AI Discussion (dev branch)

[haleks 8212]

Can we please get an official response indicating BIS plans regarding the AI?

Dreamer!:rolleyes:

[old_painless 182]

I have come to realise that as long as DLCs can be released and function with the current AI, no incitament for fixing the AI exists. The only situation where poor AI would be detrimental to BI themselves is if they affected official missions and the campaign. And the campaign does not rely on the AI being able to drive, use indoors environments, display self-preservation on the battlefield etc. All the situations where the AI is not up to par seem to be conveniently avoided in official content.

Not that the campaign is bad, on the contrary, it is brilliant and very entertaining. But the situations where the AI must do stuff seem to be carefully chosen and/or scripted (this is from playing through the first two episodes of the campaign)

Edited October 2, 2014 by Old_Painless

[stang725 10]

I have been trying to elicit a BI response to what is and isn't possible in terms of an improvement in terms of overall AI scaling. Not much luck... Someone, somewhere at BI knows exactly what is and isn't possible for improving AI... they just aren't being given a pass to engage the community on the level required to really explain things. Once you acknowledge there is a "real" problem, there is no longer plausible deniability that the statement that AI works "good enough" is kind of BS.

http://forums.bistudio.com/showthread.php?183822-Would-you-donate-to-a-bounty-to-expedite-fixing-some-major-in-game-issues-via-a-Patch

[vegeta897 13]

Someone, somewhere at BI knows exactly what is and isn't possible for improving AI...

Why would you think it is so black and white? Anything is possible since they can change the engine however they want. The feasibility is the question, and this is a gradient that isn't so easy to commit to a position on (which is what you're asking for). They don't want to say no to possibilities, and saying yes when they're not absolutely sure is a huge risk.

[Bucic 1]

I have come to realise that as long as DLCs can be released and function with the current AI, no incitament for fixing the AI exists. The only situation where poor AI would be detrimental to BI themselves is if they affected official missions and the campaign. And the campaign does not rely on the AI being able to drive, use indoors environments, display self-preservation on the battlefield etc. All the situations where the AI is not up to par seem to be conveniently avoided in official content.

Not that the campaign is bad, on the contrary, it is brilliant and very entertaining. But the situations where the AI must do stuff seem to be carefully chosen and/or scripted (this is from playing through the first two episodes of the campaign)

You've hit the nail on the head!

To be honest I'd rather see this thread (along with few others, on damage system, sound etc.) closed right now. It would at least address one problem that is solvable right now. Hope/delusion and fanboyism.

[chortles 263]

Not that the campaign is bad, on the contrary, it is brilliant and very entertaining. But the situations where the AI must do stuff seem to be carefully chosen and/or scripted (this is from playing through the first two episodes of the campaign)

And here I was thinking that 'carefully chosen and/or scripted around the AI's limitations' was a good thing...

[Alwarren 2767]

And here I was thinking that 'carefully chosen and/or scripted around the AI's limitations' was a good thing...

It would be preferable if the AI was smart enough to act on their own, so no it isn't a good thing in this case.

[old_painless 182]

It would be preferable if the AI was smart enough to act on their own, so no it isn't a good thing in this case.

Yeah, that was my point, thanks :) Anyway, I think we need to get past the era of new DLCs+content (FFV, Marksman, Helos, Karts (sic), sling loading) before someone at BI starts looking seriously at the AI again.

[oukej 2910]

You see, AI doing everything on their own or being scripted for a specific scenario situation. It's one of the major design problems and you are always going to end up with compromises.

Let's have an example - autonomously rearming AI. Players do that. If you see a better weapon or an AT, you grab it. But what if you want to have a military-proper scenario? AI looting stuff around - not very immersive. Not talking about situations where you need the AI to run away from a suppressed location. But it sees a nice launcher, goes for it, eats mortar. For every added functionality, you have to prioritize the behavior and put relations and exceptions to functions that are already in place. You can easily end up with a unmaintainable system.

There's always great amount of situations, where some behavior is not desired. And it's nearly impossible to account for all those situations. So you can either have a modular system which in return places higher workload on the scenario designer and still requires a crazy amount of maintenance. Or you can just put in a set of features you think are the most beneficial to the gameplay and fit most of the situations. But once you have that and want or need to change it, you're entering a game of Jenga Tower :)

And I think this is also somewhat related to the issues mentioned beforehand and why it may seem that some of the development is taking too long.

Bottom line - AI universality and backwards compatibility

From the AI point of view there are "direct" fixes continuously being done. DLCs mainly add some sexy functionality. And then we can steal some more maneuverability (overhaulin') space within the bigger package of an expansion.

[killzone_kid 1330]

You see, AI doing everything on their own or being scripted for a specific scenario situation. It's one of the major design problems and you are always going to end up with compromises.

Let's have an example - autonomously rearming AI. Players do that. If you see a better weapon or an AT, you grab it. But what if you want to have a military-proper scenario? AI looting stuff around - not very immersive. Not talking about situations where you need the AI to run away from a suppressed location. But it sees a nice launcher, goes for it, eats mortar. For every added functionality, you have to prioritize the behavior and put relations and exceptions to functions that are already in place. You can easily end up with a unmaintainable system.

There's always great amount of situations, where some behavior is not desired. And it's nearly impossible to account for all those situations. So you can either have a modular system which in return places higher workload on the scenario designer and still requires a crazy amount of maintenance. Or you can just put in a set of features you think are the most beneficial to the gameplay and fit most of the situations. But once you have that and want or need to change it, you're entering a game of Jenga Tower :)

And I think this is also somewhat related to the issues mentioned beforehand and why it may seem that some of the development is taking too long.

Bottom line - AI universality and backwards compatibility

From the AI point of view there are "direct" fixes continuously being done. DLCs mainly add some sexy functionality. And then we can steal some more maneuverability (overhaulin') space within the bigger package of an expansion.

AIs, unlike player, can have all the information about battlefield, positions and armament. So if there is an AT on the ground and enemy has a tank, why not have it? Or even better, enemy has a tank...is there AT on the ground?

[Tonci87 163]

You see, AI doing everything on their own or being scripted for a specific scenario situation. It's one of the major design problems and you are always going to end up with compromises.

Let's have an example - autonomously rearming AI. Players do that. If you see a better weapon or an AT, you grab it. But what if you want to have a military-proper scenario? AI looting stuff around - not very immersive. Not talking about situations where you need the AI to run away from a suppressed location. But it sees a nice launcher, goes for it, eats mortar. For every added functionality, you have to prioritize the behavior and put relations and exceptions to functions that are already in place. You can easily end up with a unmaintainable system.

There's always great amount of situations, where some behavior is not desired. And it's nearly impossible to account for all those situations. So you can either have a modular system which in return places higher workload on the scenario designer and still requires a crazy amount of maintenance. Or you can just put in a set of features you think are the most beneficial to the gameplay and fit most of the situations. But once you have that and want or need to change it, you're entering a game of Jenga Tower :)

And I think this is also somewhat related to the issues mentioned beforehand and why it may seem that some of the development is taking too long.

Bottom line - AI universality and backwards compatibility

From the AI point of view there are "direct" fixes continuously being done. DLCs mainly add some sexy functionality. And then we can steal some more maneuverability (overhaulin') space within the bigger package of an expansion.

I completely understand. I just hope that one day you will have the AI tank drivers keep their front towards the enemy instead of showing their rear all the time....

[Bouben 3]

You see, AI doing everything on their own or being scripted for a specific scenario situation. It's one of the major design problems and you are always going to end up with compromises.

Let's have an example - autonomously rearming AI. Players do that. If you see a better weapon or an AT, you grab it. But what if you want to have a military-proper scenario? AI looting stuff around - not very immersive. Not talking about situations where you need the AI to run away from a suppressed location. But it sees a nice launcher, goes for it, eats mortar. For every added functionality, you have to prioritize the behavior and put relations and exceptions to functions that are already in place. You can easily end up with a unmaintainable system.

There's always great amount of situations, where some behavior is not desired. And it's nearly impossible to account for all those situations. So you can either have a modular system which in return places higher workload on the scenario designer and still requires a crazy amount of maintenance. Or you can just put in a set of features you think are the most beneficial to the gameplay and fit most of the situations. But once you have that and want or need to change it, you're entering a game of Jenga Tower :)

And I think this is also somewhat related to the issues mentioned beforehand and why it may seem that some of the development is taking too long.

Bottom line - AI universality and backwards compatibility

From the AI point of view there are "direct" fixes continuously being done. DLCs mainly add some sexy functionality. And then we can steal some more maneuverability (overhaulin') space within the bigger package of an expansion.

The example with AI rearming colliding with running for their life would be easy to fix if a player has a simple command "run for your lifes and don't look back". Such a command would simply disabled everything unwanted at the moment and AI commanding would be super-predictable - that means good. New commands could solve many many issues with AI.

In cases where AI is needed to work without player input come scripts from mission makers. Again, just disable what you don't want to happen. It is much better to have something to disable than no way how to enable something.

Give commanding players more control over the AI.

Also, I am sure that a lot of current AI priority checks could be improved immensely and lot of AI features could be fine-tuned. Look at the bCombat, which is unfortunately limited by the current AI code. And I am not that kind of an uncritical player that would praise any AI mod out there. Actually, bCombat is the first AI mod that really improves the behaviour of units and keeps everything else compatible (apart from AI fleeing from armored enemy, there is some problem with that). Every other mod, in my opinion, does/did it wrong with too many vanilla compatibility problems, with too many compromises in other AI areas.

Edited October 4, 2014 by Bouben

[Alwarren 2767]

Give commanding players more control over the AI.

This. If I tell them to regroup or move to a specific spot. they should either automatically do it pronto, or at least I should have a method of telling them that it is important. There is one mission in Adapt where I tried unsuccessfully to get my guys out of mortar fire but since they cannot keep up, they just get a shell to the face every time. The player should have more control over them. In the end, it doesn't feel right to lose all of your guys just because they don't seem to have a will to live.

[Bouben 3]

This. If I tell them to regroup or move to a specific spot. they should either automatically do it pronto, or at least I should have a method of telling them that it is important. There is one mission in Adapt where I tried unsuccessfully to get my guys out of mortar fire but since they cannot keep up, they just get a shell to the face every time. The player should have more control over them. In the end, it doesn't feel right to lose all of your guys just because they don't seem to have a will to live.

Yes. I mean, this is so absurd that I cannot stress out how much I am surprised that it (unresponsive AI) has not been addressed yet.

For movement orders alone, I want different options. Move carefully with bounding overwatch, move fast but react to enemy fire, move fast and ignore everything around you, move in crouch but immediatelly go prone when you stop, move in standing postition and then immediatelly go prone when stop etc. The AI is unable to decide this for themselves at the moment. We need more options with orders.

Also, we need a "never, under whatever circumstances fire" command and many others...

Rearming is also major pain with AI as player has no easy way of telling them how to do it properly.

And there are also those clearly left-alone features like "stealth" mode which is completely, utterly useless. Or the never explained difference between the stop command and the wait command. There is so many issues with AI and commanding. For years.

Edited October 4, 2014 by Bouben

[oukej 2910]

The disobedience of AI subordinates is no less important issue to us.

This has already been discussed back in the thread. And it's still a valid point, no doubt.

Look at the bCombat, which is unfortunately limited by the current AI code.

And also made possible by it ;)

Anything that is not actually present in the code should not be taken as if it should have been there.

Edited October 4, 2014 by oukej

[Variable 322]

Thanks for responding Oukej, it is encouraging!

The disobedience of AI subordinates is no less important issue to us.

This has already been discussed back in the thread. And it's still a valid point, no doubt.

I am disappointed to see that the discussion has yet again derailed for additional functionality (AI auto rearm) instead of concentrating on existing AI behavior that is destroying the game. I think most of us will agree (and correct me if I'm wrong guys) that the most pressing issue with the AI is it's unresponsiveness and slowness, especially under fire.

You have quoted yourself there but this message of yours is 6 months old. How hard it is to fix that? Without getting into detailed design I suppose it's not that hard to add an option/player order to bypass that horrible AI ritual of slowly bounding and covering when it automatically goes into this horrendous state. It's just needs to be finally prioritized!

Changing this alone will improve the SP experience significantly, so let me ask this one more time, do you guys have any plans to fix that? Or is it something that you have no plans in changing in the near future?

[Smurf 12]

Since the discussion reached this point, little OT but if you are going to discuss "Human-to-AI orders" you HAVE to take in consideration the UI when doing so;

Current design is either too simple or too complex\specific, a smarter aproach can bring more benefits than it seems: better game flow, less player frustration = happier players.

AI grouping and simple set of behaviours like:

Open Field ----------- Urban ------------ Vehicles

Safe ------ Aware ------ Combat ----- Stealth (setting also ROE, stances and more)

Stay with Me (aka DO WHAT I'M TELLING OU TO NO MATTER WHAT FOR ONCE IN A LIFETIME!!!!!!) ------ Stay in Formation -------- Free Will

While keeping the in depth micromanegement we have now as an option.

I know you Devs know that so, I won't go deeper; Just the same old annoying reminder.

Now you have the credit, the CZKredit, find some time. :)

[froggyluv 2136]

It would be nice to hear what the DEV's would like to see for the future of this series AI's even if it may not be possible. Are you guys happy with the status quo or have bigger plans? Dev responses in this area have always been cryptic for as long as I've been here so it'd be nice to just have a frank discussion for once as this topic is arguably one of the most important and exciting to many of us. We hear rumors like "the AI code is built on binarized pre-Aramaic language unreadable by all but the equally ancient Suma Lord..."

That "streamlining old code is the agenda and any new behaviour could potentially break everything". The last new cool feature I remember was the AI ability to detect fences and crawl under them -can we not continue in this direction with other objects *cough* windows and doors?

Also +1 to an overide AT All Cost order to simply force an action or a full balls to the walls sprint.

Rearming is still a pain when you have a list of 12 rifles with no idea which one is where. I've ordered my guy right next to the corpse of an AT soldier but when ordered 'Inventory' he kept checking some damn jeep :p

[Variable 322]

Rearming is still a pain when you have a list of 12 rifles with no idea which one is where. I've ordered my guy right next to the corpse of an AT soldier but when ordered 'Inventory' he kept checking some damn jeep :p

Yeah, but before venturing into the realms of the highly complex "auto AI rearming" this can be easily improved by re-ordering of the options in the AI action menu and adding a range indication next to each ammo/rearming option.

The most pressing issue remains the AI unresponsiveness.

[Bouben 3]

This has already been discussed back in the thread. And it's still a valid point, no doubt.

And also made possible by it ;)

Anything that is not actually present in the code should not be taken as if it should have been there.

Yes, definitely possible because of the code. I have not meant to be ungrateful for the AI that we have now (and for the devs). I just feel there is so much potential that have not been exploited yet because it just isn't a priority for selling the game. I am honest.

---------- Post added at 17:43 ---------- Previous post was at 17:38 ----------

And of course I agree with Variable. There is a huge need to fine-tune and fix current issues with the AI before adding new features.

[Variable 322]

YI just feel there is so much potential that have not been exploited yet because it just isn't a priority for selling the game. I am honest.

Exactly. My heart breaks with each mission maker that have to scrap his awesome (and yet not so complicated) ideas just because the AI is unable to respond to the players orders. There's so much potential laying there, just need to allow the player bypass the AI automatic "danger behavior", and the mission makers will bring the game to a whole new level.

[chortles 263]

We hear rumors like "the AI code is built on binarized pre-Aramaic language unreadable by all but the equally ancient Suma Lord..."

"Who may or may not even remember (or care) enough about Arma to help translate for anyone else..."

I'd also note that the whole idea of "just streamline new code, don't add any new features" is belied by the dev leadership's own statements about "AI compatibility with new features as part of delivering said features", something that even VBS3 doesn't necessarily guarantee (i.e. AI can't fire-from-vehicles in VBS3) and which lends credibility to Variable's concern about AI development being ongoing but prioritizing DLC-coinciding (I won't say related because hey, free "platform update") functionality. Not that we mind it in and of itself, of course, but when it's believed to mean less attention paid to "more basic" stuff like orders prioritizing or AI "knowsAbout"...

EDIT: The worst part from the end user end (especially for newcomers) is the inability to tell why (or what) the AI is prioritizing (though by now I guess the veteran players can tell) and, if "prioritize this above all else" exists in the engine, a UI that doesn't let us actually select that.

Edited October 4, 2014 by Chortles

[Alwarren 2767]

the most pressing issue with the AI is it's unresponsiveness and slowness, especially under fire.

My point exactly. There is certain doom waiting if you have to rely on the AI to react to your commands if they are in Danger mode. Making a fast getaway is almost always impossible, which is especially troublesome in a campaign that is mostly centered around guerrilla warfare where a fast getaway is a standard tactic. Just take one of the early missions in Adapt, the one which has you attack a convoy. I wanted to disengage quickly, get into the truck and get away. I ended up with only one guy left because the others just did their usual breakdance (face to the dirt - get up - run a bit - face to the dirt).

If I come across as repetitive here, it is because there are just so many incidences were this not only kills your team but also any enjoyment of the game. Managing AI is a nightmare, getting them to do what you want is an exercise in frustration. This one single aspect alone would be a tremendous improvement over the current behavior.

[Bouben 3]

My point exactly. There is certain doom waiting if you have to rely on the AI to react to your commands if they are in Danger mode. Making a fast getaway is almost always impossible, which is especially troublesome in a campaign that is mostly centered around guerrilla warfare where a fast getaway is a standard tactic. Just take one of the early missions in Adapt, the one which has you attack a convoy. I wanted to disengage quickly, get into the truck and get away. I ended up with only one guy left because the others just did their usual breakdance (face to the dirt - get up - run a bit - face to the dirt).

If I come across as repetitive here, it is because there are just so many incidences were this not only kills your team but also any enjoyment of the game. Managing AI is a nightmare, getting them to do what you want is an exercise in frustration. This one single aspect alone would be a tremendous improvement over the current behavior.

Very well written. I completely agree.

[Variable 322]

Just take one of the early missions in Adapt, the one which has you attack a convoy. I wanted to disengage quickly, get into the truck and get away. I ended up with only one guy left because the others just did their usual breakdance (face to the dirt - get up - run a bit - face to the dirt)

And that was a campaign that was clearly designed with a lot of effort to wrork-around the AI inability to move fast. Without that problem, I bet BIS mission designers could have created a much more immersive and interesting campaign. So if the official campaign suffered from that, at least fix the problem now, for the sake of future community (and BIS?) missions.

Edited October 4, 2014 by Variable