Jump to content

Tuning Down Ai Detecting?


Recommended Posts

Posted

Currently the AI detecting aircraft seems to be too good. They call in enemies that are invisible for the human eye. In my opinion this should be tuned down. Other thing is the capability to tell friend from foe in a head on attack, this should also be tuned down I think. They open fire already at a large distance.

  • Upvote 1
Posted

Spotting, object recognition, field of vision and overall situational awareness are some of the hardest things to model realistically for AI. It's not something you simple 'tune up or down'.

Posted

You should be able to program distances where AI can recognize aircraft and can tell friend from foe.

  • Upvote 1
Posted

Yes, but then you're faced with further complications: How should the AI react to aircraft it can 'see' but not yet identify?

=VARP=Cygann
Posted

Why is that a complication? If it see and don't recognize it should do what humans do, try to close in for recognition while gaining E advantage if possible at the same time... 

 

Seems very simple and logical without any complications from my POW.

Posted

And the result will be, what we had in RoF for a loooong time: The AI constantly going on wild goose chases and getting sidetracked from their mission until they run out of fuel and crash.

 

Look: I'm not saying things like these can't or shouldn't be done, in fact I'm all for building ever more realism into the AI. But you have to understand, that such changes are not simple to make and often affect other parts of the AIs behavior in unexpected ways. Make a small change in one place and you might have to make 10 other changes in other places.

Guest deleted@30725
Posted

The AI is not too bad except when they get stuck in turn fights or try and fly their bomber like a fighter plane.

Posted

The AI for IL/2 Sturmovik and Pe-3 are a bad joke right now - total immersion killers. Bat turns, zoom climbs - flying just like Yak would fly. It's terrible.

=VARP=Cygann
Posted

@Finkeren - programming is not as complicated as some are led to believe.

 

I am not a game developer so I will not pretend I know how they did their coding and why they did it the way they did it. But I am software architect on some very complex business solutions and every time something was too difficult to modify or complex to add a new feature, it was directly associated with my decisions and it was entirely my fault (not the fault of the programmer that made implementation, but mine since it was me designing and directing solutions).

Even if someone else has made a mistake, it was my job to notice and make him fix that in time.

 

So if touching some code break 10 other things, I'd just blame man in charge and I am very confident I would not be wrong by doing so. That is plain and simple bad work if such things can happen and if there are no unit tests written for critical product sections. I don't really think that is the case here or you would see much more bugs in IL2. And since there are not many, I find it reasonable to assume they did their job properly and in that case, it's is ok to ask small changes like one in question here.

 

Also I will add one other thing not related to the topic, but still related to the way 777 spend resources. I think it is a mistake to try so hard to make AI fly with all real control surfaces. I still did not see in ROF or BOS AI that flat spins, stalls or do any human alike mistake in heat of dogfight. So I don't see the reason to have AI flying with 1000 parameters and then our current hardware can't handle more then 32. Other game developers proved that with good scripting, player can be fooled good enough without breaking laws of physics. I'd rather have 60-70 planes in the air that behave like AI in old IL2:1946, then up to maximum 30ish that usually just do spiral turn like in ROF. Fact that AI actually fly the planes same way player does does not contribute in any way if it is not more convincing then well scripted AI that does not compute so much variables but uses common tactics of areal combat better.

Also the fact that AI spawns in air on triggers in order to save resources (as opposed to taking of like in old IL2) says enough about how maybe it is a bit too early for an AI that controls a plane by actual control surfaces same way human would. It will be a nice thing in the future probably when average CPU power will be many times more then what we have now. 

  • Upvote 1
Posted

I'm just going to say one thing:

 

If programming believable AI was easy, we would have had it a long time ago.

Guest deleted@1562
Posted

Probably beating a dead horse, but BoB2's AI design was posted by the creator and it didn't seem to be complicated to implement.

=VARP=Cygann
Posted (edited)

I'm just going to say one thing:

 

If programming believable AI was easy, we would have had it a long time ago.

 

That is my point, we are not ready with current CPU power to have AI at home PC yet even if they could make it true to the definition. But well scripted bot can be quite good opponent that can fool you well enough.

 

It does not have to be anywhere near real AI, but it does have to be less predictable and repetitive then it is in ROF/BOS to be a challenge (by that I don't mean accuracy, I speak of flight maneuvers, both offensive and defensive and algorithm to randomize it a little so not every AI plane fights the same as they do now).

 

That along with AI spawn on trigger instead of taking off and having it "out there" somewhere at all times is only thing that kind of kills SP for me. I know they won't (probably even can't at this stage) change that, but it should be something to think about for 3rd engine version sometime in the future.

Edited by EAF19_Cyclops
  • Upvote 1
  • LukeFF locked this topic
Guest
This topic is now closed to further replies.
×
×
  • Create New...