Just some thoughts about AI

[D-Scythe]

Well, got a new patch, so I guess somebody should write what they think about the AI in V1.11/12. I realize that probably very little is going to be done on the AI for V1.2, but I'm going to post this here anyway ;)

 

Much of the stuff has already been pointed out by others, but there some new things I'd like to see too.

 

A/A Combat

 

There are still problems with AI fighters chasing their targets all across the map, whether it be a player or another AI aircraft. I set up a mission where a MiG-29 would enter a F/A-18's reaction range, then promptly turn around and run. Here's what happened:

 

The initial baiting. Reaction range was around 100 km.

Final part of the chase. Note the position of the Hornet relative to its starting position

 

There are several ways to minimize this. The easiest and most effective way I would think is to adjust the speed the AI flies at. For example, during a pursuit, the AI just lights the burner and runs, as you can see above. This is the main problem with the AI right now, and not only leads to random pursuits, but also heavy fuel consumption.

 

So, to eliminate two problems at once, I'd recommend tweaking when the AI can use afterburner. The only times, IMO, the AI should use afterburner is during take-off, during BVR engagements, dogfighting or when running away. Currently, the AI's AB usage is all over the place.

 

Fox example, look at these two screens:

 

As seen, the AI speeds up (a little too soon) then actually slows down just prior to firing that first BVR missile. That makes no sense - they waste all that "smash." It makes sense to switch these "slow" and "fast" parts around - i.e. the speed in the first screen above should be at 800 kmph, and the one after 1293 kmph.

 

Another example of how wierd it is right now:

 

The F/A-18C above is bingo fuel (I set it up that way) and is trying to get away...at 400 kmph? Incidentally, the Su-33 is going supersonic. Again, switch the two speeds - the F/A-18C should be the one supersonic in afterburner, and the Su-33 in mil-power, tops.

 

Another way to limit this problem would be to reduce the reaction ranges of aircraft on CAP, Fighter Sweep or Intercept to like 30 nm (55 km), 35 nm (64 km), and 40 nm (73 km). Just for comparison, the reaction range of AI in F4 is no more than 25 nm (46 km) normally.

 

The desired effect is that since the aircraft being chased is in afterburner, it would eventually be extended outside the reaction range of its pursuers, whom will hopefully find something else to run down or continue on with its waypoints.

 

~~~~

 

Another realism-killing problem is the perma-padlock the AI always has on the player and each other, usually resulting in instant reaction time. My suggestion is to at least incorporate a very simplified search function (similar to what ARH missiles have now when mad-dogged), and at the very least, just put in a delay time in between the various things the AI does to slow it down.

 

~~~~

 

Not so important...active radar capable platforms should engage multiple targets. It's sort of pointless when they are using them as SARH missiles.

 

Missile Evasion

 

I would like to see the V1.00 missile evasion technique incorporated in some way back into LOMAC (the missile pump, where the target just runs away). It not only gaurantees the survival of the target at long-ranges, but also is just more practical in some cases. Look at the screens below.

 

 

By just only beaming and diving, the targets never leave the engagement zone of the SAM, for example. So unless the SAM is destroyed, they're going to be eventually shot down as they get closer and closer by "stepping" through the threat ring.

 

However, if the target pumps the SAM, it turns 180 degrees and flies out of the threat ring. Problem solved.

 

I'm not saying that we should replace the current evasion technique, but just add a new one to the AI's repertoire. There should be a random chance or probability that the AI will execute the pump - obviously, it should happen more often at higher AI skill levels. It would also be nice if the player's AI wingmen performs this maneuver, as it increases their chances of survival.

 

And it should always happen against SAMs.

 

~~~~

 

SEAD aircraft also have the tendency to chase objects around the map, only this time it's SAMs instead of aircraft. Again, some realistic reaction range should be incorporated (like 60 km), but in this this case, the aircraft should fly back to its "attack" waypoint and recommence searching for targets there. That way, SEAD jets won't jump from one SAM to the next (i.e. won't attack one SAM 60 km away, then another 60 km away from the first SAM, etc.).

 

~~~~

 

Other cosmetic changes:

- Taxi speed should be increased

- Wingmen should be able to land with leader in formation

- Aircraft should be able to take-off/land when others are taxiing (should be the mission builder’s job to deconflict, not LOMAC’s)

 

What do you guys think?

[Yellonet]

Good post, and yes, the AI is mind-numbingly stupid.

And wingmen can land with you in formation... although it looks a bit weird.

[S77th-GOYA]

What do you guys think?

 

I think the AI is maddeningly stupid. A complete overhaul is needed.

 

Here's a recent example:

 

Tornados tasked to drop GBUs on a command center. First drop, a-ok, second drop, the Tornados always dropped down and ran into their own bomb right after drop.

 

Torando doing SEAD. Fires at a KUB and its ALARM gets shot down. Does he fire a second? Does he pickle two off? Of course not. He flies low and slow toward the SAM. Maybe he's trying for a guns run. It doesn't matter that he's got three more ALARMs on his pylons.

 

Examples of idiot AI are really too numerous to mention.

[MBot]

Good post.

 

A little trick for the reaction ranges of aircraft. The 'escort' task has the shortest reaction range of all tasks ( 15 or 20 nm IIRC ) and I use this task for almost all AI Air-Air missions.

[D-Scythe]

Good post.

 

A little trick for the reaction ranges of aircraft. The 'escort' task has the shortest reaction range of all tasks ( 15 or 20 nm IIRC ) and I use this task for almost all AI Air-Air missions.

 

I do too, but for fighters with longer ranged missiles, like the MiG-31 and Su-27 Flanker series, this is too low (and the normal reaction range is too high).

 

Escorting planes will also try to run you down on full burner ;)

[Guest ThomasDWeiss]

Good post, and yes, the AI is mind-numbingly stupid.

And wingmen can land with you in formation... although it looks a bit weird.

 

I beg to differ ;)

 

I seldom employ the F/A-18 because it tends to turn away from combat, I stick to a few fighters , whose behavior I know like the Su-17, Su-24 and Su-25 for ground attack - and MiG-23/29 for air to air work.

 

You have to position it in a way that it will react to an event in the way you want.

 

I am finishing updating a pack that I made with 1.02, then remade for 1.1 and now for 1.1.1 and it was for me, like a tutorial on how the AI evolved.

 

It is not dumb - it is stubborn.

[D-Scythe]

Caretaker put it best: "The biggest obstacle for me is that the AI is too autonomous, but not very clever at that. "

 

If we can stop the AI from just randomly chasing things across the map, then it would at least resolve some issues (survivability, controllability, etc.).

[Guest ThomasDWeiss]

That's it: "The biggest obstacle for me is that the AI is too autonomous, but not very clever at that. "

 

perfect analysis of the problem.