The AI is much Improved

[troopie]

I'm playing generated campaign, Philippines vs Japan. The Japanese have just handed me my head! I haven't been so totally crushed in SP since the US-USSR campaign in the unpatched SP2.

troopie

[Recon_slith]

I've noticed a tendency twords the AI trying to sneak small units along the board edge trying for some cheap rearwards victory hexes. SPWW2 does this too.
AI controlled artillery can't be baited into firing as easily as it used too either.
At least you can usually spot the Japanese infantry in the open in this game- in SPWW2 they are practically invisible.


------------------
Wait for Death. There's a choice?
Recon

[Charles22]

Let me contradict that a moment. Firstly, I'm fighting as Gerry against the Poles with Gerry ordered to advance, with the Poles holding all the objectives.

I sent a recon team with accompying infantry to the foremost southern hill behind enemy lines. It was unoccupied.

I also sent a flanking team of 3 PZIB's north, along with 1 PZIIIE and 1 PZ38T. There was a concentrated effort to retake the southern area with tanks, and at a certain portion of this battle, a good deal of the Polish Army went berserk to try to recapture this. With their 19 man squads, it's a challenge I'm still dealing with.

The northern flank team has ran into a very small tank team, which isn't apparent to me whether they were sent there to wipe out my team or to infiltrate my lines. In any case, I've yet to see a verifiable instance, in playing against the Poles, the three or four battles I've played to date, where they will actually make for my rear. All I can say, is if my northern flank team had run against a team of the best Polish tanks, or indeed some T34's in Russia, my flank team would have been dust.

[Larry Holt]

All in all the AIP (AI player) is much improved. In previous versions, an AI tank would just sit out in the open turn after turn with its rear facing me. There did not seem to be any provisions for "If I can't figure out what to do either advance or hide." The AIP did some neat path finding on me. In one scenario it attacked on a broad front then the next scenario, it massed forces on the southern half of the board.

HOWEVER, it still does dumb things like leave numerous (10 - 15) AFV in its rear area just sitting around, not advancing, not guarding arty, just sitting there!?

------------------
An old soldier but not yet a faded one.

[Charles22]

I have to agree. When the computer will sometimes attack only portions of the screen, and at others the entire front, it certainly makes at least the first few turns quite dramatic.

[Wild Bill]

There have been definite improvements in the AI in many areas. I'm glad it is noticed. Michael Wood will be happy to see that too.

Wild Bill

------------------
Wild Bill Wilder
Coordinator, Scenario Design
Matrix Games

[Wild Bill]

Troopie, did you get that special send I pushed off your way last night?...WB

------------------
Wild Bill Wilder
Coordinator, Scenario Design
Matrix Games

[troopie]

Yes I did, Bill.

troopie

[Arralen]

Originally posted by Wild Bill:
There have been definite improvements in the AI in many areas. I'm glad it is noticed. Michael Wood will be happy to see that too.

Wild Bill

Sorry, but I played SP for DOS extensivly those days and I don't find the AI in SP:W@W very much improved, sorry.

It seems to me that it still

- heads blindly for the victory hexes,
so that you can easily tell the enemy approach path

- runs into "death trap hex" with 3..5 units one after another instead of going 'round it only one hex left or right

- stills deploys it's units in a seemingly most random manner

- is unable to concentrate forces on a weak spot

- send units wading through shallow water with a bridge only 1 hex to the left

... to be continued ...

Arralen

PS: don't get me wrong - I'll stick to this game as long as someone says "o.k., I'll work on a improvement" ... and I hope this will be for a long time to come, as I feel the game deserves it.

[This message has been edited by Arralen (edited 05-25-2000).]

[This message has been edited by Arralen (edited 05-25-2000).]

[Larry Holt]

[QUOTE]Originally posted by Arralen:
Sorry, but I played SP for DOS extensivly those days and I don't find the AI in SP:W@W very much improved, sorry.

It seems to me that it still

- heads blindly for the victory hexes,
so that you can easily tell the enemy approach path
- runs into "death trap hex" with 3..5 units one after another instead of going 'round it only one hex left or right
- stills deploys it's units in a seemingly most random manner
- is unable to concentrate forces on a weak spot
[snip]
Arralen
[snip]
I haven't seen this but I've only played a few scenarios yet. If I may ramble on about AI and not insult Michael Wood, it seems to me that most of this is path-finding related. The AIP needs to "see" the battlefield in terms of where to move forces not only to arrive at the victory hexes but also to be in a position to out maneuver the human player. (a program can only see the battlefield in terms of a look up table defining the current status of tiles, terrain type, units located there, etc.). But what is out maneuver? I could briefly define it to mean having more combat power (i.e. accurate, killing shots) than the human does. Choosing a path to reach the victory hexes is not all that hard because the map is fixed and the movement points are known. Choosing a path to avoid blundering into the human's kill zones is more difficult as this is not known ahead of time. I don't know anything about the SPWAW AI code but it seems to me that one way to influence the path finding is to adjust the hex movement points based on combat. If an AIP unit is shot at or killed, add more movement points to those and surrounding hexes and the AIP will calculate a path that avoids those areas.

But its not all path related. Should the AIP advance or defend? Its usually advantageous to maneuver offensively but fight defensively. The AIP could keep track of the kill ratio and hit ratio vs. the human. If the ratio favors the human, the AIP will halt and shoot back as it is losing. If the ratio favors the AIP, it moves forward as it is gaining fire superiority and can move with less risk of having units killed.

While programing an AI is difficult as it can not make the leaps of awareness that humans can. IRL, troops employ predefined battle drills in pre-determined situations. Humans don't make up what they will do from scratch in each tactial situation. It seems that some of the flaws quoted above can have battle drills programed in. For example...

It should also be possible to program in some minor tactics like when a formation (platoon/section) is to cross an open area larger than some predefined size, it will not do so until another formation is positioned on the edge of the open area to provide overwatch. As the AIP does its pathfinding, it will know that there is an open area on the path that matches this criteria and start to move an overwatch formation into place before the moving formation reaches the open area. Basically this is bounding overwatch.

OK, its a lot more complicated than just this (how does the AIP determine which forces will bound and which will overwatch?) but this post is too long as it is.

------------------
An old soldier but not yet a faded one.

[Paul Vebber]

We discussed early on major changes in the AI. Problem is AI programming takes time. Lots of time. And it takes a "system" that is built with a certain AI in mind.

To make the AI "minimally smart" about things would have taken a solid year of programming - we did some improvements to what was there, but it takes essentially redoing the game from scratch to add the "hooks" to let the units keep track of "triggers" that can cause changes in unit behavior.

Essentially we could be probably halfway to producing a "miminally competant" AI now, with NO other changes.

We felt most people would appreciate other changes more, so "improved" rather than "overhauled" the AI.

To us, internet play cpability and a "live' opponant list to help you find a "human AI" when you need it was preferable to taking a year or moe to produce one that folks would still complain about <BG>

[Arralen]

So you're telling me that you neither can't nor are willing ot improve the AI substantially ??

:...-(((((

He, I fire this game up when I have no-one to play with - and a lots of other player do so, too - think of all those who said they love the long campaigns !!

And than there's a big drawback to PBEM - as someone already pointed out - you cannot watch your opponents full turn but only your own firing phases ... so IMHO only hotseat make really sense .. .

But if I'm looking for a "life" game against another player, I mostly prefer board games

So PLEASE, PLEASE do something about the AI ...

Arralen

[kfbaker]

I would say there is no doubt that the AI is far better than the original SP.

I think one major issue in the old game,(I have not yet played enougth to see if this is the case in the new one), is that while we develope crew skills and expererince with the resultants improoved battle skills, the computer AI player does not. The result is that the game handicapps itself, for while the AI can usualy give you a god fight early on, it just can not compete on a level with your elite crews.

In order to address this I thing the level of the enemy crews should be pre set for each campaign battle witha small random factor to reflect historical capabilties. Secondly the odd elite and green crew also being injected.

This would then probably even up the playing field, as I feel it is simply that as the campaign progressers we are playing with more and more of a loaded deck in our favour.

[This message has been edited by kfbaker (edited 05-25-2000).]

[Larry Holt]

Originally posted by kfbaker:
I would say there is no doubt that the AI is far better than the original SP.

Agreed!! I don't see the AIP going into "shock" and just letting its units sit in the open, rear facing me to be shot up as in SP1.

{B]I think one major issue in the old game,(I have not yet played enougth to see if this is the case in the new one), is that while we develope crew skills and expererince with the resultants improoved battle skills, the computer AI player does not.[snip][/B]

Disagree. Regardless of the quality level of its crews, the AIP still can not mold them into a good fighing unit using tactics the way a human can. Of course I realize the limitations placed on an AI given a PC's resources. I think the AIP is worthwhile fighting and I mean no critism of the outstanding job Matrix has done in giving this to us for free, I just mean to honestly critique its strengths and shortcomings. I've taught tactics and I know how hard it is to get humans (with real intelligence) to gather and analyze all the data, making connections between current enemy locations, relative combat power, intervisibilities, absolute and relative movement rates, etc. required to make a decision as what to do. I don't know if the AI uses expert rules or true, adaptive artifical intelligence but either way I wish it could handle basic tactical situations (like don't drive into a smoke hex when the last 4 units that did died there) better. I read Paul V's post above and understand the limits as to what can be done in what time (& without pay) but perhaps Matrix can consider this when they have the oppertunity to develop an AI from the ground up.

In the meantime, great job Matrix!!

------------------
An old soldier but not yet a faded one.

[Paul Vebber]

Well said Larry! AI is always going to be a problem. Take Combat Mission as a ferinstance. They have been in development for several years, with a tightly integrated AI that works on several different levels and seems about the best I've seen in a tactical game.

Now go check the usenet.

They are STILL getting reamed by folks for whom its not good enough...

As we prioritze the development schedules of our games, AI will figure differently in each one. Frank Hunter is working some VERY innovative AI concepts into Wars of Napoleon, being a multiplayer game, there is a need for really good AI to play the non-human sides! The upcoming Typhoon operational level game will likely not spend as much time on AI as its concept is more as a 2 player human vs human emphasis. If the game design is a success, then AI improvement can occur from game to game in the series.

You have to weight the tradeoffs and make tough decisions or every game takes years and years to finish and you can't survive as a business! To accumulate the resources to support a "Magnum Opus" you have to have something to keep the cash flow positive.

Matrix intends to offer a wide variety of types of games with a variety of design choices. Its a very dynamic niche and expect things will chages, title's release dates will flop back and forth, delays are inevitable, etc., so hang on and enjoy the ride!

I know we are!

[Dave R]

I think that there is one thing that we need to remember here guys. The SP engine is now quite old tech. No wI don't know didly squat about coding, but I'd guess that there's only so much tweeking that can be done to the original game engine.
I for one will be glad to see the patch to take out the worst of the bugs. But after that I'd rather throw my support into 'doing my bit' to support the Guys at Matrix develope a 'Beyond Steel Panthers' game. Then perhaps we can get the AI routines that we'd like to see

[Paul Vebber]

Appreciate the sentiment! Just tempering enthusiasm aboout what's in the realm of the possible - even in a "new game".

Chalres Moylan posted an EXCELLENT usenet article on AI building in the "war-historical" group that deserves searching out on Deja and taking a gander at.

The whole thread there is why game designers get depressed when players start talking about their wants for "AI"...

We've about 10 more years to go to get there. Hopefully we'll still be around and we'll have what you are looking for by then