missing info swapping targets Ai Petrovich

[Robo76]

Ai Petrovic has a problem if he loses his aim and goes behind the terrain. If something else appears near his line of sight at that moment, he continues to lock on to that new target without letting it know he is pointing at another target. This is how I selected and confirmed an Apache (NOT the yellow helicopter item) in Ai Petrovich, which flew behind terrain. Normally Ai would report losing the target (spinning right), but this time Ai did not break tracking, and started tracking and homing in on Kamov (spinning left), which was between the Mi-24 and the Apache. It was in multiplayer, the footage is here (time 13:28:00)
https://drive.google.com/file/d/19A8SUtv53AOuCq30PJCYUkgwZgNDomDA/view?usp=drive_link

Translated with www.DeepL.com/Translator (free version)

[Robo76]

One more note - the Apache that was further away, Petrovich found ....at the moment when I was selecting Apache in the menu, he didn't see the closer Kamov, it wasn't in the menu.

[zerO_crash]

11 hours ago, Robo76 said:

One more note - the Apache that was further away, Petrovich found ....at the moment when I was selecting Apache in the menu, he didn't see the closer Kamov, it wasn't in the menu.

Any ED AI (Petrovich, George, etc...) has realistic spotting algorithms, not only based on field of view, but contrast, background, point of interest (designated target), etc... This is based on real studies performed on human awereness and observation. There will be situations that might seem odd, but are realistic in that the Ah-64 at distance might have been easier to see, than the Ka-50 close by.

 

As an example, consider a WWII-scenario. If you fly by using BnZ (boom and zoom) tactics, you will have a much easier task spotting targets further away, albeit contrasted against somewhat bland in color landscape. That, vs. trying to spot targets straight below you, which will be camouflaged by terrain, shadows (mostly from clouds), and the fact that you cover less area with your peripherial vision. This is just an example, but it's perfectly normal and as designed, that AI isn't acting like a radar. He is given human deficiencies, he will make mistakes.

 

As to the first post, it's clearly a bug.

Edited February 8, 2024 by zerO_crash

[Robo76]

I don't have a problem with the shift to more realism, quite possibly Kamov found it, but the menu was already closed after the target was selected. Mistakes that a real operator wouldn't make still need to be addressed. From the situation above, I get the feeling that the target tracking algorithm is trying to keep the selected target by repeatedly searching for it, but not checking that it's still the same target to reliably detect target loss. But that's more of a question for Ai Petrovich programmers. 

[zerO_crash]

The AI is, and will, constantly be under development. There are many improvements to come. I cannot really tell you whether the issue is with the algorithm, or most likely the selection process of aerial units (this is not an issue with ground targets). It would be best if you could upload a .miz (trk) file, in order for this issue to be recognized and solved. The .acmi (Tacview) does sometimes produce artifacts in the form of desync, position of units, time frame, etc... This certainly is a bug, so no doubt there.

[Robo76]

I don't have the trk file unfortunately, for BF I automatically deleted them before starting as they are too big and if forgotten they fill up the disk and DCS crashes. It might be a good idea to make the DCS settings automatically delete trk files older than x days, or keep the last x files. I now set this in the boot bat file.

[Robo76]

On 2/9/2024 at 3:06 AM, zerO_crash said:

The AI is, and will, constantly be under development. There are many improvements to come. I cannot really tell you whether the issue is with the algorithm, or most likely the selection process of aerial units (this is not an issue with ground targets). It would be best if you could upload a .miz (trk) file, in order for this issue to be recognized and solved. The .acmi (Tacview) does sometimes produce artifacts in the form of desync, position of units, time frame, etc... This certainly is a bug, so no doubt there.

vygenerujte acmi pomocí souboru trk DCSWorld (dogsofwarvu.com)

[zerO_crash]

2 hours ago, Robo76 said:

vygenerujte acmi pomocí souboru trk DCSWorld (dogsofwarvu.com)

DCS .trk-files are have their own issues as well. Typically, the longer the session is, the more inaccurate the .trk becomes. Which of ACMI/TRK however, are more accurate, is a technical question irrelevant of this thread. The point is, ED uses TRK for all their intents and purposes. I suppose this has to do with having an accurate reproduction of mission events and user experienced bug/glitch in the same space of time. Another thing that I can think of, is that as far as I know, ACMI only records metrics and units, not actual events taking place in the mission (triggers, module systems info (switches, etc...).That's why ED specifically asks for TRK, not ACMI. Beyond that, a ED dev would better be able to explain why exactly the distinction.

[Robo76]

I Petrovitch has been absolutely tragic since the "improvements". If the infantry is near a forest, Ai searches all the trees in the vicinity. Even if the infantry is not covered by trees, it takes Ai the same amount of time to find it as if it were in the middle of the forest. It's absolutely tragic when infantry near a forest is marked by smoke from a Reaper. Ai ignores the smoke though, even when the troops are shooting at the helicopter. Ai doesn't have priority detection that something is firing at her, and you can see the light trails of the shots.  If the firing units are between the trucks, Ai can see the trucks but not the Bofors firing at her. I don't have the nerve. Get it right finally, or better yet, bring back the original Ai search. If you want footage again, sit in your creation on the MP server at least once and you'll be amazed at how unrealistic it is.

Translated with www.DeepL.com/Translator