Page 2 of 45 FirstFirst 12345678910111227 ... LastLast
Results 21 to 40 of 912

Thread: ReallyBadAI Battle System +Hardcore + Settlements v5.7 >Oct 17< ###quality taken to next level###

Hybrid View

Previous Post Previous Post   Next Post Next Post
  1. #1
    Gorrrrrn's Avatar Citizen
    Join Date
    Jan 2007
    Location
    here
    Posts
    5,546

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    Have DL'd the SS radical version to see how that plays.

    Had a quick look at the .xml code -can't say that much of it makes a lot of sense at first glance.
    (that;s just me not undertanding it.)

    If anyone wants to do some comparative testing I suggest we decide on a few agreed custom battle scenarios.

    Would need to include crossing, town siege, 3-tier castle siege, open field and difficult terrain scenarios.

    (Either vh or med difficulty as I expect those are the most likely options people play.)

    Both with RR/RC and without.
    (with v.72 rad or not rad as desired.)

    Full (20 unit) and part (say 10 unit) armies. Probably easier to try just 1v1 at this stage.

    (Human and AI play both sides, x3 each if you can manage it.)

    ---------------

    On a different angle should we be looking at the battle formations as well as part of this, they determine how the AI and human armies are laid out.

    (for example it would be nice to have marching columns have proper light cavalry screens and heavy infantry on the inside, rather than single file.)
    Last edited by Gorrrrrn; April 24, 2009 at 11:14 AM. Reason: spelling etc.

  2. #2
    Germanicu5's Avatar Will buy spare time...
    Join Date
    Feb 2009
    Location
    Not Zee Germany
    Posts
    2,101

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    @Rozanov
    A quick thought for start, 1vs1 unit testing doesn't work properly for testing AI behaviour, coz this one unit turns into general and is too tempted to go into melee if it's a skirmishing unit. So if I want to test something like skirmish cav reaction etc. I use 1 additional foot unit on each side, these 2 units fight while you chase after skirmish cav with my heavy cav or the other way round.
    I most often use such testing setup for open battles (just enough units to see proper maneuvering or lack of it, I put my troops in a "horseshoe" formation most of time, due to harder implementation of outflanking):
    Map - grass plain (flat); Difficulty - VH
    No cav
    AI - 3+3+2 heavy melee infantry (foot men at arms plus order foot soldiers, one unit is a general) + 3 halberd men at arms + 1-2 missile units
    Human - 4 pikemen, 2 urban spear militia or such, foot men at arms as general 1-2 missile units
    Cav
    AI - 3+3 heavy melee inf (one is a general) 3 halberd men at arms 2 heavy cav 1-2 missile units (or you can just add a proper bodyguard to that)
    Human - 4 pike, 2 spear, heavy cav, 1-2 missile units
    Last edited by Germanicu5; April 24, 2009 at 11:47 AM.
    I have no memory of this place.

  3. #3
    MiiKLL's Avatar Semisalis
    Join Date
    May 2006
    Location
    Spokane, Washington
    Posts
    451

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    It is a proven thing that the AI can be influenced and improved. Examples of this abound...Darth's...Lusted's...Condo's...GrandViz's...GrandDukeVytas's...Taiji...Xeryx's and Germanicu5's current attempts.
    I know this because...I have played every one of these extensively. Each one is able to make the BAI play and perform better by trying NEW and DIFFERENT approaches. It saddens me that some wish to expend sooo much energy on disproving others modders work. It would be nice when people take the time to critique, if they would also give some examples of items that they have found to work, whether those findings come from their own work or from their testing of others work.

    The " I have a Dream... " from the speech by MLK is quite appropriate here, I have a dream too....that everyone who is attempting to improve the playibilty of this game would set aside their egos and try to work together instead of against each other.



    PS...sorry for posting the same thing in two different threads.

    "Nothing is more destructive than the charge of artillery on a crowd."
    Napoleon Bonaparte

  4. #4
    Germanicu5's Avatar Will buy spare time...
    Join Date
    Feb 2009
    Location
    Not Zee Germany
    Posts
    2,101

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    @MiiKll
    Now I gotta post twice too ;P... If I got to conclusion AI code couldn't be improved I wouldn't be developing my own.

    It all started with me disproving my own custom skirmishing code, then I looked how others did it and found your acknowledged AI, I came to conclusion some of that code looked disputable. Then I started thinking of other reliable testing methods while developing my own AI, I created or transferred countless lines of NEW and DIFFERENT code - but it turned out their functionality was close to "0", so I abandoned most of them.

    So after that I was mostly sure of what wasn't possible, I shared this knowledge when starting my new AI thread. If I happen to discover anything I can prove without doubt, (I still have lots of untested code) I'll share that too.

    I suggested in another thread XBAI skirmishing code wasn't working, but I got rather ignored. I don't think it's good for the community if we focus on adjusting unprovable code instead of finding resourceful ways of improving AI while accepting we can't rewrite the code from scratch. I wouldn't be worried if you had some additional lines of seemingly helpful code, but skirmish and engage distance settings in XBAI make the core revert to some "default" values, so instead of something new we end up with active prehistorical settings.

    The main post here is so long and I took so much effort, because I care about correctness of my statements, particularly given the short time I've officially been around.

    Regards
    I have no memory of this place.

  5. #5
    /|\/|\/|\/|\/|\/|\/
    Join Date
    Jun 2005
    Posts
    10,770

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    Nice work and well presented, Germanicu5.

    DLV+BB's BAI would not be as good as it is now if it weren't for your well thought out criticism.

    You're an asset to the community, keep it up! +rep

  6. #6

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    I'm playing with your version of AI in the latest RR/RC and am enjoying it as playing against a cavalry heavy moors they handle it very well (picking on my weak points) and had my derrier handed to me on my last battle, nicely done!
    member of S.I.N.

  7. #7

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    Checking it as we speak.
    Meanwhile i can prove right the theory of cavalry do not pull back in time, Pathfinding is an issue but it seems that the cavalry's choice always to attack archers and infantry should be worked on aswell:
    is there any way to make them retreat if unit type of spearmen is in their radius or anything?

    A human player will pull back and try to flank at a far point other units, to make the spearmen warry or scramble. Aswell he may take a group of infantry \ missile cavalry to kill those spearmen.
    So that's another thing (if possible) to be working on: some "behaviour" or priority. it's obvoius to us we wont front-charge a spearmen, but the question is if it's nearly possible to add \ modify valus \ battle behaviour for units of type X to attack\retreat unit Y.
    ?

  8. #8
    Germanicu5's Avatar Will buy spare time...
    Join Date
    Feb 2009
    Location
    Not Zee Germany
    Posts
    2,101

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    @ginger_hammer
    I'm glad you're enjoying the AI so far.

    @ShadowBane
    There's no trigger that would make cavalry withdraw when engaged by a specific unit type, there are some other indirect settings that could influence that a bit, but I'm really sceptical about obtaining perfect results.
    I dropped modifying sieges for now (still I've had pretty decent results with changes I've made so far) and I'll be looking into improving open battles again, so please be patient .
    I have no memory of this place.

  9. #9

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    "Thanks to Point Blank who featured it as default in RR/RC/CBUR(...)"
    ...Little bit confused, isnt "ThesavageAI" the standard in the actual version of RR/RC??

    Edit: Ah...this is BattleAI only, i see... noob out lol
    Last edited by algorath; April 25, 2009 at 01:41 AM.

  10. #10
    wolfslayer's Avatar Senator
    Join Date
    Oct 2007
    Location
    Lexington, South Carolina
    Posts
    1,170

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    I should mention file cache settings may be a factor with making BAI changes ingame, without adding this to your .cfg, the 2 existing BAI files will load and cache on game start.

    [io]
    file_first = true
    disable_file_cache = true

    With this entry added, the BAI files only be accessed on battle start/restart. With the entry completely removed, the game defaults to disable_file_cache = false.
    ______________________________________________________________


    Viewing and editing MTW2 textures with MWthumb and DXTbmp









  11. #11
    Germanicu5's Avatar Will buy spare time...
    Join Date
    Feb 2009
    Location
    Not Zee Germany
    Posts
    2,101

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    @wolfslayer
    Yea, log show both files are loaded only before battle with this setting.
    I quickly tested skirmish distances (5 custom battles, usual setup) - changes weren't commenced again. I'd prefer the game to be more predictable :/.

    Regards
    I have no memory of this place.

  12. #12
    Gorrrrrn's Avatar Citizen
    Join Date
    Jan 2007
    Location
    here
    Posts
    5,546

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    Did a selection of open field - grassy plain battles over the weekend.

    I'm using SS6.2 RC4 with 110AD sub-mod for english roster.

    Used 2 full (20 unit armies) - so I get to see if the AI attacks different types of units etc. 10,000 florins exactly, no upgrades or experience. all eras.

    Roster 1: English: 1 late GB; 4 yeoman archers; 2 heavy billmen; 2 dismounted kings men (men-at-arms); 4 heavy spears; 2 crossbow militia; 2 hobilars; 2 arquebusiers, 1 city guards (voulge miltia)

    Roster 2: Cumans: 1 late GB; 4 horse archers; 2 mounted cossack gunners; 2 archer militia; 2 nomad archers; 5 spear militia; 2 cuman heavy infantry; 2 pecheneg guard (spear)

    I played both English and Cumans once using the 0.71 and restarted game to test 0.72 radical, both medium and very hard level.

    I laid out the forces more or less the same way each time:

    English with 4 heavy spears in centre, 2 yeoman archers on each flank of them with stakes, outside them arquebusiers. in front on heavy spears crossbow militia. Behind heavy spear city guard and GB, behind yeoman archers 1 dis KM, 1 Heavy billman on each flank. Hobilars back behind arquebusiers. All skirmishing troops on skirmish (as AI usually starts that way too.)

    Cumans with 5 spear miltia in front with cossak gunners and horse archers on flanks. archers start in front on spear militia on skirmish. the cuman heavies and pecheneg guard 1 per flank. 1 GB centrally in rear.

    AI draws up its forces in symetrically regards cavalry but in order they are listed regards infantry and archers -so you get nomads on one flank and archer militia on other.

    Weather - I restarted from scratch every time weather was either too foggy or thunderstorm as these might affect comparisons.

    In every battle the AI attacked, throwing its archers etc forwards quite quickly and following up with infantry at walking pace. AI deploys its horse archers further away from infantry than I do, but as they are skirmishing they usually end up about the same distance away. AI Infantry seems to go for its opposite numbers, but usually has a couple of units in reserve. AI GB charges the archers (as do english hobilars when the AI uses them.) But they withdraw after contact with spears. (There is a constant problem with ordering skirmishers to move - if you order a line of 4 units to retire behind other troops at least will just stand there and get clobbered. I always turn skirmishing off once they have retired behind heavier infantry, or else they retire so far they do nothing. AI keeps its archers well back but active - seems to use fire arrows a lot more than I do.) Several times the AI GB outflanked my infantry and caught my back-line of archers and it sometimes managed to collapse a flank of my infantry as well. It seems the survival of the GB is key to winning the battle if it is closely fought. The one battle I lost was down to the AI GB having the general as sole survivor whilst my GB which had engaged it was destroyed. AI General then attacked my routing troops stopping them from re-grouping. Equally other battles swung in my favour once the AI General had died. AI GB was very good at destroying lines of arquebusiers - but they sometimes survived well enough to shoot the AI GB. Cuman horse archers and cossack gunners were the only troops I had left in one battle. I actually lost more troops but somehow the HA and gunners managed to see off the remaining english infantry. (The AI seems more reluctant than I am to order HA to attack routing troops, rather than keep a distance and fire at them. The only difference with VH vs M is the AI gets a morale boost as far as I can see.) As for difference between 0.71 and 0.72 radical - I think the 0.72 does a better job at outflanking - but how easy it is to outflank is difficult to tell when one side has horse archers on the flanks - I tried to keep my cossack gunners close to my infantry but they sometimes get very spread out. (Not a problem with this BAI - it happens with all of them I think.) The AI loses between 10-15% of its missile troops simply by advancing into contact. If the AI waited for the human to advance it might do better in some battles.

    I'll post a spreadsheet with numbers lost etc a bit later once I've done it.

    And here it is - used 7-zip and saved in.xls format (done using open office)
    Attachment 39335
    Last edited by Gorrrrrn; April 27, 2009 at 06:44 AM. Reason: update, spelling

  13. #13
    Germanicu5's Avatar Will buy spare time...
    Join Date
    Feb 2009
    Location
    Not Zee Germany
    Posts
    2,101

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    @Rozanov
    Wow... nice determination there, but I knew I could count on that since you first posted in this thread, thanks for help.

    I'll post new AI with minor bugfixes later on.

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

    I've been researching AI scripting possibilities for the past 2 days, we can surely script some conditions for general unit (like I_IsUnitUnderFire, I_IsUnitEngaged etc) and make it take some appropriate action. Or we could spawn custom battles (like Pavia) for human players during the campaign.
    For now I'm posting a useful campaign-script addition, extension-ready and tested.
    ***********************
    AI stakes deployment script v0.1:

    - executed over multiple battles and turns, doesn't slow down the game (monitor_event used)
    - AI deploys stakes during defensive day open battles provided that it's not succesfully ambushed (incl crossing battles and sally-out batles too)*
    - no siege defense stakes**
    - timer for stakes deployment (1 minute atm)
    *conditions editable, I could include stakes in night batles, that'd be some fun, we could feel like dumb AI for once
    ** although they really rock - a pity AI's own cav impales on them so easily and that there isn't much space to deploy them inside a settlement, cities are pretty decent though

    If anyone thinks it's interesting and would like to integrate it with a campaign, just ask. Ofc no limits on private use or integrating it with basic SS or RR/RC.

    Installation: Copy the code to your campaign script file anywhere between "script" and "wait_monitors+end_script" lines, preferably in the bottom of the file. Needs campaign restart.

    known limitations: stakes not deployed on roads

    Regards

    New versions downloadable via 1st post.
    Last edited by Germanicu5; May 08, 2009 at 08:49 PM.
    I have no memory of this place.

  14. #14
    blbrotto's Avatar Libertus
    Join Date
    Apr 2009
    Location
    Curitiba, Brazil
    Posts
    68

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    Hi Germanicu!

    I'm using your BAI and I'm loving it.. despite somethings, like when the AI uses the ballistas and cattapults on a siege but doesn't withdrawn with them if lose the battle and, when the gate is open (or openned by catapults or rams) the enemy drops all their siege equipment (sometimes they don't drop the ladders and attack the walls too...) and run to the gate like crazy! If you have good spearmen soldiers there, is an easy task to kill them... ah, and it's sad because the AI sends first the cavalry (even the general) in charge to the gate... could be a good ideia, if the infantry wasn't too far away to suport the cavalry....

    On open battles, the AI is great, you have to choice carefull what kind of infantry you send to engage, because the AI is using now the Elite units to outflank you, or hold the middle if it's outnumbered, while their cavalry combats your own cavalry.. if you lose the batlle cav vs cav, then they will charge into your rear or flank... and you will be *&%¨... XD

    Ah, I have a sugestion too.... I was thinking if is possible to your units lose morale when your city is under siege and the enemy hit your buldings (like houses, markets...) I think this would represent better the "fear" of your soldiers from an enemy with a strong tecnology...

    Please! Sorry about my english! XD, I'm from Brazil...

    PS: And about the AI, I know no one can do a miracle, but I apreciate your work! Keep going

    PS 2: And to the guys who think the AI should be invencible, try to "help" the AI, if you are under siege, and the enemy is winning the battle for the gates, think like the medieval general on the battle (if the general is not good with comand, try to play more "easy") ... but like a medieval soldier too... what you would do if the cavalry is charging uppon you and you are only a levy spearmen??? of course you will run to the central square and try to regroup, or to another position more tight! XD Of course this is my opinion and way to play... you will lose some battles... but the game will be very more fun... XD!

    Bye!
    Last edited by blbrotto; April 27, 2009 at 08:27 PM.

  15. #15
    Caesar Clivus's Avatar SS Forum Moderator
    Join Date
    May 2007
    Location
    Australia
    Posts
    12,693

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    Wait...you got the AI to deploy stakes??? Wow that is going to totally change battlefield tactics. Nice work.

    BftB2 UPDATED 22nd DECEMBER. Member of the Complete Byzantine Unit Roster team

  16. #16
    /|\/|\/|\/|\/|\/|\/
    Join Date
    Jun 2005
    Posts
    10,770

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    Stakes for AI... wow man, very impressive!

  17. #17
    Germanicu5's Avatar Will buy spare time...
    Join Date
    Feb 2009
    Location
    Not Zee Germany
    Posts
    2,101

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    Update: stakes deployment dependant on AI general's distance from the nearest player unit. We can actually make archers approach our units and stick their stakes under player's nose now.

    @CC
    Dooh... tyvm
    @blbrotto
    Thanks, AI behaviour is really hard to model, there aren't many settings available via 2 battle AI files, I'm currently working on changing that by making scripts.
    As for morale related to building damage... that's doable I think.

    Regards
    I have no memory of this place.

  18. #18
    Gorrrrrn's Avatar Citizen
    Join Date
    Jan 2007
    Location
    here
    Posts
    5,546

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    Re AI using stakes.

    Quick question - given the AI puts its cavalry on wings (usually) but attacks missile troops (even if in the centre)
    where are you going to get it put its archers with stakes ability?

    If a human counters advancing archers by advancing with its own archers, then its probable that the AI will charge the human archers - through its own stakes?

    Can we get the AI to advance at a walk until much close to their target - if the target is stationary anyway. You can walk through stakes without casualties.

    (btw - the AI gets to plant stakes AFTER the set-up phase (ie during the battle?))

  19. #19
    Caesar Clivus's Avatar SS Forum Moderator
    Join Date
    May 2007
    Location
    Australia
    Posts
    12,693

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    Quote Originally Posted by Rozanov View Post
    (btw - the AI gets to plant stakes AFTER the set-up phase (ie during the battle?))
    Oh really? Ouch!

    BftB2 UPDATED 22nd DECEMBER. Member of the Complete Byzantine Unit Roster team

  20. #20
    Germanicu5's Avatar Will buy spare time...
    Join Date
    Feb 2009
    Location
    Not Zee Germany
    Posts
    2,101

    Default Re: ReallyBadAI - Stainless Steel, RR/RC battle AI development and discussion

    @Rozanov and CC
    Yea, the stakes are deployed 60 secs after battle starts in the first version (I mean real battle time, excluding pauses) and as soon as any enemy unit reaches 100m distance from AI's general unit in the alternative one.
    This is due to AI often adjusting formation in the beginning of the battle, like reforming to a certain distance during sally-out battles or even in the beginning of normal open battles.
    Stakes are currently deployed by missile units according to their place in formation, so in the middle, cav is put on flanks and it's possible it may impale on them while maneuvering under some circumstances, but that's a matter of battle files adjustment and maybe writing a script (I've been looking into that already, it may not be that easy, but it's doable).

    ad. "Can we get AI (...)" I believe the alternative script somehow answers that, actually it'd suit attacking AI too, so I'll make some amendments and release another version soon.

    Btw, I named it v0.1 only because this is the first released version, I've been testing it a lot and I've tried many scripting possibilities and these two are the most compact and reliable ones.

    Regards
    Last edited by Germanicu5; April 28, 2009 at 08:11 AM.
    I have no memory of this place.

Tags for this Thread

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •