List your AI complaints (AI suggestions - Thunder is watching) Master list on page 1

[UberBugz]

And what's so wrong with having the AI cheat a bit? Do you folks think that Relic or even the best genius in the world can come up with an AI that plays like a human?

I can recall putting some cheat code into the 'Stock_Infantry_Only' mod. If the AI saw that jeeps/motorcycles didn't exist, the AI would end up getting a huge push to build snipers. Of course, this was coupled with a random factor. If things went bad for the human player with no jeeps, they'd probably end up with a heavy axis sniper onslaught. And even then, pure randomness would have caused this regardless. Some games would be sniper heavy by default, other games would need the lack of jeeps/motorcycles to produce the sniper onslaught. Got milk?

Nothing wrong with having the AI cheat if it's controlled. At least controlled to the point it gives a mostly 'human-like' play experience. After all, if you saw your human opponent lacking in units that could detect a sniper, would you not be inclined to build a sniper to exploit that weakness? Yes/No? Maybe Relics dynamic AI is supposed to handle this sort of thing, but it clearly needs a bit of help.

Hopefully, Relic won't do something stupid like "hard-coding" the AI changes they might make. Color me paranoid, but I smell this kind of thing coming. Yep!

[BlackOmne]

Keggan,
Next you'll be wishing the AI was controlled by a HAL9000. The AI just isn't that smart to coordinate units, and unless Relic puts some serious work into the AI, I doubt you'll ever see. The best that can be hoped for is that a mod or a future Relic patch achieves building a good combined arms force, well balanced with no spamming from any type of unit including doctrine call-ins, and that force falls into nicely synced mass of units going after the same objective at the same time.

[Muad'Dib]

Can't the AI be programmed to consider mathematical analysis and priorities? (I know didly-squat about programming BTW, ergo the question) Like for example, whatever the AI sees through the FoW, it accounts into percentages. Like if there are 5 enemy squads and 10 enemy tanks, it'll know that AT weapons have to be made since there are 66.33% more tanks in the field.

Can't it have a list of priorities as to what combinations to use upon detecting a certain percentage of enemy units?

As for combined arms, can't it be programed to always group together certain units that we all know go better with each other. Like, MG42s and volks for example.

I know the AI can't be programmed to think like us but I'm (sorta) sure that mathematical probabilities aren't that impossible, just so that the AI won't make units that are only good for fodder.

Think chess AI...the AI actually responds to your moves and considers the best and most probable counter to it. Even the simplest chess AI doesn't try to go for a check without calculating its losses in the process.

[cf_nz]

I don't see why an algorithm couldn't be written to allow the AI to respond appropriately to certain situations (mimic a humans responses), the main problem I think is going to be the time and resources it would take to do this.

[Muad'Dib]

I don't see why an algorithm couldn't be written to allow the AI to respond appropriately to certain situations (mimic a humans responses), the main problem I think is going to be the time and resources it would take to do this.

Definitely, but Relic is a game development company. They should have the time and resources to pull something like this off. If they don't, then they should at least freelance it to someone who can. I suppose it would be like expanding the chess AI to about a 1000x more in term of situations, calculations and priorities...

If ever this does happen, whereas the AI can actually react accordingly to the enemy's movement, then team games will need to be re-thought. Any crappy player will suddenly get the boot since the AI can play better than him....

[BlackOmne]

Come on guys, you can't compare a 3D game to chess. Chess is very, very, very limited compared to CoH. You have a board or "map" that has only 64 distinct locations; CoH has maps that have billlions of distinct locations - pick any point on the map, either 1 virtual foot away or 12 or whatever. That alone greatly simplifies programming an AI for chess. Muad, it's more like expanding the chess AI by 1,000,000,000,000x, which would bog your computer down to a snail's pace or slower. Then there are the units. Every game has exactly the same 16 pieces for each side; they never change; one game doesn't allow a player to spam 12 pawns plus a king, queen, biskop and rook. CoH unit combination possibilities are practically infinite. in chess, the "map" is always the same; in CoH, the maps could be endless between official ones and ones built by mappers.

Now, CoH AI does seem to have the capability to recognize that if the opposing player (AI or human) builds a tank, it needs AT capability if it doesn't have any already or a scout unit if a sniper appears. One of the AI modders can probably confirm this. The build priorities and limits can be programmed into the AI and modded. It's conceivable that Relic could code some ability for the AI to build small groups of combined arms units (I imagine it could be done along the lines of attaching an LT to a tommie squad), but that doesn't exist except for the brit situation (and I'm not even certain it does it there - perhaps the LT and command tank just happen to be moving in the same direction as there subordinate units).

But, you'll never see within the near future or this game the ability of the AI to recognize the map with relation to enemy units and use it effectively. From my experience, the AI player moves it units by issuing attack-move orders to either cap a point or attack a unit seen by another unit. It also issues order to build buildings, emplacements, bunkers, OPs, etc. Building mines can be done around points but the AI doesn't know how to recognize a gap in a wall or hedge or whatever to place mines there. The AI sees the dozen or so points on the map and buildings and enemy units and lets the pathfinding routine take care of getting the units there. It issues orders to units on a unit by unit basis with no coordination. Only when the orders are the same for the units equidistant (by both distance and unit speed) from the objective do the unit "fall into sync" and appear to be attacking the same area together. Once they get there, there is little tactical coordination or thought such as flanking manuevers. Yes, the AI knows how to skirt the edge of your units visibile range and will try to avoid any cone of fire (eg, MG, ATG, etc) that your unit may have, if possible, but it just as likely or more so to send two units around the same side of your units sight range as to send one unit around each side (in flanking).

The AI doesn't attack you. It mobs you. And that is why mp bonuses help the AI to win, it builds a bigger mob sooner within the pop cap.

Overall, within the bounds what AI could accomplish on the fastest PCs, Relic has done a decent job with the AI. Now, they only need to improve the build order and unit restrictions (and tweak other AI factors like aggressiveness that modders like Warnstaff have), and we can hope that they could add code for grouping units and simple coordination between two units (like sending one around side of target and another unit around the other) for most basic of attack manuevers, flanking. That is about all we can expect. Even expecting the Brit AI to intelligently (considering fields of fire and likely routes of attack) build a good defense network around a VP or high-resouce point it probably expecting too much.

[UberBugz]

BlackOmne,

So what are you trying to do? Kill this thread completely?. Stop creating posts that are near impossible to argue with. LoL!!!

Anyway, Relic might consider adding an extra 'add-on' to (Axis/Ally) halftrack units that give them easy access to becoming an offensive weapon during skirmish games. Much the same way the CW bren carrier can be upgraded with a vickers. I'm not sure about anyone else, but I've never seen the AI actually load a halftrack with units within a skirmish game so that it would become an offensive unit (Frontal MGs for Ally/Two LMG42s for axis). Usually, halftracks would bunch up near the HQ waiting for the more costly offensive upgrades (Walking stuka/Quad). If the halftracks could be upgraded more readily with simple weapons, then at least they'd be used by the AI as probably nothing more then a stronger version of the jeep. It'd definitely reduce the HQ "bunch up" effect.

Heh, I've personal experience on why this would work!. I should also mention that this addition won't hurt the online community at all. Either you'd spend the time and munitions to make the upgrade, or just load your halftrack with infantry. Either way, the Skirmish AI can get what it needs, and not overpower the human on human side at the same time. Yes/No?

NOTE: This simple (MG) upgrade wouldn't collide with the more powerful upgrades either. Just because a halftrack is armed with MGs, doesn't mean it can't be armed with a Quad or Flamethrower. See?

[BlackOmne]

LOL! I'm not saying Relic doesn't have room for improvement with what they have already, just that we can't expect any major code capabilities like the AI understanding the "lay of the land" for each map or putting together grand tactical schemes.

You make a good example of something Relic could improve. The PE MG HTs work just fine. Why not make the US and Wehr HTs not require a adding another unit to use available MGs on the vehicle? Give them a built in gunner like the PE HT; alternatively, make the built-in gunner a cheap upgrade (eg, 50mp). The AI doesn't seem to know how to use transports, thus preventing AIs to use plain US and Wehr AIs until they get an upgrade (with the exception that an Armor doctrine US AI will use HTs to cap points once it gets that ability). Alternatively, Relic could add the capability for infantry to use transports, but this would probably require more coding and be harder to implement.

I have ideas on how Relic could implement certain other abilities that the AI can't seem to accomplish also. For example, I can think of a way that the AI could build defenses (traps, wire, mines, bunkers, emplacements) effectively (as in facing the best possible direction and in the best locations), and even possibly recognize the importance of choke points, but it would require some significant changes to all official maps (plus mappers would have to fix their files too) and changes to the map editor, but this is probably too much work to expect from a patch (Relic could possibly do it with an expansion or the next COH, COH2). I don't have time to go into now, but I'll try to post later.

[Mackle]

Because of the way the AI is now, it is possible to get results like this :-


I'm not sure why there are Union Jacks displayed in the Axis stats; I played against 2 Panzer Elite and 1 Wehrmacht.

I would of saved a replay, but the option wasn't there (as you can see in the picture) - maybe it's because I had to save the game so that I could go to work, and then continued it later on?

Anyway, I am awful at player Vs player online. I've joined games which were "noobs only" according to the title, and had people ranked level 1 and 2 in them, only to get completely trounced quickly and easily. So it's fair to say I suck at this game. But despite the fact that I suck at the game, I am able to beat 3 CPU opponents at the highest setting the AI offers :-/ That doesn't say much for the competence of the AI if I'm able to do that, surely?

(I admit, it did take me well over 3 hours of gametime to do this though!).

[sweeten2213]

Taking the advice of 1SSPzGrenLehr, I threw a list that I complied back on page 9 in the original post. I going to take a look at all the pages past that list, and add them as well. As time permits, I'll add to page one any new or original complaints/suggestions.

Edit : most things added.

There aren't that many more things to mention - we have 18 pages, 259 responses, and almost 10,500 views. This thread is near the top if you sort by views and especially replies (it's in the middle of page 1 in replies). It's not a lack of information we lack, that's for sure. Maybe, even if people just reply, "It's bad - fix it," it would help. For those picking up on the thread at this point, check out page 1 for a master list. But, again, even if you have nothing to add, the numbers of views and responses may help. And now we have self proclaimed bad players beating 3 AI opponents - sheesh!!

[sweeten2213]

OK - I just checked the changelog for the automatch beta. It not only has changes for online and automatch, but it has even more balance fixes.

Again - no fixes for the AI.

Now, I understand that right now this quasi-patch is meant to create the automatch feature that the online community has been begging for. But still - it may be even more evidence that the offline SP community (especially outside of the campaign) is not prioritized. Maybe I'm wrong - maybe adding anything other than features that would affect online play would be silly for this particular patch/beta test. Maybe someone could help me out with that one.

As possibly more evidence - notice that there is no fix for the Brit AI simengine error. They have said they know what the problem is, but are still not fixing it yet. Funny that the simengine error more or less never affects online play. For an offline player, it pretty much makes one entire faction unplayable as an AI controlled ally or opponent. That, to me, is a major bug. One large enough that if it affected the online community to that degree, there would be rioting in the CoH streets.

I don't know - maybe I'm missing something. Maybe this won't be the changelog for the 2.3 patch. Maybe they're just listing the things that apply to the automatch beta, and when the official patch comes out it will include other things. It just doesn't seem right.

[cf_nz]

It's entirely possible that the change log/patch is limited to multiplayer issues simply because the team automatch is multiplayer only. No point in listing other changes that don't apply, also it's probably highly subject to change, as the OF beta was.

[sweeten2213]

Yeah, I thought about that as I was writing the post - especially as it's a beta type patch. And no offense to the online players - you are important to the game, and deserve to have your needs addressed. I hate to go off on mini-rants like that, especially when it's not warranted, but it's almost like I'm sitting back waiting for the next patch to leave out AI issues or the simengine error. I just haven't seen any indication that they will be addressed up to this point, and I have a sneaky suspicion that this, along with any MP issues discovered during the Beta, will be the next patch.

Relic - Prove me wrong

[elite]

When I order any kind of bombardment against AI (expert), they move out from there immedietly as I click. Before the red smoke hits the ground. This is cheating!

[sweeten2213]

I hate to sound negative, and I know that the patch changelog is far from final, but...

Buggo just put in the patch status thread that the 2.3 patch is in Beta testing, making it look at least at little more likely that the changelog in the automatch thread is going to be the basic jist if the 2.3 patch.

Buggo also posted this in the automatch thread

Hi everyone, there have been a few posts before but I just wanted to clarify some frequent questions I have been getting.

Yes, Team Automatch beta is now working. Feel free to download and play!

Yes, it is only available for English versions of the game. There are workarounds posted in this thread you can use for the other languages unofficially. Of course, when it is patched into the official game, it will be localized to all versions.

Also, there are a number of improvements you'll see in the beta that we are testing before bringing it over to the actual game. For example, better NAT negotiation etc (meaning you will probably be able to connect better with your friends).

We released a server side patch early this afternoon, that fixed some scripts (preventing Team Automatch from working properly). But yes, there will also be a client patch out later today that will include a number of changes and fixes to bugs, as well as some balance fixes.

Thanks!

Now, I know that it is also possible that Relic just wanted to concentrate on one thing at a time, as well.

Again, we'll see. Hopefully they will throw some non MP and balance issues in there - maybe where she mentions fixing bugs near the end. It doesn't sound like it to me, but we'll see.

Buggo - maybe in your infinate kindness you can shed some light on the issue

[Mackle]

Buggo - maybe in your infinate kindness you can shed some light on the issue

Infinate kindness? I've never met an infinately kind woman I admit though, I've not met Mother Teresa so maybe I'm being a harsh judge ;-)

[Soviet779]

MAJOR AI FAULT

Dont know if its been mentioned or not but this is a major fault with it, nothing to do with what it builds etc. The AI only kicks in about 50% of the time when a player drops, so basically half the time it does nothing and dosent build a thing. This is a major fault as i can actually direct an AI player pretty well as long as im not against pros. Funny thing is that when the AI HQ is destroyed, but then rebuilt by myself, it will start pumping out pioneers etc and building a full base.

Its a great feature that the AI takes over for dropped players, CoH is the first game ive played that does this (apart from maybe a few starcraft custom maps) so when this feature is 100% working it will be great.

Minor faults

1. It blocks bridges from time to time

2. There is no way (that i know of) to set it back it its default status after its been given a map command. For example when i give a map beacon such as defend point A, the ai will defend point A for the rest of the game, and only take direction from further map beacons. It would be nice if there was a beacon or something that returned it to its default state of attacking/expanding randomly.

[Relic]

OK, somehow we need to compile these all into some sort of "fixable" list to make this a bit easier to tackle. Any ideas?

[Mackle]

I thought someone had done this on the first page? Collated them all?

[sweeten2213]

Yes, Buggo, I did a list of all the things listed on the first page. If there is something specific you need me to do, PLEASE, let me know and I will do what I can.

Part of the problem is the vast array of problems. There are some photos in the thread for examples, but posting replays would be very difficult. To go through each point on the list and to go into appropriate detail for each point would take several pages itself. Within the thread, most of the points have been taken into more detail. But, if you need something more specific, and I am able to do it in any way that will get something changed, let me know.

Also, asking modders like Capt Combat and Warnstaff, and testers like Blackomne would help (maybe through a PM or e-mail). They are the experts within the community and can help with not only what is specifically wrong in the code, but what changes they have made.

Edit : I just went through every page and added some more details to the master list on page 1. Again, Buggo, let me know if there is something more specific I/we can do. If anyone can think of new things to add, or some specifics to add to the list, go ahead and post and I'll add it. If you just want to just chime in and add that you would also like the AI to be fixed for some or all of the above reasons, do so as well - it would be like adding your signature to a petition in way.

[Mackle]

And now we have self proclaimed bad players beating 3 AI opponents - sheesh!!

You missed a bit out... self proclaimed bad players beating 3 EXPERT AI opponents ;-)

I was very proud of myself. If I can do that to 3 expert AI opponents, imagine what these level 1 & 2 players who hand me my ass on a plate when I play online, could do to 3 expert AI's

[Pershing720]

I didn't see these specifically on the 1st page.

- AI being able to see you in the fog of war, for example... ATs sniping you, American/British artillery dropping on you out of nowhere, strafing runs, mortars. There REALLY needs to be a limit how far in the comp can see you.

- Hard and Expert Wehrmacht comps do not upgrade there veterancy. They should upgrade slowly.

- After a Scorched Earth PE's HQ is destroyed it can still drop sector arty on you.

- Game not ending after every British structure is destroyed. The British keep pumping out HQ trucks.

[sweeten2213]

@ mackle - sorry to use you as an example like that I apologize. The reality is, even an "expert" player should not be able to beat 3 "expert" AI opponents.

@ Pershing 720 - the brit HQ problem and the PE sector arty problem are more bugs than a poor AI issue, but I'll add them. The not upgrading to vet is in there :

11) Not building Komkraft center or getting vet for Wehr

and the FOW is there, though it is in the 2nd section, and I have to admit, I'm OK with it as the AI sucks bad even when being able to see into the FOW - the reality is increasing MP increase rate (as is done in Warnstaff's mod) and being able to see into the FOW may be the only way, even with fixes, the AI can compete. To be honest, the driving force behind the thread is to get a more challenging AI.

3) AI "cheats" like seeing into FOW or knowing your units/doctrine (although this was disputed by several posters as being necessary, and I have to agree with that)

[Pershing720]

11) Not building Komkraft center or getting vet for Wehr

I saw this one but you didn't mention...

- Hard and Expert Wehrmacht comps do not upgrade there veterancy. They should upgrade slowly.

I don't agree that the comps should see you so far into the FOW. I get tired of all my infantry being mowed down by strafing runs. Or AT guns sniping my tanks from extreme ranges. Or mortars bombing my infantry half way across the level.

[Relic]

Yes, Buggo, I did a list of all the things listed on the first page. If there is something specific you need me to do, PLEASE, let me know and I will do what I can.

Hmm, yeah I think the problem is we need more specifics. At the moment, the list is really hard for Dev Test to test. We need specific steps about how to get that problem, and then maybe some ideas on how to fix them.

Aside from this list though, we're constantly looking at ways to improve the AI and it's an ongoing project, outside of this forum thread.

[dreddnott]

I think part of the problem, Buggo, is that the AI is very inconsistent in its behaviour. Sometimes it's half-competent, other times it's completely pathetic. It's frustrating to not be able to get a consistent game out of it. I don't mean the same tactics or teching every time, I just want to be able to expect a certain level of competitiveness from it.

[sweeten2213]

Buggo,

This may sound too simple, but if your testers play a game or two against the AI in skrimish, they will see many of the things listed.

There's no way to get the AI to "do" some of the things listed, it just does them.

Spamming can be more obvious as you cut the AI opponent off from parts of the map or from resources, but it tends to spam something every game.

Many of the things are impossible to get the AI to do, since the fact is it doesn't do them. Such things like not building Brit emplacements (trenches, 17lbers, 25 lbers), not using off map abilities, not using unit abilities (like button), not building infantry, etc..., are impossible to prove they don't do other than to play and see that they don't do them.

Some things are tactical, and you would have to play to see what the AI should or souldn't do (like protect a certain point, attack a certain point, not attack a certain point, defend his base, etc...) and have some sort of feel for it.

Some things are done occasionally, but rarely (and often not well), like using built artillery. It is impossible to "make" the AI build them, but if you play a little you'll see they will but not fire them.

Some of the AI experts (which does not include me ) may be able to help with what is wrong with the code to cause the problems and what suggestions they have to fix them.

In all honesty, if any of the testers who are familier with the code play several games on several maps using and against several factions, they will see everything we're talking about. Like Dreddnott said, it can be inconsistant.

I really just don't how to tell you how to get it to do things in ways you would to find a bug - especially as most of the things are what it doesn't or rarely does. A "when I do this it always does this" doesn't really apply to getting an AI to act more competantly. It's not really bugs per-se - it's that the AI as an opponent sucks and seems clueless for all of the reasons mentioned.

Maybe someone else could help or explain things a little better, or the AI guys on the forums.

[Chuckwagon]

I would tend to agree that much of the brain dead behavior needs to be seen by just playing the game. But it may help if folks give some examples of types of games where they saw certain behavior.

For example, my friends and I play a lot of 3 vs 4 games, with the 3 of us going up against 4 Expert AI comps. At some point during the games we get enough control and enough force to attack one of the AI bases. Let's say an overwhelming arty barage. We blow the base to pieces, and most of that AI's forces are toast. How do the other AI forces respond? By sending just about every vehicle they have into the now blasted base, where they sit about and do nothing until we destroy them. And they will doggedly defend that empty useless base until we attack another one, and often even after that. That's just silly. It's a pile of wasted resources. If you play a few 4v4 games against 4 comps I'm sure you'll see something like this.

I've also seen similar behavior where resource points are concerned, as if the AI decides a particular point is the most important thing it's ever found and will defend it to the last man.

It's as if the AI decision tree is one way, and broken at that. Once it's decided a threat exists it'll be damned if it lets it go. Even if the threat isn't to anything of value. It also seems very reactive, only using key abilities when a player does something to annoy the AI into doing so. For example, we'll be going along nicely, no off map attacks from the AI, and one of us will build an arty piece, and suddenly 12 air strikes in less then a minute, blasting any arty built, then back to nothing for a while.

Another thing that may help show off some odd behaviors is playing maps that can be "choked" so the AI has nowhere to go. Once they have all of the points on their side, they just bunch up and mill about until a player does something to provoke them. Only minor effort to break the choke hold happens, and not very well thought out effort. The AI will repair bridges for example, but won't cut barbed wire to get to the bridge if it's been blocked off. Also, these choked off games are where the ability for the AI to "see" what it can't see becomes apparent, as it will happily attack well behind the front lines where it should not be able to.

I'm pretty sure any of these odd behaviors will show up in just a few games of 4v4. I'll try to keep track of any behaviors I see more than once in a while and what type of game it was in and post more if I can.

[cf_nz]

Is it possible to break the list down into things like:

- Bugs. Include steps to reproduce if possible.
- Suggestions for change, to improve gameplay/enjoyment (eg. AI not seeing into FOW).
- Suggestions for change, to imrpove challenge/difficulty.

It's going to be tough to do and somewhat subjective based on individual tastes but if if helps...

[Relic]

Is it possible to break the list down into things like:

- Bugs. Include steps to reproduce if possible.
- Suggestions for change, to improve gameplay/enjoyment (eg. AI not seeing into FOW).
- Suggestions for change, to imrpove challenge/difficulty.

It's going to be tough to do and somewhat subjective based on individual tastes but if if helps...

Now THIS would be very helpful!

[BlackOmne]

"They know not what they ask..."

Relic/Buggo,
Please, ignore requests for the AI Player (vice AI units) not seeing into the FOW. As you know and as they would know if they had any inkling how the AI achieves some of its sneakier behaviour, having the AI Player see into the FOW is necessary. They don't realize that removing this ability would make the AI look considerably dumber.

I have other, supportive ideas about the rest of the list but don't have time atm to post further.

[Panzer Jager]

*cough..*
AI should build Grenadiers.
This can be done very easily by setting the ai_info variables for the stormtrooper_kar98k to be higher, especially that for the aiclass_infantry.

[Chuckwagon]

"They know not what they ask..."

Relic/Buggo,
Please, ignore requests for the AI Player (vice AI units) not seeing into the FOW. As you know and as they would know if they had any inkling how the AI achieves some of its sneakier behaviour, having the AI Player see into the FOW is necessary. They don't realize that removing this ability would make the AI look considerably dumber.

I have other, supportive ideas about the rest of the list but don't have time atm to post further.

If you want to accept a brain dead AI I suppose it may "need" to see into the FOW. But properly done, the AI does not need to see into the FOW any more than an expert player would. Give me one example of a situation where the AI "needs" the ability to see into the FOW. I don't need a sneaky cheater AI, I want an AI that will challenge me the way a human player would, but without all the whining and carping that humans do. Hell, if you think having the ability to see into the FOW is good, because it makes the AI sneakier, why not just have random death strike some of the human players units? That'd be sneaky too, and give the AI an advantage.

The whole point of the FOW is to keep you in the dark about what your opponent is doing, unless you scout or recon, and allows you to hide your intentions. If the AI can see me all the time, then it can always counter my moves perfectly, if it isn't brain dead. And that's what we're trying to fix. If we make the AI not brain dead, then seeing into the FOW is a bad thing.

[Cpt Combat]

Buggo:

As far as specifics goes, I have fixed several issues related to various aspects of the AI. Such as:

1) All units will fire artillery and smoke barrages
2) No spamming of units (this includes off map units)
3) Building of emplacements
4) Use of all abilities (expect hull down, still working on that one)
5) correctly calling in gliders for british
6) Correct priority in captureing points and VPs
7) reinforcement of units
8) Improved base defense
9) Correct Doctrine selection
10) Building of Kampfkraft center and upgrading veterancy
11) calling in reinforcement engineers when HQ destroyed
12) Deploying Panther battle group for PE
13) AC221 securing points
14) Correct usage of cover

And a host of other modifications.

However, code for deployment of british HQs need to be looked at. I've modified it, but its still not very good

Like to see adding a stub of code that would allow the AI to capture team weapons


Here's the link to my mod so you can take a look at it.

http://forums.relicnews.com/showthread.php?t=173948

[Cherni_95]

Wehr AI to stop pumping out so many Pumas and concentrate on the heavier tanks.

[sweeten2213]

Thanks for your help, guys. I think the best thing said so far (other than agreeing with BlackOmne about the FOW) was by Capt Combat. He not only knows of the problems, but what causes them AND HOW TO FIX THEM. I know I can't add that sort of knowledge. Like I said earlier, Buggo, it's people like Capt Combat, Warnstaff, and BlackOmne, who can truly help. Maybe your guys and gals could DL their mods and see what they've done. I am merely a messenger and a gamer who sees things and wants a challenge, but I have no skills as far as coding (whether the codes that are required to make an AI, what codes are in the game, and what changes can be made). The modders have solved many of the problems listed - so at least we know it can be done. I also requested Warnstaff's help in his thread (hopefully he doesn't get upset with that ).

Since I got myself into this, I will TRY and take each point and break it down as best I can when I have more time. In the mean time, hopefully we'll get some wisdom from others who actually know what they are doing more than me - which would probably be just about everyone

[BlackOmne]

If you want to accept a brain dead AI I suppose it may "need" to see into the FOW. But properly done, the AI does not need to see into the FOW any more than an expert player would. Give me one example of a situation where the AI "needs" the ability to see into the FOW. I don't need a sneaky cheater AI, I want an AI that will challenge me the way a human player would, but without all the whining and carping that humans do. Hell, if you think having the ability to see into the FOW is good, because it makes the AI sneakier, why not just have random death strike some of the human players units? That'd be sneaky too, and give the AI an advantage.

The whole point of the FOW is to keep you in the dark about what your opponent is doing, unless you scout or recon, and allows you to hide your intentions. If the AI can see me all the time, then it can always counter my moves perfectly, if it isn't brain dead. And that's what we're trying to fix. If we make the AI not brain dead, then seeing into the FOW is a bad thing.

Examples of why the AI needs to see into the FOW:
1) Have its units skirt the field of vision of your units to penetrate the frontline as away to disrupt your resources by capping points in your rear. How often have you seen engineers/pioneers suddenly appear in your rear and do this? Probably, lots unless you have solid wall of wire/traps that the AI units can't sneak thru. This allows the AI to simulate similar behaviour from a human. The AI Player has to know what your units can see if its gonna use the waypoint routine to avoid your units seeing its units.
2) Ever wonder how the AI occasionally has its units unexpectantly approach your MGs and ATGs from outside their deployed fields of fire? By knowing where your units are and how they're facing and sending its units skirting around your field of vision. This is the closest the AI will likely ever come to accomplishing a flanking manuever.
3) By knowing where your units are heading and where the majority of your forces are, the AI can send a large group of units at and attack the weaker parts of your line. This simulates how a human player might anticipate where you are weak and exploiting it.

You want to remove this ability because:
a) You don't like the AI Player targetting your units with doctrine abilities like strafing run and off-map artillery. But the AI uses these abilities so seldom that it rarely affects the outcome.
OR
b) You think wrongly that the AI has its units fire at units beyond the field of their vision (eg, ATGs) since the AI Player can see into the FOW. This doesn't happen; AI units can only fire on what they can see or what friendly units can see in a forward observer fashion. I can gurantee that if you review any replay from the AI's perspective (with FOW on) that you'll see the unit that revealed your targetted unit so that it could be fired on.
OR
c) You think you can't setup up a proper ambush. I setup ambushes/kill-zones all the time when I setup a wall of wire/traps across the map leaving only a couple holes covered by "ambushing" units. I kill AI units in droves like this because the AI Player funnels his units through the only paths I allow it to reach objectives on my side of the map. Ambushes can be achieved - I would rather see Relic give the AI units the ability to cut unguarded wire and blowing traps and fixing bridges by knowing that you don't have any units guarding those areas thru seeing into the FOW.

I want an AI that will challenge me the way a human player would,

Your desires for a human-like AI, although I agree it would be great to see, is far removed from the reality of what they can accomplish in real-time on a 3D map with today's technology. Today's AIs can't learn (eg, learn from the mistake of sending a unit where you have ambushed it) and AI will likely never become able to learn in our lifetimes. Don't expect them to, especially in CoH's lifetime. Higher thinking like probing and coordinated flanking attacks with several combined arms units and feints and baiting would likely bogged down even the fastest PCs to a snails pace, even if they could find some way of accomplishing it. If you think these capabilities are possible, please tell us how exactly you would instruct the AI to accomplish it.

Certain minimal cheating by the AI is necessary if it's gonna have a chance at offering a challenge to a truly clever human player. Being able to multi-task (ordering all of its units over the entire map at the same time) faster than a human and build more efficiently (the instant sufficient resources exist) is not enough of an advantage to counter the human's ability to learn and use complex superior tactics. Allowing this FOW ability and giving the AI resource bonuses is a step in the direction of balancing this unequal match of intellilect. It helps give the AI the illusion of intelligence. I hate cheating too (especially by humans) but we have to allow a minimal amount by the AI if it's gonna have any chance of beating us.

So, for all intents and purposes, the AI is brain dead and will remain so and we can only expect Relic to improve slightly on what they have (along the lines of Warnstaff's mod) and therefore improve the illusion of "intelligence".

I should have spent this time enhancing Sweeten's list....but again I run out of time.

[Pershing720]

You want to remove this ability because:
a) You don't like the AI Player targetting your units with doctrine abilities like strafing run and off-map artillery. But the AI uses these abilities so seldom that it rarely affects the outcome.
OR
b) You think wrongly that the AI has its units fire at units beyond the field of their vision (eg, ATGs) since the AI Player can see into the FOW. This doesn't happen; AI units can only fire on what they can see or what friendly units can see in a forward observer fashion. I can gurantee that if you review any replay from the AI's perspective (with FOW on) that you'll see the unit that revealed your targetted unit so that it could be fired on.

I see the British/Americans using arty/strafing runs all the time. If Relic could tone down the arty/strafe surprise to where it doesn't happen as often behind your lines that would be better.

The comps don't have the restrictions of FOW as we do. So yes AT guns CAN snipe your tanks within your FOW. I've played hundreds of games against the AI and know this for a fact.

[Chuckwagon]

Examples of why the AI needs to see into the FOW:
1) Have its units skirt the field of vision of your units to penetrate the frontline as away to disrupt your resources by capping points in your rear. How often have you seen engineers/pioneers suddenly appear in your rear and do this? Probably, lots unless you have solid wall of wire/traps that the AI units can't sneak thru. This allows the AI to simulate similar behaviour from a human. The AI Player has to know what your units can see if its gonna use the waypoint routine to avoid your units seeing its units.
2) Ever wonder how the AI occasionally has its units unexpectantly approach your MGs and ATGs from outside their deployed fields of fire? By knowing where your units are and how they're facing and sending its units skirting around your field of vision. This is the closest the AI will likely ever come to accomplishing a flanking manuever.
3) By knowing where your units are heading and where the majority of your forces are, the AI can send a large group of units at and attack the weaker parts of your line. This simulates how a human player might anticipate where you are weak and exploiting it.

None of these are needs. Human players accomplish the same things without special knowledge. How? Trial and error perhaps. The AI can do it the same way. Scouting and recon AI routines are not that tough. There are units for those tasks. There is no NEED for the AI to cheat around doing things the same way humans do. It just makes it easier to program the AI if it cheats.

b) You think wrongly that the AI has its units fire at units beyond the field of their vision (eg, ATGs) since the AI Player can see into the FOW. This doesn't happen; AI units can only fire on what they can see or what friendly units can see in a forward observer fashion. I can gurantee that if you review any replay from the AI's perspective (with FOW on) that you'll see the unit that revealed your targetted unit so that it could be fired on.

This one I'm pretty sure about. On a map divided by a river, with the only bridge across destroyed so no units from either side can pass the river until it is repaired, the AI has routinely attacked units at the back of the map, well outside any of its units range to detect. This has occured repeatedly, and there are witnesses. The AI even attacked moving targets in the FOW, nice trick that. So something fishy is about for sure.

Making the AI better isn't simple, I know that. But it isn't as difficult as you make out. The AI doesn't have to learn, it does need to have options and a complex decision tree. And it needs well thought out feedback loops. It's not about learning, it's about setting conditions and checking if those conditions are met, and resetting conditions based on feedback. The system COH uses may not ba able to ever accomplish a more realistic AI. If not, I doubt that Relic has the commitment to re-architecting it so it can, cost issues and such, manpower, whatever, and I wouldn't blame them. But that doesn't mean I can't ask. Older games played on lesser hardware had smarter AI systems, so it isn't unreasonable. And the current system can certainly be improved, as the modders have shown.

In any case, I'll keep trying to point out things I think are particularly odd, and hope changes get made.

[cf_nz]

This is my take on sweetens master list, along with my own thoughts. I’ve only commented on the ones I know about or have an opinion on.

I play at expert level and primarily 1v1, I would like to play team games but the AI doesn’t make a good team-mate. I would like a competent and challenging AI at the hardest levels and a fun to play against AI at lower levels.

Bugs.

• Wehrmacht and Americans not/rarely building infantry (Volks, Grenadiers, and Riflemen squads):
Something changed between CoH and OF; the build orders seem to be the same yet it is quite rare to see these units. I do not know why or how to duplicate it outside of just playing a game.

• Wehrmacht expert AI does not build Kampfkraft centre:
Again, it’s something that always happens, I’ve seen other people post that easy and normal do build and use the Kampfkraft centre/axis vet but hard and expert don’t.

• Simengine.dll crash:
Known bug, fix pending. (http://forums.relicnews.com/showpost...3&postcount=15)

Suggestions for improvements – gameplay.

• Unit spam:
I’m ok with this provided the units that are spammed are useful, i.e. If go heavy rifle spam it’s acceptable for the AI to spam Pumas. If I have tanks then it’s not.

• AI "cheats" like seeing into FOW or knowing your units/doctrine:
I’m ok with this as well; it’s a crude parallel to a humans ability to listen into the FOW.

• Not re-building/repairing base buildings after damaged/destroyed. Also, not making units from HQ when repaired:
Not sure about this, I’ve seen the AI repair and rebuild base buildings regularly.

• Not repairing bridges:
I’ve seen the AI repair bridges but don’t play on maps with bridges often enough to comment properly.

• Not using Bergetiger:
I don’t think the Bergetiger is currently enabled for use by the AI.

• Using too much fake arty:
Is there such a thing?

Suggestions for improvements – challenge/difficulty (making the AI play better).

• Observation posts:
Try to mimic human, i.e. minimise the number of OPs build; do not build on +5 resource points early in a game. Manpower is better spent on combat units unless at pop cap or late in the game and floating lots of manpower.

• Brit trucks - poor choice of deployment location:
Try to mimic human. Again favour higher resource points. On Angoville for example a skilled Brit player will cap the nearest +16 munitions or +16 fuel and move their HQ truck there; the AI will deploy on the +5 fuel next to the base. Also the AI should avoid clumping truck together on one sector.

• More difficulty levels:
I don’t think 4 difficulty levels are sufficient to cater to the wide variety of skills and playstyles CoH players have.

• Build orders:
More specialised for both map (size, number of players, urban/rural environment) and difficulty level ('general' for lower levels, 'optimised' for higher levels).

• Poor usage of emplacements:
Whether it is not building in places where it should, building them where they shouldn't, or not building them at all (includes Brits: slit trenches, 17lb, 25 lb, and mortars; Americans: MG nests; Wehrmacht: bunkers).

• AI absent in early game:
Not sure about this, on some difficulty levels the AI is designed to delay the attack.

• AI mortar usage - whether it be charging to death or just not used correctly:
The AI appears to prioritise targets well if you are within range of its base otherwise the next best target is the players base, it will charge mortars recklessly into range and bombard base buildings. If a mortar is lost a new one is built.

• Some units, like stukas, are obsessed with attacking your base rather than key troop or defence positions on the rare occasion they do fire:
Similar to above. Player base buildings must have too high of a priority as a target.

• Not knowing how to break a defense. Will ram its head into it time and time again without any chance of success (artillery, non or off map would help):
Tied to the above poor usage of indirect fire/artillery pieces.

• Retreat weapons teams sooner:
This applies primarily to American and Wehrmacht mortar and HMG teams. If under direct close quarters fire they short retreat much sooner then they currently do.

• Units using cover, especially garrison cover more:
Units under attack or upon detection of a nearby threat should move into cover as a first option unless superior to the unit they are fighting. Applies to both infantry and HMGs.

• Idle units in base:
The ones I see most regularly are American and Wehrmacht halftracks. Without an upgrade (Quad .50, flammerwerfer, stuka) the AI does not use the, kind of understandable as they are non-combat units. Could the AI build them only if sufficient munitions are available for the upgrade; learn to use them as a transport or for field reinforcement; garrison an engineer or pioneer unit inside to man the MGs and use as a combat unit?

• AI value system:
AI needs a better ‘counter’ system to allow it to build the best unit or upgrade to counter a situation in game, e.g. mortars aren’t always needed and should perhaps only be built to counter garrisoned units, emplacements or static weapons (MG, AT gun).

• Build high tier units sooner:
I frequently see the AI with A Panzer Command or Tank Depot yet it continues to build lower tier units, e.g. building Pumas when saving for a Panzer IV or Ostwind would be a better choice.

• AI not using abilities, for both command tree and other abilities:
Will try to follow this up with examples.

Other AI related issues, not related to skirmish opponents.

• General vehicle pathfinding.

• Units stuck on barbed wire.

• Tanks getting in closer than necessary when told to attack.

• Units running out of cover when they shouldn't or not "listening" to orders.


I’ll edit this when I get more time/think of more things to add.

[BlackOmne]

Scouting and recon AI routines are not that tough.

The AI doesn't have to learn, it does need to have options and a complex decision tree. And it needs well thought out feedback loops. It's not about learning, it's about setting conditions and checking if those conditions are met, and resetting conditions based on feedback.

Please, provide detailed logic on how you think this will work if you think it's that easy. It doesn't need to be in programming language. Then, apply that to dozens of units simultaneously, all moving and reacting in real time without bogging down the fastest PC processors. You make it sound so simple. So, you should be able to lay down the logic if its that simple. I believe you overestimate the magic/intelligence of any AI programmer in the world under these conditions.

This one I'm pretty sure about. On a map divided by a river, with the only bridge across destroyed so no units from either side can pass the river until it is repaired, the AI has routinely attacked units at the back of the map, well outside any of its units range to detect. This has occured repeatedly, and there are witnesses. The AI even attacked moving targets in the FOW, nice trick that. So something fishy is about for sure.

I dare you to upload a replay proving this. I don't think you can.

Older games played on lesser hardware had smarter AI systems, so it isn't unreasonable.

Here I agree with you, but I think you're forgetting or glossing over a couple factors in this statement. This better AI existed in turn-based 2D games using some type of grid system imposed on the 2D map, akin to a chess board, greatly limiting the number of locations for units in the game. Turn based games allow the AI time to think - it might take the AI upwards of 30 seconds or more (a long time for a computer) to decide on its moves & attacks after you finished your turn, time a RTS doesn't have. And a 2D map with a grid system is extremely more simple for a programmer to code conditions for than 3D map with no grid system. If you can provide an example of a 3D RTS with this kind of AI that you describe, please do.

I see the British/Americans using arty/strafing runs all the time. If Relic could tone down the arty/strafe surprise to where it doesn't happen as often behind your lines that would be better.

The comps don't have the restrictions of FOW as we do. So yes AT guns CAN snipe your tanks within your FOW. I've played hundreds of games against the AI and know this for a fact.

Pershing,
Yes, the AI Player can use doctrine abilities/attacks like strafing run and off-map artillery to attack your units that are not visible to any of its units, but I rarely see this occur and I play the AI a lot.

As for on-board units attacking units that can't be seen by any AI units, that doesn't happen and like with Chuckwagon, I dare you to upload a replay proving it. It should be easy to do if you say it happens so often.

[Reverie]

Has anyone tried the beta skirmish AI yet? Just played a game versus a defensive AI and it made StuGs, Panthers and Stukas fairly fast and pushed pretty well. It even made good use of a mortar and registered arty. Is it just me or does it seem improved?

[cf_nz]

I played one game tonight as US, a 1v1 vs the expert Axis AI on Angoville, didn't notice any difference.

[Chuckwagon]

Yes, the AI Player can use doctrine abilities/attacks like strafing run and off-map artillery to attack your units that are not visible to any of its units, but I rarely see this occur and I play the AI a lot.

Heh heh, minor misunderstanding; in your earlier post where you said units, I assumed (my bad) that meant the AI player, as in the AI players are the Computer's units. It seems you actually meant the individual units, like tanks and such. That being the case, I agree, I've never seen any UNIT fire beyond it's FOV into the FOW. However, that wasn't what I was referring to. I was referring to the way the AI attacks things in the FOW in a manner that I cannot do in return. I assumed such attacks were off map abilities, not units. Sorry for the misunderstanding on that issue.

As for the AI, we'll just have to agree to disagree. It seems you think the AI is some DARPA project and will consume enormous CPU resources and be able to out think a human and have autonomous thought. I don't. I assume the AI is complex and time consuming, and that the devs will be creative enough to get it to work better. But in the end it's more like a very complex set of scripts, questions, answers, and pre-set actions and variables. No real intelligence is needed, because a human has already done the complex thinking and the AI is only responding as it's been told to respond based on what the human decided. I don't expect real AI in any game, it's not cost effective I'm sure. But more complex scripting, adaptive scripting, etc. wouldn't be out of the realm of the possible.

In any case it isn't worth belaboring the point. The devs will do what they can. That's all I expect. And I'll whine about things I think are silly. And if they can fix them great. If not, oh well.



Edit: I also meant to point out that this is not really a 3d game. It's a 2d game with 3d graphics. No units in the game have the ability to alter their altitude, so for every X,Y coordinate in the game there is only ever one Z coord in relation, and that is defined by the map. Even though the Z axis is variable, no unit will ever be anywhere other than the map defined Z coord for any given X,Y. Even if the map is deformable, there is only ever one Z for any X,Y. Even the flying units only have one Z until shot down, and then only a landing location is needed, the path taken to that location isn't important. So it isn't really 3d in the same way Homeworld was. So while it may have a couple more considerations for map positition than Starcraft would have, it's still essentially 2d. There really isn't that much more to consider for unit management.

[Pershing720]

Took a bunch of pics to prove my point that the comps CAN AT snipe you inside the FOW without any spotters. But not sure how to post them.

[Chuckwagon]

When I hit reply I get an additional options pane at the bottom that looks like what I have attached. Do you not get this option?



Edit - Since you said you do get the option I deleted my attachment. No need for it.

[Pershing720]

Yes I get the same thing and I hit the "Insert Image" and it says.

Script Prompt:

Enter the text to be formatted
[IMG]xxx[/IMG]

Not sure what to do with that.

[Chuckwagon]

Is your pop up blocker stopping it from opening the attachment management window? I get a whole new window to manage from when I hit the button.



Also, what browser are you using, mine is Explorer.

[retroholyfire]

I've read the master list but couldn't find this one so here goes. I find that when I'm as Brits with a US ally vs someone like wehr, my US ally is ALWAYS miles behind everyone else even if our team is winning by a good marjin. It has been same difficulty for ally comps yet whilst the Wehr did decide to go and build a Panther and other stuff, the US ally just spammed jeeps. It's top unit was the M8. I havn't played much against US comp though but it seems to me that the ally comp is always worse than the enemy ones.

[Pershing720]

This is what I'm talking about. AT sniping inside the FOW without any spotter. It becomes very frustrating when the comps spam ATs and snipe your tanks and vehicles.