Engine 5 Wishlist

[PEB14]

1 minute ago, Artkin said:

I cant remember. Something about the integrity of the spirit of the game, they dont want people using the game as a toy. Not like it's a simulator or anything. 

These arguments sound more like pretext than real arguments…

[A Canadian Cat]

11 minutes ago, PEB14 said:

14 minutes ago, Artkin said:

I cant remember. Something about the integrity of the spirit of the game, they dont want people using the game as a toy. Not like it's a simulator or anything. 

These arguments sound more like pretext than real arguments…

Call it whatever you like. Steve has stated before that they have no plans to open up the vehicle and weapons systems' attributes for tweaking. His justification is two fold:

On the business side they want to be the one's making content for the game and they want it to be as historically accurate as possible. Having to compete with people that didn't develop the game would not be in their interest. Having to allow people that think <insert favourite uber tank here> should be the uberest would suck (not just for them). Their game, their market, their call.

On the technical side they don't want the support headaches or to fracture the multi player community. People playing head to head are a small group in a niche market if there were also 15 different fan mods for basic kit there would probably be just two guys left playing.

[PEB14]

I probably wasn't clear: I don't ask for tweaking the vehicle and weapons systems' attributes. BFC rightfully never would. My main request was to open the battalion structure OOB to modders. This way they would be able to tweak the squad's staff and equipment. Of course, you would get delirious mods with battalions of flamethrowers. I don't think it would attract lots of people. On the other hand, removing the Panzefaust from German squads would allow to mod Stalingrad accurately, by example.

Having to compete with people that didn't develop the game obviously would not be in BFC interest. But if BFC doesn't intend to go early war, they've nothing to lose and could let it to the community. There is indeed an active (Steve would call it a "vocal") minority of  people asking for an historically accurate content related to early WW2. Indeed CM is BFC's game and if they don't wanna do early WW2 they won't. Their game, their market, their call, as you rightfully point out. So I feel that they could at least open this era to modders. A niche inside a niche. Nothing to gain, nothing to lose.

 

[Vacillator]

15 minutes ago, PEB14 said:

There is indeed an active (Steve would call it a "vocal") minority of  people asking for an historically accurate content related to early WW2.

Yes please.  I'll tag in Steve @Battlefront.com as I know he really agrees and secretly wants to do early WW2 stuff.  If I'm banished to the wilderness I wish you well...

[HunterICX]

Playing it for a couple of years , mainly SF2 and CW I have adjusted to how the engine plays like but I mainly would like to see optimizations happening, would love to see it play smoother and bring the engine up to standards which would be more appealing for new players to this franchiset to keep it going for many more years to come.

All in due time and I would welcome any step of improvement in that area as a big plus.
 

[Cirrus]

Back for this game for long time with engine 4. I am pure ”hobbyist” gamer for these type of games.

Now trying to play CMBN ”Panzer Marsch!” Campaign and scenario ”Le Dezert” I believe. I do not think I can finish it. Try to make vechiles move rationally in narrow roads just make my head explode.

I really wish for and miss simple order I could give as army transportation NCO back in the day: ”Follow that vehicle”

[RMM]

4 hours ago, Cirrus said:

Back for this game for long time with engine 4. I am pure ”hobbyist” gamer for these type of games.

Now trying to play CMBN ”Panzer Marsch!” Campaign and scenario ”Le Dezert” I believe. I do not think I can finish it. Try to make vechiles move rationally in narrow roads just make my head explode.

I really wish for and miss simple order I could give as army transportation NCO back in the day: ”Follow that vehicle”

Oh mercy. Can we give, like two thumbs up and fireworks!

[Codreanu]

On 5/6/2023 at 2:59 AM, Cirrus said:

Back for this game for long time with engine 4. I am pure ”hobbyist” gamer for these type of games.

Now trying to play CMBN ”Panzer Marsch!” Campaign and scenario ”Le Dezert” I believe. I do not think I can finish it. Try to make vechiles move rationally in narrow roads just make my head explode.

I really wish for and miss simple order I could give as army transportation NCO back in the day: ”Follow that vehicle”

Some scenarios I just won't play because they involve babying a gigantic column of vehicles down a single lane road. It takes so much time and clicking that it just isn't worth it, especially since you have to save every turn to make sure a vehicle's pathfinding won't break and create a 20 car pileup. A working convoy command is one of my biggest asks too.

[RMM]

Units that are completely out of ammo and become adjacent to enemy units should surrender.

[BornGinger]

On 5/8/2023 at 7:42 PM, Codreanu said:

Some scenarios I just won't play because they involve babying a gigantic column of vehicles down a single lane road.

If there will ever be a new game engine for Combat Mission or at least a better updated version of the current one, it would help a lot if there was a similar movement order as there is in the Mius-Front game where you choose to move vehicles to a certain point with the order to "follow the road" to that point.

Edited June 6, 2023 by BornGinger

[SlowMotion]

It would be great if one day the the game AI that plots turns could be taught by feeding it turn files of human players playing the game. So instead of doing things according to AI plans made in scenario editor, it would play the game by creating those plans by itself. 

And if dreaming this further maybe there could be different "AI players" that would play differently because they had been taught to play from different set of H2H games. People could choose which AI they would play against. Now that AI development is advancing fast I'm sure this kind of thing will become reality in some games some day, but I'm not sure if this will happen in CM series.

[Ithikial_AU]

1 hour ago, SlowMotion said:

It would be great if one day the the game AI that plots turns could be taught by feeding it turn files of human players playing the game. So instead of doing things according to AI plans made in scenario editor, it would play the game by creating those plans by itself.

It's a great idea on paper but then drill down and you quickly run into some problems.

1. How do you import and export the "Player as AI" plans and distribute to the broader single player audience.

2. Most players are single players.

3. What happens when the scenario or game is first released and there are no/few MP battles?

4. How would new players feel? Could raise the difficulty bar too high and turns new players off the game if the AI is 'too good'. Scenarios designed to be easy entry points could become very difficult over time.

Two cents plus sales tax.

[SlowMotion]

3 hours ago, Ithikial_AU said:

It's a great idea on paper but then drill down and you quickly run into some problems.

1. How do you import and export the "Player as AI" plans and distribute to the broader single player audience.

2. Most players are single players.

3. What happens when the scenario or game is first released and there are no/few MP battles?

4. How would new players feel? Could raise the difficulty bar too high and turns new players off the game if the AI is 'too good'. Scenarios designed to be easy entry points could become very difficult over time.

Two cents plus sales tax.

1. I did not mean they would be scenario specific, but more general purpose AIs. Those that could play any battle of some CM game. How those would be included in the game, not sure. But if they wouldn't take many gigabytes per AI, I think it could be possible.
2. Yes and exactly for this purpose my idea would be good. You would still play against an AI, but it would be closer to human player because it would have learned from how real human players play.
3. Scenarios could be made just like now, except scenario authors would not need to make AI plans if these general purpose CM AIs would play well enough.
4. For that purpose you could have different kind of AIs. Some would get little training and not from very good players, so they would play like rookie human players. Others would be taught tricks from more experienced players and they would play better.  The human player could choose the AI he wants to play against, just like now you select difficulty level when starting to play a battle.

This may be science fiction at the moment, but I think this kind of things will happen during the next 20 years maybe.  Like now there are self driving taxi services - would you have believed it during 2000?
https://waymo.com/

 

Edited November 1, 2023 by SlowMotion

[Redwolf]

9 hours ago, SlowMotion said:

It would be great if one day the the game AI that plots turns could be taught by feeding it turn files of human players playing the game. So instead of doing things according to AI plans made in scenario editor, it would play the game by creating those plans by itself. 

And if dreaming this further maybe there could be different "AI players" that would play differently because they had been taught to play from different set of H2H games. People could choose which AI they would play against. Now that AI development is advancing fast I'm sure this kind of thing will become reality in some games some day, but I'm not sure if this will happen in CM series.

 

I don't think we can gather enough complete battle runthroughs from the player base to use machine learning on this. Given the wide variety of terrain and situations you'd need many thousands if not a million such playthroughs.

[CarlXII]

I don't think that  a self thinking AI is neccesary to wastly improve the combat mission single player experience.

Give us a trigger/scripting system simular to that of DCS world/ Eagle Dynaics and Combat Mission singel player gameplay will be elevated beyond recognition....

 

[SlowMotion]

5 hours ago, CarlXII said:

I don't think that  a self thinking AI is neccesary to wastly improve the combat mission single player experience.

Give us a trigger/scripting system simular to that of DCS world/ Eagle Dynaics and Combat Mission singel player gameplay will be elevated beyond recognition....

 

My limited experience of AI plans in current Scenario Editor is that making those AI plans work well takes a lot of work and that probably has a lot to do with how many 3rd party scenarios there are available.

Once you know how the current AI works, the experience is quite far from playing against a human player.  In H2H things do not proceed according to a plan that has been scheduled to proceed in certain way by the scenario designer. But like Redwolf wrote currently teaching the AI things like I suggested takes lots of input and that could be difficult to arrange.
Unless the workload could be somehow shared with a larger community. People could somehow store the turn files of their H2H game and if they agree sharing it with the company they could upload it for Battlefront's AI training purposes. If BF would agree that the game includes stuff that is useful for training the AI the game turns could be used. But certainly this won't happen during next few years.

Another way of using AI, which might be easier to achieve than the player AI, could be using AI in Scenario Editor. Something like it is done in this NVidia Canvas which is currently in beta testing phase. See title AI Assistant.  Scenario Designer could design things at large level and the AI would somehow fill the details so that you would not need to choose every map tile yourself.  And somehow make things like hills and roads faster. Maybe something like this could be done to make map making easier and faster.
https://www.nvidia.com/en-us/studio/canvas/

Edited November 1, 2023 by SlowMotion

[CarlXII]

29 minutes ago, SlowMotion said:

My limited experience of AI plans in current Scenario Editor is that making those AI plans work well takes a lot of work and that probably has a lot to do with how many 3rd party scenarios there are available.

You certainly are right on point here ...

The current CM scripting tools though are very, very, VERY limited and by todays standars...to be honest...very bad.

This results in a very limited AI performance ones we play the game and unneccesarely long developing times when it comes to designing scenarios....Requiring endless testing and tweaking to get the AI to perform as desired...Atleast for anything except for the simplest of scenarios.

This does not have to be the case though.

With decent scripting tools both scenario design and gameplay vs the AI will improve drastically. As i suggested above...The scripting tools from DCS world for example does not compare in any way with what we currently have in CM. It's a completally different world and the results can be absolutelly amazing scenarious with very high level AI forces.

These scripting tools are by no means rocket sience. They are simply just clever and fairly easy to work with...allowing for amazing flexebility in scenario design. CM is i'm sorry to say the complete opposit.

IIRC the few triggers  we currently have in CM today were ment to be expanded on over the years but that has not happened at all...sadly. We are left with a very limited system.

49 minutes ago, SlowMotion said:

Once you know how the current AI works, the experience is quite far from playing against a human player.  In H2H things do not proceed according to a plan that has been scheduled to proceed in certain way by the scenario designer. But like Redwolf wrote currently teaching the AI things like I suggested takes lots of input and that could be difficult to arrange.
Unless the workload could be somehow shared with a larger community. People could somehow store the turn files of their H2H game and if they agree sharing it with the company they could upload it for Battlefront's AI training purposes. If BF would agree that the game includes stuff that is useful for training the AI the game turns could be used. But certainly this won't happen during next few years.
 

Hopefully we will see an AI that is able to act on its own (in tactical combat computergames)...SOMEDAY...But i fear that that SOMEDAY is quite some years away yet...

Personally i don't want to wait many years yet until we get an improved AI and updated scenario editor in combat mission.

I would like to see that rather soon ...Adding a set of top notch scripting tools would be comparably simple i'm guessing.

56 minutes ago, SlowMotion said:

See title AI Assistant.  Scenario Designer could design things at large level and the AI would somehow fill the details so that you would not need to choose every map tile yourself.  And somehow make things like hills and roads faster. Maybe something like this could be done to make map making easier and faster.
https://www.nvidia.com/en-us/studio/canvas/

Improvements to things like map creating is pretty much vital i belive if we are to see any major increase in scenario output.

I agree...That ought to be doable.

 

[Centurian52]

3 minutes ago, SlowMotion said:

they could upload it for Battlefront's AI training purposes

I don't think Battlefront has that kind of AI. The kind of AI that you feed training data to is called a neural network (because the data structures involved have a passing resemblance to a very basic understanding of how biological neurons work). A well designed neural network, fed a sufficient amount of training data, can do things like accurately identify hand-written letters, or guess which pictures are and are not pictures of bees. Very large neural networks, which are fed massive amounts of training data, are called deep-learning networks. Some well known deep-learning networks have gotten very good at specific tasks such as, in the case of chatGPT, convincingly mimicking human language.

A neural network would probably not be the most efficient way to create a good wargaming AI. A neural net can't think multiple steps into the future. It can only make the decision with the lowest cost at this particular moment ("cost" in this context refers to the mathematical punishment/reward system that was used to train the AI). That works just fine for things like chatGPT, since all it takes to convincingly mimic human language is to respond appropriately to the most recent prompt. No memory of past prompts nor anticipation of future prompts is required. But a good tactical AI needs to do more than just anticipate the immediate consequences of a decision. It needs to be able to think several steps into the future. It needs to be able to plan. Neural networks (as they exist today) can't plan*.

A better approach might be to do something like the General Staff: Black Powder AI. It analyzes the battlefield (using pre-programmed methods (if you watch the video the narrator mentions a spanning-tree algorithm used to calculate frontages), not training data), breaks the situation down into a series of logical statements, and then deduces which courses of action it should take. In the video I linked the General Staff AI was able to identify an exposed flank and assign a unit to conduct a flank attack. For something like Combat Mission the AI would, for example, need to have a concept of fire-superiority. It would need to recognize whether or not it had fire-superiority, and know not to attempt to advance without fire-superiority (that would stop a lot of AI lemming charges).

https://www.youtube.com/watch?v=Y0K8DnS414o

*Which is not to say that neural nets don't have some really exciting possible applications. They can be trained to be extremely good at recognizing objects, so they are ideal for tasks such as spotting and identifying targets. And they will do that way faster than any human ever could. But they will have an error rate, and they will be stumped by any object that wasn't in their training data, so we'll still want a human in the loop to approve/disapprove targeting decisions for at least the next few years until all the kinks are worked out. You could also send such systems into areas that are known to contain no friendly or neutral targets, allowing it to engage targets without waiting for human approval.

[W1ll1am]

1 hour ago, Centurian52 said:

I don't think Battlefront has that kind of AI. The kind of AI that you feed training data to is called a neural network (because the data structures involved have a passing resemblance to a very basic understanding of how biological neurons work). A well designed neural network, fed a sufficient amount of training data, can do things like accurately identify hand-written letters, or guess which pictures are and are not pictures of bees. Very large neural networks, which are fed massive amounts of training data, are called deep-learning networks. Some well known deep-learning networks have gotten very good at specific tasks such as, in the case of chatGPT, convincingly mimicking human language.

A neural network would probably not be the most efficient way to create a good wargaming AI. A neural net can't think multiple steps into the future. It can only make the decision with the lowest cost at this particular moment ("cost" in this context refers to the mathematical punishment/reward system that was used to train the AI). That works just fine for things like chatGPT, since all it takes to convincingly mimic human language is to respond appropriately to the most recent prompt. No memory of past prompts nor anticipation of future prompts is required. But a good tactical AI needs to do more than just anticipate the immediate consequences of a decision. It needs to be able to think several steps into the future. It needs to be able to plan. Neural networks (as they exist today) can't plan*.

A better approach might be to do something like the General Staff: Black Powder AI. It analyzes the battlefield (using pre-programmed methods (if you watch the video the narrator mentions a spanning-tree algorithm used to calculate frontages), not training data), breaks the situation down into a series of logical statements, and then deduces which courses of action it should take. In the video I linked the General Staff AI was able to identify an exposed flank and assign a unit to conduct a flank attack. For something like Combat Mission the AI would, for example, need to have a concept of fire-superiority. It would need to recognize whether or not it had fire-superiority, and know not to attempt to advance without fire-superiority (that would stop a lot of AI lemming charges).

https://www.youtube.com/watch?v=Y0K8DnS414o

*Which is not to say that neural nets don't have some really exciting possible applications. They can be trained to be extremely good at recognizing objects, so they are ideal for tasks such as spotting and identifying targets. And they will do that way faster than any human ever could. But they will have an error rate, and they will be stumped by any object that wasn't in their training data, so we'll still want a human in the loop to approve/disapprove targeting decisions for at least the next few years until all the kinks are worked out. You could also send such systems into areas that are known to contain no friendly or neutral targets, allowing it to engage targets without waiting for human approval.

Another option (next to DL and conventional terrain analysis, as used in General Staff) is reinforcement learning. It has a more narrow application, but is easier to fit to a game like Combat Mission.
Here is an example, as shown at the ConnectionsUS 2020 on-line conference: 'Course of Action Generation with ML and a COTS Wargame', using the Flashpoint Campaigns game.
https://drive.google.com/file/d/1C9nfWwwlsTMkNN-au6j5UrTfDMr9v4KF/view?usp=drive_link

[Cepheus76]

I don't know if this was added to the wishlist before, but I would like to see that medics/ corpsmen added, and perhaps even introduce some kind of casevac system, where it would make sense in a scenario, like setting up a  casualty collecting point where casualties must be taken to to be collected. 

[Anthony P.]

Another couple of issues with AI learning for CM:

It would likely struggle to comprehend certain actions and applying them correctly. Faints, scouting and other sorts of "independent" actions where units go off on specialised missions away from their parent units would be things which the AI would struggle to place into context.

Learning from human players and/or drawing its own conclusions might cause the AI to adopt gamey, unrealistic tactics (e.g. one recent H2H game I played included facing the questionable tactic of sending a wave of empty halftracks just a few paces ahead of a tank attack to serve as a human shield/reveal positions by drawing fire).

 

AI learning might be useful, but more so in a "controlled environment" where BFC can teach it realistic tactics appropriate for different nations represented in the CM series.

[Centurian52]

27 minutes ago, Anthony P. said:

gamey, unrealistic tactics (e.g. one recent H2H game I played included facing the questionable tactic of sending a wave of empty halftracks just a few paces ahead of a tank attack to serve as a human shield/reveal positions by drawing fire).

That's actually not as unrealistic as you might think. I don't know of any examples of someone IRL trying that exact tactic. But people have tried all sorts of things to get the enemy to reveal their positions. I know of at least one instance in which a jeep was sent speeding down the main road of a town for the express purpose of seeing if any Germans would shoot at them, which is the exact sort of thing that you might call gamey and unrealistic if someone tried it in Combat Mission. But the truth is that Combat Mission is realistic enough that if you've thought to try it in game, odds are that someone has tried it in real life.

[Anthony P.]

A pet peeve of mine is certain aspects of armoured combat which could be improved upon.

Armour quality ought to be a factor. Soviet, and especially German tanks later during WW2 were well known for having brittle armour owing to a scarcity of resources. Tiger and Panther tanks routinely saw frontal armour which on paper should been impenetrable to 75mm projectiles fail catastrophically due to this: in CM, armour values seem to be taken at face value.

Targeting of speculative contacts (really this goes for infantry as well). Whether or not this is something that should only happen as player directed targeting, AI initiative or a mix of both I don't know, but realistically speaking tankers have frequently targeted suspected enemy vehicles. If you're fairly sure that there's a tank destroyer "over there", you're not gonna hold back on engaging it until you've confirmed whether it's a StuG or a Jagdpanzer: you'd start sending AP shells its way. Right now the best that can be managed is area firing 99% ineffective HE and MG fire towards it... which brings me to my last point:

Non penetrating hits should cause crew suppression. Historically this is another fact often seen in the case of well armoured tanks in cases when their armour did hold up to incoming fire, but the repeated heavy caliber hits slamming into the armour, causing sudden loud noises, reverberations, reminding the crew that the next hit just might find a weak spot, etc. often caused "significant emotional events". See T-34 and KV crews in Barbarossa panicking under a hail of technically ineffective 37mm fire, or Tiger and Panther crews retreating or even abandoning perfectly functional tanks when pummeled by 75mm HE and WP shells.
Heavy weapons hits should absolutely cause suppression. Currently tanks may reverse away after taking a hit from an unknown/dangerous opponent, but it never causes suppression. The whole "getting the first shot off" thing doesn't work out against opponents with capable armour: in fact, this is reversed from reality because getting the first shot off against a tank with superior armour only increases the odds of being spotted and engaged by it, when real life lessons conclusively indicate that 75mm Shermans/76mm T-34s getting the first shot off against Panthers and Tigers or 37/50mm Panzer IIIs getting the first shot off against T-34s and KV tanks would typically come out on top even if they failed to penetrate their opponent's armour.

The closest you can get to that is degrading the enemy tank's optics, but that only results in a relatively minor spotting penalty and requires a lot of hits. The idea that hits shouldn't suppress AFV crews unless they penetrate the vehicle is akin to if infantry would only be suppressed by enemy fire that wounds/kills someone in their team, and be entirely unaffected by the bullets and shells which misses them.

Edited November 14, 2023 by Anthony P.

[Anthony P.]

16 minutes ago, Centurian52 said:

That's actually not as unrealistic as you might think. I don't know of any examples of someone IRL trying that exact tactic. But people have tried all sorts of things to get the enemy to reveal their positions. I know of at least one instance in which a jeep was sent speeding down the main road of a town for the express purpose of seeing if any Germans would shoot at them, which is the exact sort of thing that you might call gamey and unrealistic if someone tried it in Combat Mission. But the truth is that Combat Mission is realistic enough that if you've thought to try it in game, odds are that someone has tried it in real life.

That's rather different though. The scenario you're describing is a fairly run of the mill kind of thing intended to simply figure out if there's an enemy precense in a fairly large/significant area at all (not a common scenario in CM), whereas I'm describing a scenario in which the presence of a British infantry company and a tank company was known beyond a shadow of a doubt, and the halftracks were sent literally mere paces ahead of the tanks, 1-2 halftracks dedicated per Panzer to draw fire away from them and reveal the specific location of individual British tanks to aid the Panzers in immediately engaging them.

Edited November 14, 2023 by Anthony P.

[Butschi]

1 hour ago, Anthony P. said:

Learning from human players and/or drawing its own conclusions might cause the AI to adopt gamey, unrealistic tactics

I wonder if I mentioned this elsewhere already... Anyway, a former AI dev for Stellaris once said when people were again complaining about "dumb AI" that it would be easy to build an AI that beats human players, e.g. by overwhelming the player with more attacks than he can process in parallel and stuff like that (works for sn RTS). What people want is not a smart and competent AI but a credible AI. One that behaves like a human instead of an AI. One that doesn't play gamey but like the human player imagines an admiral with certain personality traits would. Or in our case like a commander of a Soviet formation which having as an opponent should feel distinctly different from a US commander.

So it's often more about roleplaying. If you want a challenging/competitive AI, it has to play gamey in some cases because that's what humans will do.