More centralised AI-groups for better game play

[BornGinger]

There have been quite a few people on this forum, me included, requesting more AI-groups than the 16 for each side in the scenario editor to be able to hopefully make the home made scenarios more interesting and more challenging for those who prefer not to play H2H battles.

 

A few days ago when I began doing the AI-movements for the scenario I'm a bit busy with now, I got the idea to see how squads in the platoons are moving around on the battle scenario area when painting the same AI-movements and giving the same AI-orders in the editor for in one case two separate company groups of a battalion and in another case two companies in one battalion group.

 

When I painted the movement orders I didn't paint two or three large yellow areas. Instead of that I decided to paint six different yellow areas and thus try to give each platoon its own area (two companies with three platoons each) just to see how the AI would behave when facing that challenge.

 

In the first test I had a battalion consisting of two companies, where one company was group A1 and the other was group A2 and where each group (company) were given three painted areas each (three platoons). In the second test I had the same battalion with the same two companies where the whole battalion was group A2. The AI-orders were the same with the difference that this time the AI-movements which earlier belonged to group A1 had been copied and added to the larger group A2. So this time the battalion had six separate painted areas to use.

 

When there were two company groups the squads in each platoon were sticking together quite well although they sometimes spread out a bit. When the whole battalion was one group the squads in each platoon spread out even more and didn't really stick together very much.

 

The picture below shows an example of 1 Platoon, 1 Company in these tests. The left column shows 1 Platoon when each company was a separate group and the right column shows 1 Platoon, 1 Company when the two companies belonged to the same group.

 

 

I think it would be great, and hopefully not too much work, if the game engine could keep track on each squad in the platoons so they are staying closer to each other when they are following the movement orders that are given in the editor. That way the platoon leader would stay closer to the squads he's supposed to influence and the whole AI-platoon would maybe fight better.

 

If it was possible for the game engine to keep the squads closer to their leader it would maybe be easier to make an interesting scenario where one can paint more separate movement areas with the trust that a platoon would move there together. 

 

Instead of having a battalion consisting of three companies with three platoons each divided into nine AI-groups in a large or huge scenario, one AI-group for each platoon, it would be possible to for example divide that battalion into three groups. And if we wanted to have eight tanks in that scenario moving separately we could make them into eight different groups and still have five AI-groups left. Those five groups could be used for whatever we'd find would make the scenario more fun to play.

Edited August 21, 2020 by BornGinger

[MikeyD]

My old eyeballs are having difficulty seeing what's happening in those screenshots. In the AI plans try inserting an intermediate waypoint order between the start point and final destination. In theory they should gather at the intermediate point before moving on. The game's also less tolerant these days of open-ended movement orders without specific times attached. instead of keeping the 'Exit Between/ and..' times at 0:00 give then an achievable timeframe. Exit between 5:30 and 8:00, next movement order Exit between 8:30 and 10:00.

[Bulletpoint]

Instead of asking for more AI groups so we could micromanage more, it would be nicer if the game just parsed the movement orders a bit more intelligently - keeping platoons and platoon leaders reasonably close together for example, and not advancing with the PL in front.

Currently, the game just sees the whole AI group as a random collection of units, and when given a move order, it randomly assigns each of these units an order to move to a randomly chosen spot in the yellow painted zone.

Edited August 21, 2020 by Bulletpoint

[BornGinger]

1 hour ago, MikeyD said:

My old eyeballs are having difficulty seeing what's happening in those screenshots. In the AI plans try inserting an intermediate waypoint order between the start point and final destination.

It's possible to enlarge the picture by clicking on it.

I did several waypoints from the start till the end line. If I remember correctly I painted movement areas for every five minutes or similar of game play.

[Glubokii Boy]

Imho it is not even remotely realistic to hope for a vastly improved AI in CM2...

One that is able to read the terrain, asses dangers and advantages, fullt understand the pros and cons as well as the purpose of individual teams of various kinds.

To get this it would require massive changes to the AI...

It's not going to happen ! If we ever see CM3...sure...maybe

But not in CM2...NO !

On the other hand...more AI groups...

That sounds like something that might actually be possible within CM2

As the only thing needed would be a small UI change in the editor...

 

 

 

 

[Combatintman]

Agreed that the AI is painful when you're assigning platoon-sized or larger units to AI groups, their movements are illogical.  In both of my scenarios for Fire and Rubble that I'm putting together I have had examples where the extreme left hand unit on the previous waypoint wants to become the extreme right hand unit on the next waypoint while predictably the extreme right hand unit on the previous waypoint wants to become the extreme left hand hand unit on the next waypoint, despite the distance between waypoints being less than 100m and there being no intervening terrain features, or any enemy activity that would cause the Tac AI to kick in that would prevent them from just moving directly ahead from their last waypoint to the next waypoint.

[Bulletpoint]

13 minutes ago, RepsolCBR said:

Imho it is not even remotely realistic to hope for a vastly improved AI in CM2...

One that is able to read the terrain, asses dangers and advantages, fullt understand the pros and cons as well as the purpose of individual teams of various kinds.

To get this it would require massive changes to the AI...

But I didn't ask for a vastly improved AI that could do all those things. I asked for small improvements, instead of just increasing the number of AI groups. How many scenario designers even use the amount of AI groups we have now?

You're asking for more horses when you should be asking for a train.

Edited August 21, 2020 by Bulletpoint

[Glubokii Boy]

Well...

I've had very simular experience to what combatintman describes above...

Trying to get an AI group to advance on both sides of a road...result...

The various squads zig-zagging over the OPEN road again and again as they advanced...

Not pretty !!

A very simple solution to this problem would be...to have 2 AI groups ...one advancing on each side of the road...

If the AI could be made to understant notto do that...great ! But i don't think that such a fix is all that easy...and this is just ONE of MANY pretty simple....'stupid' things the AI does if left all alone...

A starter AI would be great...but more AI groups is an easier, Quick fix imo..

[Glubokii Boy]

Should be....a SMARTER AI

[Bulletpoint]

22 minutes ago, RepsolCBR said:

A starter AI would be great...but more AI groups is an easier, Quick fix imo..

True.. I wouldn't complain if they added more AI groups. It wouldn't make me go back to making scenarios either though.

[Anxel Torrente]

One way in which this behaviour of the AI could be bad is probably when the mortar teams might run out of ammo too early because those who carry ammo are running wild several hundred meters away or maybe in the situation when FOs can't request mortar fire because the only officer with a radio is too far away to tell the mortar team where to shoot.

Edited August 27, 2020 by Anxel Torrente

[Macisle]

On 8/22/2020 at 5:18 AM, RepsolCBR said:

Well...

I've had very simular experience to what combatintman describes above...

Trying to get an AI group to advance on both sides of a road...result...

The various squads zig-zagging over the OPEN road again and again as they advanced...

Not pretty !!

A very simple solution to this problem would be...to have 2 AI groups ...one advancing on each side of the road...

If the AI could be made to understant notto do that...great ! But i don't think that such a fix is all that easy...and this is just ONE of MANY pretty simple....'stupid' things the AI does if left all alone...

A starter AI would be great...but more AI groups is an easier, Quick fix imo..

Yup. This.

It's been said many times by many people for many years.

The AI isn't going to get smarter with CMx2 and I doubt much smarter with CMx3. The only way to mitigate that is for scenario designers to have more groups. That actually makes the workload less, not more. Less groups = more time spent with workarounds and testing, testing, testing. More groups = less time as the designer can get something organized, test a bit and move on.

32+ Groups is still at the top of my desired feature list by a country mile.

[benpark]

This is a good test. I can submit this specifically if I can get a save.

The thing to look for is unintended consequences from fixing something- does removing that seeming randomness (also applied to the AI here) suddenly make the AI units move in a suddenly unconvincing way? Then a new system needs to be created to make that look "right".

I'd wager- More AI slots makes more of the what we have from an existing system, rather than taking apart some existing system for reconfiguring. The only consequence for us would be keeping track of all these plans (I have a Photoshop tool for tracking AI plans for post release of the RT module, and an idea for something better submitted). Probably less easy from BFC's perspective.

There's a lot once can do with the existing tools, but this is one area I think we all bump our heads on.

 

Edited August 27, 2020 by benpark

[Glubokii Boy]

1 hour ago, Macisle said:

Yup. This.

It's been said many times by many people for many years.

The AI isn't going to get smarter with CMx2 and I doubt much smarter with CMx3. 

This is probaly true...

The battlefield is simply to complex i fear and the situational awareness of the AI is to limited to be able to handle it...

This is in no way something that is limited to combat mission though...many games struggle with the AI.

As a counter argument for more AI groups i have many times heard that it would be better to improve the AI...

If it only was that easy...

To examples of very, very SIMPLE ! misstakes made by the AI and two problems that BFC have been struggeling big time with trying to fix...

- units using the right door when entering and exiting buildings.

- the hedgerow bug ( actually fixad i belive)

This two examples are rediculesly simple to 'get right' with a human brain...but for the AI....apparently not.

I hear as an argument for smarter AI that the AI does not neccesarely have to be massively improved...only slightely improved will be good enough...

Imo...the current AI is lackning two very fundamental things to even become slightly better...

- the ability to read and understand the terrain both in the short and long distance combined with the threatsituation...

- the AI is not aware how to use the various teams and different weapons in a good enough way...not even seperated on their own and most certainly not as a combined arms force.

Lackning these two very fundamental things it makes it very hard to improve the AI even slightelly i would belive.

And making the AI capable to perform these two tasks requires...NO small effort.

The AI...NEEDS ! a human touch for some guidance ...

More AI groups might not be a perfekt solution but a truely belive it is the easiest way forward...and like Macisle mentioned above...

Ut actually makes scenariodesign...

Easier and faster

[Macisle]

9 hours ago, benpark said:

...The thing to look for is unintended consequences from fixing something- does removing that seeming randomness (also applied to the AI here) suddenly make the AI units move in a suddenly unconvincing way? Then a new system needs to be created to make that look "right".

I'd wager- More AI slots makes more of the what we have from an existing system, rather than taking apart some existing system for reconfiguring. The only consequence for us would be keeping track of all these plans (I have a Photoshop tool for tracking AI plans for post release of the RT module, and an idea for something better submitted). Probably less easy from BFC's perspective.

There's a lot once can do with the existing tools, but this is one area I think we all bump our heads on.

Agree and I'm fine with just having more slots with the current system. IMO, they're really not hard to keep track of once you get a feel for things.  I also had a Photoshop tool for awhile, but ended up going back to paper and pencil.

The built-in randomness of the AI is exactly what you want in some situations and exactly what you don't want in others. Having more groups allows the designer a good chance to avoid the situations where randomness is unnatural and/or suicidal with a minimized workload.  Another big help would be in facilitating things like triggered local counterattacks and keeping ambush groups shielded from area fire/arty until they are triggered to take their intended positions. FO use is another big one. Having enough group slots to be able to use a number of single-team groups would be huge. (On my urban map, I'm finding the AI much less aggressive with arty than usual. It seems to only reliably call it when it has a TRP. So, having enough group slots to control individual FO teams to keep them in good positions with views to TRPs would solve the problem). 

Basically, as you say, there is quite a lot that can be done with current tools. The problem is, there aren't enough group slots available to make full use of the existing tools. There's a whole heap of cool stuff that could be done with what we have now if we didn't have to worry about running out of group slots.

8 hours ago, RepsolCBR said:

...This is in no way something that is limited to combat mission though...many games struggle with the AI. ...

Definitely. I'm not making any digs against BF here. It's just the nature of the beast. Steve has said on the forums that AI programming is one of the areas you readily hit diminishing returns. So, when I say I'm not expecting a smarter AI, it's not said with any piss and vinegar -- just matter of fact.

Like I said above, I'm actually fine with the current AI if you give me more groups to avoid its rough edges and facilitate my devilish schemes.

Having said that, what I'd really like to see is the current system buffed up with 32+ groups and a host of things like terrain objective unlocks, triggered reinforcement groups, point-neutral exits, more reinforcement groups, all-arty-asset AI area fire, and...

[Macisle]

Here's another example, very relevant to Fire and Rubble, where not having to worry about conserving group slots allows the designer to make the current AI functionality much more human-like and effective with minimal testing:

The situation:

A platoon-sized element of AI troops is defending an area of multi-story buildings with some space separations between them (street, etc.). The higher floors offer very good defensive positions. Enemy artillery and/or HE direct prep is expected.

A human player would likely stay hidden on the bottom floors during enemy prep and use a few scouts to know when to occupy the higher floors after the enemy begins his actual advance.

Now, what to do with the AI?

Well, of course you can hide the troops on the bottom floors during setup and trigger them when the enemy advances. But what happens if you only have one group slot available? Well, if their setup uses more than one building, they are going to crisscross between buildings, waste time, give away intel and likely get shot to pieces. In other words, an immersion-blowing chitshow.

Now, what if you had, say, a group slot for each squad and team? Well, not only could you spread them out to cover lots more buildings and make them unhide and go up the stairs when triggered, but you could also periodically have them move up or down in the same building (tip: to make this work, the painted tile needs to change between orders. Just select an adjacent tile in the same building). So, in addition to the other benefits, you can make the AI more human-like by having AI units return to levels they've displaced from and/or generally sneak around and keep the human player guessing.

All that with no additional AI functionality needed from BF and very easy to set up in the editor. I just tested it in CMFB with six groups and it worked like a charm. Well, except for one group of two teams that I put in different buildings as a test. They crossed the street and got shot up (as did the whole platoon when I tested with one group).

All the others did exactly what I wanted them to inside their own buildings. And it took a whopping 10-15 minutes to put together. Compare that with hours of testing trying to make something similar work with less groups. Ugh.

More groups = better AI = happier designers = faster, more productive designers = better CM.

[Glubokii Boy]

Another small example

Situation...a german rifle platoon supported by an HMG, a medium mortar and a Stummel halftrack.

Put all these units in one AI group and have them adance....the result will not be pretty !

On the other hand if the rifle platoon could be its own AI group (or even two) and the HMG, mortar and halftrack where placed in seperate AI groups...

- The mortarteam could provide indirect HE fire at the desired time and locations as the platoon advances...

- the HMG could be made to occupy good base of fire possitions and provide both suppretionfire (areafire) as well as engaging spotted enemies. And then move forward at desired times.

- simularely the Stummel could be held back and provide direct fire HE using both areafire as well as targeting spotted enemies...advancing carefully behind the infantry using shoot and scoot tactics...

The difference in AI performance between using one or multiple AI groups in the example above would be...night and day

I realize that very few designes will ever put these units into one single AI group even now...but it kind of shows the difference between one and multiple groups

[Anxel Torrente]

On 8/27/2020 at 3:45 PM, benpark said:

I can submit this specifically if I can get a save.

Any success with that save Mr. Benpark?

[benpark]

Don't see the file- is the new UI here doing something different with attachments?

[Bulletpoint]

4 hours ago, Macisle said:

A human player would likely stay hidden on the bottom floors

Or on the top floor, since that is safer in this game Even 150mm hitting the roof does no damage to the guys in the room below, as far as I've seen.

Edited August 29, 2020 by Bulletpoint

[Bulletpoint]

I've only done two scenarios, but especially the last one was very detailed with lots of AI groups even for a small German force. Lots of fallback triggers etc too. So I can recognise that more AI triggers could be good to have. But, I'd much rather have

An undo function

A copy/paste option for terrain

A better way to choose building configurations instead of clicking to cycle through all options for doors, windows, etc

Reinforcements by trigger

AI artillery fire plans

Option to disable AI surrender without giving them reinforcements that will never arrive.. I had several players ask me why the end battle screen said there were more enemies when they cleared out the map

More and better vegetation options so I can make more immersive and realistic maps

Etc.

[BornGinger]

2 hours ago, Bulletpoint said:

Option to disable AI surrender without giving them reinforcements that will never arrive..

What do you mean by that, Bullet?

[Bulletpoint]

35 minutes ago, BornGinger said:

What do you mean by that, Bullet?

In scenarios, the AI is coded to automatically surrender when taking a certain amount of casualties. Often, the designer doesn't want them to surrender - it's more fun for the player to finally actually take the objective than to just get a total win from a surrender before setting foot on the objective zone. The designer can prevent the auto surrender by giving the AI hidden reinforcements set to arrive after the battle ends. These extra troops make the AI feel braver so it won't surrender. But they still show up on the end game tally.

[Glubokii Boy]

3 hours ago, Bulletpoint said:

A copy/paste option for terrain

I agree that a copy, rotate, paste option with the ability to save (and share) the selected mapsections as templates would be a great improvement...

One that would significantelly speed up mapdesign...

Just imagine if you had something like 6 - 8 large farms and a simular number of small and medium farms as templates...ready to simply click into place on the map...

Designing a map with 10+ detailed farms on it could be done within minutes...

Simularelly having templates for detailed Woods, streams and roads with surrounding detail, parts of cities and villages etc, etc...

That would indeed be good and such a feature is high on my wishlist...

But imho the without any comparison number one area that needs the most attention...as a singel player game...is improvements to the AI...

Having the option to face of agaist a decent/improved opponent (AI) would bring far more enjoyment, challange than any cosmetic or userfriendlyness changes made to the game...

 

 

[Macisle]

11 hours ago, Bulletpoint said:

Or on the top floor, since that is safer in this game Even 150mm hitting the roof does no damage to the guys in the room below, as far as I've seen.

That's not my experience. I've done a lot of testing with arty strikes on my urban map. The safest place to be during arty is hidden on the bottom floor. And, the safest, safest place to be is on the bottom floor of a tall building with connected sections surrounding yours so you have a shield of solid walls between you and any landing shells. Of course, if the occupied section collapses, your team is dead.  Also, I haven't tested with the new patches, but as of current CMRT, arty very often drives AI troops down to the ground level anyway. So, the designer is back to needing a way to get the AI back up to the upper levels without a clusterfrack of AI keystone coppers caused by having too few groups. A human can do this easily. Without having enough groups, the AI can't. This means that, as things stand, the human player simply needs to hit an urban area with arty to permanently flush the AI from higher building positions for the duration of a scenario. That puts the AI at a huge disadvantage when it doesn't have to be and kinda kills the point of having all those tall buildings...

7 hours ago, RepsolCBR said:

...But imho the without any comparison number one area that needs the most attention...as a singel player game...is improvements to the AI...

Having the option to face of agaist a decent/improved opponent (AI) would bring far more enjoyment, challange than any cosmetic or userfriendlyness changes made to the game...

Exactly. Steve has said that the vast majority of CM is played in SP mode.  So, most players are playing against the AI most of the time. Therefore, it is self-evident that maximizing the SP experience should be a top priority. Having a better, more human-like AI would absolutely enhance the SP experience. However, Steve has also said that AI programming is difficult and quickly produces diminishing returns. So, why not just chuck the "smarter AI" pipe dream and do what can be done now with the tools available now and git 'er done? Just need more groups (32 is just a minimum. Unlimited would be nice ).

Of course, the SP experience would also be enhanced by many other things like a better AI arty plan system, better AI area fire (optional fire rates), triggered reinforcements, point-neutral exits, objective unlocks and on and on.  And, adding tools for faster map production would yield more maps --which would be great. However, as pointed out, you're still fighting the same AI. The AI is at the core of the experience.

At this point, nothing would make the AI more human-like than designers being able to apply their creativity fully by not having to worry about conserving group slots. And, unless there is some code issue making it prohibitive, adding the functionality should be straightforward and thus low-cost. Group limit unchained = better AI on the cheap = better SP experience = biggest boost to CM = big happy .