13/05/09 Taiji Battle AI for TA:TW+RC 1.1

[Taiji]

Hi all, I made some battle AI for DLV and I hear it's quite good, so here's my approach to battle AI adjusted to TA:TW for your enjoyment

Features in brief:
- Uses flanking well. Cav will typically avoid stakes and frontal charges vs anticav.

- Very aggressive, usually will not try to outshoot you before engaging in melee.

- In most circumstances on the field it guards missile troops and tries to counter flankers.

- Siege defence is the best I have seen, particularly with larger settlements that have large garrisons. Many BAI's use settings which cause stages to pass too quickly or even be missed resulting a boring 'get through the gate then fight them in the square' everytime, I've tried to avoid this and to encourage the AI to go on the offensive when it thinks it is winning.

- Siege offence is aggressive and does not stall, ladders and siege-towers are more effective but the focus is on the gate or breach.

- An unusual point with the siege mechanics is that in most cases the "hold the square" timer is disabled and as a result AI forces will often defend the city defences rather than fall back. This prevents many exploits or so-called "clever tactics" that use an unrealistc gameplay element - the timer. The approach also encourages the AI to make maximum use of towers to thin your ranks.

- It raises wall and tower HP to allow more time for the AI to react to breaches. It also increases the manning radius for gateways.

- Another important feature is that freezing during sieges has been eliminated.

____________________________

I've adjusted it to fit RC which offers the best battle mechanics available so you will need to install Real Combat for TA:TW submod before you install this: Attachment 41832 Just stick the contents in "...\mods\Third_Age\data".

Changelog:
v0.2 - 07/05/09: Campaign relevant variables better adjusted to suit TA:TW, based on the assumption that the previous BAI's values had been tuned for the CAI.
v0.3 - 08/05/09: Adjustments to improve flexibility, an attempt to better cope with the huge variation in army composition TA:TW is capable of.
v0.4 - 10/05/09: Alteration to better accomodate the awesome Germincu5 'AI stakes deployment script' (now in TA:TW+RC). Plus some fixes and tweaks.
v0.5 - 11/05/09: Adjustmented to better fit present RC. Reduced wall HP a touch.
v0.6 - 12/05/09: Keeping up with PB's bugfixing. Sally-out routine made more stable.
v0.7 - 13/05/09: Numerous tweaks. Revised siege defence and made it more flexible and responsive. Reduced wall/defence hp, roughly balanced to AI response times. Compatible with RC for 1.1.

Let me know if you have any problems or questions, feedback will be much appreciated

[Ragnarök]

Thank you, I'll try it. +rep

[Tharros]

Hey Taiji! Will be definitely trying this out on the weekend!
+rep

[Taiji]

Thanks very much guys, I'd appreciate any feedback you might have on it. With the amount of variation in army composition this mod provides I am expecting some problems but I can't seem to find any big ones so far

[Taiji]

Updated, changed some variables I forgot to adjust in order to correctly import my BAI from DLV to TA:TW. Stuff that helps the CAI makes decisions is stuff my AI tries to avoid messing with, so I have restored certain parameters to the values found in the default TA:TW BAI.

Could one of the 40 or so people who downloaded it yesterday give me some feedback please? Testing it all by myself is so... lonely

... Well there is so much to test! I'd love to be able to imagine that with it working so well in DLV it probably works well in TA:TW too but I can't, it just wouldn't make sense! There must be something glaringly obvious wrong with it for at least one faction is my suspicion, but so far it only does good things...

[HUtH]

- An unusual point is that in most cases the "hold the square" timer is disabled and AI forces will often defend the cities defences rather than fall back. Which prevents many exploits or so-called "clever tactics" that use a unrealistc gameplay element - the timer.

I'm not fully understand. it means that timer during assaults was totally disabled? So if I am outnumbered 10:1 and have 1 unit, I don't have to defend the center and my unit can stay on the wall?

[JonSnow]

Has anybody tried to compare this to germanicus'? I'm curious as to which works better. I'm loth to replace his with this one though until I hear a review because his works pretty well taking into account the crappy hardcoding the AI has...

[Taiji]

HUtH, Up to a point yes, although outnumbered 10:1 is not really what it is meant for. Basically the attacker will have to get most of his 20 stack army into the square before the timer will start so in situations where the attacker greatly outnumbers the defender the timer can be used... to save time.

If an attacker tries some 'clever' tactic of getting a unit into the square with another unit to block the defending army from taking the square back, it will definately not work. This way players are forced to fight more realistically most of the time when attacking settlements.

JonSnow, I've tested Germincu5's BAI and I think it's good. As to which is better, it's a tough question which I think personal choice will come into. You should test it out for yourself because of this, noone else will know what you will make of it.

BAI is very subjective, effectively I am trying to use my BAI to beat a certain kind of player and if you're that type then you're bound to like it.

I'd suggest you back up what you're using and give it a shot

[Hellenikon]

Looks very good Taiji!

I remember using this AI in DLV and it was pretty challenging, specially on sieges as you said. With this ai you have to actually fight on the walls to take the town, instead of going directly for the town square.

I don´t know much about germanicus one, to be fair. I guess all is up to test both and see what works better for you.

I´ll try this when RC releases (to have all the unit stats similar), and start a new campaign to avoid any problems.

As i said i´ve already played with Taiji´s ai and liked the experience, so i´ll definetively give this a shot to see how it goes.

[Radagast the Brown]

Im testing now ... I realy like RC from Mith and Point Blank so if you re working to make the AI fit into theire mod great. What I noticed without your BAI the AI do not focus on bringing Trolls and Mumakil down with theire archers and ballista units. What realy makes winning such games easy.

Feedback posted later thx and +rep

[Lap]

I'd really like to see how Germanicus's ReallyBadAI and this compare to each other. I simply don't have time to test this stuff out myself right now so any experiences comparing the two would be appreciated.

[Taiji]

Thanks Bucefalo and Radagast, I appreciate the praise

Radagast is that really true? I was expecting it with trolls but I'm very pleased to hear about it happening with the Mumakils. To be honest controlling the AI's missile attack choices is very hit and miss... (sorry bad joke) but with DLV I had it attacking guys with no shields pretty consistantly. My ideal would be that they consistantly went through a clearly defined list of priorities with generals at the very top but there seems to be no way I can find of implementing this.

Lap, actually I too would be very interested in the comparison but I would like to stress that I don't want to see heavy criticism of someone elses work in my thread.

I think it's great for the BAI community to have an open discussion of all the potential BAI's we could have available for TA:TW but once someone gets defensive or resentful it's all liable to fall apart. All of us that work on BAI have probably put in a hell of a lot of hours testing and experimenting, it can be very annoying to notice there is a problem with the work you've done and to notice it by way of people insulting it makes it feel worse.

We're trying to improve things and it doesn't matter how 'right' we are if things do not improve. I really hope it's not going to turn into a competition where people decide who has the best AI overall, that would be a cop out. For me the useful part of any comparison is in noting the pros and the cons of whichever AI is mentioned. It's not feasible that any one AI is going to be better at everything, sometimes even if the AI is broken (file is not being read due to errors) it will provide a better challenge than any other AI. The situations and variables of battles range so wide that that the restrictions we try to enforce in the AI will sometimes be inappropriate.

There's been some talk from Germanicu5 about scripting alternate AI's to be in use at the same time, very interesting talk. If we can have different AI used on a per faction or army composition basis then we'll definately have the best BAI system for any TW game. What Germancu5 has achieved so far, with scripting the deployment of stakes, has impressed the hell out me and certainly gives me some faith that we can alter BAI on the fly using the campaign script.

I can appreciate that some people don't have the time to be testing BAIs but I really recommend those that do to make the effort and give the feeback, comparisons and criticism for all of us BAI makers if possible. It will help us improve the BAI faster to have all our faults and successes highlighted but please try to be respectful

[JonSnow]

That would be awesome if we can have it randomize between your ai, germ's ai, darths AI and maybe XBAI I guess...

[Point Blank]

Randomization is definitely possible.

[Taiji]

Yeah you're right, even just randomised would be cool. I enter a state of game design ecstacy imagining the BAI being 'intelligently' selected on the basis of the battles characteristics so I'm really hoping for that

Point Blank, nice to have that confirmed

[Lap]

Taiji-If you're that worried about this degrading into insults and what not perhaps a new thread comparing AI's would work better. At least the battleground wouldn't be in someone's home thread should it somehow come to that. Gotta do it somewhere though as us players can currently only decide on one Battle AI mod.

Doing randomization is really the only alternative to everyone working together on a single AI. I know everyone puts a lot of work in their AI's, but as far as the player is concerned, if AI X works better than AI Y in 51% of situations then the player will exclusively use X. AI Y might as well not exist. If we didn't have to overwrite the same files I'm sure most people would prefer to add on AI's instead of doing global replacements. Randomization would probably be the best option because of this.

[Taiji]

I'm only really specifying this thread as off-limits for heavy criticism of other peoples work because this thread is the only environment where I feel an obligation to try to control other peoples behaviour in any way. On the other hand, I want criticism, even heavy criticism, for my own work and I don't care where I find it. I really appreciate the efforts of those that take the time to write something.

So it's not that I don't want to see peoples work assassinated just in my thread, I don't want to see it happening to others at all. My work is another matter, I really want criticism and I'm willing to suffer insults to get it

It's a good point you make about the current situation, there is no simple way to be switching BAIs around atm. The current contenders are all very good in their own ways, I won't go into detail because I don't want to color your perspective. You want a simple choice and I want unbiased feedback... sorry for being cagey about sharing my comparison, I'm honestly not just trying to frustrate you for no reason.

I think if one works better in 49% of scenarios and one in 51% but the 49% does some interesting things that forces you to change tactics then you'll want to use both. I really think we are talking about small differences in overall effectiveness but sometimes with a marked difference in flavor.

[Point Blank]

I'd also like to know if we can randomise the formations file. That combined with BAI randomistaion would result in a huge number of possible tactical situations to face.

If formation randomisation is possible, it could be synchronised with the config_ai_battle that is chosen, eg if the config_ai_battle file has been altered so that it has a 'defensive' emphasis, then the available formations could also be configured with defense in mind.

[Germanicu5]

I'd also like to know if we can randomise the formations file. That combined with BAI randomistaion would result in a huge number of possible tactical situations to face.

If formation randomisation is possible, it could be synchronised with the config_ai_battle that is chosen, eg if the config_ai_battle file has been altered so that it has a 'defensive' emphasis, then the available formations could also be configured with defense in mind.

I remember my proposal was to expand formations file and then just use switcher to swap battle AI files with different formation types named inside (although, as we know, some of settings don't reload - that needs actual checking before we start doing anything else).

Regards

[Taiji]

I've noticed that the game seems to load any files of a similar name to files it expects, e.g. copy of export_descr_units.txt, I figured it would be for scripted switching of files. DLV does it in places I think... anyway descr_formations is cached in the same way and so "Copy of descr_formations_ai.txt" gets cached at game launch. Presumably any file being cached like that can be switched to.

I think it looks possible so far

edit: It is a really nice idea to have defensive and offensive formations alternating with the engagement type BTW, lots of possibilities there.