      If only I could save my skirmish games to use for bug reports....


      Sorian said... / April 2, 2010 at 3:00 PM  

      @Brian: Running the game at +10 should not have any issues.

      Not sure why you are seeing queued units. The AI does not queue factory commands unless the bp calls for more than 1 unit (and none do, that I know of). Again, this may simply be fixed on my end already.

      Brian said... / April 2, 2010 at 4:27 PM  

      strange, i thought there was no factory queue at first too.

      There is one unit that actually does build more than one, the artillery units build 3. Not sure if that is intentional or not. Not really an issue anyways.

      When i SetFocusArmy to the AI, I don't actually see any units queued for the factories, but when i force self-destruct them, suddenly the AIs bank shows the extra mass, which can be huge amounts(like 20k).

      This is what led me to believe the transport/engineer spam was due to it queuing the factories up before a the first units were done being built, bypassing the limits.

      I guess we will see once all your awesome new fixes come out in a patch.
      Thanks for the hard work!

      Sorian said... / April 3, 2010 at 9:01 AM  

      I managed to get some more stuff in. Can't wait for it all to be released.

      jbencevet said... / April 5, 2010 at 9:48 AM  

      Please hurry! Oh how I love AI skirmish battles, but the AI at the moment is no challenge, for the main reasons that it doesn't seem to react to what I'm doing on a strategic level, and it keeps doing the same repeating actions over and over, in vain. However your changelog seems too suggest this might be about to change significantly?

      Murazor said... / April 5, 2010 at 4:36 PM  

      Hi Sorian,

      First thanks for a great game.

      Can you tell me why on earth you removed the save-load option in skirmish mode for SC2 ? What is the logic reasoning in removing something which is basic stuff in all stragetic games ?

      Like many strategy players I enjoy battles on large maps with many AIs lasting for many hours. This means that I VERY much like the option to save my games for various reasons:

      1)If only limited time is available, I can save and continue playing the game at a later stage

      2)If I make some wrong strategic choices, I can reload the game at an earlier stage instead of restarting the entire game

      3)If the game suddenly crashes (like I experienced a few times in the Campagn mode) I can still relaod the game at last save stage.


      Brian said... / April 6, 2010 at 6:30 PM  

      Save wasn't removed from skirmish as much as it was never finished.

      The campaign AI is totally different from skirmish which is why there are save games in campaign. There are other differences from campaign to skirmish which affect this.

      In the end I'm assuming it came down to GPG not having enough time/money to get it in.

      A while back I modded in the save/load to skirmish, and it worked fine, except it would crash everytime the game had AI in it. So I can confirm that GPG is not just blocking this feature from us.


      Post a Comment