Jump to content

About CMX2 AI?


legend42
 Share

Recommended Posts

A ways back I remember an article I read somewhere about the next generation ai.I recall that the "next ai" will learn from its own mistakes,possibly building its own file of "hey I better not do that again".Example,ai is dug in,squad in a trench in woods,it decides to charge for no reason and gets slaughtered,that mistake goes into its learn file and will not do it again.Is this true or not and what kind of jump in caliber can we expect from the next ai?

Link to comment
Share on other sites

Originally posted by legend42:

A ways back I remember an article I read somewhere about the next generation ai.I recall that the "next ai" will learn from its own mistakes,possibly building its own file of "hey I better not do that again".Example,ai is dug in,squad in a trench in woods,it decides to charge for no reason and gets slaughtered,that mistake goes into its learn file and will not do it again.Is this true or not and what kind of jump in caliber can we expect from the next ai?

So if I buy the game six months after you, the Tac AI on your computer will be doing things better when your computer calculates the PBEM files, and worse when mine is calculating the files?

Seems unlikely...

Link to comment
Share on other sites

My impression is that the AI in CMx2 won't be a 'learning AI'. That would be too involved and probably not practical for a computer game. It's not like the computer could take lessons from every game or situation it played and apply them to the next game. That would take too much memory and would be horribly complex to program for something as variable as CM is.

I believe that the major difference in CMx2's AI will be that units will have a 'memory' of what happened in previous turns to build decisions on (for just that game, rather than a database of 'mistakes' and other data carried from game to game). Currently CMx1 doesn't do this and units have to react to what they see and experience in the current turn... so they'll forget that a Tiger just hid behind some buildings unless it moves again and takes a shot at them during that turn. CMx1 attempted to fix this problem in an user controlled manner with 'covered arcs' (which also served other purposes). With CMx2's AI, hopefully, an unit that just broke will remember that they received fire from the direction of the woods to their right and avoid retreating into that particular line of fire again, etc.

Plus there'll need to be a lot of changes to accomodate whatever method of relative spotting that Charles and Steve decide on.

[ January 10, 2005, 08:15 AM: Message edited by: Schrullenhaft ]

Link to comment
Share on other sites

More practical type of 'learning' AI is that where the AI has a bunch of different basic strategies on its plate, and then by trying them out it learns which work on forested maps with armour-heavy troops, which work better on plains maps, which are preferable for infantry in cities etc. But making it truly capable of learning is impossible, as it can't understand the tactics it is programmed to use.

Link to comment
Share on other sites

Originally posted by Sergei:

More practical type of 'learning' AI is that where the AI has a bunch of different basic strategies on its plate, and then by trying them out it learns which work on forested maps with armour-heavy troops, which work better on plains maps, which are preferable for infantry in cities etc. But making it truly capable of learning is impossible, as it can't understand the tactics it is programmed to use.

Why wouldn't you just preprogram such an AI before shipping it to the customer?

If it takes learning, you could simply run the simulation in the programmer's computer 10 billion times, let it learn what it can, then ship it...

I suspect I am missing something here?

What if you wanted the Tac AI to simulate a "green" opponent - not at the individual squad level, but at the level of the company and battalion commanders giving the orders? Wouldn't you need to preprogram the AI in order to be able to simulate at will this kind of quality differentiation?

I mean, once the AI has "taught" itself to be a Veteran commander, could it go back and pretend to be "green"?

Link to comment
Share on other sites

Back in the what do we want thread part II, when we discussed Hoolaman idea of command zones, there were talks about having a feature to let scenario designers plan certain moves for the computer, a feature that could be disabled by the player. It could work in a number of ways, from assigning certain priority/threat/objectives zones within some sort of AI on-map graphic interface to actually giving orders to units. While this option is obviously less than perfect and not exactly related to AI core improvment, IMO it does provide some alternate approaches to this problem.

First it would provide far more elaborate plans for the computer, at least in the initial stages of a battle. The replayability would not be much reduced since this particular feature is but an enhancement of the "stick to scenario default" setup setting. Supposing that this "plan" feature is imbeded in a scenario in a modular way, there could be several "plans" for a given scenario, adding the fun of comparing different views on a specific problem and replayability.

It could also let historical scenario designers implement realistic plans and tactics based on accounts, reports and various archives, as it would be for people sharing ideas about various doctrine topics.

Since no one knows for sure were CMx2 is going, this is more than probably irrelevant, but still, I though the idea worth exploring.

Cheers

Link to comment
Share on other sites

this is STILL a GREAT IDEA

and I hope we see something like it in CMx2......

"Hoolaman

Member

Member # 12153

posted November 28, 2004 05:36 PM                         

------------------------------------------------------------------------

It would be good if scenario designers could use SOPs, designate areas of advance and lay out a few different options for the AI to produce more realistic behaviour before the AI calculations even come into it.

It is possible to prod and coax the AI to attack through sensible routes now by making trails of flags or using no flags or putting reinforcements in strange unexpected places. If this could be factored in to the game properly through the behaviour of AI, the work would be half done.

If you think about how a human plans an attack, firstly they will identify where the enemy will be likely to deploy. Then they will split their forces to advance through planned routes, to deal with the areas that he hopes the enemy might be in.

If they see enemy units, they will try to assign them to a likely formation, and note that formation on the map. If the AI responds to all these factors, and identifies enemy units as "blocks" just like a human, then that might help simplify tactics."

Link to comment
Share on other sites

I recall there was discussion (okay, maybe it was mostly by me) of something equivalent to 'invisible flags' during initial scenario design, a concept pretty close to Tarkus's command zones. If you've got a bridge in the scenario I suppose you might be able to plant either a 'red crosses bridge' or 'blue defends bridge' flag at scenario design stage depending on your needs. Maybe you'll be able to designate a flag's zone of influence much like cover arcs now. I guess a 'red avoids this area' flag could be used to keep AI troop from walking into obvious ambush territory or keep them off the previously mentioned bridge. Behavior modifier flags wouldn't be as unbreakable a rule as scripting but it would help the designer shape the battlefield.

I don't have a clue about how the new AI is really going to work, but this idea had been brought up some time ago. Maybe somebody was listening tongue.gif

Link to comment
Share on other sites

Originally posted by MikeyD:

...I recall there was discussion of something equivalent to 'invisible flags' during initial scenario design...

That would be a big help for the AI. Instead of getting up (leaving a perfectly good defensive or tactical position) and walking around, they could stay put and guard their assigned area (without the visible give-away flag to blow their cover).

Ken

Link to comment
Share on other sites

I am hoping there will be something FAR more ingenious and "satisfactory" than simply "invisible flags" to make this happen.

We are talking about CMx2 here, and seeing as it is a WHOLE new engine written from the ground up and keeping in mind that, as has been mentioned, units will have "memory" of past events (yes no? is that a done deal?) within the framework of a single game, I am truly hoping that some form of SOP orders can be issued to specific units.

I am confident that the good and kind folks at BFC (some of them actually being scenario designers them selves) WILL see to it that the defender's options for setting up a scenario and designing a scenario will give the AI in the ALL the oppotunities and advantages possible by allowing as much latitude as technically possible in the programing of the units and "terrain" with SOP's of the mission or scenario, so that a wise and shrewd scenario designer will have ALL the tools/options needed to "HINT" the AI in the direction or "mindset" (if possible) intended by the designer of the scenario. I am thinking here specifically of those defend scenarios BUT all the same tools could be employed for a GREAT attack or assault scenario as well.

I am guessing that the NEW engine will have something FAR superior to invisible flags to let scenario designers invest LOTS of time and energy into creating GREAT and wonderful challenging scenarios that will let the AI in the game live up to its FULL potential when is it appropriately "hinted" or programed (units with SOPS to recon or defend or hold at all costs or RUN and retreat at the first shot, off the top of my head) by the scenario designer.

Maybe I have my hopes up way TOO high BUT I know they have said they like to make games that THEY actually like to play and if programing units with SOP orders so as to enable the FULL potential of the AI is something they think they (the Designers Steve and Charles et. al.) would like to play with then I hope we can look forward to something truly Revolutionary, ground breaking and innovative!

Let's see what they come up with a year from how....

-tom w

Originally posted by kenfedoroff:

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

...I recall there was discussion of something equivalent to 'invisible flags' during initial scenario design...

That would be a big help for the AI. Instead of getting up (leaving a perfectly good defensive or tactical position) and walking around, they could stay put and guard their assigned area (without the visible give-away flag to blow their cover).

Ken </font>

Link to comment
Share on other sites

I think the whole point is to somewhat "connect" AI with scenario designer so to benefit from human intelligence in some aspect of the COA the AI uses, which is not always impressive. It is still the best around that I know, but it will strangely rotate a Panther on top of a crest under the mount of spotted IS-2s every now and then...

"Invisible flags" isn't bad at all IMO, since it would be a workable interface for scenario designers. I mean you can have the best AI there is, if suggesting COAs to it is too complicated, we wont be able maximize the "benefits" (read: having our butt handed to us real good, as the saying goes) from such a good AI.

And it need not being "flags". Like MikeyD put it, you could paint areas, buildings, structures, draw arcs, whatever fits, so long as the designer can assign more accurate objectives, course of actions, deployments and moves.

That being said, now that BFC has clearly stated they are on it, I feel like shuting it up tight and wait for what *they* think about these things. Bones will come our way eventually, the sooner the better.

Best.

Link to comment
Share on other sites

hi all,

one of the most important things cmx2 must have is an option to direct a hole compagny of division

or group by one set of orders not having to buy a new mouse after each turn (from klicking a million times)

I like big scenario's so you got a lot of people(playing pied peiper 40000pts or higher )you get frustrated by klicking every single squad and tank to make them move in the same direction.

This is very needed while the most you want to advance is over roads. otherwise your hole division is bugged in snow or mud (imho)

So plz can someone add that

When i give an order to the tankcommander the rest will automaticly take the same order (until i want them to take a different route then i must manually give that tank another order.

I hope the designers know what i mean otherwise plz play the pied peiper after a few turns you know what i mean.

maus

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

 Share

×
×
  • Create New...