Cheating AI.

[XarBat]

What is weird is that enemy choppers seem much more difficult to spot than ground vehicles.

[Bushmanni]

Here's the track. This time I turned around at the other end of the hill and flew back to my starting point. The enemy chopper followed my movements without a fail. I understand that responding to a lost enemy contact situation in a smart manner is hard for even the smartest of the humans as thats just the kind of situation belonging to the art side of war instead of science. But solving the situation through cheating just isn't fun.

 

In the FPS world Delta Force AI couldn't know where player was when there was no LOS so they would proceed to the last know location and shoot suppressive fire at that location once in a while. While it wasn't the smartest move it wasn't completely dumb and it was believable. In Rainbow six: Las Vegas the enemies know players location all the time after first contact. This makes using real tactics ie. flanking and surprising movement useless and breaks immersion big time. Cheating AI can be beat by shooting skills alone instead of proper tactics which is boring.

 

My proposed solution is to remove the cheat knowledge and give the AI two options set in mission editor to either fly to the last known position or return to mission. This way the mission creator can choose whether to allow the AI to pursue and possibly get ambushed or stick to the mission.

 

Here's a more detailed description of the proposed behavior:

 

Pursue after lost contact:

-Fly to the last know position

-Orbit for one and half turn and return to mission if nothing is found

-If target is found engage as usual

 

Do not pursue after lost contact:

-If target is closer than 6km fly to the last known position anyway, else do the following

-Fly to the direction of last know position for 10 seconds at combat speed

-Fly to the direction of last know position for 20 seconds at endurance speed

-After that orbit for one and half turn

-After orbiting return to mission

-After turning back check six at 20 second intervals until two minutes has passed from turning back

-After that check six at 40 second intervals until 5 minutes has passed since turning back

-If target is found engage as usual and forget about six checks

 

Times and distances should vary depending on the AI aircraft and the target type. F-15 has no need to check six if he's been flying away from helicopter for 2 minutes at cruise speed as it's far beyond sensors and weapons range already and has no change of catching up while helicopter should check six for maybe 30 minutes after contact as he can't be sure if and when the fighter is coming back. It would be realistic and more effective also if there would be slight random variation to the these numbers to make the AI more unpredictable in order to prevent exploiting rigid behavior.

AI_test_lost_contact_behavior.trk

Edited November 26, 2011 by Bushmanni
Added some improvements to the idea

[159th_Viper]

Here's the track.....

 

Ta for the track :)

 

OK, I've got you now.

 

A.I. will continue to track you, notwithstanding the fact that the A.I. LOS has been obstructed by, in this case, a large chunk of rock.

 

Reported to Bug factory with a polite request to the Devs that the matter be looked at as a matter of urgency. Will attempt to, within the confines of reasonableness, to keep you up to date as to how things develop. As stated before, no promises though.

[nomdeplume]

A.I. will continue to track you, notwithstanding the fact that the A.I. LOS has been obstructed by, in this case, a large chunk of rock.

Maybe a related thing - I've noticed RWR spikes (in A-10C) and LWS warnings (in BS2) persist for some time after you've put terrain between yourself and the enemy unit.

 

For example in Black Shark, if you put a Roland ADS and then play pop-up games with it, you'll get the warning as soon as you put your head above the hill and it gets LoS to you (as expected). If you then drop down behind the hill and keep resetting the LWS, you'll continue to receive new warnings for some time even though there's no way it should be able to light you up if you've got a hill between you and it. I haven't actually timed it, but it's long enough to be easily noticable.

 

I've been thinking maybe it's an optimisation to avoid having to continually check LoS, and instead only do it periodically, so the warnings persist until the next check cycle. But it does seem as if you get lit up the moment you exit cover, which implies those checks are done continuously.

 

So, maybe it's a general AI 'hack' in that the AI still "sees" targets for a short time if they break LoS, in order for it to not instantly forget all about the target. Easier than trying to implement a "search mode" whereby they try to 'guess' where a target may have gone - that'd require some pretty impressive machine logic. It does have some unfortunate side-effects, though.

[159th_Viper]

Maybe a related thing - I've noticed RWR spikes (in A-10C) and LWS warnings (in BS2) persist for some time after you've put terrain between yourself and the enemy unit.....

 

Ta for the heads-up. That was a known issue a ways back that was/should have been rectified IIRC?.

 

I'll look into that

[159th_Viper]

...I haven't actually timed it, but it's long enough to be easily noticable.....

 

OK - Had a look. There is a slight delay from breaking LOS until warnings cease (6-8 secs) which is quite normal. Otherwise all is fine - as soon as you break LOS the site does not paint you at all. Tested and verified with A10C and Ka-50 so no problems there.

 

Tested with Roland vs Ka-50 and A-10C vs SA-6. If you have tracks that show the behaviour you experienced kindly post and I'll have a further look.

[Luigi Gorgonzola]

OK - Had a look. There is a slight delay from breaking LOS until warnings cease (6-8 secs) which is quite normal. Otherwise all is fine - as soon as you break LOS the site does not paint you at all. Tested and verified with A10C and Ka-50 so no problems there.

 

Tested with Roland vs Ka-50 and A-10C vs SA-6. If you have tracks that show the behaviour you experienced kindly post and I'll have a further look.

 

Just out of curiousity, why is this behaviour considered normal? Sort of cool down period for the LWR?

[nomdeplume]

OK - Had a look. There is a slight delay from breaking LOS until warnings cease (6-8 secs) which is quite normal.

Probably just a misunderstanding on my part. The Ka-50 does continue to generate 'new' laser warnings for a few seconds after breaking the LoS, but it seems to only do so in the same quadrant as the original warning. I just did some more testing and spun the helo immediately after breaking the LoS and it didn't generate any additional warnings on either the 'new' or the 'old' quadrant during those few seconds.

 

So, it's a bit of an odd (to me) behaviour, but not what I thought was happening. When flying I'd always had my nose pointing at the threat when I ducked from it, so it was a bit annoying to keep getting 'new' warnings after breaking the line of sight. Guess I should just be a bit more patient before resetting the warning... it's only a few seconds, after all. :D

 

Otherwise all is fine - as soon as you break LOS the site does not paint you at all. Tested and verified with A10C and Ka-50 so no problems there.

When I really noticed it in Warthog was with an SA-10 at Gudauta and I was flying through the valley with the dam in it north of Senaki. The blip on the RWR continued for a very long time after breaking the line of sight. However in a quick attempt to replicate that scenario I can't get the SA-10 to appear on the RWR at all, so not sure what's going on there.

 

I tried with the SA-6 and it showed up fine, and disappeared pretty quickly after breaking its LoS. So, I think the issue is fixed. One weird thing I did note though, is that if I performed a roll after breaking the LoS but before the RWR warnings stopped, it would disappear as soon as it moved into the RWR's blind spot and then reappear once it moved back, despite their being no LoS.

 

By the way, is "quite normal" normal as in real life, or normal as in the sim? i.e. do they really continue to warn for a few seconds after they stop receiving the energy? It makes sense they would (better safe than sorry), but then I'm not sure why the threat is immediately removed from the RWR when you put it in the blind spot, and then returns afterwards, even though the energy is no longer being received. Likewise, why the Ka-50's system stops generating new alerts if you spin the chopper (shouldn't it keep generating 'new' alerts on the original quadrant?).