Promising news about the ai

[Lusted]

As some modders at TWC have been able to crack the first demo to allow unscripted custom battles of a sort, there have been some very good results reported in relation to the ai. Check this thread out for more info.

http://www.twcenter.net/forums/showt...6&page=1&pp=20

[Phil17]

That dpes sound good Lets hope it wasnt just a fluke on the AI's part lol

[Cheetah]

So basically what was said in the developers blog is potentially right.

I wonder whether you guys could trust Palamedes without cracking the demo ....

[ELITEofWARMANGINGERYBREADMEN88]

Hmm Looks good. Hope the AI is better in the "Finsihed" game though..

[shifty157]

I wonder whether you guys could trust Palamedes without cracking the demo ....

Well CA was saying some nice things about the AI before RTW's release.

[hoom]

Also listen to the 5th podcast on the official site, talks about some technical AI stuff that sounded pretty good :)

[Spino]

I was really tempted to post a new thread just to make sure everyone got the news but it makes more sense for me to post in here.

I just read an incredibly insightful post by one of the developers (Palamedes) over at the official forums. He was posting in reaction to the negative reaction people have had over the AI in the older (Build 47) demo.

This is amazing stuff; amazing as in this is the single most informative post on the AI I have ever seen coming from one of the developers...

http://p223.ezboard.com/fshoguntotal...art=41&stop=59

Ok guys to be fair to those with concerns……………..

I will give you a slight insight into what we have done with the current AI. In previous titles AI's focus was target acquisition and movement, they would choose appropriate targets and move toward them. M2TW’s AI system adds a preference for maintaining formation.

From a developers point of view a balance has to be found between perceived intelligence and actual intelligence. For example a lone infantry unit guarding a village outnumbered by horse archers may withdraw to the plaza and hold its ground. It will continue to stand there and not move even though its being shot. This is actually intelligent because better to be shot in its plaza than run after horse archers in the open. However the lack of movement was perceived dumb or worse broken. To address this particular example a little movement back and forth was added to give the impression of thought.

As far as I know all you veterans out there should know that the way to beat battle AI in the past is to use a few units to draw off many AI units. You then stack the units that weren’t drawn away, route them and then move on the units now chasing your “bait”. Movement equals perceived intelligence so previously staying as a group was secondary to chasing a target that you knew you could kill.

It would have been very easy and safe to leave the AI in this state, however we took the risk to do what no previous title has done with battle AI. Once all battle mechanics were finalized Iain, Artem and myself worked like crazy to employ logic many multiplayer’s would know and appreciate. For human players it’s all about getting more men to the engagement first and maintaining a formation with the rest of the army. The players that don’t keep their units together loose, those that do keep them in formation win.

When we first introduced the new logic the AI became very cautious, players could not draw any units away from the main army but the army stopped and assessed its next move too slowly. This was then sped up but caused it to digress into not considering army formation any longer. We kept refining this until a good balance was found. Once this was achieved all related exploits had to be addressed. The last of which was that when the AI group moved toward its primary target you could keep charging lone cavalry units into the army’s rear and it wouldn’t respond at all. Unfortunately this last fix generated another exploit that made it into build 52 which is Gold. If you interact with the AI it does fine but if you don’t it will move to charge distance and wait for you to make a move, indefinitely.

This was fixed a few days later with many other fixes that will be available for the first update that as far as I know will be available on release.

I know for many what us developers say maybe seen as lip service so my best advice is listen to Epistolary Richard from the .ORG. We have flown him to our studio for 10 days as a community representative here to help us grasp how we can best support modding. He played build 52 today and while doing custom battle tests found the AI defect in question. I then asked him to try using the build with the update fixes installed, as far as I know he is yet to win a battle.

Anyway I picked him up at the airport at 6:30am today and his head hit the table with fatigue by 2:00pm, so we carried him to his hotel where I am assuming he is sleeping like a baby. Anyway I will ask him if he would like to respond to this thread tomorrow if he finds his way to the studio (man should have given him breadcrumbs or pebbles hey it's Australia at most he will be an hour late drunk wearing shorts, singlet and thongs).

What you will love and ER is yet to see, is that when the AI has more missile strength than you it will ensure that its non-skirmish units all skirmish as a battle group so that it can continue to shoot you to pieces. For those that know TW battle AI you know that this change is major, previously when you lost the shootout you could just charge and the AI would oblige and charge back not using its missile superiority. Anyway tomorrow I will set up a custom battle between a Mongol hit and run army and a slow moving infantry heavy European army, ER can play it and I will ask if he can give you an accurate assessment of where AI is at.

So to conclude build 47 was still in very basic overly defensive form, 48-50 smoothed it out, 51 & 52 fixed exploits and introduced a silly one that you wont notice unless you go answer the phone come back and realize “hey, what’s it waiting for?”. The first update fixes this and smoothes out the movement even further. The battle AI has so much good stuff in it and more importantly it will continue to evolve. Already we are designing shootout phase logic with cavalry skirmishing, and better use of elevation and vegetation. This will continue indefinately untill we have a battle AI system we can say is the best out there.

Man this turned into a long post………… please excuse any lack of clarity its getting late and my Wife will kick my ass soon.

Jason

[Bob the Insane]

Wow...

I work in business software as a support manager and if one of our developers did that the Product Manager would gut him...

Still, that was cool...

[Spino]

Wow...

I work in business software as a support manager and if one of our developers did that the Product Manager would gut him...

Still, that was cool...

I know, CA has never done (or allowed?) this kind of posting in the past. Either Palamedes crossed the line of non-disclosure or this truly is a sign that CA is going to great lengths to reach out to the hardcore TW community. I wonder if this newfound dialogue between CA and the TW community is the direct result of CA being owned by Sega instead of Activision?

Now I'd be positively giddy if CA commented on how they improved the strategic AI as well...

I must say Palamedes' post has done wonders to boost my optimistic expectations for MTW2. I think I'm perilously close to jumping down from the fence...

[LadyAnn]

I like this version of CA :D

Annie

[Vladimir]

Way to go ER!!!

My *MTW* experience with AI is slightly different. I didn't think the AI was stupid by standing and taking shots from my HAs but that at maximum range they not only turned their backs but they bunched up as well! That was stupid, find some trees and give each other some space!

I still think that with this build I can use the Jinettespam. I doubt the AI will be cav heavy and it'll HAVE to react to half a dozen units charging the line and firing javelins. It will take too many missile units to counter them. Using units like them and mounted X-bows it is possible to confuse the units and lead to the creation of stragglers. How will their army react to being pincered *and* anviled?

Edit: I do like the work they have done, however and I have an understanding of how difficult such a task can be, especially considering that AI programming is one of the last things to be done and doesn’t get much time.

[fallen851]

I like this version of CA :D

Annie

For along time I cursed CA in ways that wouldn't get me warnings on this forum, upset that a game with so much potential turned out so wrong (RTW).

Palamedes post may make me rethink CA.

[edyzmedieval]

Wow, that's awesome news. Really incredible.

[Zatoichi]

That was a great post - kudos to Jason for going into so much detail.

I look forward to seeing EP's AAR post tomorrow - jet lag permitting!

Sounds like we get a patch on the day of release, which will be a first (excluding the Multiplayer patch for Rome). That's a bit of a shame, but still pretty encouraging in terms of support!

[frogbeastegg]

I like this version of CA :D

Agreed. It's this kind of detail which makes it easier for me to believe something without adding a pinch of salt. I'm a sceptical frog at heart.

This sounds fabulous. I'm now really looking forward to playing the game at the Kew Bridge event and seeing this in action. I wonder which version will be on display there?

[Lord_Phan]

I like everything he said except the part about standing around getting shot being better then charging.

A lone infantry unit standing around getting shot has a 0% chance of living, as opposed to your 1 in a million shot with Charging the Horse Archers. Thus, You must Charge.

[Kor Khan]

I'm not sure I agree: Imagine you are one of a unit of infantrymen and you have a big shield. Surely the best way of surviving against a horde of horse archers would be to cower behind your shield and armor until they run out of arrows. Then you and your mates could bunch up into a tight formation and hold your ground, forcing the horse archers to either engage you in melee (certain death for them) or to retreat.
Ideally though, you should have a few archers and cavalry units with you.

[Zawath]

Indeed, there's no point in charging at horse archers because infantry can't catch them. Hopefully the AI infantry uses the same behaviour when dealing with regular cavalry. Way too often in RTW the enemy starts pursuiting your lone cavalry unit with half of their soldiers.

[Lord_Phan]

Let me get this straight, you're surrounded by horse archers so you're going to stand there and get shot at until you're all dead? You're one unit against an army of Horse Archers. You somehow think that these thousands of arrows arn't going to cripple, maim and kill your soldiers and add to that you think you'll have the advantage in the melee against their charge and subsequent flank? What school did you guys learn game theory from?

[Puzz3D]

I like everything he said except the part about standing around getting shot being better then charging.

A lone infantry unit standing around getting shot has a 0% chance of living, as opposed to your 1 in a million shot with Charging the Horse Archers. Thus, You must Charge.

Standing and being shot might be the best tactic. The HA may run out of arrows before the infantry unit is dead. Chasing a unit that you can't catch isn't smart. You'll loose your shield protection and will get exhausted. Either way in that example given by Palamedes, the infantry unit is not going to win. If the infantry unit could corner one of the HA quickly, then it would be better to charge, but I don't think the AI is sophisticated tenough o figure out that it can corner a unit.

Also, if the enemy cav are not shooters, an infantry unit should wait and make the cav charge. It was relatively easy to exploit the AI's tendency to chase enemy cav units with infantry in the previous games.

Palamedes post is very encouraging concerning the battle AI. It shows some innovation in this area, and considerable effort to address possible exploits of this new AI. I think the concensus around here was that RTW battle AI was very similar to STW, but it didn't work as well given the new battle engine. They either had to change the AI or return the gameplay to a stronger rock, paper, scissors system to get more challenging SP battles without resorting to simply giving the AI more combat bonuses.

[TB666]

I agree with Kor Khan.
You have a greater chance of survial if you stay put and take cover behind your shield.
If you chase them you will only die tired.

[Lord_Phan]

If it was 1 unit vs 1 unit I'd agree with you.

If this was out in the field you should fall back. If you're surrounded in your city center and they're surrounding you shooting at you then it's time to gamble. Standing there dying is not preferable to charging and dying.

Anyway it's highly situational. I think it was a bad example. Ideally the AI wouldn't put itself in that position. Strategic AI should be thinking 3 moves ahead based on it's possible enemies. I was merely pointing out the risk to reward situation involved, aswell as the fact that the cavalry charge and subsquent flank would decimate the infantry unit. If it was pikeman it'd be best to back into a corner assuming of course the amount of horsearchers shooting at you isn't mathematically going to reduce you unit all but a few men.

[wraithdt]

What you will love and ER is yet to see, is that when the AI has more missile strength than you it will ensure that its non-skirmish units all skirmish as a battle group so that it can continue to shoot you to pieces. For those that know TW battle AI you know that this change is major, previously when you lost the shootout you could just charge and the AI would oblige and charge back not using its missile superiority. Anyway tomorrow I will set up a custom battle between a Mongol hit and run army and a slow moving infantry heavy European army, ER can play it and I will ask if he can give you an accurate assessment of where AI is at.

This is awesome. I'm happy and fearful at the same time. I'm really gonna dread fighting the Mongols or any army with a large HA force in M2TW. Sucks to be at the receiving end of an arrow storm.

[Geoffrey S]

Standing there dying is preferable in almost all cases. The longer they hold out, the longer they've got chance of reinforcements of some kind arriving to support what is presumably a useful position.

Anyway, thanks to Palamedes for such an extensive and insightful post; it's precisely this kind of thing that is making me hopeful for Medieval 2.

[Puzz3D]

If this was out in the field you should fall back. If you're surrounded in your city center and they're surrounding you shooting at you then it's time to gamble. Standing there dying is not preferable to charging and dying.

I don't see any difference. You dead anyway and you didn't inflict any casualties. In fact, you're dead sooner. So if there are reinforcements coming, you're less likely to be alive by the time they get there. In anycase, I think it's too much to expect AI of the sophistication you want. The game is either going to give you an AI that always chases cav with infantry or doesn't chase. Take your pick, but you can come up with lots of examples where chasing is bad.

Anyway it's highly situational. I think it was a bad example. Ideally the AI wouldn't put itself in that position. Strategic AI should be thinking 3 moves ahead based on it's possible enemies. I was merely pointing out the risk to reward situation involved, aswell as the fact that the cavalry charge and subsquent flank would decimate the infantry unit. If it was pikeman it'd be best to back into a corner assuming of course the amount of horsearchers shooting at you isn't mathematically going to reduce you unit all but a few men.

The whole idea of holding the city by protecting the central plaza is artificial. If the infantry unit backed up aganst a wall or into a corner to protect its flanks, the plaza falls after 3 minutes and the city is lost. Even if you charge after one of the HA, the city falls 3 minutes later while the infantry unit is chasing the HA down one of the streets.

[Martok]

Wow. I have to say I'm pretty impressed by Jason's post. I too concur with LadyAnn that I really appreciate it when CA shares that kind of information with us. I think I'm actually starting to feel some optimism for the game again, which is a pretty strange sensation at this point. (It could just be a side effect from sleep deprivation, however, so I can't be too sure yet. )

Regardless, kudos to Palamedes and CA for reaching out to us veterans and explaining the combat AI in more detail.

[Bob the Insane]

I guess this discussion demonstrates why the AI can never be perfect... There is no perfect in tactics and battles. Decisions can only be judged good or bad in hindsight once the battle is done.

And even then historians and armchair generals with analyze and critique battle for centuries or even millenia...

[wilddog]

Personally that was an excellent post. It was very useful to understand some of the 'thinking' behind the AI. I hope this sor of dialogue continues and extends to cover both Campaign AI and siege AI - both attacking and defending.

[Barkhorn1x]

Excellent post by Jason - excellent detail. And from that detail one now understands that CA understands what "we" are looking for in a TAC AI.

Barkhorn.

[Prince of the Poodles]

Great news!

It seems like they've really put a lot more thought into everything about this game.

8)