How on earth do AI planes detect my Sidewinder launch?

[PoorOldSpike]

What drives me crazy is the fact that when I fire a Sidewinder at an AI plane, it instantly begins popping flares even if I was literally miles away when I fired!

How the heck did the enemy plane know I'd fired?

It happens no matter what type of enemy plane it is (transport/fighter-bomber etc), and even his helis do it, sheesh!

Surely it can't be realistic?

[silverdevil]

@aimbotdefender... seriously though i see this too. i have tried to get them to prematurely drop flares and there seems to be a delay for them to do again. thats when i fire. honestly they seem to know when they are in your grasp because many times, in guns-only ACM, they pop flares too. so maybe not the detection of a missile firing.

Edited September 13, 2020 by silverdevil

[PoorOldSpike]

@aimbotdefender... seriously though i see this too. i have tried to get them to prematurely drop flares and there seems to be a delay for them to do again. thats when i fire. honestly they seem to know when they are in your grasp because many times, in guns-only ACM, they pop flares too. so maybe not the detection of a missile firing.

 

 

Good point, i'll try switching off my radar in case radar sweeps wake them up and makes them stand by with their fingers on the flares/chaff button, and report back to this thread later.

Edited September 13, 2020 by PoorOldSpike

[kreisch]

Because this is the way the AI currently works - simple as that.

Try engaging an "ace ai during the night, without radar, only using your NVG to engage with a heatseeker - they will use flares the moment you launch, even if they have no missile warning system at all.

 

Gesendet von meinem VOG-L29 mit Tapatalk

[Mr. Big.Biggs]

Cmon guys, it’s a computer program. When x=1 the flares pop and the ai dodge its that simple. I would think the only other way would be to either put a random generator to say “yes ai will react” or “no, they will ignore” And its an odds thing or perhaps put a timer before they react after launch is generated but they still “know” exactly what to do to dodge.

The computer and everything on it knows exactly when you launch, what you launch and under what particular parameters. You can’t fool it, it only pretends to be beaten. AI is pretend after all, they have no real abilities to think...

[stormrider]

Cmon guys, it’s a computer program. When x=1 the flares pop and the ai dodge its that simple. I would think the only other way would be to either put a random generator to say “yes ai will react” or “no, they will ignore” And its an odds thing or perhaps put a timer before they react after launch is generated but they still “know” exactly what to do to dodge.

The computer and everything on it knows exactly when you launch, what you launch and under what particular parameters. You can’t fool it, it only pretends to be beaten. AI is pretend after all, they have no real abilities to think...

 

 

No really. The computer knows everything, yes, but the entities in it can be configured to have its own senses and respond to them alone, nothing new here. The problem is DCS AI. It's way too simple and predictable. It needs more complexity. Let's take for example Assetto Corsa AI. Stefano literally taught his AI how to race a car like almost human. The AI in there follows the same rules of physics a any human player, no cheats, clairvoyant or artificial blindness and each AI on track is an individual entity. It took quite some time before it was perfected, but its achievable.

 

x not always equals 1; it can have ramifications depending on the perception of the entity times a random variable.

[Doum76]

What drives me crazy is the fact that when I fire a Sidewinder at an AI plane, it instantly begins popping flares even if I was literally miles away when I fired!

How the heck did the enemy plane know I'd fired?

It happens no matter what type of enemy plane it is (transport/fighter-bomber etc), and even his helis do it, sheesh!

Surely it can't be realistic?

 

 

This bugs you, but the fact they constantly use after burners and never run out of fuel never bugged you? :) That what made me stop fighting AI before my break last night, when i made a test because i got a little fed up to see them keeping going vertical nonstop with burners on infinite and gave 10% fuel on the AI and the ran away from me full burners for over 50 nm, with 10% fuel remaining....

[PoorOldSpike]

At the moment I'm finding the best way to overcome the AI's flare-popping frenzy is to give myself unlimited weapons so I can fire repeated volleys of Sidewinders at the AI planes, his flares/chaff will decoy some but not all, hehe.

Sure, to Alice watching from the ground the skies will look like a cross between a fireworks display and a nuthouse but hey that's war..:)

 

"Oh I do so wish something would make sense for a change"- Alice in Wonderland

[Headwarp]

I mean.. I see your point but at the same time.. even with AI popping flares the only times my heatseekers tend to miss AI are when I fire at a at a bad angle or too far away/too close. As long as I'm firing from a good angle and proper range they tend to hit regardless of flares.. AI doesn't maneuver hard enough to throw them off.

[PoorOldSpike]

I ran some tests today in which I'd switched off my radar so the AI pilots wouldn't know I was buzzing around, and I also made them all 'Average' skill level, but it had no effect, they still popped flares the instant I fired Sidewinders from any angle and any range, it's highly unrealistic and doesn't do the reputation of DCS and the devs any good at all, grrr..

[Raisuli]

This bugs you, but the fact they constantly use after burners and never run out of fuel never bugged you? :) That what made me stop fighting AI before my break last night, when i made a test because i got a little fed up to see them keeping going vertical nonstop with burners on infinite and gave 10% fuel on the AI and the ran away from me full burners for over 50 nm, with 10% fuel remaining....

 

 

I've run them out of fuel a few times. Always considered it a failure because they should be dead before then.

 

 

The F/A-18 doing vertical rolls with an F-16 until I finally clipped the wing and ran it out of gas was entertaining. I know the Hornet much better than the Viper, and I can tell you there's no way it had that much energy. Same deal fighting an F-5 in an F-18. They put energy drinks in the tanks or something... :thumbup:

 

 

 

Just the nature of the beast.

[Nealius]

The AI are poorly programmed. Simple as that. They're omniscient and know exactly when you've fired a missile when a human player in the same scenario wouldn't know, they can detect you visually in the dark, their radars overperform, and they're known to defy physics. I can't even fly formation turns with them because they slew laterally in 3D space to get back on flightpath. The AI is just bad.

[probad]

its scripted, the ai aircraft will react and pop flares autonomously even after its pilot has ejected

 

the ai can pull 2g over the top with negative aoa. nuff said.

[pete_auau]

thats why i let my ai wing man to deal with them, what they can do, my wing man knows what they are planning to do in advanced :)

[PoorOldSpike]

Hmm..so apparently I paid a whopping 62 GB pounds (75 US dollars) for DCS F/A-18 Hornet and find it contains a number of highly unrealistic features that much cheaper flight sims to their credit don't have, grrrr..

[Ramsay]

Hmm..so apparently I paid a whopping 62 GB pounds (75 US dollars) for DCS F/A-18 Hornet and find it contains a number of highly unrealistic features that much cheaper flight sims to their credit don't have, grrrr..

Yes, DCS World is imperfect and it's why I always advise players to explore the sim with the free Su-25T and TF-51D first.

 

The AI planes and logic you complain of are part of the base game, rather than a specific issue with the Hornet.

 

IIRC ED recently said they plan (hope ?) to improve the logic so "AI pilots" don't have the same 360° vision they have today but AFAIK there wasn't a timeline/road map when such a change might happen.

[BuzzU]

AI have been a problem no matter what the sim. Combat flight sim, auto racing, or 1st person shooters. AI are always a problem.

 

If those who develop the AI will just ask the questions..........would I do that? or could I do that? If the answer is no. Then don't let the AI do it either.

[Nealius]

Hmm..so apparently I paid a whopping 62 GB pounds (75 US dollars) for DCS F/A-18 Hornet and find it contains a number of highly unrealistic features that much cheaper flight sims to their credit don't have, grrrr..

 

The 62GBP were for the Hornet specifically, which doesn't have unrealistic features. The unrealistic "features" are issues with the 0GBP base game.

[BIGNEWY]

AI will defend against missiles within 10nm or half the missile range.

 

We do have to compromise in some places with AI so we don't have to require everyone to have a NASA super computer to play DCS.

 

Thanks

[captain_dalan]

AI will defend against missiles within 10nm or half the missile range.

 

We do have to compromise in some places with AI so we don't have to require everyone to have a NASA super computer to play DCS.

 

Thanks

 

Or? Does that mean which ever is bigger of those two?

And by missile range, you mean some fixed value or a value derived from the DLZ?

[Nealius]

AI will defend against missiles within 10nm or half the missile range.

 

We do have to compromise in some places with AI so we don't have to require everyone to have a NASA super computer to play DCS.

 

Thanks

 

Many combat sims over the years have better AI defense reactions without requiring a NASA supercomputer, and the only reason DCS would require that is due to "innovation inflation" where we get so many performance-hitting visual upgrades and such to the game while it's still bound to a single CPU core while the GPU does hardly any work.

[some1]

Or? Does that mean which ever is bigger of those two?

And by missile range, you mean some fixed value or a value derived from the DLZ?

 

Fixed value, whichever is bigger. Just observe how ridiculous AI reacts to Aim-54s.

 

The AI has mostly the same routines as 15 years ago, when computers had fraction of power of today's machines. So the NASA computer excuse is pretty weak.

[Fri13]

AI will defend against missiles within 10nm or half the missile range.

 

We do have to compromise in some places with AI so we don't have to require everyone to have a NASA super computer to play DCS.

 

Thanks

 

Simulation is required for the AI to avoid processing unnecessary actions, but the problem is that DCS doesn't do that, it simply has the simplified system that is run constantly.

 

In the old simulators, when there is no player watching/observing, then there are no 3D models, there is no checking LOS, there is no timers, scripts or anything performed. It is unnecessary and waste of precious processing time.

 

It is very very simple dice rolling, a unit on map is nothing else than a table of features and capabilities, like a RPG game character stats card. Then when the another unit challenges (engage) the other unit, the check is made from the table against corresponding feature. Example "awareness" is done first. If no awareness is happening, no engagement. If unit is aware of the another unit, then spotting is required. When spotting happens, the unit with YES value starts first. Check is made for the range that can spotting unit engage the spotted unit. At this moment the spotted unit can still be unaware of the spotting unit.

When the spotting unit performs a check is it in the range, it needs to make checking for propabilities for success as well is it acquired to engage. Unit can avoid engagement by hiding, or try to sneak around. If the unit mission is to get stealthy inside a castle, it avoids all the confrontation as possible. If the mission has a time limit, unit does check how much it can route or delay movement or engagement.

The spotting unit is that does all the decisions as it is aware of something. The spotted unit can be unaware and minding own business, performing own task like "Patrol the waypoints A to B".

 

If a unit decided to engage, it is in the range, then dice is rolled with the engagement. The unit can have a change to roll a check for attack every X turn, where a turn can be like a 5 seconds.

If the unit to be engaged becomes aware of the engagement, it can roll a check against weapon can it counter it or not, or does it do something else like try to flee, dodge or something.

 

The winner can be quickly and easily decided just with propabilities, like "In range, Solid Lock, Position Rear hemisphere, Target unaware = Kill".

 

On the strategic map the engagement has happened in a milliseconds. The winner has been decided and simulation of engagement can be ended.

Lost unit is removed from the map and player is not any wiser that what really happened.

 

The critical component that DCS lacks is the actual simulation for checking. Every unit calls a checking X times a second that what units are in it's detection range. Every unit doing that constantly against all units.

There is already the Green and Red state, but AI will know that enemy is behind the mountain and in range.

So when the enemy pop-up behind the ridge, AI starts turning cannons immediately at it, if not already turned.

 

In a such scenarios the only thing that should be added is that every unit receives a "radar". Just like a F-18 has radar beam modeled that sweeps sky etc, same needs to be on every single unit. In a MBT that is 3-4 such ones, driver, gunner, commander and loader. Each is given a scan zones and FOV periodic checking. Like driver sees only forward and scans narrow zone. Loader has own small zone to check sometimes if not searching.

Gunner has very narrow but long range FOV and can rotate it around with turret.

Commander has wider and so on....

 

Unit is given a multiple levels of awareness and states. "Air Threat" is different from "long range combat". The various elements has different tasks and capabilities that change the unit performance and functions. Like if unit is not aware of air threat, no one is looking up in the sky. In close-combat mode where enemy infantry is near by and RPG risks exists etc, no one looks up in the air.

 

A more complex modeling and simulation is done only when a player is watching the scene. Otherwise it is just simulated comparing a card tables and rolled dice and be done with it. Lots of checks does go "can't see" or "uncertain hit probability" and so on combat becomes more realistic as it can last long times.

Add fear and moral factors and units would move in cover, retreat etc regardless their commands.

 

Like no commander is going to command to drive their MBT over a hill if there is a MBT platoon waiting. An ambush is highest risk that is to be avoided. Instead the commander informs the situation and calls air support or try to find an alternative route.

No matter how much is someone commanding "move there". No matter if they would need to move through that path to support own troops behind the ambush.

 

This risk evaluation and checking is missing in DCS totally. The units will march to their deaths no matter what. A MBT platoon drives toward their death while enemy MBT platoon snipes them one by one in 1 km range. On the first hit the advancing MBT platoon leader would call retreat to cover when ambushed, as they have no idea who and from where took a shot. Find the enemy, analyze the positions and plan tactics etc.

 

And one of the problems for this is that AI has no flexibility to maneuver away from the waypoint path. It is strictly denied from moving away from it and to change time of arrival.

The AI lacks basic military tactics, like send a scout first ahead the main force . Form a proper formation to counter enemy formation strength by using its weakness.

If enemy has a arrows, spread units to large open spaces formation.

If enemy has a line defense, form a spearhead.

If enemy has split positions, flank and engage only other from opposite side.

 

Another problem in DCS is the exact units location. The fog of war is missing.

The idea is that player knows all the time exactly where every single unit is and their status.

That is not so in real military. All is based to assumptions and expectations.

It becomes a dice rolling even in real world. 3:1 ratio for attack in forest, 6-9:1 to attack in urban. A unit is expected to able move X kilometres per hour on road, a build a A2 level defensive position in X hours. A MBT platoon is capable to perform it's attack duties with a 1 of 4 MBT disabled, if it becomes 2.5 of 4 that are combat ready then that MBT platoon doesn't anymore go anywhere before back to at least 3 of 4 status.

2 men killed from 7 men squad renders them ineffective, a secondary reserve unit.

In combat one commands unit to move from A to B and expects them to be at B in X hours or minutes. It doesn't matter where they are or what they do, only thing that matters is that they are time on target. Their last position was known and their expected position is known and time of arrival, and it is up to unit perform their expectations to complete that order. If status changes that they can't reach a position in time, they update their status to command.

 

And that is major feature that is missing as dog of war and command structure allows to start simulating units quickly and effectively without wasting processing resources to run 3D modeling and all checking all the time.

 

It is cheated in big scale, if a AI knows that on unit Alpha route from A to B has nowhere near enemy troops, then it is never required to start performing any checking for anything. It is just "Unit Alpha moving". On map there shouldn't be anything about unit exact position or status, just a command from Alpha to move to B from A with time table and commanded route. If a player wants to know where they really are, they need to get in that location to see it.

 

And this is where a major problem does appear, a free camera. In a combat simulator there shouldn't be possible to have a free camera jumping capability to anywhere than what really has confirmed location.

As if unit location is unknown, no one should be able jump there in free camera or see them in free camera.

 

Such a expectation for that capability requires that everything is known all the time, that their rendering and all is done continually and it eats a lot of processing resources, that would otherwise be just "rollíng a dice".

 

Many thinks that Falcon game series dynamic campaign is amazing, but they don't really know how simple it is and how well it is cheated all the time, that renders it laughable as DCS does a lot more as it doesn't cheat in clever manner like those other dynamic campaigns does.

 

Simulation means that something is not done, it is just checked what would happen if expected things works as should.

At some point it is required to have a uncertain randomness, that allows to simulate complex large scale scenarios as nothing is required to run with tiniest details, but it is done in large scale batch processing.

 

A full combat can be executed and simulated in matter of second, then it is just playing the end result as a movie for the viewer. So while makers know what is going to happen, for the first time viewer the whole thing is beautiful experience, even when story end is well known.

 

Problems appears if a viewer is given a change to intervene action middle of it. As on that moment simulated end game doesn't happen and it needs to be done again from that point with modified new board.

 

There is no need for a NASA level computers when players gets cheated properly, by simply really simulating only the necessary.

[captain_dalan]

Fixed value, whichever is bigger. Just observe how ridiculous AI reacts to Aim-54s.

 

The AI has mostly the same routines as 15 years ago, when computers had fraction of power of today's machines. So the NASA computer excuse is pretty weak.

 

Yeah, i mean i did test just that a while ago.

 

 

I was asking this because, while the AI may "actively" engage in defensive maneuvers at certain range from the missile, it starts dispensing CM immediately after being shot at.

 

Wouldn't it be not all that complex. but much more authentic if the AI's counter measure response be tied to it's defensive maneuvers? Nothing Nobel Prize level, just simple IF branches:

1. Fox 1 medium or long, AI dispenses CM when locked (for role playing reasons, even if they don't work like that in DCS) or when fired upon.

2. Fox 3 medium or long, AI dispenses CM when missile goes active. Shouldn't be an issue, as the AI is completely aware of the missile and it's parameters at any given time.

3. Fox 2 medium, the AI doesn't defend or dispense CM until within visual range for that missile and-or launch

4. Fox 1, 2 and 3 short, AI defends in full measure right away.

[Thor_H]

They have the fighter pilot instinct, or "gut feeling" that you need in this business.