Jump to content

AI plans and a more responsive AI


Recommended Posts

9 hours ago, MikeyD said:

When doing an attacker AI need you need to design-in logical tactics. Supporting fire, suppression of likely hot points, use of defilade, shifting position to avoid potential mortaring, etc. If you just tell 'em to 'assault' across an open field the AI will fail as assuredly as a human player doing the same thing would fail. Just now I was playing through the Beta of a CMRT module scenario, watching the Russian AI enter a town and capture a defended POW camp. That plan requires a LOT of moving parts. you can't just tell the AI 'BE SUCCESSFUL!'

A few things that imo makes designing attacking AI difficult...

Number one ! The inability of the AI to reevaluate the situation. The AI will NEVER change its attackplan...ever. Not on its own atleast. It has one way forward and one way forward only regardless of how the player defences are set-up. A skilled designer may be able to design the AI plans in such a way that it seeems as if the AI is adjusting its original plan when running into strong resistance. To be able to do this the designer will pretty much have to 'guess' right though as to what the players defence set-up will look like. If he guess wrong the outcome will simply look wierd...if the AI abandons a succesful attack !

Number two...The limited number of AI groups. For a reinforced company sized attack 16 AI groups may well be enough but for a reinforced battalion it is a bit on the low side. Unfortunatelly...If the player commands something like a company sized force then the attacking AI will pretty much need atleast a battalion to provide much of a challange...unless the forces are very unballanced. I belive that one of the reasons for the AI attacks currently often looking like suiccidal human wawe attacks are indeed the lack of sufficient AI groups.

Number three...perhaps not so much of a problem...but to a degree atleast...the low tempo of an AI attack...atleast when conducted over somewhat larger distances. imo ones the AI units gets pinned down it can often take quite some time for them to recover and move forward again...Far longer then it would take a human player to get the same units moving again...They sort of 'get stuck' it seems...

number four...The limited ability of the AI to get HE and smoke on the right location at the right time...(has improved somewhat with the on-map mortar trick)

 

Link to comment
Share on other sites

21 hours ago, Freyberg said:

In an attack-defend scenario, this will give you a very good static defence, and with a suitable map, will give you a fun and challenging Quick Battle. The AI will allocate the groups very intelligently and will create an integrated network of defensive positions, there may be interlocked fields of fire, AT guns will be well sited, avenues of approach will be covered, and it may place units as bait. It seems to have an excellent 'understanding' of the relationship between terrain, objectives and setup zones.

I'm sorry, but that is just not true. The AI has no understanding of any of those things. It does not understand how set up AT guns in good positions, make interlocking fields of fire, use bait, or anything else like that. It doesn't even keep squads from the same platoon close to each other or in contact with their HQ.

The only thing that the AI does when deciding when to place units in the painted setup zone is that it will look for positions with cover. So if you paint a whole field, including a hedgerow, infantry will always get placed along the hedge, and never out in the open.

But of course, if you have a lot of units and paint big setup zones, you'll get a good spread of enemy positions. Some of them will be in great locations, but that is completely down to chance. And especially if you set them to ambush orders, you can get a randomised setup that will pose a challenge and might even surprise you. However, you are likely to still meet AT guns in the middle of forests, etc.

Link to comment
Share on other sites

21 minutes ago, Bulletpoint said:

I'm sorry, but that is just not true. The AI has no understanding of any of those things. It does not understand how set up AT guns in good positions, make interlocking fields of fire, use bait, or anything else like that. It doesn't even keep squads from the same platoon close to each other or in contact with their HQ.

The only thing that the AI does when deciding when to place units in the painted setup zone is that it will look for positions with cover. So if you paint a whole field, including a hedgerow, infantry will always get placed along the hedge, and never out in the open.

But of course, if you have a lot of units and paint big setup zones, you'll get a good spread of enemy positions. Some of them will be in great locations, but that is completely down to chance. And especially if you set them to ambush orders, you can get a randomised setup that will pose a challenge and might even surprise you. However, you are likely to still meet AT guns in the middle of forests, etc.

Yeah...this sounds more like the AI i'm used to seeing in regular QBs...🥴

Freybergs suggestion might provide a slight improvement over regular QBs though as the 'designer" has some control over wich unit he places in wich AI group...

The downside being stationary units...

Link to comment
Share on other sites

1 hour ago, Bulletpoint said:

I'm sorry, but that is just not true. The AI has no understanding of any of those things. It does not understand how set up AT guns in good positions, make interlocking fields of fire, use bait, or anything else like that.

We must be playing a different game...

I see these things all the time. It does depend on setting up a QB map the way I described it though. One of the reasons I started playing with AI plans was that I was dissatisfied with some of the QB maps I'd played on.

Edited by Freyberg
Link to comment
Share on other sites

33 minutes ago, Freyberg said:

I see these things all the time. It does depend on setting up a QB map the way I described it though. 

One way to expand on your original idea might be...

Instead of painting pretty much the entire defensive zone for each AI-group...One could perhaps be a little more 'picky' 😎

That is...

Make one AI group (or a few)for AT-guns, one group (or a few) for HMGs, simular for perhaps armour, tankhunters and finally regular rifle squads...

Instead of painting the entire defensive zone...for the AT-gun group you only paint every good and decent area of the defensive zone sutable to deploy AT-guns in...And perhaps if you want to even some less good possitions.

Do the same for the HMG group...paint pretty much all of the good and decent possitions to site and HMG in.

And go on like this for all the groups...

This will give the AI a pretty high level of freedom but still help it avoid the really stupid, wierd set-up choises for the different weapons.

 

Link to comment
Share on other sites

13 minutes ago, RepsolCBR said:

One way to expand on your original idea might be...

Instead of painting pretty much the entire defensive zone for each AI-group...One could perhaps be a little more 'picky' 😎

That is...

Make one AI group (or a few)for AT-guns, one group (or a few) for HMGs, simular for perhaps armour, tankhunters and finally regular rifle squads...

Instead of painting the entire defensive zone...for the AT-gun group you only paint every good and decent area of the defensive zone sutable to deploy AT-guns in...And perhaps if you want to even some less good possitions.

Do the same for the HMG group...paint pretty much all of the good and decent possitions to site and HMG in.

And go on like this for all the groups...

This will give the AI a pretty high level of freedom but still help it avoid the really stupid, wierd set-up choises for the different weapons.

 

I tried that at first, but it didn't seem to be necessary.

As long as you give the AI plenty of groups with different settings, it really does seem to make excellent choices about where to position units.

I think one of the problems with some QB maps is using only a small number of groups - I don't know why, but at least 4 or 5, preferably more (I usually have at least one of each of: active, normal, cautious, ambush 1000m and maybe a hide group or some other ambush groups) even if they are very general, does seem to give a much improved game.

 

Link to comment
Share on other sites

1 hour ago, Freyberg said:

We must be playing a different game...

I see these things all the time. It does depend on setting up a QB map the way I described it though. One of the reasons I started playing with AI plans was that I was dissatisfied with some of the QB maps I'd played on.

That was my reaction when I read your post, too.. "We must be playing two different games" :)

I think it's like seeing faces in the clouds. Where I see random enemy placements, where one team is randomly in a good position, and another randomly placed in a bad position, you might see one position controlling the high ground, while the other team is deliberately placed too far forward as bait.. etc. We humans like to find patterns in chaos.

Edited by Bulletpoint
Link to comment
Share on other sites

5 hours ago, RepsolCBR said:

Number one ! The inability of the AI to reevaluate the situation. The AI will NEVER change its attackplan...ever. Not on its own atleast. It has one way forward and one way forward only regardless of how the player defences are set-up. A skilled designer may be able to design the AI plans in such a way that it seeems as if the AI is adjusting its original plan when running into strong resistance. To be able to do this the designer will pretty much have to 'guess' right though as to what the players defence set-up will look like. If he guess wrong the outcome will simply look wierd...if the AI abandons a succesful attack !



yes, this is the point I was trying, and failing, to make. I don't find it a constant issue in the WW2 games as an aggressive attack can succeed, however in the modern games it essentially dooms many attacks before they even really begin.



RE: playing different games

I wonder if its not related to the maps being played? If anyone played FEAR, a first person shooter from 2005 or so, they'll likely have heard that the AI is superlative. However, in reality the AI isn't anymore advance than just about any other FPS. What the designers did is specifically design the maps for the AI to take advantage of. If placed in any other terrain they would be hopeless but in the very specific maps made for FEAR the AI was stellar.

Within the CM context specific map design might give you much "smarter" ai deployments than others. I notice in Freyberg's post for example there aren't any deep woods that I can see and the roads appear to be relatively open. Perhaps that is influencing how good the deployment is in some way. Because when I play QBs I often see absolutely terrible deployments. Critical AT guns in the middle of deep forest, directly behind a house with no LOS, etc... It might also have to do with the size of the battles being played. If you give the AI more units then its going to have more chances to get it right and the fact that the 88mm gun is in the ass end of nowhere isn't as noticeable or critical.


https://en.wikipedia.org/wiki/F.E.A.R._(video_game)

Link to comment
Share on other sites

QB is the lowest form of AI orders design for the reason that you don't know if you're writing movement orders for a platoon of infantry, a company of supertanks or a half dozen flamethrower half-tracks. QB AI orders necessarily need to be broad strokes, get the unknown unit to cross the map and come in contact with the enemy then let the chips fall where they may. Scenario AI orders design is another matter.

A particular scenario may play badly because its a bad scenario, not because the AI engine is faulty. When the old Shock Force game got refreshed for CMSF2 some of those 1st generation scenarios were from the dawn of the CM2 game engine. Back then nobody was experienced with the editor. There was no precedent to follow and no gurus to consult. A few early scenarios were little more than a single 'attack' order into the center of the map. Now there's a deep knowledge base on how to do proper AI editor work. Today if you do sloppy AI orders for a scenario with the tools you got that's on your head.

Link to comment
Share on other sites

8 hours ago, RepsolCBR said:

A few things that imo makes designing attacking AI difficult...

Number one ! The inability of the AI to reevaluate the situation. The AI will NEVER change its attackplan...ever. Not on its own atleast. It has one way forward and one way forward only regardless of how the player defences are set-up. A skilled designer may be able to design the AI plans in such a way that it seeems as if the AI is adjusting its original plan when running into strong resistance. To be able to do this the designer will pretty much have to 'guess' right though as to what the players defence set-up will look like. If he guess wrong the outcome will simply look wierd...if the AI abandons a succesful attack !

Number two...The limited number of AI groups. For a reinforced company sized attack 16 AI groups may well be enough but for a reinforced battalion it is a bit on the low side. Unfortunatelly...If the player commands something like a company sized force then the attacking AI will pretty much need atleast a battalion to provide much of a challange...unless the forces are very unballanced. I belive that one of the reasons for the AI attacks currently often looking like suiccidal human wawe attacks are indeed the lack of sufficient AI groups.

Number three...perhaps not so much of a problem...but to a degree atleast...the low tempo of an AI attack...atleast when conducted over somewhat larger distances. imo ones the AI units gets pinned down it can often take quite some time for them to recover and move forward again...Far longer then it would take a human player to get the same units moving again...They sort of 'get stuck' it seems...

number four...The limited ability of the AI to get HE and smoke on the right location at the right time...(has improved somewhat with the on-map mortar trick)

 

These are problems yes but can you name a game with AI that does any of these things? Everything listed here is an example of abstract thinking the sort of which no machine in existence is capable of doing yet or presumably the world would be a very different place right now. What you may think of as examples of games which can do the above concepts probably don't do anything like evaluation involving the sort of abstract thinking only a human mind can do. Any video game you've played that can do these things is in fact obscuring its inability to actually perform these concepts through the rest of the game's tools. 

Video game developers use all sorts of tricks to create compelling and intriguing challenges especially when the opponent is a human who possesses the insurmountable advantage of his human brain. For instance lots of games give the artificial intelligence huge stat buffs like a big health pool, damage buffs, borg communications, spawning enemies out of the player's line of sight, spawning enemies endlessly (ie: "clown car'ing"), granting them knowledge of player dispositions and stats, +2 on every dice roll etc. Basically the favorite method in many games is to build cheats into the AI. In Combat Mission this tends to lead to the infamous cases of overpacked maps with huge, in fact ridiculous defender headcounts vs the human attacker. While there is a time and a place for this and it's not necessarily a bad thing it is none the less a rather crude and inelegant solution to the AI's deficiencies. There are other ways to create challenge and intrigue.  

So no the AI cannot conduct self assessments, but you can use tools in the games to emulate that. You can restrict some plans to semi-completion, or have some other plans exist to implement what is describable as a "sub optimal" solution to the other side's plan. If you use the tools presented by the game successfully, it won't be necessary to think of such highly granular concepts like situational awareness or the limited AI plans available. Think of the game's many facets, like how the scenario is scored, how the terrain of the map influences each side's thinking, how the ToEs and tools available to each side can influence the computer's plan vs a player's plan. I'm not trying to suggest that there is any ideal or "right" way to go about designing scenarios but it's a bit much to expect BFC to make the Combat Mission games and also make us Skynet. 

Edited by SimpleSimon
Link to comment
Share on other sites

18 hours ago, MikeyD said:

you can't just tell the AI 'BE SUCCESSFUL!'

Could we add it to the list?  ;)

My last gasp for good old CM:SF, 'Al Hutumah' (The Wrecker), is an attempt to produce a sustained, realistic and convincing AI attack, using just eight AI slots and no triggers or exits.....Still working on it.  :rolleyes:

Link to comment
Share on other sites

40 minutes ago, SimpleSimon said:

These are problems yes but can you name a game with AI that does any of these things? 

While not having the capabilities of a supercomputer....one game that comes to mind that has an absolutelly amazing  missioneditor (AI) would be...

DCS World by Eagle Dynamics...the greatest flightsim every made 😁...

The capabilities of that missioneditor is just...WOW !!

The game uses a very wide range of various triggers, flags, conditions and results designed in a very clever way to allow the entire gameworld to act and react in a very belivable and intresting way...

The game does not relly on the AI to be able to make all these decissions on its own...because of the multitude of various triggers and resulting actions the scenario designers are able to 'help' the AI to act in a fairly realistic, reactive and challeging way...😎

It it could be at all possible to get something simular in CM...the capabilities of the AI would increase massively...a whole new level !

All without the AI needing to grow a superbrain...

The editor works decently...yes ! But theres always room for improvements...

A simular AI programing feature would indeed be...amazing !

Link to comment
Share on other sites

Fundamentally the issue is that the strategic AI does not exist. Only tactical AI and that AI is only reactive never proactive. What we all call the strategic AI is simply the designer creating from scratch a per scenario AI that is going to follow a rote path. The real limiter on scenario design is that the designer is pretty limited on their ability to build scenario specific AI. Cool things can be done with it but its still limited.

2 hours ago, SimpleSimon said:

Everything listed here is an example of abstract thinking the sort of which no machine in existence is capable of doing yet or presumably the world would be a very different place right now.

Its not abstract thinking - you just consider it abstract thinking because @RepsolCBR described it using human language. You can achieve RepsolCBR's "abstract thinking" via logic gates. The scenario designer is doing this right now its just that the options are limited. Below for example is a series of options that can be done in-game currently. #5 could be construed by some people as abstract thinking but its really just a trigger.

1. The AI is to move at a cautious speed
2. AI is to attack starting from POINT Alfa to Point Bravo with GROUP 1
3. The AI may call artillery on LOCATION November
4. After arriving the AI will delay 20 minutes
5. The AI will attack from Point Bravo to Point Delta if enemy troops detected at LOCATION NOVEMBER


Now if you had more detailed and granular options with the inclusion of SOPs and some free flow from the AI you could get more natural reactions.

1. The AI is to move at a cautious speed
2. AI is to attack starting from POINT Alfa to Point Bravo with GROUP 1
3A: If GROUP 1 sustains 50% casualties reroute GROUP 1 and GROUP 2 TO POINT CHARLIE.
3B: Once POINT CHARLIE is reached proceed to POINT BRAVO
3. The AI may call artillery on LOCATION November
4. After arriving the GROUP 1 will delay 20 minutes
4A: If GROUP 1 casulties > 40% cancel all orders proceed to DEFENSE POSITION
5. GROUP 1 will attack from Point Bravo to Point NOVEMBER if enemy troops detected at LOCATION NOVEMBER

All of the above could again be done within the bounds of the scenario designer system in the game currently. Just a bunch of triggers.


And again if you gave the AI some free hand you might not even need to do all of this.

1. The AI is to move at a cautious speed
2. AI is to attack starting from POINT Alfa to Point Bravo with GROUP 1
AI LOGIC: If GROUP 1 engaged by infantry ClOSER than 200 meters TARGET BRIEFLY 50% of buildings within 200 Meters along route of march. 100% of buildings that contain contact marker

Now you have the AI automatically reacting to a group threat. A single unit in the group has detected an infantry threat at close range and now the entire group is going to react via searching fire. If the AI suspects that a unit is there (a contact marker) it will definitely fire at the position.


AI LOGIC: If GROUP 1 Casualties > 50% reroute remainder 400 meters Left/Right to area of greater cover and attempt to move to AI POINT CHARLIE and then to POINT BRAVO

And the AI will no longer continue a suicidal attack down a single avenue of approach but attempt another route. All it does it say "where is more cover" and then add its own intermediate waypoint between ALFA and BRAVO


-----

All of these examples have something in them that you might consider "abstract thinking" but none of it actually is. Its all just a series of decision points that the AI arrives at. The difference between CM and many other games is that there are no decision points for the AI at all. Any decision is inputed by the scenario designer. You don't necessarily need the AI to make its own decisions and the CM system is fine. But as we;ve seen over the years the more power you give the designer the smarter the AI can appear.

I've done both some Arma scenario design and Combat Mission and the Arma system has no real strategic AI either and is handled by player triggers. The system is far more complex and as a result you can get far more complex results from it if you put in the time.

Edit: I will say I've done some game AI design and none of its simple. But you do not need an AI capable of abstract thinking. You just need to spend enough time working with it so that the maths behind each decision makes sense most of the time. Abstract thinking really only comes into play when you want a universal AI but no one here is talking about that. This is all just bespoke AI that exists solely to play Combat Mission - you can do that and do it quite well with the technology on hand. The issue is that building it is going to eat up time. Which is, I assume, why we have the scenario designer doing AI programming. Which again is fine, but the more options you give the designer the better the resulting AI can be. Any of the recommendations people have made could be done with a designer system there are just insufficient options for it currently.

Edited by com-intern
Edit
Link to comment
Share on other sites

Yepp...things like what com-intern mentions above is in line with what i'm hoping for...

Things like a UNIT IN ZONE trigger gives simular options as a casulty-level trigger...

An  AI T-34 platoon advances towards a river crossing...Suddenly a  player JagdPanther is spotted in reverse slope possition on top of a hill overseeing the crossing...

Instead of simply continuing on its path across the river...The player JagdPantherer has been mark by a unit objective and ones spotted in the terrain-zone covering the hill the UNIT IN ZONE trigger instructs the AI T-34 platoon to halt its advance and instead head for a secondary crossing point...

The  AI will not need to understand what is happening on its own...The trigger that has been put in place by the designer 'helps'...the AI to make a rather good decition 😎...

 

 

 

Link to comment
Share on other sites

It's me Again 😉...

Another thing that perhaps could be possible with a decent number of various triggers would be the possibility to get the AI off to a good start...

As of now every scenario can have multiple AI plans...what if the AI had the capability to 'chose'  AI plan at the start of each game...after the player has finisished his deployment and clicked GO !

The scenario designer lists one or a few conditions that needs to true in order for the AI plan to be selected by the AI...

Using avaliable trigger options...

Things like Unit in zone, unit out of zone, unit value in zone higher then (using for example QB battle purchase values for units)...armour in zone, armour value in zone higher then etc, etc...

If the conditions for the first AI plan is met then that plan is chosen by the AI...if not then the conditions for the second AI plan is checked...And so on...if no plan clears the check the the first plan is selected...or perhaps a random one...

This would allow the AI to do some nice 'cheating'...to help it decide if the overall plan should be a left flank attack or a right flank attack or an attack up the middle...

Dependant on how the player has set up his defences.

Simular in an AI defence scenario it will give the AI some needed initial help...

It's cheating...yes ! But...comes on...It's the AI...it needs it 🙂...it's not a H2H game...

This way the AI might be a somewhat more challeging opponent...

Not that the AI cant be challeging right now !

It can...but a little more help never hurt the guy i belive 😁...the level of help the AI will get can be pretty well controlled by the designer to not be...over the top..

 

 

Link to comment
Share on other sites

1 hour ago, com-intern said:

Its not abstract thinking - you just consider it abstract thinking because @RepsolCBR described it using human language. You can achieve RepsolCBR's "abstract thinking" via logic gates. The scenario designer is doing this right now its just that the options are limited. Below for example is a series of options that can be done in-game currently. #5 could be construed by some people as abstract thinking but its really just a trigger.

That's fair enough. I just think it's worth asking how badly we need the tools capable of programming in more sophisticated AI routines when it seems that just featuring more of what we've got would suffice. Your own point seems to reinforce that as well unless i'm misunderstanding something. I certainly have nothing against better mechanisms for influencing the AI, I would just like to know what the expense of these features are and how that would fit into an overall development cycle. Workarounds are not ideal, but how inconvenient are they? 

1 hour ago, com-intern said:

But you do not need an AI capable of abstract thinking.

Which is exactly my point as well. Is it entirely necessary for CM to feature such sophistication? How much of how many scenarios are essentially "move here shoot that"? 

Link to comment
Share on other sites

 

17 hours ago, SimpleSimon said:

I just think it's worth asking how badly we need the tools capable of programming in more sophisticated AI routines when it seems that just featuring more of what we've got would suffice.

Essentially yea. The game doesn't need any actual AI but if you give designer more options then you could have much more competent scenario AI.

Triggers based on % casualties for an AI group, AI group SOPs and so on. Essentially ways for the scenario designer to say IF THIS THEN THAT for their plans. It'd make the AI more resilient and make scenario design far easier. As right now good design requires a ton of play through to get a feel for "most likely outcomes" and sort of prediction of what players will do. For example, I made an AI attack scenario a few years ago and essentially watched and recorded the results of a ton of slight variations of an attack. Like most of the AI creation time was just trying to figure out how well the attack went and then trying to see into the players mind. If I could have just said "DO this attack until X then do Y" it would have made a more competent opponent and not taking so many hours.


---

If you look at my second example in my long post you can kind of see what I'm talking about. Essentially a trigger for % casualties suffered by AI group opens up a lot of options we don't current;y have. It isn't extremely different but each additional trigger allows for increasingly complex scenario AI design.

Edited by com-intern
edit
Link to comment
Share on other sites

We don't even need very sophisticated AI routines. We just need the current AI to be improved slightly in certain key areas.

Example: In the editor, I can give an AI platoon orders to advance in bounding leaps. However, I can't decide how long each leap will be (and they are usually way too long). Also, teams won't stop in places of cover along the way. They will run right past a low wall or hedgerow to go prone out in a field.

So, the code that gives the teams the individual movement orders should do a quick check to see if the movement path intersects any good cover. If so, the movement should stop at that cover until next movement order.

Compared to what else goes on in the game code, such a check is easy to do. The code for the AI to evaluate cover is actually already in the game, it's just not used very much. You can see this for yourself if you do a scenario and paint a setup zone for infantry across open ground that includes some cover. The truppen will deploy in places of cover.

Edited by Bulletpoint
Link to comment
Share on other sites

2 hours ago, Bulletpoint said:

 

Example: In the editor, I can give an AI platoon orders to advance in bounding leaps. However, I can't decide how long each leap will be (and they are usually way too long). Also, teams won't stop in places of cover along the way. They will run right past a low wall or hedgerow to go prone out in a field.

If the AI can be improved to handle the terrain and various threaths from varios directions that would be great !

Another solution that i have been advocating for for several years now is....More AI groups 😎

Let the scenariodesigner 'help' the AI with these kind of tasks...

Instead of the platoon being ONE single AI group...let it be two...let the platoon/company supporting weapons be their own AI group (or several) etc and not be mixed up with regular rifle squads...

It's not only infantry that has trouble doing tactical movements in a 'very good' way if asigned to the same AI group....The same thing applies to armour also...

as well as other supporting vehicles...Stummels and the likes...

If such assets also could be their onw AI group it would be a good thing.

An armour platoon consisting of 2, 3 or even 4 AI groups would be able to do a far better job then if asigned to a single group imo...

Especially now with the addition of areafire, face and withdraw commands.

Imo it would also be far easier to program the AI if we where allowed to use a greater number of AI groups...less tweaking, less testing 

 

Link to comment
Share on other sites

3 hours ago, Bulletpoint said:

We don't even need very sophisticated AI routines. We just need the current AI to be improved slightly in certain key areas.

Example: In the editor, I can give an AI platoon orders to advance in bounding leaps. However, I can't decide how long each leap will be (and they are usually way too long). Also, teams won't stop in places of cover along the way. They will run right past a low wall or hedgerow to go prone out in a field.

So, the code that gives the teams the individual movement orders should do a quick check to see if the movement path intersects any good cover. If so, the movement should stop at that cover until next movement order.

Compared to what else goes on in the game code, such a check is easy to do. The code for the AI to evaluate cover is actually already in the game, it's just not used very much. You can see this for yourself if you do a scenario and paint a setup zone for infantry across open ground that includes some cover. The truppen will deploy in places of cover.

The AIP/TacAI is only concerned with paths of least resistance where it can go from point A to B the fastest. Its appreciation of cover is also a little different. It highly prefers cover as given from terrain mesh. Craters, road ditches, dips and bumps in terrain and such. Same for the Evade routine which seeks to put cover (LOS/LOF obstruction) terrain between enemy threats and evading unit. Think distance where the TacAI infantry starts leapfroggin is at ~80m something (advance/assault modes). Hard to work with, but for Plt size formations (and groups) one at least can achieve some predictable results. Unlike vehicle groups/formations which remain a major annoyance in the AIP´s hands.

Link to comment
Share on other sites

On 5/27/2020 at 11:28 PM, Bulletpoint said:

That was my reaction when I read your post, too.. "We must be playing two different games" :)

I think it's like seeing faces in the clouds. Where I see random enemy placements, where one team is randomly in a good position, and another randomly placed in a bad position, you might see one position controlling the high ground, while the other team is deliberately placed too far forward as bait.. etc. We humans like to find patterns in chaos.

I get your point :). It may well be that the AI algorithms are not complex, but they are certainly clever.

Here is a very typical result from a game I'm playing against the AI now:

AI_setup_crossroads.jpg?dl=1

My scouts easily saw the tank just behind the crossroads (now a burning wreck) - to which there were two good avenues of approach. Expecting a trap, I threw everything at it and took both. I don't know if I've seen everything yet, because the game is still in progress, but each group of tanks had to come over a slight rise, leaving them vulnerable to temporary numerical disadvantage and visibility disadvantage.

The ambush was cleverly done. I expected it, threw everything at it, and still lost two tanks and an immobilisation just to kill one tank and take out a lightweight AA gun (so far).

I know that none of this was engineered by the designer, because I was the map designer and I gave no thought at all to any of this when I created the map or selected the slice, and the AI plan was one of the very simple ones I mentioned above. This very enjoyable and quite difficult game is all the result of the AI. This kind of clever set-up is the norm. In the maps I've been doing myself I never see a random defence that is easily overcome.*

Now it's possible that the AI just has a very simple set of rules for placing units. I don't believe that - my guess (and it is just a guess) is that it's quite sophisticated. Either way, the end result is very good. And my original point was that the AI wouldn't need major changes to be made more responsive. If it simply used the calculations it's already doing, but directed them actively at general areas of enemy concentrations, I think it would most formidable.

(*Having said that, whatever the algorithm is doing, it seems to work best with multiple AI groups, even very simple ones.)

 

Link to comment
Share on other sites

This was another fun battle on an urban ruins map. :)

Here is just a small part of the action, but the whole thing kept me engaged for hours and was actually very very difficult, with a string of mutually supporting enemy positions - every time I dealt with one and went to assault the other, another would reveal itself.

In this small slice of action, I was trying to get a good angle on a minor strongpoint...

urban_ruins_ai.jpg?dl=1

It was fun and infuriating. In the end, I couldn't beat the AI with just infantry tactics, and I had to resort to using on-map 81mm mortars, which I had hoped not to have to use.

When I designed the map, it was purely to try make it 'look cool' - I gave no thought whatsoever to what the AI or players might do. In the event, it surprised me over and over again.

So it's possible the AI is placing units more-or-less at random, but it doesn't feel like it to me :)

(I didn't take any screenshots while playing, so this is from memory with screenshots of the map in the editor)

Edited by Freyberg
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
Unfortunately, your content contains terms that we do not allow. Please edit your content to remove the highlighted words below.
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.

×
×
  • Create New...