AI loop problem

[Quip]

I found this clip on YouTube, the title says it all.

[ame]

[/ame]

[myHelljumper]

Not new at all....

[razo+r]

you need to know, intelligent AI isn't really easy to programm and every game has his own AI issues and supidity

[myHelljumper]

Yeah for sure, but there it really affect the game experience...

[Jowen G. Bruère-Dawson]

IA is always a problem for all games. It's not new, and like said before, really hard to programm

[Frenetique]

Making an effective IA is hard, true. But the one from DCS could really benefit from an upgrade. As long as you go against ground target, it doesn't really matter, but flying IA is just dumb as a brick and not reliable or realistic to play with.

 

I am quite impressed by the fact that BMS has a better IA, even though the game is super old and is supported fans.

[razo+r]

Can you tell me at which skill level those AI got set?

 

makes also a difference

[myHelljumper]

They were average, but I don't think it change anything here.

[razo+r]

It will, try it with the maximum skill level

[Quip]

Not my video, but IMO the Skill setting only affects the AI's overall aggressiveness, not its "intelligence", i.e: it'll behave in the same way, but more or less aggressively.

 

Besides, you really don't want to have AI above "High" in multiplayer missions; the AI gets too aggressive and behaves very irrationally.

[mvsgas]

Is this kind of AI behavior really acceptable?

Yes

[MagicBra]

I wonder if it goes all the way with SAMs on the enemy AB.

Anyway I can add my REXP to the subject, setting the IA skill level doesn't change anything on its A/A missions but it seems to do on its A/G missions sometimes.

 

If only CAP AI could maintain a FAOR. it would be so simple to design missions. Just put a FAOR on the map assign groups to it and that's all.

It's a real pain to design effective air missions with current AI because this kind of behavior happens a lot.

 

If someone has an effective way to make effective CAPs and defending an airspace without them going SWEEP i'll be glad to know the trick.

[Quip]

The editor lets you control many parameters which in fact act as limiters; they're there to bypass the AI's natural behavior.

It could be argued most of these are there simply because the game's AI isn't predictable or trustworthy.

 

The problem is that the results of using these limiters (Advanced actions) itself is unpredictable and untrustworthy. It's more likely than not that what the game ends up doing is rather not what you wanted. And any one outcome can only be predictable in an isolated situation. Something that works in a test may or may not end up working in a mission once you add more units and dependencies.

[Grimes]

At least from the video it looks like a somewhat rare occurrence of the AI getting stuck on a logic loop and is never able to correctly engage. I've seen behavior like that with engaging ground targets, never aircraft.

 

If you have a short track featuring that behavior it would be appreciated. I can't report a bug with a video, partially because it appears to have been recorded with the use of a potato, but mostly because tracks contain data that the developers can use to see precisely what is going on with the AI.

[zxarkov]

I hate when people say things like oh its too hard to program AI so why bother with it just play multiplayer?

 

DCS needs a halfway decent AI overhaul before we need more planes and other features. I know it is possible, look at IL-2 or Falcon BMS..much older games. It effects game play in a major way! At least some simple damage modeling on AI..

Edited March 22, 2016 by zxarkov

[Quip]

At least from the video it looks like a somewhat rare occurrence of the AI getting stuck on a logic loop and is never able to correctly engage. I've seen behavior like that with engaging ground targets, never aircraft.

 

If you have a short track featuring that behavior it would be appreciated. I can't report a bug with a video, partially because it appears to have been recorded with the use of a potato, but mostly because tracks contain data that the developers can use to see precisely what is going on with the AI.

 

As I said, I didn't create that. But why not ask the poster on YT?

 

Edit: But it's not the looping behavior which is the most important lesson from that video, IMHO.

Rather: Why is the MiG23 RTB-ing while under attack? Shouldn't it defend itself?

 

On the same note: If I tell a pilot (IRL) to "attack only using AA missiles", would he not use the guns to defend himself? Given the restrictions "attack using", the AI in DCS will not use the Guns even for defense. As such, the AI is unaware of whether it's in an attack or defense mode.

Edited March 23, 2016 by Quip

[Quip]

I hate when people say things like oh its too hard to program AI so why bother with it just play multiplayer?

 

DCS needs a halfway decent AI overhaul before we need more planes and other features. I know it is possible, look at IL-2 or Falcon BMS..much older games. It effects game play in a major way! At least some simple damage modeling on AI..

 

Agreed. I've been flying DCS in Multiplayer mode exclusively for the last 3 years, and I'd go as far as saying that AI is even more important in MP than Offline missions, simply for the complexities you run into with several humans misbehaving at once in the same world.

But even further: with a good AI it would be simple® to create interesting MP missions (see BMS). Today, with DCS, creating vast, complex missions with varying types of targets and a dynamic FLOT is impossible.

[Grimes]

Edit: But it's not the looping behavior which is the most important lesson from that video, IMHO.

Rather: Why is the MiG23 RTB-ing while under attack? Shouldn't it defend itself?

 

I don't know what is going on before the Mig gets hit. It looks vaguely like a bug I've seen before but its hard to say just based on video evidence. Once the aircraft gets hit to the extent it was the behavior to RTB is normal and expected. Basically the AI will try to be defensive or fight it out, but once they take a critical hit to the point where enough damage has been done to render them combat ineffective but still flyable, the AI will try and RTB. When they get into the RTB mode like that the AI's "tasking" pretty much turns off and they only care about returning to whichever base they are going to. As you can see in the video the Mig has quite a smoke trail following him, so he is definitely in that RTB mode.

[Quip]

[i had typed a long answer that the forum lost by logging me out as I was typing!! This is an abbreviated version.]

 

Grimes, let's say the behavior is correct: to RTB when you're damaged (I argue it's not unless you use an escape window, which the AI knows nothing about, but I digress)

 

Note: I'm addressing you Grimes, but of course you're not responsible for the situation nor for fixing it.

So the AI has put the MiG is "I'm damaged/RTB mode", ie bugging out, ie I'm dead. So why isn't the AI applying the same rules to the F5? Why doesn't the F5 get the same signal and stops the attack?

 

Further... the MiG RTB's, bringing him into red territory. It's not in that video, but I know I've seen the following: the chasing AC (the F5) will chase the fleeing AC (the MiG) even if it brings the chasing AC into defended territory (either surface to air- or air to air-defenses). There seems to be no understanding of "this is a bad tactical situation for me" in the AI.

 

So the fact that the AI can't dog-fight worth a hot potato (to use your expression), isn't a problem: it goes both ways and most of the time it'll sort itself out.

But what is a problem is the fact that the AI isn't anywhere near being predictable or trustworthy. And that (as proven above) it plays by different rules in the same setting.

[Tomhatter]

AI is a huge issue for sure. I'm trying to setup a multiple SEAD Op with the Su-25T and I have experimented with many settings but the AI pilots continue crashing in the water or the mountains. They're not even being locked up, just painted. They refuse to follow the altitude settings or stay on task. I'm not sure what the problem is with changing the altitude setting from Mean Sea Level to Above Ground Level but when I tried to change from MSL to AGL, that brought about a whole new set of issues-AI won't even follow WPoints. Any ideas?

[Grimes]

Tomhatter, change the "Reaction To Threat" option for the AI's task. One of the options should make it so the AI only go evasive when fired upon. The difference between MSL and AGL is with AGL the AI will try to stay at the specified altitude above ground level. So they will rise and climb more or less with the terrain. It has the best results if the altitude is lower than 3000m or so, otherwise its kind of pointless and you should use MSL. I'm not aware of any WP issues with regard to using AGL, I'd need a track file.

 

Quip, AI chasing AI like that is an age old problem of an AI being told "destroy target X" and not having additional rules to override that behavior. Well there are additional rules but those rules are more to do with self preservation as the issues arise. The F-5 continuing to try and attack the fleeing unit is an extension of that behavior.

 

Also the 'potato' comment was toward the quality of the video being at such a low resolution.

[Quip]

So if the bug is "age old", it proves it's known. So:

Fix it.

 

;)

 

(Hadn't noted the video was 360 only)