Cv90/30 ai won't shoot, even at close range.

[Juusp]

Here is the scenario:

 

https://drive.google.com/file/d/1ot1eLHNBOOOJhBOX9-jIO0RBJQ6l8z25/view?usp=sharing

 

Here is a video of the issue:

 

 

[TankHunter]

They will not shoot at things that they do not expect to penetrate. That is intended.

[Juusp]

2 hours ago, TankHunter said:

They will not shoot at things that they do not expect to penetrate. That is intended.

I asked the same question on reddit and gave them the mission file. A user on there was able to make them fire at close range.

[Ssnake]

MAYBE they should open fire the moment that, in the video, the M1 turns around first the hull, then also the turret. Zapping a few into the buste rack would take away most of the ammo, and I suppose shooting the rear would also immobilize the M1. So, in that sense I agree that there's something fishy.

We had a bug where AI units would always use the frontal armor values of a target for their shoot/no shoot decisions. But that was supposedly fixed a few years ago. I suppose we'll look into the matter.

[Juusp]

20 hours ago, Ssnake said:

MAYBE they should open fire the moment that, in the video, the M1 turns around first the hull, then also the turret. Zapping a few into the buste rack would take away most of the ammo, and I suppose shooting the rear would also immobilize the M1. So, in that sense I agree that there's something fishy.

We had a bug where AI units would always use the frontal armor values of a target for their shoot/no shoot decisions. But that was supposedly fixed a few years ago. I suppose we'll look into the matter.

I did some testing. The same issue persists on the cv90/35 variants. The cv 90/40 variants will engage anything as inteded. The cv90/30 and 90/35 variants do not engage any leo2 or m1 variants under any circumstances. They do however engage the leo1 variants, all t72 variants and even the t90. No clue what could be causing this.

[Volcano]

Well, it is because that CV9040 gun is more powerful (170mm KE RHA penetration versus 110). There is an actual literal cut off point where the AI decides "no".

 

And even if the AI fired on the vehicle that it could *damage but not kill*,  there would be a complaint that the AI damaged the vehicle to the point of all it could accomplish (all the damages it could achieve), yet continues to fire and not kill the threat, using up all the ammo. So then the AI would have to be aware (cheat) of whether or not it damaged the target sufficiently, then stop wasting ammo. Or the AI could be cleared to fire a portion of its ammo, not re-engaging again unless the target fires, perhaps. Or it could be that the AI is allowed to fire a % of its ammo in self defense like this, but save the rest for targets it can kill. In the latter case, the user would not know what is going on, and would think that is a bug.

 

There is no easy solution here, as either behavior causes a problem - and the current behavior is designed to be the lesser of evils, but not perfect of course. 

 

The current behavior is that these vehicles are expected to operate with other entities that can kill the tank (whether it is infantry with RPG and ATGMs, or friendly tanks). The AI doesn't waste any ammo on things it cannot kill, but yes, you end up with edge cases where the vehicle is alone and the AI could damage a threat but not kill it, and it doesn't attempt it at all, nor is it able to ascertain that a threat is damaged sufficiently enough to stop wasting ammo (most humans do not know that either, granted, but they usually have a gut feeling on when they are wasting ammo). 

 

In order to make the AI "smart" in that regard, it would take a lot of conditions to evaluate, which would not always be the same in every situation. It is certainly a difficult subject.

[Juusp]

21 hours ago, Volcano said:

Well, it is because that CV9040 gun is more powerful (170mm KE RHA penetration versus 110). There is an actual literal cut off point where the AI decides "no".

 

And even if the AI fired on the vehicle that it could *damage but not kill*,  there would be a complaint that the AI damaged the vehicle to the point of all it could accomplish (all the damages it could achieve), yet continues to fire and not kill the threat, using up all the ammo. So then the AI would have to be aware (cheat) of whether or not it damaged the target sufficiently, then stop wasting ammo. Or the AI could be cleared to fire a portion of its ammo, not re-engaging again unless the target fires, perhaps. Or it could be that the AI is allowed to fire a % of its ammo in self defense like this, but save the rest for targets it can kill. In the latter case, the user would not know what is going on, and would think that is a bug.

 

There is no easy solution here, as either behavior causes a problem - and the current behavior is designed to be the lesser of evils, but not perfect of course. 

 

The current behavior is that these vehicles are expected to operate with other entities that can kill the tank (whether it is infantry with RPG and ATGMs, or friendly tanks). The AI doesn't waste any ammo on things it cannot kill, but yes, you end up with edge cases where the vehicle is alone and the AI could damage a threat but not kill it, and it doesn't attempt it at all, nor is it able to ascertain that a threat is damaged sufficiently enough to stop wasting ammo (most humans do not know that either, granted, but they usually have a gut feeling on when they are wasting ammo). 

 

In order to make the AI "smart" in that regard, it would take a lot of conditions to evaluate, which would not always be the same in every situation. It is certainly a difficult subject.

Would be nice if it was an option in the mission editor so it would be up to the mission maker to decide.

[Ssnake]

Oh great, let's develop two A.I.s to toggle between. What could possibly go wrong?