Detection probabilities

[MBot]

Keep in mind that the visual detection model for air and ground AI are not necessary the same, at least they weren't until some months ago. AI aircraft got a quite advanced detection model (for air and ground targets) some when around the release of Black Shark, featuring a variety of factors such as target background, target movement, sensor type, field of search etc. Ground units on the other hand have had a very simplistic detection model of aircraft and other ground units. Visual detection was instant as soon as they were in range and in LOS.

 

That seemed to have changed a while ago though. A recent test I made showed a variation in detection distance of a fast approaching aircraft as seen from a SA-18 unit. That was total new to me, as the insta-detection model was still in place last summer when the Gazelle was released. The exact extent of the different ground AI detection model is still unclear to me.

[Fri13]

No, they don't. It is in the Detection.lua as well.

If you come from behind they "get aware" of your presence much later. (It's a factor).

The issue is that a Group "knows" all what Group Units know (talking to each other, like IRL) so a Group of tanks in a circle formation has all angles covered and thus spots you from every side.

 

That is the thing, a single group is a single AI. Once one see, everyone know.

It try to be like a real group where individuals notify other group members, but there is no good delays etc. Like radio allows to notify everyone at once, but that everyone is turning and finding target easily.

 

There was a good testing about popping up to group of 10 units how long it toke to spot and aim etc depending the skill level, distance, background etc. At fastest it was a 1-2 seconds regardless of any other parameter, the slowest unit in group was sometimes 10 seconds.

 

The findings was, you can't do a pop-up or slide attacks as once you are inside detection ring (firing range) then you are followed and tracked as nothing.

Even when doing a correct pop-up maneuvers, you should have at least 30-120 seconds before anyone even notices you (in armors) less than 1km when doing NOE. At further distances you should be able to hang there for minutes even as long target doesn't have a FLIR and ain't looking at you.

 

The same goes for a air attack, a single pop-up and even passing after firing (causing chaos by destruction) should be possible without anyone really noticing most cases when buttoned down.

 

Of course this puts tremendous requirements to AI as if there is infantry little further, they can hear you fairly well if at height or close.

 

A real issue is, once they've spotted you they track you no matter where you hide until you leave the detection distance.

 

EDIT and the damn time factor means in the end they will spot you, it is just a clock ticking down. You have LOS (through trees) it is just a matter of the time running out and they "spot" you...

It is not too unrealistic, as the longer you scan an area, the more likely you are to detect the enemy, but in combination with trees and buildings obstructing your view, but not the AI's it is pretty unfair sometimes.

 

The thing works in theory, just like the detection LUA says, but it is very very unrealistic as AI is scanning all the time all directions.

 

And it doesn't have reaction times or a changes to miss or look elsewhere.

 

And just as you say too, add the trees and anything else than ground and AI see you.

 

Ie. Once I was in situation where I was hovering behind a 30 floor building in a KA-50 to wait that AI AH-64 would turn.

 

When I thought so, I started a rapid pop-up to do the engagement. I didn't even have a change to get over the building before AI had fired a burst with autocannon to calculated impact point just above the building.

 

The playback was funny that just when my rotordisk came visible, shells hit my top fuselage and destroyed me. AI fired from a 2.5km distance a burst without ever even giving me a change to see it, or it to see me.

 

It is all just way too.... Cheating as the detection system doesn't really work.

 

Ie. I had old training mission with Mig-21 when it was released, starting behind a Su-25 from 3km distance. And it was annoying as everytime I launched a IR missile, AI pulled perfect maneuvers in the launch moment and flared.

 

No matter of distance or altitude difference etc, AI always knew missile is coming.

 

The same was with C-130, but that I suppose has a MWS and automatic flaring system in place, so you can't sneak to them ever.

 

But the AI takes a lot of fun away from combat part in DCS as it is all knowing God.

[Fri13]

Anecdotally, 30 years ago I walked into a main battle tank in the dark when it was switched off and covered up (i was looking for it in a navigation excercise and actually hit it and it hurt)

 

And if you turn the buggers on they make a racket you can hear for miles.

 

So moving, not moving should/could be a massive difference in detection for armoured vehicles.

Agree as stationary unit is very difficult to spot.

The main reasons to spot something are movement, reflections, tracks, sound, light and heat (smoke too).

 

I don't think that DCS can use any of those than a movement for the detection measurement.

And it could be used for humans too.

 

Meaning if a unit is stationary, make it partially transparent so it's spotting becomes far more difficult. Render the shadow normally.

Once unit starts moving, make it opaque so it is clearer to spot.

 

Tracks are something that we should get in fields or grass and sand. AI can't use those, but humans could.

 

Reflections I would just forget, and then just take sound and heat. So even an AI could detect noise. In ground units it is easy as those move slowly, fast movers the sound would come with a slight latency. And this would be mainly for infantry benefit. Heat is easy, if unit has FLIR, it helps if ground unit ain't going in time, or ain't cover.

 

A mission designers would get far more out with small differences here. As different units has different means to detect and be detected.

 

If we get dynamic random ground foliage, the transparent trick ain't needed from close distances (<2-5km). As we would need to spot them among rocks, bushes, individual trees etc. So closer we get, more dynamic clutter there is.

[shagrat]

That is the thing, a single group is a single AI. Once one see, everyone know.

It try to be like a real group where individuals notify other group members, but there is no good delays etc. Like radio allows to notify everyone at once, but that everyone is turning and finding target easily.

 

There was a good testing about popping up to group of 10 units how long it toke to spot and aim etc depending the skill level, distance, background etc. At fastest it was a 1-2 seconds regardless of any other parameter, the slowest unit in group was sometimes 10 seconds.

 

The findings was, you can't do a pop-up or slide attacks as once you are inside detection ring (firing range) then you are followed and tracked as nothing.

Even when doing a correct pop-up maneuvers, you should have at least 30-120 seconds before anyone even notices you (in armors) less than 1km when doing NOE. At further distances you should be able to hang there for minutes even as long target doesn't have a FLIR and ain't looking at you.

 

The same goes for a air attack, a single pop-up and even passing after firing (causing chaos by destruction) should be possible without anyone really noticing most cases when buttoned down.

 

Of course this puts tremendous requirements to AI as if there is infantry little further, they can hear you fairly well if at height or close.

 

 

 

The thing works in theory, just like the detection LUA says, but it is very very unrealistic as AI is scanning all the time all directions.

 

And it doesn't have reaction times or a changes to miss or look elsewhere.

 

And just as you say too, add the trees and anything else than ground and AI see you.

 

Ie. Once I was in situation where I was hovering behind a 30 floor building in a KA-50 to wait that AI AH-64 would turn.

 

When I thought so, I started a rapid pop-up to do the engagement. I didn't even have a change to get over the building before AI had fired a burst with autocannon to calculated impact point just above the building.

 

The playback was funny that just when my rotordisk came visible, shells hit my top fuselage and destroyed me. AI fired from a 2.5km distance a burst without ever even giving me a change to see it, or it to see me.

 

It is all just way too.... Cheating as the detection system doesn't really work.

 

Ie. I had old training mission with Mig-21 when it was released, starting behind a Su-25 from 3km distance. And it was annoying as everytime I launched a IR missile, AI pulled perfect maneuvers in the launch moment and flared.

 

No matter of distance or altitude difference etc, AI always knew missile is coming.

 

The same was with C-130, but that I suppose has a MWS and automatic flaring system in place, so you can't sneak to them ever.

 

But the AI takes a lot of fun away from combat part in DCS as it is all knowing God.

I am with you. That is the biggest issue. After they "spotted" you, they won't lose you. Buildings are not blocking LOS for that part, just the AI doesn't fire before your "boundary box" (the collision model around your aircraft to check collisions) is visible...

The detection.lua only define how long it takes to find you. Only if you leave the detection radius for a while they "lose sight". :(

[71st_AH Rob]

Agree as stationary unit is very difficult to spot.

The main reasons to spot something are movement, reflections, tracks, sound, light and heat (smoke too).

 

You missed a few:

 

Fieldcraft B-GL-392-009/FP-0003

CHAPTER 2 FINDING THE ENEMY

SECTION 1 DETECTION

 

WHY THINGS ARE SEEN

 

1. The ability to observe effectively is an acquired skill. Thingsare seen because they contrast with their surroundings in one way or another. There are 11 reasons why things are seen:

 

shape;

 

shadow;

 

silhouette;

 

movement;

 

spacing;

 

position;

 

texture;

 

colour;

 

scale;

 

noise; and

 

shine

 

You may wish to review the following training film to brush up your skills that I used way back when I was a Pl Comd at the Battle School:

 

 

Reflections (shine) would be a massive improvement in DCS World, both on the ground and in the air, it would be game changing. Let's hope that becomes a reality in 2.5. It could be simulated for AI by taking lighting conditions into account and apply a factor to spotting but

[Fri13]

You missed a few:

 

Fieldcraft B-GL-392-009/FP-0003

CHAPTER 2 FINDING THE ENEMY

SECTION 1 DETECTION

 

WHY THINGS ARE SEEN

 

1. The ability to observe effectively is an acquired skill. Thingsare seen because they contrast with their surroundings in one way or another. There are 11 reasons why things are seen:

 

shape;

 

shadow;

 

silhouette;

 

movement;

 

spacing;

 

position;

 

texture;

 

colour;

 

scale;

 

noise; and

 

shine

 

You may wish to review the following training film to brush up your skills that I used way back when I was a Pl Comd at the Battle School:

 

 

Reflections (shine) would be a massive improvement in DCS World, both on the ground and in the air, it would be game changing. Let's hope that becomes a reality in 2.5. It could be simulated for AI by taking lighting conditions into account and apply a factor to spotting but

Oh I wasn't talking about real life and having a lecture for that ;-)

 

Just the few simple things that should or could be in DCS level. As we are not a infantry simulation or fps game, so many are outside of the scope as we are focused to Air-Ground combat mainly, some extension to Ground-Ground combat with CA module.

 

Like we have reflections from cockpits, but how far distances?

 

We have a slight feature like afterburners visible from a 10km.

 

Yet we have so simple things missing like capability sneak with a Mig-21 to close range because so tiny profile compared others (findings from Project Donugh etc).

 

And yet with a 4K display I can spot a C-130 size aircraft from a 30+ km distance against clear sky, spot a F-15C from a 17-20km.

Or spot a T-55 from a 15-17km on typical Caucasus map terrain, while it is stationary.

 

So many unrealistic things and yet some realistic.

 

While I should be unable to spot a stationary MBT from air further than couple km if not camouflaged, when I could just walk on it to find it.

 

So mission creators are having a huge limitations to get a realistic situations where a single SAM site can be huge threat to aircrafts if not aware such being there (intelligence).

 

Like now the viggen capabilities to do electronic recon is something amazing, to let us fly and recon radars for estimated areas and analyze the signal for type etc.

 

Something mind blowing for missions as we can gather intelligence that doesn't require direct visual spotting!

 

Having a friend to fly viggen for CAS is now like a friend being a JTAC for A-10C.

 

And yet we can't have an helicopters operate realistic manner.

 

So many things are about spotting, that the normal infantry is required to do in normal daily warfare that makes DCS so much like a "chess game" is to real war.

 

Basic simple rules, lots of possibilities and challenges, yet nothing compared to real thing where a single soldier can reveal whole platoon position to enemy because single simple mistake.

 

So many things could be done with a random generator checks (does it see it? 7% probability for yes) and then same code makes situation where something should be very clear to spot and AI just doesn't.

 

So to find the balance, is just huge work if not possible.

[shagrat]

You missed a few:

 

Fieldcraft B-GL-392-009/FP-0003

CHAPTER 2 FINDING THE ENEMY

SECTION 1 DETECTION

 

WHY THINGS ARE SEEN

 

1. The ability to observe effectively is an acquired skill. Thingsare seen because they contrast with their surroundings in one way or another. There are 11 reasons why things are seen:

 

shape;

 

shadow;

 

silhouette;

 

movement;

 

spacing;

 

position;

 

texture;

 

colour;

 

scale;

 

noise; and

 

shine

 

You may wish to review the following training film to brush up your skills that I used way back when I was a Pl Comd at the Battle School:

 

 

Reflections (shine) would be a massive improvement in DCS World, both on the ground and in the air, it would be game changing. Let's hope that becomes a reality in 2.5. It could be simulated for AI by taking lighting conditions into account and apply a factor to spotting but

Windshields of Humvees already reflect sunlight...

[Fri13]

Windshields of Humvees already reflect sunlight...

Any other one with larger windows?

It has been long time since I have been against humvee in such lighting conditions....

[shagrat]

I need to test, which. Propably the newer more detailed models.

KAMAZ Truck, Ural, GAZ... Tigr, the German Fuchs with the protection up?

Best go in F7 view and use the mouse to find the reflection angle with sun at morning or afternoon.

Doesn't show very far but the glass gets really bright.

Where there is definitely nothing is the old pixelated Busses, VAZ car, 3307 and 3308 trucks.

 

And of course most combat vehicles have a matte finish and obscured glass, for the reason of stealth.

[Fri13]

I need to test, which. Propably the newer more detailed models.

KAMAZ Truck, Ural, GAZ... Tigr, the German Fuchs with the protection up?

Best go in F7 view and use the mouse to find the reflection angle with sun at morning or afternoon.

Doesn't show very far but the glass gets really bright.

Where there is definitely nothing is the old pixelated Busses, VAZ car, 3307 and 3308 trucks.

 

And of course most combat vehicles have a matte finish and obscured glass, for the reason of stealth.

Sure from a close range, I am more thinking to fly at altitude of couple km and see a reflection of sun. So you could spot something that way.

[shagrat]

There was a glimpse of this as WIP in one of the Nevada Update videos... Looked a bit overdone (Carpark like dozens of flashlights), but the general effect and approach was really cool and useful.

[FlightControl]

Tree height

 

Any view on what assumptions i take as the default height of the bounding box around a "forest" area?

 

3m, 2m, 5m?

[shagrat]

Any view on what assumptions i take as the default height of the bounding box around a "forest" area?

 

3m, 2m, 5m?

 

definitely 15-20m!!! Trees average height is from 20+ to 40+ meter depending on age, so a thick normal grown wood definitely will block LOS up to a height of 20m easily, before the branches are thinning out considerably.

 

Another question, would be a "border-box" of 10-15m on the forrest edges, where you may spot a vehicle more easily, even if partly concealed by trees (especially when moving).

 

Though we can just choose a box, a tad bit smaller than the forrest... :smartass:

[Fri13]

And not to forget that ground units view angle to sky past foremost trees is fairly high so the visibility behind trees goes fairly high quickly. So 20-30m is a lot in that sense, while little when just thinking a one tree.

[FlightControl]

definitely 15-20m!!! Trees average height is from 20+ to 40+ meter depending on age, so a thick normal grown wood definitely will block LOS up to a height of 20m easily, before the branches are thinning out considerably.

 

Another question, would be a "border-box" of 10-15m on the forrest edges, where you may spot a vehicle more easily, even if partly concealed by trees (especially when moving).

 

Though we can just choose a box, a tad bit smaller than the forrest... :smartass:

 

Hi Shagrat,

 

That is exactly what i did in the early prototype demo test mission :-) Drew a polygon around the forest area, but about 5 to 10 meters (some more) distance from the forest border... Note that if you would put the vehicles at the back outside of the forest, they would become "visible", although in real life you wouldn't see them. For that a ray casting algorihm is in the make (=not so easy as one would think from the first sight). I underestimated the complexity a bit, but still. Working on it.

 

Trying to get a simple model in place, not an accurate to life forest visibility simulation. But something that works better for most that what DCS provides out of the box, and it may dissatisfy some.

We have to keep in mind: DCS world runs on 1 CPU which is already overloaded with a complete simulation, and on top this logic ...

 

Sven

[shagrat]

Hi Shagrat,

 

That is exactly what i did in the early prototype demo test mission :-) Drew a polygon around the forest area, but about 5 to 10 meters (some more) distance from the forest border... Note that if you would put the vehicles at the back outside of the forest, they would become "visible", although in real life you wouldn't see them. For that a ray casting algorihm is in the make (=not so easy as one would think from the first sight). I underestimated the complexity a bit, but still. Working on it.

 

Trying to get a simple model in place, not an accurate to life forest visibility simulation. But something that works better for most that what DCS provides out of the box, and it may dissatisfy some.

We have to keep in mind: DCS world runs on 1 CPU which is already overloaded with a complete simulation, and on top this logic ...

 

Sven

I wouldnt go for the raytracing, may be check vector against the box to determine "in front"/"behind" the forest...?

[FlightControl]

I wouldnt go for the raytracing, may be check vector against the box to determine "in front"/"behind" the forest...?

 

No raytracing. Ray casting.

 

Basically what is done is to test if the "observation" vector between the recce and the target is "touching" one of the edges on one of the polygons that are in the bounding box of that observation vector. If it finds a hit, I know the object may be invisible.

 

Of course, an efficient and slick algorithm is needed to do that testing, so the fastest and least CPU invasive way. Trying to keep it as simple as possible, and as efficient as possible. If you're interested, get on slack and we can further discuss it. It is interesting stuff to make.

 

Sven

[FlightControl]

@ED,

 

Some friends have advised me to hold the development on the forest and village zones los problems, since these are likely to be resolved in future maps.

So, NTTR, has LOS on trees. Will other maps have LOS on trees also and will the caucasus map eventually also have tree and villages incorporated in LOS detection?

[DarkFire]

@ED,

 

Some friends have advised me to hold the development on the forest and village zones los problems, since these are likely to be resolved in future maps.

So, NTTR, has LOS on trees. Will other maps have LOS on trees also and will the caucasus map eventually also have tree and villages incorporated in LOS detection?

 

From what Wags has posted it appears that the Caucasus map is getting a much more comprehensive upgrade than a more detailed terrain mesh, and as the LOS technology exists already in the NTTR map, I would expect it to feature in all future maps, though this is just a guess.

[FlightControl]

From what Wags has posted it appears that the Caucasus map is getting a much more comprehensive upgrade than a more detailed terrain mesh, and as the LOS technology exists already in the NTTR map, I would expect it to feature in all future maps, though this is just a guess.

 

@Ed or somebody else, can clarity be brought on this question please?

Ill also post on another forum.

 

Fc

[FlightControl]

Okay ... Remember this thread?

 

I have a question, and it involves detection. Got in MOOSE this slick DETECTION classes...

 

I have been working on it and improving the mechanisms how the DETECTION classes "recognize" the objects it detects. And I've noticed something really strange...

 

It concerns the detection of ground objects.

There are basically two kinds of ground objects:

- The ones that emit radar signals.

- The ones that don't emit radar signals.

 

Targets can be detected in many ways:

- Visually

- Through RADAR

- Through RWR

- Through IR

- Through Data Link

- Through Optical Equipment

 

An MQ-9 Reaper is able to detect ground targets from very far, like 120km distance, knowing exactly its type name, distance etc.

This is the case when it detects targets using all available detection methods. So I went to search, why that is. Can't be.

 

The Reaper seems to be able to detect all targets from a far distance, when it uses the RWR detection method.

Now I think that is a bit of bollocks, how on earth can it detect a BMP-1 at 120km distance using RWR? A BMP does not emit radar.

 

So what I am going to do, is to check that if the target distance is over 15km, AND, if the unit is not emitting radar, that it can't be detected by a Unit using only the RWR detection method ...

 

FC

[vicx]

Flightcontrol, this confirms what I found via testing too.

 

A few years ago I found the Abstract RWR device in most aircraft was a bit too effective.

https://forums.eagle.ru/showthread.php?t=133111

 

Interesting thing from this old thread was a mention I made of MSF which is a Lua mission framework from the Russian speaking Forums. MSF had/has some very cool ideas in it. At the time I really liked the implementation of communications networks with unit/static object dependencies. There were other very strong elements in the framework that you might like to borrow for MOOSE. Automatic text to speech generation and other cool stuff.

 

I was also looking through my old files recently and found lua snippets I wrote for exporting live mission data to a Tacview log. My goal at the time was to leverage Tacview to visualise as much as possible what the sim was doing during a mission (like detections and targets).

 

I think a MOOSE class that exports data to Tacview could be very valuable for advanced mission development. Might be able to provide a gods eye view for mission designers including what the AI is thinking at any point in time.

 

Will keep looking for videos and pictures I produced in Tacview because pictures speak louder than words.

Edited August 15, 2017 by vicx
MSF not SSE

[FlightControl]

Yup. thanks!

[exil]

@ED,

 

Some friends have advised me to hold the development on the forest and village zones los problems, since these are likely to be resolved in future maps.

So, NTTR, has LOS on trees. Will other maps have LOS on trees also and will the caucasus map eventually also have tree and villages incorporated in LOS detection?

 

Um, from what i understand is that trees don't have LOS. They are just collideable. But still, enemy forces can "spot" through it if you're in the detection range. At least, when the Gazelle was released i tried a couple of manouvers with only the viviane sight sticking out of the treetops and was immediatelly spottet.

[Fri13]

There are some scripts about the detection. Like target distance, angle, speed, background and percentage of the unit visibility.

 

Yet the AI has 360* spherical vision constantly spotting everything in their engagement range with few factors to lower the speed of reaction like skill level.

 

--

I usually post from my phone so please excuse any typos, inappropriate punctuation and capitalization, missing words and general lack of cohesion and sense in my posts.....