BVR Match Play AI does a weird initial crank putting it at disadvantage
-
So I found this weird behavior:
Using Match Play with 4 radar missiles and 2 heatseekers with one veteran or ace opponent. All default except that initial separation is 30 to 45 nm and strategy set to Random or Pure Pursuit. In 4 out of 5 cases the AI craft at the start does this weird crank before any missiles are fired and then recommits. This puts it at quite the disadvantage against the human who flies straight at it, as it is at a lower speed then. I don’t know what this is supposed to do for it.
Looking at the Tacview wondering why it was so easy revealed this to me. The one time where it shoot me down was the one where it didn’t do this initial crank, but just flew straight at me as it should until firing it’s own missile.Screenshot attached below.
Purple is the direct line between the two starting spots. You can see this rightward (leftward happens too sometimes) crank and then the recommit. After that it will behave somewhat sensibly but will be at a disadvantage all the way thorough as it is slower.Is this a bug? What am I missing?
-
I usually don’t get too excited about issues within the DF module as I don’t think it has been a priority with the dev team for quite some time.
Maybe it’s trying to beam your radar if you have him hard locked, or perhaps the AI thinks it is part of a 4 ship and is doing a pincer move with his “ghost friends”.
TE would probably behave better, but I do understand that DF mode is much easier for practice. I would just keep playing it over and over again and see if behavior changes over time.
-
there is a section in the comms manual that explains this maneuver, they are trying to Notch you not a bug
-
This is not useful at all though, as being 45 degrees off my flight path never results in me loosing the radar lock. And they recommit after about 20 seconds which is still before either of us has fired a missile.
schnirdman, what is TE? All I want is to test in a mode that allows me to train against an AI opponent that will fly straight at me gaining speed until they have fired their first missile. I am willing to use a less convenient mode for that.
-
TE = Tactical Engagement (like the training missions … but you can tweak those or create your own)
TR#18 BARCAP is a possible starting point … it’s more of a 2v2 setup iirc but perhaps you can adapt it.
-
This is not useful at all though, as being 45 degrees off my flight path never results in me loosing the radar lock. And they recommit after about 20 seconds which is still before either of us has fired a missile.
however his Flightpath was perpendicular to yours when he started the maneuver he is still trying to notch you, you would of lost radar returns at that point
-
AI behavior close the ground isn’t optimal yet…
Hopefully, someone will do something about it in the future.
-
Most of the tactics except pure pursuit or grinder will involve a notch
Obviously those tactics are more adapted to 2vsx or 4vsx scenarios
So select pure pursuit and not random in a 1vsx scenario
-
AI behavior close the ground isn’t optimal yet…
Hopefully, someone will do something about it in the future.
Is this kind of behavior set at compile time of Falcon, or would I be able to help out/ change some ini-like files myself to make or change a behavior myself?
AI behavior close the ground isn’t optimal yet…
Altitude was set to 16000ft for all tests.
however his Flightpath was perpendicular to yours when he started the maneuver he is still trying to notch you, you would of lost radar returns at that point
No of course not. Scenarios are set up so that we are nose on at the start.
Most of the tactics except pure pursuit or grinder will involve a notch
Obviously those tactics are more adapted to 2vsx or 4vsx scenarios
So select pure pursuit and not random in a 1vsx scenario
Turning right after being nose on with the target and then recommitting ~15 seconds later is no useful notch. This is done before either craft is in missle range of the other.
This behavior happens on “pure pursuit”, too.I have uploaded a Tacview recording here: https://www.mediafire.com/file/73zc21lz13pdu09/TAPE0025.zip/file
(Uploading the .zip as an attachment fails, why is that?)
For the first few flights I turned off my radar till I got a missile warning to see if that would change the behavior.
You still can see that the AI craft always fails to increase its speed initially and sometimes does this short rightward crank and recommit before missiles are fired.
About failing to increase speed this might be because the fuel states at the start are rather low I think, but this still would be wrong behavior then.
Parameters for the Tacview-session are attached as an image, those were also used on the other tests where the weird initial crank happens. only changes there were using strategies like “pure pursuit”, “offensive pump and dump” and “grinder” and varying ranges between 35 and 50 nm (50 for testing against a F-14).[Also I feel like the AI is making a mistake turning cold to defend, as that gives me doing just a strong crank the ability to keep it on it’s back foot and then fire missiles again whenever it wants to turn back into me. This behavior occurs on all available AI Strategy settings.]
-
Flying the stock BARCAP TE the mig23s are pretty good at notching me. I lose lock, then once they’re dragging and I reacquire, my missiles track only 1 of the 2 bandits.
-
Altitude was set to 16000ft for all tests.
No of course not. Scenarios are set up so that we are nose on at the start.
Turning right after being nose on with the target and then recommitting ~15 seconds later is no useful notch. This is done before either craft is in missle range of the other.
This behavior happens on “pure pursuit”, too.oh I agree with you about the notch now. However you set it up for a Beam Deploy the AI did what was asked for by you… so still not a bug!
-
AI behavior close the ground isn’t optimal yet…
Hopefully, someone will do something about it in the future.
Hey Max, thanks a lot for mentioning the altitude here.
As opposed to 40nm, 16000ft where I feel I have the Ace AI thoroughly figured out (see attached Tacview above), as it turns cold where it doesn’t need to and fails to increase its speed, with an initial separation of 50nm at 30000ft I haven’t found a pattern with which I can reliably beat it yet and the fight feels a lot more like a challenge.
So while the assertion that AI behavior isn’t great in certain situations still stands, this at least gives me some more to think about and learn from. Many thanks.If you or anyone else with more insight wants to share the parameters where they feel the AI really is at its best, I would be happy to have them.
The one I liked here where it can hand my ass to me reasonably often I have attached as an image below.Cheers!
-
Hey Max, thanks a lot for mentioning the altitude here.
As opposed to 40nm, 16000ft where I feel I have the Ace AI thoroughly figured out (see attached Tacview above), as it turns cold where it doesn’t need to and fails to increase its speed, with an initial separation of 50nm at 30000ft I haven’t found a pattern with which I can reliably beat it yet and the fight feels a lot more like a challenge.
So while the assertion that AI behavior isn’t great in certain situations still stands, this at least gives me some more to think about and learn from. Many thanks.If you or anyone else with more insight wants to share the parameters where they feel the AI really is at its best, I would be happy to have them.
The one I liked here where it can hand my ass to me reasonably often I have attached as an image below.Cheers!
Once again
In 1 vs 1 the only tactic that would be efficient is pure intercept
The tactics are set up for 2vs x or 3/4 vs x scénarios
So don’t set random or any other tactics for a 1 vs 1 fight
If you fight vs a very weak opponent like A4 it would likely simply try to escape and not fight in BVr