Jump to content

Object size and AI detection.


Hedgehog

Recommended Posts

So this happened today:


 

Quote

 

SB AI in a nutshell: Imma ignore the 4 tanks frontally attacking me and go for the FO guy in my left rear who is frantically crawling for his life.

 

SS_15_24_27.thumb.png.2e0703b387fa13e7e8b6284f027ca0da.png

SS_15_40_53.thumb.png.466d5ca1ad5bd3ad318fda62f7aaea20.png

SS_15_43_28.thumb.png.2e3ca6766e384d9eb1cade2cfc6b5e0f.png


 

 

And it got me wondering, given that the AI has "eSim Laser Vision(tm)" aided by thermals, radar and general omnipotence (When it wants to) etc etc

 

Does the object size come into play when determining the detection of units?

In the above example I'd have thought 4 x T-55s were bigger / more immediate threat than a lone FO dude running away.

 

 

 

 

Link to comment
Share on other sites

  • Members

Target prioritization is a notoriously difficult topic. I think that most of the time the AI vehicle commander is relatively competent. But of course, the few cases where he acts really stupid are the memorable ones. As there is no perfect formula that can consider the perfect solution for every possible constellation of target arrays - a complete enumeration would create about a fantastillion permutations, probably more - we have to rely on heuristics, rules of thumb, applied by brain-dead zombies.

 

If it is of any consolation, we'll try a new approach in version 5. I can't promise a solution in version 4.x

Link to comment
Share on other sites

  • Members
6 hours ago, Hedgehog said:

given that the AI has "eSim Laser Vision(tm)" aided by thermals, radar and general omnipotence (When it wants to) etc etc

Actually, it doesn't.

 

6 hours ago, Hedgehog said:

Does the object size come into play when determining the detection of units?

Yes.

But it plays less of a role than you might think. Once detected, a (hostile) unit goes into the AI target matrix, and every unit uses that same matrix to determine its individual priority order of all the targets in range (based on available munitions), based on range and where the threat is based on the unit's own orientation.

With FO units, the thing is that AI units are being denied the information what specialty observed infantry is. It's all "infantry" and they all might carry a potent RPG or even ATGM. So an infantry unit at about 400m range in your deep flank potentially is a tank killer. IOW, for the AI you were surrounded by five tank killers. Which one should have taken priority? It doesn't really matter much, in all likelihood you're toast in such a situation, anyway.

Link to comment
Share on other sites

10 hours ago, Ssnake said:

Actually, it doesn't.

 

Yes.

But it plays less of a role than you might think. Once detected, a (hostile) unit goes into the AI target matrix, and every unit uses that same matrix to determine its individual priority order of all the targets in range (based on available munitions), based on range and where the threat is based on the unit's own orientation.

With FO units, the thing is that AI units are being denied the information what specialty observed infantry is. It's all "infantry" and they all might carry a potent RPG or even ATGM. So an infantry unit at about 400m range in your deep flank potentially is a tank killer. IOW, for the AI you were surrounded by five tank killers. Which one should have taken priority? It doesn't really matter much, in all likelihood you're toast in such a situation, anyway.

I think the problem is beyond the target matrix problrmatic and into the magic infantry spotting problem domain. I find it highly unlikely that a tank would even spot a single infantry soldier, 400m away, in its deep flank while all the gunners are pointing their TIR sights to the front arcs where MBT were already identified.

Link to comment
Share on other sites

  • Members

Yeah, but, well, this isn't how it works in the code; once a hostile unit's position is known, it's in the hive mind; the individual unit can't "un-know" the location.

Maybe we should have different code. But it should also perform well. So, it's all about striking a balance that, most of the time, delivers acceptable results. I was merely reporting what the situation is to explain how it can happen. I'm not here to tell you that your eyes are lying and you didn't see the problems that you described.

Link to comment
Share on other sites

My post below is a bit rambly, I've kinda just text dumped on to the page, sorry.

 

TL:DR / pertinent points:

I'd like to see Red Map Diamonds with question marks in them at long range (depending on the unit but 2k plus), unless the Red Diamonds start shooting at the Blue Map Rectangles Circles.

And if an enemy unit is not seen for I dunno 5* mins (depending on circumstances) he drops off the hive mind until spotted again, and a bit of code to tell said unit to cut his losses and bugger off (If allowed).

Increasing the AI spotting area from 1 pixel to 3x3 or 5x5 pixels.

*Arbitrary figure

 

 

50 minutes ago, Ssnake said:

can't "un-know" the location.

This is probably where SB AI has it's greatest failing.

 

in my opinion.

 

I think there should be a matrix as to how a unit can "Forget" where it has seen an enemey unit.

In my example, a single infantry man sloping off should be "forgotten" quicker than 3 x T-55s that are attacking in force.

I'd make it simple and use a timer with modifiers (I.E. moving further away = quicker forgotten outcome.) and statrt the timer when the unit cannot be seen by friendly units.

Of course if the Infantry man gets spotted again, he goes to the top.

 

You could also have a global function of "movement to not be seen / move from last seen location"

I am primarily thinking of the Jockeying AFVs do in a "hold/defend/guard tactic" and expanding it (Mainly so Infantry move around to avoid arty fire, if you want them to.)

 

In the infantry for example a unit without any determined tactic, the training could have a input, the better training they have dictates what they do.

Untrained could just sit there fat dumb and happy, Militia might runaway from the last seen unit ASAP, Regulars could move laterally side to side of a "Frontline trace" to avoid artillery fire, Elite might try and sneak in the direction of last seen enemy

 

"Frontline trace" could be set as a line in mission editor or determined using center of mass of party units E.G. "blob of blue" on one side and "blob of red" on the other front line down the middle.

I would add the caveate of having a methoid of ignoring all this so the mission designer can specify what they want the AI to do.

 

15 hours ago, Ssnake said:
22 hours ago, Hedgehog said:

given that the AI has "eSim Laser Vision(tm)" aided by thermals, radar and general omnipotence (When it wants to) etc etc

Actually, it doesn't.

The AI instantly know everything about a unit as soon as it sees one pixel comes very close.

I'd suggest increasing the spotting area 3x3 or 5x5 pixels and expanding the "infantry type determination delay"

 

E.G. "Something over there" >

Delay (unless it does something aggresive like fire a 100m gun in your direction, in which case: "Ping! Tank T-55/T-62! Aggressive! Priority target! Disseminate to hive mind units within 2000m") >

Vehicle or infantry*? >

If vehicle: Tank / PC, SPG, etc / Truck >

What's it doing? > etc.

 

*The existing determination code seems to work OK.

 

And a matrix to modifiy the time taken, such as.

Is it in front something or obscured by vegetaion?

Is it swwking hull down? or taking aggresive action?

Is it trying to hide? or taking defensive action?

How far away is it? <- this one "Should" have a significant impact

Is it skylined?

 

I am hoping this is coming over as constructive.

I am trying to help you make a better product after all. :)

I hope the ideas I am suggesting make it into Ver 5.127 (or sooner, but happy with later.)

 

Anyway I'm going back to simulating CAS Sperwers with rockets.

Link to comment
Share on other sites

I had a similar case, quite a long time ago. When assaulting a T-80U platoon hiding in the woods just 500 meters away, the AI saw fit to turn the M1A1 turrets to the rear and engage a lone BMP-1 in a hull-down position at 2000 meters range. That BMP was pretty passive, didn't launch a single Sagger, but the T-80s were more potent and soon the whole M1A1 platoon was wiped out.

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...