How is the AI? (Full Version)

All Forums >> [Current Games From Matrix.] >> [World War II] >> WW2: Road to Victory



Message


Forwarn45 -> How is the AI? (7/26/2008 4:02:52 PM)

Since there is no current multiplayer option, the AI is especially important. So I was wondering how the AI was? I'd be interested to hear from some players with reference to how they are doing against the AI so far in games.




Zakhal -> RE: How is the AI? (7/26/2008 8:34:13 PM)

You can play multiplayer too. Just choose human players for the nations and then email/hot-seat the savegame. Kind of crude but it should work. 




Jim D Burns -> RE: How is the AI? (7/26/2008 10:05:29 PM)


quote:

ORIGINAL: Forwarn45

Since there is no current multiplayer option, the AI is especially important. So I was wondering how the AI was? I'd be interested to hear from some players with reference to how they are doing against the AI so far in games.



I posted this in another thread:



The AI so far feels adequate, but Iíve only just played into Barbarossa, so I havenít yet finished a game. The one problem I see with the AI is it defends every city (as it should), making it easy for you to bypass non-VP cities without having to fight those units.

I donít view this is an AI weakness, but rather a scenario design weakness. There are few enough cities on map, that I believe each one should have at least one VP, more for critical production cities and capitals of course.

Then change it so non-winter turns are one week (gives you time to take all the extra VP locations) and I think the AI would be a lot tougher as youíd now have to fight all is units, not just those defending VP sites.

There is one AI problem I view as a bug and not a weakness. For some reason the AI sometimes abandons critical VP cities. I saw the AI pullout of Paris once as one of my armor units moved close enough to grab it in one move. I guess the AI failed to recognize the danger and was moving the defending unit towards the *front*. I saw this again as I closed on the Yugoslavian capital too.

The AI needs to keep a permanent garrison in every VP city no matter what. Perhaps creating fixed garrison corps that cannot be moved would fix this. Or simply tweak the AI so its first order of business is to make sure every VP city is garrisoned before it moves any other units.

Of course I restarted the game and this time around Iíve had to fight for every VP location Iíve captured so far.

Jim




comrade -> RE: How is the AI? (7/27/2008 10:44:13 PM)


quote:

ORIGINAL: Jim D Burns

There is one AI problem I view as a bug and not a weakness. For some reason the AI sometimes abandons critical VP cities. I saw the AI pullout of Paris once as one of my armor units moved close enough to grab it in one move. I guess the AI failed to recognize the danger and was moving the defending unit towards the *front*. I saw this again as I closed on the Yugoslavian capital too.

The AI needs to keep a permanent garrison in every VP city no matter what. Perhaps creating fixed garrison corps that cannot be moved would fix this. Or simply tweak the AI so its first order of business is to make sure every VP city is garrisoned before it moves any other units.



Garrisoning algorithm has been improved and in 1.20 AI should no longer abandon cities.




Vypuero -> RE: How is the AI? (7/28/2008 12:22:53 AM)

Is the 1.2 patch available yet?




JMass -> RE: How is the AI? (7/28/2008 4:25:21 AM)

I am playing as Soviet in the 1941's scenario but the AI seems to be not so smart, it is very simple cut off from supply advancing german units and destroy them the turn after. Around Odessa, held by one my unit, it is a enormous concentration of axis units that remain in place turn after turn without to do nothing, I'll take one screenshot.
Definetively I hope in a patch to have a secure pbem.




Vypuero -> RE: How is the AI? (7/28/2008 6:26:37 AM)

Yes the British in North Africa seem to leave their cities when I approach, making their capture kind of easy.  Otherwise it was tough - one motor unit held out in Jerusalem for a long time.

No oil (resources) in the ME?




JMass -> RE: How is the AI? (7/29/2008 4:30:07 PM)

quote:

ORIGINAL: JMass

Around Odessa, held by one my unit, it is a enormous concentration of axis units that remain in place turn after turn without to do nothing, I'll take one screenshot.



Here's the situation, it it one full year that the ring is around Odessa, some units have left, some are arrived as substitutes.

[image]local://upfiles/20885/B0D598B39E094F2C8DA48DBDB6D24C6E.jpg[/image]




JudgeDredd -> RE: How is the AI? (7/29/2008 5:20:47 PM)

I was very, very close to buying this...but looking at that shot, I'll wait. Clearly it needs a little tlc.




Erik Rutins -> RE: How is the AI? (7/29/2008 5:29:11 PM)

That is quite a bizarre situation there, I agree. Do you have a save for this that you could post or e-mail to me?




JMass -> RE: How is the AI? (7/29/2008 5:51:20 PM)

quote:

ORIGINAL: Erik Rutins

That is quite a bizarre situation there, I agree. Do you have a save for this that you could post or e-mail to me?


Yes, here are the screenshot's savegame and another some turns later.




comrade -> RE: How is the AI? (7/29/2008 7:17:15 PM)

The reason for this is that it's not possible to assemble an attack with units from different countries, and AI won't attack on poor odds, so it just sends more and more units to the area. During our tests we've had the same problems with Gibraltar and Sevastopol where AI could not conquer the city. Events were added (Siege of Gibraltar and Siege of Sevastopol) to allow the AI to overcome this problem - if the city is blocked for 3 turns, garrison will surrender. There is an event for Odessa in 1.20, this should do the work.




winky51 -> RE: How is the AI? (7/29/2008 8:18:59 PM)

Just FYI for all you gamers.  Making a good AI for a game this complex is very very VERY difficult. 




JMass -> RE: How is the AI? (7/29/2008 8:24:06 PM)


quote:

ORIGINAL: comrade

The reason for this is that it's not possible to assemble an attack with units from different countries, and AI won't attack on poor odds


Thanks, so for now I think the better thing to do is sea transport the unit away to unblock the situation.
Could you answer me if in the future it will be possible have a pbem's modality?




Plain Ian -> RE: How is the AI? (7/29/2008 8:43:09 PM)


quote:

ORIGINAL: winky51

Just FYI for all you gamers.† Making a good AI for a game this complex is very very VERY difficult.†


Which is why a lot of players graduate to PBEM or even prefer to play these type of games as pbem. The AI is usually only for training and learning the game mechanics. It will never be a good opponent. This is probably why Malta was left off the map, as the AI will have to deal with it.





Plain Ian -> RE: How is the AI? (7/29/2008 8:53:40 PM)

quote:

ORIGINAL: comrade

The reason for this is that it's not possible to assemble an attack with units from different countries, and AI won't attack on poor odds, so it just sends more and more units to the area. During our tests we've had the same problems with Gibraltar and Sevastopol where AI could not conquer the city. Events were added (Siege of Gibraltar and Siege of Sevastopol) to allow the AI to overcome this problem - if the city is blocked for 3 turns, garrison will surrender. There is an event for Odessa in 1.20, this should do the work.


But 3 turns can be differing lengths in game time depending on weather/season? Will this be taken into consideration? Other considerations would also be nice, eg who has naval control of the sea area. This could shorten the time period.

I like the idea of 'lack of cooperation' between different powers but how about allowing 'major powers' be able to 'borrow' or take control of limited numbers of 'minor countries' so that they could move and attack with them?

eg Germany with Rumania/Hungary....nad also Italy? That would also help in situations like this plus add further historical content? Too difficult to code?




JudgeDredd -> RE: How is the AI? (7/29/2008 9:41:36 PM)


quote:

ORIGINAL: winky51

Just FYI for all you gamers. Making a good AI for a game this complex is very very VERY difficult.

Well, we all know that's the case (or most of us)...but seeing so many units round a city because it's not taken seems a bit raw.

Seems to me the AI code is "if there is a city nearby send troops"....next turn is exactly the same and so on. No check to see if there is space round the city. No check to see if units round the city need replacing/relieving. No check to see how many units are around the city.

It just seems to me there are too few checks and that cities are magnets for units, regardless of requirement. I am, of course, only going on this screenshot and I haven't got the game as yet.




Widell -> RE: How is the AI? (7/30/2008 2:11:09 PM)

quote:

ORIGINAL: comrade
Events were added (Siege of Gibraltar and Siege of Sevastopol) to allow the AI to overcome this problem - if the city is blocked for 3 turns, garrison will surrender. There is an event for Odessa in 1.20, this should do the work.


This sounds like a reasonable quick fix, but not something you'd want to maintain over the life of the product. IMHO you can't fix all strange AI behaviours with events as the playability of the game will gradually be reduced as events take over rather than the player actually achieving military successes.

May I propose a piece of code that checks surrounded and cut of units every turn against some kind of factor (Attack vs Defence strength, number of turns surrounded, morale, whatever) and based on the result of the check, the surrounded unit may, or may not, surrender. The surrender may then of course trigger one or more events if that is the correct thing to do in the particular scenario.




comrade -> RE: How is the AI? (7/30/2008 3:32:02 PM)

quote:

ORIGINAL: Widell

quote:

ORIGINAL: comrade
Events were added (Siege of Gibraltar and Siege of Sevastopol) to allow the AI to overcome this problem - if the city is blocked for 3 turns, garrison will surrender. There is an event for Odessa in 1.20, this should do the work.


This sounds like a reasonable quick fix, but not something you'd want to maintain over the life of the product. IMHO you can't fix all strange AI behaviours with events as the playability of the game will gradually be reduced as events take over rather than the player actually achieving military successes.

May I propose a piece of code that checks surrounded and cut of units every turn against some kind of factor (Attack vs Defence strength, number of turns surrounded, morale, whatever) and based on the result of the check, the surrounded unit may, or may not, surrender. The surrender may then of course trigger one or more events if that is the correct thing to do in the particular scenario.


I very much like this idea, coz it's a) simple b) generic. It may work in a way that if the check fails surrounded unit level is reduced by 1. If check fails for lev 1 unit - it surrenders. Apart from the factors you mentioned, I'd add city supply value and an air superiority for the hex with the besieged unit (higher supply and ability to reinforce from air would make the siege last longer).




Widell -> RE: How is the AI? (7/30/2008 5:15:54 PM)

quote:

ORIGINAL: comrade
I very much like this idea, coz it's a) simple b) generic. It may work in a way that if the check fails surrounded unit level is reduced by 1. If check fails for lev 1 unit - it surrenders. Apart from the factors you mentioned, I'd add city supply value and an air superiority for the hex with the besieged unit (higher supply and ability to reinforce from air would make the siege last longer).


Agree, and to add to this idea, you may want to store the impact of the different values in one of the files so they are easily moddable later on.




comrade -> RE: How is the AI? (7/30/2008 5:44:42 PM)


quote:

ORIGINAL: Widell

quote:

ORIGINAL: comrade
I very much like this idea, coz it's a) simple b) generic. It may work in a way that if the check fails surrounded unit level is reduced by 1. If check fails for lev 1 unit - it surrenders. Apart from the factors you mentioned, I'd add city supply value and an air superiority for the hex with the besieged unit (higher supply and ability to reinforce from air would make the siege last longer).


Agree, and to add to this idea, you may want to store the impact of the different values in one of the files so they are easily moddable later on.


Like i said in the other thread, everything in this game (except the map, which is a binary file) is stored in csv files, and you can be sure that next patches will stick to this good practice.




DD696 -> RE: How is the AI? (7/30/2008 9:32:09 PM)

My experience with the game so far is that the AI will only defend cities. There appears to be no effort made at establishing a frontline or making use of river/terrain defensive positions (unless they are within one hex of a city). As a long time wargamer since the sixties, this lack of a defensive strategy concerns me. The game appears to be based on city defense/offense AI routines and I feel that this should really be improved upon and addressed.




Widell -> RE: How is the AI? (7/30/2008 10:24:01 PM)

quote:

ORIGINAL: comrade
...you can be sure that next patches will stick to this good practice.


[&o]

DD696: Can you perhaps provide a few examples of games which base their AI routines on other basics? Would be interesting as a reference even if I'm not at all involved in this, or any other game programming.




DD696 -> RE: How is the AI? (7/30/2008 11:31:13 PM)

Take a look at War in Russia or Second Front, War in Europe or any of the old SPI games (or AH games for that matter) - old wargames to be sure, but ones that attempt to create a line of defense that is not based on simply a few points on a map. If only the cities are defended and a great many hexes of open space in between them, then it is all too easy to surround them and defeat them - especially with the new changes the developers have stated they are implementing with the next patch.

I feel fairly confident in stating that in most modern (within the last 100 years or so) military campaigns a contiguous and active line of defense is imperative if you wish to defend against an enemy. Leaving great gaps in a defensive line would be frowned upon by the commander of the theatre. Look at World War I or World War II - you just cannot leave gaping gaps of a great many miles in the line if you wish to survive contact with the enemy.

Or am I simply too realistic?




AZ Gung Ho -> RE: How is the AI? (7/31/2008 12:36:42 AM)

Does the AI do this on all fronts? Or just the Russian front?

I can see a few gaps in the Russian Front, as it was so large that there were many sections lightly held. But if the AI does that all the time, then I think there's a MAJOR problem here the needs to be fixed before I'll buy it!




Widell -> RE: How is the AI? (7/31/2008 12:01:33 PM)

quote:

ORIGINAL: DD696
Take a look at War in Russia or Second Front, War in Europe or any of the old SPI games (or AH games for that matter) - old wargames to be sure, but ones that attempt to create a line of defense that is not based on simply a few points on a map.


OK, let's hope these games are looked into by the design team if/when they take on this task.

quote:

ORIGINAL: DD696
If only the cities are defended and a great many hexes of open space in between them, then it is all too easy to surround them and defeat them - especially with the new changes the developers have stated they are implementing with the next patch.


Without having bought the game yet, I agree, I guess this becomes very obvious in Russia where there's lot's of space without cities to defend for the AI. In more densely populated areas, the issue is maybe smaller?

quote:

ORIGINAL: DD696
I feel fairly confident in stating that in most modern (within the last 100 years or so) military campaigns a contiguous and active line of defense is imperative if you wish to defend against an enemy. Leaving great gaps in a defensive line would be frowned upon by the commander of the theatre. Look at World War I or World War II - you just cannot leave gaping gaps of a great many miles in the line if you wish to survive contact with the enemy.


Well, no one will disagree about what history and reality shows us, but it may be hard to represent that reality in a smart way as implemented via AI programming [;)]

quote:

ORIGINAL: DD696
Or am I simply too realistic?


No, definitely not in your critique of the AI implementation, but the question is if the designers are willing to look into the AI algorithms, and if such changes are possible in a near future? The AI behaviour is a huge source for discussion in any game, but since this one seems to be very heavily tuned to Human vs AI, it is a very important topic for this game indeed.

quote:

ORIGINAL: AZ Gung Ho
But if the AI does that all the time, then I think there's a MAJOR problem here the needs to be fixed before I'll buy it!


Well, technically, it's not a problem if it is WAD, but the implementation of the AI may not meet our expectations of a WWII game... Subtle difference maybe, but important to understand in the perspective of what can and will change in future patches.




db99 -> RE: How is the AI? (7/31/2008 4:07:06 PM)


quote:

ORIGINAL: DD696

My experience with the game so far is that the AI will only defend cities. There appears to be no effort made at establishing a frontline or making use of river/terrain defensive positions (unless they are within one hex of a city).


In addition to this another thing I've noted is that the AI doesn't seem to make any preference based on the VP or PP value of the cities. In taking Yugoslavia I found all the Yugoslav forces clustered around Sarajevo and only a single unit in Beograd and one in Split which are the 2 VP cities you need to get Yugoslavia to surrender. Seems more weight should be given to VP and high PP cities by the AI, in that order.






DD696 -> RE: How is the AI? (7/31/2008 7:17:59 PM)

I stated in another thread that they have a real "diamond in the rough" here - the sheer size of the game, a viable production system, nice use of ground units, and nicely thought out. Having much of the data user modifiable is a huge plus in favor of the game. However, the AI simply does not hold up to the flavor of the operational and strategic flow of World War II. If this can be upgraded so that lines of defense (and offense) become viable then this game will be one of the great ones. Simply battling for the next city with no thought given to gaps in the line is not going to give this game the status that it should deserve.

Of course, if it were adapted for PBEM it could excel without a good AI. The problem with that, for the huge majority of us, is that we are AI players and prefer to spend our time actually playing the game than sitting and waiting, perhaps for days, for a turn to arrive.

All this is for the developers to decide. All I can do is give an answer to "How is the AI?".




Plain Ian -> RE: How is the AI? (7/31/2008 7:37:25 PM)

Well the viable production system would be more viable if there was a production delay built into the system, so you 'ordered' units rather than intantly produced them but I'll admit that saving PP's or whatever achieves much the same effect...although I prefer the former system.

I do however agree that the 'play on my ownsome brigade' may have ruined the game. I suspect that a lot of things done in the game are designed so that progamming an AI will be easier -

- reduced number of units (see comments about OOB's in other threads)
- supply being tied to cities only (is this why the number of cities looks to be low?)
- missing islands (Malta and Cretes port)

Possibly I'm totally wrong here?




comrade -> RE: How is the AI? (8/1/2008 9:15:26 AM)

Thanks for all opinions we're following this thread very carefully. I can assure you that the AI will be constantly improving in next patches.

1.20 (which is coming very soon) among many new features and fixes will have an improved AI garrisoning algorithm (In 1.19 AI can unexpectedly abandon important city)

We're considering implementing more advanced AI (using lines of defence, not only "points of defence") in the following patch, i.e. 1.21, though can't promise for 100% yet. 

There are also some rumors about PBEM [;)]




Page: [1] 2   next >   >>

Valid CSS!




Forum Software © ASPPlayground.NET Advanced Edition 2.4.5 ANSI
3.320313E-02