CMSF AI - Your opinions please?

[Bob_the_Fish]

Ok I'm a big CMAK fan and have mostly been playing PBEM.

I've downloaded the CMSF demo, and I like it! For me it's retained the essence of CM but with exciting new weapons.

My only gripe is the AI is seems rather scatty. I'm playing the demon mission "going to town" as red team. The blue team can't even mount an attack, and seems to be enjoying driving up and down the highway.

Please tell me that you can have a good playing experience versus the AI in the full game. Is is capable of mouting a serious attack?

[SgtMuhammed]

The AI in CMSF is much more dependent on the scenario designer. It also has multiple plans. You might try another run through to see if the AI does anything different. I can't remember how many plans eash side is supposed to have in the demo.

[Lethaface]

Yes. All depends on the scenario creator in this case.

[Paper Tiger]

Bob the fish

Please tell me that you can have a good playing experience versus the AI in the full game. Is is capable of mouting a serious attack?

The answer is a resounding 'yes' to both questions. I haven't played the demo scenarios because I only downloaded it to see if it worked okay with Vista. Once it powered up and everything seemed fine I bought the game the same day. I wouldn't expect too much from the demo.

If you buy the full game, be sure to visit CMMODS and download some of the scenarios there. There are a number of them with good AI attacks.

[Michael Dorosh]

Well, I'm a scenario designer, and while I agree that the AI as designed can only operate up to maximum in the hands of a competent scenario designer, that maximum is still not comparable to the AI designed for CMX1 - in my opinion.

Which isn't to say it is worse, or better, just not comparable. They are apples and oranges.

The AI in CMX2 is scripted; you can write multiple scripts, but at the end of the day, all forces are reliant on what they are told to do before the first shot is fired or the first move plotted. Individual scenarios will work better with this system than others, and individual players will have differing thresholds as far as willing suspension of disbelief.

In neither case - CM:AK or CM:SF, for example - will the AI use fire and movement effectively on the attack, for example, though in the latter it can be partially programmed in a most general, scripted sense to do so. In CM:SF, there are no "triggers" and at no time will the AI respond to actual events on the battlefield rather than the timed sequence of events preplotted before the game starts. The Tac AI, however, will carry out limited tactical self-preservation moves in either game.

[the Fighting Seabee]

Yes, I agree. The ai has its limitations, but the good scenarios have good ai plans that will challenge you. I bought the game not long after it came out and I still play. If you're into this sort of thing, I highly recommend the full game. There is simply no other that can match it.

[Darkmath]

I've just played a WEGO meeting engagement map in full version (Trident Valley) as Syrian.

Well, while I'm not a good player and it's hard to kill bradley with BMP1, the AI had a total victory!

As for the AI one thing amazed me : when one of my squad shoot at a bradley with RPG7, the Bradley moved by its own around the squad to spot it ! And it was in WEGO mode.

[Darkmath]

I should have added that AI planning as it is in CMX2 allows more complex plan than in CMX1. The planning design fits well for an AI defender.

Therefore, a defending AI could act better in CMX2 than in CMX1, except for artillery handling.

As for an AI attacker, time driven design reaches some limitations that there weren't in CMX1 dynamic AI.

However, CMX1 AI isn't that perfect : there was some patterns used by AI such as rushing to the flag at the last turns or HQs in front of an attacking wave, or artillery support on the first enemy spotted or on the flag only...

Time driven designing however could be compensated by making multiple plans, so the AI randomly initially choose between these different and complex plans.

It offers a better game replayability.

Obviously, time driven AI in CMX2 performs less succesfully in QBs than CMX1, but here you always can do multiple plans for QBs maps. Better than nothing.

Overall, AI in CMSF could do few but sophisticated plans, while AI in CMX1 is more dynamic but more rudimentary.

But I avoid playing against an attacking AI since CMX1 (except for Meeting engagement), so I would say CMSF AI is better as defender.

[Secondbrooks]

AI in CMSF can't handle change, it has zero iniative to player's moves. And designing and testing AI plans currently is too cubersome task. It would be nice to be able to see how my plan works out without playing whole scenario. Simpliest would be to be able to watch AI'plans fulfilled on map in accelerated time for each side (no fighting or anything, units just moving roughly like they would move and take their positions in game). I could see bugs in my plans much more easily and in faster time.

But even with this tool AI is still too predictable, it can't function without orders, it has zero iniative (if squad B was assigned to seize objective #3, but they got killed on their way, then nobody really cares). Hopefully BFC will add dynamicthings to AI-plans.

Infact i would say that AI is better on attack It has the iniative and player has to react to that. AI on defence, then terrain has to be such that AI's keyweapons can affect to player's units in almost all situations. Player has iniative and AI should react to that, but sadly right now it can't react but with shooting.

[ March 02, 2008, 11:38 AM: Message edited by: Secondbrooks ]

[Darkmath]

I would like to know how CMX1 AI works, how it reacts to player's move on the fly .

What could the AI in CMX1 do that the AI in CMSF could not, in specifical situations?

There is a theoritical difference, but don't CMX1 AI would follow some patterns just like a script for CMSF?

For example, AI in CMX1 systematically counter attack to retake a flag then sometime get unecessarily slaughtered.

I remember a CMBO scenario in which several german squads were rushing to a house one after the another through the bocage (and makes my US squad get 30 kills. )

(I haven't played CMX1 for a while, so I can hardly compare )

[ March 02, 2008, 01:47 PM: Message edited by: Darkmath ]

[Londoner]

Originally posted by Secondbrooks:

designing and testing AI plans currently is too cubersome task.

Find a play tester or two sir!

[dan/california]

Steve has said that reactive/conditional AI orders are high on their list now that the basic game mechanics seem functional. Which I agree that the basics are working quite well now.

People are down to complaining about weapon effects and accuracy. That is a very good sign.

[dan/california]

The ability to have the AI plans run both sides would be a godsend for scenario builders but I have the general impression that the coding involved is significant.

[Paper Tiger]

Darkmath:

"I would like to know how CMX1 AI works, how it reacts to player's move on the fly .

What could the AI in CMX1 do that the AI in CMSF could not, in specifical situations?"

I'm no expert on CMx1 AI behaviour (or CMx2 for that matter) but here's what I learned about it in CMx1. It appeared to be entirely driven by those VP flags. On the attack, the AI would advance towards the closest one first so a scenario designer could set it up so that they attacked location A first and then flag B afterwards etc, or he could place the flags so that the AI would attack both locations simultaneously etc. Since the TAC AI functioned quite realistically at that level of abstraction it certainly looked to be better than CMx2 AI.

As long as there was at least one flag on the board that was neutral or enemy controlled, the AI would DO something. When the AI controlled the lot, it pretty much stopped in the vicinity of the flag. When flags changed from friendly controlled to enemy controlled, other AI forces would then mount a counterattack, usually at the expense of defending their own VP location. However, this kind of behaviour is not unrealistic in terms of WW2 battles.

With CMx2 the AI will do absolutely nothing without an AI plan even when there are lots of VP locations being controlled by the human player. The VP locations mean nothing to the AI at all, neither do casualties or indeed anything else except it's current order. And it will plough on towards that objective or die in the process. However, a scenario designer can at least try and anticipate some of the player's reactions and make the AI at least appear to react or move from an attacking stance to a defensive stance etc.

The real problems the AI has, without reactive scripting anyway, is that there are too few orders available for the scenario designer to use. Therefore, most of the AI attacks will just be 'move to this location in x minutes or die trying'. With more orders, it would be possible to make this non-reactive AI system at least appear a bit more reactive.

[Darkmath]

Bump for this thread , just to draw Steve's attention to STratAI (at least I would have tried).

Will changes/features (such as triggers) be included in the future patches, or later in the next releases?

And now, to reply to PaperTiger:

I think AI in CMX1 was programmed by putting VL on the map.

But the agressive counter attacking of CMX1 AI defender might be unrealistic , ie the rest of the units rushing to the lost VL out from their foxholes.

Also, while it's hard to time right a retreat of the defender in CMSF, it was simply impossible to do that in CMX1 only with VLs.

[Secondbrooks]

Originally posted by Londoner:

</font><blockquote>quote:</font><hr />Originally posted by Secondbrooks:

designing and testing AI plans currently is too cubersome task.

Find a play tester or two sir! </font>

[Paper Tiger]

secondbrooks. In answer to your questions...

1. For CMBB I designed and played my own scenarios but never released them as I didn't think they were good enough.

2. When I'm not working on my own creations I much prefer to play the community produced scenarios. I don't know why but I rarely play the stock ones that came with the game although I did like Ceorge Mc's "Al Amarah". And I liked 'Al Huqf engagement' too. I HATED 'ATGM ambush' with a passion though.

3. So far, only one called 'In Harms's Way' but that was a LONG time ago 1.04 I think. However, if you include campaigns, my current effort contains 10 scenarios which is quite a substantial effort.

[kevinkins]

For many years designers have been asking for scripted AI. Now we have it for CMSF. I guess some other wargames have it but I can't think of any. None of the ones I owned over the year did. It is a great development. That said, it is time consuming and many designers have little time to actually play the game. Some accelerated mode would be a big help. Even the old AI vs AI mode would help - that mode was in many games in the past.

[ March 09, 2008, 07:17 AM: Message edited by: Kevin Kinscherff ]

[thewood]

Most of the scripted AIs I have seen (TOW comes to mind) are extremely dependent on the scripting. CMSF I think has a good balance of scripting and organic AI. In TOW, it seems if you don't explicitly say to do soemthing, the AI won't do it.

CMSF now seems to have a decent TacAI. The only thing lacking right now is more tools in the AI plan beyond times scripts.

[kevinkins]

I still think the time for a designer to test the draft scenario could be shortened. I am all for more tools of any kind.

Kevin

[Omenowl]

Warcraft 3 and I believe Starcraft had the ability for scripted AI. Seeing what some people have done with WC3 I am very impressed. Battletanks being an ideal mod with fairly good AI (in most cases superior to inexperienced human players).

This is one reason I look forward to triggers. It requires a lot more coding, but the ability for the AI to suddenly change its attack plan or to retreat and rearm and then counter attack would be nice. Also suddenly have an unexpected force appear at an inopportune time, but not magically appear next to your units because it was trigger based on enemy placement rather than purely time.

Also something along the lines of the bridge over remagen where depending on the length of the scenario when your players reach the bridge determines if explosives go off and whether you succeed in your objective.