AI Discussion (dev branch)

[FallujahMedic -FM- 867]

draws line in the sand

This thread has essentially derailed from what it was intended and Requested to be. See: klamacz's original post 

It was not intended as a "moan and complain" thread. As was stated by the OP: 
 

Quote

It really helps us in this thread when you provide considered feedback to the changes in progress, or clean repros missions that highlight some repeatable AI behaviour.

Therefore, lets return to its original purpose, lest it be locked. Further spammy and smart-alecky posts will be dealt with accordingly. 
 

[lex__1 422]

1 hour ago, FallujahMedic -FM- said:

Encore update. Of the 181 items addressed, only 14 were DLC. The other 167 items were fixes, tweaks, updates and NON-DLC additions. 

Are we arguing that the doesn't need to improve? No, we are not. 
Has the Ai gotten better over successive iterations? Yes it has. 
Have BI taken a few steps back at times with relation to Ai? Yes, they have. 
Are they acting as fast as we want them to? Probably not. 

Lets be objective (rather than subjective) about the issue. If you feel that you can do the job better,  https://careers.bohemia.net/positions

In 2013 when I watched the preview of Arma3, there nothing was said that the most part of content, and the general development of Arma3 will be directed to improvement only of style of a game as PvP.
I understand that many AI take away many personal computers resources that visual improvements consume the personal computers resources too. Imposing of these requirements to the personal computer likely aren't located. But even in these restrictions one AI has to perform the functions correctly.

If to consider each special case where AI loses control, or is beyond not planned behavior, one and same problem of traveling points is traced.
It is the ticket, and a problem of AI https://feedback.bistudio.com/T82570  

The type of transport influences behavior of AI in traveling points.

Vehicles break the system of creation of the AI convoy

If BIS waits for this guy here https://careers.bohemia.net/positions , but he hasn't appeared, - there has to be own initiative and responsibility in this question?

[mickeymen 324]

On 03.09.2018 at 9:58 AM, FallujahMedic -FM- said:

Encore update. Of the 181 items addressed, only 14 were DLC. The other 167 items were fixes, tweaks, updates and NON-DLC additions. 

 

So what? You want to say that the standard updates never broke the game? Of course they broke. DLC brought a greater chance for breakdowns, while standard updates also had every chances broke the game... Perhaps in the developers' studio, there was a lot of staff turnover. Some people started the previous version of the game, while the next version was finished completely different people. For this reason, the docking of the two versions of the game always does not go smoothly

 

I see that while no one has any desire to create a petition that I proposed:

 

On 02.09.2018 at 10:35 PM, mickeymen said:

Dear users! All who are concerned about the fact of the destruction of AI in the game!

 

I think we need to come up with some kind of mechanism, an impact on the developers. Otherwise, the developers will not react in any way, to the desire to make AI more adequate. Every day the chances are increasingly melting! The remaining developers spend their efforts to create useless things (such as encore), which, moreover, are likely to add additional AI-errors. 

 

I do not even dream of having a new capabilities added to the AI.

I only dream of having everything that is stated in ArmA3 worked fine!

I mean every command for player subordinates in the command menu and every waypoint type and its settings, which is stated the game editor.

All this details must work 100%! Unfortunately, it seems to me, now 40% of those details in Arma3 is not functional or semi-functional.

 

All this is a sad experience, which makes me angry because I remember the moments when it all worked fine!

 

I propose to collect a petition

Let's think together, How it's best can be done?

A petition with a large number of signatures may have a chance

 

No any answer about.

Instead of wanting to create a petition, just chatter. 

Let's create a petition! Or Most likely this date is September 3, 2018, the final death for arma3 AI?

 

[R0adki11 3949]

5 hours ago, mickeymen said:

Let's create a petition! Or Most likely this date is September 3, 2018, the final death for arma3 AI?

 

 

If you feel the need to create a petition, why don't you ahead and create one?

[Tankbuster 1744]

@lex__1, reference your second video, that behaviour is more due to the roadpieces not joining up (roadsconnectedto) than AI, but yes, your point stands all the same.

[mickeymen 324]

15 hours ago, R0adki11 said:

 

If you feel the need to create a petition, why don't you ahead and create one?

 

I do not know how to do it. Where to create this post? In addition, I wanted to see interested users in such a petition, but it seems that they are not present in the nature.

 I'll have to face it ...

 

 

 

[The Man Without Qualities 110]

21 hours ago, mickeymen said:

 

I do not know how to do it. Where to create this post? In addition, I wanted to see interested users in such a petition, but it seems that they are not present in the nature.

 I'll have to face it ...

 

 

 

Ummm, one purpose of a petition IS to find out if someone cares...and then show the result to the addressed party of the petition.

Relevant decision makers should be aware about the condition of A3. If not, this company has bigger problems then AI issues.

 

[The Man Without Qualities 110]

A simple statement by BIS could close this:

 

Yes, we confirm the AI status as described in lex's videos - no, we won't fix it in A3.

[FallujahMedic -FM- 867]

Quote

 provide considered feedback to the changes in progress, or clean repros missions that highlight some repeatable AI behaviour.

All other comments will be removed. 

[fn_Quiksilver 1636]

AI "cost" config is wrong on many post-2013 assets

 

_cost = (getNumber (configFile >> 'CfgVehicles' >> (typeOf cursorobject) >> 'cost'));

 

https://community.bistudio.com/wiki/CfgVehicles_Config_Reference#cost

 

Quote

This value reflects the attractiveness of the target to the enemy AI. When all other considerations are equal. A soldier eg is not interested in Air, despite it's highly attractive cost values. On the other hand, he is interested in a medic, since a medic (in normal configs) has a (while small) a higher cost, than others in the group, including the officer.

 

B_soldier_F                                       // 100,000

B_MBT_01_cannon_F                     // 2,500,000

 

B_AFV_wheeled_01_cannon_F (rhino)     // 40,000 ???

 

 

[.kju 3244]

BI tweaked them in recent updates; https://pastebin.ubuntu.com/p/23xvRJyyG2/

[Variable 322]

AI Guros,

Could you help us understand the direction of each of the AI pref skills, meaning, a higher value means better, or otherwise? Of course, an answer for each specific value will be appreciated.

 

        0.1,        // aiming speed
        0.1, //0.03,        // aiming accuracy
        0.03,        // aiming shake
        0.05,        // spotting time
        0.05,        // spotting distance
        0.6,        // commanding ability
        0.6,        // general AI skill
        0            // allow fleeing. 0 means they won't flee

 

Thank you!

[reyhard 2082]

19 minutes ago, Variable said:

AI Guros,

Could you help us understand the direction of each of the AI pref skills, meaning, a higher value means better, or otherwise? Of course, an answer for each specific value will be appreciated.

 

        0.1,        // aiming speed
        0.1, //0.03,        // aiming accuracy
        0.03,        // aiming shake
        0.05,        // spotting time
        0.05,        // spotting distance
        0.6,        // commanding ability
        0.6,        // general AI skill
        0            // allow fleeing. 0 means they won't flee

 

Thank you!

https://community.bistudio.com/wiki/AI_Sub-skills

Do you mean this page?

[Variable 322]

Thank you @Reyhard!

[The Man Without Qualities 110]

On 19.9.2018 at 6:59 AM, .kju said:

BI tweaked them in recent updates; https://pastebin.ubuntu.com/p/23xvRJyyG2/

 

That brings me back to a question that wasn't answered to me since 2003:

 

If a soldier is facing 2 threats: enemy soldier and enemy tank but owns just a rifle, no AT, how is the AI behavior calculated?

Which parameters make that he might flee the (how?) asumed threat range of the tank? What is the preference if AI can attack both soldier and tank from (assumed) hidden position?

[pierremgi 4850]

Hi all,

It seems a command misses to grab all playable units (disabled AIs in lobby or not).

playableUnits returns the intended array, ONLY if AIs are not disabled in lobby.

if disabled, it's just similar to allPlayers.

 

The aim is to apply some scripts on playable units, regardless if played or not.

The problem is the disabled units simply "exist" in lobby (at least, you can choose a slot), but variable name and all init field, loadout or else can't be used, just existing in mission.sqm. All disabled unplayed units are nil.

get3denEntity doesn't work during mission...

Thanks

[nkenny 1057]

Expanding Suppression

 

I have been testing the effects of suppression on the AI and found the results to be impressive. Suppressing AI adds depth to the simulation and greatly enhances the firefight. Player survivability is vastly increased by the proper application of firepower. My experiments have used the getSuppression and setSuppression commands. Conclusions and numbers are explored HERE. The shortest description is that successfully suppressing AI by even a small amount reduces killing power considerably. 

 

A reoccurring shortcoming is that Bohemia has done a poor job of providing feedback that the AI is in a suppressed or panicked state.  There are no indicators, animation, audio, or visualization to indicate a soldier is in peril. No wonder players are ignorant of the feature!  With this in mind, I have a few small suggestions to enhance and expand the suppression feature. 

 

1. Add a Suppression Eventhandler,  this will allow mission designers an easy way to attach special scripted behavior to the condition. 

 

2. Track Suppression for players. One limitation of suppression values is that it is not tracked for players. Given that it seems possible to do so for all AI, without any particular loss of performance, additionally tracking it for players would open an easy and performance friendly way for modders and mission designers to modify player suppression effects or other scripted behaviors. 

 

3. Add a background suppression level or support for radiating suppression values . Possibly the biggest suggestion I will be making. This suggestion also marks the turn to more speculative features. Active armoured vehicles are dangerous. Being in close proximity to such a vehicle should set a default suppression level of some figure (I would suggest 0.25; roughly a close burst of fire).   This allows for better modelling of the shock-and-awe factor of tracked vehicle assaults. 

 

4. Tweak suppression regeneration. Currently suppression values regenerate very quickly. This limits their utility as a means of modeling the effect of combat stress. Increasing this time slightly or enable tweaking it, would help fine-tune the suppression mechanic as a whole. Once I set up my Arma3 rig again I will provide some numbers in this regards. Loss of suppression state is currently very rapid. 

 

There are bigger more fundamental issues in regards to the lack of visualization-- player feedback-- which I wish that Bohemia would explore. However in Arma3's current old man state, it is unrealistic to propose any major overhauls or feature introductions.  For future projects:  introducing  animations, sounds, distinct behavior, or other indicators-- in short any feedback at all is essential. 

 

-- 

 

To give an example of how I have leveraged suppression and eventhandlers: By attaching a 'GetOut' eventhandler to vehicle crew and with a triggered condition on damaged or immobilized vehicles, and setting AI suppression to 1 (100%)-- I have created a simple, performance friendly fix, to the laser-accurate-dismounting-from-burning-wreck crew syndrome. 

 

 

-k 

 

[lex__1 422]

Spoiler

1 hour ago, nkenny said:

Expanding Suppression

 

I have been testing the effects of suppression on the AI and found the results to be impressive. Suppressing AI adds depth to the simulation and greatly enhances the firefight. Player survivability is vastly increased by the proper application of firepower. My experiments have used the getSuppression and setSuppression commands. Conclusions and numbers are explored HERE. The shortest description is that successfully suppressing AI by even a small amount reduces killing power considerably. 

 

A reoccurring shortcoming is that Bohemia has done a poor job of providing feedback that the AI is in a suppressed or panicked state.  There are no indicators, animation, audio, or visualization to indicate a soldier is in peril. No wonder players are ignorant of the feature!  With this in mind, I have a few small suggestions to enhance and expand the suppression feature. 

 

1. Add a Suppression Eventhandler,  this will allow mission designers an easy way to attach special scripted behavior to the condition. 

 

2. Track Suppression for players. One limitation of suppression values is that it is not tracked for players. Given that it seems possible to do so for all AI, without any particular loss of performance, additionally tracking it for players would open an easy and performance friendly way for modders and mission designers to create a player suppression or other scripted behaviors. 

 

3. Add a background suppression level or support for radiating suppression values . Possibly the biggest suggestion I will be making. This suggestion also marks the turn to more speculative features. Active armoured vehicles are dangerous. Being in close proximity to such a vehicle should set a default suppression level of some figure (I would suggest 0.25; roughly a close burst of fire).   This allows for better modelling of the shock-and-awe factor of tracked vehicle assaults. 

 

4. Tweak suppression regeneration. Currently suppression values regenerate very quickly. This limits their utility as a means of modeling the effect of combat stress. Increasing this time slightly or enable tweaking it, would help fine-tune the suppression mechanic as a whole. Once I set up my Arma3 rig again I will provide some numbers in this regards. Loss of suppression state is currently very rapid. 

 

There are bigger more fundamental issues in regards to the lack of visualization-- player feedback-- which I wish that Bohemia would explore. However in Arma3's current old man state, it is unrealistic to propose any major overhauls or feature introductions.  For future projects:  introducing  animations, sounds, distinct behavior, or other indicators-- in short any feedback at all is essential. 

 

-- 

 

To give an example of how I have leveraged suppression and eventhandlers: By attaching a 'GetOut' eventhandler to vehicle crew and with a triggered condition on damaged or immobilized vehicles, and setting AI suppression to 1 (100%)-- I have created a simple, performance friendly fix, to the laser-accurate-dismounting-from-burning-wreck crew syndrome.  

 

 

-k 

 

Addition 3den Enhanced makes access to similar settings in the editor. It expands access to inaccessible settings in the vanilla editor.

 

[lex__1 422]

1 hour ago, nkenny said:

Expanding Suppression

 

I have been testing the effects of suppression on the AI and found the results to be impressive. Suppressing AI adds depth to the simulation and greatly enhances the firefight. Player survivability is vastly increased by the proper application of firepower. My experiments have used the getSuppression and setSuppression commands. Conclusions and numbers are explored HERE. The shortest description is that successfully suppressing AI by even a small amount reduces killing power considerably. 

 

A reoccurring shortcoming is that Bohemia has done a poor job of providing feedback that the AI is in a suppressed or panicked state.  There are no indicators, animation, audio, or visualization to indicate a soldier is in peril. No wonder players are ignorant of the feature!  With this in mind, I have a few small suggestions to enhance and expand the suppression feature. 

 

1. Add a Suppression Eventhandler,  this will allow mission designers an easy way to attach special scripted behavior to the condition. 

 

2. Track Suppression for players. One limitation of suppression values is that it is not tracked for players. Given that it seems possible to do so for all AI, without any particular loss of performance, additionally tracking it for players would open an easy and performance friendly way for modders and mission designers to create a player suppression or other scripted behaviors. 

 

3. Add a background suppression level or support for radiating suppression values . Possibly the biggest suggestion I will be making. This suggestion also marks the turn to more speculative features. Active armoured vehicles are dangerous. Being in close proximity to such a vehicle should set a default suppression level of some figure (I would suggest 0.25; roughly a close burst of fire).   This allows for better modelling of the shock-and-awe factor of tracked vehicle assaults. 

 

4. Tweak suppression regeneration. Currently suppression values regenerate very quickly. This limits their utility as a means of modeling the effect of combat stress. Increasing this time slightly or enable tweaking it, would help fine-tune the suppression mechanic as a whole. Once I set up my Arma3 rig again I will provide some numbers in this regards. Loss of suppression state is currently very rapid. 

 

There are bigger more fundamental issues in regards to the lack of visualization-- player feedback-- which I wish that Bohemia would explore. However in Arma3's current old man state, it is unrealistic to propose any major overhauls or feature introductions.  For future projects:  introducing  animations, sounds, distinct behavior, or other indicators-- in short any feedback at all is essential. 

 

-- 

 

To give an example of how I have leveraged suppression and eventhandlers: By attaching a 'GetOut' eventhandler to vehicle crew and with a triggered condition on damaged or immobilized vehicles, and setting AI suppression to 1 (100%)-- I have created a simple, performance friendly fix, to the laser-accurate-dismounting-from-burning-wreck crew syndrome.  

 

 

-k 

 

Addition 3den Enhanced makes access to similar settings in the editor. It expands access to inaccessible settings in the vanilla editor.

[nkenny 1057]

Some information in regards to AI suppression: 

1. Suppression is measured 0 to 1. 

1 is max suppression. 0 is the default state. 

 

2. Suppression reduces effective AI accuracy. 

This has been tested in some detail here here. 

 

3. The time an AI is suppressed is affected only by Skill and not rank. 

A max skilled soldier will remain suppressed for less than 1 second

A 50% skilled soldier will remain suppressed for around 7 seconds

A 20% skilled soldier will remain suppressed for 11 or more seconds. 

 

The chart reads like this:

edit: Suppression here indicates how much time in seconds it takes a soldier to go from Full to no suppression. 

 

4. Suppression seems unaffected by caliber. 

It was my belief that higher caliber weapons would increase Suppression values. However in testing with a tank I did not find this to be the case. 120mm overhead had the same effect as a 7.62x51mm.  Driving the tank closer to the AI changed nothing. Explosives seem to have no special suppressive qualities.  edit: Further testing does seem to indicate some difference between near hits by 5.56x45 and 9.3x64. However the brief period of suppression means that this makes for little if any difference in practical terms. 

 

Testing suppression yourself

One way of testing suppression values is by pasting the following code into your own mission. (NB: works best for smaller ops, and only in scenarios were EAST is the enemy)

Spoiler

// Check suppression 
while {true} do 
  {
  
  // define
  _txt = format ["SUPPRESSION:<br/>(count %1 knows %2)<br/>",count (allUnits select {side _x == EAST }),east knowsAbout player];
  
  // check it!
  {
  _msg = format ["- %1 suppression %2<br/>",(getText (configFile >> "CfgVehicles" >> typeOf _x >> "displayName")),(getSuppression _x) toFixed 2];
  _txt = (_txt + _msg); 
  } count (allUnits select {side _x == EAST && {getSuppression _x > 0}}); 
  
  //  wait
  sleep 0.5; 
  
  // display!
  hintSilent parsetext _txt;
  
  }; 

 

 

Another is simply by playing the small scenario I've put together for that purpose: LINK

 

All tests performed on the latest Devbranch, 145791

[Cenwulf 40]

From my admittedly brief tests, it seems that passengers in vehicles receive zero suppression (or are at least are very resistant to it). This makes sense for some vehicles like tanks, less so for an open topped transport truck.

I agree that a suppression event handler would be very useful to allow for fine control of this mechanic.

I was hoping to use this for an ambush scenario where a get out event handler would be used to amplify a units suppression when exiting a vehicle (to simulate jumping out of your nice relatively safe vehicle into a hail of bullets) but as they weren’t receiving suppression in the first place will need to revert to less elegant solutions like setting suppression to a fixed value upon exiting.

Fully agree that suppression is a fantastic mechanic, it’s just under utilised and needs be be fleshed out out a bit.

[fn_Quiksilver 1636]

The internal function is already there, please gib us a script command to return "currently attacked target" (dont confuse this with "assignedTarget").

 

 

please 🙂

[Dedmen 2696]

On 6/24/2019 at 9:32 AM, fn_Quiksilver said:

please 🙂

Code for that was written 18 days ago. But there are currently bigger priorities, and pushing your request every week won't make it go faster.

[Varanon 892]

On 6/21/2019 at 10:31 AM, Cenwulf said:

From my admittedly brief tests, it seems that passengers in vehicles receive zero suppression (or are at least are very resistant to it). This makes sense for some vehicles like tanks, less so for an open topped transport truck.

Or a static weapon like a machine gun. This basically means that AI in static weapons are not affected at all from suppression, which is rather nonsensical

[Alwarren 2767]

2 hours ago, Varanon said:

Or a static weapon like a machine gun. This basically means that AI in static weapons are not affected at all from suppression, which is rather nonsensical

 

In static weapons, units are only getting suppressed IF the bullets hit nearby, while normal units get suppressed when the bullets fly by close enough OR they hit the ground close by.