CMBB-What it REALLY needs!(better Strat AI)

[aka_tom_w]

good thread but it needs to have AI in the title other wise it is likely to be over looked my most folks here

-tom w

[Captain Wacky]

How bout seeing a couple of these babies roll across your screen at full speed?

BTS hire that man!

[Panzer Leader]

A Panzergrau Panther F?? They are making CMBB not CM'46

[chris talpas]

Actually one group that has generally produced games with superior AI is the boys from down under at Strategic Studies Group (SSG).

Although not the simplest to use, their "War Room" AI editor for their Carrier's At War series probably has not seen its match since.

Perhaps Steve and Charles might consider subcontracting the strategic AI to these boys for CM2.

Edit:I realize this post is about CMBB, but I was thinking further ahead to the game engine rewrite since redoing the StratAI would probably be no mean task

Chris

[ April 06, 2002, 11:53 PM: Message edited by: chris talpas ]

[Fred]

yes, SSG had some good AI routines, but I think that it is possible to enhance the ones in CM (look what they did from version 1.0 to 1.12!).

It takes time, sure, but it would be worth the effort.

Fred

[Smiler]

Originally posted by Fred:

The argument, 'that it is not possible to make a good AI with today techniques' does not hit. Just take a look at the already implemented TAC AI!

Some other folks have already mentioned some of the things I was going to say here, but TAC AI and strat AI are two completely different things!

I mean, look at chess. Chess is a much simpler game than CMBO, requires no initial strategy (there's no setup), and has had many more man-hours of work on it. Yet, my understanding is that chess programs are still largely tactical (chess grogs feel free to correct me). The strategy is derived by using existing strategy (opening book moves), not internally generated on the fly (which is required in a variable-setup game like CMBO). Computers are sufficiently better at tactics that in games that rely on tactics enough, computers can win. Chess appears to be right at the edge, as the best human chess players can still use superior strategy to compete with the better computer tactics.

And if all programmers stop working on AI enhancements, because 'it will not be better', well, then, indeed, nothing will ever happen, like a self-fulfilling prophecy.

Fred

Oh I agree whole heartedley here. Being a naturally surly and cynical person (as my handle implies), I'd just as soon play vs. the AI instead of real, annoying humans (current PBEM opps. excluded, of course ). So I'd like to see the AI improved, too. Sometimes, I just think people underestimate how hard strategic AI is.

[xerxes]

I'd like to see the addition of A.I. "markers" for scenario design. By this I mean, the ability to put markers on the map that the A.I. uses but the human never sees.

Two basic types of markers, attract and repel. An attract marker gives the area a high interest for the a.i. (like a flag), the repel does just the opposite.

Optionally the flag could be set to infantry, armor, spotter or all. It would only apply to that class of units.

Using markers, the designer could basically give high level (strategic) guidance to the a.i..

Replaying a scenario that is marked wouldn't work too well though.

-marc s

[Smiler]

Originally posted by xerxes:

I'd like to see the addition of A.I. "markers" for scenario design.

...

Replaying a scenario that is marked wouldn't work too well though.

-marc s

Sounds like a good (and workable) idea to me. As for replay, well, replay is always a bit less enjoyable than the first time though anyway. And if you could turn off the markers' effect, you could still get variable (though probably less effective) outcomes.

[Michael Dorosh]

Originally posted by xerxes:

I'd like to see the addition of A.I. "markers" for scenario design. By this I mean, the ability to put markers on the map that the A.I. uses but the human never sees.

Two basic types of markers, attract and repel. An attract marker gives the area a high interest for the a.i. (like a flag), the repel does just the opposite.

Optionally the flag could be set to infantry, armor, spotter or all. It would only apply to that class of units.

Using markers, the designer could basically give high level (strategic) guidance to the a.i..

Replaying a scenario that is marked wouldn't work too well though.

-marc s

Program it with variable markers. Give the computer 2 or 3 or more choices of which markers or sets of markers it will use. One playthrough, the computer might opt to take the high ground; the next, they roll on the crossroads.

[aka_tom_w]

Originally posted by xerxes:

I'd like to see the addition of A.I. "markers" for scenario design. By this I mean, the ability to put markers on the map that the A.I. uses but the human never sees.

Two basic types of markers, attract and repel. An attract marker gives the area a high interest for the a.i. (like a flag), the repel does just the opposite.

Optionally the flag could be set to infantry, armor, spotter or all. It would only apply to that class of units.

Using markers, the designer could basically give high level (strategic) guidance to the a.i..

Replaying a scenario that is marked wouldn't work too well though.

-marc s

this sounds sort of like matter and anti-matter

just joking

it would be VERY nice if a scenario designer could give some hints or clues to the AI in this regard.

The idea that these markers be placed by the designer in such a way as to have them invisible to the player makes a GREAT deal of sense.

Helping the AI along in this way can also be accomplised already by setting up the AI in a defensive scenario in good well thought out defensive positions and a requesting that the player stict to scenario default positions. Perhaps a similiar arrangement could be made for meeting engagements and attack scenario's with the use of insivible attract and repel markers that the designer could set up ALONG with a time line as to at what turns in the game these invisible attract and repel markers would be effective or "active" for the AI

NOW that would be COOL!

TOO late for CMBB I guess

oh well interesting idea none the less....

-tom w

[aka_tom_w]

any comments?

-tom w

[Slappy]

I have one.

I do think that markers as described, or other triggers could make scenario design much more flexible and rewarding. As it is, you can do this to some extent with placement of VLs and setup zones. You can use setup zones to force the AI into a layered defence rather than a one line or set them up to attack from different angles by placing setup zones closer to the defending lines and in less of a line than traditional.

You can also 'encourage' certain lines of attack by placing intermediate VLs and then adjusting the victory percents to 'erase' any extra points given to the AI from holding those points. The balance is tough, but you can do it.

The problem is that all of these are essentially what has been refered to in this post as 'cheating'. They make a solo scenario much more playable and exciting, but you have still 'told' the AI how to execute the attack.

Example- a non WWII Example, forgive me: You want to have the AI play the CSA on Day 2 of the Battle of Gettysburg. The action on this day in reality was all around Little Big Top. The Confederates realized that if they could take this hill and get artillery on top, they could make the entire Union position untennable. You set up the battle, and the AI just keeps slamming into the front lines taking the shortest path to the VLs. Essentially, they play day 3 of the battle (Pickett's Charge) on day 2. In order to correct this, you put a big VL on top of Little Roundtop. The AI now sends units out that way to try to take that hill and things go a bit more like you wanted them to. The problem is that the AI didn't do anything different, didn't 'figure it out'. It still has no concept of taking that ground in order to rain fire down on an adjacent valley, it's just taking a VL. In other situations, QBs come to mind, the AI will always fail to take the flank or the hill, or whatever unless a human puts a VL on the hill. What you have is better scenario design, not better AI.

Better scenarios are always a great outcome. I especially like a well designed v. the AI scenario given that I rarely have the time to commit to playing a human, but I don't confuse a good scenario with a better AI which should be able to handle a much wider variety of map and setup types creatively and flexibly.

[SlowMotion]

This idea that scenario designers could place markers/flags to help the AI play more intelligently has been discussed many times in those CMxx Wish List threads. I think the AI could be made a better opponent in battles where it plays ok at the moment. It could also be possible to define some tasks where the AI is doing very badly in CMBO, like one group trying to protect another retreating group. I'm too lazy to do a search right now to check what has been written before, but here are three things that could be useful:

1.

It would be cool if the scenario designer could define groups of units (like group1 = 2 tanks, group2 = some infantry), so that you could specify that some marker is valid only for some group(s).

2.

The markers should have many kinds of meanings, like 'Keep the enemy away from this area', 'move own troops here' or 'hold this area'.

3.

Many kinds of conditions like 'enemy troops near markerX' or 'own troops near markerY' that would cause a marker to be considered the selected way.

To make the retreating scenario work, the designer

could place Marker1 that is valid for group2 to define a place where the infantry should retreat. Then place marker2 that is only valid for group1 (the tanks). The meaning would be 'keep enemy away from the area' and condition 'if not own troops at marker1'. Then place marker3 to some place where tanks should retreat in the end, valid for group1, meaning 'move own troops here', condition 'if own troops at marker1'.

During the battle the infantry would try to reach Marker1 while the tanks would protect them. Once the infantry reaches Marker1, the tanks could stop firing the enemy near Marker2 and retreat to Marker3.

So I see two big benefits here:

the AI could behave more like human players and things would be much more difficult for the human player because he wouldn't know how many markers there are, where they are and how the AI views them during the battle. Now he can immediately see the target flags on the map and predict that the AI will try to get those flags.

[gredeker]

Originally posted by Michael Dorosh:

Program it with variable markers. Give the computer 2 or 3 or more choices of which markers or sets of markers it will use. One playthrough, the computer might opt to take the high ground; the next, they roll on the crossroads.

I really like this idea. As scenario designer, have the option to construct up to a half-dozen plans for the computer to pick from using these markers, perhaps making the markers specific to infantry, vehicles, and heavy SW. Heck, the various plans using the markers could be adapted from playings during the beta testing process for the scenario.

As a corollary, it would be great if there could be several default setups for each side, rather than just one.

[aka_tom_w]

These all are great ideas.

I think the KEY here is the idea of the invisible marker (Attract/Repel Whatever) that the scenario design can set for the AI.

Flags do the job now, but for the most part they are always visible and completely obvious to both the AI and the player.

Invisible markers would allow scenario designers to design scenarios with some underhanded, diabolical (Mu ha ha) sneaky-ness involved in unit movement and organization tactical and strategic 'thinking'. This sure would help the AI and serve to make better games against the AI when you can't find time to play a REAL person.

In fact with this system given a certain specific set of criteria two players could challenge each other to build an AI scenario that would challenge the other. Suppose I say to

Rune (All Hale Rune the Great Scenario Designer), hey, lets specify some parameters (Hills, Trees, Date, Combined Arms number of points = 1500 side = Heer etc.) and using the editor build we each will design a scenario that would challenge the other player.

Then we both send each other our scenario's and play them to see who could beat the other's design.

To make it interesting we could both use the SAME map?? In this fantasy the "other player" might be able to then pick or "buy" his own player controled units.

We could do that now (in CMBO with its current limitations) I suppose, but we need more control to help the AI out to actually make it challenging.

Its still fun to dream

-tom w

[ April 09, 2002, 03:07 PM: Message edited by: aka_tom_w ]

[SlowMotion]

Originally posted by redeker:

I really like this idea. As scenario designer, have the option to construct up to a half-dozen plans for the computer to pick from using these markers, perhaps making the markers specific to infantry, vehicles, and heavy SW. Heck, the various plans using the markers could be adapted from playings during the beta testing process for the scenario.

As a corollary, it would be great if there could be several default setups for each side, rather than just one.

one more:

And several alternative places for reinforcements as well.

[Talon xBMCx]

This sounds a lot like the scenario builder for Steel Beasts. You can put units on routes with different SOP's, have conditonal routes, variable routes, triggered routes, etc. It does lead to some very good scenarios and depending on the scenario designer, has a lot of replay value because of the variable routes.

Something like this would help the Strat AI work as well as the Tactical AI.

Talon

[ April 09, 2002, 03:40 PM: Message edited by: Talon xBMCx ]

[aka_tom_w]

Originally posted by Talon xBMCx:

This sounds a lot like the scenario builder for Steel Beasts. You can put units on routes with different SOP's, have conditonal routes, variable routes, triggered routes, etc. It does lead to some very good scenarios and depending on the scenario designer, has a lot of replay value because of the variable routes.

Something like this would help the Strat AI work as well as the Tactical AI.

Talon

Maybe thats like a HINT and Steve and Charles and Matt might look into how the works in Steel Beasts??

I'm on a Mac so I have not played the game or looked at its scenario editor.

Could you post some screen shots of the interface on the editor from Steel Beasts to see what it looks like.

Has anyone else from this thread ever played with the scenario builder from Steel Beasts?

just curious

-tom w

[Talon xBMCx]

At work right now .... but I will post some screenshots when I get home

Talon

[Redwolf]

[not sure whether it has been noted in this thread]

TacOps has a programmed opponent that is driven by premade plans, a number of different plans for each scenario. Unfortunately, it is not exported to the user, user-made scenarios cannot be played by the PO at all, even if the scenario designer was willing to direct the AI.

But on the other hand this is quite flexible, since you get the full power of the programming language to express your plan. TacOps can kick quite a bit, and on the attack that is.

Which leads me back to my bitching about loadable *.dll files so that we can do stuff like this as a user.

[WWB]

Uh, just how many scenario designers can write C++ AI code redwolf? Even if they could, how many could write it well? I suspect no more than a handful, if any.

That said, I would really like to see a couple ways for a designer to 'manipulate' the strat AI without having to resort to unbalancing battles or ruining immersion. Unlikely, but it would be kinda cool to see some sort of 'hidden' flags so one could encourage the StratAI to take the long way around.

I, for one, am against the 'planned' AI approach that one sees in TOAW, because it gets stale fast unless the designer is real, real slick at his plans. SOPs would be very, very cool however.

WWB

[Talon xBMCx]

OK ... here is an example of what you can do with the Steel Beasts scenario designer. In fact, this is one of the ways you can plan the movements of your units within a scenario ...

As you can see ... there are many options ... and with the availability to set random events, triggers, etc, you could really extend the life of a scenario or operation. All depends on how much work you want to put into it.

Now ... if we could get the TacOps SOP orders in here as well ... WOW ... what a game!!

But ... I think all these features makes it a different game. CM is a tactical game ... adding in features like this is probably out of the projects scope ... but ... maybe a future project???

Talon

Edited for HORRIBLE spelling ... lol

[ April 10, 2002, 08:41 PM: Message edited by: Talon xBMCx ]

[Panzer76]

All this is well and good for the sceanario designer, but would be of no value in a QB. Instead of having to resort to micro management BECAUSE of bad StratAI, it would be better to just have better StratAI. Unfortanly this is not so easy to do (or what do you say Charles?) and the micromangment thingy could be the solution in a phase where they were trying to make better AI.

[Talon xBMCx]

Agreed ... this is probably not the best solution for a QB. ALthough a Strat AI needs to be built on some algorithm ... this might not be a bad place to start.

The real problem is that technology still hasnt caught up to the human thought process. We may be asking for more than can be provided for.

Talon

[Sergei]

Having waypoints etc. AI parameters for scenario designers would be great, but it wouldn't improve QB performance in any sort of way. Scenario designers can already tweak their scenarios by positioning of troops, victory flags and reinforcements so that the AI acts in a reasonable way.

I guess the answer to this is, "play QB's against human opponents only". But that takes so much away. And in operations the AI can't follow just one set of orders from battle to battle, because the situation is different at the beginning of each clash.