How realistic is the AI in a fight?

[DCS FIGHTER PILOT]

Simply curious because I tend not to like multiplayer.

[Mars Exulte]

In general, not very. It's supposed to receive an overhaul soonTM

[DCS FIGHTER PILOT]

Do we know how soon?

[Raven (Elysian Angel)]

AI in DCS uses Simple Flight Model, and thus can pull manoeuvres that are impossible for a human-controlled Professional Flight Model-module.

Also, the amount of available manoeuvres they have in their "skill-database" is quite low. The latter will be expanded at some point in the future (no official ETA).

We're told that AI using PFMs would be too taxing for people's computers, and therefor only SFMs are used.

[skouras]

AI in DCS uses Simple Flight Model, and thus can pull manoeuvres that are impossible for a human-controlled Professional Flight Model-module.

Also, the amount of available manoeuvres they have in their "skill-database" is quite low. The latter will be expanded at some point in the future (no official ETA).

We're told that AI using PFMs would be too taxing for people's computers, and therefor only SFMs are used.

that..

[DCS FIGHTER PILOT]

How will the new AI improvement be an improvement?

 

Recently people have said that the AI is getting an update soon. This is good for me because I generally like to stay away from MP. On occasion though I will join a game. Besides the point, how will the AI be overhauled? Will it be more realistic?

[BlacleyCole]

I haven’t heard about an ai improvement but I think your missing out on a Lott by not findding a squad of like minded flyers. I’ve seen strick by the books squads that you can. Learn a lot from to squads. That allow you to progress at your own pace. Some are aiircraafr restrictive others arre open to any dcs aircraft and get too getter for fun. The assistance you get from such squads usually relate to what you put in the Moore you put in. The more you get. Out exponentially.

 

I was in the military for 20. Years and when I retired except for a few friends that retired give or take a few years before or after me I felt lost.. I had this suppport system, and the comrade of my felllow service members before retiring then. I had thee va system and Tri care after Tri care. In my area wanted to send me to doctors in three local area I was lost with the va everything was in one location they covered transportation costs provided a. Room for the night if required or desired. I stumbled in to aa cowboy shooting sport. Game and after a while found a great bunch people that made up for my lost I had from. Retiring. Then. Last year I had to give up competitive shooting do to health issues. Then last November I stumbled in to dcs. I’ve be flying FlightSim apps since the mid 80’s and never joined a virtual flying organization. But I flew dcs for a few months and decided to try the map side. I. Really felt I found something in my life that was missing. The first couple didn’t work out as. I. Hoped between picking the hardest aircraft to fly and some technical isssuues with my equipment the firstlet meek goo initially for. Lack ooff progresssiion same. With the second plus. They felt tThat I wasn’t a goood t since I was. Asking questions about other squads. Then I found a unit that let you put in what you wanted and would help you when you asked for it. I really like this squad and I’m sstill a member but it’s been a month or more since I was aablle to fly with them because of a combination Orr hheeaalth issues and tearing my pit down to raise it up to make it easier. Toggetin and out of. I was recruited in this other squad and really found a home there. I. Let them know up front I was having both tech and physical problems it didn’t matter.. as. I’m. Learning the new squad I saw they could use some of the information I had gathered in my dcs career I. Started assisting them with. This and that and was quickly. premoted twice once in rank and secondly in position. The squad has transformed allot ssiince my recruitment and is turning into a real military organization.

 

So ass you can see the multiplayer side has a lot more to offer than just a group of flyers flying together. So take another look into it find a squad with like minded ideas. And give thhem a try.

[BlacleyCole]

Dcs fighter pilot check my post in your other topic where you we’re asking about something and. Mentioned not liking my.

[Raven (Elysian Angel)]

I already told you in the other thread you created on this topic.

[RaisedByWolves]

Do we know how soon?

 

 

2 weeks.

[GunnyCook]

2 weeks.

LOL

[DCS FIGHTER PILOT]

Well thanks everyone. I probably will try mp a bit more now but I should probably get some things first, like a new headset. Any suggestions on some good hardware? How do the comms work in MP? Is it through the radio in the aircraft or is it some type of chat room?

[TheBamse]

There's no general way of communicating in MP but very commonly used is SRS, Simple Radio Standalone by Ciribob and MorpheusXAUT found here: https://forums.eagle.ru/showthread.php?t=169387

In the end it's up to the server owner so check the briefing or server description on whatever server you're connecting to for details. Some use TS3, some use Discord, some use SRS ... or a combination.

[Cape]

AI in DCS uses Simple Flight Model, and thus can pull manoeuvres that are impossible for a human-controlled Professional Flight Model-module.

Also, the amount of available manoeuvres they have in their "skill-database" is quite low. The latter will be expanded at some point in the future (no official ETA).

We're told that AI using PFMs would be too taxing for people's computers, and therefor only SFMs are used.

 

Or ED could improve their engine and take advantage of multicore processors, using a couple of cores for the simulated AI in PFM. Current AI kills the immersion

[jef32]

The AI flying realism in DCS ?

 

 

[Gierasimov]

AI is not perfect but not totally bad...

 

While I agree that AI A/A needs improvement, and having in mind that this is something that ED already said was recognized and work in progress, I still think it is not as bad as it could have been.

 

It does the job of killing you if you are under-performing, and it reacts defensively if you gain an advantage in the dogfight. I have seen worse implementations of AI A/A combat.

[Exorcet]

While I agree that AI A/A needs improvement, and having in mind that this is something that ED already said was recognized and work in progress, I still think it is not as bad as it could have been.

 

It does the job of killing you if you are under-performing, and it reacts defensively if you gain an advantage in the dogfight. I have seen worse implementations of AI A/A combat.

The AI can do somethings well and it is a threat, but the biggest problem with it for me is that it's not convincing. People like to point out the SFM but that's not really the main issue, it's AI tactics and limitations. The AI knows your airspeed exactly when you're dogfighting it and instantly reacts based on this knowledge. It will also never lose sight of you once it engages which leads to things like instantly locking you despite diving into clouds or firing a missile on you the moment it turns around because it doesn't have to search for a target.

 

 

The AI works as it is, to varying degrees, but whenever we do get improved AI logic that addresses the current issues, DCS will improve by bounds.

[p1t1o]

AI for FPS is different to AI for tic-tac-toe is different to AI for combat aircraft.

 

I can imagine its hard to write a good AI combat pilot.

 

Its all very well and good saying that the DCS AI is rubbish, but what if it is the best-in-class?

 

Its the same result but I think its an important distinction, and I for one, dont know what the state of the art is, exactly.

 

 

Has anyone seen a better air combat AI?

[kengou]

Has anyone seen a better air combat AI?

I've played other sims that are not allowed to be mentioned on this forum that feature better AI, yes. One popular mod in particular.

[The Falcon]

Maybe with more cores etc .. but the problem is not if the AI uses the pfm or an sfm, the point is how realistic the smf is. For realistic i mean to verify that ai does not exceed the limits by doing unlikely things, in a nutshell make the ai less strong, in essence you have to do a specific check for each aircraft assigning them the possibility to fail random.

[Fri13]

Or ED could improve their engine and take advantage of multicore processors, using a couple of cores for the simulated AI in PFM. Current AI kills the immersion

 

As can be proven, DCS currently runs on one core (two threads, audio and everything else) so there is rest available for everything else.

 

What I know about threaded programming, it is extremely difficult if you try to get a one process to work on multiple. But if you have multiple processes, then you can run all of those on separate cores.

 

DCS should be split to multiple threads such way that there is clear split between ground, air, blue and red.

 

RED has separate AI from the BLUE. This means that BLUE AI can't see nor control or know what the RED does, and vice versa. It is one way of the Anti-Cheat for AI. As there is no AI that knows what otherside is doing.

 

Ground is separated from the Air units. Meaning, all the ground units are running on dedicated CPU core. The ground units includes all the MBT, APC, Infantry, AAA, SAM, ships etc. It is their line of sight calculations, communications, networking, radars etc.

 

The AIR units are all about the AI controlled aircrafts. Be it helicopter, be it a fixed wing, you name it, that one CPU core is running all those.

 

And then there are two dedicated CPU cores that are left for two things, two things only.

One is the player own aircraft, now the current DCS runs that, and everything else. But we leave everything else out of that single core. This is where all of the player aircraft engine, flight modeling etc is done.

 

And then the last dedicated CPU core is for missiles and shells. Everything that any unit launches, shoot or releases. It is own dedicated core that simulates missile flight modeling, the bomb impact zones, the bullet impact calculations, fragmentation etc. It does nothing else.

 

That is six cores/threads at optimal running. And minimum CPU would be quad-core. That means everything since 2007 when quad-core processors (or dual-core with Hyper Threading) processors came available. And if someone is running TODAY a DCS on such old CPU that is not quad core, they need to upgrade their computer ASAP to something modern. Sorry.

 

The DCS could then be running on 4 core computer:

 

Core #1 = Player aircraft

Core #2 = Blue + Red ground units

Core #3 = Red + Blue air units

Core #4 = Missiles, shells, all ammunition.

 

The whole DCS should be very modular, a server-client architecture, where all those dedicated areas under control of AI are clients, and the DCS itself is the player aircraft that includes the map, weather, and the core functions and the player aircraft itself at minimum. The weather can be then as well separated to own core as it is one program more to run.

These things could even be set for a external computer as server. So one sets a own dedicated server to other room to run DCS that handles map and weather and then ground units. While the player computer runs player aircraft and air units and ammunition.

 

The AI modeling could as well be set as own dedicated CPU core per side that means the AI aircrafts flight modeling can be more accurately modeled. And this should anyways be dynamic, so when the AI aircraft is far, far away. It doesn't need to be running at all by the flight modeling any other than its location, heading and speed is known. Nothing else really by default. It is in realm of simulation even when no flaps, thrust, wind affection etc is calculated. And when the AI aircraft is getting affected by something, then its simulation is leveled up dynamically. Meaning, when the AI aircrafts comes to fly 25 kilometers from the enemy aircraft or missile, its flight modeling is leveled up to SFM. When it is in 5-10 kilometers, it is leveled up to PFM level. Now the AI aircraft behaves with all the limitations and benefits when fighting against other aircrafts or avoiding a missile etc. A flight of 4 fighters flying in close formation is totally useless to be simulated in PFM level or even SFM level when there is no enemy nearby, they are not targeting or getting targeted by anything. They should be just a information that somewhere is such flight going around.

 

It doesn't matter if a operational aircraft is performing landing 200 kilometers from player location, no need to simulate the flight dynamics etc for that aircraft at all. It should be nothing else than the statics that aircraft X is about to land in 3min 41 seconds. And only if a player camera is jumped on that location, is that simulation flight dynamics etc puffed up a notch or two.

 

The flight dynamics should be very dynamically calculated. That is required already by a dynamic campaign. As we need to be able have a tens of thousands of units on the map all the time, all operating in their parameters as expected. And in war, most of the time is just waiting or moving. Only when there is a possibility that there comes a engagement, is there requirement to start simulating more accurately that what the unit can see, what it can't see and what is its condition.

 

This as well means dynamic change like if a player turns Radar On, then all units it paints in the 400km range will get a upgrade in simulation. Checking their attitude, that does the radar signal return to radar or not (are you notching etc) or is there a RWR warning from that active radar.

 

At this moment it looks that DCS is made in 100% full simulation. All the time the ground units are calculating LOS, all the time the AI is simulating the ground units tracks turning and wheels rolling regardless there is no one seeing it. All the time the flying missile is calculated every millisecond that should it see something or not etc. It is extremely heavy calculation load that means when you double the amount of units, you double the calculations. So going from 15 active units to 30 is heavy load.

 

When player now goes to 1 vs 1 with AI or 2 vs 2 where there is 3 AI, it shouldn't be at all a problem for a multi-core CPU. All those should be possible be ran by one player offline and yet at highest level of simulation.

 

But when our current counter measurements are even just a virtual dice rolling that does the missile get locked to one, not much can be expected from the close combat or BVR. The chaff is like non-existing in the DCS, just a roll of a dice, it is basically just like a flare but for radar missiles.

 

In PC games there needs to be cheating. simplifications etc. But it needs to be done smartly. Like now when the AI pilot is all knowing, all seeing, has perfect control and timing with SFM in their posession, it makes just all unfair.

 

The AI pilots would need reaction times increased, they should have line of sight with very tiny 2 degree cone of vision that they need to use to acquire a target on sky. They would be needed to make check randomly at different periods their instruments for a second and then return back to visually looking the target. If they can't find it, they need to perform search pattern for the possible areas. Separate the AI from the other aircrafts. Make them more trust their maneuvers and tactics and then try to find the targets. Have them a moral system and stress level. Firing long burst missing them would increase their stress level, make them making stupid moves, increase their reaction times or loosen then capability spot visually targets. Having 2 vs 1 or 3 vs 1 situation would put the AI in stressful moment where their surviving level would increase and they would flee instead attack without fear, thinking they can shoot all down one by one.

 

RANT

 

This is the same thing that is missing in the multiplayer, virtual pilots has infinite lives. There is no responsibility to return to base with aircraft intact and alive. There is no penalty from getting shot at or even worse, losing the aircraft or dying. The multiplayer servers should have limited lives for virtual pilots, if you waste example your 3 virtual lives or virtual aircrafts, you can't fly anymore on that day. It is 12 hour ban to fly on that server. Welcome back on next day, with little bit more fear and skill to avoid stupid moves, crashing while landing, trying to challenge other pilots in stupid ways etc.

 

That is something that even a Top Gun movie tried to show even in the start, that every active duty pilot has their stress levels and fear limits. Once you reach your limit, you know to get away.

 

One of the things that many doesn't get is that USA has always trusted to numbers instead technology. People just say "Oh, but the western tech is so much better that one fighter equals to three soviets". While in fact it is always multiple times more USA troops against the enemy. 3:1 or 4:1 ratios. If there is enemy 2 fighters, you send there 6 or 8 fighters. You circle them and you deny their strategies and tactics and you win by the numbers.

 

In ground combat it is common 3:1 ratio, but in urban combat it is 9:1 ratio that is needed to overcome the defending units. Meaning if you want to win a defensive platoon, you need to come with three platoons and with nine platoons in urban.

 

So what can we expect from even a Dog Fight between two fighters, when the mindset for the pilots is already thinking that they should easily win by the technology and the fighter pilot job is easy as driving a bus and dropping few bombs to ground as there is anyways just a one SAM in the 100 km area without any other challenge. Even DCS AAA doesn't know how to shoot effectively, and AI pilots even less know how to fly as they just love to do some simple predictable move that you can abuse as much you want to get them down, regardless their SFM limitation (this at least has changed somewhat with the new AI flight maneuvers added in this month).

[Fri13]

Maybe with more cores etc .. but the problem is not if the AI uses the pfm or an sfm, the point is how realistic the smf is. For realistic i mean to verify that ai does not exceed the limits by doing unlikely things, in a nutshell make the ai less strong, in essence you have to do a specific check for each aircraft assigning them the possibility to fail random.

 

Yeah. The SFM ain't the problem, the AI flying the aircraft is. We need to separate that 3D aircraft and its flight modeling from the AI pilot that is flying it. As once if we would get an AI pilot that has such features like:

 

- Reaction times (like F-14 Jester that pushes buttons with delays instead instantly input them)

- FOV cone and scanning time (head turning, area of focus / interest), limits in the angles and times to observe.

- Stress/fear level (how quickly to scan area and possibly "skip" a object there or when to escape the situation and not push)

- Physical condition (breathing, G-forces tolerance)

- Randomness to control inputs (not pulling straight perfect curve, but have some relaxing and harder pulls so generate more dynamic turns and rolls)

- Communication requirements, if one AI spots something, it is not known right away by everyone. It needs to be relayed and communicated to each others in the same channel.

 

And then of course have some dynamics in the flying, so example when AI flies in formation with you, they do not perfectly follow your inputs so they stay on with you. But if you pull slight up and down, they have "dampening" effect. And they might slightly change heading to get more separation or react to your changes in time. And to fix he common problem that AI has in formation flying because SFM and AFM modeling difference at fuel consumption and drag, if the AI is flying SFM and in the formation or near the AFM aircraft (same) at the same speed and similar weapon loadout, just match the amount of fuel between them. Yes, it is cheating but it would fix the huge problem where AI simply turns away middle of the mission because they ran out of fuel way too early by following human pilot and trying to keep up. Once the AI extends from formation to own flight maneuvering (combat, attack etc) then own calculation for that fuel flow is done.

[metzger]

Would be good if AI wingman can be improved in all aspects of flying in both rotary and fixed wing:

 

1. Be able to departure in formation

2. Be able to execute different approaches, straight in, overhead brake and so on.

3. Be able to do basic wingman responsibilities(scan the area,cover you, call lunches etc..) without having to tell him everything with F buttons

4. Generally make him better helicopter pilot as of now, he is horrible

5. Be able to execute certain ACM tactics a.k.a Fighting wing, loose deuce etc and reports his status e.g. "2 is offensive" when engaging offensively and so on

6. Make him possible to execute different A2G attacks correctly, use Final attack heading cone, correctly use stand off weapons, low angle, high angle dive bombing and rocket/gun runs etc...

7. Make it possible to pre-brief him e.g. Today we will travel in finger 4, once Fence IN switch to combat spread, if enemy is detected, loose deuce tactic to be applied. Then he follows the plan unless commanded to do something different..

[Fisherman82]

Would be good if AI wingman can be improved in all aspects of flying in both rotary and fixed wing:

 

1. Be able to departure in formation

2. Be able to execute different approaches, straight in, overhead brake and so on.

..

 

I totally agree with you just let me add some stuff:

1. Depart AND LAND in formation

2. Be willing to fly as low as his flight lead, when I fly at 10m above sea with my Viggen as the SWAF did then he hangs above afraid to come down.

3. Mimic the external light settings that the human player uses on his aircraft, that way you would have simple control of the wingmen so that they dont behave like idiots in this regard.

4. Report his fuel state

5. Report his position in relation to you if he has visual on you (usefull in combat)

6. Preserve his cool when locked by radars running in to targets, I dont know how to solve this but he has to be painted to fire a HARM he cant go defensive right away when he is locked.

 

And in general for newer modules like Viggen, Hornet and Harrier I hope that "engage my target" would be fixed so that it worked, it doesent now for me atleast. I mean if I have a designated target with my TGP or Maverick it would be good to give that to him with that command since there is no other way of coordinating a attack. Best would be if you could hand him the target and then not have him go for it right away until you give him a order to execute, that way you could plan with the TGP and then go in together coordinated for maximum effect in one pass. I have still not gotten a AI Hornet wingman to engage and destroy a target, I hope that is worked on in general. I now all these modules I mentioned are Early Access though so Im not complaining just saying what Im hoping for them.

[JIMJAM]

Just had a typical AI dogfight that has been this way forever. Say I have a P-51 light on fuel vs any of the German fighters say a 109 loaded with fuel on easy. I will skip the enemy's UFO flight model and ability to zoom climb and turn with no energy bleed for another time. So I get on his six and open up. Bits and pieces smoke and debri but he keeps going. Repeat this until my wrist is tired. On the other hand, he does a head on pass and with a quick blip, every bullet is that golden BB that hits every critical system I have. Wing rips off, pilot is dead, plane looks like swiss cheese.

I just flew with 3 aces against ONE "easy" 190 and he shot us all down. Pleeeeeze fix this.