I just read the v4 Scenario Editor

[markshot]

It seems that all plans are strictly linear:

* You can control when an order executes by trigger.

* You can control when an order executes by time.

But there are no branching or case orders.  And, of course, if one had branching or case orders, then one would likely include random factor as being one of the conditions.

So for example,

IF ARMOR(AREA(5)) THEN

  EXECUTE ORDER 10

ELSE

  EXECUTE ORDER 20

ENDIF

Just curious.  Thanks.

PS:  I now see how you can have units on the map without the chain of command required.  You simply put them in reinforcement group that never actually shows up.  The explains why Bn is missing from of the scenarios.

[George MC]

Ongoing thread discussing the strategic AI limitations and methods for making the best of the tools we have which you might find useful?

 

And this one

 

Edited October 1, 2020 by George MC

[A Canadian Cat]

Correct there is no branching. I'd like to see it however there are things you can do that come close. The thread @George MC linked has a bunch of ideas.

1 hour ago, markshot said:

PS:  I now see how you can have units on the map without the chain of command required.  You simply put them in reinforcement group that never actually shows up.  The explains why Bn is missing from of the scenarios.

Yes, that is a thing that people do etc. However WRT battalion HQ you don't need to have them in a never arriving reinforcement group for them to show up on the C2 list. They just will appear as in C2 to the Company HQs when they have their radios up and running and out when they do not.

 

[MikeyD]

Early on, when triggers were first being designed, there were all sorts of ideas on how to implement them. Steve rejected all suggestions that would have put a 'programming language' burden on the end-user. I don't think they used the term but their concept was 'KISS' (Keep It Simple, Stupid). It would have been easy enough to code-up an AI orders feature that only 20% of users would be able to figure out. I wonder what percentage of users can figure out the AI as it currently stands.

[George MC]

2 minutes ago, MikeyD said:

Early on, when triggers were first being designed, there were all sorts of ideas on how to implement them. Steve rejected all suggestions that would have put a 'programming language' burden on the end-user. I don't think they used the term but their concept was 'KISS' (Keep It Simple, Stupid). It would have been easy enough to code-up an AI orders feature that only 20% of users would be able to figure out. I wonder what percentage of users can figure out the AI as it currently stands.

Good point Mikey, as having the strategic AI plan options to branch rather than be linear would mean keeping track of multiple variations. I'm not sure that would make AI planning easier, and given that currently the AI planning process is a block for designers making play against AI scenarios (though does not prevent making play H2H only).

[Erwin]

15 minutes ago, MikeyD said:

It would have been easy enough to code-up an AI orders feature that only 20% of users would be able to figure out.

Normally, that would be a very sensible approach.  However, we are blessed with a CM2 fan base that has astonishing talents and willingness to do hundreds of man-hours of work to produce scenarios, campaigns and mods to improve the game.  These talented and dedicated folks may comprise far less than 20% of the market.  But, just look at what they have accomplished. 

Based on that, hopefully BF may reconsider that approach and make provision so that those dedicated folks (such as yourself Mikey) can have access to and utilize these extra features.

[Glubokii Boy]

18 minutes ago, George MC said:

Good point Mikey, as having the strategic AI plan options to branch rather than be linear would mean keeping track of multiple variations. I'm not sure that would make AI planning easier, 

Simply because you have the option to use a braching feature does not mean that you have to use it...every time ! with every AI group.

In many would be scenarios it might be enough to use the brancing option ones or twice for a few of the AI groups...

I certainly would find that easier as opposed to having to come up with some tweek/trick to achive a simular result...

[George MC]

17 minutes ago, Erwin said:

Normally, that would be a very sensible approach.  However, we are blessed with a CM2 fan base that has astonishing talents and willingness to do hundreds of man-hours of work to produce scenarios, campaigns and mods to improve the game.  These talented and dedicated folks may comprise far less than 20% of the market.  But, just look at what they have accomplished. 

Based on that, hopefully BF may reconsider that approach and make provision so that those dedicated folks (such as yourself Mikey) can have access to and utilize these extra features.

Aye be interesting where this goes in the future. In an ideal world the AI player will be self aware and will generate attack and defensive plans on the fly with zero in-put from the player (if desired). Although the most likely outcome from an AI having that degree of ability is the end of humanity...

Personally I'd like the ability to create conditional triggers but I also know that will create a **** tonne of work for me when creating an AI plan... Especially if I have my wished for 32 AI unit slots...

Then again I'd not want to see a return to the old CMX1 AI plan flag rush (although that did make AI planning ever so simple).

Edited October 1, 2020 by George MC

[George MC]

2 minutes ago, RepsolCBR said:

Simply because you have the option to use a braching feature does not mean that you have to use it...every time ! with every AI group.

In many would be scenarios it might be enough to use the brancing option ones or twice for a few of the AI groups...

I certainly would find that easier as opposed to having to come up with some tweek/trick to achive a simular result...

What a new feature and not overuse it. Nah...!

It would still quickly escalate. Lets say accompany (I tend to use three AI groups to run them) so three strands to co-ordinate this unit. If I add one conditional trigger to them all that's doubled it to six strands. If I'm using the full 32 way point options it becomes a lot of managing. Doable and I know other games use this but they use an interface that allows this. That's why I can't see this happening within CMX2 unless there is a massive redesign and reengineer of the current AI plan designing interface in the editor.

So if at some point in the future we go into CMX3 then yup I'm on board with a whole new AI planning interface that allows me to do it in 3D with the unit and then can show me run the whole plan in 2D whilst i watch it. And lets me intervene and tweak stuff. In short if I can paint AI plans I'm up for that!

Edited October 1, 2020 by George MC

[Glubokii Boy]

3 minutes ago, George MC said:

What a new feature and not overuse it. Nah...!

Got a point

[markshot]

Scenario and content modelers are the creme of any games customer crop.  No, I don't think it is too much to ask that they understand the three basic concepts of programming:

* Linear flow

* Branching

* Looping

I would not be surprised if this much programming is covered in elementary school these days.  Obviously, with the right predicates and group behaviors some very sophisticated scenario behavior could be achieved.

---

Of course, I would still argue that time would be better invested in a smart game engine, than smart individual scenarios.

It is not so hard to quantify basic rules of battlefield plans.

If you have an initial map maker that can start with a topo map from Google Earth, and just needs some painting of vegetation and built up areas.  If you have a force editor.  If you have an objective editor.  Then, you have something simple enough for the average joe to make scenarios, and those scenarios can then be analyzed and animated by your tactical engine.

So, the way to create lots of content creators is keep it really very, very basic.  Make the tools smart, and let the people just read a 5 page tutorial on how to make a scenario.

[Glubokii Boy]

Another reason to try and improve/simplyfy the current editor would imho be...

New designers !

Sure...at precent the threads with the tips, tricks AND tweeks are quite active and new members might read them...

But in 3 months...they may be somewhat lost a few pages down...

Having an editor that do not rely this much on tweeks but rather on included features would be desirable...

 

[Kevin2k]

Let me add this example:

In CMSF2 Quick Battle I was flanking the main enemy force (computer) on the side of a hill. Now the enemy force was also split without my knowledge, and enemy vehicles came over the top of that hill, onto my position there: Surprised me. Caused havoc. But the computer then drove these vehicle 30 meters further and waited there with their backsides facing the survivors of my force. My survivors could acquire some AT weapons and finish them off one by one.

AI plan was to go to that spot apparently. It found that spot more important then my survivors. 

I don't really expect scenario designers to be able to solve such. It seems more like the Tac AI is reasonable when troops are in a position, as it considers nearby cover and opposition. But when moving between positions the computer's behavior about that movement is simple minded: Keep going and shooting until dead or until you can't take it anymore.

Would love to see that aspect of the AI get smarter. Would also love to get the auto mix force selection get smarter. Glad to see the v4.03 patches finally making the Engine 4 Tac AI on defense get smarter. I suppose it is all about when the programmer has the time, skill and inspiration to attend to it. But then smarter may also mean more complicated, more entangled and therefor prone to undesirable side effects; bugs. 

[Freyberg]

A possible 'simple' improvement would be the ability to:
1) import and export AI plans;
2) name AI groups and objectives with recognisable names ('infantry recon'; 'phase line one'; 'long range fire position' and so on)

Then the top designers could design cool generic plans, that just needed tweaking and having suitable objectives applied, and we less skilled map & scenario makers could exploit them.

[markshot]

I do have a software background, but just informally in AI.

Let say we have 2 forms of narrow AI and general AI.  Narrow AI solves very domain specific problem, and does not transfer; like chess engines.  General AI is like Google's neural net which is blowing through problems.

General AI is not possible for lack of expertise, funds, computing power ... ultimately if it becomes available to games it will need a training system/database, and be sold as a packaged module to companies.

But for CM narrow AI is possible in 2 forms:

* Rule based where specific behaviors are coded for based on your forces, enemy response, objectives and terrain options.  This rapidly gets hard due to the many, many cases.

* A chess engine approach.  Chess engines have the open, middle, and end game.  Open is just a statistical DB of the 6 move games results over 500 years of recorded game.  Moves are picked by percentage.  End game is an absolute lookup table of next move for a limited number of remaining pieces.  It is done by tables which go down to checkmate.  The result of the tables are simply perfect chess.  Middle game is done by looking ahead and scoring all possible board positions, and then picking the high scorer giving 10 moves into the future and what it is the best next move.  (There are many heuristics done to make this more efficient.)

---

How do you write a middle engine for CM?  You divide every CM map into a equal number of sectors rated on terrain for a small number of force types.  You only partially lift the FOW for the AI.  Just enough that the engine may make assumption about enemy forces and their locations.  You then run various minutes into future different positions of forces moving with estimates of kills.  A chess board rates position on the basis of 1/100 of pawn being lost.  Here the rating might be a number representing the ratio of purchase points remaining.  You look into all force movement maybe 5 minutes ahead.  You then pick the movement plans that lead to the objective which have the best ratio.  You tweak the rating by making tanks ahead of infantry have a lower potential rating than infantry ahead of tanks.  This is how chess engines do it so that bishops and knights which have the same point value may generate different board ratings based:  2 bishops remaining (cover the board), knights in an early game or pawn wall are more powerful than bishops, late game bishops shine with open space ... you tweak the scores and those tweaks give behavior based play without the engine understanding behavior and plans.

---

That is my prognosis for a CM which put ups a good fight without plans.

For the would be CM content creator ...

You make fast map creation possible by getting base maps from other sources.

The forces editor is fine as it stands.

Player picks setup zone and objectives.  (Battlefield engine does the rest)

---

I am not blowing smoke.  We have plenty of fine examples of open source chess engines of they this are coded that they can easily beat the best human players and can hit standardized ratings 20% better than the best human.  After 20 years, of coding such narrow AI, Google's Neural net blew away the best engine in 2 days just by playing games against itself to learn --- neural net.  But what powers the traditional middle game is no PhD stuff anymore ... any decent programmer can follow the approach.  It just needs to be adapted for CM.

CLICK ON GO.

[Glubokii Boy]

Well CM has far more factors to considder compared to a chessboard...

Elevation, different weather and time of day, a wide variaty of different terrain...

I chessboard has non of those...

Also a standard size CM map has many, many times the grid squares compared to a chessboard...

Things like moral, experience, fatigue ammo load and supession also needs to be considdered...not so with chess.

Concealment, protection are other things..

Changable LOS (smoke) and changable terrain (destroyed/damaged buildings etc..)

Maybe what you are suggesting is indeed possible but to me CM if far, far more complex compared to a standard chessgame.

But if it is doable...hell yeah ! Do it

Edited October 2, 2020 by RepsolCBR

[markshot]

Chess proves such a framework is feasible.  Also, it easier to tune ratings than lines of code.  Both will alter behavior, but assigning values is are easier once you have a framework and can be done by a non-programmer.  Given the nature of combat, you don't need 10 level deep look ahead; maybe 5 minutes.

The CM board does not need to match action squares.  So CM evaluation square just needs to have a cover rating based tree or structure coverage.  Let the TACAI find the actually cover to use on the square.

Which makes more sense?  To have 20 people trying to work out blind plans for scenario writing or have 1 person code a move evaluation engine and have 2 or 3 people working on rating values for different positions until they get it tuned just right.  You can even leave those table values open to modders.  So, if you want a CM with aggressive armor, you can tweak the values to make it so.

I have no data, but I suspect the majority of customers are SP.  I think improved AI will translate into sales.  Too many games have been released with superb graphics and million features, but brain dead computer opposition.

[Glubokii Boy]

1 hour ago, markshot said:

  I think improved AI will translate into sales.  Too many games have been released with superb graphics and million features, but brain dead computer opposition.

Agreed...

[Sgt.Squarehead]

On 10/1/2020 at 8:41 PM, Kevin2k said:

But when moving between positions the computer's behavior about that movement is simple minded: Keep going and shooting until dead or until you can't take it anymore.

This is exactly the issue, but it can be somewhat mitigated with forethought by the designer.

In the instance you describe, with the enemy force over-running your position, this could probably have been solved had the designer split the move into several smaller moves, and used the Maximum Assault Stance for one of them (the one where they are in what is likely to prove the most favourable position, should enemy units be in that area).

With the Maximum Assault Stance units will not move to their next order location until they have eliminated all enemies in contact.

Edited October 2, 2020 by Sgt.Squarehead

[Bulletpoint]

On 10/2/2020 at 11:25 PM, Sgt.Squarehead said:

With the Maximum Assault Stance units will not move to their next order location until they have eliminated all enemies in contact.

I thought max assault meant that they would keep going no matter what? So that they will continue by crawling if they get suppressed.

[Sgt.Squarehead]

Quoting from the Engine 4 manual:

"Max Assault – the opposite of Dash, Max Assault tells the Group to stop and engage with maximum firepower whenever each unit sees an opportunity to do so.
This is generally a poor choice for getting a Group to stay on the move if a lot of enemy is expected in the vicinity. It can also be a bad idea if the area being moved over is a poor place to stop for a firefight. The best use is for short moves where there is good cover and enemy activity is highly expected."

Edited October 6, 2020 by Sgt.Squarehead

[Erwin]

In my 4.0 engine manual I didn't see any mention of "Max Assault".  What page or is there a newer engine version?  

[Howler]

23 minutes ago, Erwin said:

In my 4.0 engine manual I didn't see any mention of "Max Assault".  What page or is there a newer engine version?  

Bottom of page 104.

[Sgt.Squarehead]

Why would you care anyway? 

As you so often remind us.....You don't do scenario design. 

[Erwin]

Ah.  Thanks...  I thought it was referring to the in-game ASSAULT order.  Freaked me out as it's tiresome to reread the manuals and relearn game features every time a new patch comes out.  Phew...

Edited October 6, 2020 by Erwin